Index: head/sys/dev/mpr/mpr.c =================================================================== --- head/sys/dev/mpr/mpr.c (revision 349848) +++ head/sys/dev/mpr/mpr.c (revision 349849) @@ -1,4025 +1,4030 @@ /*- * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2016 Avago Technologies * Copyright 2000-2020 Broadcom Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Broadcom Inc. (LSI) MPT-Fusion Host Adapter FreeBSD * */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT3 */ /* TODO Move headers to mprvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int mpr_diag_reset(struct mpr_softc *sc, int sleep_flag); static int mpr_init_queues(struct mpr_softc *sc); static void mpr_resize_queues(struct mpr_softc *sc); static int mpr_message_unit_reset(struct mpr_softc *sc, int sleep_flag); static int mpr_transition_operational(struct mpr_softc *sc); static int mpr_iocfacts_allocate(struct mpr_softc *sc, uint8_t attaching); static void mpr_iocfacts_free(struct mpr_softc *sc); static void mpr_startup(void *arg); static int mpr_send_iocinit(struct mpr_softc *sc); static int mpr_alloc_queues(struct mpr_softc *sc); static int mpr_alloc_hw_queues(struct mpr_softc *sc); static int mpr_alloc_replies(struct mpr_softc *sc); static int mpr_alloc_requests(struct mpr_softc *sc); static int mpr_alloc_nvme_prp_pages(struct mpr_softc *sc); static int mpr_attach_log(struct mpr_softc *sc); static __inline void mpr_complete_command(struct mpr_softc *sc, struct mpr_command *cm); static void mpr_dispatch_event(struct mpr_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *reply); static void mpr_config_complete(struct mpr_softc *sc, struct mpr_command *cm); static void mpr_periodic(void *); static int mpr_reregister_events(struct mpr_softc *sc); static void mpr_enqueue_request(struct mpr_softc *sc, struct mpr_command *cm); static int mpr_get_iocfacts(struct mpr_softc *sc, MPI2_IOC_FACTS_REPLY *facts); static int mpr_wait_db_ack(struct mpr_softc *sc, int timeout, int sleep_flag); static int mpr_debug_sysctl(SYSCTL_HANDLER_ARGS); static int mpr_dump_reqs(SYSCTL_HANDLER_ARGS); static void mpr_parse_debug(struct mpr_softc *sc, char *list); SYSCTL_NODE(_hw, OID_AUTO, mpr, CTLFLAG_RD, 0, "MPR Driver Parameters"); MALLOC_DEFINE(M_MPR, "mpr", "mpr driver memory"); /* * Do a "Diagnostic Reset" aka a hard reset. This should get the chip out of * any state and back to its initialization state machine. */ static char mpt2_reset_magic[] = { 0x00, 0x0f, 0x04, 0x0b, 0x02, 0x07, 0x0d }; /* * Added this union to smoothly convert le64toh cm->cm_desc.Words. * Compiler only supports uint64_t to be passed as an argument. * Otherwise it will throw this error: * "aggregate value used where an integer was expected" */ typedef union _reply_descriptor { u64 word; struct { u32 low; u32 high; } u; } reply_descriptor, request_descriptor; /* Rate limit chain-fail messages to 1 per minute */ static struct timeval mpr_chainfail_interval = { 60, 0 }; /* * sleep_flag can be either CAN_SLEEP or NO_SLEEP. * If this function is called from process context, it can sleep * and there is no harm to sleep, in case if this fuction is called * from Interrupt handler, we can not sleep and need NO_SLEEP flag set. * based on sleep flags driver will call either msleep, pause or DELAY. * msleep and pause are of same variant, but pause is used when mpr_mtx * is not hold by driver. */ static int mpr_diag_reset(struct mpr_softc *sc,int sleep_flag) { uint32_t reg; int i, error, tries = 0; uint8_t first_wait_done = FALSE; mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); /* Clear any pending interrupts */ mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* * Force NO_SLEEP for threads prohibited to sleep * e.a Thread from interrupt handler are prohibited to sleep. */ #if __FreeBSD_version >= 1000029 if (curthread->td_no_sleeping) #else //__FreeBSD_version < 1000029 if (curthread->td_pflags & TDP_NOSLEEPING) #endif //__FreeBSD_version >= 1000029 sleep_flag = NO_SLEEP; mpr_dprint(sc, MPR_INIT, "sequence start, sleep_flag=%d\n", sleep_flag); /* Push the magic sequence */ error = ETIMEDOUT; while (tries++ < 20) { for (i = 0; i < sizeof(mpt2_reset_magic); i++) mpr_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, mpt2_reset_magic[i]); /* wait 100 msec */ if (mtx_owned(&sc->mpr_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mpr_mtx, 0, "mprdiag", hz/10); else if (sleep_flag == CAN_SLEEP) pause("mprdiag", hz/10); else DELAY(100 * 1000); reg = mpr_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); if (reg & MPI2_DIAG_DIAG_WRITE_ENABLE) { error = 0; break; } } if (error) { mpr_dprint(sc, MPR_INIT, "sequence failed, error=%d, exit\n", error); return (error); } /* Send the actual reset. XXX need to refresh the reg? */ reg |= MPI2_DIAG_RESET_ADAPTER; mpr_dprint(sc, MPR_INIT, "sequence success, sending reset, reg= 0x%x\n", reg); mpr_regwrite(sc, MPI2_HOST_DIAGNOSTIC_OFFSET, reg); /* Wait up to 300 seconds in 50ms intervals */ error = ETIMEDOUT; for (i = 0; i < 6000; i++) { /* * Wait 50 msec. If this is the first time through, wait 256 * msec to satisfy Diag Reset timing requirements. */ if (first_wait_done) { if (mtx_owned(&sc->mpr_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mpr_mtx, 0, "mprdiag", hz/20); else if (sleep_flag == CAN_SLEEP) pause("mprdiag", hz/20); else DELAY(50 * 1000); } else { DELAY(256 * 1000); first_wait_done = TRUE; } /* * Check for the RESET_ADAPTER bit to be cleared first, then * wait for the RESET state to be cleared, which takes a little * longer. */ reg = mpr_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); if (reg & MPI2_DIAG_RESET_ADAPTER) { continue; } reg = mpr_regread(sc, MPI2_DOORBELL_OFFSET); if ((reg & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_RESET) { error = 0; break; } } if (error) { mpr_dprint(sc, MPR_INIT, "reset failed, error= %d, exit\n", error); return (error); } mpr_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 0x0); mpr_dprint(sc, MPR_INIT, "diag reset success, exit\n"); return (0); } static int mpr_message_unit_reset(struct mpr_softc *sc, int sleep_flag) { int error; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); error = 0; mpr_regwrite(sc, MPI2_DOORBELL_OFFSET, MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET << MPI2_DOORBELL_FUNCTION_SHIFT); if (mpr_wait_db_ack(sc, 5, sleep_flag) != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Doorbell handshake failed\n"); error = ETIMEDOUT; } mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); return (error); } static int mpr_transition_ready(struct mpr_softc *sc) { uint32_t reg, state; int error, tries = 0; int sleep_flags; MPR_FUNCTRACE(sc); /* If we are in attach call, do not sleep */ sleep_flags = (sc->mpr_flags & MPR_FLAGS_ATTACH_DONE) ? CAN_SLEEP : NO_SLEEP; error = 0; mpr_dprint(sc, MPR_INIT, "%s entered, sleep_flags= %d\n", __func__, sleep_flags); while (tries++ < 1200) { reg = mpr_regread(sc, MPI2_DOORBELL_OFFSET); mpr_dprint(sc, MPR_INIT, " Doorbell= 0x%x\n", reg); /* * Ensure the IOC is ready to talk. If it's not, try * resetting it. */ if (reg & MPI2_DOORBELL_USED) { mpr_dprint(sc, MPR_INIT, " Not ready, sending diag " "reset\n"); mpr_diag_reset(sc, sleep_flags); DELAY(50000); continue; } /* Is the adapter owned by another peer? */ if ((reg & MPI2_DOORBELL_WHO_INIT_MASK) == (MPI2_WHOINIT_PCI_PEER << MPI2_DOORBELL_WHO_INIT_SHIFT)) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "IOC is under the " "control of another peer host, aborting " "initialization.\n"); error = ENXIO; break; } state = reg & MPI2_IOC_STATE_MASK; if (state == MPI2_IOC_STATE_READY) { /* Ready to go! */ error = 0; break; } else if (state == MPI2_IOC_STATE_FAULT) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "IOC in fault " "state 0x%x, resetting\n", state & MPI2_DOORBELL_FAULT_CODE_MASK); mpr_diag_reset(sc, sleep_flags); } else if (state == MPI2_IOC_STATE_OPERATIONAL) { /* Need to take ownership */ mpr_message_unit_reset(sc, sleep_flags); } else if (state == MPI2_IOC_STATE_RESET) { /* Wait a bit, IOC might be in transition */ mpr_dprint(sc, MPR_INIT|MPR_FAULT, "IOC in unexpected reset state\n"); } else { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "IOC in unknown state 0x%x\n", state); error = EINVAL; break; } /* Wait 50ms for things to settle down. */ DELAY(50000); } if (error) mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Cannot transition IOC to ready\n"); mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); return (error); } static int mpr_transition_operational(struct mpr_softc *sc) { uint32_t reg, state; int error; MPR_FUNCTRACE(sc); error = 0; reg = mpr_regread(sc, MPI2_DOORBELL_OFFSET); mpr_dprint(sc, MPR_INIT, "%s entered, Doorbell= 0x%x\n", __func__, reg); state = reg & MPI2_IOC_STATE_MASK; if (state != MPI2_IOC_STATE_READY) { mpr_dprint(sc, MPR_INIT, "IOC not ready\n"); if ((error = mpr_transition_ready(sc)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "failed to transition ready, exit\n"); return (error); } } error = mpr_send_iocinit(sc); mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); return (error); } static void mpr_resize_queues(struct mpr_softc *sc) { u_int reqcr, prireqcr, maxio, sges_per_frame, chain_seg_size; /* * Size the queues. Since the reply queues always need one free * entry, we'll deduct one reply message here. The LSI documents * suggest instead to add a count to the request queue, but I think * that it's better to deduct from reply queue. */ prireqcr = MAX(1, sc->max_prireqframes); prireqcr = MIN(prireqcr, sc->facts->HighPriorityCredit); reqcr = MAX(2, sc->max_reqframes); reqcr = MIN(reqcr, sc->facts->RequestCredit); sc->num_reqs = prireqcr + reqcr; sc->num_prireqs = prireqcr; sc->num_replies = MIN(sc->max_replyframes + sc->max_evtframes, sc->facts->MaxReplyDescriptorPostQueueDepth) - 1; /* Store the request frame size in bytes rather than as 32bit words */ sc->reqframesz = sc->facts->IOCRequestFrameSize * 4; /* * Gen3 and beyond uses the IOCMaxChainSegmentSize from IOC Facts to * get the size of a Chain Frame. Previous versions use the size as a * Request Frame for the Chain Frame size. If IOCMaxChainSegmentSize * is 0, use the default value. The IOCMaxChainSegmentSize is the * number of 16-byte elelements that can fit in a Chain Frame, which is * the size of an IEEE Simple SGE. */ if (sc->facts->MsgVersion >= MPI2_VERSION_02_05) { chain_seg_size = htole16(sc->facts->IOCMaxChainSegmentSize); if (chain_seg_size == 0) chain_seg_size = MPR_DEFAULT_CHAIN_SEG_SIZE; sc->chain_frame_size = chain_seg_size * MPR_MAX_CHAIN_ELEMENT_SIZE; } else { sc->chain_frame_size = sc->reqframesz; } /* * Max IO Size is Page Size * the following: * ((SGEs per frame - 1 for chain element) * Max Chain Depth) * + 1 for no chain needed in last frame * * If user suggests a Max IO size to use, use the smaller of the * user's value and the calculated value as long as the user's * value is larger than 0. The user's value is in pages. */ sges_per_frame = sc->chain_frame_size/sizeof(MPI2_IEEE_SGE_SIMPLE64)-1; maxio = (sges_per_frame * sc->facts->MaxChainDepth + 1) * PAGE_SIZE; /* * If I/O size limitation requested then use it and pass up to CAM. * If not, use MAXPHYS as an optimization hint, but report HW limit. */ if (sc->max_io_pages > 0) { maxio = min(maxio, sc->max_io_pages * PAGE_SIZE); sc->maxio = maxio; } else { sc->maxio = maxio; maxio = min(maxio, MAXPHYS); } sc->num_chains = (maxio / PAGE_SIZE + sges_per_frame - 2) / sges_per_frame * reqcr; if (sc->max_chains > 0 && sc->max_chains < sc->num_chains) sc->num_chains = sc->max_chains; /* * Figure out the number of MSIx-based queues. If the firmware or * user has done something crazy and not allowed enough credit for * the queues to be useful then don't enable multi-queue. */ if (sc->facts->MaxMSIxVectors < 2) sc->msi_msgs = 1; if (sc->msi_msgs > 1) { sc->msi_msgs = MIN(sc->msi_msgs, mp_ncpus); sc->msi_msgs = MIN(sc->msi_msgs, sc->facts->MaxMSIxVectors); if (sc->num_reqs / sc->msi_msgs < 2) sc->msi_msgs = 1; } mpr_dprint(sc, MPR_INIT, "Sized queues to q=%d reqs=%d replies=%d\n", sc->msi_msgs, sc->num_reqs, sc->num_replies); } /* * This is called during attach and when re-initializing due to a Diag Reset. * IOC Facts is used to allocate many of the structures needed by the driver. * If called from attach, de-allocation is not required because the driver has * not allocated any structures yet, but if called from a Diag Reset, previously * allocated structures based on IOC Facts will need to be freed and re- * allocated bases on the latest IOC Facts. */ static int mpr_iocfacts_allocate(struct mpr_softc *sc, uint8_t attaching) { int error; Mpi2IOCFactsReply_t saved_facts; uint8_t saved_mode, reallocating; mpr_dprint(sc, MPR_INIT|MPR_TRACE, "%s entered\n", __func__); /* Save old IOC Facts and then only reallocate if Facts have changed */ if (!attaching) { bcopy(sc->facts, &saved_facts, sizeof(MPI2_IOC_FACTS_REPLY)); } /* * Get IOC Facts. In all cases throughout this function, panic if doing * a re-initialization and only return the error if attaching so the OS * can handle it. */ if ((error = mpr_get_iocfacts(sc, sc->facts)) != 0) { if (attaching) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Failed to get " "IOC Facts with error %d, exit\n", error); return (error); } else { panic("%s failed to get IOC Facts with error %d\n", __func__, error); } } MPR_DPRINT_PAGE(sc, MPR_XINFO, iocfacts, sc->facts); snprintf(sc->fw_version, sizeof(sc->fw_version), "%02d.%02d.%02d.%02d", sc->facts->FWVersion.Struct.Major, sc->facts->FWVersion.Struct.Minor, sc->facts->FWVersion.Struct.Unit, sc->facts->FWVersion.Struct.Dev); mpr_dprint(sc, MPR_INFO, "Firmware: %s, Driver: %s\n", sc->fw_version, MPR_DRIVER_VERSION); mpr_dprint(sc, MPR_INFO, "IOCCapabilities: %b\n", sc->facts->IOCCapabilities, "\20" "\3ScsiTaskFull" "\4DiagTrace" "\5SnapBuf" "\6ExtBuf" "\7EEDP" "\10BiDirTarg" "\11Multicast" "\14TransRetry" "\15IR" "\16EventReplay" "\17RaidAccel" "\20MSIXIndex" "\21HostDisc" "\22FastPath" "\23RDPQArray" "\24AtomicReqDesc" "\25PCIeSRIOV"); /* * If the chip doesn't support event replay then a hard reset will be * required to trigger a full discovery. Do the reset here then * retransition to Ready. A hard reset might have already been done, * but it doesn't hurt to do it again. Only do this if attaching, not * for a Diag Reset. */ if (attaching && ((sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY) == 0)) { mpr_dprint(sc, MPR_INIT, "No event replay, resetting\n"); mpr_diag_reset(sc, NO_SLEEP); if ((error = mpr_transition_ready(sc)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Failed to " "transition to ready with error %d, exit\n", error); return (error); } } /* * Set flag if IR Firmware is loaded. If the RAID Capability has * changed from the previous IOC Facts, log a warning, but only if * checking this after a Diag Reset and not during attach. */ saved_mode = sc->ir_firmware; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID) sc->ir_firmware = 1; if (!attaching) { if (sc->ir_firmware != saved_mode) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "new IR/IT mode " "in IOC Facts does not match previous mode\n"); } } /* Only deallocate and reallocate if relevant IOC Facts have changed */ reallocating = FALSE; sc->mpr_flags &= ~MPR_FLAGS_REALLOCATED; if ((!attaching) && ((saved_facts.MsgVersion != sc->facts->MsgVersion) || (saved_facts.HeaderVersion != sc->facts->HeaderVersion) || (saved_facts.MaxChainDepth != sc->facts->MaxChainDepth) || (saved_facts.RequestCredit != sc->facts->RequestCredit) || (saved_facts.ProductID != sc->facts->ProductID) || (saved_facts.IOCCapabilities != sc->facts->IOCCapabilities) || (saved_facts.IOCRequestFrameSize != sc->facts->IOCRequestFrameSize) || (saved_facts.IOCMaxChainSegmentSize != sc->facts->IOCMaxChainSegmentSize) || (saved_facts.MaxTargets != sc->facts->MaxTargets) || (saved_facts.MaxSasExpanders != sc->facts->MaxSasExpanders) || (saved_facts.MaxEnclosures != sc->facts->MaxEnclosures) || (saved_facts.HighPriorityCredit != sc->facts->HighPriorityCredit) || (saved_facts.MaxReplyDescriptorPostQueueDepth != sc->facts->MaxReplyDescriptorPostQueueDepth) || (saved_facts.ReplyFrameSize != sc->facts->ReplyFrameSize) || (saved_facts.MaxVolumes != sc->facts->MaxVolumes) || (saved_facts.MaxPersistentEntries != sc->facts->MaxPersistentEntries))) { reallocating = TRUE; /* Record that we reallocated everything */ sc->mpr_flags |= MPR_FLAGS_REALLOCATED; } /* * Some things should be done if attaching or re-allocating after a Diag * Reset, but are not needed after a Diag Reset if the FW has not * changed. */ if (attaching || reallocating) { /* * Check if controller supports FW diag buffers and set flag to * enable each type. */ if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_DIAG_TRACE_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_TRACE]. enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_SNAPSHOT_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_SNAPSHOT]. enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EXTENDED_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_EXTENDED]. enabled = TRUE; /* * Set flags for some supported items. */ if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EEDP) sc->eedp_enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_TLR) sc->control_TLR = TRUE; if ((sc->facts->IOCCapabilities & MPI26_IOCFACTS_CAPABILITY_ATOMIC_REQ) && (sc->mpr_flags & MPR_FLAGS_SEA_IOC)) sc->atomic_desc_capable = TRUE; mpr_resize_queues(sc); /* * Initialize all Tail Queues */ TAILQ_INIT(&sc->req_list); TAILQ_INIT(&sc->high_priority_req_list); TAILQ_INIT(&sc->chain_list); TAILQ_INIT(&sc->prp_page_list); TAILQ_INIT(&sc->tm_list); } /* * If doing a Diag Reset and the FW is significantly different * (reallocating will be set above in IOC Facts comparison), then all * buffers based on the IOC Facts will need to be freed before they are * reallocated. */ if (reallocating) { mpr_iocfacts_free(sc); mprsas_realloc_targets(sc, saved_facts.MaxTargets + saved_facts.MaxVolumes); } /* * Any deallocation has been completed. Now start reallocating * if needed. Will only need to reallocate if attaching or if the new * IOC Facts are different from the previous IOC Facts after a Diag * Reset. Targets have already been allocated above if needed. */ error = 0; while (attaching || reallocating) { if ((error = mpr_alloc_hw_queues(sc)) != 0) break; if ((error = mpr_alloc_replies(sc)) != 0) break; if ((error = mpr_alloc_requests(sc)) != 0) break; if ((error = mpr_alloc_queues(sc)) != 0) break; break; } if (error) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Failed to alloc queues with error %d\n", error); mpr_free(sc); return (error); } /* Always initialize the queues */ bzero(sc->free_queue, sc->fqdepth * 4); mpr_init_queues(sc); /* * Always get the chip out of the reset state, but only panic if not * attaching. If attaching and there is an error, that is handled by * the OS. */ error = mpr_transition_operational(sc); if (error != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Failed to " "transition to operational with error %d\n", error); mpr_free(sc); return (error); } /* * Finish the queue initialization. * These are set here instead of in mpr_init_queues() because the * IOC resets these values during the state transition in * mpr_transition_operational(). The free index is set to 1 * because the corresponding index in the IOC is set to 0, and the * IOC treats the queues as full if both are set to the same value. * Hence the reason that the queue can't hold all of the possible * replies. */ sc->replypostindex = 0; mpr_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); mpr_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, 0); /* * Attach the subsystems so they can prepare their event masks. * XXX Should be dynamic so that IM/IR and user modules can attach */ error = 0; while (attaching) { mpr_dprint(sc, MPR_INIT, "Attaching subsystems\n"); if ((error = mpr_attach_log(sc)) != 0) break; if ((error = mpr_attach_sas(sc)) != 0) break; if ((error = mpr_attach_user(sc)) != 0) break; break; } if (error) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Failed to attach all subsystems: error %d\n", error); mpr_free(sc); return (error); } /* * XXX If the number of MSI-X vectors changes during re-init, this * won't see it and adjust. */ if (attaching && (error = mpr_pci_setup_interrupts(sc)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Failed to setup interrupts\n"); mpr_free(sc); return (error); } return (error); } /* * This is called if memory is being free (during detach for example) and when * buffers need to be reallocated due to a Diag Reset. */ static void mpr_iocfacts_free(struct mpr_softc *sc) { struct mpr_command *cm; int i; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); if (sc->free_busaddr != 0) bus_dmamap_unload(sc->queues_dmat, sc->queues_map); if (sc->free_queue != NULL) bus_dmamem_free(sc->queues_dmat, sc->free_queue, sc->queues_map); if (sc->queues_dmat != NULL) bus_dma_tag_destroy(sc->queues_dmat); if (sc->chain_frames != NULL) { bus_dmamap_unload(sc->chain_dmat, sc->chain_map); bus_dmamem_free(sc->chain_dmat, sc->chain_frames, sc->chain_map); } if (sc->chain_dmat != NULL) bus_dma_tag_destroy(sc->chain_dmat); if (sc->sense_busaddr != 0) bus_dmamap_unload(sc->sense_dmat, sc->sense_map); if (sc->sense_frames != NULL) bus_dmamem_free(sc->sense_dmat, sc->sense_frames, sc->sense_map); if (sc->sense_dmat != NULL) bus_dma_tag_destroy(sc->sense_dmat); if (sc->prp_page_busaddr != 0) bus_dmamap_unload(sc->prp_page_dmat, sc->prp_page_map); if (sc->prp_pages != NULL) bus_dmamem_free(sc->prp_page_dmat, sc->prp_pages, sc->prp_page_map); if (sc->prp_page_dmat != NULL) bus_dma_tag_destroy(sc->prp_page_dmat); if (sc->reply_busaddr != 0) bus_dmamap_unload(sc->reply_dmat, sc->reply_map); if (sc->reply_frames != NULL) bus_dmamem_free(sc->reply_dmat, sc->reply_frames, sc->reply_map); if (sc->reply_dmat != NULL) bus_dma_tag_destroy(sc->reply_dmat); if (sc->req_busaddr != 0) bus_dmamap_unload(sc->req_dmat, sc->req_map); if (sc->req_frames != NULL) bus_dmamem_free(sc->req_dmat, sc->req_frames, sc->req_map); if (sc->req_dmat != NULL) bus_dma_tag_destroy(sc->req_dmat); if (sc->chains != NULL) free(sc->chains, M_MPR); if (sc->prps != NULL) free(sc->prps, M_MPR); if (sc->commands != NULL) { for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; bus_dmamap_destroy(sc->buffer_dmat, cm->cm_dmamap); } free(sc->commands, M_MPR); } if (sc->buffer_dmat != NULL) bus_dma_tag_destroy(sc->buffer_dmat); mpr_pci_free_interrupts(sc); free(sc->queues, M_MPR); sc->queues = NULL; } /* * The terms diag reset and hard reset are used interchangeably in the MPI * docs to mean resetting the controller chip. In this code diag reset * cleans everything up, and the hard reset function just sends the reset * sequence to the chip. This should probably be refactored so that every * subsystem gets a reset notification of some sort, and can clean up * appropriately. */ int mpr_reinit(struct mpr_softc *sc) { int error; struct mprsas_softc *sassc; sassc = sc->sassc; MPR_FUNCTRACE(sc); mtx_assert(&sc->mpr_mtx, MA_OWNED); mpr_dprint(sc, MPR_INIT|MPR_INFO, "Reinitializing controller\n"); if (sc->mpr_flags & MPR_FLAGS_DIAGRESET) { mpr_dprint(sc, MPR_INIT, "Reset already in progress\n"); return 0; } /* * Make sure the completion callbacks can recognize they're getting * a NULL cm_reply due to a reset. */ sc->mpr_flags |= MPR_FLAGS_DIAGRESET; /* * Mask interrupts here. */ mpr_dprint(sc, MPR_INIT, "Masking interrupts and resetting\n"); mpr_mask_intr(sc); error = mpr_diag_reset(sc, CAN_SLEEP); if (error != 0) { panic("%s hard reset failed with error %d\n", __func__, error); } /* Restore the PCI state, including the MSI-X registers */ mpr_pci_restore(sc); /* Give the I/O subsystem special priority to get itself prepared */ mprsas_handle_reinit(sc); /* * Get IOC Facts and allocate all structures based on this information. * The attach function will also call mpr_iocfacts_allocate at startup. * If relevant values have changed in IOC Facts, this function will free * all of the memory based on IOC Facts and reallocate that memory. */ if ((error = mpr_iocfacts_allocate(sc, FALSE)) != 0) { panic("%s IOC Facts based allocation failed with error %d\n", __func__, error); } /* * Mapping structures will be re-allocated after getting IOC Page8, so * free these structures here. */ mpr_mapping_exit(sc); /* * The static page function currently read is IOC Page8. Others can be * added in future. It's possible that the values in IOC Page8 have * changed after a Diag Reset due to user modification, so always read * these. Interrupts are masked, so unmask them before getting config * pages. */ mpr_unmask_intr(sc); sc->mpr_flags &= ~MPR_FLAGS_DIAGRESET; mpr_base_static_config_pages(sc); /* * Some mapping info is based in IOC Page8 data, so re-initialize the * mapping tables. */ mpr_mapping_initialize(sc); /* * Restart will reload the event masks clobbered by the reset, and * then enable the port. */ mpr_reregister_events(sc); /* the end of discovery will release the simq, so we're done. */ mpr_dprint(sc, MPR_INIT|MPR_XINFO, "Finished sc %p post %u free %u\n", sc, sc->replypostindex, sc->replyfreeindex); mprsas_release_simq_reinit(sassc); mpr_dprint(sc, MPR_INIT, "%s exit error= %d\n", __func__, error); return 0; } /* Wait for the chip to ACK a word that we've put into its FIFO * Wait for seconds. In single loop wait for busy loop * for 500 microseconds. * Total is [ 0.5 * (2000 * ) ] in miliseconds. * */ static int mpr_wait_db_ack(struct mpr_softc *sc, int timeout, int sleep_flag) { u32 cntdn, count; u32 int_status; u32 doorbell; count = 0; cntdn = (sleep_flag == CAN_SLEEP) ? 1000*timeout : 2000*timeout; do { int_status = mpr_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); if (!(int_status & MPI2_HIS_SYS2IOC_DB_STATUS)) { mpr_dprint(sc, MPR_TRACE, "%s: successful count(%d), " "timeout(%d)\n", __func__, count, timeout); return 0; } else if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) { doorbell = mpr_regread(sc, MPI2_DOORBELL_OFFSET); if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) { mpr_dprint(sc, MPR_FAULT, "fault_state(0x%04x)!\n", doorbell); return (EFAULT); } } else if (int_status == 0xFFFFFFFF) goto out; /* * If it can sleep, sleep for 1 milisecond, else busy loop for * 0.5 milisecond */ if (mtx_owned(&sc->mpr_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mpr_mtx, 0, "mprdba", hz/1000); else if (sleep_flag == CAN_SLEEP) pause("mprdba", hz/1000); else DELAY(500); count++; } while (--cntdn); out: mpr_dprint(sc, MPR_FAULT, "%s: failed due to timeout count(%d), " "int_status(%x)!\n", __func__, count, int_status); return (ETIMEDOUT); } /* Wait for the chip to signal that the next word in its FIFO can be fetched */ static int mpr_wait_db_int(struct mpr_softc *sc) { int retry; for (retry = 0; retry < MPR_DB_MAX_WAIT; retry++) { if ((mpr_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET) & MPI2_HIS_IOC2SYS_DB_STATUS) != 0) return (0); DELAY(2000); } return (ETIMEDOUT); } /* Step through the synchronous command state machine, i.e. "Doorbell mode" */ static int mpr_request_sync(struct mpr_softc *sc, void *req, MPI2_DEFAULT_REPLY *reply, int req_sz, int reply_sz, int timeout) { uint32_t *data32; uint16_t *data16; int i, count, ioc_sz, residual; int sleep_flags = CAN_SLEEP; #if __FreeBSD_version >= 1000029 if (curthread->td_no_sleeping) #else //__FreeBSD_version < 1000029 if (curthread->td_pflags & TDP_NOSLEEPING) #endif //__FreeBSD_version >= 1000029 sleep_flags = NO_SLEEP; /* Step 1 */ mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* Step 2 */ if (mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) return (EBUSY); /* Step 3 * Announce that a message is coming through the doorbell. Messages * are pushed at 32bit words, so round up if needed. */ count = (req_sz + 3) / 4; mpr_regwrite(sc, MPI2_DOORBELL_OFFSET, (MPI2_FUNCTION_HANDSHAKE << MPI2_DOORBELL_FUNCTION_SHIFT) | (count << MPI2_DOORBELL_ADD_DWORDS_SHIFT)); /* Step 4 */ if (mpr_wait_db_int(sc) || (mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) == 0) { mpr_dprint(sc, MPR_FAULT, "Doorbell failed to activate\n"); return (ENXIO); } mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); if (mpr_wait_db_ack(sc, 5, sleep_flags) != 0) { mpr_dprint(sc, MPR_FAULT, "Doorbell handshake failed\n"); return (ENXIO); } /* Step 5 */ /* Clock out the message data synchronously in 32-bit dwords*/ data32 = (uint32_t *)req; for (i = 0; i < count; i++) { mpr_regwrite(sc, MPI2_DOORBELL_OFFSET, htole32(data32[i])); if (mpr_wait_db_ack(sc, 5, sleep_flags) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout while writing doorbell\n"); return (ENXIO); } } /* Step 6 */ /* Clock in the reply in 16-bit words. The total length of the * message is always in the 4th byte, so clock out the first 2 words * manually, then loop the rest. */ data16 = (uint16_t *)reply; if (mpr_wait_db_int(sc) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout reading doorbell 0\n"); return (ENXIO); } data16[0] = mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); if (mpr_wait_db_int(sc) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout reading doorbell 1\n"); return (ENXIO); } data16[1] = mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* Number of 32bit words in the message */ ioc_sz = reply->MsgLength; /* * Figure out how many 16bit words to clock in without overrunning. * The precision loss with dividing reply_sz can safely be * ignored because the messages can only be multiples of 32bits. */ residual = 0; count = MIN((reply_sz / 4), ioc_sz) * 2; if (count < ioc_sz * 2) { residual = ioc_sz * 2 - count; mpr_dprint(sc, MPR_ERROR, "Driver error, throwing away %d " "residual message words\n", residual); } for (i = 2; i < count; i++) { if (mpr_wait_db_int(sc) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout reading doorbell %d\n", i); return (ENXIO); } data16[i] = mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); } /* * Pull out residual words that won't fit into the provided buffer. * This keeps the chip from hanging due to a driver programming * error. */ while (residual--) { if (mpr_wait_db_int(sc) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout reading doorbell\n"); return (ENXIO); } (void)mpr_regread(sc, MPI2_DOORBELL_OFFSET); mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); } /* Step 7 */ if (mpr_wait_db_int(sc) != 0) { mpr_dprint(sc, MPR_FAULT, "Timeout waiting to exit doorbell\n"); return (ENXIO); } if (mpr_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) mpr_dprint(sc, MPR_FAULT, "Warning, doorbell still active\n"); mpr_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); return (0); } static void mpr_enqueue_request(struct mpr_softc *sc, struct mpr_command *cm) { request_descriptor rd; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_TRACE, "SMID %u cm %p ccb %p\n", cm->cm_desc.Default.SMID, cm, cm->cm_ccb); if (sc->mpr_flags & MPR_FLAGS_ATTACH_DONE && !(sc->mpr_flags & MPR_FLAGS_SHUTDOWN)) mtx_assert(&sc->mpr_mtx, MA_OWNED); if (++sc->io_cmds_active > sc->io_cmds_highwater) sc->io_cmds_highwater++; KASSERT(cm->cm_state == MPR_CM_STATE_BUSY, ("command not busy\n")); cm->cm_state = MPR_CM_STATE_INQUEUE; if (sc->atomic_desc_capable) { rd.u.low = cm->cm_desc.Words.Low; mpr_regwrite(sc, MPI26_ATOMIC_REQUEST_DESCRIPTOR_POST_OFFSET, rd.u.low); } else { rd.u.low = cm->cm_desc.Words.Low; rd.u.high = cm->cm_desc.Words.High; rd.word = htole64(rd.word); mpr_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_LOW_OFFSET, rd.u.low); mpr_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_HIGH_OFFSET, rd.u.high); } } /* * Just the FACTS, ma'am. */ static int mpr_get_iocfacts(struct mpr_softc *sc, MPI2_IOC_FACTS_REPLY *facts) { MPI2_DEFAULT_REPLY *reply; MPI2_IOC_FACTS_REQUEST request; int error, req_sz, reply_sz; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); req_sz = sizeof(MPI2_IOC_FACTS_REQUEST); reply_sz = sizeof(MPI2_IOC_FACTS_REPLY); reply = (MPI2_DEFAULT_REPLY *)facts; bzero(&request, req_sz); request.Function = MPI2_FUNCTION_IOC_FACTS; error = mpr_request_sync(sc, &request, reply, req_sz, reply_sz, 5); mpr_dprint(sc, MPR_INIT, "%s exit, error= %d\n", __func__, error); return (error); } static int mpr_send_iocinit(struct mpr_softc *sc) { MPI2_IOC_INIT_REQUEST init; MPI2_DEFAULT_REPLY reply; int req_sz, reply_sz, error; struct timeval now; uint64_t time_in_msec; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); /* Do a quick sanity check on proper initialization */ if ((sc->pqdepth == 0) || (sc->fqdepth == 0) || (sc->reqframesz == 0) || (sc->replyframesz == 0)) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Driver not fully initialized for IOCInit\n"); return (EINVAL); } req_sz = sizeof(MPI2_IOC_INIT_REQUEST); reply_sz = sizeof(MPI2_IOC_INIT_REPLY); bzero(&init, req_sz); bzero(&reply, reply_sz); /* * Fill in the init block. Note that most addresses are * deliberately in the lower 32bits of memory. This is a micro- * optimzation for PCI/PCIX, though it's not clear if it helps PCIe. */ init.Function = MPI2_FUNCTION_IOC_INIT; init.WhoInit = MPI2_WHOINIT_HOST_DRIVER; init.MsgVersion = htole16(MPI2_VERSION); init.HeaderVersion = htole16(MPI2_HEADER_VERSION); init.SystemRequestFrameSize = htole16((uint16_t)(sc->reqframesz / 4)); init.ReplyDescriptorPostQueueDepth = htole16(sc->pqdepth); init.ReplyFreeQueueDepth = htole16(sc->fqdepth); init.SenseBufferAddressHigh = 0; init.SystemReplyAddressHigh = 0; init.SystemRequestFrameBaseAddress.High = 0; init.SystemRequestFrameBaseAddress.Low = htole32((uint32_t)sc->req_busaddr); init.ReplyDescriptorPostQueueAddress.High = 0; init.ReplyDescriptorPostQueueAddress.Low = htole32((uint32_t)sc->post_busaddr); init.ReplyFreeQueueAddress.High = 0; init.ReplyFreeQueueAddress.Low = htole32((uint32_t)sc->free_busaddr); getmicrotime(&now); time_in_msec = (now.tv_sec * 1000 + now.tv_usec/1000); init.TimeStamp.High = htole32((time_in_msec >> 32) & 0xFFFFFFFF); init.TimeStamp.Low = htole32(time_in_msec & 0xFFFFFFFF); init.HostPageSize = HOST_PAGE_SIZE_4K; error = mpr_request_sync(sc, &init, &reply, req_sz, reply_sz, 5); if ((reply.IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) error = ENXIO; mpr_dprint(sc, MPR_INIT, "IOCInit status= 0x%x\n", reply.IOCStatus); mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); return (error); } void mpr_memaddr_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { bus_addr_t *addr; addr = arg; *addr = segs[0].ds_addr; } void mpr_memaddr_wait_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mpr_busdma_context *ctx; int need_unload, need_free; ctx = (struct mpr_busdma_context *)arg; need_unload = 0; need_free = 0; mpr_lock(ctx->softc); ctx->error = error; ctx->completed = 1; if ((error == 0) && (ctx->abandoned == 0)) { *ctx->addr = segs[0].ds_addr; } else { if (nsegs != 0) need_unload = 1; if (ctx->abandoned != 0) need_free = 1; } if (need_free == 0) wakeup(ctx); mpr_unlock(ctx->softc); if (need_unload != 0) { bus_dmamap_unload(ctx->buffer_dmat, ctx->buffer_dmamap); *ctx->addr = 0; } if (need_free != 0) free(ctx, M_MPR); } static int mpr_alloc_queues(struct mpr_softc *sc) { struct mpr_queue *q; int nq, i; nq = sc->msi_msgs; mpr_dprint(sc, MPR_INIT|MPR_XINFO, "Allocating %d I/O queues\n", nq); sc->queues = malloc(sizeof(struct mpr_queue) * nq, M_MPR, M_NOWAIT|M_ZERO); if (sc->queues == NULL) return (ENOMEM); for (i = 0; i < nq; i++) { q = &sc->queues[i]; mpr_dprint(sc, MPR_INIT, "Configuring queue %d %p\n", i, q); q->sc = sc; q->qnum = i; } return (0); } static int mpr_alloc_hw_queues(struct mpr_softc *sc) { bus_addr_t queues_busaddr; uint8_t *queues; int qsize, fqsize, pqsize; /* * The reply free queue contains 4 byte entries in multiples of 16 and * aligned on a 16 byte boundary. There must always be an unused entry. * This queue supplies fresh reply frames for the firmware to use. * * The reply descriptor post queue contains 8 byte entries in * multiples of 16 and aligned on a 16 byte boundary. This queue * contains filled-in reply frames sent from the firmware to the host. * * These two queues are allocated together for simplicity. */ sc->fqdepth = roundup2(sc->num_replies + 1, 16); sc->pqdepth = roundup2(sc->num_replies + 1, 16); fqsize= sc->fqdepth * 4; pqsize = sc->pqdepth * 8; qsize = fqsize + pqsize; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ qsize, /* maxsize */ 1, /* nsegments */ qsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->queues_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate queues DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->queues_dmat, (void **)&queues, BUS_DMA_NOWAIT, &sc->queues_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate queues memory\n"); return (ENOMEM); } bzero(queues, qsize); bus_dmamap_load(sc->queues_dmat, sc->queues_map, queues, qsize, mpr_memaddr_cb, &queues_busaddr, 0); sc->free_queue = (uint32_t *)queues; sc->free_busaddr = queues_busaddr; sc->post_queue = (MPI2_REPLY_DESCRIPTORS_UNION *)(queues + fqsize); sc->post_busaddr = queues_busaddr + fqsize; mpr_dprint(sc, MPR_INIT, "free queue busaddr= %#016jx size= %d\n", (uintmax_t)sc->free_busaddr, fqsize); mpr_dprint(sc, MPR_INIT, "reply queue busaddr= %#016jx size= %d\n", (uintmax_t)sc->post_busaddr, pqsize); return (0); } static int mpr_alloc_replies(struct mpr_softc *sc) { int rsize, num_replies; /* Store the reply frame size in bytes rather than as 32bit words */ sc->replyframesz = sc->facts->ReplyFrameSize * 4; /* * sc->num_replies should be one less than sc->fqdepth. We need to * allocate space for sc->fqdepth replies, but only sc->num_replies * replies can be used at once. */ num_replies = max(sc->fqdepth, sc->num_replies); rsize = sc->replyframesz * num_replies; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 4, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->reply_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate replies DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->reply_dmat, (void **)&sc->reply_frames, BUS_DMA_NOWAIT, &sc->reply_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate replies memory\n"); return (ENOMEM); } bzero(sc->reply_frames, rsize); bus_dmamap_load(sc->reply_dmat, sc->reply_map, sc->reply_frames, rsize, mpr_memaddr_cb, &sc->reply_busaddr, 0); mpr_dprint(sc, MPR_INIT, "reply frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->reply_busaddr, rsize); return (0); } static void mpr_load_chains_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mpr_softc *sc = arg; struct mpr_chain *chain; bus_size_t bo; int i, o, s; if (error != 0) return; for (i = 0, o = 0, s = 0; s < nsegs; s++) { for (bo = 0; bo + sc->chain_frame_size <= segs[s].ds_len; bo += sc->chain_frame_size) { chain = &sc->chains[i++]; chain->chain =(MPI2_SGE_IO_UNION *)(sc->chain_frames+o); chain->chain_busaddr = segs[s].ds_addr + bo; o += sc->chain_frame_size; mpr_free_chain(sc, chain); } if (bo != segs[s].ds_len) o += segs[s].ds_len - bo; } sc->chain_free_lowwater = i; } static int mpr_alloc_requests(struct mpr_softc *sc) { struct mpr_command *cm; int i, rsize, nsegs; rsize = sc->reqframesz * sc->num_reqs; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->req_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate request DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->req_dmat, (void **)&sc->req_frames, BUS_DMA_NOWAIT, &sc->req_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate request memory\n"); return (ENOMEM); } bzero(sc->req_frames, rsize); bus_dmamap_load(sc->req_dmat, sc->req_map, sc->req_frames, rsize, mpr_memaddr_cb, &sc->req_busaddr, 0); mpr_dprint(sc, MPR_INIT, "request frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->req_busaddr, rsize); sc->chains = malloc(sizeof(struct mpr_chain) * sc->num_chains, M_MPR, M_NOWAIT | M_ZERO); if (!sc->chains) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate chain memory\n"); return (ENOMEM); } rsize = sc->chain_frame_size * sc->num_chains; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ howmany(rsize, PAGE_SIZE), /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->chain_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate chain DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->chain_dmat, (void **)&sc->chain_frames, BUS_DMA_NOWAIT | BUS_DMA_ZERO, &sc->chain_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate chain memory\n"); return (ENOMEM); } if (bus_dmamap_load(sc->chain_dmat, sc->chain_map, sc->chain_frames, rsize, mpr_load_chains_cb, sc, BUS_DMA_NOWAIT)) { mpr_dprint(sc, MPR_ERROR, "Cannot load chain memory\n"); bus_dmamem_free(sc->chain_dmat, sc->chain_frames, sc->chain_map); return (ENOMEM); } rsize = MPR_SENSE_LEN * sc->num_reqs; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 1, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->sense_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate sense DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->sense_dmat, (void **)&sc->sense_frames, BUS_DMA_NOWAIT, &sc->sense_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate sense memory\n"); return (ENOMEM); } bzero(sc->sense_frames, rsize); bus_dmamap_load(sc->sense_dmat, sc->sense_map, sc->sense_frames, rsize, mpr_memaddr_cb, &sc->sense_busaddr, 0); mpr_dprint(sc, MPR_INIT, "sense frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->sense_busaddr, rsize); /* * Allocate NVMe PRP Pages for NVMe SGL support only if the FW supports * these devices. */ if ((sc->facts->MsgVersion >= MPI2_VERSION_02_06) && (sc->facts->ProtocolFlags & MPI2_IOCFACTS_PROTOCOL_NVME_DEVICES)) { if (mpr_alloc_nvme_prp_pages(sc) == ENOMEM) return (ENOMEM); } nsegs = (sc->maxio / PAGE_SIZE) + 1; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 1, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT,/* maxsize */ nsegs, /* nsegments */ BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */ BUS_DMA_ALLOCNOW, /* flags */ busdma_lock_mutex, /* lockfunc */ &sc->mpr_mtx, /* lockarg */ &sc->buffer_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate buffer DMA tag\n"); return (ENOMEM); } /* * SMID 0 cannot be used as a free command per the firmware spec. * Just drop that command instead of risking accounting bugs. */ sc->commands = malloc(sizeof(struct mpr_command) * sc->num_reqs, M_MPR, M_WAITOK | M_ZERO); if (!sc->commands) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate command memory\n"); return (ENOMEM); } for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; cm->cm_req = sc->req_frames + i * sc->reqframesz; cm->cm_req_busaddr = sc->req_busaddr + i * sc->reqframesz; cm->cm_sense = &sc->sense_frames[i]; cm->cm_sense_busaddr = sc->sense_busaddr + i * MPR_SENSE_LEN; cm->cm_desc.Default.SMID = i; cm->cm_sc = sc; cm->cm_state = MPR_CM_STATE_BUSY; TAILQ_INIT(&cm->cm_chain_list); TAILQ_INIT(&cm->cm_prp_page_list); callout_init_mtx(&cm->cm_callout, &sc->mpr_mtx, 0); /* XXX Is a failure here a critical problem? */ if (bus_dmamap_create(sc->buffer_dmat, 0, &cm->cm_dmamap) == 0) { if (i <= sc->num_prireqs) mpr_free_high_priority_command(sc, cm); else mpr_free_command(sc, cm); } else { panic("failed to allocate command %d\n", i); sc->num_reqs = i; break; } } return (0); } /* * Allocate contiguous buffers for PCIe NVMe devices for building native PRPs, * which are scatter/gather lists for NVMe devices. * * This buffer must be contiguous due to the nature of how NVMe PRPs are built * and translated by FW. * * returns ENOMEM if memory could not be allocated, otherwise returns 0. */ static int mpr_alloc_nvme_prp_pages(struct mpr_softc *sc) { int PRPs_per_page, PRPs_required, pages_required; int rsize, i; struct mpr_prp_page *prp_page; /* * Assuming a MAX_IO_SIZE of 1MB and a PAGE_SIZE of 4k, the max number * of PRPs (NVMe's Scatter/Gather Element) needed per I/O is: * MAX_IO_SIZE / PAGE_SIZE = 256 * * 1 PRP entry in main frame for PRP list pointer still leaves 255 PRPs * required for the remainder of the 1MB I/O. 512 PRPs can fit into one * page (4096 / 8 = 512), so only one page is required for each I/O. * * Each of these buffers will need to be contiguous. For simplicity, * only one buffer is allocated here, which has all of the space * required for the NVMe Queue Depth. If there are problems allocating * this one buffer, this function will need to change to allocate * individual, contiguous NVME_QDEPTH buffers. * * The real calculation will use the real max io size. Above is just an * example. * */ PRPs_required = sc->maxio / PAGE_SIZE; PRPs_per_page = (PAGE_SIZE / PRP_ENTRY_SIZE) - 1; pages_required = (PRPs_required / PRPs_per_page) + 1; sc->prp_buffer_size = PAGE_SIZE * pages_required; rsize = sc->prp_buffer_size * NVME_QDEPTH; if (bus_dma_tag_create( sc->mpr_parent_dmat, /* parent */ 4, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->prp_page_dmat)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate NVMe PRP DMA " "tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->prp_page_dmat, (void **)&sc->prp_pages, BUS_DMA_NOWAIT, &sc->prp_page_map)) { mpr_dprint(sc, MPR_ERROR, "Cannot allocate NVMe PRP memory\n"); return (ENOMEM); } bzero(sc->prp_pages, rsize); bus_dmamap_load(sc->prp_page_dmat, sc->prp_page_map, sc->prp_pages, rsize, mpr_memaddr_cb, &sc->prp_page_busaddr, 0); sc->prps = malloc(sizeof(struct mpr_prp_page) * NVME_QDEPTH, M_MPR, M_WAITOK | M_ZERO); for (i = 0; i < NVME_QDEPTH; i++) { prp_page = &sc->prps[i]; prp_page->prp_page = (uint64_t *)(sc->prp_pages + i * sc->prp_buffer_size); prp_page->prp_page_busaddr = (uint64_t)(sc->prp_page_busaddr + i * sc->prp_buffer_size); mpr_free_prp_page(sc, prp_page); sc->prp_pages_free_lowwater++; } return (0); } static int mpr_init_queues(struct mpr_softc *sc) { int i; memset((uint8_t *)sc->post_queue, 0xff, sc->pqdepth * 8); /* * According to the spec, we need to use one less reply than we * have space for on the queue. So sc->num_replies (the number we * use) should be less than sc->fqdepth (allocated size). */ if (sc->num_replies >= sc->fqdepth) return (EINVAL); /* * Initialize all of the free queue entries. */ for (i = 0; i < sc->fqdepth; i++) { sc->free_queue[i] = sc->reply_busaddr + (i * sc->replyframesz); } sc->replyfreeindex = sc->num_replies; return (0); } /* Get the driver parameter tunables. Lowest priority are the driver defaults. * Next are the global settings, if they exist. Highest are the per-unit * settings, if they exist. */ void mpr_get_tunables(struct mpr_softc *sc) { char tmpstr[80], mpr_debug[80]; /* XXX default to some debugging for now */ sc->mpr_debug = MPR_INFO | MPR_FAULT; sc->disable_msix = 0; sc->disable_msi = 0; sc->max_msix = MPR_MSIX_MAX; sc->max_chains = MPR_CHAIN_FRAMES; sc->max_io_pages = MPR_MAXIO_PAGES; sc->enable_ssu = MPR_SSU_ENABLE_SSD_DISABLE_HDD; sc->spinup_wait_time = DEFAULT_SPINUP_WAIT; sc->use_phynum = 1; sc->max_reqframes = MPR_REQ_FRAMES; sc->max_prireqframes = MPR_PRI_REQ_FRAMES; sc->max_replyframes = MPR_REPLY_FRAMES; sc->max_evtframes = MPR_EVT_REPLY_FRAMES; /* * Grab the global variables. */ bzero(mpr_debug, 80); if (TUNABLE_STR_FETCH("hw.mpr.debug_level", mpr_debug, 80) != 0) mpr_parse_debug(sc, mpr_debug); TUNABLE_INT_FETCH("hw.mpr.disable_msix", &sc->disable_msix); TUNABLE_INT_FETCH("hw.mpr.disable_msi", &sc->disable_msi); TUNABLE_INT_FETCH("hw.mpr.max_msix", &sc->max_msix); TUNABLE_INT_FETCH("hw.mpr.max_chains", &sc->max_chains); TUNABLE_INT_FETCH("hw.mpr.max_io_pages", &sc->max_io_pages); TUNABLE_INT_FETCH("hw.mpr.enable_ssu", &sc->enable_ssu); TUNABLE_INT_FETCH("hw.mpr.spinup_wait_time", &sc->spinup_wait_time); TUNABLE_INT_FETCH("hw.mpr.use_phy_num", &sc->use_phynum); TUNABLE_INT_FETCH("hw.mpr.max_reqframes", &sc->max_reqframes); TUNABLE_INT_FETCH("hw.mpr.max_prireqframes", &sc->max_prireqframes); TUNABLE_INT_FETCH("hw.mpr.max_replyframes", &sc->max_replyframes); TUNABLE_INT_FETCH("hw.mpr.max_evtframes", &sc->max_evtframes); /* Grab the unit-instance variables */ snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.debug_level", device_get_unit(sc->mpr_dev)); bzero(mpr_debug, 80); if (TUNABLE_STR_FETCH(tmpstr, mpr_debug, 80) != 0) mpr_parse_debug(sc, mpr_debug); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.disable_msix", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->disable_msix); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.disable_msi", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->disable_msi); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_msix", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_msix); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_chains", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_chains); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_io_pages", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_io_pages); bzero(sc->exclude_ids, sizeof(sc->exclude_ids)); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.exclude_ids", device_get_unit(sc->mpr_dev)); TUNABLE_STR_FETCH(tmpstr, sc->exclude_ids, sizeof(sc->exclude_ids)); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.enable_ssu", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->enable_ssu); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.spinup_wait_time", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->spinup_wait_time); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.use_phy_num", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->use_phynum); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_reqframes", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_reqframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_prireqframes", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_prireqframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_replyframes", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_replyframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mpr.%d.max_evtframes", device_get_unit(sc->mpr_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_evtframes); } static void mpr_setup_sysctl(struct mpr_softc *sc) { struct sysctl_ctx_list *sysctl_ctx = NULL; struct sysctl_oid *sysctl_tree = NULL; char tmpstr[80], tmpstr2[80]; /* * Setup the sysctl variable so the user can change the debug level * on the fly. */ snprintf(tmpstr, sizeof(tmpstr), "MPR controller %d", device_get_unit(sc->mpr_dev)); snprintf(tmpstr2, sizeof(tmpstr2), "%d", device_get_unit(sc->mpr_dev)); sysctl_ctx = device_get_sysctl_ctx(sc->mpr_dev); if (sysctl_ctx != NULL) sysctl_tree = device_get_sysctl_tree(sc->mpr_dev); if (sysctl_tree == NULL) { sysctl_ctx_init(&sc->sysctl_ctx); sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_hw_mpr), OID_AUTO, tmpstr2, CTLFLAG_RD, 0, tmpstr); if (sc->sysctl_tree == NULL) return; sysctl_ctx = &sc->sysctl_ctx; sysctl_tree = sc->sysctl_tree; } SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "debug_level", CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0, mpr_debug_sysctl, "A", "mpr debug level"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "disable_msix", CTLFLAG_RD, &sc->disable_msix, 0, "Disable the use of MSI-X interrupts"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_msix", CTLFLAG_RD, &sc->max_msix, 0, "User-defined maximum number of MSIX queues"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "msix_msgs", CTLFLAG_RD, &sc->msi_msgs, 0, "Negotiated number of MSIX queues"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_reqframes", CTLFLAG_RD, &sc->max_reqframes, 0, "Total number of allocated request frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_prireqframes", CTLFLAG_RD, &sc->max_prireqframes, 0, "Total number of allocated high priority request frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_replyframes", CTLFLAG_RD, &sc->max_replyframes, 0, "Total number of allocated reply frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_evtframes", CTLFLAG_RD, &sc->max_evtframes, 0, "Total number of event frames allocated"); SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "firmware_version", CTLFLAG_RW, sc->fw_version, strlen(sc->fw_version), "firmware version"); SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "driver_version", CTLFLAG_RW, MPR_DRIVER_VERSION, strlen(MPR_DRIVER_VERSION), "driver version"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "io_cmds_active", CTLFLAG_RD, &sc->io_cmds_active, 0, "number of currently active commands"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "io_cmds_highwater", CTLFLAG_RD, &sc->io_cmds_highwater, 0, "maximum active commands seen"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_free", CTLFLAG_RD, &sc->chain_free, 0, "number of free chain elements"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_free_lowwater", CTLFLAG_RD, &sc->chain_free_lowwater, 0,"lowest number of free chain elements"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_chains", CTLFLAG_RD, &sc->max_chains, 0,"maximum chain frames that will be allocated"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_io_pages", CTLFLAG_RD, &sc->max_io_pages, 0,"maximum pages to allow per I/O (if <1 use " "IOCFacts)"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "enable_ssu", CTLFLAG_RW, &sc->enable_ssu, 0, "enable SSU to SATA SSD/HDD at shutdown"); SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_alloc_fail", CTLFLAG_RD, &sc->chain_alloc_fail, "chain allocation failures"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "spinup_wait_time", CTLFLAG_RD, &sc->spinup_wait_time, DEFAULT_SPINUP_WAIT, "seconds to wait for " "spinup after SATA ID error"); SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "dump_reqs", CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_SKIP, sc, 0, mpr_dump_reqs, "I", "Dump Active Requests"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "use_phy_num", CTLFLAG_RD, &sc->use_phynum, 0, "Use the phy number for enumeration"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "prp_pages_free", CTLFLAG_RD, &sc->prp_pages_free, 0, "number of free PRP pages"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "prp_pages_free_lowwater", CTLFLAG_RD, &sc->prp_pages_free_lowwater, 0,"lowest number of free PRP pages"); SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "prp_page_alloc_fail", CTLFLAG_RD, &sc->prp_page_alloc_fail, "PRP page allocation failures"); } static struct mpr_debug_string { char *name; int flag; } mpr_debug_strings[] = { {"info", MPR_INFO}, {"fault", MPR_FAULT}, {"event", MPR_EVENT}, {"log", MPR_LOG}, {"recovery", MPR_RECOVERY}, {"error", MPR_ERROR}, {"init", MPR_INIT}, {"xinfo", MPR_XINFO}, {"user", MPR_USER}, {"mapping", MPR_MAPPING}, {"trace", MPR_TRACE} }; enum mpr_debug_level_combiner { COMB_NONE, COMB_ADD, COMB_SUB }; static int mpr_debug_sysctl(SYSCTL_HANDLER_ARGS) { struct mpr_softc *sc; struct mpr_debug_string *string; struct sbuf *sbuf; char *buffer; size_t sz; int i, len, debug, error; sc = (struct mpr_softc *)arg1; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf = sbuf_new_for_sysctl(NULL, NULL, 128, req); debug = sc->mpr_debug; sbuf_printf(sbuf, "%#x", debug); sz = sizeof(mpr_debug_strings) / sizeof(mpr_debug_strings[0]); for (i = 0; i < sz; i++) { string = &mpr_debug_strings[i]; if (debug & string->flag) sbuf_printf(sbuf, ",%s", string->name); } error = sbuf_finish(sbuf); sbuf_delete(sbuf); if (error || req->newptr == NULL) return (error); len = req->newlen - req->newidx; if (len == 0) return (0); buffer = malloc(len, M_MPR, M_ZERO|M_WAITOK); error = SYSCTL_IN(req, buffer, len); mpr_parse_debug(sc, buffer); free(buffer, M_MPR); return (error); } static void mpr_parse_debug(struct mpr_softc *sc, char *list) { struct mpr_debug_string *string; enum mpr_debug_level_combiner op; char *token, *endtoken; size_t sz; int flags, i; if (list == NULL || *list == '\0') return; if (*list == '+') { op = COMB_ADD; list++; } else if (*list == '-') { op = COMB_SUB; list++; } else op = COMB_NONE; if (*list == '\0') return; flags = 0; sz = sizeof(mpr_debug_strings) / sizeof(mpr_debug_strings[0]); while ((token = strsep(&list, ":,")) != NULL) { /* Handle integer flags */ flags |= strtol(token, &endtoken, 0); if (token != endtoken) continue; /* Handle text flags */ for (i = 0; i < sz; i++) { string = &mpr_debug_strings[i]; if (strcasecmp(token, string->name) == 0) { flags |= string->flag; break; } } } switch (op) { case COMB_NONE: sc->mpr_debug = flags; break; case COMB_ADD: sc->mpr_debug |= flags; break; case COMB_SUB: sc->mpr_debug &= (~flags); break; } return; } struct mpr_dumpreq_hdr { uint32_t smid; uint32_t state; uint32_t numframes; uint32_t deschi; uint32_t desclo; }; static int mpr_dump_reqs(SYSCTL_HANDLER_ARGS) { struct mpr_softc *sc; struct mpr_chain *chain, *chain1; struct mpr_command *cm; struct mpr_dumpreq_hdr hdr; struct sbuf *sb; uint32_t smid, state; int i, numreqs, error = 0; sc = (struct mpr_softc *)arg1; if ((error = priv_check(curthread, PRIV_DRIVER)) != 0) { printf("priv check error %d\n", error); return (error); } state = MPR_CM_STATE_INQUEUE; smid = 1; numreqs = sc->num_reqs; if (req->newptr != NULL) return (EINVAL); if (smid == 0 || smid > sc->num_reqs) return (EINVAL); if (numreqs <= 0 || (numreqs + smid > sc->num_reqs)) numreqs = sc->num_reqs; sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req); /* Best effort, no locking */ for (i = smid; i < numreqs; i++) { cm = &sc->commands[i]; if (cm->cm_state != state) continue; hdr.smid = i; hdr.state = cm->cm_state; hdr.numframes = 1; hdr.deschi = cm->cm_desc.Words.High; hdr.desclo = cm->cm_desc.Words.Low; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain1) hdr.numframes++; sbuf_bcat(sb, &hdr, sizeof(hdr)); sbuf_bcat(sb, cm->cm_req, 128); TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain1) sbuf_bcat(sb, chain->chain, 128); } error = sbuf_finish(sb); sbuf_delete(sb); return (error); } int mpr_attach(struct mpr_softc *sc) { int error; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); mtx_init(&sc->mpr_mtx, "MPR lock", NULL, MTX_DEF); callout_init_mtx(&sc->periodic, &sc->mpr_mtx, 0); callout_init_mtx(&sc->device_check_callout, &sc->mpr_mtx, 0); TAILQ_INIT(&sc->event_list); timevalclear(&sc->lastfail); if ((error = mpr_transition_ready(sc)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Failed to transition ready\n"); return (error); } sc->facts = malloc(sizeof(MPI2_IOC_FACTS_REPLY), M_MPR, M_ZERO|M_NOWAIT); if (!sc->facts) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "Cannot allocate memory, exit\n"); return (ENOMEM); } /* * Get IOC Facts and allocate all structures based on this information. * A Diag Reset will also call mpr_iocfacts_allocate and re-read the IOC * Facts. If relevant values have changed in IOC Facts, this function * will free all of the memory based on IOC Facts and reallocate that * memory. If this fails, any allocated memory should already be freed. */ if ((error = mpr_iocfacts_allocate(sc, TRUE)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "IOC Facts allocation " "failed with error %d\n", error); return (error); } /* Start the periodic watchdog check on the IOC Doorbell */ mpr_periodic(sc); /* * The portenable will kick off discovery events that will drive the * rest of the initialization process. The CAM/SAS module will * hold up the boot sequence until discovery is complete. */ sc->mpr_ich.ich_func = mpr_startup; sc->mpr_ich.ich_arg = sc; if (config_intrhook_establish(&sc->mpr_ich) != 0) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot establish MPR config hook\n"); error = EINVAL; } /* * Allow IR to shutdown gracefully when shutdown occurs. */ sc->shutdown_eh = EVENTHANDLER_REGISTER(shutdown_final, mprsas_ir_shutdown, sc, SHUTDOWN_PRI_DEFAULT); if (sc->shutdown_eh == NULL) mpr_dprint(sc, MPR_INIT|MPR_ERROR, "shutdown event registration failed\n"); mpr_setup_sysctl(sc); sc->mpr_flags |= MPR_FLAGS_ATTACH_DONE; mpr_dprint(sc, MPR_INIT, "%s exit error= %d\n", __func__, error); return (error); } /* Run through any late-start handlers. */ static void mpr_startup(void *arg) { struct mpr_softc *sc; sc = (struct mpr_softc *)arg; mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); mpr_lock(sc); mpr_unmask_intr(sc); /* initialize device mapping tables */ mpr_base_static_config_pages(sc); mpr_mapping_initialize(sc); mprsas_startup(sc); mpr_unlock(sc); mpr_dprint(sc, MPR_INIT, "disestablish config intrhook\n"); config_intrhook_disestablish(&sc->mpr_ich); sc->mpr_ich.ich_arg = NULL; mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); } /* Periodic watchdog. Is called with the driver lock already held. */ static void mpr_periodic(void *arg) { struct mpr_softc *sc; uint32_t db; sc = (struct mpr_softc *)arg; if (sc->mpr_flags & MPR_FLAGS_SHUTDOWN) return; db = mpr_regread(sc, MPI2_DOORBELL_OFFSET); if ((db & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) { if ((db & MPI2_DOORBELL_FAULT_CODE_MASK) == IFAULT_IOP_OVER_TEMP_THRESHOLD_EXCEEDED) { panic("TEMPERATURE FAULT: STOPPING."); } mpr_dprint(sc, MPR_FAULT, "IOC Fault 0x%08x, Resetting\n", db); mpr_reinit(sc); } callout_reset(&sc->periodic, MPR_PERIODIC_DELAY * hz, mpr_periodic, sc); } static void mpr_log_evt_handler(struct mpr_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event) { MPI2_EVENT_DATA_LOG_ENTRY_ADDED *entry; MPR_DPRINT_EVENT(sc, generic, event); switch (event->Event) { case MPI2_EVENT_LOG_DATA: mpr_dprint(sc, MPR_EVENT, "MPI2_EVENT_LOG_DATA:\n"); if (sc->mpr_debug & MPR_EVENT) hexdump(event->EventData, event->EventDataLength, NULL, 0); break; case MPI2_EVENT_LOG_ENTRY_ADDED: entry = (MPI2_EVENT_DATA_LOG_ENTRY_ADDED *)event->EventData; mpr_dprint(sc, MPR_EVENT, "MPI2_EVENT_LOG_ENTRY_ADDED event " "0x%x Sequence %d:\n", entry->LogEntryQualifier, entry->LogSequence); break; default: break; } return; } static int mpr_attach_log(struct mpr_softc *sc) { uint8_t events[16]; bzero(events, 16); setbit(events, MPI2_EVENT_LOG_DATA); setbit(events, MPI2_EVENT_LOG_ENTRY_ADDED); mpr_register_events(sc, events, mpr_log_evt_handler, NULL, &sc->mpr_log_eh); return (0); } static int mpr_detach_log(struct mpr_softc *sc) { if (sc->mpr_log_eh != NULL) mpr_deregister_events(sc, sc->mpr_log_eh); return (0); } /* * Free all of the driver resources and detach submodules. Should be called * without the lock held. */ int mpr_free(struct mpr_softc *sc) { int error; mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); /* Turn off the watchdog */ mpr_lock(sc); sc->mpr_flags |= MPR_FLAGS_SHUTDOWN; mpr_unlock(sc); /* Lock must not be held for this */ callout_drain(&sc->periodic); callout_drain(&sc->device_check_callout); if (((error = mpr_detach_log(sc)) != 0) || ((error = mpr_detach_sas(sc)) != 0)) { mpr_dprint(sc, MPR_INIT|MPR_FAULT, "failed to detach " "subsystems, error= %d, exit\n", error); return (error); } mpr_detach_user(sc); /* Put the IOC back in the READY state. */ mpr_lock(sc); if ((error = mpr_transition_ready(sc)) != 0) { mpr_unlock(sc); return (error); } mpr_unlock(sc); if (sc->facts != NULL) free(sc->facts, M_MPR); /* * Free all buffers that are based on IOC Facts. A Diag Reset may need * to free these buffers too. */ mpr_iocfacts_free(sc); if (sc->sysctl_tree != NULL) sysctl_ctx_free(&sc->sysctl_ctx); /* Deregister the shutdown function */ if (sc->shutdown_eh != NULL) EVENTHANDLER_DEREGISTER(shutdown_final, sc->shutdown_eh); mtx_destroy(&sc->mpr_mtx); mpr_dprint(sc, MPR_INIT, "%s exit\n", __func__); return (0); } static __inline void mpr_complete_command(struct mpr_softc *sc, struct mpr_command *cm) { MPR_FUNCTRACE(sc); if (cm == NULL) { mpr_dprint(sc, MPR_ERROR, "Completing NULL command\n"); return; } if (cm->cm_flags & MPR_CM_FLAGS_POLLED) cm->cm_flags |= MPR_CM_FLAGS_COMPLETE; if (cm->cm_complete != NULL) { mpr_dprint(sc, MPR_TRACE, "%s cm %p calling cm_complete %p data %p reply %p\n", __func__, cm, cm->cm_complete, cm->cm_complete_data, cm->cm_reply); cm->cm_complete(sc, cm); } if (cm->cm_flags & MPR_CM_FLAGS_WAKEUP) { mpr_dprint(sc, MPR_TRACE, "waking up %p\n", cm); wakeup(cm); } if (sc->io_cmds_active != 0) { sc->io_cmds_active--; } else { mpr_dprint(sc, MPR_ERROR, "Warning: io_cmds_active is " "out of sync - resynching to 0\n"); } } static void mpr_sas_log_info(struct mpr_softc *sc , u32 log_info) { union loginfo_type { u32 loginfo; struct { u32 subcode:16; u32 code:8; u32 originator:4; u32 bus_type:4; } dw; }; union loginfo_type sas_loginfo; char *originator_str = NULL; sas_loginfo.loginfo = log_info; if (sas_loginfo.dw.bus_type != 3 /*SAS*/) return; /* each nexus loss loginfo */ if (log_info == 0x31170000) return; /* eat the loginfos associated with task aborts */ if ((log_info == 30050000) || (log_info == 0x31140000) || (log_info == 0x31130000)) return; switch (sas_loginfo.dw.originator) { case 0: originator_str = "IOP"; break; case 1: originator_str = "PL"; break; case 2: originator_str = "IR"; break; } mpr_dprint(sc, MPR_LOG, "log_info(0x%08x): originator(%s), " "code(0x%02x), sub_code(0x%04x)\n", log_info, originator_str, sas_loginfo.dw.code, sas_loginfo.dw.subcode); } static void mpr_display_reply_info(struct mpr_softc *sc, uint8_t *reply) { MPI2DefaultReply_t *mpi_reply; u16 sc_status; mpi_reply = (MPI2DefaultReply_t*)reply; sc_status = le16toh(mpi_reply->IOCStatus); if (sc_status & MPI2_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE) mpr_sas_log_info(sc, le32toh(mpi_reply->IOCLogInfo)); } void mpr_intr(void *data) { struct mpr_softc *sc; uint32_t status; sc = (struct mpr_softc *)data; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); /* * Check interrupt status register to flush the bus. This is * needed for both INTx interrupts and driver-driven polling */ status = mpr_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); if ((status & MPI2_HIS_REPLY_DESCRIPTOR_INTERRUPT) == 0) return; mpr_lock(sc); mpr_intr_locked(data); mpr_unlock(sc); return; } /* * In theory, MSI/MSIX interrupts shouldn't need to read any registers on the * chip. Hopefully this theory is correct. */ void mpr_intr_msi(void *data) { struct mpr_softc *sc; sc = (struct mpr_softc *)data; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); mpr_lock(sc); mpr_intr_locked(data); mpr_unlock(sc); return; } /* * The locking is overly broad and simplistic, but easy to deal with for now. */ void mpr_intr_locked(void *data) { MPI2_REPLY_DESCRIPTORS_UNION *desc; MPI2_DIAG_RELEASE_REPLY *rel_rep; mpr_fw_diagnostic_buffer_t *pBuffer; struct mpr_softc *sc; uint64_t tdesc; struct mpr_command *cm = NULL; uint8_t flags; u_int pq; sc = (struct mpr_softc *)data; pq = sc->replypostindex; mpr_dprint(sc, MPR_TRACE, "%s sc %p starting with replypostindex %u\n", __func__, sc, sc->replypostindex); for ( ;; ) { cm = NULL; desc = &sc->post_queue[sc->replypostindex]; /* * Copy and clear out the descriptor so that any reentry will * immediately know that this descriptor has already been * looked at. There is unfortunate casting magic because the * MPI API doesn't have a cardinal 64bit type. */ tdesc = 0xffffffffffffffff; tdesc = atomic_swap_64((uint64_t *)desc, tdesc); desc = (MPI2_REPLY_DESCRIPTORS_UNION *)&tdesc; flags = desc->Default.ReplyFlags & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK; if ((flags == MPI2_RPY_DESCRIPT_FLAGS_UNUSED) || (le32toh(desc->Words.High) == 0xffffffff)) break; /* increment the replypostindex now, so that event handlers * and cm completion handlers which decide to do a diag * reset can zero it without it getting incremented again * afterwards, and we break out of this loop on the next * iteration since the reply post queue has been cleared to * 0xFF and all descriptors look unused (which they are). */ if (++sc->replypostindex >= sc->pqdepth) sc->replypostindex = 0; switch (flags) { case MPI2_RPY_DESCRIPT_FLAGS_SCSI_IO_SUCCESS: case MPI25_RPY_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO_SUCCESS: case MPI26_RPY_DESCRIPT_FLAGS_PCIE_ENCAPSULATED_SUCCESS: cm = &sc->commands[le16toh(desc->SCSIIOSuccess.SMID)]; KASSERT(cm->cm_state == MPR_CM_STATE_INQUEUE, ("command not inqueue\n")); cm->cm_state = MPR_CM_STATE_BUSY; cm->cm_reply = NULL; break; case MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY: { uint32_t baddr; uint8_t *reply; /* * Re-compose the reply address from the address * sent back from the chip. The ReplyFrameAddress * is the lower 32 bits of the physical address of * particular reply frame. Convert that address to * host format, and then use that to provide the * offset against the virtual address base * (sc->reply_frames). */ baddr = le32toh(desc->AddressReply.ReplyFrameAddress); reply = sc->reply_frames + (baddr - ((uint32_t)sc->reply_busaddr)); /* * Make sure the reply we got back is in a valid * range. If not, go ahead and panic here, since * we'll probably panic as soon as we deference the * reply pointer anyway. */ if ((reply < sc->reply_frames) || (reply > (sc->reply_frames + (sc->fqdepth * sc->replyframesz)))) { printf("%s: WARNING: reply %p out of range!\n", __func__, reply); printf("%s: reply_frames %p, fqdepth %d, " "frame size %d\n", __func__, sc->reply_frames, sc->fqdepth, sc->replyframesz); printf("%s: baddr %#x,\n", __func__, baddr); /* LSI-TODO. See Linux Code for Graceful exit */ panic("Reply address out of range"); } if (le16toh(desc->AddressReply.SMID) == 0) { if (((MPI2_DEFAULT_REPLY *)reply)->Function == MPI2_FUNCTION_DIAG_BUFFER_POST) { /* * If SMID is 0 for Diag Buffer Post, * this implies that the reply is due to * a release function with a status that * the buffer has been released. Set * the buffer flags accordingly. */ rel_rep = (MPI2_DIAG_RELEASE_REPLY *)reply; if ((le16toh(rel_rep->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_DIAGNOSTIC_RELEASED) { pBuffer = &sc->fw_diag_buffer_list[ rel_rep->BufferType]; pBuffer->valid_data = TRUE; pBuffer->owned_by_firmware = FALSE; pBuffer->immediate = FALSE; } } else mpr_dispatch_event(sc, baddr, (MPI2_EVENT_NOTIFICATION_REPLY *) reply); } else { cm = &sc->commands[ le16toh(desc->AddressReply.SMID)]; - if (cm->cm_state != MPR_CM_STATE_TIMEDOUT) - cm->cm_state = MPR_CM_STATE_BUSY; - cm->cm_reply = reply; - cm->cm_reply_data = - le32toh(desc->AddressReply. - ReplyFrameAddress); + if (cm->cm_state == MPR_CM_STATE_INQUEUE) { + cm->cm_reply = reply; + cm->cm_reply_data = + le32toh(desc->AddressReply. + ReplyFrameAddress); + } else { + mpr_dprint(sc, MPR_RECOVERY, + "Bad state for ADDRESS_REPLY status," + " ignoring state %d cm %p\n", + cm->cm_state, cm); + } } break; } case MPI2_RPY_DESCRIPT_FLAGS_TARGETASSIST_SUCCESS: case MPI2_RPY_DESCRIPT_FLAGS_TARGET_COMMAND_BUFFER: case MPI2_RPY_DESCRIPT_FLAGS_RAID_ACCELERATOR_SUCCESS: default: /* Unhandled */ mpr_dprint(sc, MPR_ERROR, "Unhandled reply 0x%x\n", desc->Default.ReplyFlags); cm = NULL; break; } if (cm != NULL) { // Print Error reply frame if (cm->cm_reply) mpr_display_reply_info(sc,cm->cm_reply); mpr_complete_command(sc, cm); } } if (pq != sc->replypostindex) { mpr_dprint(sc, MPR_TRACE, "%s sc %p writing postindex %d\n", __func__, sc, sc->replypostindex); mpr_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, sc->replypostindex); } return; } static void mpr_dispatch_event(struct mpr_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *reply) { struct mpr_event_handle *eh; int event, handled = 0; event = le16toh(reply->Event); TAILQ_FOREACH(eh, &sc->event_list, eh_list) { if (isset(eh->mask, event)) { eh->callback(sc, data, reply); handled++; } } if (handled == 0) mpr_dprint(sc, MPR_EVENT, "Unhandled event 0x%x\n", le16toh(event)); /* * This is the only place that the event/reply should be freed. * Anything wanting to hold onto the event data should have * already copied it into their own storage. */ mpr_free_reply(sc, data); } static void mpr_reregister_events_complete(struct mpr_softc *sc, struct mpr_command *cm) { mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); if (cm->cm_reply) MPR_DPRINT_EVENT(sc, generic, (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply); mpr_free_command(sc, cm); /* next, send a port enable */ mprsas_startup(sc); } /* * For both register_events and update_events, the caller supplies a bitmap * of events that it _wants_. These functions then turn that into a bitmask * suitable for the controller. */ int mpr_register_events(struct mpr_softc *sc, uint8_t *mask, mpr_evt_callback_t *cb, void *data, struct mpr_event_handle **handle) { struct mpr_event_handle *eh; int error = 0; eh = malloc(sizeof(struct mpr_event_handle), M_MPR, M_WAITOK|M_ZERO); if (!eh) { mpr_dprint(sc, MPR_EVENT|MPR_ERROR, "Cannot allocate event memory\n"); return (ENOMEM); } eh->callback = cb; eh->data = data; TAILQ_INSERT_TAIL(&sc->event_list, eh, eh_list); if (mask != NULL) error = mpr_update_events(sc, eh, mask); *handle = eh; return (error); } int mpr_update_events(struct mpr_softc *sc, struct mpr_event_handle *handle, uint8_t *mask) { MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; MPI2_EVENT_NOTIFICATION_REPLY *reply = NULL; struct mpr_command *cm = NULL; struct mpr_event_handle *eh; int error, i; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); if ((mask != NULL) && (handle != NULL)) bcopy(mask, &handle->mask[0], 16); memset(sc->event_mask, 0xff, 16); TAILQ_FOREACH(eh, &sc->event_list, eh_list) { for (i = 0; i < 16; i++) sc->event_mask[i] &= ~eh->mask[i]; } if ((cm = mpr_alloc_command(sc)) == NULL) return (EBUSY); evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; evtreq->MsgFlags = 0; evtreq->SASBroadcastPrimitiveMasks = 0; #ifdef MPR_DEBUG_ALL_EVENTS { u_char fullmask[16]; memset(fullmask, 0x00, 16); bcopy(fullmask, (uint8_t *)&evtreq->EventMasks, 16); } #else bcopy(sc->event_mask, (uint8_t *)&evtreq->EventMasks, 16); #endif cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = NULL; error = mpr_request_polled(sc, &cm); if (cm != NULL) reply = (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply; if ((reply == NULL) || (reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) error = ENXIO; if (reply) MPR_DPRINT_EVENT(sc, generic, reply); mpr_dprint(sc, MPR_TRACE, "%s finished error %d\n", __func__, error); if (cm != NULL) mpr_free_command(sc, cm); return (error); } static int mpr_reregister_events(struct mpr_softc *sc) { MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; struct mpr_command *cm; struct mpr_event_handle *eh; int error, i; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); /* first, reregister events */ memset(sc->event_mask, 0xff, 16); TAILQ_FOREACH(eh, &sc->event_list, eh_list) { for (i = 0; i < 16; i++) sc->event_mask[i] &= ~eh->mask[i]; } if ((cm = mpr_alloc_command(sc)) == NULL) return (EBUSY); evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; evtreq->MsgFlags = 0; evtreq->SASBroadcastPrimitiveMasks = 0; #ifdef MPR_DEBUG_ALL_EVENTS { u_char fullmask[16]; memset(fullmask, 0x00, 16); bcopy(fullmask, (uint8_t *)&evtreq->EventMasks, 16); } #else bcopy(sc->event_mask, (uint8_t *)&evtreq->EventMasks, 16); #endif cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = NULL; cm->cm_complete = mpr_reregister_events_complete; error = mpr_map_command(sc, cm); mpr_dprint(sc, MPR_TRACE, "%s finished with error %d\n", __func__, error); return (error); } int mpr_deregister_events(struct mpr_softc *sc, struct mpr_event_handle *handle) { TAILQ_REMOVE(&sc->event_list, handle, eh_list); free(handle, M_MPR); return (mpr_update_events(sc, NULL, NULL)); } /** * mpr_build_nvme_prp - This function is called for NVMe end devices to build a * native SGL (NVMe PRP). The native SGL is built starting in the first PRP entry * of the NVMe message (PRP1). If the data buffer is small enough to be described * entirely using PRP1, then PRP2 is not used. If needed, PRP2 is used to * describe a larger data buffer. If the data buffer is too large to describe * using the two PRP entriess inside the NVMe message, then PRP1 describes the * first data memory segment, and PRP2 contains a pointer to a PRP list located * elsewhere in memory to describe the remaining data memory segments. The PRP * list will be contiguous. * The native SGL for NVMe devices is a Physical Region Page (PRP). A PRP * consists of a list of PRP entries to describe a number of noncontigous * physical memory segments as a single memory buffer, just as a SGL does. Note * however, that this function is only used by the IOCTL call, so the memory * given will be guaranteed to be contiguous. There is no need to translate * non-contiguous SGL into a PRP in this case. All PRPs will describe contiguous * space that is one page size each. * * Each NVMe message contains two PRP entries. The first (PRP1) either contains * a PRP list pointer or a PRP element, depending upon the command. PRP2 contains * the second PRP element if the memory being described fits within 2 PRP * entries, or a PRP list pointer if the PRP spans more than two entries. * * A PRP list pointer contains the address of a PRP list, structured as a linear * array of PRP entries. Each PRP entry in this list describes a segment of * physical memory. * * Each 64-bit PRP entry comprises an address and an offset field. The address * always points to the beginning of a PAGE_SIZE physical memory page, and the * offset describes where within that page the memory segment begins. Only the * first element in a PRP list may contain a non-zero offest, implying that all * memory segments following the first begin at the start of a PAGE_SIZE page. * * Each PRP element normally describes a chunck of PAGE_SIZE physical memory, * with exceptions for the first and last elements in the list. If the memory * being described by the list begins at a non-zero offset within the first page, * then the first PRP element will contain a non-zero offset indicating where the * region begins within the page. The last memory segment may end before the end * of the PAGE_SIZE segment, depending upon the overall size of the memory being * described by the PRP list. * * Since PRP entries lack any indication of size, the overall data buffer length * is used to determine where the end of the data memory buffer is located, and * how many PRP entries are required to describe it. * * Returns nothing. */ void mpr_build_nvme_prp(struct mpr_softc *sc, struct mpr_command *cm, Mpi26NVMeEncapsulatedRequest_t *nvme_encap_request, void *data, uint32_t data_in_sz, uint32_t data_out_sz) { int prp_size = PRP_ENTRY_SIZE; uint64_t *prp_entry, *prp1_entry, *prp2_entry; uint64_t *prp_entry_phys, *prp_page, *prp_page_phys; uint32_t offset, entry_len, page_mask_result, page_mask; bus_addr_t paddr; size_t length; struct mpr_prp_page *prp_page_info = NULL; /* * Not all commands require a data transfer. If no data, just return * without constructing any PRP. */ if (!data_in_sz && !data_out_sz) return; /* * Set pointers to PRP1 and PRP2, which are in the NVMe command. PRP1 is * located at a 24 byte offset from the start of the NVMe command. Then * set the current PRP entry pointer to PRP1. */ prp1_entry = (uint64_t *)(nvme_encap_request->NVMe_Command + NVME_CMD_PRP1_OFFSET); prp2_entry = (uint64_t *)(nvme_encap_request->NVMe_Command + NVME_CMD_PRP2_OFFSET); prp_entry = prp1_entry; /* * For the PRP entries, use the specially allocated buffer of * contiguous memory. PRP Page allocation failures should not happen * because there should be enough PRP page buffers to account for the * possible NVMe QDepth. */ prp_page_info = mpr_alloc_prp_page(sc); KASSERT(prp_page_info != NULL, ("%s: There are no PRP Pages left to be " "used for building a native NVMe SGL.\n", __func__)); prp_page = (uint64_t *)prp_page_info->prp_page; prp_page_phys = (uint64_t *)(uintptr_t)prp_page_info->prp_page_busaddr; /* * Insert the allocated PRP page into the command's PRP page list. This * will be freed when the command is freed. */ TAILQ_INSERT_TAIL(&cm->cm_prp_page_list, prp_page_info, prp_page_link); /* * Check if we are within 1 entry of a page boundary we don't want our * first entry to be a PRP List entry. */ page_mask = PAGE_SIZE - 1; page_mask_result = (uintptr_t)((uint8_t *)prp_page + prp_size) & page_mask; if (!page_mask_result) { /* Bump up to next page boundary. */ prp_page = (uint64_t *)((uint8_t *)prp_page + prp_size); prp_page_phys = (uint64_t *)((uint8_t *)prp_page_phys + prp_size); } /* * Set PRP physical pointer, which initially points to the current PRP * DMA memory page. */ prp_entry_phys = prp_page_phys; /* Get physical address and length of the data buffer. */ paddr = (bus_addr_t)(uintptr_t)data; if (data_in_sz) length = data_in_sz; else length = data_out_sz; /* Loop while the length is not zero. */ while (length) { /* * Check if we need to put a list pointer here if we are at page * boundary - prp_size (8 bytes). */ page_mask_result = (uintptr_t)((uint8_t *)prp_entry_phys + prp_size) & page_mask; if (!page_mask_result) { /* * This is the last entry in a PRP List, so we need to * put a PRP list pointer here. What this does is: * - bump the current memory pointer to the next * address, which will be the next full page. * - set the PRP Entry to point to that page. This is * now the PRP List pointer. * - bump the PRP Entry pointer the start of the next * page. Since all of this PRP memory is contiguous, * no need to get a new page - it's just the next * address. */ prp_entry_phys++; *prp_entry = htole64((uint64_t)(uintptr_t)prp_entry_phys); prp_entry++; } /* Need to handle if entry will be part of a page. */ offset = (uint32_t)paddr & page_mask; entry_len = PAGE_SIZE - offset; if (prp_entry == prp1_entry) { /* * Must fill in the first PRP pointer (PRP1) before * moving on. */ *prp1_entry = htole64((uint64_t)paddr); /* * Now point to the second PRP entry within the * command (PRP2). */ prp_entry = prp2_entry; } else if (prp_entry == prp2_entry) { /* * Should the PRP2 entry be a PRP List pointer or just a * regular PRP pointer? If there is more than one more * page of data, must use a PRP List pointer. */ if (length > PAGE_SIZE) { /* * PRP2 will contain a PRP List pointer because * more PRP's are needed with this command. The * list will start at the beginning of the * contiguous buffer. */ *prp2_entry = htole64( (uint64_t)(uintptr_t)prp_entry_phys); /* * The next PRP Entry will be the start of the * first PRP List. */ prp_entry = prp_page; } else { /* * After this, the PRP Entries are complete. * This command uses 2 PRP's and no PRP list. */ *prp2_entry = htole64((uint64_t)paddr); } } else { /* * Put entry in list and bump the addresses. * * After PRP1 and PRP2 are filled in, this will fill in * all remaining PRP entries in a PRP List, one per each * time through the loop. */ *prp_entry = htole64((uint64_t)paddr); prp_entry++; prp_entry_phys++; } /* * Bump the phys address of the command's data buffer by the * entry_len. */ paddr += entry_len; /* Decrement length accounting for last partial page. */ if (entry_len > length) length = 0; else length -= entry_len; } } /* * mpr_check_pcie_native_sgl - This function is called for PCIe end devices to * determine if the driver needs to build a native SGL. If so, that native SGL * is built in the contiguous buffers allocated especially for PCIe SGL * creation. If the driver will not build a native SGL, return TRUE and a * normal IEEE SGL will be built. Currently this routine supports NVMe devices * only. * * Returns FALSE (0) if native SGL was built, TRUE (1) if no SGL was built. */ static int mpr_check_pcie_native_sgl(struct mpr_softc *sc, struct mpr_command *cm, bus_dma_segment_t *segs, int segs_left) { uint32_t i, sge_dwords, length, offset, entry_len; uint32_t num_entries, buff_len = 0, sges_in_segment; uint32_t page_mask, page_mask_result, *curr_buff; uint32_t *ptr_sgl, *ptr_first_sgl, first_page_offset; uint32_t first_page_data_size, end_residual; uint64_t *msg_phys; bus_addr_t paddr; int build_native_sgl = 0, first_prp_entry; int prp_size = PRP_ENTRY_SIZE; Mpi25IeeeSgeChain64_t *main_chain_element = NULL; struct mpr_prp_page *prp_page_info = NULL; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); /* * Add up the sizes of each segment length to get the total transfer * size, which will be checked against the Maximum Data Transfer Size. * If the data transfer length exceeds the MDTS for this device, just * return 1 so a normal IEEE SGL will be built. F/W will break the I/O * up into multiple I/O's. [nvme_mdts = 0 means unlimited] */ for (i = 0; i < segs_left; i++) buff_len += htole32(segs[i].ds_len); if ((cm->cm_targ->MDTS > 0) && (buff_len > cm->cm_targ->MDTS)) return 1; /* Create page_mask (to get offset within page) */ page_mask = PAGE_SIZE - 1; /* * Check if the number of elements exceeds the max number that can be * put in the main message frame (H/W can only translate an SGL that * is contained entirely in the main message frame). */ sges_in_segment = (sc->reqframesz - offsetof(Mpi25SCSIIORequest_t, SGL)) / sizeof(MPI25_SGE_IO_UNION); if (segs_left > sges_in_segment) build_native_sgl = 1; else { /* * NVMe uses one PRP for each physical page (or part of physical * page). * if 4 pages or less then IEEE is OK * if > 5 pages then we need to build a native SGL * if > 4 and <= 5 pages, then check the physical address of * the first SG entry, then if this first size in the page * is >= the residual beyond 4 pages then use IEEE, * otherwise use native SGL */ if (buff_len > (PAGE_SIZE * 5)) build_native_sgl = 1; else if ((buff_len > (PAGE_SIZE * 4)) && (buff_len <= (PAGE_SIZE * 5)) ) { msg_phys = (uint64_t *)(uintptr_t)segs[0].ds_addr; first_page_offset = ((uint32_t)(uint64_t)(uintptr_t)msg_phys & page_mask); first_page_data_size = PAGE_SIZE - first_page_offset; end_residual = buff_len % PAGE_SIZE; /* * If offset into first page pushes the end of the data * beyond end of the 5th page, we need the extra PRP * list. */ if (first_page_data_size < end_residual) build_native_sgl = 1; /* * Check if first SG entry size is < residual beyond 4 * pages. */ if (htole32(segs[0].ds_len) < (buff_len - (PAGE_SIZE * 4))) build_native_sgl = 1; } } /* check if native SGL is needed */ if (!build_native_sgl) return 1; /* * Native SGL is needed. * Put a chain element in main message frame that points to the first * chain buffer. * * NOTE: The ChainOffset field must be 0 when using a chain pointer to * a native SGL. */ /* Set main message chain element pointer */ main_chain_element = (pMpi25IeeeSgeChain64_t)cm->cm_sge; /* * For NVMe the chain element needs to be the 2nd SGL entry in the main * message. */ main_chain_element = (Mpi25IeeeSgeChain64_t *) ((uint8_t *)main_chain_element + sizeof(MPI25_IEEE_SGE_CHAIN64)); /* * For the PRP entries, use the specially allocated buffer of * contiguous memory. PRP Page allocation failures should not happen * because there should be enough PRP page buffers to account for the * possible NVMe QDepth. */ prp_page_info = mpr_alloc_prp_page(sc); KASSERT(prp_page_info != NULL, ("%s: There are no PRP Pages left to be " "used for building a native NVMe SGL.\n", __func__)); curr_buff = (uint32_t *)prp_page_info->prp_page; msg_phys = (uint64_t *)(uintptr_t)prp_page_info->prp_page_busaddr; /* * Insert the allocated PRP page into the command's PRP page list. This * will be freed when the command is freed. */ TAILQ_INSERT_TAIL(&cm->cm_prp_page_list, prp_page_info, prp_page_link); /* * Check if we are within 1 entry of a page boundary we don't want our * first entry to be a PRP List entry. */ page_mask_result = (uintptr_t)((uint8_t *)curr_buff + prp_size) & page_mask; if (!page_mask_result) { /* Bump up to next page boundary. */ curr_buff = (uint32_t *)((uint8_t *)curr_buff + prp_size); msg_phys = (uint64_t *)((uint8_t *)msg_phys + prp_size); } /* Fill in the chain element and make it an NVMe segment type. */ main_chain_element->Address.High = htole32((uint32_t)((uint64_t)(uintptr_t)msg_phys >> 32)); main_chain_element->Address.Low = htole32((uint32_t)(uintptr_t)msg_phys); main_chain_element->NextChainOffset = 0; main_chain_element->Flags = MPI2_IEEE_SGE_FLAGS_CHAIN_ELEMENT | MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR | MPI26_IEEE_SGE_FLAGS_NSF_NVME_PRP; /* Set SGL pointer to start of contiguous PCIe buffer. */ ptr_sgl = curr_buff; sge_dwords = 2; num_entries = 0; /* * NVMe has a very convoluted PRP format. One PRP is required for each * page or partial page. We need to split up OS SG entries if they are * longer than one page or cross a page boundary. We also have to insert * a PRP list pointer entry as the last entry in each physical page of * the PRP list. * * NOTE: The first PRP "entry" is actually placed in the first SGL entry * in the main message in IEEE 64 format. The 2nd entry in the main * message is the chain element, and the rest of the PRP entries are * built in the contiguous PCIe buffer. */ first_prp_entry = 1; ptr_first_sgl = (uint32_t *)cm->cm_sge; for (i = 0; i < segs_left; i++) { /* Get physical address and length of this SG entry. */ paddr = segs[i].ds_addr; length = segs[i].ds_len; /* * Check whether a given SGE buffer lies on a non-PAGED * boundary if this is not the first page. If so, this is not * expected so have FW build the SGL. */ if ((i != 0) && (((uint32_t)paddr & page_mask) != 0)) { mpr_dprint(sc, MPR_ERROR, "Unaligned SGE while " "building NVMe PRPs, low address is 0x%x\n", (uint32_t)paddr); return 1; } /* Apart from last SGE, if any other SGE boundary is not page * aligned then it means that hole exists. Existence of hole * leads to data corruption. So fallback to IEEE SGEs. */ if (i != (segs_left - 1)) { if (((uint32_t)paddr + length) & page_mask) { mpr_dprint(sc, MPR_ERROR, "Unaligned SGE " "boundary while building NVMe PRPs, low " "address: 0x%x and length: %u\n", (uint32_t)paddr, length); return 1; } } /* Loop while the length is not zero. */ while (length) { /* * Check if we need to put a list pointer here if we are * at page boundary - prp_size. */ page_mask_result = (uintptr_t)((uint8_t *)ptr_sgl + prp_size) & page_mask; if (!page_mask_result) { /* * Need to put a PRP list pointer here. */ msg_phys = (uint64_t *)((uint8_t *)msg_phys + prp_size); *ptr_sgl = htole32((uintptr_t)msg_phys); *(ptr_sgl+1) = htole32((uint64_t)(uintptr_t) msg_phys >> 32); ptr_sgl += sge_dwords; num_entries++; } /* Need to handle if entry will be part of a page. */ offset = (uint32_t)paddr & page_mask; entry_len = PAGE_SIZE - offset; if (first_prp_entry) { /* * Put IEEE entry in first SGE in main message. * (Simple element, System addr, not end of * list.) */ *ptr_first_sgl = htole32((uint32_t)paddr); *(ptr_first_sgl + 1) = htole32((uint32_t)((uint64_t)paddr >> 32)); *(ptr_first_sgl + 2) = htole32(entry_len); *(ptr_first_sgl + 3) = 0; /* No longer the first PRP entry. */ first_prp_entry = 0; } else { /* Put entry in list. */ *ptr_sgl = htole32((uint32_t)paddr); *(ptr_sgl + 1) = htole32((uint32_t)((uint64_t)paddr >> 32)); /* Bump ptr_sgl, msg_phys, and num_entries. */ ptr_sgl += sge_dwords; msg_phys = (uint64_t *)((uint8_t *)msg_phys + prp_size); num_entries++; } /* Bump the phys address by the entry_len. */ paddr += entry_len; /* Decrement length accounting for last partial page. */ if (entry_len > length) length = 0; else length -= entry_len; } } /* Set chain element Length. */ main_chain_element->Length = htole32(num_entries * prp_size); /* Return 0, indicating we built a native SGL. */ return 0; } /* * Add a chain element as the next SGE for the specified command. * Reset cm_sge and cm_sgesize to indicate all the available space. Chains are * only required for IEEE commands. Therefore there is no code for commands * that have the MPR_CM_FLAGS_SGE_SIMPLE flag set (and those commands * shouldn't be requesting chains). */ static int mpr_add_chain(struct mpr_command *cm, int segsleft) { struct mpr_softc *sc = cm->cm_sc; MPI2_REQUEST_HEADER *req; MPI25_IEEE_SGE_CHAIN64 *ieee_sgc; struct mpr_chain *chain; int sgc_size, current_segs, rem_segs, segs_per_frame; uint8_t next_chain_offset = 0; /* * Fail if a command is requesting a chain for SIMPLE SGE's. For SAS3 * only IEEE commands should be requesting chains. Return some error * code other than 0. */ if (cm->cm_flags & MPR_CM_FLAGS_SGE_SIMPLE) { mpr_dprint(sc, MPR_ERROR, "A chain element cannot be added to " "an MPI SGL.\n"); return(ENOBUFS); } sgc_size = sizeof(MPI25_IEEE_SGE_CHAIN64); if (cm->cm_sglsize < sgc_size) panic("MPR: Need SGE Error Code\n"); chain = mpr_alloc_chain(cm->cm_sc); if (chain == NULL) return (ENOBUFS); /* * Note: a double-linked list is used to make it easier to walk for * debugging. */ TAILQ_INSERT_TAIL(&cm->cm_chain_list, chain, chain_link); /* * Need to know if the number of frames left is more than 1 or not. If * more than 1 frame is required, NextChainOffset will need to be set, * which will just be the last segment of the frame. */ rem_segs = 0; if (cm->cm_sglsize < (sgc_size * segsleft)) { /* * rem_segs is the number of segements remaining after the * segments that will go into the current frame. Since it is * known that at least one more frame is required, account for * the chain element. To know if more than one more frame is * required, just check if there will be a remainder after using * the current frame (with this chain) and the next frame. If * so the NextChainOffset must be the last element of the next * frame. */ current_segs = (cm->cm_sglsize / sgc_size) - 1; rem_segs = segsleft - current_segs; segs_per_frame = sc->chain_frame_size / sgc_size; if (rem_segs > segs_per_frame) { next_chain_offset = segs_per_frame - 1; } } ieee_sgc = &((MPI25_SGE_IO_UNION *)cm->cm_sge)->IeeeChain; ieee_sgc->Length = next_chain_offset ? htole32((uint32_t)sc->chain_frame_size) : htole32((uint32_t)rem_segs * (uint32_t)sgc_size); ieee_sgc->NextChainOffset = next_chain_offset; ieee_sgc->Flags = (MPI2_IEEE_SGE_FLAGS_CHAIN_ELEMENT | MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR); ieee_sgc->Address.Low = htole32(chain->chain_busaddr); ieee_sgc->Address.High = htole32(chain->chain_busaddr >> 32); cm->cm_sge = &((MPI25_SGE_IO_UNION *)chain->chain)->IeeeSimple; req = (MPI2_REQUEST_HEADER *)cm->cm_req; req->ChainOffset = (sc->chain_frame_size - sgc_size) >> 4; cm->cm_sglsize = sc->chain_frame_size; return (0); } /* * Add one scatter-gather element to the scatter-gather list for a command. * Maintain cm_sglsize and cm_sge as the remaining size and pointer to the * next SGE to fill in, respectively. In Gen3, the MPI SGL does not have a * chain, so don't consider any chain additions. */ int mpr_push_sge(struct mpr_command *cm, MPI2_SGE_SIMPLE64 *sge, size_t len, int segsleft) { uint32_t saved_buf_len, saved_address_low, saved_address_high; u32 sge_flags; /* * case 1: >=1 more segment, no room for anything (error) * case 2: 1 more segment and enough room for it */ if (cm->cm_sglsize < (segsleft * sizeof(MPI2_SGE_SIMPLE64))) { mpr_dprint(cm->cm_sc, MPR_ERROR, "%s: warning: Not enough room for MPI SGL in frame.\n", __func__); return(ENOBUFS); } KASSERT(segsleft == 1, ("segsleft cannot be more than 1 for an MPI SGL; segsleft = %d\n", segsleft)); /* * There is one more segment left to add for the MPI SGL and there is * enough room in the frame to add it. This is the normal case because * MPI SGL's don't have chains, otherwise something is wrong. * * If this is a bi-directional request, need to account for that * here. Save the pre-filled sge values. These will be used * either for the 2nd SGL or for a single direction SGL. If * cm_out_len is non-zero, this is a bi-directional request, so * fill in the OUT SGL first, then the IN SGL, otherwise just * fill in the IN SGL. Note that at this time, when filling in * 2 SGL's for a bi-directional request, they both use the same * DMA buffer (same cm command). */ saved_buf_len = sge->FlagsLength & 0x00FFFFFF; saved_address_low = sge->Address.Low; saved_address_high = sge->Address.High; if (cm->cm_out_len) { sge->FlagsLength = cm->cm_out_len | ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_HOST_TO_IOC | MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << MPI2_SGE_FLAGS_SHIFT); cm->cm_sglsize -= len; /* Endian Safe code */ sge_flags = sge->FlagsLength; sge->FlagsLength = htole32(sge_flags); sge->Address.High = htole32(sge->Address.High); sge->Address.Low = htole32(sge->Address.Low); bcopy(sge, cm->cm_sge, len); cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); } sge->FlagsLength = saved_buf_len | ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_LAST_ELEMENT | MPI2_SGE_FLAGS_END_OF_LIST | MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << MPI2_SGE_FLAGS_SHIFT); if (cm->cm_flags & MPR_CM_FLAGS_DATAIN) { sge->FlagsLength |= ((uint32_t)(MPI2_SGE_FLAGS_IOC_TO_HOST) << MPI2_SGE_FLAGS_SHIFT); } else { sge->FlagsLength |= ((uint32_t)(MPI2_SGE_FLAGS_HOST_TO_IOC) << MPI2_SGE_FLAGS_SHIFT); } sge->Address.Low = saved_address_low; sge->Address.High = saved_address_high; cm->cm_sglsize -= len; /* Endian Safe code */ sge_flags = sge->FlagsLength; sge->FlagsLength = htole32(sge_flags); sge->Address.High = htole32(sge->Address.High); sge->Address.Low = htole32(sge->Address.Low); bcopy(sge, cm->cm_sge, len); cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); return (0); } /* * Add one IEEE scatter-gather element (chain or simple) to the IEEE scatter- * gather list for a command. Maintain cm_sglsize and cm_sge as the * remaining size and pointer to the next SGE to fill in, respectively. */ int mpr_push_ieee_sge(struct mpr_command *cm, void *sgep, int segsleft) { MPI2_IEEE_SGE_SIMPLE64 *sge = sgep; int error, ieee_sge_size = sizeof(MPI25_SGE_IO_UNION); uint32_t saved_buf_len, saved_address_low, saved_address_high; uint32_t sge_length; /* * case 1: No room for chain or segment (error). * case 2: Two or more segments left but only room for chain. * case 3: Last segment and room for it, so set flags. */ /* * There should be room for at least one element, or there is a big * problem. */ if (cm->cm_sglsize < ieee_sge_size) panic("MPR: Need SGE Error Code\n"); if ((segsleft >= 2) && (cm->cm_sglsize < (ieee_sge_size * 2))) { if ((error = mpr_add_chain(cm, segsleft)) != 0) return (error); } if (segsleft == 1) { /* * If this is a bi-directional request, need to account for that * here. Save the pre-filled sge values. These will be used * either for the 2nd SGL or for a single direction SGL. If * cm_out_len is non-zero, this is a bi-directional request, so * fill in the OUT SGL first, then the IN SGL, otherwise just * fill in the IN SGL. Note that at this time, when filling in * 2 SGL's for a bi-directional request, they both use the same * DMA buffer (same cm command). */ saved_buf_len = sge->Length; saved_address_low = sge->Address.Low; saved_address_high = sge->Address.High; if (cm->cm_out_len) { sge->Length = cm->cm_out_len; sge->Flags = (MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR); cm->cm_sglsize -= ieee_sge_size; /* Endian Safe code */ sge_length = sge->Length; sge->Length = htole32(sge_length); sge->Address.High = htole32(sge->Address.High); sge->Address.Low = htole32(sge->Address.Low); bcopy(sgep, cm->cm_sge, ieee_sge_size); cm->cm_sge = (MPI25_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + ieee_sge_size); } sge->Length = saved_buf_len; sge->Flags = (MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR | MPI25_IEEE_SGE_FLAGS_END_OF_LIST); sge->Address.Low = saved_address_low; sge->Address.High = saved_address_high; } cm->cm_sglsize -= ieee_sge_size; /* Endian Safe code */ sge_length = sge->Length; sge->Length = htole32(sge_length); sge->Address.High = htole32(sge->Address.High); sge->Address.Low = htole32(sge->Address.Low); bcopy(sgep, cm->cm_sge, ieee_sge_size); cm->cm_sge = (MPI25_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + ieee_sge_size); return (0); } /* * Add one dma segment to the scatter-gather list for a command. */ int mpr_add_dmaseg(struct mpr_command *cm, vm_paddr_t pa, size_t len, u_int flags, int segsleft) { MPI2_SGE_SIMPLE64 sge; MPI2_IEEE_SGE_SIMPLE64 ieee_sge; if (!(cm->cm_flags & MPR_CM_FLAGS_SGE_SIMPLE)) { ieee_sge.Flags = (MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR); ieee_sge.Length = len; mpr_from_u64(pa, &ieee_sge.Address); return (mpr_push_ieee_sge(cm, &ieee_sge, segsleft)); } else { /* * This driver always uses 64-bit address elements for * simplicity. */ flags |= MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_64_BIT_ADDRESSING; /* Set Endian safe macro in mpr_push_sge */ sge.FlagsLength = len | (flags << MPI2_SGE_FLAGS_SHIFT); mpr_from_u64(pa, &sge.Address); return (mpr_push_sge(cm, &sge, sizeof sge, segsleft)); } } static void mpr_data_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mpr_softc *sc; struct mpr_command *cm; u_int i, dir, sflags; cm = (struct mpr_command *)arg; sc = cm->cm_sc; /* * In this case, just print out a warning and let the chip tell the * user they did the wrong thing. */ if ((cm->cm_max_segs != 0) && (nsegs > cm->cm_max_segs)) { mpr_dprint(sc, MPR_ERROR, "%s: warning: busdma returned %d " "segments, more than the %d allowed\n", __func__, nsegs, cm->cm_max_segs); } /* * Set up DMA direction flags. Bi-directional requests are also handled * here. In that case, both direction flags will be set. */ sflags = 0; if (cm->cm_flags & MPR_CM_FLAGS_SMP_PASS) { /* * We have to add a special case for SMP passthrough, there * is no easy way to generically handle it. The first * S/G element is used for the command (therefore the * direction bit needs to be set). The second one is used * for the reply. We'll leave it to the caller to make * sure we only have two buffers. */ /* * Even though the busdma man page says it doesn't make * sense to have both direction flags, it does in this case. * We have one s/g element being accessed in each direction. */ dir = BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD; /* * Set the direction flag on the first buffer in the SMP * passthrough request. We'll clear it for the second one. */ sflags |= MPI2_SGE_FLAGS_DIRECTION | MPI2_SGE_FLAGS_END_OF_BUFFER; } else if (cm->cm_flags & MPR_CM_FLAGS_DATAOUT) { sflags |= MPI2_SGE_FLAGS_HOST_TO_IOC; dir = BUS_DMASYNC_PREWRITE; } else dir = BUS_DMASYNC_PREREAD; /* Check if a native SG list is needed for an NVMe PCIe device. */ if (cm->cm_targ && cm->cm_targ->is_nvme && mpr_check_pcie_native_sgl(sc, cm, segs, nsegs) == 0) { /* A native SG list was built, skip to end. */ goto out; } for (i = 0; i < nsegs; i++) { if ((cm->cm_flags & MPR_CM_FLAGS_SMP_PASS) && (i != 0)) { sflags &= ~MPI2_SGE_FLAGS_DIRECTION; } error = mpr_add_dmaseg(cm, segs[i].ds_addr, segs[i].ds_len, sflags, nsegs - i); if (error != 0) { /* Resource shortage, roll back! */ if (ratecheck(&sc->lastfail, &mpr_chainfail_interval)) mpr_dprint(sc, MPR_INFO, "Out of chain frames, " "consider increasing hw.mpr.max_chains.\n"); cm->cm_flags |= MPR_CM_FLAGS_CHAIN_FAILED; mpr_complete_command(sc, cm); return; } } out: bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir); mpr_enqueue_request(sc, cm); return; } static void mpr_data_cb2(void *arg, bus_dma_segment_t *segs, int nsegs, bus_size_t mapsize, int error) { mpr_data_cb(arg, segs, nsegs, error); } /* * This is the routine to enqueue commands ansynchronously. * Note that the only error path here is from bus_dmamap_load(), which can * return EINPROGRESS if it is waiting for resources. Other than this, it's * assumed that if you have a command in-hand, then you have enough credits * to use it. */ int mpr_map_command(struct mpr_softc *sc, struct mpr_command *cm) { int error = 0; if (cm->cm_flags & MPR_CM_FLAGS_USE_UIO) { error = bus_dmamap_load_uio(sc->buffer_dmat, cm->cm_dmamap, &cm->cm_uio, mpr_data_cb2, cm, 0); } else if (cm->cm_flags & MPR_CM_FLAGS_USE_CCB) { error = bus_dmamap_load_ccb(sc->buffer_dmat, cm->cm_dmamap, cm->cm_data, mpr_data_cb, cm, 0); } else if ((cm->cm_data != NULL) && (cm->cm_length != 0)) { error = bus_dmamap_load(sc->buffer_dmat, cm->cm_dmamap, cm->cm_data, cm->cm_length, mpr_data_cb, cm, 0); } else { /* Add a zero-length element as needed */ if (cm->cm_sge != NULL) mpr_add_dmaseg(cm, 0, 0, 0, 1); mpr_enqueue_request(sc, cm); } return (error); } /* * This is the routine to enqueue commands synchronously. An error of * EINPROGRESS from mpr_map_command() is ignored since the command will * be executed and enqueued automatically. Other errors come from msleep(). */ int mpr_wait_command(struct mpr_softc *sc, struct mpr_command **cmp, int timeout, int sleep_flag) { int error, rc; struct timeval cur_time, start_time; struct mpr_command *cm = *cmp; if (sc->mpr_flags & MPR_FLAGS_DIAGRESET) return EBUSY; cm->cm_complete = NULL; cm->cm_flags |= (MPR_CM_FLAGS_WAKEUP + MPR_CM_FLAGS_POLLED); error = mpr_map_command(sc, cm); if ((error != 0) && (error != EINPROGRESS)) return (error); // Check for context and wait for 50 mSec at a time until time has // expired or the command has finished. If msleep can't be used, need // to poll. #if __FreeBSD_version >= 1000029 if (curthread->td_no_sleeping) #else //__FreeBSD_version < 1000029 if (curthread->td_pflags & TDP_NOSLEEPING) #endif //__FreeBSD_version >= 1000029 sleep_flag = NO_SLEEP; getmicrouptime(&start_time); if (mtx_owned(&sc->mpr_mtx) && sleep_flag == CAN_SLEEP) { error = msleep(cm, &sc->mpr_mtx, 0, "mprwait", timeout*hz); if (error == EWOULDBLOCK) { /* * Record the actual elapsed time in the case of a * timeout for the message below. */ getmicrouptime(&cur_time); timevalsub(&cur_time, &start_time); } } else { while ((cm->cm_flags & MPR_CM_FLAGS_COMPLETE) == 0) { mpr_intr_locked(sc); if (sleep_flag == CAN_SLEEP) pause("mprwait", hz/20); else DELAY(50000); getmicrouptime(&cur_time); timevalsub(&cur_time, &start_time); if (cur_time.tv_sec > timeout) { error = EWOULDBLOCK; break; } } } if (error == EWOULDBLOCK) { if (cm->cm_timeout_handler == NULL) { mpr_dprint(sc, MPR_FAULT, "Calling Reinit from %s, timeout=%d," " elapsed=%jd\n", __func__, timeout, (intmax_t)cur_time.tv_sec); rc = mpr_reinit(sc); mpr_dprint(sc, MPR_FAULT, "Reinit %s\n", (rc == 0) ? "success" : "failed"); } else cm->cm_timeout_handler(sc, cm); if (sc->mpr_flags & MPR_FLAGS_REALLOCATED) { /* * Tell the caller that we freed the command in a * reinit. */ *cmp = NULL; } error = ETIMEDOUT; } return (error); } /* * This is the routine to enqueue a command synchonously and poll for * completion. Its use should be rare. */ int mpr_request_polled(struct mpr_softc *sc, struct mpr_command **cmp) { int error, rc; struct timeval cur_time, start_time; struct mpr_command *cm = *cmp; error = 0; cm->cm_flags |= MPR_CM_FLAGS_POLLED; cm->cm_complete = NULL; mpr_map_command(sc, cm); getmicrouptime(&start_time); while ((cm->cm_flags & MPR_CM_FLAGS_COMPLETE) == 0) { mpr_intr_locked(sc); if (mtx_owned(&sc->mpr_mtx)) msleep(&sc->msleep_fake_chan, &sc->mpr_mtx, 0, "mprpoll", hz/20); else pause("mprpoll", hz/20); /* * Check for real-time timeout and fail if more than 60 seconds. */ getmicrouptime(&cur_time); timevalsub(&cur_time, &start_time); if (cur_time.tv_sec > 60) { mpr_dprint(sc, MPR_FAULT, "polling failed\n"); error = ETIMEDOUT; break; } } if (error) { mpr_dprint(sc, MPR_FAULT, "Calling Reinit from %s\n", __func__); rc = mpr_reinit(sc); mpr_dprint(sc, MPR_FAULT, "Reinit %s\n", (rc == 0) ? "success" : "failed"); if (sc->mpr_flags & MPR_FLAGS_REALLOCATED) { /* * Tell the caller that we freed the command in a * reinit. */ *cmp = NULL; } } return (error); } /* * The MPT driver had a verbose interface for config pages. In this driver, * reduce it to much simpler terms, similar to the Linux driver. */ int mpr_read_config_page(struct mpr_softc *sc, struct mpr_config_params *params) { MPI2_CONFIG_REQUEST *req; struct mpr_command *cm; int error; if (sc->mpr_flags & MPR_FLAGS_BUSY) { return (EBUSY); } cm = mpr_alloc_command(sc); if (cm == NULL) { return (EBUSY); } req = (MPI2_CONFIG_REQUEST *)cm->cm_req; req->Function = MPI2_FUNCTION_CONFIG; req->Action = params->action; req->SGLFlags = 0; req->ChainOffset = 0; req->PageAddress = params->page_address; if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { MPI2_CONFIG_EXTENDED_PAGE_HEADER *hdr; hdr = ¶ms->hdr.Ext; req->ExtPageType = hdr->ExtPageType; req->ExtPageLength = hdr->ExtPageLength; req->Header.PageType = MPI2_CONFIG_PAGETYPE_EXTENDED; req->Header.PageLength = 0; /* Must be set to zero */ req->Header.PageNumber = hdr->PageNumber; req->Header.PageVersion = hdr->PageVersion; } else { MPI2_CONFIG_PAGE_HEADER *hdr; hdr = ¶ms->hdr.Struct; req->Header.PageType = hdr->PageType; req->Header.PageNumber = hdr->PageNumber; req->Header.PageLength = hdr->PageLength; req->Header.PageVersion = hdr->PageVersion; } cm->cm_data = params->buffer; cm->cm_length = params->length; if (cm->cm_data != NULL) { cm->cm_sge = &req->PageBufferSGE; cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION); cm->cm_flags = MPR_CM_FLAGS_SGE_SIMPLE | MPR_CM_FLAGS_DATAIN; } else cm->cm_sge = NULL; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete_data = params; if (params->callback != NULL) { cm->cm_complete = mpr_config_complete; return (mpr_map_command(sc, cm)); } else { error = mpr_wait_command(sc, &cm, 0, CAN_SLEEP); if (error) { mpr_dprint(sc, MPR_FAULT, "Error %d reading config page\n", error); if (cm != NULL) mpr_free_command(sc, cm); return (error); } mpr_config_complete(sc, cm); } return (0); } int mpr_write_config_page(struct mpr_softc *sc, struct mpr_config_params *params) { return (EINVAL); } static void mpr_config_complete(struct mpr_softc *sc, struct mpr_command *cm) { MPI2_CONFIG_REPLY *reply; struct mpr_config_params *params; MPR_FUNCTRACE(sc); params = cm->cm_complete_data; if (cm->cm_data != NULL) { bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); } /* * XXX KDM need to do more error recovery? This results in the * device in question not getting probed. */ if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { params->status = MPI2_IOCSTATUS_BUSY; goto done; } reply = (MPI2_CONFIG_REPLY *)cm->cm_reply; if (reply == NULL) { params->status = MPI2_IOCSTATUS_BUSY; goto done; } params->status = reply->IOCStatus; if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { params->hdr.Ext.ExtPageType = reply->ExtPageType; params->hdr.Ext.ExtPageLength = reply->ExtPageLength; params->hdr.Ext.PageType = reply->Header.PageType; params->hdr.Ext.PageNumber = reply->Header.PageNumber; params->hdr.Ext.PageVersion = reply->Header.PageVersion; } else { params->hdr.Struct.PageType = reply->Header.PageType; params->hdr.Struct.PageNumber = reply->Header.PageNumber; params->hdr.Struct.PageLength = reply->Header.PageLength; params->hdr.Struct.PageVersion = reply->Header.PageVersion; } done: mpr_free_command(sc, cm); if (params->callback != NULL) params->callback(sc, params); return; } Index: head/sys/dev/mpr/mpr_sas.c =================================================================== --- head/sys/dev/mpr/mpr_sas.c (revision 349848) +++ head/sys/dev/mpr/mpr_sas.c (revision 349849) @@ -1,3929 +1,3931 @@ /*- * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2016 Avago Technologies * Copyright 2000-2020 Broadcom Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Broadcom Inc. (LSI) MPT-Fusion Host Adapter FreeBSD * */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT3 */ /* TODO Move headers to mprvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if __FreeBSD_version >= 900026 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #define MPRSAS_DISCOVERY_TIMEOUT 20 #define MPRSAS_MAX_DISCOVERY_TIMEOUTS 10 /* 200 seconds */ /* * static array to check SCSI OpCode for EEDP protection bits */ #define PRO_R MPI2_SCSIIO_EEDPFLAGS_CHECK_REMOVE_OP #define PRO_W MPI2_SCSIIO_EEDPFLAGS_INSERT_OP #define PRO_V MPI2_SCSIIO_EEDPFLAGS_INSERT_OP static uint8_t op_code_prot[256] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; MALLOC_DEFINE(M_MPRSAS, "MPRSAS", "MPR SAS memory"); static void mprsas_remove_device(struct mpr_softc *, struct mpr_command *); static void mprsas_remove_complete(struct mpr_softc *, struct mpr_command *); static void mprsas_action(struct cam_sim *sim, union ccb *ccb); static void mprsas_poll(struct cam_sim *sim); static void mprsas_scsiio_timeout(void *data); static void mprsas_abort_complete(struct mpr_softc *sc, struct mpr_command *cm); static void mprsas_action_scsiio(struct mprsas_softc *, union ccb *); static void mprsas_scsiio_complete(struct mpr_softc *, struct mpr_command *); static void mprsas_action_resetdev(struct mprsas_softc *, union ccb *); static void mprsas_resetdev_complete(struct mpr_softc *, struct mpr_command *); static int mprsas_send_abort(struct mpr_softc *sc, struct mpr_command *tm, struct mpr_command *cm); static void mprsas_async(void *callback_arg, uint32_t code, struct cam_path *path, void *arg); #if (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) static void mprsas_check_eedp(struct mpr_softc *sc, struct cam_path *path, struct ccb_getdev *cgd); static void mprsas_read_cap_done(struct cam_periph *periph, union ccb *done_ccb); #endif static int mprsas_send_portenable(struct mpr_softc *sc); static void mprsas_portenable_complete(struct mpr_softc *sc, struct mpr_command *cm); #if __FreeBSD_version >= 900026 static void mprsas_smpio_complete(struct mpr_softc *sc, struct mpr_command *cm); static void mprsas_send_smpcmd(struct mprsas_softc *sassc, union ccb *ccb, uint64_t sasaddr); static void mprsas_action_smpio(struct mprsas_softc *sassc, union ccb *ccb); #endif //FreeBSD_version >= 900026 struct mprsas_target * mprsas_find_target_by_handle(struct mprsas_softc *sassc, int start, uint16_t handle) { struct mprsas_target *target; int i; for (i = start; i < sassc->maxtargets; i++) { target = &sassc->targets[i]; if (target->handle == handle) return (target); } return (NULL); } /* we need to freeze the simq during attach and diag reset, to avoid failing * commands before device handles have been found by discovery. Since * discovery involves reading config pages and possibly sending commands, * discovery actions may continue even after we receive the end of discovery * event, so refcount discovery actions instead of assuming we can unfreeze * the simq when we get the event. */ void mprsas_startup_increment(struct mprsas_softc *sassc) { MPR_FUNCTRACE(sassc->sc); if ((sassc->flags & MPRSAS_IN_STARTUP) != 0) { if (sassc->startup_refcount++ == 0) { /* just starting, freeze the simq */ mpr_dprint(sassc->sc, MPR_INIT, "%s freezing simq\n", __func__); #if (__FreeBSD_version >= 1000039) || \ ((__FreeBSD_version < 1000000) && (__FreeBSD_version >= 902502)) xpt_hold_boot(); #endif xpt_freeze_simq(sassc->sim, 1); } mpr_dprint(sassc->sc, MPR_INIT, "%s refcount %u\n", __func__, sassc->startup_refcount); } } void mprsas_release_simq_reinit(struct mprsas_softc *sassc) { if (sassc->flags & MPRSAS_QUEUE_FROZEN) { sassc->flags &= ~MPRSAS_QUEUE_FROZEN; xpt_release_simq(sassc->sim, 1); mpr_dprint(sassc->sc, MPR_INFO, "Unfreezing SIM queue\n"); } } void mprsas_startup_decrement(struct mprsas_softc *sassc) { MPR_FUNCTRACE(sassc->sc); if ((sassc->flags & MPRSAS_IN_STARTUP) != 0) { if (--sassc->startup_refcount == 0) { /* finished all discovery-related actions, release * the simq and rescan for the latest topology. */ mpr_dprint(sassc->sc, MPR_INIT, "%s releasing simq\n", __func__); sassc->flags &= ~MPRSAS_IN_STARTUP; xpt_release_simq(sassc->sim, 1); #if (__FreeBSD_version >= 1000039) || \ ((__FreeBSD_version < 1000000) && (__FreeBSD_version >= 902502)) xpt_release_boot(); #else mprsas_rescan_target(sassc->sc, NULL); #endif } mpr_dprint(sassc->sc, MPR_INIT, "%s refcount %u\n", __func__, sassc->startup_refcount); } } /* * The firmware requires us to stop sending commands when we're doing task * management. * use. * XXX The logic for serializing the device has been made lazy and moved to * mprsas_prepare_for_tm(). */ struct mpr_command * mprsas_alloc_tm(struct mpr_softc *sc) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpr_command *tm; MPR_FUNCTRACE(sc); tm = mpr_alloc_high_priority_command(sc); if (tm == NULL) return (NULL); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->Function = MPI2_FUNCTION_SCSI_TASK_MGMT; return tm; } void mprsas_free_tm(struct mpr_softc *sc, struct mpr_command *tm) { int target_id = 0xFFFFFFFF; MPR_FUNCTRACE(sc); if (tm == NULL) return; /* * For TM's the devq is frozen for the device. Unfreeze it here and * free the resources used for freezing the devq. Must clear the * INRESET flag as well or scsi I/O will not work. */ if (tm->cm_targ != NULL) { tm->cm_targ->flags &= ~MPRSAS_TARGET_INRESET; target_id = tm->cm_targ->tid; } if (tm->cm_ccb) { mpr_dprint(sc, MPR_INFO, "Unfreezing devq for target ID %d\n", target_id); xpt_release_devq(tm->cm_ccb->ccb_h.path, 1, TRUE); xpt_free_path(tm->cm_ccb->ccb_h.path); xpt_free_ccb(tm->cm_ccb); } mpr_free_high_priority_command(sc, tm); } void mprsas_rescan_target(struct mpr_softc *sc, struct mprsas_target *targ) { struct mprsas_softc *sassc = sc->sassc; path_id_t pathid; target_id_t targetid; union ccb *ccb; MPR_FUNCTRACE(sc); pathid = cam_sim_path(sassc->sim); if (targ == NULL) targetid = CAM_TARGET_WILDCARD; else targetid = targ - sassc->targets; /* * Allocate a CCB and schedule a rescan. */ ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mpr_dprint(sc, MPR_ERROR, "unable to alloc CCB for rescan\n"); return; } if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { mpr_dprint(sc, MPR_ERROR, "unable to create path for rescan\n"); xpt_free_ccb(ccb); return; } if (targetid == CAM_TARGET_WILDCARD) ccb->ccb_h.func_code = XPT_SCAN_BUS; else ccb->ccb_h.func_code = XPT_SCAN_TGT; mpr_dprint(sc, MPR_TRACE, "%s targetid %u\n", __func__, targetid); xpt_rescan(ccb); } static void mprsas_log_command(struct mpr_command *cm, u_int level, const char *fmt, ...) { struct sbuf sb; va_list ap; char str[192]; char path_str[64]; if (cm == NULL) return; /* No need to be in here if debugging isn't enabled */ if ((cm->cm_sc->mpr_debug & level) == 0) return; sbuf_new(&sb, str, sizeof(str), 0); va_start(ap, fmt); if (cm->cm_ccb != NULL) { xpt_path_string(cm->cm_ccb->csio.ccb_h.path, path_str, sizeof(path_str)); sbuf_cat(&sb, path_str); if (cm->cm_ccb->ccb_h.func_code == XPT_SCSI_IO) { scsi_command_string(&cm->cm_ccb->csio, &sb); sbuf_printf(&sb, "length %d ", cm->cm_ccb->csio.dxfer_len); } } else { sbuf_printf(&sb, "(noperiph:%s%d:%u:%u:%u): ", cam_sim_name(cm->cm_sc->sassc->sim), cam_sim_unit(cm->cm_sc->sassc->sim), cam_sim_bus(cm->cm_sc->sassc->sim), cm->cm_targ ? cm->cm_targ->tid : 0xFFFFFFFF, cm->cm_lun); } sbuf_printf(&sb, "SMID %u ", cm->cm_desc.Default.SMID); sbuf_vprintf(&sb, fmt, ap); sbuf_finish(&sb); mpr_print_field(cm->cm_sc, "%s", sbuf_data(&sb)); va_end(ap); } static void mprsas_remove_volume(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; struct mprsas_target *targ; uint16_t handle; MPR_FUNCTRACE(sc); reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; targ = tm->cm_targ; if (reply == NULL) { /* XXX retry the remove after the diag reset completes? */ mpr_dprint(sc, MPR_FAULT, "%s NULL reply resetting device " "0x%04x\n", __func__, handle); mprsas_free_tm(sc, tm); return; } if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mpr_dprint(sc, MPR_ERROR, "IOCStatus = 0x%x while resetting " "device 0x%x\n", le16toh(reply->IOCStatus), handle); } mpr_dprint(sc, MPR_XINFO, "Reset aborted %u commands\n", le32toh(reply->TerminationCount)); mpr_free_reply(sc, tm->cm_reply_data); tm->cm_reply = NULL; /* Ensures the reply won't get re-freed */ mpr_dprint(sc, MPR_XINFO, "clearing target %u handle 0x%04x\n", targ->tid, handle); /* * Don't clear target if remove fails because things will get confusing. * Leave the devname and sasaddr intact so that we know to avoid reusing * this target id if possible, and so we can assign the same target id * to this device if it comes back in the future. */ if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SUCCESS) { targ = tm->cm_targ; targ->handle = 0x0; targ->encl_handle = 0x0; targ->encl_level_valid = 0x0; targ->encl_level = 0x0; targ->connector_name[0] = ' '; targ->connector_name[1] = ' '; targ->connector_name[2] = ' '; targ->connector_name[3] = ' '; targ->encl_slot = 0x0; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; targ->devinfo = 0x0; targ->flags = 0x0; targ->scsi_req_desc_type = 0; } mprsas_free_tm(sc, tm); } /* * No Need to call "MPI2_SAS_OP_REMOVE_DEVICE" For Volume removal. * Otherwise Volume Delete is same as Bare Drive Removal. */ void mprsas_prepare_volume_remove(struct mprsas_softc *sassc, uint16_t handle) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpr_softc *sc; struct mpr_command *cm; struct mprsas_target *targ = NULL; MPR_FUNCTRACE(sassc->sc); sc = sassc->sc; targ = mprsas_find_target_by_handle(sassc, 0, handle); if (targ == NULL) { /* FIXME: what is the action? */ /* We don't know about this device? */ mpr_dprint(sc, MPR_ERROR, "%s %d : invalid handle 0x%x \n", __func__,__LINE__, handle); return; } targ->flags |= MPRSAS_TARGET_INREMOVAL; cm = mprsas_alloc_tm(sc); if (cm == NULL) { mpr_dprint(sc, MPR_ERROR, "%s: command alloc failure\n", __func__); return; } mprsas_rescan_target(sc, targ); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)cm->cm_req; req->DevHandle = targ->handle; req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; if (!targ->is_nvme || sc->custom_nvme_tm_handling) { /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; } else { /* PCIe Protocol Level Reset*/ req->MsgFlags = MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE; } cm->cm_targ = targ; cm->cm_data = NULL; cm->cm_complete = mprsas_remove_volume; cm->cm_complete_data = (void *)(uintptr_t)handle; mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n", __func__, targ->tid); mprsas_prepare_for_tm(sc, cm, targ, CAM_LUN_WILDCARD); mpr_map_command(sc, cm); } /* * The firmware performs debounce on the link to avoid transient link errors * and false removals. When it does decide that link has been lost and a * device needs to go away, it expects that the host will perform a target reset * and then an op remove. The reset has the side-effect of aborting any * outstanding requests for the device, which is required for the op-remove to * succeed. It's not clear if the host should check for the device coming back * alive after the reset. */ void mprsas_prepare_remove(struct mprsas_softc *sassc, uint16_t handle) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpr_softc *sc; struct mpr_command *tm; struct mprsas_target *targ = NULL; MPR_FUNCTRACE(sassc->sc); sc = sassc->sc; targ = mprsas_find_target_by_handle(sassc, 0, handle); if (targ == NULL) { /* FIXME: what is the action? */ /* We don't know about this device? */ mpr_dprint(sc, MPR_ERROR, "%s : invalid handle 0x%x \n", __func__, handle); return; } targ->flags |= MPRSAS_TARGET_INREMOVAL; tm = mprsas_alloc_tm(sc); if (tm == NULL) { mpr_dprint(sc, MPR_ERROR, "%s: command alloc failure\n", __func__); return; } mprsas_rescan_target(sc, targ); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; memset(req, 0, sizeof(*req)); req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; tm->cm_targ = targ; tm->cm_data = NULL; tm->cm_complete = mprsas_remove_device; tm->cm_complete_data = (void *)(uintptr_t)handle; mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n", __func__, targ->tid); mprsas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD); mpr_map_command(sc, tm); } static void mprsas_remove_device(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SAS_IOUNIT_CONTROL_REQUEST *req; struct mprsas_target *targ; struct mpr_command *next_cm; uint16_t handle; MPR_FUNCTRACE(sc); reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for remove of " "handle %#04x! This should not happen!\n", __func__, tm->cm_flags, handle); } if (reply == NULL) { /* XXX retry the remove after the diag reset completes? */ mpr_dprint(sc, MPR_FAULT, "%s NULL reply resetting device " "0x%04x\n", __func__, handle); mprsas_free_tm(sc, tm); return; } if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mpr_dprint(sc, MPR_ERROR, "IOCStatus = 0x%x while resetting " "device 0x%x\n", le16toh(reply->IOCStatus), handle); } mpr_dprint(sc, MPR_XINFO, "Reset aborted %u commands\n", le32toh(reply->TerminationCount)); mpr_free_reply(sc, tm->cm_reply_data); tm->cm_reply = NULL; /* Ensures the reply won't get re-freed */ /* Reuse the existing command */ req = (MPI2_SAS_IOUNIT_CONTROL_REQUEST *)tm->cm_req; memset(req, 0, sizeof(*req)); req->Function = MPI2_FUNCTION_SAS_IO_UNIT_CONTROL; req->Operation = MPI2_SAS_OP_REMOVE_DEVICE; req->DevHandle = htole16(handle); tm->cm_data = NULL; tm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; tm->cm_complete = mprsas_remove_complete; tm->cm_complete_data = (void *)(uintptr_t)handle; mpr_map_command(sc, tm); mpr_dprint(sc, MPR_INFO, "clearing target %u handle 0x%04x\n", targ->tid, handle); if (targ->encl_level_valid) { mpr_dprint(sc, MPR_INFO, "At enclosure level %d, slot %d, " "connector name (%4s)\n", targ->encl_level, targ->encl_slot, targ->connector_name); } TAILQ_FOREACH_SAFE(tm, &targ->commands, cm_link, next_cm) { union ccb *ccb; mpr_dprint(sc, MPR_XINFO, "Completing missed command %p\n", tm); ccb = tm->cm_complete_data; mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); mprsas_scsiio_complete(sc, tm); } } static void mprsas_remove_complete(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SAS_IOUNIT_CONTROL_REPLY *reply; uint16_t handle; struct mprsas_target *targ; struct mprsas_lun *lun; MPR_FUNCTRACE(sc); reply = (MPI2_SAS_IOUNIT_CONTROL_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_XINFO, "%s: cm_flags = %#x for remove of " "handle %#04x! This should not happen!\n", __func__, tm->cm_flags, handle); mprsas_free_tm(sc, tm); return; } if (reply == NULL) { /* most likely a chip reset */ mpr_dprint(sc, MPR_FAULT, "%s NULL reply removing device " "0x%04x\n", __func__, handle); mprsas_free_tm(sc, tm); return; } mpr_dprint(sc, MPR_XINFO, "%s on handle 0x%04x, IOCStatus= 0x%x\n", __func__, handle, le16toh(reply->IOCStatus)); /* * Don't clear target if remove fails because things will get confusing. * Leave the devname and sasaddr intact so that we know to avoid reusing * this target id if possible, and so we can assign the same target id * to this device if it comes back in the future. */ if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SUCCESS) { targ = tm->cm_targ; targ->handle = 0x0; targ->encl_handle = 0x0; targ->encl_level_valid = 0x0; targ->encl_level = 0x0; targ->connector_name[0] = ' '; targ->connector_name[1] = ' '; targ->connector_name[2] = ' '; targ->connector_name[3] = ' '; targ->encl_slot = 0x0; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; targ->devinfo = 0x0; targ->flags = 0x0; targ->scsi_req_desc_type = 0; while (!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPR); } } mprsas_free_tm(sc, tm); } static int mprsas_register_events(struct mpr_softc *sc) { uint8_t events[16]; bzero(events, 16); setbit(events, MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_SAS_DISCOVERY); setbit(events, MPI2_EVENT_SAS_BROADCAST_PRIMITIVE); setbit(events, MPI2_EVENT_SAS_INIT_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_SAS_INIT_TABLE_OVERFLOW); setbit(events, MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST); setbit(events, MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST); setbit(events, MPI2_EVENT_IR_VOLUME); setbit(events, MPI2_EVENT_IR_PHYSICAL_DISK); setbit(events, MPI2_EVENT_IR_OPERATION_STATUS); setbit(events, MPI2_EVENT_TEMP_THRESHOLD); setbit(events, MPI2_EVENT_SAS_DEVICE_DISCOVERY_ERROR); if (sc->facts->MsgVersion >= MPI2_VERSION_02_06) { setbit(events, MPI2_EVENT_ACTIVE_CABLE_EXCEPTION); if (sc->mpr_flags & MPR_FLAGS_GEN35_IOC) { setbit(events, MPI2_EVENT_PCIE_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_PCIE_ENUMERATION); setbit(events, MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST); } } mpr_register_events(sc, events, mprsas_evt_handler, NULL, &sc->sassc->mprsas_eh); return (0); } int mpr_attach_sas(struct mpr_softc *sc) { struct mprsas_softc *sassc; cam_status status; int unit, error = 0, reqs; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_INIT, "%s entered\n", __func__); sassc = malloc(sizeof(struct mprsas_softc), M_MPR, M_WAITOK|M_ZERO); if (!sassc) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SAS subsystem memory\n"); return (ENOMEM); } /* * XXX MaxTargets could change during a reinit. Since we don't * resize the targets[] array during such an event, cache the value * of MaxTargets here so that we don't get into trouble later. This * should move into the reinit logic. */ sassc->maxtargets = sc->facts->MaxTargets + sc->facts->MaxVolumes; sassc->targets = malloc(sizeof(struct mprsas_target) * sassc->maxtargets, M_MPR, M_WAITOK|M_ZERO); if (!sassc->targets) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SAS target memory\n"); free(sassc, M_MPR); return (ENOMEM); } sc->sassc = sassc; sassc->sc = sc; reqs = sc->num_reqs - sc->num_prireqs - 1; if ((sassc->devq = cam_simq_alloc(reqs)) == NULL) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SIMQ\n"); error = ENOMEM; goto out; } unit = device_get_unit(sc->mpr_dev); sassc->sim = cam_sim_alloc(mprsas_action, mprsas_poll, "mpr", sassc, unit, &sc->mpr_mtx, reqs, reqs, sassc->devq); if (sassc->sim == NULL) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Cannot allocate SIM\n"); error = EINVAL; goto out; } TAILQ_INIT(&sassc->ev_queue); /* Initialize taskqueue for Event Handling */ TASK_INIT(&sassc->ev_task, 0, mprsas_firmware_event_work, sc); sassc->ev_tq = taskqueue_create("mpr_taskq", M_NOWAIT | M_ZERO, taskqueue_thread_enqueue, &sassc->ev_tq); taskqueue_start_threads(&sassc->ev_tq, 1, PRIBIO, "%s taskq", device_get_nameunit(sc->mpr_dev)); mpr_lock(sc); /* * XXX There should be a bus for every port on the adapter, but since * we're just going to fake the topology for now, we'll pretend that * everything is just a target on a single bus. */ if ((error = xpt_bus_register(sassc->sim, sc->mpr_dev, 0)) != 0) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Error %d registering SCSI bus\n", error); mpr_unlock(sc); goto out; } /* * Assume that discovery events will start right away. * * Hold off boot until discovery is complete. */ sassc->flags |= MPRSAS_IN_STARTUP | MPRSAS_IN_DISCOVERY; sc->sassc->startup_refcount = 0; mprsas_startup_increment(sassc); callout_init(&sassc->discovery_callout, 1 /*mpsafe*/); /* * Register for async events so we can determine the EEDP * capabilities of devices. */ status = xpt_create_path(&sassc->path, /*periph*/NULL, cam_sim_path(sc->sassc->sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); if (status != CAM_REQ_CMP) { mpr_dprint(sc, MPR_INIT|MPR_ERROR, "Error %#x creating sim path\n", status); sassc->path = NULL; } else { int event; #if (__FreeBSD_version >= 1000006) || \ ((__FreeBSD_version >= 901503) && (__FreeBSD_version < 1000000)) event = AC_ADVINFO_CHANGED | AC_FOUND_DEVICE; #else event = AC_FOUND_DEVICE; #endif /* * Prior to the CAM locking improvements, we can't call * xpt_register_async() with a particular path specified. * * If a path isn't specified, xpt_register_async() will * generate a wildcard path and acquire the XPT lock while * it calls xpt_action() to execute the XPT_SASYNC_CB CCB. * It will then drop the XPT lock once that is done. * * If a path is specified for xpt_register_async(), it will * not acquire and drop the XPT lock around the call to * xpt_action(). xpt_action() asserts that the caller * holds the SIM lock, so the SIM lock has to be held when * calling xpt_register_async() when the path is specified. * * But xpt_register_async calls xpt_for_all_devices(), * which calls xptbustraverse(), which will acquire each * SIM lock. When it traverses our particular bus, it will * necessarily acquire the SIM lock, which will lead to a * recursive lock acquisition. * * The CAM locking changes fix this problem by acquiring * the XPT topology lock around bus traversal in * xptbustraverse(), so the caller can hold the SIM lock * and it does not cause a recursive lock acquisition. * * These __FreeBSD_version values are approximate, especially * for stable/10, which is two months later than the actual * change. */ #if (__FreeBSD_version < 1000703) || \ ((__FreeBSD_version >= 1100000) && (__FreeBSD_version < 1100002)) mpr_unlock(sc); status = xpt_register_async(event, mprsas_async, sc, NULL); mpr_lock(sc); #else status = xpt_register_async(event, mprsas_async, sc, sassc->path); #endif if (status != CAM_REQ_CMP) { mpr_dprint(sc, MPR_ERROR, "Error %#x registering async handler for " "AC_ADVINFO_CHANGED events\n", status); xpt_free_path(sassc->path); sassc->path = NULL; } } if (status != CAM_REQ_CMP) { /* * EEDP use is the exception, not the rule. * Warn the user, but do not fail to attach. */ mpr_printf(sc, "EEDP capabilities disabled.\n"); } mpr_unlock(sc); mprsas_register_events(sc); out: if (error) mpr_detach_sas(sc); mpr_dprint(sc, MPR_INIT, "%s exit, error= %d\n", __func__, error); return (error); } int mpr_detach_sas(struct mpr_softc *sc) { struct mprsas_softc *sassc; struct mprsas_lun *lun, *lun_tmp; struct mprsas_target *targ; int i; MPR_FUNCTRACE(sc); if (sc->sassc == NULL) return (0); sassc = sc->sassc; mpr_deregister_events(sc, sassc->mprsas_eh); /* * Drain and free the event handling taskqueue with the lock * unheld so that any parallel processing tasks drain properly * without deadlocking. */ if (sassc->ev_tq != NULL) taskqueue_free(sassc->ev_tq); /* Make sure CAM doesn't wedge if we had to bail out early. */ mpr_lock(sc); while (sassc->startup_refcount != 0) mprsas_startup_decrement(sassc); /* Deregister our async handler */ if (sassc->path != NULL) { xpt_register_async(0, mprsas_async, sc, sassc->path); xpt_free_path(sassc->path); sassc->path = NULL; } if (sassc->flags & MPRSAS_IN_STARTUP) xpt_release_simq(sassc->sim, 1); if (sassc->sim != NULL) { xpt_bus_deregister(cam_sim_path(sassc->sim)); cam_sim_free(sassc->sim, FALSE); } mpr_unlock(sc); if (sassc->devq != NULL) cam_simq_free(sassc->devq); for (i = 0; i < sassc->maxtargets; i++) { targ = &sassc->targets[i]; SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) { free(lun, M_MPR); } } free(sassc->targets, M_MPR); free(sassc, M_MPR); sc->sassc = NULL; return (0); } void mprsas_discovery_end(struct mprsas_softc *sassc) { struct mpr_softc *sc = sassc->sc; MPR_FUNCTRACE(sc); if (sassc->flags & MPRSAS_DISCOVERY_TIMEOUT_PENDING) callout_stop(&sassc->discovery_callout); /* * After discovery has completed, check the mapping table for any * missing devices and update their missing counts. Only do this once * whenever the driver is initialized so that missing counts aren't * updated unnecessarily. Note that just because discovery has * completed doesn't mean that events have been processed yet. The * check_devices function is a callout timer that checks if ALL devices * are missing. If so, it will wait a little longer for events to * complete and keep resetting itself until some device in the mapping * table is not missing, meaning that event processing has started. */ if (sc->track_mapping_events) { mpr_dprint(sc, MPR_XINFO | MPR_MAPPING, "Discovery has " "completed. Check for missing devices in the mapping " "table.\n"); callout_reset(&sc->device_check_callout, MPR_MISSING_CHECK_DELAY * hz, mpr_mapping_check_devices, sc); } } static void mprsas_action(struct cam_sim *sim, union ccb *ccb) { struct mprsas_softc *sassc; sassc = cam_sim_softc(sim); MPR_FUNCTRACE(sassc->sc); mpr_dprint(sassc->sc, MPR_TRACE, "ccb func_code 0x%x\n", ccb->ccb_h.func_code); mtx_assert(&sassc->sc->mpr_mtx, MA_OWNED); switch (ccb->ccb_h.func_code) { case XPT_PATH_INQ: { struct ccb_pathinq *cpi = &ccb->cpi; struct mpr_softc *sc = sassc->sc; cpi->version_num = 1; cpi->hba_inquiry = PI_SDTR_ABLE|PI_TAG_ABLE|PI_WIDE_16; cpi->target_sprt = 0; #if (__FreeBSD_version >= 1000039) || \ ((__FreeBSD_version < 1000000) && (__FreeBSD_version >= 902502)) cpi->hba_misc = PIM_NOBUSRESET | PIM_UNMAPPED | PIM_NOSCAN; #else cpi->hba_misc = PIM_NOBUSRESET | PIM_UNMAPPED; #endif cpi->hba_eng_cnt = 0; cpi->max_target = sassc->maxtargets - 1; cpi->max_lun = 255; /* * initiator_id is set here to an ID outside the set of valid * target IDs (including volumes). */ cpi->initiator_id = sassc->maxtargets; strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN); strlcpy(cpi->hba_vid, "Avago Tech", HBA_IDLEN); strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN); cpi->unit_number = cam_sim_unit(sim); cpi->bus_id = cam_sim_bus(sim); /* * XXXSLM-I think this needs to change based on config page or * something instead of hardcoded to 150000. */ cpi->base_transfer_speed = 150000; cpi->transport = XPORT_SAS; cpi->transport_version = 0; cpi->protocol = PROTO_SCSI; cpi->protocol_version = SCSI_REV_SPC; cpi->maxio = sc->maxio; mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); break; } case XPT_GET_TRAN_SETTINGS: { struct ccb_trans_settings *cts; struct ccb_trans_settings_sas *sas; struct ccb_trans_settings_scsi *scsi; struct mprsas_target *targ; cts = &ccb->cts; sas = &cts->xport_specific.sas; scsi = &cts->proto_specific.scsi; KASSERT(cts->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_GET_TRAN_SETTINGS\n", cts->ccb_h.target_id)); targ = &sassc->targets[cts->ccb_h.target_id]; if (targ->handle == 0x0) { mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); break; } cts->protocol_version = SCSI_REV_SPC2; cts->transport = XPORT_SAS; cts->transport_version = 0; sas->valid = CTS_SAS_VALID_SPEED; switch (targ->linkrate) { case 0x08: sas->bitrate = 150000; break; case 0x09: sas->bitrate = 300000; break; case 0x0a: sas->bitrate = 600000; break; case 0x0b: sas->bitrate = 1200000; break; default: sas->valid = 0; } cts->protocol = PROTO_SCSI; scsi->valid = CTS_SCSI_VALID_TQ; scsi->flags = CTS_SCSI_FLAGS_TAG_ENB; mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); break; } case XPT_CALC_GEOMETRY: cam_calc_geometry(&ccb->ccg, /*extended*/1); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); break; case XPT_RESET_DEV: mpr_dprint(sassc->sc, MPR_XINFO, "mprsas_action " "XPT_RESET_DEV\n"); mprsas_action_resetdev(sassc, ccb); return; case XPT_RESET_BUS: case XPT_ABORT: case XPT_TERM_IO: mpr_dprint(sassc->sc, MPR_XINFO, "mprsas_action faking success " "for abort or reset\n"); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); break; case XPT_SCSI_IO: mprsas_action_scsiio(sassc, ccb); return; #if __FreeBSD_version >= 900026 case XPT_SMP_IO: mprsas_action_smpio(sassc, ccb); return; #endif default: mprsas_set_ccbstatus(ccb, CAM_FUNC_NOTAVAIL); break; } xpt_done(ccb); } static void mprsas_announce_reset(struct mpr_softc *sc, uint32_t ac_code, target_id_t target_id, lun_id_t lun_id) { path_id_t path_id = cam_sim_path(sc->sassc->sim); struct cam_path *path; mpr_dprint(sc, MPR_XINFO, "%s code %x target %d lun %jx\n", __func__, ac_code, target_id, (uintmax_t)lun_id); if (xpt_create_path(&path, NULL, path_id, target_id, lun_id) != CAM_REQ_CMP) { mpr_dprint(sc, MPR_ERROR, "unable to create path for reset " "notification\n"); return; } xpt_async(ac_code, path, NULL); xpt_free_path(path); } static void mprsas_complete_all_commands(struct mpr_softc *sc) { struct mpr_command *cm; int i; int completed; MPR_FUNCTRACE(sc); mtx_assert(&sc->mpr_mtx, MA_OWNED); /* complete all commands with a NULL reply */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_state == MPR_CM_STATE_FREE) continue; cm->cm_state = MPR_CM_STATE_BUSY; cm->cm_reply = NULL; completed = 0; if (cm->cm_flags & MPR_CM_FLAGS_SATA_ID_TIMEOUT) { MPASS(cm->cm_data); free(cm->cm_data, M_MPR); cm->cm_data = NULL; } if (cm->cm_flags & MPR_CM_FLAGS_POLLED) cm->cm_flags |= MPR_CM_FLAGS_COMPLETE; if (cm->cm_complete != NULL) { mprsas_log_command(cm, MPR_RECOVERY, "completing cm %p state %x ccb %p for diag reset\n", cm, cm->cm_state, cm->cm_ccb); cm->cm_complete(sc, cm); completed = 1; } else if (cm->cm_flags & MPR_CM_FLAGS_WAKEUP) { mprsas_log_command(cm, MPR_RECOVERY, "waking up cm %p state %x ccb %p for diag reset\n", cm, cm->cm_state, cm->cm_ccb); wakeup(cm); completed = 1; } if ((completed == 0) && (cm->cm_state != MPR_CM_STATE_FREE)) { /* this should never happen, but if it does, log */ mprsas_log_command(cm, MPR_RECOVERY, "cm %p state %x flags 0x%x ccb %p during diag " "reset\n", cm, cm->cm_state, cm->cm_flags, cm->cm_ccb); } } sc->io_cmds_active = 0; } void mprsas_handle_reinit(struct mpr_softc *sc) { int i; /* Go back into startup mode and freeze the simq, so that CAM * doesn't send any commands until after we've rediscovered all * targets and found the proper device handles for them. * * After the reset, portenable will trigger discovery, and after all * discovery-related activities have finished, the simq will be * released. */ mpr_dprint(sc, MPR_INIT, "%s startup\n", __func__); sc->sassc->flags |= MPRSAS_IN_STARTUP; sc->sassc->flags |= MPRSAS_IN_DISCOVERY; mprsas_startup_increment(sc->sassc); /* notify CAM of a bus reset */ mprsas_announce_reset(sc, AC_BUS_RESET, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); /* complete and cleanup after all outstanding commands */ mprsas_complete_all_commands(sc); mpr_dprint(sc, MPR_INIT, "%s startup %u after command completion\n", __func__, sc->sassc->startup_refcount); /* zero all the target handles, since they may change after the * reset, and we have to rediscover all the targets and use the new * handles. */ for (i = 0; i < sc->sassc->maxtargets; i++) { if (sc->sassc->targets[i].outstanding != 0) mpr_dprint(sc, MPR_INIT, "target %u outstanding %u\n", i, sc->sassc->targets[i].outstanding); sc->sassc->targets[i].handle = 0x0; sc->sassc->targets[i].exp_dev_handle = 0x0; sc->sassc->targets[i].outstanding = 0; sc->sassc->targets[i].flags = MPRSAS_TARGET_INDIAGRESET; } } static void mprsas_tm_timeout(void *data) { struct mpr_command *tm = data; struct mpr_softc *sc = tm->cm_sc; mtx_assert(&sc->mpr_mtx, MA_OWNED); mprsas_log_command(tm, MPR_INFO|MPR_RECOVERY, "task mgmt %p timed " "out\n", tm); KASSERT(tm->cm_state == MPR_CM_STATE_INQUEUE, ("command not inqueue\n")); tm->cm_state = MPR_CM_STATE_BUSY; mpr_reinit(sc); } static void mprsas_logical_unit_reset_complete(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; unsigned int cm_count = 0; struct mpr_command *cm; struct mprsas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_RECOVERY|MPR_ERROR, "%s: cm_flags = %#x for LUN reset! " "This should not happen!\n", __func__, tm->cm_flags); mprsas_free_tm(sc, tm); return; } if (reply == NULL) { mpr_dprint(sc, MPR_RECOVERY, "NULL reset reply for tm %p\n", tm); if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing " "reset, ignoring NULL LUN reset reply\n"); targ->tm = NULL; mprsas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on " "LUN reset attempt, resetting controller\n"); mpr_reinit(sc); } return; } mpr_dprint(sc, MPR_RECOVERY, "logical unit reset status 0x%x code 0x%x count %u\n", le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); /* * See if there are any outstanding commands for this LUN. * This could be made more efficient by using a per-LU data * structure of some sort. */ TAILQ_FOREACH(cm, &targ->commands, cm_link) { if (cm->cm_lun == tm->cm_lun) cm_count++; } if (cm_count == 0) { mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "Finished recovery after LUN reset for target %u\n", targ->tid); mprsas_announce_reset(sc, AC_SENT_BDR, targ->tid, tm->cm_lun); /* * We've finished recovery for this logical unit. check and * see if some other logical unit has a timedout command * that needs to be processed. */ cm = TAILQ_FIRST(&targ->timedout_commands); if (cm) { mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "More commands to abort for target %u\n", targ->tid); mprsas_send_abort(sc, tm, cm); } else { targ->tm = NULL; mprsas_free_tm(sc, tm); } } else { /* if we still have commands for this LUN, the reset * effectively failed, regardless of the status reported. * Escalate to a target reset. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "logical unit reset complete for target %u, but still " "have %u command(s), sending target reset\n", targ->tid, cm_count); if (!targ->is_nvme || sc->custom_nvme_tm_handling) mprsas_send_reset(sc, tm, MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET); else mpr_reinit(sc); } } static void mprsas_target_reset_complete(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mprsas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for target " "reset! This should not happen!\n", __func__, tm->cm_flags); mprsas_free_tm(sc, tm); return; } if (reply == NULL) { mpr_dprint(sc, MPR_RECOVERY, "NULL target reset reply for tm %p TaskMID %u\n", tm, le16toh(req->TaskMID)); if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing " "reset, ignoring NULL target reset reply\n"); targ->tm = NULL; mprsas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on " "target reset attempt, resetting controller\n"); mpr_reinit(sc); } return; } mpr_dprint(sc, MPR_RECOVERY, "target reset status 0x%x code 0x%x count %u\n", le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); if (targ->outstanding == 0) { /* * We've finished recovery for this target and all * of its logical units. */ mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "Finished reset recovery for target %u\n", targ->tid); mprsas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid, CAM_LUN_WILDCARD); targ->tm = NULL; mprsas_free_tm(sc, tm); } else { /* * After a target reset, if this target still has * outstanding commands, the reset effectively failed, * regardless of the status reported. escalate. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "Target reset complete for target %u, but still have %u " "command(s), resetting controller\n", targ->tid, targ->outstanding); mpr_reinit(sc); } } #define MPR_RESET_TIMEOUT 30 int mprsas_send_reset(struct mpr_softc *sc, struct mpr_command *tm, uint8_t type) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mprsas_target *target; int err, timeout; target = tm->cm_targ; if (target->handle == 0) { mpr_dprint(sc, MPR_ERROR, "%s null devhandle for target_id " "%d\n", __func__, target->tid); return -1; } req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(target->handle); req->TaskType = type; if (!target->is_nvme || sc->custom_nvme_tm_handling) { timeout = MPR_RESET_TIMEOUT; /* * Target reset method = * SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; } else { timeout = (target->controller_reset_timeout) ? ( target->controller_reset_timeout) : (MPR_RESET_TIMEOUT); /* PCIe Protocol Level Reset*/ req->MsgFlags = MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE; } if (type == MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET) { /* XXX Need to handle invalid LUNs */ MPR_SET_LUN(req->LUN, tm->cm_lun); tm->cm_targ->logical_unit_resets++; mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "Sending logical unit reset to target %u lun %d\n", target->tid, tm->cm_lun); tm->cm_complete = mprsas_logical_unit_reset_complete; mprsas_prepare_for_tm(sc, tm, target, tm->cm_lun); } else if (type == MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET) { tm->cm_targ->target_resets++; mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "Sending target reset to target %u\n", target->tid); tm->cm_complete = mprsas_target_reset_complete; mprsas_prepare_for_tm(sc, tm, target, CAM_LUN_WILDCARD); } else { mpr_dprint(sc, MPR_ERROR, "unexpected reset type 0x%x\n", type); return -1; } if (target->encl_level_valid) { mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "At enclosure level %d, slot %d, connector name (%4s)\n", target->encl_level, target->encl_slot, target->connector_name); } tm->cm_data = NULL; tm->cm_complete_data = (void *)tm; callout_reset(&tm->cm_callout, timeout * hz, mprsas_tm_timeout, tm); err = mpr_map_command(sc, tm); if (err) mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY, "error %d sending reset type %u\n", err, type); return err; } static void mprsas_abort_complete(struct mpr_softc *sc, struct mpr_command *tm) { struct mpr_command *cm; MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mprsas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_RECOVERY|MPR_ERROR, "cm_flags = %#x for abort %p TaskMID %u!\n", tm->cm_flags, tm, le16toh(req->TaskMID)); mprsas_free_tm(sc, tm); return; } if (reply == NULL) { mpr_dprint(sc, MPR_RECOVERY, "NULL abort reply for tm %p TaskMID %u\n", tm, le16toh(req->TaskMID)); if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mpr_dprint(sc, MPR_RECOVERY, "Hardware undergoing " "reset, ignoring NULL abort reply\n"); targ->tm = NULL; mprsas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "NULL reply on " "abort attempt, resetting controller\n"); mpr_reinit(sc); } return; } mpr_dprint(sc, MPR_RECOVERY, "abort TaskMID %u status 0x%x code 0x%x count %u\n", le16toh(req->TaskMID), le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); cm = TAILQ_FIRST(&tm->cm_targ->timedout_commands); if (cm == NULL) { /* * if there are no more timedout commands, we're done with * error recovery for this target. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "Finished abort recovery for target %u\n", targ->tid); targ->tm = NULL; mprsas_free_tm(sc, tm); } else if (le16toh(req->TaskMID) != cm->cm_desc.Default.SMID) { /* abort success, but we have more timedout commands to abort */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "Continuing abort recovery for target %u\n", targ->tid); mprsas_send_abort(sc, tm, cm); } else { /* * we didn't get a command completion, so the abort * failed as far as we're concerned. escalate. */ mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "Abort failed for target %u, sending logical unit reset\n", targ->tid); mprsas_send_reset(sc, tm, MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET); } } #define MPR_ABORT_TIMEOUT 5 static int mprsas_send_abort(struct mpr_softc *sc, struct mpr_command *tm, struct mpr_command *cm) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mprsas_target *targ; int err, timeout; targ = cm->cm_targ; if (targ->handle == 0) { mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY, "%s null devhandle for target_id %d\n", __func__, cm->cm_ccb->ccb_h.target_id); return -1; } mprsas_log_command(cm, MPR_RECOVERY|MPR_INFO, "Aborting command %p\n", cm); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_ABORT_TASK; /* XXX Need to handle invalid LUNs */ MPR_SET_LUN(req->LUN, cm->cm_ccb->ccb_h.target_lun); req->TaskMID = htole16(cm->cm_desc.Default.SMID); tm->cm_data = NULL; tm->cm_complete = mprsas_abort_complete; tm->cm_complete_data = (void *)tm; tm->cm_targ = cm->cm_targ; tm->cm_lun = cm->cm_lun; if (!targ->is_nvme || sc->custom_nvme_tm_handling) timeout = MPR_ABORT_TIMEOUT; else timeout = sc->nvme_abort_timeout; callout_reset(&tm->cm_callout, timeout * hz, mprsas_tm_timeout, tm); targ->aborts++; mprsas_prepare_for_tm(sc, tm, targ, tm->cm_lun); err = mpr_map_command(sc, tm); if (err) mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY, "error %d sending abort for cm %p SMID %u\n", err, cm, req->TaskMID); return err; } static void mprsas_scsiio_timeout(void *data) { sbintime_t elapsed, now; union ccb *ccb; struct mpr_softc *sc; struct mpr_command *cm; struct mprsas_target *targ; cm = (struct mpr_command *)data; sc = cm->cm_sc; ccb = cm->cm_ccb; now = sbinuptime(); MPR_FUNCTRACE(sc); mtx_assert(&sc->mpr_mtx, MA_OWNED); mpr_dprint(sc, MPR_XINFO|MPR_RECOVERY, "Timeout checking cm %p\n", cm); /* * Run the interrupt handler to make sure it's not pending. This * isn't perfect because the command could have already completed * and been re-used, though this is unlikely. */ mpr_intr_locked(sc); - if (cm->cm_state != MPR_CM_STATE_INQUEUE) { + if (cm->cm_flags & MPR_CM_FLAGS_ON_RECOVERY) { mprsas_log_command(cm, MPR_XINFO, "SCSI command %p almost timed out\n", cm); return; } if (cm->cm_ccb == NULL) { mpr_dprint(sc, MPR_ERROR, "command timeout with NULL ccb\n"); return; } targ = cm->cm_targ; targ->timeouts++; elapsed = now - ccb->ccb_h.qos.sim_data; mprsas_log_command(cm, MPR_INFO|MPR_RECOVERY, "Command timeout on target %u(0x%04x), %d set, %d.%d elapsed\n", targ->tid, targ->handle, ccb->ccb_h.timeout, sbintime_getsec(elapsed), elapsed & 0xffffffff); if (targ->encl_level_valid) { mpr_dprint(sc, MPR_INFO|MPR_RECOVERY, "At enclosure level %d, slot %d, connector name (%4s)\n", targ->encl_level, targ->encl_slot, targ->connector_name); } /* XXX first, check the firmware state, to see if it's still * operational. if not, do a diag reset. */ mprsas_set_ccbstatus(cm->cm_ccb, CAM_CMD_TIMEOUT); - cm->cm_state = MPR_CM_STATE_TIMEDOUT; + cm->cm_flags |= MPR_CM_FLAGS_ON_RECOVERY | MPR_CM_FLAGS_TIMEDOUT; TAILQ_INSERT_TAIL(&targ->timedout_commands, cm, cm_recovery); if (targ->tm != NULL) { /* target already in recovery, just queue up another * timedout command to be processed later. */ mpr_dprint(sc, MPR_RECOVERY, "queued timedout cm %p for " "processing by tm %p\n", cm, targ->tm); } else if ((targ->tm = mprsas_alloc_tm(sc)) != NULL) { /* start recovery by aborting the first timedout command */ mpr_dprint(sc, MPR_RECOVERY|MPR_INFO, "Sending abort to target %u for SMID %d\n", targ->tid, cm->cm_desc.Default.SMID); mpr_dprint(sc, MPR_RECOVERY, "timedout cm %p allocated tm %p\n", cm, targ->tm); mprsas_send_abort(sc, targ->tm, cm); } else { /* XXX queue this target up for recovery once a TM becomes * available. The firmware only has a limited number of * HighPriority credits for the high priority requests used * for task management, and we ran out. * * Isilon: don't worry about this for now, since we have * more credits than disks in an enclosure, and limit * ourselves to one TM per target for recovery. */ mpr_dprint(sc, MPR_ERROR|MPR_RECOVERY, "timedout cm %p failed to allocate a tm\n", cm); } } /** * mprsas_build_nvme_unmap - Build Native NVMe DSM command equivalent * to SCSI Unmap. * Return 0 - for success, * 1 - to immediately return back the command with success status to CAM * negative value - to fallback to firmware path i.e. issue scsi unmap * to FW without any translation. */ static int mprsas_build_nvme_unmap(struct mpr_softc *sc, struct mpr_command *cm, union ccb *ccb, struct mprsas_target *targ) { Mpi26NVMeEncapsulatedRequest_t *req = NULL; struct ccb_scsiio *csio; struct unmap_parm_list *plist; struct nvme_dsm_range *nvme_dsm_ranges = NULL; struct nvme_command *c; int i, res; uint16_t ndesc, list_len, data_length; struct mpr_prp_page *prp_page_info; uint64_t nvme_dsm_ranges_dma_handle; csio = &ccb->csio; #if __FreeBSD_version >= 1100103 list_len = (scsiio_cdb_ptr(csio)[7] << 8 | scsiio_cdb_ptr(csio)[8]); #else if (csio->ccb_h.flags & CAM_CDB_POINTER) { list_len = (ccb->csio.cdb_io.cdb_ptr[7] << 8 | ccb->csio.cdb_io.cdb_ptr[8]); } else { list_len = (ccb->csio.cdb_io.cdb_bytes[7] << 8 | ccb->csio.cdb_io.cdb_bytes[8]); } #endif if (!list_len) { mpr_dprint(sc, MPR_ERROR, "Parameter list length is Zero\n"); return -EINVAL; } plist = malloc(csio->dxfer_len, M_MPR, M_ZERO|M_NOWAIT); if (!plist) { mpr_dprint(sc, MPR_ERROR, "Unable to allocate memory to " "save UNMAP data\n"); return -ENOMEM; } /* Copy SCSI unmap data to a local buffer */ bcopy(csio->data_ptr, plist, csio->dxfer_len); /* return back the unmap command to CAM with success status, * if number of descripts is zero. */ ndesc = be16toh(plist->unmap_blk_desc_data_len) >> 4; if (!ndesc) { mpr_dprint(sc, MPR_XINFO, "Number of descriptors in " "UNMAP cmd is Zero\n"); res = 1; goto out; } data_length = ndesc * sizeof(struct nvme_dsm_range); if (data_length > targ->MDTS) { mpr_dprint(sc, MPR_ERROR, "data length: %d is greater than " "Device's MDTS: %d\n", data_length, targ->MDTS); res = -EINVAL; goto out; } prp_page_info = mpr_alloc_prp_page(sc); KASSERT(prp_page_info != NULL, ("%s: There is no PRP Page for " "UNMAP command.\n", __func__)); /* * Insert the allocated PRP page into the command's PRP page list. This * will be freed when the command is freed. */ TAILQ_INSERT_TAIL(&cm->cm_prp_page_list, prp_page_info, prp_page_link); nvme_dsm_ranges = (struct nvme_dsm_range *)prp_page_info->prp_page; nvme_dsm_ranges_dma_handle = prp_page_info->prp_page_busaddr; bzero(nvme_dsm_ranges, data_length); /* Convert SCSI unmap's descriptor data to NVMe DSM specific Range data * for each descriptors contained in SCSI UNMAP data. */ for (i = 0; i < ndesc; i++) { nvme_dsm_ranges[i].length = htole32(be32toh(plist->desc[i].nlb)); nvme_dsm_ranges[i].starting_lba = htole64(be64toh(plist->desc[i].slba)); nvme_dsm_ranges[i].attributes = 0; } /* Build MPI2.6's NVMe Encapsulated Request Message */ req = (Mpi26NVMeEncapsulatedRequest_t *)cm->cm_req; bzero(req, sizeof(*req)); req->DevHandle = htole16(targ->handle); req->Function = MPI2_FUNCTION_NVME_ENCAPSULATED; req->Flags = MPI26_NVME_FLAGS_WRITE; req->ErrorResponseBaseAddress.High = htole32((uint32_t)((uint64_t)cm->cm_sense_busaddr >> 32)); req->ErrorResponseBaseAddress.Low = htole32(cm->cm_sense_busaddr); req->ErrorResponseAllocationLength = htole16(sizeof(struct nvme_completion)); req->EncapsulatedCommandLength = htole16(sizeof(struct nvme_command)); req->DataLength = htole32(data_length); /* Build NVMe DSM command */ c = (struct nvme_command *) req->NVMe_Command; c->opc = NVME_OPC_DATASET_MANAGEMENT; c->nsid = htole32(csio->ccb_h.target_lun + 1); c->cdw10 = htole32(ndesc - 1); c->cdw11 = htole32(NVME_DSM_ATTR_DEALLOCATE); cm->cm_length = data_length; cm->cm_data = NULL; cm->cm_complete = mprsas_scsiio_complete; cm->cm_complete_data = ccb; cm->cm_targ = targ; cm->cm_lun = csio->ccb_h.target_lun; cm->cm_ccb = ccb; cm->cm_desc.Default.RequestFlags = MPI26_REQ_DESCRIPT_FLAGS_PCIE_ENCAPSULATED; csio->ccb_h.qos.sim_data = sbinuptime(); #if __FreeBSD_version >= 1000029 callout_reset_sbt(&cm->cm_callout, SBT_1MS * ccb->ccb_h.timeout, 0, mprsas_scsiio_timeout, cm, 0); #else //__FreeBSD_version < 1000029 callout_reset(&cm->cm_callout, (ccb->ccb_h.timeout * hz) / 1000, mprsas_scsiio_timeout, cm); #endif //__FreeBSD_version >= 1000029 targ->issued++; targ->outstanding++; TAILQ_INSERT_TAIL(&targ->commands, cm, cm_link); ccb->ccb_h.status |= CAM_SIM_QUEUED; mprsas_log_command(cm, MPR_XINFO, "%s cm %p ccb %p outstanding %u\n", __func__, cm, ccb, targ->outstanding); mpr_build_nvme_prp(sc, cm, req, (void *)(uintptr_t)nvme_dsm_ranges_dma_handle, 0, data_length); mpr_map_command(sc, cm); out: free(plist, M_MPR); return 0; } static void mprsas_action_scsiio(struct mprsas_softc *sassc, union ccb *ccb) { MPI2_SCSI_IO_REQUEST *req; struct ccb_scsiio *csio; struct mpr_softc *sc; struct mprsas_target *targ; struct mprsas_lun *lun; struct mpr_command *cm; uint8_t i, lba_byte, *ref_tag_addr, scsi_opcode; uint16_t eedp_flags; uint32_t mpi_control; int rc; sc = sassc->sc; MPR_FUNCTRACE(sc); mtx_assert(&sc->mpr_mtx, MA_OWNED); csio = &ccb->csio; KASSERT(csio->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_SCSI_IO\n", csio->ccb_h.target_id)); targ = &sassc->targets[csio->ccb_h.target_id]; mpr_dprint(sc, MPR_TRACE, "ccb %p target flag %x\n", ccb, targ->flags); if (targ->handle == 0x0) { mpr_dprint(sc, MPR_ERROR, "%s NULL handle for target %u\n", __func__, csio->ccb_h.target_id); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } if (targ->flags & MPR_TARGET_FLAGS_RAID_COMPONENT) { mpr_dprint(sc, MPR_ERROR, "%s Raid component no SCSI IO " "supported %u\n", __func__, csio->ccb_h.target_id); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } /* * Sometimes, it is possible to get a command that is not "In * Progress" and was actually aborted by the upper layer. Check for * this here and complete the command without error. */ if (mprsas_get_ccbstatus(ccb) != CAM_REQ_INPROG) { mpr_dprint(sc, MPR_TRACE, "%s Command is not in progress for " "target %u\n", __func__, csio->ccb_h.target_id); xpt_done(ccb); return; } /* * If devinfo is 0 this will be a volume. In that case don't tell CAM * that the volume has timed out. We want volumes to be enumerated * until they are deleted/removed, not just failed. */ if (targ->flags & MPRSAS_TARGET_INREMOVAL) { if (targ->devinfo == 0) mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); else mprsas_set_ccbstatus(ccb, CAM_SEL_TIMEOUT); xpt_done(ccb); return; } if ((sc->mpr_flags & MPR_FLAGS_SHUTDOWN) != 0) { mpr_dprint(sc, MPR_INFO, "%s shutting down\n", __func__); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } /* * If target has a reset in progress, freeze the devq and return. The * devq will be released when the TM reset is finished. */ if (targ->flags & MPRSAS_TARGET_INRESET) { ccb->ccb_h.status = CAM_BUSY | CAM_DEV_QFRZN; mpr_dprint(sc, MPR_INFO, "%s: Freezing devq for target ID %d\n", __func__, targ->tid); xpt_freeze_devq(ccb->ccb_h.path, 1); xpt_done(ccb); return; } cm = mpr_alloc_command(sc); if (cm == NULL || (sc->mpr_flags & MPR_FLAGS_DIAGRESET)) { if (cm != NULL) { mpr_free_command(sc, cm); } if ((sassc->flags & MPRSAS_QUEUE_FROZEN) == 0) { xpt_freeze_simq(sassc->sim, 1); sassc->flags |= MPRSAS_QUEUE_FROZEN; } ccb->ccb_h.status &= ~CAM_SIM_QUEUED; ccb->ccb_h.status |= CAM_REQUEUE_REQ; xpt_done(ccb); return; } /* For NVME device's issue UNMAP command directly to NVME drives by * constructing equivalent native NVMe DataSetManagement command. */ #if __FreeBSD_version >= 1100103 scsi_opcode = scsiio_cdb_ptr(csio)[0]; #else if (csio->ccb_h.flags & CAM_CDB_POINTER) scsi_opcode = csio->cdb_io.cdb_ptr[0]; else scsi_opcode = csio->cdb_io.cdb_bytes[0]; #endif if (scsi_opcode == UNMAP && targ->is_nvme && (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) { rc = mprsas_build_nvme_unmap(sc, cm, ccb, targ); if (rc == 1) { /* return command to CAM with success status */ mpr_free_command(sc, cm); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); xpt_done(ccb); return; } else if (!rc) /* Issued NVMe Encapsulated Request Message */ return; } req = (MPI2_SCSI_IO_REQUEST *)cm->cm_req; bzero(req, sizeof(*req)); req->DevHandle = htole16(targ->handle); req->Function = MPI2_FUNCTION_SCSI_IO_REQUEST; req->MsgFlags = 0; req->SenseBufferLowAddress = htole32(cm->cm_sense_busaddr); req->SenseBufferLength = MPR_SENSE_LEN; req->SGLFlags = 0; req->ChainOffset = 0; req->SGLOffset0 = 24; /* 32bit word offset to the SGL */ req->SGLOffset1= 0; req->SGLOffset2= 0; req->SGLOffset3= 0; req->SkipCount = 0; req->DataLength = htole32(csio->dxfer_len); req->BidirectionalDataLength = 0; req->IoFlags = htole16(csio->cdb_len); req->EEDPFlags = 0; /* Note: BiDirectional transfers are not supported */ switch (csio->ccb_h.flags & CAM_DIR_MASK) { case CAM_DIR_IN: mpi_control = MPI2_SCSIIO_CONTROL_READ; cm->cm_flags |= MPR_CM_FLAGS_DATAIN; break; case CAM_DIR_OUT: mpi_control = MPI2_SCSIIO_CONTROL_WRITE; cm->cm_flags |= MPR_CM_FLAGS_DATAOUT; break; case CAM_DIR_NONE: default: mpi_control = MPI2_SCSIIO_CONTROL_NODATATRANSFER; break; } if (csio->cdb_len == 32) mpi_control |= 4 << MPI2_SCSIIO_CONTROL_ADDCDBLEN_SHIFT; /* * It looks like the hardware doesn't require an explicit tag * number for each transaction. SAM Task Management not supported * at the moment. */ switch (csio->tag_action) { case MSG_HEAD_OF_Q_TAG: mpi_control |= MPI2_SCSIIO_CONTROL_HEADOFQ; break; case MSG_ORDERED_Q_TAG: mpi_control |= MPI2_SCSIIO_CONTROL_ORDEREDQ; break; case MSG_ACA_TASK: mpi_control |= MPI2_SCSIIO_CONTROL_ACAQ; break; case CAM_TAG_ACTION_NONE: case MSG_SIMPLE_Q_TAG: default: mpi_control |= MPI2_SCSIIO_CONTROL_SIMPLEQ; break; } mpi_control |= sc->mapping_table[csio->ccb_h.target_id].TLR_bits; req->Control = htole32(mpi_control); if (MPR_SET_LUN(req->LUN, csio->ccb_h.target_lun) != 0) { mpr_free_command(sc, cm); mprsas_set_ccbstatus(ccb, CAM_LUN_INVALID); xpt_done(ccb); return; } if (csio->ccb_h.flags & CAM_CDB_POINTER) bcopy(csio->cdb_io.cdb_ptr, &req->CDB.CDB32[0], csio->cdb_len); else { KASSERT(csio->cdb_len <= IOCDBLEN, ("cdb_len %d is greater than IOCDBLEN but CAM_CDB_POINTER " "is not set", csio->cdb_len)); bcopy(csio->cdb_io.cdb_bytes, &req->CDB.CDB32[0],csio->cdb_len); } req->IoFlags = htole16(csio->cdb_len); /* * Check if EEDP is supported and enabled. If it is then check if the * SCSI opcode could be using EEDP. If so, make sure the LUN exists and * is formatted for EEDP support. If all of this is true, set CDB up * for EEDP transfer. */ eedp_flags = op_code_prot[req->CDB.CDB32[0]]; if (sc->eedp_enabled && eedp_flags) { SLIST_FOREACH(lun, &targ->luns, lun_link) { if (lun->lun_id == csio->ccb_h.target_lun) { break; } } if ((lun != NULL) && (lun->eedp_formatted)) { req->EEDPBlockSize = htole16(lun->eedp_block_size); eedp_flags |= (MPI2_SCSIIO_EEDPFLAGS_INC_PRI_REFTAG | MPI2_SCSIIO_EEDPFLAGS_CHECK_REFTAG | MPI2_SCSIIO_EEDPFLAGS_CHECK_GUARD); if (sc->mpr_flags & MPR_FLAGS_GEN35_IOC) { eedp_flags |= MPI25_SCSIIO_EEDPFLAGS_APPTAG_DISABLE_MODE; } req->EEDPFlags = htole16(eedp_flags); /* * If CDB less than 32, fill in Primary Ref Tag with * low 4 bytes of LBA. If CDB is 32, tag stuff is * already there. Also, set protection bit. FreeBSD * currently does not support CDBs bigger than 16, but * the code doesn't hurt, and will be here for the * future. */ if (csio->cdb_len != 32) { lba_byte = (csio->cdb_len == 16) ? 6 : 2; ref_tag_addr = (uint8_t *)&req->CDB.EEDP32. PrimaryReferenceTag; for (i = 0; i < 4; i++) { *ref_tag_addr = req->CDB.CDB32[lba_byte + i]; ref_tag_addr++; } req->CDB.EEDP32.PrimaryReferenceTag = htole32(req-> CDB.EEDP32.PrimaryReferenceTag); req->CDB.EEDP32.PrimaryApplicationTagMask = 0xFFFF; req->CDB.CDB32[1] = (req->CDB.CDB32[1] & 0x1F) | 0x20; } else { eedp_flags |= MPI2_SCSIIO_EEDPFLAGS_INC_PRI_APPTAG; req->EEDPFlags = htole16(eedp_flags); req->CDB.CDB32[10] = (req->CDB.CDB32[10] & 0x1F) | 0x20; } } } cm->cm_length = csio->dxfer_len; if (cm->cm_length != 0) { cm->cm_data = ccb; cm->cm_flags |= MPR_CM_FLAGS_USE_CCB; } else { cm->cm_data = NULL; } cm->cm_sge = &req->SGL; cm->cm_sglsize = (32 - 24) * 4; cm->cm_complete = mprsas_scsiio_complete; cm->cm_complete_data = ccb; cm->cm_targ = targ; cm->cm_lun = csio->ccb_h.target_lun; cm->cm_ccb = ccb; /* * If using FP desc type, need to set a bit in IoFlags (SCSI IO is 0) * and set descriptor type. */ if (targ->scsi_req_desc_type == MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO) { req->IoFlags |= MPI25_SCSIIO_IOFLAGS_FAST_PATH; cm->cm_desc.FastPathSCSIIO.RequestFlags = MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO; if (!sc->atomic_desc_capable) { cm->cm_desc.FastPathSCSIIO.DevHandle = htole16(targ->handle); } } else { cm->cm_desc.SCSIIO.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO; if (!sc->atomic_desc_capable) cm->cm_desc.SCSIIO.DevHandle = htole16(targ->handle); } csio->ccb_h.qos.sim_data = sbinuptime(); #if __FreeBSD_version >= 1000029 callout_reset_sbt(&cm->cm_callout, SBT_1MS * ccb->ccb_h.timeout, 0, mprsas_scsiio_timeout, cm, 0); #else //__FreeBSD_version < 1000029 callout_reset(&cm->cm_callout, (ccb->ccb_h.timeout * hz) / 1000, mprsas_scsiio_timeout, cm); #endif //__FreeBSD_version >= 1000029 targ->issued++; targ->outstanding++; TAILQ_INSERT_TAIL(&targ->commands, cm, cm_link); ccb->ccb_h.status |= CAM_SIM_QUEUED; mprsas_log_command(cm, MPR_XINFO, "%s cm %p ccb %p outstanding %u\n", __func__, cm, ccb, targ->outstanding); mpr_map_command(sc, cm); return; } /** * mpr_sc_failed_io_info - translated non-succesfull SCSI_IO request */ static void mpr_sc_failed_io_info(struct mpr_softc *sc, struct ccb_scsiio *csio, Mpi2SCSIIOReply_t *mpi_reply, struct mprsas_target *targ) { u32 response_info; u8 *response_bytes; u16 ioc_status = le16toh(mpi_reply->IOCStatus) & MPI2_IOCSTATUS_MASK; u8 scsi_state = mpi_reply->SCSIState; u8 scsi_status = mpi_reply->SCSIStatus; char *desc_ioc_state = NULL; char *desc_scsi_status = NULL; u32 log_info = le32toh(mpi_reply->IOCLogInfo); if (log_info == 0x31170000) return; desc_ioc_state = mpr_describe_table(mpr_iocstatus_string, ioc_status); desc_scsi_status = mpr_describe_table(mpr_scsi_status_string, scsi_status); mpr_dprint(sc, MPR_XINFO, "\thandle(0x%04x), ioc_status(%s)(0x%04x)\n", le16toh(mpi_reply->DevHandle), desc_ioc_state, ioc_status); if (targ->encl_level_valid) { mpr_dprint(sc, MPR_XINFO, "At enclosure level %d, slot %d, " "connector name (%4s)\n", targ->encl_level, targ->encl_slot, targ->connector_name); } /* * We can add more detail about underflow data here * TO-DO */ mpr_dprint(sc, MPR_XINFO, "\tscsi_status(%s)(0x%02x), " "scsi_state %b\n", desc_scsi_status, scsi_status, scsi_state, "\20" "\1AutosenseValid" "\2AutosenseFailed" "\3NoScsiStatus" "\4Terminated" "\5Response InfoValid"); if (sc->mpr_debug & MPR_XINFO && scsi_state & MPI2_SCSI_STATE_AUTOSENSE_VALID) { mpr_dprint(sc, MPR_XINFO, "-> Sense Buffer Data : Start :\n"); scsi_sense_print(csio); mpr_dprint(sc, MPR_XINFO, "-> Sense Buffer Data : End :\n"); } if (scsi_state & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) { response_info = le32toh(mpi_reply->ResponseInfo); response_bytes = (u8 *)&response_info; mpr_dprint(sc, MPR_XINFO, "response code(0x%01x): %s\n", response_bytes[0], mpr_describe_table(mpr_scsi_taskmgmt_string, response_bytes[0])); } } /** mprsas_nvme_trans_status_code * * Convert Native NVMe command error status to * equivalent SCSI error status. * * Returns appropriate scsi_status */ static u8 mprsas_nvme_trans_status_code(uint16_t nvme_status, struct mpr_command *cm) { u8 status = MPI2_SCSI_STATUS_GOOD; int skey, asc, ascq; union ccb *ccb = cm->cm_complete_data; int returned_sense_len; uint8_t sct, sc; sct = NVME_STATUS_GET_SCT(nvme_status); sc = NVME_STATUS_GET_SC(nvme_status); status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_NO_SENSE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; switch (sct) { case NVME_SCT_GENERIC: switch (sc) { case NVME_SC_SUCCESS: status = MPI2_SCSI_STATUS_GOOD; skey = SSD_KEY_NO_SENSE; asc = SCSI_ASC_NO_SENSE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_INVALID_OPCODE: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_ILLEGAL_COMMAND; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_INVALID_FIELD: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_INVALID_CDB; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_DATA_TRANSFER_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_NO_SENSE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_ABORTED_POWER_LOSS: status = MPI2_SCSI_STATUS_TASK_ABORTED; skey = SSD_KEY_ABORTED_COMMAND; asc = SCSI_ASC_WARNING; ascq = SCSI_ASCQ_POWER_LOSS_EXPECTED; break; case NVME_SC_INTERNAL_DEVICE_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_HARDWARE_ERROR; asc = SCSI_ASC_INTERNAL_TARGET_FAILURE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_ABORTED_BY_REQUEST: case NVME_SC_ABORTED_SQ_DELETION: case NVME_SC_ABORTED_FAILED_FUSED: case NVME_SC_ABORTED_MISSING_FUSED: status = MPI2_SCSI_STATUS_TASK_ABORTED; skey = SSD_KEY_ABORTED_COMMAND; asc = SCSI_ASC_NO_SENSE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_INVALID_NAMESPACE_OR_FORMAT: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_ACCESS_DENIED_INVALID_LUN_ID; ascq = SCSI_ASCQ_INVALID_LUN_ID; break; case NVME_SC_LBA_OUT_OF_RANGE: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_ILLEGAL_BLOCK; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_CAPACITY_EXCEEDED: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_NO_SENSE; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_NAMESPACE_NOT_READY: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_NOT_READY; asc = SCSI_ASC_LUN_NOT_READY; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; } break; case NVME_SCT_COMMAND_SPECIFIC: switch (sc) { case NVME_SC_INVALID_FORMAT: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_FORMAT_COMMAND_FAILED; ascq = SCSI_ASCQ_FORMAT_COMMAND_FAILED; break; case NVME_SC_CONFLICTING_ATTRIBUTES: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_INVALID_CDB; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; } break; case NVME_SCT_MEDIA_ERROR: switch (sc) { case NVME_SC_WRITE_FAULTS: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_PERIPHERAL_DEV_WRITE_FAULT; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_UNRECOVERED_READ_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_UNRECOVERED_READ_ERROR; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_GUARD_CHECK_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_LOG_BLOCK_GUARD_CHECK_FAILED; ascq = SCSI_ASCQ_LOG_BLOCK_GUARD_CHECK_FAILED; break; case NVME_SC_APPLICATION_TAG_CHECK_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_LOG_BLOCK_APPTAG_CHECK_FAILED; ascq = SCSI_ASCQ_LOG_BLOCK_APPTAG_CHECK_FAILED; break; case NVME_SC_REFERENCE_TAG_CHECK_ERROR: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MEDIUM_ERROR; asc = SCSI_ASC_LOG_BLOCK_REFTAG_CHECK_FAILED; ascq = SCSI_ASCQ_LOG_BLOCK_REFTAG_CHECK_FAILED; break; case NVME_SC_COMPARE_FAILURE: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_MISCOMPARE; asc = SCSI_ASC_MISCOMPARE_DURING_VERIFY; ascq = SCSI_ASCQ_CAUSE_NOT_REPORTABLE; break; case NVME_SC_ACCESS_DENIED: status = MPI2_SCSI_STATUS_CHECK_CONDITION; skey = SSD_KEY_ILLEGAL_REQUEST; asc = SCSI_ASC_ACCESS_DENIED_INVALID_LUN_ID; ascq = SCSI_ASCQ_INVALID_LUN_ID; break; } break; } returned_sense_len = sizeof(struct scsi_sense_data); if (returned_sense_len < ccb->csio.sense_len) ccb->csio.sense_resid = ccb->csio.sense_len - returned_sense_len; else ccb->csio.sense_resid = 0; scsi_set_sense_data(&ccb->csio.sense_data, SSD_TYPE_FIXED, 1, skey, asc, ascq, SSD_ELEM_NONE); ccb->ccb_h.status |= CAM_AUTOSNS_VALID; return status; } /** mprsas_complete_nvme_unmap * * Complete native NVMe command issued using NVMe Encapsulated * Request Message. */ static u8 mprsas_complete_nvme_unmap(struct mpr_softc *sc, struct mpr_command *cm) { Mpi26NVMeEncapsulatedErrorReply_t *mpi_reply; struct nvme_completion *nvme_completion = NULL; u8 scsi_status = MPI2_SCSI_STATUS_GOOD; mpi_reply =(Mpi26NVMeEncapsulatedErrorReply_t *)cm->cm_reply; if (le16toh(mpi_reply->ErrorResponseCount)){ nvme_completion = (struct nvme_completion *)cm->cm_sense; scsi_status = mprsas_nvme_trans_status_code( nvme_completion->status, cm); } return scsi_status; } static void mprsas_scsiio_complete(struct mpr_softc *sc, struct mpr_command *cm) { MPI2_SCSI_IO_REPLY *rep; union ccb *ccb; struct ccb_scsiio *csio; struct mprsas_softc *sassc; struct scsi_vpd_supported_page_list *vpd_list = NULL; u8 *TLR_bits, TLR_on, *scsi_cdb; int dir = 0, i; u16 alloc_len; struct mprsas_target *target; target_id_t target_id; MPR_FUNCTRACE(sc); mpr_dprint(sc, MPR_TRACE, "cm %p SMID %u ccb %p reply %p outstanding %u\n", cm, cm->cm_desc.Default.SMID, cm->cm_ccb, cm->cm_reply, cm->cm_targ->outstanding); callout_stop(&cm->cm_callout); mtx_assert(&sc->mpr_mtx, MA_OWNED); sassc = sc->sassc; ccb = cm->cm_complete_data; csio = &ccb->csio; target_id = csio->ccb_h.target_id; rep = (MPI2_SCSI_IO_REPLY *)cm->cm_reply; /* * XXX KDM if the chain allocation fails, does it matter if we do * the sync and unload here? It is simpler to do it in every case, * assuming it doesn't cause problems. */ if (cm->cm_data != NULL) { if (cm->cm_flags & MPR_CM_FLAGS_DATAIN) dir = BUS_DMASYNC_POSTREAD; else if (cm->cm_flags & MPR_CM_FLAGS_DATAOUT) dir = BUS_DMASYNC_POSTWRITE; bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); } cm->cm_targ->completed++; cm->cm_targ->outstanding--; TAILQ_REMOVE(&cm->cm_targ->commands, cm, cm_link); ccb->ccb_h.status &= ~(CAM_STATUS_MASK | CAM_SIM_QUEUED); - if (cm->cm_state == MPR_CM_STATE_TIMEDOUT) { + if (cm->cm_flags & MPR_CM_FLAGS_ON_RECOVERY) { TAILQ_REMOVE(&cm->cm_targ->timedout_commands, cm, cm_recovery); - cm->cm_state = MPR_CM_STATE_BUSY; + KASSERT(cm->cm_state == MPR_CM_STATE_BUSY, + ("Not busy for CM_FLAGS_TIMEDOUT: %d\n", cm->cm_state)); + cm->cm_flags &= ~MPR_CM_FLAGS_ON_RECOVERY; if (cm->cm_reply != NULL) mprsas_log_command(cm, MPR_RECOVERY, "completed timedout cm %p ccb %p during recovery " "ioc %x scsi %x state %x xfer %u\n", cm, cm->cm_ccb, le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); else mprsas_log_command(cm, MPR_RECOVERY, "completed timedout cm %p ccb %p during recovery\n", cm, cm->cm_ccb); } else if (cm->cm_targ->tm != NULL) { if (cm->cm_reply != NULL) mprsas_log_command(cm, MPR_RECOVERY, "completed cm %p ccb %p during recovery " "ioc %x scsi %x state %x xfer %u\n", cm, cm->cm_ccb, le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); else mprsas_log_command(cm, MPR_RECOVERY, "completed cm %p ccb %p during recovery\n", cm, cm->cm_ccb); } else if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) { mprsas_log_command(cm, MPR_RECOVERY, "reset completed cm %p ccb %p\n", cm, cm->cm_ccb); } if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { /* * We ran into an error after we tried to map the command, * so we're getting a callback without queueing the command * to the hardware. So we set the status here, and it will * be retained below. We'll go through the "fast path", * because there can be no reply when we haven't actually * gone out to the hardware. */ mprsas_set_ccbstatus(ccb, CAM_REQUEUE_REQ); /* * Currently the only error included in the mask is * MPR_CM_FLAGS_CHAIN_FAILED, which means we're out of * chain frames. We need to freeze the queue until we get * a command that completed without this error, which will * hopefully have some chain frames attached that we can * use. If we wanted to get smarter about it, we would * only unfreeze the queue in this condition when we're * sure that we're getting some chain frames back. That's * probably unnecessary. */ if ((sassc->flags & MPRSAS_QUEUE_FROZEN) == 0) { xpt_freeze_simq(sassc->sim, 1); sassc->flags |= MPRSAS_QUEUE_FROZEN; mpr_dprint(sc, MPR_XINFO, "Error sending command, " "freezing SIM queue\n"); } } /* * Point to the SCSI CDB, which is dependent on the CAM_CDB_POINTER * flag, and use it in a few places in the rest of this function for * convenience. Use the macro if available. */ #if __FreeBSD_version >= 1100103 scsi_cdb = scsiio_cdb_ptr(csio); #else if (csio->ccb_h.flags & CAM_CDB_POINTER) scsi_cdb = csio->cdb_io.cdb_ptr; else scsi_cdb = csio->cdb_io.cdb_bytes; #endif /* * If this is a Start Stop Unit command and it was issued by the driver * during shutdown, decrement the refcount to account for all of the * commands that were sent. All SSU commands should be completed before * shutdown completes, meaning SSU_refcount will be 0 after SSU_started * is TRUE. */ if (sc->SSU_started && (scsi_cdb[0] == START_STOP_UNIT)) { mpr_dprint(sc, MPR_INFO, "Decrementing SSU count.\n"); sc->SSU_refcount--; } /* Take the fast path to completion */ if (cm->cm_reply == NULL) { if (mprsas_get_ccbstatus(ccb) == CAM_REQ_INPROG) { if ((sc->mpr_flags & MPR_FLAGS_DIAGRESET) != 0) mprsas_set_ccbstatus(ccb, CAM_SCSI_BUS_RESET); else { mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); csio->scsi_status = SCSI_STATUS_OK; } if (sassc->flags & MPRSAS_QUEUE_FROZEN) { ccb->ccb_h.status |= CAM_RELEASE_SIMQ; sassc->flags &= ~MPRSAS_QUEUE_FROZEN; mpr_dprint(sc, MPR_XINFO, "Unfreezing SIM queue\n"); } } /* * There are two scenarios where the status won't be * CAM_REQ_CMP. The first is if MPR_CM_FLAGS_ERROR_MASK is * set, the second is in the MPR_FLAGS_DIAGRESET above. */ if (mprsas_get_ccbstatus(ccb) != CAM_REQ_CMP) { /* * Freeze the dev queue so that commands are * executed in the correct order after error * recovery. */ ccb->ccb_h.status |= CAM_DEV_QFRZN; xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1); } mpr_free_command(sc, cm); xpt_done(ccb); return; } target = &sassc->targets[target_id]; if (scsi_cdb[0] == UNMAP && target->is_nvme && (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) { rep->SCSIStatus = mprsas_complete_nvme_unmap(sc, cm); csio->scsi_status = rep->SCSIStatus; } mprsas_log_command(cm, MPR_XINFO, "ioc %x scsi %x state %x xfer %u\n", le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); switch (le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) { case MPI2_IOCSTATUS_SCSI_DATA_UNDERRUN: csio->resid = cm->cm_length - le32toh(rep->TransferCount); /* FALLTHROUGH */ case MPI2_IOCSTATUS_SUCCESS: case MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR: if ((le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR) mprsas_log_command(cm, MPR_XINFO, "recovered error\n"); /* Completion failed at the transport level. */ if (rep->SCSIState & (MPI2_SCSI_STATE_NO_SCSI_STATUS | MPI2_SCSI_STATE_TERMINATED)) { mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); break; } /* In a modern packetized environment, an autosense failure * implies that there's not much else that can be done to * recover the command. */ if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_FAILED) { mprsas_set_ccbstatus(ccb, CAM_AUTOSENSE_FAIL); break; } /* * CAM doesn't care about SAS Response Info data, but if this is * the state check if TLR should be done. If not, clear the * TLR_bits for the target. */ if ((rep->SCSIState & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) && ((le32toh(rep->ResponseInfo) & MPI2_SCSI_RI_MASK_REASONCODE) == MPR_SCSI_RI_INVALID_FRAME)) { sc->mapping_table[target_id].TLR_bits = (u8)MPI2_SCSIIO_CONTROL_NO_TLR; } /* * Intentionally override the normal SCSI status reporting * for these two cases. These are likely to happen in a * multi-initiator environment, and we want to make sure that * CAM retries these commands rather than fail them. */ if ((rep->SCSIStatus == MPI2_SCSI_STATUS_COMMAND_TERMINATED) || (rep->SCSIStatus == MPI2_SCSI_STATUS_TASK_ABORTED)) { mprsas_set_ccbstatus(ccb, CAM_REQ_ABORTED); break; } /* Handle normal status and sense */ csio->scsi_status = rep->SCSIStatus; if (rep->SCSIStatus == MPI2_SCSI_STATUS_GOOD) mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); else mprsas_set_ccbstatus(ccb, CAM_SCSI_STATUS_ERROR); if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_VALID) { int sense_len, returned_sense_len; returned_sense_len = min(le32toh(rep->SenseCount), sizeof(struct scsi_sense_data)); if (returned_sense_len < csio->sense_len) csio->sense_resid = csio->sense_len - returned_sense_len; else csio->sense_resid = 0; sense_len = min(returned_sense_len, csio->sense_len - csio->sense_resid); bzero(&csio->sense_data, sizeof(csio->sense_data)); bcopy(cm->cm_sense, &csio->sense_data, sense_len); ccb->ccb_h.status |= CAM_AUTOSNS_VALID; } /* * Check if this is an INQUIRY command. If it's a VPD inquiry, * and it's page code 0 (Supported Page List), and there is * inquiry data, and this is for a sequential access device, and * the device is an SSP target, and TLR is supported by the * controller, turn the TLR_bits value ON if page 0x90 is * supported. */ if ((scsi_cdb[0] == INQUIRY) && (scsi_cdb[1] & SI_EVPD) && (scsi_cdb[2] == SVPD_SUPPORTED_PAGE_LIST) && ((csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) && (csio->data_ptr != NULL) && ((csio->data_ptr[0] & 0x1f) == T_SEQUENTIAL) && (sc->control_TLR) && (sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_SSP_TARGET)) { vpd_list = (struct scsi_vpd_supported_page_list *) csio->data_ptr; TLR_bits = &sc->mapping_table[target_id].TLR_bits; *TLR_bits = (u8)MPI2_SCSIIO_CONTROL_NO_TLR; TLR_on = (u8)MPI2_SCSIIO_CONTROL_TLR_ON; alloc_len = ((u16)scsi_cdb[3] << 8) + scsi_cdb[4]; alloc_len -= csio->resid; for (i = 0; i < MIN(vpd_list->length, alloc_len); i++) { if (vpd_list->list[i] == 0x90) { *TLR_bits = TLR_on; break; } } } /* * If this is a SATA direct-access end device, mark it so that * a SCSI StartStopUnit command will be sent to it when the * driver is being shutdown. */ if ((scsi_cdb[0] == INQUIRY) && (csio->data_ptr != NULL) && ((csio->data_ptr[0] & 0x1f) == T_DIRECT) && (sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_SATA_DEVICE) && ((sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_MASK_DEVICE_TYPE) == MPI2_SAS_DEVICE_INFO_END_DEVICE)) { target = &sassc->targets[target_id]; target->supports_SSU = TRUE; mpr_dprint(sc, MPR_XINFO, "Target %d supports SSU\n", target_id); } break; case MPI2_IOCSTATUS_SCSI_INVALID_DEVHANDLE: case MPI2_IOCSTATUS_SCSI_DEVICE_NOT_THERE: /* * If devinfo is 0 this will be a volume. In that case don't * tell CAM that the volume is not there. We want volumes to * be enumerated until they are deleted/removed, not just * failed. */ if (cm->cm_targ->devinfo == 0) mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); else mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); break; case MPI2_IOCSTATUS_INVALID_SGL: mpr_print_scsiio_cmd(sc, cm); mprsas_set_ccbstatus(ccb, CAM_UNREC_HBA_ERROR); break; case MPI2_IOCSTATUS_SCSI_TASK_TERMINATED: /* * This is one of the responses that comes back when an I/O * has been aborted. If it is because of a timeout that we * initiated, just set the status to CAM_CMD_TIMEOUT. * Otherwise set it to CAM_REQ_ABORTED. The effect on the * command is the same (it gets retried, subject to the * retry counter), the only difference is what gets printed * on the console. */ - if (cm->cm_state == MPR_CM_STATE_TIMEDOUT) + if (cm->cm_flags & MPR_CM_FLAGS_TIMEDOUT) mprsas_set_ccbstatus(ccb, CAM_CMD_TIMEOUT); else mprsas_set_ccbstatus(ccb, CAM_REQ_ABORTED); break; case MPI2_IOCSTATUS_SCSI_DATA_OVERRUN: /* resid is ignored for this condition */ csio->resid = 0; mprsas_set_ccbstatus(ccb, CAM_DATA_RUN_ERR); break; case MPI2_IOCSTATUS_SCSI_IOC_TERMINATED: case MPI2_IOCSTATUS_SCSI_EXT_TERMINATED: /* * These can sometimes be transient transport-related * errors, and sometimes persistent drive-related errors. * We used to retry these without decrementing the retry * count by returning CAM_REQUEUE_REQ. Unfortunately, if * we hit a persistent drive problem that returns one of * these error codes, we would retry indefinitely. So, * return CAM_REQ_CMP_ERROR so that we decrement the retry * count and avoid infinite retries. We're taking the * potential risk of flagging false failures in the event * of a topology-related error (e.g. a SAS expander problem * causes a command addressed to a drive to fail), but * avoiding getting into an infinite retry loop. */ mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); mpr_dprint(sc, MPR_INFO, "Controller reported %s tgt %u SMID %u loginfo %x\n", mpr_describe_table(mpr_iocstatus_string, le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK), target_id, cm->cm_desc.Default.SMID, le32toh(rep->IOCLogInfo)); mpr_dprint(sc, MPR_XINFO, "SCSIStatus %x SCSIState %x xfercount %u\n", rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); break; case MPI2_IOCSTATUS_INVALID_FUNCTION: case MPI2_IOCSTATUS_INTERNAL_ERROR: case MPI2_IOCSTATUS_INVALID_VPID: case MPI2_IOCSTATUS_INVALID_FIELD: case MPI2_IOCSTATUS_INVALID_STATE: case MPI2_IOCSTATUS_OP_STATE_NOT_SUPPORTED: case MPI2_IOCSTATUS_SCSI_IO_DATA_ERROR: case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR: case MPI2_IOCSTATUS_SCSI_RESIDUAL_MISMATCH: case MPI2_IOCSTATUS_SCSI_TASK_MGMT_FAILED: default: mprsas_log_command(cm, MPR_XINFO, "completed ioc %x loginfo %x scsi %x state %x xfer %u\n", le16toh(rep->IOCStatus), le32toh(rep->IOCLogInfo), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); csio->resid = cm->cm_length; if (scsi_cdb[0] == UNMAP && target->is_nvme && (csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); else mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); break; } mpr_sc_failed_io_info(sc, csio, rep, cm->cm_targ); if (sassc->flags & MPRSAS_QUEUE_FROZEN) { ccb->ccb_h.status |= CAM_RELEASE_SIMQ; sassc->flags &= ~MPRSAS_QUEUE_FROZEN; mpr_dprint(sc, MPR_XINFO, "Command completed, unfreezing SIM " "queue\n"); } if (mprsas_get_ccbstatus(ccb) != CAM_REQ_CMP) { ccb->ccb_h.status |= CAM_DEV_QFRZN; xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1); } mpr_free_command(sc, cm); xpt_done(ccb); } #if __FreeBSD_version >= 900026 static void mprsas_smpio_complete(struct mpr_softc *sc, struct mpr_command *cm) { MPI2_SMP_PASSTHROUGH_REPLY *rpl; MPI2_SMP_PASSTHROUGH_REQUEST *req; uint64_t sasaddr; union ccb *ccb; ccb = cm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and SMP * commands require two S/G elements only. That should be handled * in the standard request size. */ if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x on SMP " "request!\n", __func__, cm->cm_flags); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } rpl = (MPI2_SMP_PASSTHROUGH_REPLY *)cm->cm_reply; if (rpl == NULL) { mpr_dprint(sc, MPR_ERROR, "%s: NULL cm_reply!\n", __func__); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req; sasaddr = le32toh(req->SASAddress.Low); sasaddr |= ((uint64_t)(le32toh(req->SASAddress.High))) << 32; if ((le16toh(rpl->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS || rpl->SASStatus != MPI2_SASSTATUS_SUCCESS) { mpr_dprint(sc, MPR_XINFO, "%s: IOCStatus %04x SASStatus %02x\n", __func__, le16toh(rpl->IOCStatus), rpl->SASStatus); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } mpr_dprint(sc, MPR_XINFO, "%s: SMP request to SAS address %#jx " "completed successfully\n", __func__, (uintmax_t)sasaddr); if (ccb->smpio.smp_response[2] == SMP_FR_ACCEPTED) mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); else mprsas_set_ccbstatus(ccb, CAM_SMP_STATUS_ERROR); bailout: /* * We sync in both directions because we had DMAs in the S/G list * in both directions. */ bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); mpr_free_command(sc, cm); xpt_done(ccb); } static void mprsas_send_smpcmd(struct mprsas_softc *sassc, union ccb *ccb, uint64_t sasaddr) { struct mpr_command *cm; uint8_t *request, *response; MPI2_SMP_PASSTHROUGH_REQUEST *req; struct mpr_softc *sc; struct sglist *sg; int error; sc = sassc->sc; sg = NULL; error = 0; #if (__FreeBSD_version >= 1000028) || \ ((__FreeBSD_version >= 902001) && (__FreeBSD_version < 1000000)) switch (ccb->ccb_h.flags & CAM_DATA_MASK) { case CAM_DATA_PADDR: case CAM_DATA_SG_PADDR: /* * XXX We don't yet support physical addresses here. */ mpr_dprint(sc, MPR_ERROR, "%s: physical addresses not " "supported\n", __func__); mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; case CAM_DATA_SG: /* * The chip does not support more than one buffer for the * request or response. */ if ((ccb->smpio.smp_request_sglist_cnt > 1) || (ccb->smpio.smp_response_sglist_cnt > 1)) { mpr_dprint(sc, MPR_ERROR, "%s: multiple request or " "response buffer segments not supported for SMP\n", __func__); mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } /* * The CAM_SCATTER_VALID flag was originally implemented * for the XPT_SCSI_IO CCB, which only has one data pointer. * We have two. So, just take that flag to mean that we * might have S/G lists, and look at the S/G segment count * to figure out whether that is the case for each individual * buffer. */ if (ccb->smpio.smp_request_sglist_cnt != 0) { bus_dma_segment_t *req_sg; req_sg = (bus_dma_segment_t *)ccb->smpio.smp_request; request = (uint8_t *)(uintptr_t)req_sg[0].ds_addr; } else request = ccb->smpio.smp_request; if (ccb->smpio.smp_response_sglist_cnt != 0) { bus_dma_segment_t *rsp_sg; rsp_sg = (bus_dma_segment_t *)ccb->smpio.smp_response; response = (uint8_t *)(uintptr_t)rsp_sg[0].ds_addr; } else response = ccb->smpio.smp_response; break; case CAM_DATA_VADDR: request = ccb->smpio.smp_request; response = ccb->smpio.smp_response; break; default: mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } #else /* __FreeBSD_version < 1000028 */ /* * XXX We don't yet support physical addresses here. */ if (ccb->ccb_h.flags & (CAM_DATA_PHYS|CAM_SG_LIST_PHYS)) { mpr_dprint(sc, MPR_ERROR, "%s: physical addresses not " "supported\n", __func__); mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } /* * If the user wants to send an S/G list, check to make sure they * have single buffers. */ if (ccb->ccb_h.flags & CAM_SCATTER_VALID) { /* * The chip does not support more than one buffer for the * request or response. */ if ((ccb->smpio.smp_request_sglist_cnt > 1) || (ccb->smpio.smp_response_sglist_cnt > 1)) { mpr_dprint(sc, MPR_ERROR, "%s: multiple request or " "response buffer segments not supported for SMP\n", __func__); mprsas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } /* * The CAM_SCATTER_VALID flag was originally implemented * for the XPT_SCSI_IO CCB, which only has one data pointer. * We have two. So, just take that flag to mean that we * might have S/G lists, and look at the S/G segment count * to figure out whether that is the case for each individual * buffer. */ if (ccb->smpio.smp_request_sglist_cnt != 0) { bus_dma_segment_t *req_sg; req_sg = (bus_dma_segment_t *)ccb->smpio.smp_request; request = (uint8_t *)(uintptr_t)req_sg[0].ds_addr; } else request = ccb->smpio.smp_request; if (ccb->smpio.smp_response_sglist_cnt != 0) { bus_dma_segment_t *rsp_sg; rsp_sg = (bus_dma_segment_t *)ccb->smpio.smp_response; response = (uint8_t *)(uintptr_t)rsp_sg[0].ds_addr; } else response = ccb->smpio.smp_response; } else { request = ccb->smpio.smp_request; response = ccb->smpio.smp_response; } #endif /* __FreeBSD_version < 1000028 */ cm = mpr_alloc_command(sc); if (cm == NULL) { mpr_dprint(sc, MPR_ERROR, "%s: cannot allocate command\n", __func__); mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req; bzero(req, sizeof(*req)); req->Function = MPI2_FUNCTION_SMP_PASSTHROUGH; /* Allow the chip to use any route to this SAS address. */ req->PhysicalPort = 0xff; req->RequestDataLength = htole16(ccb->smpio.smp_request_len); req->SGLFlags = MPI2_SGLFLAGS_SYSTEM_ADDRESS_SPACE | MPI2_SGLFLAGS_SGL_TYPE_MPI; mpr_dprint(sc, MPR_XINFO, "%s: sending SMP request to SAS address " "%#jx\n", __func__, (uintmax_t)sasaddr); mpr_init_sge(cm, req, &req->SGL); /* * Set up a uio to pass into mpr_map_command(). This allows us to * do one map command, and one busdma call in there. */ cm->cm_uio.uio_iov = cm->cm_iovec; cm->cm_uio.uio_iovcnt = 2; cm->cm_uio.uio_segflg = UIO_SYSSPACE; /* * The read/write flag isn't used by busdma, but set it just in * case. This isn't exactly accurate, either, since we're going in * both directions. */ cm->cm_uio.uio_rw = UIO_WRITE; cm->cm_iovec[0].iov_base = request; cm->cm_iovec[0].iov_len = le16toh(req->RequestDataLength); cm->cm_iovec[1].iov_base = response; cm->cm_iovec[1].iov_len = ccb->smpio.smp_response_len; cm->cm_uio.uio_resid = cm->cm_iovec[0].iov_len + cm->cm_iovec[1].iov_len; /* * Trigger a warning message in mpr_data_cb() for the user if we * wind up exceeding two S/G segments. The chip expects one * segment for the request and another for the response. */ cm->cm_max_segs = 2; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete = mprsas_smpio_complete; cm->cm_complete_data = ccb; /* * Tell the mapping code that we're using a uio, and that this is * an SMP passthrough request. There is a little special-case * logic there (in mpr_data_cb()) to handle the bidirectional * transfer. */ cm->cm_flags |= MPR_CM_FLAGS_USE_UIO | MPR_CM_FLAGS_SMP_PASS | MPR_CM_FLAGS_DATAIN | MPR_CM_FLAGS_DATAOUT; /* The chip data format is little endian. */ req->SASAddress.High = htole32(sasaddr >> 32); req->SASAddress.Low = htole32(sasaddr); /* * XXX Note that we don't have a timeout/abort mechanism here. * From the manual, it looks like task management requests only * work for SCSI IO and SATA passthrough requests. We may need to * have a mechanism to retry requests in the event of a chip reset * at least. Hopefully the chip will insure that any errors short * of that are relayed back to the driver. */ error = mpr_map_command(sc, cm); if ((error != 0) && (error != EINPROGRESS)) { mpr_dprint(sc, MPR_ERROR, "%s: error %d returned from " "mpr_map_command()\n", __func__, error); goto bailout_error; } return; bailout_error: mpr_free_command(sc, cm); mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } static void mprsas_action_smpio(struct mprsas_softc *sassc, union ccb *ccb) { struct mpr_softc *sc; struct mprsas_target *targ; uint64_t sasaddr = 0; sc = sassc->sc; /* * Make sure the target exists. */ KASSERT(ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_SMP_IO\n", ccb->ccb_h.target_id)); targ = &sassc->targets[ccb->ccb_h.target_id]; if (targ->handle == 0x0) { mpr_dprint(sc, MPR_ERROR, "%s: target %d does not exist!\n", __func__, ccb->ccb_h.target_id); mprsas_set_ccbstatus(ccb, CAM_SEL_TIMEOUT); xpt_done(ccb); return; } /* * If this device has an embedded SMP target, we'll talk to it * directly. * figure out what the expander's address is. */ if ((targ->devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) != 0) sasaddr = targ->sasaddr; /* * If we don't have a SAS address for the expander yet, try * grabbing it from the page 0x83 information cached in the * transport layer for this target. LSI expanders report the * expander SAS address as the port-associated SAS address in * Inquiry VPD page 0x83. Maxim expanders don't report it in page * 0x83. * * XXX KDM disable this for now, but leave it commented out so that * it is obvious that this is another possible way to get the SAS * address. * * The parent handle method below is a little more reliable, and * the other benefit is that it works for devices other than SES * devices. So you can send a SMP request to a da(4) device and it * will get routed to the expander that device is attached to. * (Assuming the da(4) device doesn't contain an SMP target...) */ #if 0 if (sasaddr == 0) sasaddr = xpt_path_sas_addr(ccb->ccb_h.path); #endif /* * If we still don't have a SAS address for the expander, look for * the parent device of this device, which is probably the expander. */ if (sasaddr == 0) { #ifdef OLD_MPR_PROBE struct mprsas_target *parent_target; #endif if (targ->parent_handle == 0x0) { mpr_dprint(sc, MPR_ERROR, "%s: handle %d does not have " "a valid parent handle!\n", __func__, targ->handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } #ifdef OLD_MPR_PROBE parent_target = mprsas_find_target_by_handle(sassc, 0, targ->parent_handle); if (parent_target == NULL) { mpr_dprint(sc, MPR_ERROR, "%s: handle %d does not have " "a valid parent target!\n", __func__, targ->handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } if ((parent_target->devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) { mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent %d " "does not have an SMP target!\n", __func__, targ->handle, parent_target->handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } sasaddr = parent_target->sasaddr; #else /* OLD_MPR_PROBE */ if ((targ->parent_devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) { mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent %d " "does not have an SMP target!\n", __func__, targ->handle, targ->parent_handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } if (targ->parent_sasaddr == 0x0) { mpr_dprint(sc, MPR_ERROR, "%s: handle %d parent handle " "%d does not have a valid SAS address!\n", __func__, targ->handle, targ->parent_handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } sasaddr = targ->parent_sasaddr; #endif /* OLD_MPR_PROBE */ } if (sasaddr == 0) { mpr_dprint(sc, MPR_INFO, "%s: unable to find SAS address for " "handle %d\n", __func__, targ->handle); mprsas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } mprsas_send_smpcmd(sassc, ccb, sasaddr); return; bailout: xpt_done(ccb); } #endif //__FreeBSD_version >= 900026 static void mprsas_action_resetdev(struct mprsas_softc *sassc, union ccb *ccb) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpr_softc *sc; struct mpr_command *tm; struct mprsas_target *targ; MPR_FUNCTRACE(sassc->sc); mtx_assert(&sassc->sc->mpr_mtx, MA_OWNED); KASSERT(ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of " "bounds in XPT_RESET_DEV\n", ccb->ccb_h.target_id)); sc = sassc->sc; tm = mprsas_alloc_tm(sc); if (tm == NULL) { mpr_dprint(sc, MPR_ERROR, "command alloc failure in " "mprsas_action_resetdev\n"); mprsas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } targ = &sassc->targets[ccb->ccb_h.target_id]; req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; if (!targ->is_nvme || sc->custom_nvme_tm_handling) { /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; } else { /* PCIe Protocol Level Reset*/ req->MsgFlags = MPI26_SCSITASKMGMT_MSGFLAGS_PROTOCOL_LVL_RST_PCIE; } tm->cm_data = NULL; tm->cm_complete = mprsas_resetdev_complete; tm->cm_complete_data = ccb; mpr_dprint(sc, MPR_INFO, "%s: Sending reset for target ID %d\n", __func__, targ->tid); tm->cm_targ = targ; mprsas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD); mpr_map_command(sc, tm); } static void mprsas_resetdev_complete(struct mpr_softc *sc, struct mpr_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *resp; union ccb *ccb; MPR_FUNCTRACE(sc); mtx_assert(&sc->mpr_mtx, MA_OWNED); resp = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; ccb = tm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for reset of " "handle %#04x! This should not happen!\n", __func__, tm->cm_flags, req->DevHandle); mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } mpr_dprint(sc, MPR_XINFO, "%s: IOCStatus = 0x%x ResponseCode = 0x%x\n", __func__, le16toh(resp->IOCStatus), le32toh(resp->ResponseCode)); if (le32toh(resp->ResponseCode) == MPI2_SCSITASKMGMT_RSP_TM_COMPLETE) { mprsas_set_ccbstatus(ccb, CAM_REQ_CMP); mprsas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid, CAM_LUN_WILDCARD); } else mprsas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); bailout: mprsas_free_tm(sc, tm); xpt_done(ccb); } static void mprsas_poll(struct cam_sim *sim) { struct mprsas_softc *sassc; sassc = cam_sim_softc(sim); if (sassc->sc->mpr_debug & MPR_TRACE) { /* frequent debug messages during a panic just slow * everything down too much. */ mpr_dprint(sassc->sc, MPR_XINFO, "%s clearing MPR_TRACE\n", __func__); sassc->sc->mpr_debug &= ~MPR_TRACE; } mpr_intr_locked(sassc->sc); } static void mprsas_async(void *callback_arg, uint32_t code, struct cam_path *path, void *arg) { struct mpr_softc *sc; sc = (struct mpr_softc *)callback_arg; switch (code) { #if (__FreeBSD_version >= 1000006) || \ ((__FreeBSD_version >= 901503) && (__FreeBSD_version < 1000000)) case AC_ADVINFO_CHANGED: { struct mprsas_target *target; struct mprsas_softc *sassc; struct scsi_read_capacity_data_long rcap_buf; struct ccb_dev_advinfo cdai; struct mprsas_lun *lun; lun_id_t lunid; int found_lun; uintptr_t buftype; buftype = (uintptr_t)arg; found_lun = 0; sassc = sc->sassc; /* * We're only interested in read capacity data changes. */ if (buftype != CDAI_TYPE_RCAPLONG) break; /* * See the comment in mpr_attach_sas() for a detailed * explanation. In these versions of FreeBSD we register * for all events and filter out the events that don't * apply to us. */ #if (__FreeBSD_version < 1000703) || \ ((__FreeBSD_version >= 1100000) && (__FreeBSD_version < 1100002)) if (xpt_path_path_id(path) != sassc->sim->path_id) break; #endif /* * We should have a handle for this, but check to make sure. */ KASSERT(xpt_path_target_id(path) < sassc->maxtargets, ("Target %d out of bounds in mprsas_async\n", xpt_path_target_id(path))); target = &sassc->targets[xpt_path_target_id(path)]; if (target->handle == 0) break; lunid = xpt_path_lun_id(path); SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id == lunid) { found_lun = 1; break; } } if (found_lun == 0) { lun = malloc(sizeof(struct mprsas_lun), M_MPR, M_NOWAIT | M_ZERO); if (lun == NULL) { mpr_dprint(sc, MPR_ERROR, "Unable to alloc " "LUN for EEDP support.\n"); break; } lun->lun_id = lunid; SLIST_INSERT_HEAD(&target->luns, lun, lun_link); } bzero(&rcap_buf, sizeof(rcap_buf)); xpt_setup_ccb(&cdai.ccb_h, path, CAM_PRIORITY_NORMAL); cdai.ccb_h.func_code = XPT_DEV_ADVINFO; cdai.ccb_h.flags = CAM_DIR_IN; cdai.buftype = CDAI_TYPE_RCAPLONG; #if (__FreeBSD_version >= 1100061) || \ ((__FreeBSD_version >= 1001510) && (__FreeBSD_version < 1100000)) cdai.flags = CDAI_FLAG_NONE; #else cdai.flags = 0; #endif cdai.bufsiz = sizeof(rcap_buf); cdai.buf = (uint8_t *)&rcap_buf; xpt_action((union ccb *)&cdai); if ((cdai.ccb_h.status & CAM_DEV_QFRZN) != 0) cam_release_devq(cdai.ccb_h.path, 0, 0, 0, FALSE); if ((mprsas_get_ccbstatus((union ccb *)&cdai) == CAM_REQ_CMP) && (rcap_buf.prot & SRC16_PROT_EN)) { switch (rcap_buf.prot & SRC16_P_TYPE) { case SRC16_PTYPE_1: case SRC16_PTYPE_3: lun->eedp_formatted = TRUE; lun->eedp_block_size = scsi_4btoul(rcap_buf.length); break; case SRC16_PTYPE_2: default: lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; break; } } else { lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; } break; } #endif case AC_FOUND_DEVICE: { struct ccb_getdev *cgd; /* * See the comment in mpr_attach_sas() for a detailed * explanation. In these versions of FreeBSD we register * for all events and filter out the events that don't * apply to us. */ #if (__FreeBSD_version < 1000703) || \ ((__FreeBSD_version >= 1100000) && (__FreeBSD_version < 1100002)) if (xpt_path_path_id(path) != sc->sassc->sim->path_id) break; #endif cgd = arg; #if (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) mprsas_check_eedp(sc, path, cgd); #endif break; } default: break; } } #if (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) static void mprsas_check_eedp(struct mpr_softc *sc, struct cam_path *path, struct ccb_getdev *cgd) { struct mprsas_softc *sassc = sc->sassc; struct ccb_scsiio *csio; struct scsi_read_capacity_16 *scsi_cmd; struct scsi_read_capacity_eedp *rcap_buf; path_id_t pathid; target_id_t targetid; lun_id_t lunid; union ccb *ccb; struct cam_path *local_path; struct mprsas_target *target; struct mprsas_lun *lun; uint8_t found_lun; char path_str[64]; pathid = cam_sim_path(sassc->sim); targetid = xpt_path_target_id(path); lunid = xpt_path_lun_id(path); KASSERT(targetid < sassc->maxtargets, ("Target %d out of bounds in " "mprsas_check_eedp\n", targetid)); target = &sassc->targets[targetid]; if (target->handle == 0x0) return; /* * Determine if the device is EEDP capable. * * If this flag is set in the inquiry data, the device supports * protection information, and must support the 16 byte read capacity * command, otherwise continue without sending read cap 16. */ if ((cgd->inq_data.spc3_flags & SPC3_SID_PROTECT) == 0) return; /* * Issue a READ CAPACITY 16 command. This info is used to determine if * the LUN is formatted for EEDP support. */ ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mpr_dprint(sc, MPR_ERROR, "Unable to alloc CCB for EEDP " "support.\n"); return; } if (xpt_create_path(&local_path, xpt_periph, pathid, targetid, lunid) != CAM_REQ_CMP) { mpr_dprint(sc, MPR_ERROR, "Unable to create path for EEDP " "support.\n"); xpt_free_ccb(ccb); return; } /* * If LUN is already in list, don't create a new one. */ found_lun = FALSE; SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id == lunid) { found_lun = TRUE; break; } } if (!found_lun) { lun = malloc(sizeof(struct mprsas_lun), M_MPR, M_NOWAIT | M_ZERO); if (lun == NULL) { mpr_dprint(sc, MPR_ERROR, "Unable to alloc LUN for " "EEDP support.\n"); xpt_free_path(local_path); xpt_free_ccb(ccb); return; } lun->lun_id = lunid; SLIST_INSERT_HEAD(&target->luns, lun, lun_link); } xpt_path_string(local_path, path_str, sizeof(path_str)); mpr_dprint(sc, MPR_INFO, "Sending read cap: path %s handle %d\n", path_str, target->handle); /* * Issue a READ CAPACITY 16 command for the LUN. The * mprsas_read_cap_done function will load the read cap info into the * LUN struct. */ rcap_buf = malloc(sizeof(struct scsi_read_capacity_eedp), M_MPR, M_NOWAIT | M_ZERO); if (rcap_buf == NULL) { mpr_dprint(sc, MPR_ERROR, "Unable to alloc read capacity " "buffer for EEDP support.\n"); xpt_free_path(ccb->ccb_h.path); xpt_free_ccb(ccb); return; } xpt_setup_ccb(&ccb->ccb_h, local_path, CAM_PRIORITY_XPT); csio = &ccb->csio; csio->ccb_h.func_code = XPT_SCSI_IO; csio->ccb_h.flags = CAM_DIR_IN; csio->ccb_h.retry_count = 4; csio->ccb_h.cbfcnp = mprsas_read_cap_done; csio->ccb_h.timeout = 60000; csio->data_ptr = (uint8_t *)rcap_buf; csio->dxfer_len = sizeof(struct scsi_read_capacity_eedp); csio->sense_len = MPR_SENSE_LEN; csio->cdb_len = sizeof(*scsi_cmd); csio->tag_action = MSG_SIMPLE_Q_TAG; scsi_cmd = (struct scsi_read_capacity_16 *)&csio->cdb_io.cdb_bytes; bzero(scsi_cmd, sizeof(*scsi_cmd)); scsi_cmd->opcode = 0x9E; scsi_cmd->service_action = SRC16_SERVICE_ACTION; ((uint8_t *)scsi_cmd)[13] = sizeof(struct scsi_read_capacity_eedp); ccb->ccb_h.ppriv_ptr1 = sassc; xpt_action(ccb); } static void mprsas_read_cap_done(struct cam_periph *periph, union ccb *done_ccb) { struct mprsas_softc *sassc; struct mprsas_target *target; struct mprsas_lun *lun; struct scsi_read_capacity_eedp *rcap_buf; if (done_ccb == NULL) return; /* Driver need to release devq, it Scsi command is * generated by driver internally. * Currently there is a single place where driver * calls scsi command internally. In future if driver * calls more scsi command internally, it needs to release * devq internally, since those command will not go back to * cam_periph. */ if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) ) { done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN; xpt_release_devq(done_ccb->ccb_h.path, /*count*/ 1, /*run_queue*/TRUE); } rcap_buf = (struct scsi_read_capacity_eedp *)done_ccb->csio.data_ptr; /* * Get the LUN ID for the path and look it up in the LUN list for the * target. */ sassc = (struct mprsas_softc *)done_ccb->ccb_h.ppriv_ptr1; KASSERT(done_ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out " "of bounds in mprsas_read_cap_done\n", done_ccb->ccb_h.target_id)); target = &sassc->targets[done_ccb->ccb_h.target_id]; SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id != done_ccb->ccb_h.target_lun) continue; /* * Got the LUN in the target's LUN list. Fill it in with EEDP * info. If the READ CAP 16 command had some SCSI error (common * if command is not supported), mark the lun as not supporting * EEDP and set the block size to 0. */ if ((mprsas_get_ccbstatus(done_ccb) != CAM_REQ_CMP) || (done_ccb->csio.scsi_status != SCSI_STATUS_OK)) { lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; break; } if (rcap_buf->protect & 0x01) { mpr_dprint(sassc->sc, MPR_INFO, "LUN %d for target ID " "%d is formatted for EEDP support.\n", done_ccb->ccb_h.target_lun, done_ccb->ccb_h.target_id); lun->eedp_formatted = TRUE; lun->eedp_block_size = scsi_4btoul(rcap_buf->length); } break; } // Finished with this CCB and path. free(rcap_buf, M_MPR); xpt_free_path(done_ccb->ccb_h.path); xpt_free_ccb(done_ccb); } #endif /* (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) */ /* * Set the INRESET flag for this target so that no I/O will be sent to * the target until the reset has completed. If an I/O request does * happen, the devq will be frozen. The CCB holds the path which is * used to release the devq. The devq is released and the CCB is freed * when the TM completes. */ void mprsas_prepare_for_tm(struct mpr_softc *sc, struct mpr_command *tm, struct mprsas_target *target, lun_id_t lun_id) { union ccb *ccb; path_id_t path_id; ccb = xpt_alloc_ccb_nowait(); if (ccb) { path_id = cam_sim_path(sc->sassc->sim); if (xpt_create_path(&ccb->ccb_h.path, xpt_periph, path_id, target->tid, lun_id) != CAM_REQ_CMP) { xpt_free_ccb(ccb); } else { tm->cm_ccb = ccb; tm->cm_targ = target; target->flags |= MPRSAS_TARGET_INRESET; } } } int mprsas_startup(struct mpr_softc *sc) { /* * Send the port enable message and set the wait_for_port_enable flag. * This flag helps to keep the simq frozen until all discovery events * are processed. */ sc->wait_for_port_enable = 1; mprsas_send_portenable(sc); return (0); } static int mprsas_send_portenable(struct mpr_softc *sc) { MPI2_PORT_ENABLE_REQUEST *request; struct mpr_command *cm; MPR_FUNCTRACE(sc); if ((cm = mpr_alloc_command(sc)) == NULL) return (EBUSY); request = (MPI2_PORT_ENABLE_REQUEST *)cm->cm_req; request->Function = MPI2_FUNCTION_PORT_ENABLE; request->MsgFlags = 0; request->VP_ID = 0; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete = mprsas_portenable_complete; cm->cm_data = NULL; cm->cm_sge = NULL; mpr_map_command(sc, cm); mpr_dprint(sc, MPR_XINFO, "mpr_send_portenable finished cm %p req %p complete %p\n", cm, cm->cm_req, cm->cm_complete); return (0); } static void mprsas_portenable_complete(struct mpr_softc *sc, struct mpr_command *cm) { MPI2_PORT_ENABLE_REPLY *reply; struct mprsas_softc *sassc; MPR_FUNCTRACE(sc); sassc = sc->sassc; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * port enable commands don't have S/G lists. */ if ((cm->cm_flags & MPR_CM_FLAGS_ERROR_MASK) != 0) { mpr_dprint(sc, MPR_ERROR, "%s: cm_flags = %#x for port enable! " "This should not happen!\n", __func__, cm->cm_flags); } reply = (MPI2_PORT_ENABLE_REPLY *)cm->cm_reply; if (reply == NULL) mpr_dprint(sc, MPR_FAULT, "Portenable NULL reply\n"); else if (le16toh(reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) mpr_dprint(sc, MPR_FAULT, "Portenable failed\n"); mpr_free_command(sc, cm); /* * Done waiting for port enable to complete. Decrement the refcount. * If refcount is 0, discovery is complete and a rescan of the bus can * take place. */ sc->wait_for_port_enable = 0; sc->port_enable_complete = 1; wakeup(&sc->port_enable_complete); mprsas_startup_decrement(sassc); } int mprsas_check_id(struct mprsas_softc *sassc, int id) { struct mpr_softc *sc = sassc->sc; char *ids; char *name; ids = &sc->exclude_ids[0]; while((name = strsep(&ids, ",")) != NULL) { if (name[0] == '\0') continue; if (strtol(name, NULL, 0) == (long)id) return (1); } return (0); } void mprsas_realloc_targets(struct mpr_softc *sc, int maxtargets) { struct mprsas_softc *sassc; struct mprsas_lun *lun, *lun_tmp; struct mprsas_target *targ; int i; sassc = sc->sassc; /* * The number of targets is based on IOC Facts, so free all of * the allocated LUNs for each target and then the target buffer * itself. */ for (i=0; i< maxtargets; i++) { targ = &sassc->targets[i]; SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) { free(lun, M_MPR); } } free(sassc->targets, M_MPR); sassc->targets = malloc(sizeof(struct mprsas_target) * maxtargets, M_MPR, M_WAITOK|M_ZERO); if (!sassc->targets) { panic("%s failed to alloc targets with error %d\n", __func__, ENOMEM); } } Index: head/sys/dev/mpr/mpr_sas_lsi.c =================================================================== --- head/sys/dev/mpr/mpr_sas_lsi.c (revision 349848) +++ head/sys/dev/mpr/mpr_sas_lsi.c (revision 349849) @@ -1,1699 +1,1701 @@ /*- * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2016 Avago Technologies * Copyright 2000-2020 Broadcom Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Broadcom Inc. (LSI) MPT-Fusion Host Adapter FreeBSD */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT3 */ /* TODO Move headers to mprvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* For Hashed SAS Address creation for SATA Drives */ #define MPT2SAS_SN_LEN 20 #define MPT2SAS_MN_LEN 40 struct mpr_fw_event_work { u16 event; void *event_data; TAILQ_ENTRY(mpr_fw_event_work) ev_link; }; union _sata_sas_address { u8 wwid[8]; struct { u32 high; u32 low; } word; }; /* * define the IDENTIFY DEVICE structure */ struct _ata_identify_device_data { u16 reserved1[10]; /* 0-9 */ u16 serial_number[10]; /* 10-19 */ u16 reserved2[7]; /* 20-26 */ u16 model_number[20]; /* 27-46*/ u16 reserved3[170]; /* 47-216 */ u16 rotational_speed; /* 217 */ u16 reserved4[38]; /* 218-255 */ }; static u32 event_count; static void mprsas_fw_work(struct mpr_softc *sc, struct mpr_fw_event_work *fw_event); static void mprsas_fw_event_free(struct mpr_softc *, struct mpr_fw_event_work *); static int mprsas_add_device(struct mpr_softc *sc, u16 handle, u8 linkrate); static int mprsas_add_pcie_device(struct mpr_softc *sc, u16 handle, u8 linkrate); static int mprsas_get_sata_identify(struct mpr_softc *sc, u16 handle, Mpi2SataPassthroughReply_t *mpi_reply, char *id_buffer, int sz, u32 devinfo); static void mprsas_ata_id_timeout(struct mpr_softc *, struct mpr_command *); int mprsas_get_sas_address_for_sata_disk(struct mpr_softc *sc, u64 *sas_address, u16 handle, u32 device_info, u8 *is_SATA_SSD); static int mprsas_volume_add(struct mpr_softc *sc, u16 handle); static void mprsas_SSU_to_SATA_devices(struct mpr_softc *sc, int howto); static void mprsas_stop_unit_done(struct cam_periph *periph, union ccb *done_ccb); void mprsas_evt_handler(struct mpr_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event) { struct mpr_fw_event_work *fw_event; u16 sz; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); MPR_DPRINT_EVENT(sc, sas, event); mprsas_record_event(sc, event); fw_event = malloc(sizeof(struct mpr_fw_event_work), M_MPR, M_ZERO|M_NOWAIT); if (!fw_event) { printf("%s: allocate failed for fw_event\n", __func__); return; } sz = le16toh(event->EventDataLength) * 4; fw_event->event_data = malloc(sz, M_MPR, M_ZERO|M_NOWAIT); if (!fw_event->event_data) { printf("%s: allocate failed for event_data\n", __func__); free(fw_event, M_MPR); return; } bcopy(event->EventData, fw_event->event_data, sz); fw_event->event = event->Event; if ((event->Event == MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE || event->Event == MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST) && sc->track_mapping_events) sc->pending_map_events++; /* * When wait_for_port_enable flag is set, make sure that all the events * are processed. Increment the startup_refcount and decrement it after * events are processed. */ if ((event->Event == MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST) && sc->wait_for_port_enable) mprsas_startup_increment(sc->sassc); TAILQ_INSERT_TAIL(&sc->sassc->ev_queue, fw_event, ev_link); taskqueue_enqueue(sc->sassc->ev_tq, &sc->sassc->ev_task); } static void mprsas_fw_event_free(struct mpr_softc *sc, struct mpr_fw_event_work *fw_event) { free(fw_event->event_data, M_MPR); free(fw_event, M_MPR); } /** * _mpr_fw_work - delayed task for processing firmware events * @sc: per adapter object * @fw_event: The fw_event_work object * Context: user. * * Return nothing. */ static void mprsas_fw_work(struct mpr_softc *sc, struct mpr_fw_event_work *fw_event) { struct mprsas_softc *sassc; sassc = sc->sassc; mpr_dprint(sc, MPR_EVENT, "(%d)->(%s) Working on Event: [%x]\n", event_count++, __func__, fw_event->event); switch (fw_event->event) { case MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST: { MPI2_EVENT_DATA_SAS_TOPOLOGY_CHANGE_LIST *data; MPI2_EVENT_SAS_TOPO_PHY_ENTRY *phy; uint8_t i; data = (MPI2_EVENT_DATA_SAS_TOPOLOGY_CHANGE_LIST *) fw_event->event_data; mpr_mapping_topology_change_event(sc, fw_event->event_data); for (i = 0; i < data->NumEntries; i++) { phy = &data->PHY[i]; switch (phy->PhyStatus & MPI2_EVENT_SAS_TOPO_RC_MASK) { case MPI2_EVENT_SAS_TOPO_RC_TARG_ADDED: if (mprsas_add_device(sc, le16toh(phy->AttachedDevHandle), phy->LinkRate)) { mpr_dprint(sc, MPR_ERROR, "%s: " "failed to add device with handle " "0x%x\n", __func__, le16toh(phy->AttachedDevHandle)); mprsas_prepare_remove(sassc, le16toh( phy->AttachedDevHandle)); } break; case MPI2_EVENT_SAS_TOPO_RC_TARG_NOT_RESPONDING: mprsas_prepare_remove(sassc, le16toh( phy->AttachedDevHandle)); break; case MPI2_EVENT_SAS_TOPO_RC_PHY_CHANGED: case MPI2_EVENT_SAS_TOPO_RC_NO_CHANGE: case MPI2_EVENT_SAS_TOPO_RC_DELAY_NOT_RESPONDING: default: break; } } /* * refcount was incremented for this event in * mprsas_evt_handler. Decrement it here because the event has * been processed. */ mprsas_startup_decrement(sassc); break; } case MPI2_EVENT_SAS_DISCOVERY: { MPI2_EVENT_DATA_SAS_DISCOVERY *data; data = (MPI2_EVENT_DATA_SAS_DISCOVERY *)fw_event->event_data; if (data->ReasonCode & MPI2_EVENT_SAS_DISC_RC_STARTED) mpr_dprint(sc, MPR_TRACE,"SAS discovery start event\n"); if (data->ReasonCode & MPI2_EVENT_SAS_DISC_RC_COMPLETED) { mpr_dprint(sc, MPR_TRACE,"SAS discovery stop event\n"); sassc->flags &= ~MPRSAS_IN_DISCOVERY; mprsas_discovery_end(sassc); } break; } case MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE: { Mpi2EventDataSasEnclDevStatusChange_t *data; data = (Mpi2EventDataSasEnclDevStatusChange_t *) fw_event->event_data; mpr_mapping_enclosure_dev_status_change_event(sc, fw_event->event_data); break; } case MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST: { Mpi2EventIrConfigElement_t *element; int i; u8 foreign_config, reason; u16 elementType; Mpi2EventDataIrConfigChangeList_t *event_data; struct mprsas_target *targ; unsigned int id; event_data = fw_event->event_data; foreign_config = (le32toh(event_data->Flags) & MPI2_EVENT_IR_CHANGE_FLAGS_FOREIGN_CONFIG) ? 1 : 0; element = (Mpi2EventIrConfigElement_t *)&event_data->ConfigElement[0]; id = mpr_mapping_get_raid_tid_from_handle(sc, element->VolDevHandle); mpr_mapping_ir_config_change_event(sc, event_data); for (i = 0; i < event_data->NumElements; i++, element++) { reason = element->ReasonCode; elementType = le16toh(element->ElementFlags) & MPI2_EVENT_IR_CHANGE_EFLAGS_ELEMENT_TYPE_MASK; /* * check for element type of Phys Disk or Hot Spare */ if ((elementType != MPI2_EVENT_IR_CHANGE_EFLAGS_VOLPHYSDISK_ELEMENT) && (elementType != MPI2_EVENT_IR_CHANGE_EFLAGS_HOTSPARE_ELEMENT)) // do next element goto skip_fp_send; /* * check for reason of Hide, Unhide, PD Created, or PD * Deleted */ if ((reason != MPI2_EVENT_IR_CHANGE_RC_HIDE) && (reason != MPI2_EVENT_IR_CHANGE_RC_UNHIDE) && (reason != MPI2_EVENT_IR_CHANGE_RC_PD_CREATED) && (reason != MPI2_EVENT_IR_CHANGE_RC_PD_DELETED)) goto skip_fp_send; // check for a reason of Hide or PD Created if ((reason == MPI2_EVENT_IR_CHANGE_RC_HIDE) || (reason == MPI2_EVENT_IR_CHANGE_RC_PD_CREATED)) { // build RAID Action message Mpi2RaidActionRequest_t *action; Mpi2RaidActionReply_t *reply = NULL; struct mpr_command *cm; int error = 0; if ((cm = mpr_alloc_command(sc)) == NULL) { printf("%s: command alloc failed\n", __func__); return; } mpr_dprint(sc, MPR_EVENT, "Sending FP action " "from " "MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST " ":\n"); action = (MPI2_RAID_ACTION_REQUEST *)cm->cm_req; action->Function = MPI2_FUNCTION_RAID_ACTION; action->Action = MPI2_RAID_ACTION_PHYSDISK_HIDDEN; action->PhysDiskNum = element->PhysDiskNum; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; error = mpr_request_polled(sc, &cm); if (cm != NULL) reply = (Mpi2RaidActionReply_t *) cm->cm_reply; if (error || (reply == NULL)) { /* FIXME */ /* * If the poll returns error then we * need to do diag reset */ printf("%s: poll for page completed " "with error %d", __func__, error); } if (reply && (le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mpr_dprint(sc, MPR_ERROR, "%s: error " "sending RaidActionPage; " "iocstatus = 0x%x\n", __func__, le16toh(reply->IOCStatus)); } if (cm) mpr_free_command(sc, cm); } skip_fp_send: mpr_dprint(sc, MPR_EVENT, "Received " "MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST Reason " "code %x:\n", element->ReasonCode); switch (element->ReasonCode) { case MPI2_EVENT_IR_CHANGE_RC_VOLUME_CREATED: case MPI2_EVENT_IR_CHANGE_RC_ADDED: if (!foreign_config) { if (mprsas_volume_add(sc, le16toh(element->VolDevHandle))) { printf("%s: failed to add RAID " "volume with handle 0x%x\n", __func__, le16toh(element-> VolDevHandle)); } } break; case MPI2_EVENT_IR_CHANGE_RC_VOLUME_DELETED: case MPI2_EVENT_IR_CHANGE_RC_REMOVED: /* * Rescan after volume is deleted or removed. */ if (!foreign_config) { if (id == MPR_MAP_BAD_ID) { printf("%s: could not get ID " "for volume with handle " "0x%04x\n", __func__, le16toh(element-> VolDevHandle)); break; } targ = &sassc->targets[id]; targ->handle = 0x0; targ->encl_slot = 0x0; targ->encl_handle = 0x0; targ->encl_level_valid = 0x0; targ->encl_level = 0x0; targ->connector_name[0] = ' '; targ->connector_name[1] = ' '; targ->connector_name[2] = ' '; targ->connector_name[3] = ' '; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; mprsas_rescan_target(sc, targ); printf("RAID target id 0x%x removed\n", targ->tid); } break; case MPI2_EVENT_IR_CHANGE_RC_PD_CREATED: case MPI2_EVENT_IR_CHANGE_RC_HIDE: /* * Phys Disk of a volume has been created. Hide * it from the OS. */ targ = mprsas_find_target_by_handle(sassc, 0, element->PhysDiskDevHandle); if (targ == NULL) break; targ->flags |= MPR_TARGET_FLAGS_RAID_COMPONENT; mprsas_rescan_target(sc, targ); break; case MPI2_EVENT_IR_CHANGE_RC_PD_DELETED: /* * Phys Disk of a volume has been deleted. * Expose it to the OS. */ if (mprsas_add_device(sc, le16toh(element->PhysDiskDevHandle), 0)) { printf("%s: failed to add device with " "handle 0x%x\n", __func__, le16toh(element-> PhysDiskDevHandle)); mprsas_prepare_remove(sassc, le16toh(element-> PhysDiskDevHandle)); } break; } } /* * refcount was incremented for this event in * mprsas_evt_handler. Decrement it here because the event has * been processed. */ mprsas_startup_decrement(sassc); break; } case MPI2_EVENT_IR_VOLUME: { Mpi2EventDataIrVolume_t *event_data = fw_event->event_data; /* * Informational only. */ mpr_dprint(sc, MPR_EVENT, "Received IR Volume event:\n"); switch (event_data->ReasonCode) { case MPI2_EVENT_IR_VOLUME_RC_SETTINGS_CHANGED: mpr_dprint(sc, MPR_EVENT, " Volume Settings " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); break; case MPI2_EVENT_IR_VOLUME_RC_STATUS_FLAGS_CHANGED: mpr_dprint(sc, MPR_EVENT, " Volume Status " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); break; case MPI2_EVENT_IR_VOLUME_RC_STATE_CHANGED: mpr_dprint(sc, MPR_EVENT, " Volume State " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); u32 state; struct mprsas_target *targ; state = le32toh(event_data->NewValue); switch (state) { case MPI2_RAID_VOL_STATE_MISSING: case MPI2_RAID_VOL_STATE_FAILED: mprsas_prepare_volume_remove(sassc, event_data->VolDevHandle); break; case MPI2_RAID_VOL_STATE_ONLINE: case MPI2_RAID_VOL_STATE_DEGRADED: case MPI2_RAID_VOL_STATE_OPTIMAL: targ = mprsas_find_target_by_handle(sassc, 0, event_data->VolDevHandle); if (targ) { printf("%s %d: Volume handle " "0x%x is already added \n", __func__, __LINE__, event_data->VolDevHandle); break; } if (mprsas_volume_add(sc, le16toh(event_data-> VolDevHandle))) { printf("%s: failed to add RAID " "volume with handle 0x%x\n", __func__, le16toh( event_data->VolDevHandle)); } break; default: break; } break; default: break; } break; } case MPI2_EVENT_IR_PHYSICAL_DISK: { Mpi2EventDataIrPhysicalDisk_t *event_data = fw_event->event_data; struct mprsas_target *targ; /* * Informational only. */ mpr_dprint(sc, MPR_EVENT, "Received IR Phys Disk event:\n"); switch (event_data->ReasonCode) { case MPI2_EVENT_IR_PHYSDISK_RC_SETTINGS_CHANGED: mpr_dprint(sc, MPR_EVENT, " Phys Disk Settings " "changed from 0x%x to 0x%x for Phys Disk Number " "%d and handle 0x%x at Enclosure handle 0x%x, Slot " "%d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); break; case MPI2_EVENT_IR_PHYSDISK_RC_STATUS_FLAGS_CHANGED: mpr_dprint(sc, MPR_EVENT, " Phys Disk Status changed " "from 0x%x to 0x%x for Phys Disk Number %d and " "handle 0x%x at Enclosure handle 0x%x, Slot %d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); break; case MPI2_EVENT_IR_PHYSDISK_RC_STATE_CHANGED: mpr_dprint(sc, MPR_EVENT, " Phys Disk State changed " "from 0x%x to 0x%x for Phys Disk Number %d and " "handle 0x%x at Enclosure handle 0x%x, Slot %d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); switch (event_data->NewValue) { case MPI2_RAID_PD_STATE_ONLINE: case MPI2_RAID_PD_STATE_DEGRADED: case MPI2_RAID_PD_STATE_REBUILDING: case MPI2_RAID_PD_STATE_OPTIMAL: case MPI2_RAID_PD_STATE_HOT_SPARE: targ = mprsas_find_target_by_handle( sassc, 0, event_data->PhysDiskDevHandle); if (targ) { targ->flags |= MPR_TARGET_FLAGS_RAID_COMPONENT; printf("%s %d: Found Target " "for handle 0x%x.\n", __func__, __LINE__ , event_data-> PhysDiskDevHandle); } break; case MPI2_RAID_PD_STATE_OFFLINE: case MPI2_RAID_PD_STATE_NOT_CONFIGURED: case MPI2_RAID_PD_STATE_NOT_COMPATIBLE: default: targ = mprsas_find_target_by_handle( sassc, 0, event_data->PhysDiskDevHandle); if (targ) { targ->flags |= ~MPR_TARGET_FLAGS_RAID_COMPONENT; printf("%s %d: Found Target " "for handle 0x%x. \n", __func__, __LINE__ , event_data-> PhysDiskDevHandle); } break; } default: break; } break; } case MPI2_EVENT_IR_OPERATION_STATUS: { Mpi2EventDataIrOperationStatus_t *event_data = fw_event->event_data; /* * Informational only. */ mpr_dprint(sc, MPR_EVENT, "Received IR Op Status event:\n"); mpr_dprint(sc, MPR_EVENT, " RAID Operation of %d is %d " "percent complete for Volume with handle 0x%x", event_data->RAIDOperation, event_data->PercentComplete, le16toh(event_data->VolDevHandle)); break; } case MPI2_EVENT_TEMP_THRESHOLD: { pMpi2EventDataTemperature_t temp_event; temp_event = (pMpi2EventDataTemperature_t)fw_event->event_data; /* * The Temp Sensor Count must be greater than the event's Sensor * Num to be valid. If valid, print the temp thresholds that * have been exceeded. */ if (sc->iounit_pg8.NumSensors > temp_event->SensorNum) { mpr_dprint(sc, MPR_FAULT, "Temperature Threshold flags " "%s %s %s %s exceeded for Sensor: %d !!!\n", ((temp_event->Status & 0x01) == 1) ? "0 " : " ", ((temp_event->Status & 0x02) == 2) ? "1 " : " ", ((temp_event->Status & 0x04) == 4) ? "2 " : " ", ((temp_event->Status & 0x08) == 8) ? "3 " : " ", temp_event->SensorNum); mpr_dprint(sc, MPR_FAULT, "Current Temp in Celsius: " "%d\n", temp_event->CurrentTemperature); } break; } case MPI2_EVENT_ACTIVE_CABLE_EXCEPTION: { pMpi26EventDataActiveCableExcept_t ace_event_data; ace_event_data = (pMpi26EventDataActiveCableExcept_t)fw_event->event_data; switch(ace_event_data->ReasonCode) { case MPI26_EVENT_ACTIVE_CABLE_INSUFFICIENT_POWER: { mpr_printf(sc, "Currently a cable with " "ReceptacleID %d cannot be powered and device " "connected to this active cable will not be seen. " "This active cable requires %d mW of power.\n", ace_event_data->ReceptacleID, ace_event_data->ActiveCablePowerRequirement); break; } case MPI26_EVENT_ACTIVE_CABLE_DEGRADED: { mpr_printf(sc, "Currently a cable with " "ReceptacleID %d is not running at optimal speed " "(12 Gb/s rate)\n", ace_event_data->ReceptacleID); break; } default: break; } break; } case MPI2_EVENT_PCIE_DEVICE_STATUS_CHANGE: { pMpi26EventDataPCIeDeviceStatusChange_t pcie_status_event_data; pcie_status_event_data = (pMpi26EventDataPCIeDeviceStatusChange_t)fw_event->event_data; switch (pcie_status_event_data->ReasonCode) { case MPI26_EVENT_PCIDEV_STAT_RC_PCIE_HOT_RESET_FAILED: { mpr_printf(sc, "PCIe Host Reset failed on DevHandle " "0x%x\n", pcie_status_event_data->DevHandle); break; } default: break; } break; } case MPI2_EVENT_SAS_DEVICE_DISCOVERY_ERROR: { pMpi25EventDataSasDeviceDiscoveryError_t discovery_error_data; uint64_t sas_address; discovery_error_data = (pMpi25EventDataSasDeviceDiscoveryError_t) fw_event->event_data; sas_address = discovery_error_data->SASAddress.High; sas_address = (sas_address << 32) | discovery_error_data->SASAddress.Low; switch(discovery_error_data->ReasonCode) { case MPI25_EVENT_SAS_DISC_ERR_SMP_FAILED: { mpr_printf(sc, "SMP command failed during discovery " "for expander with SAS Address %jx and " "handle 0x%x.\n", sas_address, discovery_error_data->DevHandle); break; } case MPI25_EVENT_SAS_DISC_ERR_SMP_TIMEOUT: { mpr_printf(sc, "SMP command timed out during " "discovery for expander with SAS Address %jx and " "handle 0x%x.\n", sas_address, discovery_error_data->DevHandle); break; } default: break; } break; } case MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST: { MPI26_EVENT_DATA_PCIE_TOPOLOGY_CHANGE_LIST *data; MPI26_EVENT_PCIE_TOPO_PORT_ENTRY *port_entry; uint8_t i, link_rate; uint16_t handle; data = (MPI26_EVENT_DATA_PCIE_TOPOLOGY_CHANGE_LIST *) fw_event->event_data; mpr_mapping_pcie_topology_change_event(sc, fw_event->event_data); for (i = 0; i < data->NumEntries; i++) { port_entry = &data->PortEntry[i]; handle = le16toh(port_entry->AttachedDevHandle); link_rate = port_entry->CurrentPortInfo & MPI26_EVENT_PCIE_TOPO_PI_RATE_MASK; switch (port_entry->PortStatus) { case MPI26_EVENT_PCIE_TOPO_PS_DEV_ADDED: if (link_rate < MPI26_EVENT_PCIE_TOPO_PI_RATE_2_5) { mpr_dprint(sc, MPR_ERROR, "%s: Cannot " "add PCIe device with handle 0x%x " "with unknown link rate.\n", __func__, handle); break; } if (mprsas_add_pcie_device(sc, handle, link_rate)) { mpr_dprint(sc, MPR_ERROR, "%s: failed " "to add PCIe device with handle " "0x%x\n", __func__, handle); mprsas_prepare_remove(sassc, handle); } break; case MPI26_EVENT_PCIE_TOPO_PS_NOT_RESPONDING: mprsas_prepare_remove(sassc, handle); break; case MPI26_EVENT_PCIE_TOPO_PS_PORT_CHANGED: case MPI26_EVENT_PCIE_TOPO_PS_NO_CHANGE: case MPI26_EVENT_PCIE_TOPO_PS_DELAY_NOT_RESPONDING: default: break; } } /* * refcount was incremented for this event in * mprsas_evt_handler. Decrement it here because the event has * been processed. */ mprsas_startup_decrement(sassc); break; } case MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE: case MPI2_EVENT_SAS_BROADCAST_PRIMITIVE: default: mpr_dprint(sc, MPR_TRACE,"Unhandled event 0x%0X\n", fw_event->event); break; } mpr_dprint(sc, MPR_EVENT, "(%d)->(%s) Event Free: [%x]\n", event_count, __func__, fw_event->event); mprsas_fw_event_free(sc, fw_event); } void mprsas_firmware_event_work(void *arg, int pending) { struct mpr_fw_event_work *fw_event; struct mpr_softc *sc; sc = (struct mpr_softc *)arg; mpr_lock(sc); while ((fw_event = TAILQ_FIRST(&sc->sassc->ev_queue)) != NULL) { TAILQ_REMOVE(&sc->sassc->ev_queue, fw_event, ev_link); mprsas_fw_work(sc, fw_event); } mpr_unlock(sc); } static int mprsas_add_device(struct mpr_softc *sc, u16 handle, u8 linkrate) { char devstring[80]; struct mprsas_softc *sassc; struct mprsas_target *targ; Mpi2ConfigReply_t mpi_reply; Mpi2SasDevicePage0_t config_page; uint64_t sas_address, parent_sas_address = 0; u32 device_info, parent_devinfo = 0; unsigned int id; int ret = 1, error = 0, i; struct mprsas_lun *lun; u8 is_SATA_SSD = 0; struct mpr_command *cm; sassc = sc->sassc; mprsas_startup_increment(sassc); if (mpr_config_get_sas_device_pg0(sc, &mpi_reply, &config_page, MPI2_SAS_DEVICE_PGAD_FORM_HANDLE, handle) != 0) { mpr_dprint(sc, MPR_INFO|MPR_MAPPING|MPR_FAULT, "Error reading SAS device %#x page0, iocstatus= 0x%x\n", handle, mpi_reply.IOCStatus); error = ENXIO; goto out; } device_info = le32toh(config_page.DeviceInfo); if (((device_info & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) && (le16toh(config_page.ParentDevHandle) != 0)) { Mpi2ConfigReply_t tmp_mpi_reply; Mpi2SasDevicePage0_t parent_config_page; if (mpr_config_get_sas_device_pg0(sc, &tmp_mpi_reply, &parent_config_page, MPI2_SAS_DEVICE_PGAD_FORM_HANDLE, le16toh(config_page.ParentDevHandle)) != 0) { mpr_dprint(sc, MPR_MAPPING|MPR_FAULT, "Error reading parent SAS device %#x page0, " "iocstatus= 0x%x\n", le16toh(config_page.ParentDevHandle), tmp_mpi_reply.IOCStatus); } else { parent_sas_address = parent_config_page.SASAddress.High; parent_sas_address = (parent_sas_address << 32) | parent_config_page.SASAddress.Low; parent_devinfo = le32toh(parent_config_page.DeviceInfo); } } /* TODO Check proper endianness */ sas_address = config_page.SASAddress.High; sas_address = (sas_address << 32) | config_page.SASAddress.Low; mpr_dprint(sc, MPR_MAPPING, "Handle 0x%04x SAS Address from SAS device " "page0 = %jx\n", handle, sas_address); /* * Always get SATA Identify information because this is used to * determine if Start/Stop Unit should be sent to the drive when the * system is shutdown. */ if (device_info & MPI2_SAS_DEVICE_INFO_SATA_DEVICE) { ret = mprsas_get_sas_address_for_sata_disk(sc, &sas_address, handle, device_info, &is_SATA_SSD); if (ret) { mpr_dprint(sc, MPR_MAPPING|MPR_ERROR, "%s: failed to get disk type (SSD or HDD) for SATA " "device with handle 0x%04x\n", __func__, handle); } else { mpr_dprint(sc, MPR_MAPPING, "Handle 0x%04x SAS Address " "from SATA device = %jx\n", handle, sas_address); } } /* * use_phynum: * 1 - use the PhyNum field as a fallback to the mapping logic * 0 - never use the PhyNum field * -1 - only use the PhyNum field * * Note that using the Phy number to map a device can cause device adds * to fail if multiple enclosures/expanders are in the topology. For * example, if two devices are in the same slot number in two different * enclosures within the topology, only one of those devices will be * added. PhyNum mapping should not be used if multiple enclosures are * in the topology. */ id = MPR_MAP_BAD_ID; if (sc->use_phynum != -1) id = mpr_mapping_get_tid(sc, sas_address, handle); if (id == MPR_MAP_BAD_ID) { if ((sc->use_phynum == 0) || ((id = config_page.PhyNum) > sassc->maxtargets)) { mpr_dprint(sc, MPR_INFO, "failure at %s:%d/%s()! " "Could not get ID for device with handle 0x%04x\n", __FILE__, __LINE__, __func__, handle); error = ENXIO; goto out; } } mpr_dprint(sc, MPR_MAPPING, "%s: Target ID for added device is %d.\n", __func__, id); /* * Only do the ID check and reuse check if the target is not from a * RAID Component. For Physical Disks of a Volume, the ID will be reused * when a volume is deleted because the mapping entry for the PD will * still be in the mapping table. The ID check should not be done here * either since this PD is already being used. */ targ = &sassc->targets[id]; if (!(targ->flags & MPR_TARGET_FLAGS_RAID_COMPONENT)) { if (mprsas_check_id(sassc, id) != 0) { mpr_dprint(sc, MPR_MAPPING|MPR_INFO, "Excluding target id %d\n", id); error = ENXIO; goto out; } if (targ->handle != 0x0) { mpr_dprint(sc, MPR_MAPPING, "Attempting to reuse " "target id %d handle 0x%04x\n", id, targ->handle); error = ENXIO; goto out; } } targ->devinfo = device_info; targ->devname = le32toh(config_page.DeviceName.High); targ->devname = (targ->devname << 32) | le32toh(config_page.DeviceName.Low); targ->encl_handle = le16toh(config_page.EnclosureHandle); targ->encl_slot = le16toh(config_page.Slot); targ->encl_level = config_page.EnclosureLevel; targ->connector_name[0] = config_page.ConnectorName[0]; targ->connector_name[1] = config_page.ConnectorName[1]; targ->connector_name[2] = config_page.ConnectorName[2]; targ->connector_name[3] = config_page.ConnectorName[3]; targ->handle = handle; targ->parent_handle = le16toh(config_page.ParentDevHandle); targ->sasaddr = mpr_to_u64(&config_page.SASAddress); targ->parent_sasaddr = le64toh(parent_sas_address); targ->parent_devinfo = parent_devinfo; targ->tid = id; targ->linkrate = (linkrate>>4); targ->flags = 0; if (is_SATA_SSD) { targ->flags = MPR_TARGET_IS_SATA_SSD; } if ((le16toh(config_page.Flags) & MPI25_SAS_DEVICE0_FLAGS_ENABLED_FAST_PATH) && (le16toh(config_page.Flags) & MPI25_SAS_DEVICE0_FLAGS_FAST_PATH_CAPABLE)) { targ->scsi_req_desc_type = MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO; } if (le16toh(config_page.Flags) & MPI2_SAS_DEVICE0_FLAGS_ENCL_LEVEL_VALID) { targ->encl_level_valid = TRUE; } TAILQ_INIT(&targ->commands); TAILQ_INIT(&targ->timedout_commands); while (!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPR); } SLIST_INIT(&targ->luns); mpr_describe_devinfo(targ->devinfo, devstring, 80); mpr_dprint(sc, (MPR_INFO|MPR_MAPPING), "Found device <%s> <%s> " "handle<0x%04x> enclosureHandle<0x%04x> slot %d\n", devstring, mpr_describe_table(mpr_linkrate_names, targ->linkrate), targ->handle, targ->encl_handle, targ->encl_slot); if (targ->encl_level_valid) { mpr_dprint(sc, (MPR_INFO|MPR_MAPPING), "At enclosure level %d " "and connector name (%4s)\n", targ->encl_level, targ->connector_name); } #if ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000039)) || \ (__FreeBSD_version < 902502) if ((sassc->flags & MPRSAS_IN_STARTUP) == 0) #endif mprsas_rescan_target(sc, targ); mpr_dprint(sc, MPR_MAPPING, "Target id 0x%x added\n", targ->tid); /* * Check all commands to see if the SATA_ID_TIMEOUT flag has been set. * If so, send a Target Reset TM to the target that was just created. * An Abort Task TM should be used instead of a Target Reset, but that * would be much more difficult because targets have not been fully * discovered yet, and LUN's haven't been setup. So, just reset the * target instead of the LUN. */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_flags & MPR_CM_FLAGS_SATA_ID_TIMEOUT) { targ->timeouts++; - cm->cm_state = MPR_CM_STATE_TIMEDOUT; + cm->cm_flags |= MPR_CM_FLAGS_TIMEDOUT; if ((targ->tm = mprsas_alloc_tm(sc)) != NULL) { mpr_dprint(sc, MPR_INFO, "%s: sending Target " "Reset for stuck SATA identify command " "(cm = %p)\n", __func__, cm); targ->tm->cm_targ = targ; mprsas_send_reset(sc, targ->tm, MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET); } else { mpr_dprint(sc, MPR_ERROR, "Failed to allocate " "tm for Target Reset after SATA ID command " "timed out (cm %p)\n", cm); } /* * No need to check for more since the target is * already being reset. */ break; } } out: /* * Free the commands that may not have been freed from the SATA ID call */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_flags & MPR_CM_FLAGS_SATA_ID_TIMEOUT) { free(cm->cm_data, M_MPR); mpr_free_command(sc, cm); } } mprsas_startup_decrement(sassc); return (error); } int mprsas_get_sas_address_for_sata_disk(struct mpr_softc *sc, u64 *sas_address, u16 handle, u32 device_info, u8 *is_SATA_SSD) { Mpi2SataPassthroughReply_t mpi_reply; int i, rc, try_count; u32 *bufferptr; union _sata_sas_address hash_address; struct _ata_identify_device_data ata_identify; u8 buffer[MPT2SAS_MN_LEN + MPT2SAS_SN_LEN]; u32 ioc_status; u8 sas_status; memset(&ata_identify, 0, sizeof(ata_identify)); memset(&mpi_reply, 0, sizeof(mpi_reply)); try_count = 0; do { rc = mprsas_get_sata_identify(sc, handle, &mpi_reply, (char *)&ata_identify, sizeof(ata_identify), device_info); try_count++; ioc_status = le16toh(mpi_reply.IOCStatus) & MPI2_IOCSTATUS_MASK; sas_status = mpi_reply.SASStatus; switch (ioc_status) { case MPI2_IOCSTATUS_SUCCESS: break; case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR: /* No sense sleeping. this error won't get better */ break; default: if (sc->spinup_wait_time > 0) { mpr_dprint(sc, MPR_INFO, "Sleeping %d seconds " "after SATA ID error to wait for spinup\n", sc->spinup_wait_time); msleep(&sc->msleep_fake_chan, &sc->mpr_mtx, 0, "mprid", sc->spinup_wait_time * hz); } } } while (((rc && (rc != EWOULDBLOCK)) || (ioc_status && (ioc_status != MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR)) || sas_status) && (try_count < 5)); if (rc == 0 && !ioc_status && !sas_status) { mpr_dprint(sc, MPR_MAPPING, "%s: got SATA identify " "successfully for handle = 0x%x with try_count = %d\n", __func__, handle, try_count); } else { mpr_dprint(sc, MPR_MAPPING, "%s: handle = 0x%x failed\n", __func__, handle); return -1; } /* Copy & byteswap the 40 byte model number to a buffer */ for (i = 0; i < MPT2SAS_MN_LEN; i += 2) { buffer[i] = ((u8 *)ata_identify.model_number)[i + 1]; buffer[i + 1] = ((u8 *)ata_identify.model_number)[i]; } /* Copy & byteswap the 20 byte serial number to a buffer */ for (i = 0; i < MPT2SAS_SN_LEN; i += 2) { buffer[MPT2SAS_MN_LEN + i] = ((u8 *)ata_identify.serial_number)[i + 1]; buffer[MPT2SAS_MN_LEN + i + 1] = ((u8 *)ata_identify.serial_number)[i]; } bufferptr = (u32 *)buffer; /* There are 60 bytes to hash down to 8. 60 isn't divisible by 8, * so loop through the first 56 bytes (7*8), * and then add in the last dword. */ hash_address.word.low = 0; hash_address.word.high = 0; for (i = 0; (i < ((MPT2SAS_MN_LEN+MPT2SAS_SN_LEN)/8)); i++) { hash_address.word.low += *bufferptr; bufferptr++; hash_address.word.high += *bufferptr; bufferptr++; } /* Add the last dword */ hash_address.word.low += *bufferptr; /* Make sure the hash doesn't start with 5, because it could clash * with a SAS address. Change 5 to a D. */ if ((hash_address.word.high & 0x000000F0) == (0x00000050)) hash_address.word.high |= 0x00000080; *sas_address = (u64)hash_address.wwid[0] << 56 | (u64)hash_address.wwid[1] << 48 | (u64)hash_address.wwid[2] << 40 | (u64)hash_address.wwid[3] << 32 | (u64)hash_address.wwid[4] << 24 | (u64)hash_address.wwid[5] << 16 | (u64)hash_address.wwid[6] << 8 | (u64)hash_address.wwid[7]; if (ata_identify.rotational_speed == 1) { *is_SATA_SSD = 1; } return 0; } static int mprsas_get_sata_identify(struct mpr_softc *sc, u16 handle, Mpi2SataPassthroughReply_t *mpi_reply, char *id_buffer, int sz, u32 devinfo) { Mpi2SataPassthroughRequest_t *mpi_request; Mpi2SataPassthroughReply_t *reply; struct mpr_command *cm; char *buffer; int error = 0; buffer = malloc( sz, M_MPR, M_NOWAIT | M_ZERO); if (!buffer) return ENOMEM; if ((cm = mpr_alloc_command(sc)) == NULL) { free(buffer, M_MPR); return (EBUSY); } mpi_request = (MPI2_SATA_PASSTHROUGH_REQUEST *)cm->cm_req; bzero(mpi_request,sizeof(MPI2_SATA_PASSTHROUGH_REQUEST)); mpi_request->Function = MPI2_FUNCTION_SATA_PASSTHROUGH; mpi_request->VF_ID = 0; mpi_request->DevHandle = htole16(handle); mpi_request->PassthroughFlags = (MPI2_SATA_PT_REQ_PT_FLAGS_PIO | MPI2_SATA_PT_REQ_PT_FLAGS_READ); mpi_request->DataLength = htole32(sz); mpi_request->CommandFIS[0] = 0x27; mpi_request->CommandFIS[1] = 0x80; mpi_request->CommandFIS[2] = (devinfo & MPI2_SAS_DEVICE_INFO_ATAPI_DEVICE) ? 0xA1 : 0xEC; cm->cm_sge = &mpi_request->SGL; cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION); cm->cm_flags = MPR_CM_FLAGS_DATAIN; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = buffer; cm->cm_length = htole32(sz); /* * Use a custom handler to avoid reinit'ing the controller on timeout. * This fixes a problem where the FW does not send a reply sometimes * when a bad disk is in the topology. So, this is used to timeout the * command so that processing can continue normally. */ cm->cm_timeout_handler = mprsas_ata_id_timeout; error = mpr_wait_command(sc, &cm, MPR_ATA_ID_TIMEOUT, CAN_SLEEP); /* mprsas_ata_id_timeout does not reset controller */ KASSERT(cm != NULL, ("%s: surprise command freed", __func__)); reply = (Mpi2SataPassthroughReply_t *)cm->cm_reply; if (error || (reply == NULL)) { /* FIXME */ /* * If the request returns an error then we need to do a diag * reset */ mpr_dprint(sc, MPR_INFO|MPR_FAULT|MPR_MAPPING, "Request for SATA PASSTHROUGH page completed with error %d\n", error); error = ENXIO; goto out; } bcopy(buffer, id_buffer, sz); bcopy(reply, mpi_reply, sizeof(Mpi2SataPassthroughReply_t)); if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mpr_dprint(sc, MPR_INFO|MPR_MAPPING|MPR_FAULT, "Error reading device %#x SATA PASSTHRU; iocstatus= 0x%x\n", handle, reply->IOCStatus); error = ENXIO; goto out; } out: /* * If the SATA_ID_TIMEOUT flag has been set for this command, don't free * it. The command and buffer will be freed after sending an Abort * Task TM. */ if ((cm->cm_flags & MPR_CM_FLAGS_SATA_ID_TIMEOUT) == 0) { mpr_free_command(sc, cm); free(buffer, M_MPR); } return (error); } static void mprsas_ata_id_timeout(struct mpr_softc *sc, struct mpr_command *cm) { mpr_dprint(sc, MPR_INFO, "%s ATA ID command timeout cm %p sc %p\n", __func__, cm, sc); /* * The Abort Task cannot be sent from here because the driver has not * completed setting up targets. Instead, the command is flagged so - * that special handling will be used to send the abort. + * that special handling will be used to send the abort. Now that + * this command has timed out, it's no longer in the queue. */ cm->cm_flags |= MPR_CM_FLAGS_SATA_ID_TIMEOUT; + cm->cm_state = MPR_CM_STATE_BUSY; } static int mprsas_add_pcie_device(struct mpr_softc *sc, u16 handle, u8 linkrate) { char devstring[80]; struct mprsas_softc *sassc; struct mprsas_target *targ; Mpi2ConfigReply_t mpi_reply; Mpi26PCIeDevicePage0_t config_page; Mpi26PCIeDevicePage2_t config_page2; uint64_t pcie_wwid, parent_wwid = 0; u32 device_info, parent_devinfo = 0; unsigned int id; int error = 0; struct mprsas_lun *lun; sassc = sc->sassc; mprsas_startup_increment(sassc); if ((mpr_config_get_pcie_device_pg0(sc, &mpi_reply, &config_page, MPI26_PCIE_DEVICE_PGAD_FORM_HANDLE, handle))) { printf("%s: error reading PCIe device page0\n", __func__); error = ENXIO; goto out; } device_info = le32toh(config_page.DeviceInfo); if (((device_info & MPI26_PCIE_DEVINFO_PCI_SWITCH) == 0) && (le16toh(config_page.ParentDevHandle) != 0)) { Mpi2ConfigReply_t tmp_mpi_reply; Mpi26PCIeDevicePage0_t parent_config_page; if ((mpr_config_get_pcie_device_pg0(sc, &tmp_mpi_reply, &parent_config_page, MPI26_PCIE_DEVICE_PGAD_FORM_HANDLE, le16toh(config_page.ParentDevHandle)))) { printf("%s: error reading PCIe device %#x page0\n", __func__, le16toh(config_page.ParentDevHandle)); } else { parent_wwid = parent_config_page.WWID.High; parent_wwid = (parent_wwid << 32) | parent_config_page.WWID.Low; parent_devinfo = le32toh(parent_config_page.DeviceInfo); } } /* TODO Check proper endianness */ pcie_wwid = config_page.WWID.High; pcie_wwid = (pcie_wwid << 32) | config_page.WWID.Low; mpr_dprint(sc, MPR_INFO, "PCIe WWID from PCIe device page0 = %jx\n", pcie_wwid); if ((mpr_config_get_pcie_device_pg2(sc, &mpi_reply, &config_page2, MPI26_PCIE_DEVICE_PGAD_FORM_HANDLE, handle))) { printf("%s: error reading PCIe device page2\n", __func__); error = ENXIO; goto out; } id = mpr_mapping_get_tid(sc, pcie_wwid, handle); if (id == MPR_MAP_BAD_ID) { mpr_dprint(sc, MPR_ERROR | MPR_INFO, "failure at %s:%d/%s()! " "Could not get ID for device with handle 0x%04x\n", __FILE__, __LINE__, __func__, handle); error = ENXIO; goto out; } mpr_dprint(sc, MPR_MAPPING, "%s: Target ID for added device is %d.\n", __func__, id); if (mprsas_check_id(sassc, id) != 0) { mpr_dprint(sc, MPR_MAPPING|MPR_INFO, "Excluding target id %d\n", id); error = ENXIO; goto out; } mpr_dprint(sc, MPR_MAPPING, "WWID from PCIe device page0 = %jx\n", pcie_wwid); targ = &sassc->targets[id]; targ->devinfo = device_info; targ->encl_handle = le16toh(config_page.EnclosureHandle); targ->encl_slot = le16toh(config_page.Slot); targ->encl_level = config_page.EnclosureLevel; targ->connector_name[0] = ((char *)&config_page.ConnectorName)[0]; targ->connector_name[1] = ((char *)&config_page.ConnectorName)[1]; targ->connector_name[2] = ((char *)&config_page.ConnectorName)[2]; targ->connector_name[3] = ((char *)&config_page.ConnectorName)[3]; targ->is_nvme = device_info & MPI26_PCIE_DEVINFO_NVME; targ->MDTS = config_page2.MaximumDataTransferSize; if (targ->is_nvme) targ->controller_reset_timeout = config_page2.ControllerResetTO; /* * Assume always TRUE for encl_level_valid because there is no valid * flag for PCIe. */ targ->encl_level_valid = TRUE; targ->handle = handle; targ->parent_handle = le16toh(config_page.ParentDevHandle); targ->sasaddr = mpr_to_u64(&config_page.WWID); targ->parent_sasaddr = le64toh(parent_wwid); targ->parent_devinfo = parent_devinfo; targ->tid = id; targ->linkrate = linkrate; targ->flags = 0; if ((le16toh(config_page.Flags) & MPI26_PCIEDEV0_FLAGS_ENABLED_FAST_PATH) && (le16toh(config_page.Flags) & MPI26_PCIEDEV0_FLAGS_FAST_PATH_CAPABLE)) { targ->scsi_req_desc_type = MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO; } TAILQ_INIT(&targ->commands); TAILQ_INIT(&targ->timedout_commands); while (!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPR); } SLIST_INIT(&targ->luns); mpr_describe_devinfo(targ->devinfo, devstring, 80); mpr_dprint(sc, (MPR_INFO|MPR_MAPPING), "Found PCIe device <%s> <%s> " "handle<0x%04x> enclosureHandle<0x%04x> slot %d\n", devstring, mpr_describe_table(mpr_pcie_linkrate_names, targ->linkrate), targ->handle, targ->encl_handle, targ->encl_slot); if (targ->encl_level_valid) { mpr_dprint(sc, (MPR_INFO|MPR_MAPPING), "At enclosure level %d " "and connector name (%4s)\n", targ->encl_level, targ->connector_name); } #if ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000039)) || \ (__FreeBSD_version < 902502) if ((sassc->flags & MPRSAS_IN_STARTUP) == 0) #endif mprsas_rescan_target(sc, targ); mpr_dprint(sc, MPR_MAPPING, "Target id 0x%x added\n", targ->tid); out: mprsas_startup_decrement(sassc); return (error); } static int mprsas_volume_add(struct mpr_softc *sc, u16 handle) { struct mprsas_softc *sassc; struct mprsas_target *targ; u64 wwid; unsigned int id; int error = 0; struct mprsas_lun *lun; sassc = sc->sassc; mprsas_startup_increment(sassc); /* wwid is endian safe */ mpr_config_get_volume_wwid(sc, handle, &wwid); if (!wwid) { printf("%s: invalid WWID; cannot add volume to mapping table\n", __func__); error = ENXIO; goto out; } id = mpr_mapping_get_raid_tid(sc, wwid, handle); if (id == MPR_MAP_BAD_ID) { printf("%s: could not get ID for volume with handle 0x%04x and " "WWID 0x%016llx\n", __func__, handle, (unsigned long long)wwid); error = ENXIO; goto out; } targ = &sassc->targets[id]; targ->tid = id; targ->handle = handle; targ->devname = wwid; TAILQ_INIT(&targ->commands); TAILQ_INIT(&targ->timedout_commands); while (!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPR); } SLIST_INIT(&targ->luns); #if ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000039)) || \ (__FreeBSD_version < 902502) if ((sassc->flags & MPRSAS_IN_STARTUP) == 0) #endif mprsas_rescan_target(sc, targ); mpr_dprint(sc, MPR_MAPPING, "RAID target id %d added (WWID = 0x%jx)\n", targ->tid, wwid); out: mprsas_startup_decrement(sassc); return (error); } /** * mprsas_SSU_to_SATA_devices * @sc: per adapter object * * Looks through the target list and issues a StartStopUnit SCSI command to each * SATA direct-access device. This helps to ensure that data corruption is * avoided when the system is being shut down. This must be called after the IR * System Shutdown RAID Action is sent if in IR mode. * * Return nothing. */ static void mprsas_SSU_to_SATA_devices(struct mpr_softc *sc, int howto) { struct mprsas_softc *sassc = sc->sassc; union ccb *ccb; path_id_t pathid = cam_sim_path(sassc->sim); target_id_t targetid; struct mprsas_target *target; char path_str[64]; int timeout; mpr_lock(sc); /* * For each target, issue a StartStopUnit command to stop the device. */ sc->SSU_started = TRUE; sc->SSU_refcount = 0; for (targetid = 0; targetid < sc->max_devices; targetid++) { target = &sassc->targets[targetid]; if (target->handle == 0x0) { continue; } /* * The stop_at_shutdown flag will be set if this device is * a SATA direct-access end device. */ if (target->stop_at_shutdown) { ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mpr_dprint(sc, MPR_FAULT, "Unable to alloc CCB " "to stop unit.\n"); return; } if (xpt_create_path(&ccb->ccb_h.path, xpt_periph, pathid, targetid, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { mpr_dprint(sc, MPR_ERROR, "Unable to create " "path to stop unit.\n"); xpt_free_ccb(ccb); return; } xpt_path_string(ccb->ccb_h.path, path_str, sizeof(path_str)); mpr_dprint(sc, MPR_INFO, "Sending StopUnit: path %s " "handle %d\n", path_str, target->handle); /* * Issue a START STOP UNIT command for the target. * Increment the SSU counter to be used to count the * number of required replies. */ mpr_dprint(sc, MPR_INFO, "Incrementing SSU count\n"); sc->SSU_refcount++; ccb->ccb_h.target_id = xpt_path_target_id(ccb->ccb_h.path); ccb->ccb_h.ppriv_ptr1 = sassc; scsi_start_stop(&ccb->csio, /*retries*/0, mprsas_stop_unit_done, MSG_SIMPLE_Q_TAG, /*start*/FALSE, /*load/eject*/0, /*immediate*/FALSE, MPR_SENSE_LEN, /*timeout*/10000); xpt_action(ccb); } } mpr_unlock(sc); /* * Timeout after 60 seconds by default or 10 seconds if howto has * RB_NOSYNC set which indicates we're likely handling a panic. */ timeout = 600; if (howto & RB_NOSYNC) timeout = 100; /* * Wait until all of the SSU commands have completed or time * has expired. Pause for 100ms each time through. If any * command times out, the target will be reset in the SCSI * command timeout routine. */ while (sc->SSU_refcount > 0) { pause("mprwait", hz/10); if (SCHEDULER_STOPPED()) xpt_sim_poll(sassc->sim); if (--timeout == 0) { mpr_dprint(sc, MPR_ERROR, "Time has expired waiting " "for SSU commands to complete.\n"); break; } } } static void mprsas_stop_unit_done(struct cam_periph *periph, union ccb *done_ccb) { struct mprsas_softc *sassc; char path_str[64]; if (done_ccb == NULL) return; sassc = (struct mprsas_softc *)done_ccb->ccb_h.ppriv_ptr1; xpt_path_string(done_ccb->ccb_h.path, path_str, sizeof(path_str)); mpr_dprint(sassc->sc, MPR_INFO, "Completing stop unit for %s\n", path_str); /* * Nothing more to do except free the CCB and path. If the command * timed out, an abort reset, then target reset will be issued during * the SCSI Command process. */ xpt_free_path(done_ccb->ccb_h.path); xpt_free_ccb(done_ccb); } /** * mprsas_ir_shutdown - IR shutdown notification * @sc: per adapter object * * Sending RAID Action to alert the Integrated RAID subsystem of the IOC that * the host system is shutting down. * * Return nothing. */ void mprsas_ir_shutdown(struct mpr_softc *sc, int howto) { u16 volume_mapping_flags; u16 ioc_pg8_flags = le16toh(sc->ioc_pg8.Flags); struct dev_mapping_table *mt_entry; u32 start_idx, end_idx; unsigned int id, found_volume = 0; struct mpr_command *cm; Mpi2RaidActionRequest_t *action; target_id_t targetid; struct mprsas_target *target; mpr_dprint(sc, MPR_TRACE, "%s\n", __func__); /* is IR firmware build loaded? */ if (!sc->ir_firmware) goto out; /* are there any volumes? Look at IR target IDs. */ // TODO-later, this should be looked up in the RAID config structure // when it is implemented. volume_mapping_flags = le16toh(sc->ioc_pg8.IRVolumeMappingFlags) & MPI2_IOCPAGE8_IRFLAGS_MASK_VOLUME_MAPPING_MODE; if (volume_mapping_flags == MPI2_IOCPAGE8_IRFLAGS_LOW_VOLUME_MAPPING) { start_idx = 0; if (ioc_pg8_flags & MPI2_IOCPAGE8_FLAGS_RESERVED_TARGETID_0) start_idx = 1; } else start_idx = sc->max_devices - sc->max_volumes; end_idx = start_idx + sc->max_volumes - 1; for (id = start_idx; id < end_idx; id++) { mt_entry = &sc->mapping_table[id]; if ((mt_entry->physical_id != 0) && (mt_entry->missing_count == 0)) { found_volume = 1; break; } } if (!found_volume) goto out; if ((cm = mpr_alloc_command(sc)) == NULL) { printf("%s: command alloc failed\n", __func__); goto out; } action = (MPI2_RAID_ACTION_REQUEST *)cm->cm_req; action->Function = MPI2_FUNCTION_RAID_ACTION; action->Action = MPI2_RAID_ACTION_SYSTEM_SHUTDOWN_INITIATED; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; mpr_lock(sc); mpr_wait_command(sc, &cm, 5, CAN_SLEEP); mpr_unlock(sc); /* * Don't check for reply, just leave. */ if (cm) mpr_free_command(sc, cm); out: /* * All of the targets must have the correct value set for * 'stop_at_shutdown' for the current 'enable_ssu' sysctl variable. * * The possible values for the 'enable_ssu' variable are: * 0: disable to SSD and HDD * 1: disable only to HDD (default) * 2: disable only to SSD * 3: enable to SSD and HDD * anything else will default to 1. */ for (targetid = 0; targetid < sc->max_devices; targetid++) { target = &sc->sassc->targets[targetid]; if (target->handle == 0x0) { continue; } if (target->supports_SSU) { switch (sc->enable_ssu) { case MPR_SSU_DISABLE_SSD_DISABLE_HDD: target->stop_at_shutdown = FALSE; break; case MPR_SSU_DISABLE_SSD_ENABLE_HDD: target->stop_at_shutdown = TRUE; if (target->flags & MPR_TARGET_IS_SATA_SSD) { target->stop_at_shutdown = FALSE; } break; case MPR_SSU_ENABLE_SSD_ENABLE_HDD: target->stop_at_shutdown = TRUE; break; case MPR_SSU_ENABLE_SSD_DISABLE_HDD: default: target->stop_at_shutdown = TRUE; if ((target->flags & MPR_TARGET_IS_SATA_SSD) == 0) { target->stop_at_shutdown = FALSE; } break; } } } mprsas_SSU_to_SATA_devices(sc, howto); } Index: head/sys/dev/mpr/mprvar.h =================================================================== --- head/sys/dev/mpr/mprvar.h (revision 349848) +++ head/sys/dev/mpr/mprvar.h (revision 349849) @@ -1,989 +1,990 @@ /*- * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2016 Avago Technologies * Copyright 2000-2020 Broadcom Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Broadcom Inc. (LSI) MPT-Fusion Host Adapter FreeBSD * * $FreeBSD$ */ #ifndef _MPRVAR_H #define _MPRVAR_H #include #include #define MPR_DRIVER_VERSION "23.00.00.00-fbsd" #define MPR_DB_MAX_WAIT 2500 #define MPR_REQ_FRAMES 2048 #define MPR_PRI_REQ_FRAMES 128 #define MPR_EVT_REPLY_FRAMES 32 #define MPR_REPLY_FRAMES MPR_REQ_FRAMES #define MPR_CHAIN_FRAMES 16384 #define MPR_MAXIO_PAGES (-1) #define MPR_SENSE_LEN SSD_FULL_SIZE #define MPR_MSI_MAX 1 #define MPR_MSIX_MAX 96 #define MPR_SGE64_SIZE 12 #define MPR_SGE32_SIZE 8 #define MPR_SGC_SIZE 8 #define MPR_DEFAULT_CHAIN_SEG_SIZE 8 #define MPR_MAX_CHAIN_ELEMENT_SIZE 16 /* * PCIe NVMe Specific defines */ //SLM-for now just use the same value as a SAS disk #define NVME_QDEPTH MPR_REQ_FRAMES #define PRP_ENTRY_SIZE 8 #define NVME_CMD_PRP1_OFFSET 24 /* PRP1 offset in NVMe cmd */ #define NVME_CMD_PRP2_OFFSET 32 /* PRP2 offset in NVMe cmd */ #define NVME_ERROR_RESPONSE_SIZE 16 /* Max NVME Error Response */ #define HOST_PAGE_SIZE_4K 12 #define MPR_FUNCTRACE(sc) \ mpr_dprint((sc), MPR_TRACE, "%s\n", __func__) #define CAN_SLEEP 1 #define NO_SLEEP 0 #define MPR_PERIODIC_DELAY 1 /* 1 second heartbeat/watchdog check */ #define MPR_ATA_ID_TIMEOUT 5 /* 5 second timeout for SATA ID cmd */ #define MPR_MISSING_CHECK_DELAY 10 /* 10 seconds between missing check */ #define IFAULT_IOP_OVER_TEMP_THRESHOLD_EXCEEDED 0x2810 #define MPR_SCSI_RI_INVALID_FRAME (0x00000002) #define DEFAULT_SPINUP_WAIT 3 /* seconds to wait for spinup */ #include /* * host mapping related macro definitions */ #define MPR_MAPTABLE_BAD_IDX 0xFFFFFFFF #define MPR_DPM_BAD_IDX 0xFFFF #define MPR_ENCTABLE_BAD_IDX 0xFF #define MPR_MAX_MISSING_COUNT 0x0F #define MPR_DEV_RESERVED 0x20000000 #define MPR_MAP_IN_USE 0x10000000 #define MPR_MAP_BAD_ID 0xFFFFFFFF typedef uint8_t u8; typedef uint16_t u16; typedef uint32_t u32; typedef uint64_t u64; typedef struct _MPI2_CONFIG_PAGE_MAN_11 { MPI2_CONFIG_PAGE_HEADER Header; /* 0x00 */ U8 FlashTime; /* 0x04 */ U8 NVTime; /* 0x05 */ U16 Flag; /* 0x06 */ U8 RFIoTimeout; /* 0x08 */ U8 EEDPTagMode; /* 0x09 */ U8 AWTValue; /* 0x0A */ U8 Reserve1; /* 0x0B */ U8 MaxCmdFrames; /* 0x0C */ U8 Reserve2; /* 0x0D */ U16 AddlFlags; /* 0x0E */ U32 SysRefClk; /* 0x10 */ U64 Reserve3[3]; /* 0x14 */ U16 AddlFlags2; /* 0x2C */ U8 AddlFlags3; /* 0x2E */ U8 Reserve4; /* 0x2F */ U64 opDebugEnable; /* 0x30 */ U64 PlDebugEnable; /* 0x38 */ U64 IrDebugEnable; /* 0x40 */ U32 BoardPowerRequirement; /* 0x48 */ U8 NVMeAbortTO; /* 0x4C */ U8 Reserve5; /* 0x4D */ U16 Reserve6; /* 0x4E */ U32 Reserve7[3]; /* 0x50 */ } MPI2_CONFIG_PAGE_MAN_11, MPI2_POINTER PTR_MPI2_CONFIG_PAGE_MAN_11, Mpi2ManufacturingPage11_t, MPI2_POINTER pMpi2ManufacturingPage11_t; #define MPI2_MAN_PG11_ADDLFLAGS2_CUSTOM_TM_HANDLING_MASK (0x0010) /** * struct dev_mapping_table - device mapping information * @physical_id: SAS address for drives or WWID for RAID volumes * @device_info: bitfield provides detailed info about the device * @phy_bits: bitfields indicating controller phys * @dpm_entry_num: index of this device in device persistent map table * @dev_handle: device handle for the device pointed by this entry * @id: target id * @missing_count: number of times the device not detected by driver * @hide_flag: Hide this physical disk/not (foreign configuration) * @init_complete: Whether the start of the day checks completed or not * @TLR_bits: Turn TLR support on or off */ struct dev_mapping_table { u64 physical_id; u32 device_info; u32 phy_bits; u16 dpm_entry_num; u16 dev_handle; u16 reserved1; u16 id; u8 missing_count; u8 init_complete; u8 TLR_bits; u8 reserved2; }; /** * struct enc_mapping_table - mapping information about an enclosure * @enclosure_id: Logical ID of this enclosure * @start_index: index to the entry in dev_mapping_table * @phy_bits: bitfields indicating controller phys * @dpm_entry_num: index of this enclosure in device persistent map table * @enc_handle: device handle for the enclosure pointed by this entry * @num_slots: number of slots in the enclosure * @start_slot: Starting slot id * @missing_count: number of times the device not detected by driver * @removal_flag: used to mark the device for removal * @skip_search: used as a flag to include/exclude enclosure for search * @init_complete: Whether the start of the day checks completed or not */ struct enc_mapping_table { u64 enclosure_id; u32 start_index; u32 phy_bits; u16 dpm_entry_num; u16 enc_handle; u16 num_slots; u16 start_slot; u8 missing_count; u8 removal_flag; u8 skip_search; u8 init_complete; }; /** * struct map_removal_table - entries to be removed from mapping table * @dpm_entry_num: index of this device in device persistent map table * @dev_handle: device handle for the device pointed by this entry */ struct map_removal_table{ u16 dpm_entry_num; u16 dev_handle; }; typedef struct mpr_fw_diagnostic_buffer { size_t size; uint8_t extended_type; uint8_t buffer_type; uint8_t force_release; uint32_t product_specific[23]; uint8_t immediate; uint8_t enabled; uint8_t valid_data; uint8_t owned_by_firmware; uint32_t unique_id; } mpr_fw_diagnostic_buffer_t; struct mpr_softc; struct mpr_command; struct mprsas_softc; union ccb; struct mprsas_target; struct mpr_column_map; MALLOC_DECLARE(M_MPR); typedef void mpr_evt_callback_t(struct mpr_softc *, uintptr_t, MPI2_EVENT_NOTIFICATION_REPLY *reply); typedef void mpr_command_callback_t(struct mpr_softc *, struct mpr_command *cm); struct mpr_chain { TAILQ_ENTRY(mpr_chain) chain_link; void *chain; uint64_t chain_busaddr; }; struct mpr_prp_page { TAILQ_ENTRY(mpr_prp_page) prp_page_link; uint64_t *prp_page; uint64_t prp_page_busaddr; }; /* * This needs to be at least 2 to support SMP passthrough. */ #define MPR_IOVEC_COUNT 2 struct mpr_command { TAILQ_ENTRY(mpr_command) cm_link; TAILQ_ENTRY(mpr_command) cm_recovery; struct mpr_softc *cm_sc; union ccb *cm_ccb; void *cm_data; u_int cm_length; u_int cm_out_len; struct uio cm_uio; struct iovec cm_iovec[MPR_IOVEC_COUNT]; u_int cm_max_segs; u_int cm_sglsize; void *cm_sge; uint8_t *cm_req; uint8_t *cm_reply; uint32_t cm_reply_data; mpr_command_callback_t *cm_complete; void *cm_complete_data; struct mprsas_target *cm_targ; MPI2_REQUEST_DESCRIPTOR_UNION cm_desc; u_int cm_lun; u_int cm_flags; #define MPR_CM_FLAGS_POLLED (1 << 0) #define MPR_CM_FLAGS_COMPLETE (1 << 1) #define MPR_CM_FLAGS_SGE_SIMPLE (1 << 2) #define MPR_CM_FLAGS_DATAOUT (1 << 3) #define MPR_CM_FLAGS_DATAIN (1 << 4) #define MPR_CM_FLAGS_WAKEUP (1 << 5) #define MPR_CM_FLAGS_USE_UIO (1 << 6) #define MPR_CM_FLAGS_SMP_PASS (1 << 7) #define MPR_CM_FLAGS_CHAIN_FAILED (1 << 8) #define MPR_CM_FLAGS_ERROR_MASK MPR_CM_FLAGS_CHAIN_FAILED #define MPR_CM_FLAGS_USE_CCB (1 << 9) #define MPR_CM_FLAGS_SATA_ID_TIMEOUT (1 << 10) +#define MPR_CM_FLAGS_ON_RECOVERY (1 << 12) +#define MPR_CM_FLAGS_TIMEDOUT (1 << 13) u_int cm_state; #define MPR_CM_STATE_FREE 0 #define MPR_CM_STATE_BUSY 1 -#define MPR_CM_STATE_TIMEDOUT 2 -#define MPR_CM_STATE_INQUEUE 3 +#define MPR_CM_STATE_INQUEUE 2 bus_dmamap_t cm_dmamap; struct scsi_sense_data *cm_sense; uint64_t *nvme_error_response; TAILQ_HEAD(, mpr_chain) cm_chain_list; TAILQ_HEAD(, mpr_prp_page) cm_prp_page_list; uint32_t cm_req_busaddr; bus_addr_t cm_sense_busaddr; struct callout cm_callout; mpr_command_callback_t *cm_timeout_handler; }; struct mpr_column_map { uint16_t dev_handle; uint8_t phys_disk_num; }; struct mpr_event_handle { TAILQ_ENTRY(mpr_event_handle) eh_list; mpr_evt_callback_t *callback; void *data; uint8_t mask[16]; }; struct mpr_busdma_context { int completed; int abandoned; int error; bus_addr_t *addr; struct mpr_softc *softc; bus_dmamap_t buffer_dmamap; bus_dma_tag_t buffer_dmat; }; struct mpr_queue { struct mpr_softc *sc; int qnum; MPI2_REPLY_DESCRIPTORS_UNION *post_queue; int replypostindex; #ifdef notyet ck_ring_buffer_t *ringmem; ck_ring_buffer_t *chainmem; ck_ring_t req_ring; ck_ring_t chain_ring; #endif bus_dma_tag_t buffer_dmat; int io_cmds_highwater; int chain_free_lowwater; int chain_alloc_fail; struct resource *irq; void *intrhand; int irq_rid; }; struct mpr_softc { device_t mpr_dev; struct cdev *mpr_cdev; u_int mpr_flags; #define MPR_FLAGS_INTX (1 << 0) #define MPR_FLAGS_MSI (1 << 1) #define MPR_FLAGS_BUSY (1 << 2) #define MPR_FLAGS_SHUTDOWN (1 << 3) #define MPR_FLAGS_DIAGRESET (1 << 4) #define MPR_FLAGS_ATTACH_DONE (1 << 5) #define MPR_FLAGS_GEN35_IOC (1 << 6) #define MPR_FLAGS_REALLOCATED (1 << 7) #define MPR_FLAGS_SEA_IOC (1 << 8) u_int mpr_debug; int msi_msgs; u_int reqframesz; u_int replyframesz; u_int atomic_desc_capable; int tm_cmds_active; int io_cmds_active; int io_cmds_highwater; int chain_free; int max_chains; int max_io_pages; u_int maxio; int chain_free_lowwater; uint32_t chain_frame_size; int prp_buffer_size; int prp_pages_free; int prp_pages_free_lowwater; u_int enable_ssu; int spinup_wait_time; int use_phynum; uint64_t chain_alloc_fail; uint64_t prp_page_alloc_fail; struct sysctl_ctx_list sysctl_ctx; struct sysctl_oid *sysctl_tree; char fw_version[16]; struct mpr_command *commands; struct mpr_chain *chains; struct mpr_prp_page *prps; struct callout periodic; struct callout device_check_callout; struct mpr_queue *queues; struct mprsas_softc *sassc; TAILQ_HEAD(, mpr_command) req_list; TAILQ_HEAD(, mpr_command) high_priority_req_list; TAILQ_HEAD(, mpr_chain) chain_list; TAILQ_HEAD(, mpr_prp_page) prp_page_list; TAILQ_HEAD(, mpr_command) tm_list; int replypostindex; int replyfreeindex; struct resource *mpr_regs_resource; bus_space_handle_t mpr_bhandle; bus_space_tag_t mpr_btag; int mpr_regs_rid; bus_dma_tag_t mpr_parent_dmat; bus_dma_tag_t buffer_dmat; MPI2_IOC_FACTS_REPLY *facts; int num_reqs; int num_prireqs; int num_replies; int num_chains; int fqdepth; /* Free queue */ int pqdepth; /* Post queue */ uint8_t event_mask[16]; TAILQ_HEAD(, mpr_event_handle) event_list; struct mpr_event_handle *mpr_log_eh; struct mtx mpr_mtx; struct intr_config_hook mpr_ich; uint8_t *req_frames; bus_addr_t req_busaddr; bus_dma_tag_t req_dmat; bus_dmamap_t req_map; uint8_t *reply_frames; bus_addr_t reply_busaddr; bus_dma_tag_t reply_dmat; bus_dmamap_t reply_map; struct scsi_sense_data *sense_frames; bus_addr_t sense_busaddr; bus_dma_tag_t sense_dmat; bus_dmamap_t sense_map; uint8_t *chain_frames; bus_dma_tag_t chain_dmat; bus_dmamap_t chain_map; uint8_t *prp_pages; bus_addr_t prp_page_busaddr; bus_dma_tag_t prp_page_dmat; bus_dmamap_t prp_page_map; MPI2_REPLY_DESCRIPTORS_UNION *post_queue; bus_addr_t post_busaddr; uint32_t *free_queue; bus_addr_t free_busaddr; bus_dma_tag_t queues_dmat; bus_dmamap_t queues_map; uint8_t *fw_diag_buffer; bus_addr_t fw_diag_busaddr; bus_dma_tag_t fw_diag_dmat; bus_dmamap_t fw_diag_map; uint8_t ir_firmware; /* static config pages */ Mpi2IOCPage8_t ioc_pg8; Mpi2IOUnitPage8_t iounit_pg8; /* host mapping support */ struct dev_mapping_table *mapping_table; struct enc_mapping_table *enclosure_table; struct map_removal_table *removal_table; uint8_t *dpm_entry_used; uint8_t *dpm_flush_entry; Mpi2DriverMappingPage0_t *dpm_pg0; uint16_t max_devices; uint16_t max_enclosures; uint16_t max_expanders; uint8_t max_volumes; uint8_t num_enc_table_entries; uint8_t num_rsvd_entries; uint16_t max_dpm_entries; uint8_t is_dpm_enable; uint8_t track_mapping_events; uint32_t pending_map_events; /* FW diag Buffer List */ mpr_fw_diagnostic_buffer_t fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_COUNT]; /* Event Recording IOCTL support */ uint32_t events_to_record[4]; mpr_event_entry_t recorded_events[MPR_EVENT_QUEUE_SIZE]; uint8_t event_index; uint32_t event_number; /* EEDP and TLR support */ uint8_t eedp_enabled; uint8_t control_TLR; /* Shutdown Event Handler */ eventhandler_tag shutdown_eh; /* To track topo events during reset */ #define MPR_DIAG_RESET_TIMEOUT 300000 uint8_t wait_for_port_enable; uint8_t port_enable_complete; uint8_t msleep_fake_chan; /* StartStopUnit command handling at shutdown */ uint32_t SSU_refcount; uint8_t SSU_started; /* Configuration tunables */ u_int disable_msix; u_int disable_msi; u_int max_msix; u_int max_reqframes; u_int max_prireqframes; u_int max_replyframes; u_int max_evtframes; char exclude_ids[80]; struct timeval lastfail; uint8_t custom_nvme_tm_handling; uint8_t nvme_abort_timeout; }; struct mpr_config_params { MPI2_CONFIG_EXT_PAGE_HEADER_UNION hdr; u_int action; u_int page_address; /* Attributes, not a phys address */ u_int status; void *buffer; u_int length; int timeout; void (*callback)(struct mpr_softc *, struct mpr_config_params *); void *cbdata; }; struct scsi_read_capacity_eedp { uint8_t addr[8]; uint8_t length[4]; uint8_t protect; }; static __inline uint32_t mpr_regread(struct mpr_softc *sc, uint32_t offset) { uint32_t ret_val, i = 0; do { ret_val = bus_space_read_4(sc->mpr_btag, sc->mpr_bhandle, offset); } while((sc->mpr_flags & MPR_FLAGS_SEA_IOC) && (ret_val == 0) && (++i < 3)); return ret_val; } static __inline void mpr_regwrite(struct mpr_softc *sc, uint32_t offset, uint32_t val) { bus_space_write_4(sc->mpr_btag, sc->mpr_bhandle, offset, val); } /* free_queue must have Little Endian address * TODO- cm_reply_data is unwanted. We can remove it. * */ static __inline void mpr_free_reply(struct mpr_softc *sc, uint32_t busaddr) { if (++sc->replyfreeindex >= sc->fqdepth) sc->replyfreeindex = 0; sc->free_queue[sc->replyfreeindex] = htole32(busaddr); mpr_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); } static __inline struct mpr_chain * mpr_alloc_chain(struct mpr_softc *sc) { struct mpr_chain *chain; if ((chain = TAILQ_FIRST(&sc->chain_list)) != NULL) { TAILQ_REMOVE(&sc->chain_list, chain, chain_link); sc->chain_free--; if (sc->chain_free < sc->chain_free_lowwater) sc->chain_free_lowwater = sc->chain_free; } else sc->chain_alloc_fail++; return (chain); } static __inline void mpr_free_chain(struct mpr_softc *sc, struct mpr_chain *chain) { #if 0 bzero(chain->chain, 128); #endif sc->chain_free++; TAILQ_INSERT_TAIL(&sc->chain_list, chain, chain_link); } static __inline struct mpr_prp_page * mpr_alloc_prp_page(struct mpr_softc *sc) { struct mpr_prp_page *prp_page; if ((prp_page = TAILQ_FIRST(&sc->prp_page_list)) != NULL) { TAILQ_REMOVE(&sc->prp_page_list, prp_page, prp_page_link); sc->prp_pages_free--; if (sc->prp_pages_free < sc->prp_pages_free_lowwater) sc->prp_pages_free_lowwater = sc->prp_pages_free; } else sc->prp_page_alloc_fail++; return (prp_page); } static __inline void mpr_free_prp_page(struct mpr_softc *sc, struct mpr_prp_page *prp_page) { sc->prp_pages_free++; TAILQ_INSERT_TAIL(&sc->prp_page_list, prp_page, prp_page_link); } static __inline void mpr_free_command(struct mpr_softc *sc, struct mpr_command *cm) { struct mpr_chain *chain, *chain_temp; struct mpr_prp_page *prp_page, *prp_page_temp; KASSERT(cm->cm_state == MPR_CM_STATE_BUSY, ("state not busy\n")); if (cm->cm_reply != NULL) mpr_free_reply(sc, cm->cm_reply_data); cm->cm_reply = NULL; cm->cm_flags = 0; cm->cm_complete = NULL; cm->cm_complete_data = NULL; cm->cm_ccb = NULL; cm->cm_targ = NULL; cm->cm_max_segs = 0; cm->cm_lun = 0; cm->cm_state = MPR_CM_STATE_FREE; cm->cm_data = NULL; cm->cm_length = 0; cm->cm_out_len = 0; cm->cm_sglsize = 0; cm->cm_sge = NULL; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain_temp) { TAILQ_REMOVE(&cm->cm_chain_list, chain, chain_link); mpr_free_chain(sc, chain); } TAILQ_FOREACH_SAFE(prp_page, &cm->cm_prp_page_list, prp_page_link, prp_page_temp) { TAILQ_REMOVE(&cm->cm_prp_page_list, prp_page, prp_page_link); mpr_free_prp_page(sc, prp_page); } TAILQ_INSERT_TAIL(&sc->req_list, cm, cm_link); } static __inline struct mpr_command * mpr_alloc_command(struct mpr_softc *sc) { struct mpr_command *cm; cm = TAILQ_FIRST(&sc->req_list); if (cm == NULL) return (NULL); KASSERT(cm->cm_state == MPR_CM_STATE_FREE, ("mpr: Allocating busy command\n")); TAILQ_REMOVE(&sc->req_list, cm, cm_link); cm->cm_state = MPR_CM_STATE_BUSY; cm->cm_timeout_handler = NULL; return (cm); } static __inline void mpr_free_high_priority_command(struct mpr_softc *sc, struct mpr_command *cm) { struct mpr_chain *chain, *chain_temp; KASSERT(cm->cm_state == MPR_CM_STATE_BUSY, ("state not busy\n")); if (cm->cm_reply != NULL) mpr_free_reply(sc, cm->cm_reply_data); cm->cm_reply = NULL; cm->cm_flags = 0; cm->cm_complete = NULL; cm->cm_complete_data = NULL; cm->cm_ccb = NULL; cm->cm_targ = NULL; cm->cm_lun = 0; cm->cm_state = MPR_CM_STATE_FREE; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain_temp) { TAILQ_REMOVE(&cm->cm_chain_list, chain, chain_link); mpr_free_chain(sc, chain); } TAILQ_INSERT_TAIL(&sc->high_priority_req_list, cm, cm_link); } static __inline struct mpr_command * mpr_alloc_high_priority_command(struct mpr_softc *sc) { struct mpr_command *cm; cm = TAILQ_FIRST(&sc->high_priority_req_list); if (cm == NULL) return (NULL); KASSERT(cm->cm_state == MPR_CM_STATE_FREE, ("mpr: Allocating busy command\n")); TAILQ_REMOVE(&sc->high_priority_req_list, cm, cm_link); cm->cm_state = MPR_CM_STATE_BUSY; cm->cm_timeout_handler = NULL; cm->cm_desc.HighPriority.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_HIGH_PRIORITY; return (cm); } static __inline void mpr_lock(struct mpr_softc *sc) { mtx_lock(&sc->mpr_mtx); } static __inline void mpr_unlock(struct mpr_softc *sc) { mtx_unlock(&sc->mpr_mtx); } #define MPR_INFO (1 << 0) /* Basic info */ #define MPR_FAULT (1 << 1) /* Hardware faults */ #define MPR_EVENT (1 << 2) /* Event data from the controller */ #define MPR_LOG (1 << 3) /* Log data from the controller */ #define MPR_RECOVERY (1 << 4) /* Command error recovery tracing */ #define MPR_ERROR (1 << 5) /* Parameter errors, programming bugs */ #define MPR_INIT (1 << 6) /* Things related to system init */ #define MPR_XINFO (1 << 7) /* More detailed/noisy info */ #define MPR_USER (1 << 8) /* Trace user-generated commands */ #define MPR_MAPPING (1 << 9) /* Trace device mappings */ #define MPR_TRACE (1 << 10) /* Function-by-function trace */ #define MPR_SSU_DISABLE_SSD_DISABLE_HDD 0 #define MPR_SSU_ENABLE_SSD_DISABLE_HDD 1 #define MPR_SSU_DISABLE_SSD_ENABLE_HDD 2 #define MPR_SSU_ENABLE_SSD_ENABLE_HDD 3 #define mpr_printf(sc, args...) \ device_printf((sc)->mpr_dev, ##args) #define mpr_print_field(sc, msg, args...) \ printf("\t" msg, ##args) #define mpr_vprintf(sc, args...) \ do { \ if (bootverbose) \ mpr_printf(sc, ##args); \ } while (0) #define mpr_dprint(sc, level, msg, args...) \ do { \ if ((sc)->mpr_debug & (level)) \ device_printf((sc)->mpr_dev, msg, ##args); \ } while (0) #define MPR_PRINTFIELD_START(sc, tag...) \ mpr_printf((sc), ##tag); \ mpr_print_field((sc), ":\n") #define MPR_PRINTFIELD_END(sc, tag) \ mpr_printf((sc), tag "\n") #define MPR_PRINTFIELD(sc, facts, attr, fmt) \ mpr_print_field((sc), #attr ": " #fmt "\n", (facts)->attr) static __inline void mpr_from_u64(uint64_t data, U64 *mpr) { (mpr)->High = htole32((uint32_t)((data) >> 32)); (mpr)->Low = htole32((uint32_t)((data) & 0xffffffff)); } static __inline uint64_t mpr_to_u64(U64 *data) { return (((uint64_t)le32toh(data->High) << 32) | le32toh(data->Low)); } static __inline void mpr_mask_intr(struct mpr_softc *sc) { uint32_t mask; mask = mpr_regread(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET); mask |= MPI2_HIM_REPLY_INT_MASK; mpr_regwrite(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET, mask); } static __inline void mpr_unmask_intr(struct mpr_softc *sc) { uint32_t mask; mask = mpr_regread(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET); mask &= ~MPI2_HIM_REPLY_INT_MASK; mpr_regwrite(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET, mask); } int mpr_pci_setup_interrupts(struct mpr_softc *sc); void mpr_pci_free_interrupts(struct mpr_softc *sc); int mpr_pci_restore(struct mpr_softc *sc); void mpr_get_tunables(struct mpr_softc *sc); int mpr_attach(struct mpr_softc *sc); int mpr_free(struct mpr_softc *sc); void mpr_intr(void *); void mpr_intr_msi(void *); void mpr_intr_locked(void *); int mpr_register_events(struct mpr_softc *, uint8_t *, mpr_evt_callback_t *, void *, struct mpr_event_handle **); int mpr_restart(struct mpr_softc *); int mpr_update_events(struct mpr_softc *, struct mpr_event_handle *, uint8_t *); int mpr_deregister_events(struct mpr_softc *, struct mpr_event_handle *); void mpr_build_nvme_prp(struct mpr_softc *sc, struct mpr_command *cm, Mpi26NVMeEncapsulatedRequest_t *nvme_encap_request, void *data, uint32_t data_in_sz, uint32_t data_out_sz); int mpr_push_sge(struct mpr_command *, MPI2_SGE_SIMPLE64 *, size_t, int); int mpr_push_ieee_sge(struct mpr_command *, void *, int); int mpr_add_dmaseg(struct mpr_command *, vm_paddr_t, size_t, u_int, int); int mpr_attach_sas(struct mpr_softc *sc); int mpr_detach_sas(struct mpr_softc *sc); int mpr_read_config_page(struct mpr_softc *, struct mpr_config_params *); int mpr_write_config_page(struct mpr_softc *, struct mpr_config_params *); void mpr_memaddr_cb(void *, bus_dma_segment_t *, int , int ); void mpr_memaddr_wait_cb(void *, bus_dma_segment_t *, int , int ); void mpr_init_sge(struct mpr_command *cm, void *req, void *sge); int mpr_attach_user(struct mpr_softc *); void mpr_detach_user(struct mpr_softc *); void mprsas_record_event(struct mpr_softc *sc, MPI2_EVENT_NOTIFICATION_REPLY *event_reply); int mpr_map_command(struct mpr_softc *sc, struct mpr_command *cm); int mpr_wait_command(struct mpr_softc *sc, struct mpr_command **cm, int timeout, int sleep_flag); int mpr_request_polled(struct mpr_softc *sc, struct mpr_command **cm); int mpr_config_get_bios_pg3(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2BiosPage3_t *config_page); int mpr_config_get_raid_volume_pg0(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidVolPage0_t *config_page, u32 page_address); int mpr_config_get_ioc_pg8(struct mpr_softc *sc, Mpi2ConfigReply_t *, Mpi2IOCPage8_t *); int mpr_config_get_iounit_pg8(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2IOUnitPage8_t *config_page); int mpr_config_get_sas_device_pg0(struct mpr_softc *, Mpi2ConfigReply_t *, Mpi2SasDevicePage0_t *, u32 , u16 ); int mpr_config_get_pcie_device_pg0(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi26PCIeDevicePage0_t *config_page, u32 form, u16 handle); int mpr_config_get_pcie_device_pg2(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi26PCIeDevicePage2_t *config_page, u32 form, u16 handle); int mpr_config_get_dpm_pg0(struct mpr_softc *, Mpi2ConfigReply_t *, Mpi2DriverMappingPage0_t *, u16 ); int mpr_config_get_raid_volume_pg1(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidVolPage1_t *config_page, u32 form, u16 handle); int mpr_config_get_volume_wwid(struct mpr_softc *sc, u16 volume_handle, u64 *wwid); int mpr_config_get_raid_pd_pg0(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidPhysDiskPage0_t *config_page, u32 page_address); int mpr_config_get_man_pg11(struct mpr_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2ManufacturingPage11_t *config_page); void mprsas_ir_shutdown(struct mpr_softc *sc, int howto); int mpr_reinit(struct mpr_softc *sc); void mprsas_handle_reinit(struct mpr_softc *sc); void mpr_base_static_config_pages(struct mpr_softc *sc); int mpr_mapping_initialize(struct mpr_softc *); void mpr_mapping_topology_change_event(struct mpr_softc *, Mpi2EventDataSasTopologyChangeList_t *); void mpr_mapping_pcie_topology_change_event(struct mpr_softc *sc, Mpi26EventDataPCIeTopologyChangeList_t *event_data); void mpr_mapping_free_memory(struct mpr_softc *sc); int mpr_config_set_dpm_pg0(struct mpr_softc *, Mpi2ConfigReply_t *, Mpi2DriverMappingPage0_t *, u16 ); void mpr_mapping_exit(struct mpr_softc *); void mpr_mapping_check_devices(void *); int mpr_mapping_allocate_memory(struct mpr_softc *sc); unsigned int mpr_mapping_get_tid(struct mpr_softc *, uint64_t , u16); unsigned int mpr_mapping_get_tid_from_handle(struct mpr_softc *sc, u16 handle); unsigned int mpr_mapping_get_raid_tid(struct mpr_softc *sc, u64 wwid, u16 volHandle); unsigned int mpr_mapping_get_raid_tid_from_handle(struct mpr_softc *sc, u16 volHandle); void mpr_mapping_enclosure_dev_status_change_event(struct mpr_softc *, Mpi2EventDataSasEnclDevStatusChange_t *event_data); void mpr_mapping_ir_config_change_event(struct mpr_softc *sc, Mpi2EventDataIrConfigChangeList_t *event_data); void mprsas_evt_handler(struct mpr_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event); void mprsas_prepare_remove(struct mprsas_softc *sassc, uint16_t handle); void mprsas_prepare_volume_remove(struct mprsas_softc *sassc, uint16_t handle); int mprsas_startup(struct mpr_softc *sc); struct mprsas_target * mprsas_find_target_by_handle(struct mprsas_softc *, int, uint16_t); void mprsas_realloc_targets(struct mpr_softc *sc, int maxtargets); struct mpr_command * mprsas_alloc_tm(struct mpr_softc *sc); void mprsas_free_tm(struct mpr_softc *sc, struct mpr_command *tm); void mprsas_release_simq_reinit(struct mprsas_softc *sassc); int mprsas_send_reset(struct mpr_softc *sc, struct mpr_command *tm, uint8_t type); SYSCTL_DECL(_hw_mpr); /* Compatibility shims for different OS versions */ #if __FreeBSD_version >= 800001 #define mpr_kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) \ kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) #define mpr_kproc_exit(arg) kproc_exit(arg) #else #define mpr_kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) \ kthread_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) #define mpr_kproc_exit(arg) kthread_exit(arg) #endif #if defined(CAM_PRIORITY_XPT) #define MPR_PRIORITY_XPT CAM_PRIORITY_XPT #else #define MPR_PRIORITY_XPT 5 #endif #if __FreeBSD_version < 800107 // Prior to FreeBSD-8.0 scp3_flags was not defined. #define spc3_flags reserved #define SPC3_SID_PROTECT 0x01 #define SPC3_SID_3PC 0x08 #define SPC3_SID_TPGS_MASK 0x30 #define SPC3_SID_TPGS_IMPLICIT 0x10 #define SPC3_SID_TPGS_EXPLICIT 0x20 #define SPC3_SID_ACC 0x40 #define SPC3_SID_SCCS 0x80 #define CAM_PRIORITY_NORMAL CAM_PRIORITY_NONE #endif /* Definitions for SCSI unmap translation to NVMe DSM command */ /* UNMAP block descriptor structure */ struct unmap_blk_desc { uint64_t slba; uint32_t nlb; uint32_t resv; }; /* UNMAP command's data */ struct unmap_parm_list { uint16_t unmap_data_len; uint16_t unmap_blk_desc_data_len; uint32_t resv; struct unmap_blk_desc desc[0]; }; /* SCSI ADDITIONAL SENSE Codes */ #define FIXED_SENSE_DATA 0x70 #define SCSI_ASC_NO_SENSE 0x00 #define SCSI_ASC_PERIPHERAL_DEV_WRITE_FAULT 0x03 #define SCSI_ASC_LUN_NOT_READY 0x04 #define SCSI_ASC_WARNING 0x0B #define SCSI_ASC_LOG_BLOCK_GUARD_CHECK_FAILED 0x10 #define SCSI_ASC_LOG_BLOCK_APPTAG_CHECK_FAILED 0x10 #define SCSI_ASC_LOG_BLOCK_REFTAG_CHECK_FAILED 0x10 #define SCSI_ASC_UNRECOVERED_READ_ERROR 0x11 #define SCSI_ASC_MISCOMPARE_DURING_VERIFY 0x1D #define SCSI_ASC_ACCESS_DENIED_INVALID_LUN_ID 0x20 #define SCSI_ASC_ILLEGAL_COMMAND 0x20 #define SCSI_ASC_ILLEGAL_BLOCK 0x21 #define SCSI_ASC_INVALID_CDB 0x24 #define SCSI_ASC_INVALID_LUN 0x25 #define SCSI_ASC_INVALID_PARAMETER 0x26 #define SCSI_ASC_FORMAT_COMMAND_FAILED 0x31 #define SCSI_ASC_INTERNAL_TARGET_FAILURE 0x44 /* SCSI ADDITIONAL SENSE Code Qualifiers */ #define SCSI_ASCQ_CAUSE_NOT_REPORTABLE 0x00 #define SCSI_ASCQ_FORMAT_COMMAND_FAILED 0x01 #define SCSI_ASCQ_LOG_BLOCK_GUARD_CHECK_FAILED 0x01 #define SCSI_ASCQ_LOG_BLOCK_APPTAG_CHECK_FAILED 0x02 #define SCSI_ASCQ_LOG_BLOCK_REFTAG_CHECK_FAILED 0x03 #define SCSI_ASCQ_FORMAT_IN_PROGRESS 0x04 #define SCSI_ASCQ_POWER_LOSS_EXPECTED 0x08 #define SCSI_ASCQ_INVALID_LUN_ID 0x09 #endif Index: head/sys/dev/mps/mps.c =================================================================== --- head/sys/dev/mps/mps.c (revision 349848) +++ head/sys/dev/mps/mps.c (revision 349849) @@ -1,3256 +1,3267 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2015 Avago Technologies * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT2 */ /* TODO Move headers to mpsvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int mps_diag_reset(struct mps_softc *sc, int sleep_flag); static int mps_init_queues(struct mps_softc *sc); static void mps_resize_queues(struct mps_softc *sc); static int mps_message_unit_reset(struct mps_softc *sc, int sleep_flag); static int mps_transition_operational(struct mps_softc *sc); static int mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching); static void mps_iocfacts_free(struct mps_softc *sc); static void mps_startup(void *arg); static int mps_send_iocinit(struct mps_softc *sc); static int mps_alloc_queues(struct mps_softc *sc); static int mps_alloc_hw_queues(struct mps_softc *sc); static int mps_alloc_replies(struct mps_softc *sc); static int mps_alloc_requests(struct mps_softc *sc); static int mps_attach_log(struct mps_softc *sc); static __inline void mps_complete_command(struct mps_softc *sc, struct mps_command *cm); static void mps_dispatch_event(struct mps_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *reply); static void mps_config_complete(struct mps_softc *sc, struct mps_command *cm); static void mps_periodic(void *); static int mps_reregister_events(struct mps_softc *sc); static void mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm); static int mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts); static int mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag); static int mps_debug_sysctl(SYSCTL_HANDLER_ARGS); static int mps_dump_reqs(SYSCTL_HANDLER_ARGS); static void mps_parse_debug(struct mps_softc *sc, char *list); SYSCTL_NODE(_hw, OID_AUTO, mps, CTLFLAG_RD, 0, "MPS Driver Parameters"); MALLOC_DEFINE(M_MPT2, "mps", "mpt2 driver memory"); MALLOC_DECLARE(M_MPSUSER); /* * Do a "Diagnostic Reset" aka a hard reset. This should get the chip out of * any state and back to its initialization state machine. */ static char mpt2_reset_magic[] = { 0x00, 0x0f, 0x04, 0x0b, 0x02, 0x07, 0x0d }; /* Added this union to smoothly convert le64toh cm->cm_desc.Words. * Compiler only support unint64_t to be passed as argument. * Otherwise it will throw below error * "aggregate value used where an integer was expected" */ typedef union _reply_descriptor { u64 word; struct { u32 low; u32 high; } u; }reply_descriptor,address_descriptor; /* Rate limit chain-fail messages to 1 per minute */ static struct timeval mps_chainfail_interval = { 60, 0 }; /* * sleep_flag can be either CAN_SLEEP or NO_SLEEP. * If this function is called from process context, it can sleep * and there is no harm to sleep, in case if this fuction is called * from Interrupt handler, we can not sleep and need NO_SLEEP flag set. * based on sleep flags driver will call either msleep, pause or DELAY. * msleep and pause are of same variant, but pause is used when mps_mtx * is not hold by driver. * */ static int mps_diag_reset(struct mps_softc *sc,int sleep_flag) { uint32_t reg; int i, error, tries = 0; uint8_t first_wait_done = FALSE; mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); /* Clear any pending interrupts */ mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* * Force NO_SLEEP for threads prohibited to sleep * e.a Thread from interrupt handler are prohibited to sleep. */ if (curthread->td_no_sleeping != 0) sleep_flag = NO_SLEEP; mps_dprint(sc, MPS_INIT, "sequence start, sleep_flag= %d\n", sleep_flag); /* Push the magic sequence */ error = ETIMEDOUT; while (tries++ < 20) { for (i = 0; i < sizeof(mpt2_reset_magic); i++) mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, mpt2_reset_magic[i]); /* wait 100 msec */ if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, "mpsdiag", hz/10); else if (sleep_flag == CAN_SLEEP) pause("mpsdiag", hz/10); else DELAY(100 * 1000); reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); if (reg & MPI2_DIAG_DIAG_WRITE_ENABLE) { error = 0; break; } } if (error) { mps_dprint(sc, MPS_INIT, "sequence failed, error=%d, exit\n", error); return (error); } /* Send the actual reset. XXX need to refresh the reg? */ reg |= MPI2_DIAG_RESET_ADAPTER; mps_dprint(sc, MPS_INIT, "sequence success, sending reset, reg= 0x%x\n", reg); mps_regwrite(sc, MPI2_HOST_DIAGNOSTIC_OFFSET, reg); /* Wait up to 300 seconds in 50ms intervals */ error = ETIMEDOUT; for (i = 0; i < 6000; i++) { /* * Wait 50 msec. If this is the first time through, wait 256 * msec to satisfy Diag Reset timing requirements. */ if (first_wait_done) { if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, "mpsdiag", hz/20); else if (sleep_flag == CAN_SLEEP) pause("mpsdiag", hz/20); else DELAY(50 * 1000); } else { DELAY(256 * 1000); first_wait_done = TRUE; } /* * Check for the RESET_ADAPTER bit to be cleared first, then * wait for the RESET state to be cleared, which takes a little * longer. */ reg = mps_regread(sc, MPI2_HOST_DIAGNOSTIC_OFFSET); if (reg & MPI2_DIAG_RESET_ADAPTER) { continue; } reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); if ((reg & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_RESET) { error = 0; break; } } if (error) { mps_dprint(sc, MPS_INIT, "reset failed, error= %d, exit\n", error); return (error); } mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 0x0); mps_dprint(sc, MPS_INIT, "diag reset success, exit\n"); return (0); } static int mps_message_unit_reset(struct mps_softc *sc, int sleep_flag) { int error; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); error = 0; mps_regwrite(sc, MPI2_DOORBELL_OFFSET, MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET << MPI2_DOORBELL_FUNCTION_SHIFT); if (mps_wait_db_ack(sc, 5, sleep_flag) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Doorbell handshake failed\n"); error = ETIMEDOUT; } mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return (error); } static int mps_transition_ready(struct mps_softc *sc) { uint32_t reg, state; int error, tries = 0; int sleep_flags; MPS_FUNCTRACE(sc); /* If we are in attach call, do not sleep */ sleep_flags = (sc->mps_flags & MPS_FLAGS_ATTACH_DONE) ? CAN_SLEEP:NO_SLEEP; error = 0; mps_dprint(sc, MPS_INIT, "%s entered, sleep_flags= %d\n", __func__, sleep_flags); while (tries++ < 1200) { reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_dprint(sc, MPS_INIT, " Doorbell= 0x%x\n", reg); /* * Ensure the IOC is ready to talk. If it's not, try * resetting it. */ if (reg & MPI2_DOORBELL_USED) { mps_dprint(sc, MPS_INIT, " Not ready, sending diag " "reset\n"); mps_diag_reset(sc, sleep_flags); DELAY(50000); continue; } /* Is the adapter owned by another peer? */ if ((reg & MPI2_DOORBELL_WHO_INIT_MASK) == (MPI2_WHOINIT_PCI_PEER << MPI2_DOORBELL_WHO_INIT_SHIFT)) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC is under the " "control of another peer host, aborting " "initialization.\n"); error = ENXIO; break; } state = reg & MPI2_IOC_STATE_MASK; if (state == MPI2_IOC_STATE_READY) { /* Ready to go! */ error = 0; break; } else if (state == MPI2_IOC_STATE_FAULT) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC in fault " "state 0x%x, resetting\n", state & MPI2_DOORBELL_FAULT_CODE_MASK); mps_diag_reset(sc, sleep_flags); } else if (state == MPI2_IOC_STATE_OPERATIONAL) { /* Need to take ownership */ mps_message_unit_reset(sc, sleep_flags); } else if (state == MPI2_IOC_STATE_RESET) { /* Wait a bit, IOC might be in transition */ mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC in unexpected reset state\n"); } else { mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC in unknown state 0x%x\n", state); error = EINVAL; break; } /* Wait 50ms for things to settle down. */ DELAY(50000); } if (error) mps_dprint(sc, MPS_INIT|MPS_FAULT, "Cannot transition IOC to ready\n"); mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return (error); } static int mps_transition_operational(struct mps_softc *sc) { uint32_t reg, state; int error; MPS_FUNCTRACE(sc); error = 0; reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_dprint(sc, MPS_INIT, "%s entered, Doorbell= 0x%x\n", __func__, reg); state = reg & MPI2_IOC_STATE_MASK; if (state != MPI2_IOC_STATE_READY) { mps_dprint(sc, MPS_INIT, "IOC not ready\n"); if ((error = mps_transition_ready(sc)) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "failed to transition ready, exit\n"); return (error); } } error = mps_send_iocinit(sc); mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return (error); } static void mps_resize_queues(struct mps_softc *sc) { u_int reqcr, prireqcr, maxio, sges_per_frame; /* * Size the queues. Since the reply queues always need one free * entry, we'll deduct one reply message here. The LSI documents * suggest instead to add a count to the request queue, but I think * that it's better to deduct from reply queue. */ prireqcr = MAX(1, sc->max_prireqframes); prireqcr = MIN(prireqcr, sc->facts->HighPriorityCredit); reqcr = MAX(2, sc->max_reqframes); reqcr = MIN(reqcr, sc->facts->RequestCredit); sc->num_reqs = prireqcr + reqcr; sc->num_prireqs = prireqcr; sc->num_replies = MIN(sc->max_replyframes + sc->max_evtframes, sc->facts->MaxReplyDescriptorPostQueueDepth) - 1; /* Store the request frame size in bytes rather than as 32bit words */ sc->reqframesz = sc->facts->IOCRequestFrameSize * 4; /* * Max IO Size is Page Size * the following: * ((SGEs per frame - 1 for chain element) * Max Chain Depth) * + 1 for no chain needed in last frame * * If user suggests a Max IO size to use, use the smaller of the * user's value and the calculated value as long as the user's * value is larger than 0. The user's value is in pages. */ sges_per_frame = sc->reqframesz / sizeof(MPI2_SGE_SIMPLE64) - 1; maxio = (sges_per_frame * sc->facts->MaxChainDepth + 1) * PAGE_SIZE; /* * If I/O size limitation requested, then use it and pass up to CAM. * If not, use MAXPHYS as an optimization hint, but report HW limit. */ if (sc->max_io_pages > 0) { maxio = min(maxio, sc->max_io_pages * PAGE_SIZE); sc->maxio = maxio; } else { sc->maxio = maxio; maxio = min(maxio, MAXPHYS); } sc->num_chains = (maxio / PAGE_SIZE + sges_per_frame - 2) / sges_per_frame * reqcr; if (sc->max_chains > 0 && sc->max_chains < sc->num_chains) sc->num_chains = sc->max_chains; /* * Figure out the number of MSIx-based queues. If the firmware or * user has done something crazy and not allowed enough credit for * the queues to be useful then don't enable multi-queue. */ if (sc->facts->MaxMSIxVectors < 2) sc->msi_msgs = 1; if (sc->msi_msgs > 1) { sc->msi_msgs = MIN(sc->msi_msgs, mp_ncpus); sc->msi_msgs = MIN(sc->msi_msgs, sc->facts->MaxMSIxVectors); if (sc->num_reqs / sc->msi_msgs < 2) sc->msi_msgs = 1; } mps_dprint(sc, MPS_INIT, "Sized queues to q=%d reqs=%d replies=%d\n", sc->msi_msgs, sc->num_reqs, sc->num_replies); } /* * This is called during attach and when re-initializing due to a Diag Reset. * IOC Facts is used to allocate many of the structures needed by the driver. * If called from attach, de-allocation is not required because the driver has * not allocated any structures yet, but if called from a Diag Reset, previously * allocated structures based on IOC Facts will need to be freed and re- * allocated bases on the latest IOC Facts. */ static int mps_iocfacts_allocate(struct mps_softc *sc, uint8_t attaching) { int error; Mpi2IOCFactsReply_t saved_facts; uint8_t saved_mode, reallocating; mps_dprint(sc, MPS_INIT|MPS_TRACE, "%s entered\n", __func__); /* Save old IOC Facts and then only reallocate if Facts have changed */ if (!attaching) { bcopy(sc->facts, &saved_facts, sizeof(MPI2_IOC_FACTS_REPLY)); } /* * Get IOC Facts. In all cases throughout this function, panic if doing * a re-initialization and only return the error if attaching so the OS * can handle it. */ if ((error = mps_get_iocfacts(sc, sc->facts)) != 0) { if (attaching) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to get " "IOC Facts with error %d, exit\n", error); return (error); } else { panic("%s failed to get IOC Facts with error %d\n", __func__, error); } } MPS_DPRINT_PAGE(sc, MPS_XINFO, iocfacts, sc->facts); snprintf(sc->fw_version, sizeof(sc->fw_version), "%02d.%02d.%02d.%02d", sc->facts->FWVersion.Struct.Major, sc->facts->FWVersion.Struct.Minor, sc->facts->FWVersion.Struct.Unit, sc->facts->FWVersion.Struct.Dev); mps_dprint(sc, MPS_INFO, "Firmware: %s, Driver: %s\n", sc->fw_version, MPS_DRIVER_VERSION); mps_dprint(sc, MPS_INFO, "IOCCapabilities: %b\n", sc->facts->IOCCapabilities, "\20" "\3ScsiTaskFull" "\4DiagTrace" "\5SnapBuf" "\6ExtBuf" "\7EEDP" "\10BiDirTarg" "\11Multicast" "\14TransRetry" "\15IR" "\16EventReplay" "\17RaidAccel" "\20MSIXIndex" "\21HostDisc"); /* * If the chip doesn't support event replay then a hard reset will be * required to trigger a full discovery. Do the reset here then * retransition to Ready. A hard reset might have already been done, * but it doesn't hurt to do it again. Only do this if attaching, not * for a Diag Reset. */ if (attaching && ((sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY) == 0)) { mps_dprint(sc, MPS_INIT, "No event replay, reseting\n"); mps_diag_reset(sc, NO_SLEEP); if ((error = mps_transition_ready(sc)) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to " "transition to ready with error %d, exit\n", error); return (error); } } /* * Set flag if IR Firmware is loaded. If the RAID Capability has * changed from the previous IOC Facts, log a warning, but only if * checking this after a Diag Reset and not during attach. */ saved_mode = sc->ir_firmware; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID) sc->ir_firmware = 1; if (!attaching) { if (sc->ir_firmware != saved_mode) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "new IR/IT mode " "in IOC Facts does not match previous mode\n"); } } /* Only deallocate and reallocate if relevant IOC Facts have changed */ reallocating = FALSE; sc->mps_flags &= ~MPS_FLAGS_REALLOCATED; if ((!attaching) && ((saved_facts.MsgVersion != sc->facts->MsgVersion) || (saved_facts.HeaderVersion != sc->facts->HeaderVersion) || (saved_facts.MaxChainDepth != sc->facts->MaxChainDepth) || (saved_facts.RequestCredit != sc->facts->RequestCredit) || (saved_facts.ProductID != sc->facts->ProductID) || (saved_facts.IOCCapabilities != sc->facts->IOCCapabilities) || (saved_facts.IOCRequestFrameSize != sc->facts->IOCRequestFrameSize) || (saved_facts.MaxTargets != sc->facts->MaxTargets) || (saved_facts.MaxSasExpanders != sc->facts->MaxSasExpanders) || (saved_facts.MaxEnclosures != sc->facts->MaxEnclosures) || (saved_facts.HighPriorityCredit != sc->facts->HighPriorityCredit) || (saved_facts.MaxReplyDescriptorPostQueueDepth != sc->facts->MaxReplyDescriptorPostQueueDepth) || (saved_facts.ReplyFrameSize != sc->facts->ReplyFrameSize) || (saved_facts.MaxVolumes != sc->facts->MaxVolumes) || (saved_facts.MaxPersistentEntries != sc->facts->MaxPersistentEntries))) { reallocating = TRUE; /* Record that we reallocated everything */ sc->mps_flags |= MPS_FLAGS_REALLOCATED; } /* * Some things should be done if attaching or re-allocating after a Diag * Reset, but are not needed after a Diag Reset if the FW has not * changed. */ if (attaching || reallocating) { /* * Check if controller supports FW diag buffers and set flag to * enable each type. */ if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_DIAG_TRACE_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_TRACE]. enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_SNAPSHOT_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_SNAPSHOT]. enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EXTENDED_BUFFER) sc->fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_EXTENDED]. enabled = TRUE; /* * Set flag if EEDP is supported and if TLR is supported. */ if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EEDP) sc->eedp_enabled = TRUE; if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_TLR) sc->control_TLR = TRUE; mps_resize_queues(sc); /* * Initialize all Tail Queues */ TAILQ_INIT(&sc->req_list); TAILQ_INIT(&sc->high_priority_req_list); TAILQ_INIT(&sc->chain_list); TAILQ_INIT(&sc->tm_list); } /* * If doing a Diag Reset and the FW is significantly different * (reallocating will be set above in IOC Facts comparison), then all * buffers based on the IOC Facts will need to be freed before they are * reallocated. */ if (reallocating) { mps_iocfacts_free(sc); mpssas_realloc_targets(sc, saved_facts.MaxTargets + saved_facts.MaxVolumes); } /* * Any deallocation has been completed. Now start reallocating * if needed. Will only need to reallocate if attaching or if the new * IOC Facts are different from the previous IOC Facts after a Diag * Reset. Targets have already been allocated above if needed. */ error = 0; while (attaching || reallocating) { if ((error = mps_alloc_hw_queues(sc)) != 0) break; if ((error = mps_alloc_replies(sc)) != 0) break; if ((error = mps_alloc_requests(sc)) != 0) break; if ((error = mps_alloc_queues(sc)) != 0) break; break; } if (error) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to alloc queues with error %d\n", error); mps_free(sc); return (error); } /* Always initialize the queues */ bzero(sc->free_queue, sc->fqdepth * 4); mps_init_queues(sc); /* * Always get the chip out of the reset state, but only panic if not * attaching. If attaching and there is an error, that is handled by * the OS. */ error = mps_transition_operational(sc); if (error != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to " "transition to operational with error %d\n", error); mps_free(sc); return (error); } /* * Finish the queue initialization. * These are set here instead of in mps_init_queues() because the * IOC resets these values during the state transition in * mps_transition_operational(). The free index is set to 1 * because the corresponding index in the IOC is set to 0, and the * IOC treats the queues as full if both are set to the same value. * Hence the reason that the queue can't hold all of the possible * replies. */ sc->replypostindex = 0; mps_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, 0); /* * Attach the subsystems so they can prepare their event masks. * XXX Should be dynamic so that IM/IR and user modules can attach */ error = 0; while (attaching) { mps_dprint(sc, MPS_INIT, "Attaching subsystems\n"); if ((error = mps_attach_log(sc)) != 0) break; if ((error = mps_attach_sas(sc)) != 0) break; if ((error = mps_attach_user(sc)) != 0) break; break; } if (error) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to attach all " "subsystems: error %d\n", error); mps_free(sc); return (error); } /* * XXX If the number of MSI-X vectors changes during re-init, this * won't see it and adjust. */ if (attaching && (error = mps_pci_setup_interrupts(sc)) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Failed to setup " "interrupts\n"); mps_free(sc); return (error); } /* * Set flag if this is a WD controller. This shouldn't ever change, but * reset it after a Diag Reset, just in case. */ sc->WD_available = FALSE; if (pci_get_device(sc->mps_dev) == MPI2_MFGPAGE_DEVID_SSS6200) sc->WD_available = TRUE; return (error); } /* * This is called if memory is being free (during detach for example) and when * buffers need to be reallocated due to a Diag Reset. */ static void mps_iocfacts_free(struct mps_softc *sc) { struct mps_command *cm; int i; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); if (sc->free_busaddr != 0) bus_dmamap_unload(sc->queues_dmat, sc->queues_map); if (sc->free_queue != NULL) bus_dmamem_free(sc->queues_dmat, sc->free_queue, sc->queues_map); if (sc->queues_dmat != NULL) bus_dma_tag_destroy(sc->queues_dmat); if (sc->chain_frames != NULL) { bus_dmamap_unload(sc->chain_dmat, sc->chain_map); bus_dmamem_free(sc->chain_dmat, sc->chain_frames, sc->chain_map); } if (sc->chain_dmat != NULL) bus_dma_tag_destroy(sc->chain_dmat); if (sc->sense_busaddr != 0) bus_dmamap_unload(sc->sense_dmat, sc->sense_map); if (sc->sense_frames != NULL) bus_dmamem_free(sc->sense_dmat, sc->sense_frames, sc->sense_map); if (sc->sense_dmat != NULL) bus_dma_tag_destroy(sc->sense_dmat); if (sc->reply_busaddr != 0) bus_dmamap_unload(sc->reply_dmat, sc->reply_map); if (sc->reply_frames != NULL) bus_dmamem_free(sc->reply_dmat, sc->reply_frames, sc->reply_map); if (sc->reply_dmat != NULL) bus_dma_tag_destroy(sc->reply_dmat); if (sc->req_busaddr != 0) bus_dmamap_unload(sc->req_dmat, sc->req_map); if (sc->req_frames != NULL) bus_dmamem_free(sc->req_dmat, sc->req_frames, sc->req_map); if (sc->req_dmat != NULL) bus_dma_tag_destroy(sc->req_dmat); if (sc->chains != NULL) free(sc->chains, M_MPT2); if (sc->commands != NULL) { for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; bus_dmamap_destroy(sc->buffer_dmat, cm->cm_dmamap); } free(sc->commands, M_MPT2); } if (sc->buffer_dmat != NULL) bus_dma_tag_destroy(sc->buffer_dmat); mps_pci_free_interrupts(sc); free(sc->queues, M_MPT2); sc->queues = NULL; } /* * The terms diag reset and hard reset are used interchangeably in the MPI * docs to mean resetting the controller chip. In this code diag reset * cleans everything up, and the hard reset function just sends the reset * sequence to the chip. This should probably be refactored so that every * subsystem gets a reset notification of some sort, and can clean up * appropriately. */ int mps_reinit(struct mps_softc *sc) { int error; struct mpssas_softc *sassc; sassc = sc->sassc; MPS_FUNCTRACE(sc); mtx_assert(&sc->mps_mtx, MA_OWNED); mps_dprint(sc, MPS_INIT|MPS_INFO, "Reinitializing controller\n"); if (sc->mps_flags & MPS_FLAGS_DIAGRESET) { mps_dprint(sc, MPS_INIT, "Reset already in progress\n"); return 0; } /* make sure the completion callbacks can recognize they're getting * a NULL cm_reply due to a reset. */ sc->mps_flags |= MPS_FLAGS_DIAGRESET; /* * Mask interrupts here. */ mps_dprint(sc, MPS_INIT, "masking interrupts and resetting\n"); mps_mask_intr(sc); error = mps_diag_reset(sc, CAN_SLEEP); if (error != 0) { /* XXXSL No need to panic here */ panic("%s hard reset failed with error %d\n", __func__, error); } /* Restore the PCI state, including the MSI-X registers */ mps_pci_restore(sc); /* Give the I/O subsystem special priority to get itself prepared */ mpssas_handle_reinit(sc); /* * Get IOC Facts and allocate all structures based on this information. * The attach function will also call mps_iocfacts_allocate at startup. * If relevant values have changed in IOC Facts, this function will free * all of the memory based on IOC Facts and reallocate that memory. */ if ((error = mps_iocfacts_allocate(sc, FALSE)) != 0) { panic("%s IOC Facts based allocation failed with error %d\n", __func__, error); } /* * Mapping structures will be re-allocated after getting IOC Page8, so * free these structures here. */ mps_mapping_exit(sc); /* * The static page function currently read is IOC Page8. Others can be * added in future. It's possible that the values in IOC Page8 have * changed after a Diag Reset due to user modification, so always read * these. Interrupts are masked, so unmask them before getting config * pages. */ mps_unmask_intr(sc); sc->mps_flags &= ~MPS_FLAGS_DIAGRESET; mps_base_static_config_pages(sc); /* * Some mapping info is based in IOC Page8 data, so re-initialize the * mapping tables. */ mps_mapping_initialize(sc); /* * Restart will reload the event masks clobbered by the reset, and * then enable the port. */ mps_reregister_events(sc); /* the end of discovery will release the simq, so we're done. */ mps_dprint(sc, MPS_INIT|MPS_XINFO, "Finished sc %p post %u free %u\n", sc, sc->replypostindex, sc->replyfreeindex); mpssas_release_simq_reinit(sassc); mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return 0; } /* Wait for the chip to ACK a word that we've put into its FIFO * Wait for seconds. In single loop wait for busy loop * for 500 microseconds. * Total is [ 0.5 * (2000 * ) ] in miliseconds. * */ static int mps_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag) { u32 cntdn, count; u32 int_status; u32 doorbell; count = 0; cntdn = (sleep_flag == CAN_SLEEP) ? 1000*timeout : 2000*timeout; do { int_status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); if (!(int_status & MPI2_HIS_SYS2IOC_DB_STATUS)) { mps_dprint(sc, MPS_TRACE, "%s: successful count(%d), timeout(%d)\n", __func__, count, timeout); return 0; } else if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) { doorbell = mps_regread(sc, MPI2_DOORBELL_OFFSET); if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) { mps_dprint(sc, MPS_FAULT, "fault_state(0x%04x)!\n", doorbell); return (EFAULT); } } else if (int_status == 0xFFFFFFFF) goto out; /* If it can sleep, sleep for 1 milisecond, else busy loop for * 0.5 milisecond */ if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, "mpsdba", hz/1000); else if (sleep_flag == CAN_SLEEP) pause("mpsdba", hz/1000); else DELAY(500); count++; } while (--cntdn); out: mps_dprint(sc, MPS_FAULT, "%s: failed due to timeout count(%d), " "int_status(%x)!\n", __func__, count, int_status); return (ETIMEDOUT); } /* Wait for the chip to signal that the next word in its FIFO can be fetched */ static int mps_wait_db_int(struct mps_softc *sc) { int retry; for (retry = 0; retry < MPS_DB_MAX_WAIT; retry++) { if ((mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET) & MPI2_HIS_IOC2SYS_DB_STATUS) != 0) return (0); DELAY(2000); } return (ETIMEDOUT); } /* Step through the synchronous command state machine, i.e. "Doorbell mode" */ static int mps_request_sync(struct mps_softc *sc, void *req, MPI2_DEFAULT_REPLY *reply, int req_sz, int reply_sz, int timeout) { uint32_t *data32; uint16_t *data16; int i, count, ioc_sz, residual; int sleep_flags = CAN_SLEEP; if (curthread->td_no_sleeping != 0) sleep_flags = NO_SLEEP; /* Step 1 */ mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* Step 2 */ if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) return (EBUSY); /* Step 3 * Announce that a message is coming through the doorbell. Messages * are pushed at 32bit words, so round up if needed. */ count = (req_sz + 3) / 4; mps_regwrite(sc, MPI2_DOORBELL_OFFSET, (MPI2_FUNCTION_HANDSHAKE << MPI2_DOORBELL_FUNCTION_SHIFT) | (count << MPI2_DOORBELL_ADD_DWORDS_SHIFT)); /* Step 4 */ if (mps_wait_db_int(sc) || (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) == 0) { mps_dprint(sc, MPS_FAULT, "Doorbell failed to activate\n"); return (ENXIO); } mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) { mps_dprint(sc, MPS_FAULT, "Doorbell handshake failed\n"); return (ENXIO); } /* Step 5 */ /* Clock out the message data synchronously in 32-bit dwords*/ data32 = (uint32_t *)req; for (i = 0; i < count; i++) { mps_regwrite(sc, MPI2_DOORBELL_OFFSET, htole32(data32[i])); if (mps_wait_db_ack(sc, 5, sleep_flags) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout while writing doorbell\n"); return (ENXIO); } } /* Step 6 */ /* Clock in the reply in 16-bit words. The total length of the * message is always in the 4th byte, so clock out the first 2 words * manually, then loop the rest. */ data16 = (uint16_t *)reply; if (mps_wait_db_int(sc) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 0\n"); return (ENXIO); } data16[0] = mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); if (mps_wait_db_int(sc) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell 1\n"); return (ENXIO); } data16[1] = mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); /* Number of 32bit words in the message */ ioc_sz = reply->MsgLength; /* * Figure out how many 16bit words to clock in without overrunning. * The precision loss with dividing reply_sz can safely be * ignored because the messages can only be multiples of 32bits. */ residual = 0; count = MIN((reply_sz / 4), ioc_sz) * 2; if (count < ioc_sz * 2) { residual = ioc_sz * 2 - count; mps_dprint(sc, MPS_ERROR, "Driver error, throwing away %d " "residual message words\n", residual); } for (i = 2; i < count; i++) { if (mps_wait_db_int(sc) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell %d\n", i); return (ENXIO); } data16[i] = mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_DATA_MASK; mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); } /* * Pull out residual words that won't fit into the provided buffer. * This keeps the chip from hanging due to a driver programming * error. */ while (residual--) { if (mps_wait_db_int(sc) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout reading doorbell\n"); return (ENXIO); } (void)mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); } /* Step 7 */ if (mps_wait_db_int(sc) != 0) { mps_dprint(sc, MPS_FAULT, "Timeout waiting to exit doorbell\n"); return (ENXIO); } if (mps_regread(sc, MPI2_DOORBELL_OFFSET) & MPI2_DOORBELL_USED) mps_dprint(sc, MPS_FAULT, "Warning, doorbell still active\n"); mps_regwrite(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET, 0x0); return (0); } static void mps_enqueue_request(struct mps_softc *sc, struct mps_command *cm) { reply_descriptor rd; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_TRACE, "SMID %u cm %p ccb %p\n", cm->cm_desc.Default.SMID, cm, cm->cm_ccb); if (sc->mps_flags & MPS_FLAGS_ATTACH_DONE && !(sc->mps_flags & MPS_FLAGS_SHUTDOWN)) mtx_assert(&sc->mps_mtx, MA_OWNED); if (++sc->io_cmds_active > sc->io_cmds_highwater) sc->io_cmds_highwater++; rd.u.low = cm->cm_desc.Words.Low; rd.u.high = cm->cm_desc.Words.High; rd.word = htole64(rd.word); KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, ("command not busy\n")); cm->cm_state = MPS_CM_STATE_INQUEUE; /* TODO-We may need to make below regwrite atomic */ mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_LOW_OFFSET, rd.u.low); mps_regwrite(sc, MPI2_REQUEST_DESCRIPTOR_POST_HIGH_OFFSET, rd.u.high); } /* * Just the FACTS, ma'am. */ static int mps_get_iocfacts(struct mps_softc *sc, MPI2_IOC_FACTS_REPLY *facts) { MPI2_DEFAULT_REPLY *reply; MPI2_IOC_FACTS_REQUEST request; int error, req_sz, reply_sz; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); req_sz = sizeof(MPI2_IOC_FACTS_REQUEST); reply_sz = sizeof(MPI2_IOC_FACTS_REPLY); reply = (MPI2_DEFAULT_REPLY *)facts; bzero(&request, req_sz); request.Function = MPI2_FUNCTION_IOC_FACTS; error = mps_request_sync(sc, &request, reply, req_sz, reply_sz, 5); mps_dprint(sc, MPS_INIT, "%s exit error= %d\n", __func__, error); return (error); } static int mps_send_iocinit(struct mps_softc *sc) { MPI2_IOC_INIT_REQUEST init; MPI2_DEFAULT_REPLY reply; int req_sz, reply_sz, error; struct timeval now; uint64_t time_in_msec; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); /* Do a quick sanity check on proper initialization */ if ((sc->pqdepth == 0) || (sc->fqdepth == 0) || (sc->reqframesz == 0) || (sc->replyframesz == 0)) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Driver not fully initialized for IOCInit\n"); return (EINVAL); } req_sz = sizeof(MPI2_IOC_INIT_REQUEST); reply_sz = sizeof(MPI2_IOC_INIT_REPLY); bzero(&init, req_sz); bzero(&reply, reply_sz); /* * Fill in the init block. Note that most addresses are * deliberately in the lower 32bits of memory. This is a micro- * optimzation for PCI/PCIX, though it's not clear if it helps PCIe. */ init.Function = MPI2_FUNCTION_IOC_INIT; init.WhoInit = MPI2_WHOINIT_HOST_DRIVER; init.MsgVersion = htole16(MPI2_VERSION); init.HeaderVersion = htole16(MPI2_HEADER_VERSION); init.SystemRequestFrameSize = htole16((uint16_t)(sc->reqframesz / 4)); init.ReplyDescriptorPostQueueDepth = htole16(sc->pqdepth); init.ReplyFreeQueueDepth = htole16(sc->fqdepth); init.SenseBufferAddressHigh = 0; init.SystemReplyAddressHigh = 0; init.SystemRequestFrameBaseAddress.High = 0; init.SystemRequestFrameBaseAddress.Low = htole32((uint32_t)sc->req_busaddr); init.ReplyDescriptorPostQueueAddress.High = 0; init.ReplyDescriptorPostQueueAddress.Low = htole32((uint32_t)sc->post_busaddr); init.ReplyFreeQueueAddress.High = 0; init.ReplyFreeQueueAddress.Low = htole32((uint32_t)sc->free_busaddr); getmicrotime(&now); time_in_msec = (now.tv_sec * 1000 + now.tv_usec/1000); init.TimeStamp.High = htole32((time_in_msec >> 32) & 0xFFFFFFFF); init.TimeStamp.Low = htole32(time_in_msec & 0xFFFFFFFF); error = mps_request_sync(sc, &init, &reply, req_sz, reply_sz, 5); if ((reply.IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) error = ENXIO; mps_dprint(sc, MPS_INIT, "IOCInit status= 0x%x\n", reply.IOCStatus); mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return (error); } void mps_memaddr_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { bus_addr_t *addr; addr = arg; *addr = segs[0].ds_addr; } void mps_memaddr_wait_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mps_busdma_context *ctx; int need_unload, need_free; ctx = (struct mps_busdma_context *)arg; need_unload = 0; need_free = 0; mps_lock(ctx->softc); ctx->error = error; ctx->completed = 1; if ((error == 0) && (ctx->abandoned == 0)) { *ctx->addr = segs[0].ds_addr; } else { if (nsegs != 0) need_unload = 1; if (ctx->abandoned != 0) need_free = 1; } if (need_free == 0) wakeup(ctx); mps_unlock(ctx->softc); if (need_unload != 0) { bus_dmamap_unload(ctx->buffer_dmat, ctx->buffer_dmamap); *ctx->addr = 0; } if (need_free != 0) free(ctx, M_MPSUSER); } static int mps_alloc_queues(struct mps_softc *sc) { struct mps_queue *q; u_int nq, i; nq = sc->msi_msgs; mps_dprint(sc, MPS_INIT|MPS_XINFO, "Allocating %d I/O queues\n", nq); sc->queues = malloc(sizeof(struct mps_queue) * nq, M_MPT2, M_NOWAIT|M_ZERO); if (sc->queues == NULL) return (ENOMEM); for (i = 0; i < nq; i++) { q = &sc->queues[i]; mps_dprint(sc, MPS_INIT, "Configuring queue %d %p\n", i, q); q->sc = sc; q->qnum = i; } return (0); } static int mps_alloc_hw_queues(struct mps_softc *sc) { bus_addr_t queues_busaddr; uint8_t *queues; int qsize, fqsize, pqsize; /* * The reply free queue contains 4 byte entries in multiples of 16 and * aligned on a 16 byte boundary. There must always be an unused entry. * This queue supplies fresh reply frames for the firmware to use. * * The reply descriptor post queue contains 8 byte entries in * multiples of 16 and aligned on a 16 byte boundary. This queue * contains filled-in reply frames sent from the firmware to the host. * * These two queues are allocated together for simplicity. */ sc->fqdepth = roundup2(sc->num_replies + 1, 16); sc->pqdepth = roundup2(sc->num_replies + 1, 16); fqsize= sc->fqdepth * 4; pqsize = sc->pqdepth * 8; qsize = fqsize + pqsize; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ qsize, /* maxsize */ 1, /* nsegments */ qsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->queues_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate queues DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->queues_dmat, (void **)&queues, BUS_DMA_NOWAIT, &sc->queues_map)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate queues memory\n"); return (ENOMEM); } bzero(queues, qsize); bus_dmamap_load(sc->queues_dmat, sc->queues_map, queues, qsize, mps_memaddr_cb, &queues_busaddr, 0); sc->free_queue = (uint32_t *)queues; sc->free_busaddr = queues_busaddr; sc->post_queue = (MPI2_REPLY_DESCRIPTORS_UNION *)(queues + fqsize); sc->post_busaddr = queues_busaddr + fqsize; mps_dprint(sc, MPS_INIT, "free queue busaddr= %#016jx size= %d\n", (uintmax_t)sc->free_busaddr, fqsize); mps_dprint(sc, MPS_INIT, "reply queue busaddr= %#016jx size= %d\n", (uintmax_t)sc->post_busaddr, pqsize); return (0); } static int mps_alloc_replies(struct mps_softc *sc) { int rsize, num_replies; /* Store the reply frame size in bytes rather than as 32bit words */ sc->replyframesz = sc->facts->ReplyFrameSize * 4; /* * sc->num_replies should be one less than sc->fqdepth. We need to * allocate space for sc->fqdepth replies, but only sc->num_replies * replies can be used at once. */ num_replies = max(sc->fqdepth, sc->num_replies); rsize = sc->replyframesz * num_replies; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 4, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->reply_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate replies DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->reply_dmat, (void **)&sc->reply_frames, BUS_DMA_NOWAIT, &sc->reply_map)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate replies memory\n"); return (ENOMEM); } bzero(sc->reply_frames, rsize); bus_dmamap_load(sc->reply_dmat, sc->reply_map, sc->reply_frames, rsize, mps_memaddr_cb, &sc->reply_busaddr, 0); mps_dprint(sc, MPS_INIT, "reply frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->reply_busaddr, rsize); return (0); } static void mps_load_chains_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mps_softc *sc = arg; struct mps_chain *chain; bus_size_t bo; int i, o, s; if (error != 0) return; for (i = 0, o = 0, s = 0; s < nsegs; s++) { for (bo = 0; bo + sc->reqframesz <= segs[s].ds_len; bo += sc->reqframesz) { chain = &sc->chains[i++]; chain->chain =(MPI2_SGE_IO_UNION *)(sc->chain_frames+o); chain->chain_busaddr = segs[s].ds_addr + bo; o += sc->reqframesz; mps_free_chain(sc, chain); } if (bo != segs[s].ds_len) o += segs[s].ds_len - bo; } sc->chain_free_lowwater = i; } static int mps_alloc_requests(struct mps_softc *sc) { struct mps_command *cm; int i, rsize, nsegs; rsize = sc->reqframesz * sc->num_reqs; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->req_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate request DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->req_dmat, (void **)&sc->req_frames, BUS_DMA_NOWAIT, &sc->req_map)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate request memory\n"); return (ENOMEM); } bzero(sc->req_frames, rsize); bus_dmamap_load(sc->req_dmat, sc->req_map, sc->req_frames, rsize, mps_memaddr_cb, &sc->req_busaddr, 0); mps_dprint(sc, MPS_INIT, "request frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->req_busaddr, rsize); sc->chains = malloc(sizeof(struct mps_chain) * sc->num_chains, M_MPT2, M_NOWAIT | M_ZERO); if (!sc->chains) { mps_dprint(sc, MPS_ERROR, "Cannot allocate chain memory\n"); return (ENOMEM); } rsize = sc->reqframesz * sc->num_chains; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 16, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ howmany(rsize, PAGE_SIZE), /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->chain_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate chain DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->chain_dmat, (void **)&sc->chain_frames, BUS_DMA_NOWAIT | BUS_DMA_ZERO, &sc->chain_map)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate chain memory\n"); return (ENOMEM); } if (bus_dmamap_load(sc->chain_dmat, sc->chain_map, sc->chain_frames, rsize, mps_load_chains_cb, sc, BUS_DMA_NOWAIT)) { mps_dprint(sc, MPS_ERROR, "Cannot load chain memory\n"); bus_dmamem_free(sc->chain_dmat, sc->chain_frames, sc->chain_map); return (ENOMEM); } rsize = MPS_SENSE_LEN * sc->num_reqs; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR_32BIT,/* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ rsize, /* maxsize */ 1, /* nsegments */ rsize, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->sense_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate sense DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->sense_dmat, (void **)&sc->sense_frames, BUS_DMA_NOWAIT, &sc->sense_map)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate sense memory\n"); return (ENOMEM); } bzero(sc->sense_frames, rsize); bus_dmamap_load(sc->sense_dmat, sc->sense_map, sc->sense_frames, rsize, mps_memaddr_cb, &sc->sense_busaddr, 0); mps_dprint(sc, MPS_INIT, "sense frames busaddr= %#016jx size= %d\n", (uintmax_t)sc->sense_busaddr, rsize); nsegs = (sc->maxio / PAGE_SIZE) + 1; if (bus_dma_tag_create( sc->mps_parent_dmat, /* parent */ 1, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT,/* maxsize */ nsegs, /* nsegments */ BUS_SPACE_MAXSIZE_24BIT,/* maxsegsize */ BUS_DMA_ALLOCNOW, /* flags */ busdma_lock_mutex, /* lockfunc */ &sc->mps_mtx, /* lockarg */ &sc->buffer_dmat)) { mps_dprint(sc, MPS_ERROR, "Cannot allocate buffer DMA tag\n"); return (ENOMEM); } /* * SMID 0 cannot be used as a free command per the firmware spec. * Just drop that command instead of risking accounting bugs. */ sc->commands = malloc(sizeof(struct mps_command) * sc->num_reqs, M_MPT2, M_WAITOK | M_ZERO); if(!sc->commands) { mps_dprint(sc, MPS_ERROR, "Cannot allocate command memory\n"); return (ENOMEM); } for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; cm->cm_req = sc->req_frames + i * sc->reqframesz; cm->cm_req_busaddr = sc->req_busaddr + i * sc->reqframesz; cm->cm_sense = &sc->sense_frames[i]; cm->cm_sense_busaddr = sc->sense_busaddr + i * MPS_SENSE_LEN; cm->cm_desc.Default.SMID = i; cm->cm_sc = sc; cm->cm_state = MPS_CM_STATE_BUSY; TAILQ_INIT(&cm->cm_chain_list); callout_init_mtx(&cm->cm_callout, &sc->mps_mtx, 0); /* XXX Is a failure here a critical problem? */ if (bus_dmamap_create(sc->buffer_dmat, 0, &cm->cm_dmamap) == 0) if (i <= sc->num_prireqs) mps_free_high_priority_command(sc, cm); else mps_free_command(sc, cm); else { panic("failed to allocate command %d\n", i); sc->num_reqs = i; break; } } return (0); } static int mps_init_queues(struct mps_softc *sc) { int i; memset((uint8_t *)sc->post_queue, 0xff, sc->pqdepth * 8); /* * According to the spec, we need to use one less reply than we * have space for on the queue. So sc->num_replies (the number we * use) should be less than sc->fqdepth (allocated size). */ if (sc->num_replies >= sc->fqdepth) return (EINVAL); /* * Initialize all of the free queue entries. */ for (i = 0; i < sc->fqdepth; i++) sc->free_queue[i] = sc->reply_busaddr + (i * sc->replyframesz); sc->replyfreeindex = sc->num_replies; return (0); } /* Get the driver parameter tunables. Lowest priority are the driver defaults. * Next are the global settings, if they exist. Highest are the per-unit * settings, if they exist. */ void mps_get_tunables(struct mps_softc *sc) { char tmpstr[80], mps_debug[80]; /* XXX default to some debugging for now */ sc->mps_debug = MPS_INFO|MPS_FAULT; sc->disable_msix = 0; sc->disable_msi = 0; sc->max_msix = MPS_MSIX_MAX; sc->max_chains = MPS_CHAIN_FRAMES; sc->max_io_pages = MPS_MAXIO_PAGES; sc->enable_ssu = MPS_SSU_ENABLE_SSD_DISABLE_HDD; sc->spinup_wait_time = DEFAULT_SPINUP_WAIT; sc->use_phynum = 1; sc->max_reqframes = MPS_REQ_FRAMES; sc->max_prireqframes = MPS_PRI_REQ_FRAMES; sc->max_replyframes = MPS_REPLY_FRAMES; sc->max_evtframes = MPS_EVT_REPLY_FRAMES; /* * Grab the global variables. */ bzero(mps_debug, 80); if (TUNABLE_STR_FETCH("hw.mps.debug_level", mps_debug, 80) != 0) mps_parse_debug(sc, mps_debug); TUNABLE_INT_FETCH("hw.mps.disable_msix", &sc->disable_msix); TUNABLE_INT_FETCH("hw.mps.disable_msi", &sc->disable_msi); TUNABLE_INT_FETCH("hw.mps.max_msix", &sc->max_msix); TUNABLE_INT_FETCH("hw.mps.max_chains", &sc->max_chains); TUNABLE_INT_FETCH("hw.mps.max_io_pages", &sc->max_io_pages); TUNABLE_INT_FETCH("hw.mps.enable_ssu", &sc->enable_ssu); TUNABLE_INT_FETCH("hw.mps.spinup_wait_time", &sc->spinup_wait_time); TUNABLE_INT_FETCH("hw.mps.use_phy_num", &sc->use_phynum); TUNABLE_INT_FETCH("hw.mps.max_reqframes", &sc->max_reqframes); TUNABLE_INT_FETCH("hw.mps.max_prireqframes", &sc->max_prireqframes); TUNABLE_INT_FETCH("hw.mps.max_replyframes", &sc->max_replyframes); TUNABLE_INT_FETCH("hw.mps.max_evtframes", &sc->max_evtframes); /* Grab the unit-instance variables */ snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.debug_level", device_get_unit(sc->mps_dev)); bzero(mps_debug, 80); if (TUNABLE_STR_FETCH(tmpstr, mps_debug, 80) != 0) mps_parse_debug(sc, mps_debug); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msix", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->disable_msix); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msi", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->disable_msi); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_msix", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_msix); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_chains", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_chains); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_io_pages", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_io_pages); bzero(sc->exclude_ids, sizeof(sc->exclude_ids)); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.exclude_ids", device_get_unit(sc->mps_dev)); TUNABLE_STR_FETCH(tmpstr, sc->exclude_ids, sizeof(sc->exclude_ids)); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.enable_ssu", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->enable_ssu); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.spinup_wait_time", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->spinup_wait_time); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.use_phy_num", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->use_phynum); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_reqframes", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_reqframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_prireqframes", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_prireqframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_replyframes", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_replyframes); snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_evtframes", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_evtframes); } static void mps_setup_sysctl(struct mps_softc *sc) { struct sysctl_ctx_list *sysctl_ctx = NULL; struct sysctl_oid *sysctl_tree = NULL; char tmpstr[80], tmpstr2[80]; /* * Setup the sysctl variable so the user can change the debug level * on the fly. */ snprintf(tmpstr, sizeof(tmpstr), "MPS controller %d", device_get_unit(sc->mps_dev)); snprintf(tmpstr2, sizeof(tmpstr2), "%d", device_get_unit(sc->mps_dev)); sysctl_ctx = device_get_sysctl_ctx(sc->mps_dev); if (sysctl_ctx != NULL) sysctl_tree = device_get_sysctl_tree(sc->mps_dev); if (sysctl_tree == NULL) { sysctl_ctx_init(&sc->sysctl_ctx); sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_hw_mps), OID_AUTO, tmpstr2, CTLFLAG_RD, 0, tmpstr); if (sc->sysctl_tree == NULL) return; sysctl_ctx = &sc->sysctl_ctx; sysctl_tree = sc->sysctl_tree; } SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "debug_level", CTLTYPE_STRING | CTLFLAG_RW |CTLFLAG_MPSAFE, sc, 0, mps_debug_sysctl, "A", "mps debug level"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "disable_msix", CTLFLAG_RD, &sc->disable_msix, 0, "Disable the use of MSI-X interrupts"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "disable_msi", CTLFLAG_RD, &sc->disable_msi, 0, "Disable the use of MSI interrupts"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_msix", CTLFLAG_RD, &sc->max_msix, 0, "User-defined maximum number of MSIX queues"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "msix_msgs", CTLFLAG_RD, &sc->msi_msgs, 0, "Negotiated number of MSIX queues"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_reqframes", CTLFLAG_RD, &sc->max_reqframes, 0, "Total number of allocated request frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_prireqframes", CTLFLAG_RD, &sc->max_prireqframes, 0, "Total number of allocated high priority request frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_replyframes", CTLFLAG_RD, &sc->max_replyframes, 0, "Total number of allocated reply frames"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_evtframes", CTLFLAG_RD, &sc->max_evtframes, 0, "Total number of event frames allocated"); SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "firmware_version", CTLFLAG_RW, sc->fw_version, strlen(sc->fw_version), "firmware version"); SYSCTL_ADD_STRING(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "driver_version", CTLFLAG_RW, MPS_DRIVER_VERSION, strlen(MPS_DRIVER_VERSION), "driver version"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "io_cmds_active", CTLFLAG_RD, &sc->io_cmds_active, 0, "number of currently active commands"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "io_cmds_highwater", CTLFLAG_RD, &sc->io_cmds_highwater, 0, "maximum active commands seen"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_free", CTLFLAG_RD, &sc->chain_free, 0, "number of free chain elements"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_free_lowwater", CTLFLAG_RD, &sc->chain_free_lowwater, 0,"lowest number of free chain elements"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_chains", CTLFLAG_RD, &sc->max_chains, 0,"maximum chain frames that will be allocated"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "max_io_pages", CTLFLAG_RD, &sc->max_io_pages, 0,"maximum pages to allow per I/O (if <1 use " "IOCFacts)"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "enable_ssu", CTLFLAG_RW, &sc->enable_ssu, 0, "enable SSU to SATA SSD/HDD at shutdown"); SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_alloc_fail", CTLFLAG_RD, &sc->chain_alloc_fail, "chain allocation failures"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "spinup_wait_time", CTLFLAG_RD, &sc->spinup_wait_time, DEFAULT_SPINUP_WAIT, "seconds to wait for " "spinup after SATA ID error"); SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "mapping_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, mps_mapping_dump, "A", "Mapping Table Dump"); SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "encl_table_dump", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, mps_mapping_encl_dump, "A", "Enclosure Table Dump"); SYSCTL_ADD_PROC(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "dump_reqs", CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_SKIP, sc, 0, mps_dump_reqs, "I", "Dump Active Requests"); SYSCTL_ADD_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "use_phy_num", CTLFLAG_RD, &sc->use_phynum, 0, "Use the phy number for enumeration"); } static struct mps_debug_string { char *name; int flag; } mps_debug_strings[] = { {"info", MPS_INFO}, {"fault", MPS_FAULT}, {"event", MPS_EVENT}, {"log", MPS_LOG}, {"recovery", MPS_RECOVERY}, {"error", MPS_ERROR}, {"init", MPS_INIT}, {"xinfo", MPS_XINFO}, {"user", MPS_USER}, {"mapping", MPS_MAPPING}, {"trace", MPS_TRACE} }; enum mps_debug_level_combiner { COMB_NONE, COMB_ADD, COMB_SUB }; static int mps_debug_sysctl(SYSCTL_HANDLER_ARGS) { struct mps_softc *sc; struct mps_debug_string *string; struct sbuf *sbuf; char *buffer; size_t sz; int i, len, debug, error; sc = (struct mps_softc *)arg1; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf = sbuf_new_for_sysctl(NULL, NULL, 128, req); debug = sc->mps_debug; sbuf_printf(sbuf, "%#x", debug); sz = sizeof(mps_debug_strings) / sizeof(mps_debug_strings[0]); for (i = 0; i < sz; i++) { string = &mps_debug_strings[i]; if (debug & string->flag) sbuf_printf(sbuf, ",%s", string->name); } error = sbuf_finish(sbuf); sbuf_delete(sbuf); if (error || req->newptr == NULL) return (error); len = req->newlen - req->newidx; if (len == 0) return (0); buffer = malloc(len, M_MPT2, M_ZERO|M_WAITOK); error = SYSCTL_IN(req, buffer, len); mps_parse_debug(sc, buffer); free(buffer, M_MPT2); return (error); } static void mps_parse_debug(struct mps_softc *sc, char *list) { struct mps_debug_string *string; enum mps_debug_level_combiner op; char *token, *endtoken; size_t sz; int flags, i; if (list == NULL || *list == '\0') return; if (*list == '+') { op = COMB_ADD; list++; } else if (*list == '-') { op = COMB_SUB; list++; } else op = COMB_NONE; if (*list == '\0') return; flags = 0; sz = sizeof(mps_debug_strings) / sizeof(mps_debug_strings[0]); while ((token = strsep(&list, ":,")) != NULL) { /* Handle integer flags */ flags |= strtol(token, &endtoken, 0); if (token != endtoken) continue; /* Handle text flags */ for (i = 0; i < sz; i++) { string = &mps_debug_strings[i]; if (strcasecmp(token, string->name) == 0) { flags |= string->flag; break; } } } switch (op) { case COMB_NONE: sc->mps_debug = flags; break; case COMB_ADD: sc->mps_debug |= flags; break; case COMB_SUB: sc->mps_debug &= (~flags); break; } return; } struct mps_dumpreq_hdr { uint32_t smid; uint32_t state; uint32_t numframes; uint32_t deschi; uint32_t desclo; }; static int mps_dump_reqs(SYSCTL_HANDLER_ARGS) { struct mps_softc *sc; struct mps_chain *chain, *chain1; struct mps_command *cm; struct mps_dumpreq_hdr hdr; struct sbuf *sb; uint32_t smid, state; int i, numreqs, error = 0; sc = (struct mps_softc *)arg1; if ((error = priv_check(curthread, PRIV_DRIVER)) != 0) { printf("priv check error %d\n", error); return (error); } state = MPS_CM_STATE_INQUEUE; smid = 1; numreqs = sc->num_reqs; if (req->newptr != NULL) return (EINVAL); if (smid == 0 || smid > sc->num_reqs) return (EINVAL); if (numreqs <= 0 || (numreqs + smid > sc->num_reqs)) numreqs = sc->num_reqs; sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req); /* Best effort, no locking */ for (i = smid; i < numreqs; i++) { cm = &sc->commands[i]; if (cm->cm_state != state) continue; hdr.smid = i; hdr.state = cm->cm_state; hdr.numframes = 1; hdr.deschi = cm->cm_desc.Words.High; hdr.desclo = cm->cm_desc.Words.Low; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain1) hdr.numframes++; sbuf_bcat(sb, &hdr, sizeof(hdr)); sbuf_bcat(sb, cm->cm_req, 128); TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain1) sbuf_bcat(sb, chain->chain, 128); } error = sbuf_finish(sb); sbuf_delete(sb); return (error); } int mps_attach(struct mps_softc *sc) { int error; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); mtx_init(&sc->mps_mtx, "MPT2SAS lock", NULL, MTX_DEF); callout_init_mtx(&sc->periodic, &sc->mps_mtx, 0); callout_init_mtx(&sc->device_check_callout, &sc->mps_mtx, 0); TAILQ_INIT(&sc->event_list); timevalclear(&sc->lastfail); if ((error = mps_transition_ready(sc)) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "failed to transition " "ready\n"); return (error); } sc->facts = malloc(sizeof(MPI2_IOC_FACTS_REPLY), M_MPT2, M_ZERO|M_NOWAIT); if(!sc->facts) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "Cannot allocate memory, " "exit\n"); return (ENOMEM); } /* * Get IOC Facts and allocate all structures based on this information. * A Diag Reset will also call mps_iocfacts_allocate and re-read the IOC * Facts. If relevant values have changed in IOC Facts, this function * will free all of the memory based on IOC Facts and reallocate that * memory. If this fails, any allocated memory should already be freed. */ if ((error = mps_iocfacts_allocate(sc, TRUE)) != 0) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "IOC Facts based allocation " "failed with error %d, exit\n", error); return (error); } /* Start the periodic watchdog check on the IOC Doorbell */ mps_periodic(sc); /* * The portenable will kick off discovery events that will drive the * rest of the initialization process. The CAM/SAS module will * hold up the boot sequence until discovery is complete. */ sc->mps_ich.ich_func = mps_startup; sc->mps_ich.ich_arg = sc; if (config_intrhook_establish(&sc->mps_ich) != 0) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Cannot establish MPS config hook\n"); error = EINVAL; } /* * Allow IR to shutdown gracefully when shutdown occurs. */ sc->shutdown_eh = EVENTHANDLER_REGISTER(shutdown_final, mpssas_ir_shutdown, sc, SHUTDOWN_PRI_DEFAULT); if (sc->shutdown_eh == NULL) mps_dprint(sc, MPS_INIT|MPS_ERROR, "shutdown event registration failed\n"); mps_setup_sysctl(sc); sc->mps_flags |= MPS_FLAGS_ATTACH_DONE; mps_dprint(sc, MPS_INIT, "%s exit error= %d\n", __func__, error); return (error); } /* Run through any late-start handlers. */ static void mps_startup(void *arg) { struct mps_softc *sc; sc = (struct mps_softc *)arg; mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); mps_lock(sc); mps_unmask_intr(sc); /* initialize device mapping tables */ mps_base_static_config_pages(sc); mps_mapping_initialize(sc); mpssas_startup(sc); mps_unlock(sc); mps_dprint(sc, MPS_INIT, "disestablish config intrhook\n"); config_intrhook_disestablish(&sc->mps_ich); sc->mps_ich.ich_arg = NULL; mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); } /* Periodic watchdog. Is called with the driver lock already held. */ static void mps_periodic(void *arg) { struct mps_softc *sc; uint32_t db; sc = (struct mps_softc *)arg; if (sc->mps_flags & MPS_FLAGS_SHUTDOWN) return; db = mps_regread(sc, MPI2_DOORBELL_OFFSET); if ((db & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) { mps_dprint(sc, MPS_FAULT, "IOC Fault 0x%08x, Resetting\n", db); mps_reinit(sc); } callout_reset(&sc->periodic, MPS_PERIODIC_DELAY * hz, mps_periodic, sc); } static void mps_log_evt_handler(struct mps_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event) { MPI2_EVENT_DATA_LOG_ENTRY_ADDED *entry; MPS_DPRINT_EVENT(sc, generic, event); switch (event->Event) { case MPI2_EVENT_LOG_DATA: mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_DATA:\n"); if (sc->mps_debug & MPS_EVENT) hexdump(event->EventData, event->EventDataLength, NULL, 0); break; case MPI2_EVENT_LOG_ENTRY_ADDED: entry = (MPI2_EVENT_DATA_LOG_ENTRY_ADDED *)event->EventData; mps_dprint(sc, MPS_EVENT, "MPI2_EVENT_LOG_ENTRY_ADDED event " "0x%x Sequence %d:\n", entry->LogEntryQualifier, entry->LogSequence); break; default: break; } return; } static int mps_attach_log(struct mps_softc *sc) { u32 events[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS]; bzero(events, 16); setbit(events, MPI2_EVENT_LOG_DATA); setbit(events, MPI2_EVENT_LOG_ENTRY_ADDED); mps_register_events(sc, events, mps_log_evt_handler, NULL, &sc->mps_log_eh); return (0); } static int mps_detach_log(struct mps_softc *sc) { if (sc->mps_log_eh != NULL) mps_deregister_events(sc, sc->mps_log_eh); return (0); } /* * Free all of the driver resources and detach submodules. Should be called * without the lock held. */ int mps_free(struct mps_softc *sc) { int error; mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); /* Turn off the watchdog */ mps_lock(sc); sc->mps_flags |= MPS_FLAGS_SHUTDOWN; mps_unlock(sc); /* Lock must not be held for this */ callout_drain(&sc->periodic); callout_drain(&sc->device_check_callout); if (((error = mps_detach_log(sc)) != 0) || ((error = mps_detach_sas(sc)) != 0)) { mps_dprint(sc, MPS_INIT|MPS_FAULT, "failed to detach " "subsystems, exit\n"); return (error); } mps_detach_user(sc); /* Put the IOC back in the READY state. */ mps_lock(sc); if ((error = mps_transition_ready(sc)) != 0) { mps_unlock(sc); return (error); } mps_unlock(sc); if (sc->facts != NULL) free(sc->facts, M_MPT2); /* * Free all buffers that are based on IOC Facts. A Diag Reset may need * to free these buffers too. */ mps_iocfacts_free(sc); if (sc->sysctl_tree != NULL) sysctl_ctx_free(&sc->sysctl_ctx); /* Deregister the shutdown function */ if (sc->shutdown_eh != NULL) EVENTHANDLER_DEREGISTER(shutdown_final, sc->shutdown_eh); mtx_destroy(&sc->mps_mtx); mps_dprint(sc, MPS_INIT, "%s exit\n", __func__); return (0); } static __inline void mps_complete_command(struct mps_softc *sc, struct mps_command *cm) { MPS_FUNCTRACE(sc); if (cm == NULL) { mps_dprint(sc, MPS_ERROR, "Completing NULL command\n"); return; } if (cm->cm_flags & MPS_CM_FLAGS_POLLED) cm->cm_flags |= MPS_CM_FLAGS_COMPLETE; if (cm->cm_complete != NULL) { mps_dprint(sc, MPS_TRACE, "%s cm %p calling cm_complete %p data %p reply %p\n", __func__, cm, cm->cm_complete, cm->cm_complete_data, cm->cm_reply); cm->cm_complete(sc, cm); } if (cm->cm_flags & MPS_CM_FLAGS_WAKEUP) { mps_dprint(sc, MPS_TRACE, "waking up %p\n", cm); wakeup(cm); } if (cm->cm_sc->io_cmds_active != 0) { cm->cm_sc->io_cmds_active--; } else { mps_dprint(sc, MPS_ERROR, "Warning: io_cmds_active is " "out of sync - resynching to 0\n"); } } static void mps_sas_log_info(struct mps_softc *sc , u32 log_info) { union loginfo_type { u32 loginfo; struct { u32 subcode:16; u32 code:8; u32 originator:4; u32 bus_type:4; } dw; }; union loginfo_type sas_loginfo; char *originator_str = NULL; sas_loginfo.loginfo = log_info; if (sas_loginfo.dw.bus_type != 3 /*SAS*/) return; /* each nexus loss loginfo */ if (log_info == 0x31170000) return; /* eat the loginfos associated with task aborts */ if ((log_info == 30050000 || log_info == 0x31140000 || log_info == 0x31130000)) return; switch (sas_loginfo.dw.originator) { case 0: originator_str = "IOP"; break; case 1: originator_str = "PL"; break; case 2: originator_str = "IR"; break; } mps_dprint(sc, MPS_LOG, "log_info(0x%08x): originator(%s), " "code(0x%02x), sub_code(0x%04x)\n", log_info, originator_str, sas_loginfo.dw.code, sas_loginfo.dw.subcode); } static void mps_display_reply_info(struct mps_softc *sc, uint8_t *reply) { MPI2DefaultReply_t *mpi_reply; u16 sc_status; mpi_reply = (MPI2DefaultReply_t*)reply; sc_status = le16toh(mpi_reply->IOCStatus); if (sc_status & MPI2_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE) mps_sas_log_info(sc, le32toh(mpi_reply->IOCLogInfo)); } void mps_intr(void *data) { struct mps_softc *sc; uint32_t status; sc = (struct mps_softc *)data; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); /* * Check interrupt status register to flush the bus. This is * needed for both INTx interrupts and driver-driven polling */ status = mps_regread(sc, MPI2_HOST_INTERRUPT_STATUS_OFFSET); if ((status & MPI2_HIS_REPLY_DESCRIPTOR_INTERRUPT) == 0) return; mps_lock(sc); mps_intr_locked(data); mps_unlock(sc); return; } /* * In theory, MSI/MSIX interrupts shouldn't need to read any registers on the * chip. Hopefully this theory is correct. */ void mps_intr_msi(void *data) { struct mps_softc *sc; sc = (struct mps_softc *)data; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); mps_lock(sc); mps_intr_locked(data); mps_unlock(sc); return; } /* * The locking is overly broad and simplistic, but easy to deal with for now. */ void mps_intr_locked(void *data) { MPI2_REPLY_DESCRIPTORS_UNION *desc; MPI2_DIAG_RELEASE_REPLY *rel_rep; mps_fw_diagnostic_buffer_t *pBuffer; struct mps_softc *sc; struct mps_command *cm = NULL; uint64_t tdesc; uint8_t flags; u_int pq; sc = (struct mps_softc *)data; pq = sc->replypostindex; mps_dprint(sc, MPS_TRACE, "%s sc %p starting with replypostindex %u\n", __func__, sc, sc->replypostindex); for ( ;; ) { cm = NULL; desc = &sc->post_queue[sc->replypostindex]; /* * Copy and clear out the descriptor so that any reentry will * immediately know that this descriptor has already been * looked at. There is unfortunate casting magic because the * MPI API doesn't have a cardinal 64bit type. */ tdesc = 0xffffffffffffffff; tdesc = atomic_swap_64((uint64_t *)desc, tdesc); desc = (MPI2_REPLY_DESCRIPTORS_UNION *)&tdesc; flags = desc->Default.ReplyFlags & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK; if ((flags == MPI2_RPY_DESCRIPT_FLAGS_UNUSED) || (le32toh(desc->Words.High) == 0xffffffff)) break; /* increment the replypostindex now, so that event handlers * and cm completion handlers which decide to do a diag * reset can zero it without it getting incremented again * afterwards, and we break out of this loop on the next * iteration since the reply post queue has been cleared to * 0xFF and all descriptors look unused (which they are). */ if (++sc->replypostindex >= sc->pqdepth) sc->replypostindex = 0; switch (flags) { case MPI2_RPY_DESCRIPT_FLAGS_SCSI_IO_SUCCESS: cm = &sc->commands[le16toh(desc->SCSIIOSuccess.SMID)]; KASSERT(cm->cm_state == MPS_CM_STATE_INQUEUE, ("command not inqueue\n")); cm->cm_state = MPS_CM_STATE_BUSY; cm->cm_reply = NULL; break; case MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY: { uint32_t baddr; uint8_t *reply; /* * Re-compose the reply address from the address * sent back from the chip. The ReplyFrameAddress * is the lower 32 bits of the physical address of * particular reply frame. Convert that address to * host format, and then use that to provide the * offset against the virtual address base * (sc->reply_frames). */ baddr = le32toh(desc->AddressReply.ReplyFrameAddress); reply = sc->reply_frames + (baddr - ((uint32_t)sc->reply_busaddr)); /* * Make sure the reply we got back is in a valid * range. If not, go ahead and panic here, since * we'll probably panic as soon as we deference the * reply pointer anyway. */ if ((reply < sc->reply_frames) || (reply > (sc->reply_frames + (sc->fqdepth * sc->replyframesz)))) { printf("%s: WARNING: reply %p out of range!\n", __func__, reply); printf("%s: reply_frames %p, fqdepth %d, " "frame size %d\n", __func__, sc->reply_frames, sc->fqdepth, sc->replyframesz); printf("%s: baddr %#x,\n", __func__, baddr); /* LSI-TODO. See Linux Code for Graceful exit */ panic("Reply address out of range"); } if (le16toh(desc->AddressReply.SMID) == 0) { if (((MPI2_DEFAULT_REPLY *)reply)->Function == MPI2_FUNCTION_DIAG_BUFFER_POST) { /* * If SMID is 0 for Diag Buffer Post, * this implies that the reply is due to * a release function with a status that * the buffer has been released. Set * the buffer flags accordingly. */ rel_rep = (MPI2_DIAG_RELEASE_REPLY *)reply; if ((le16toh(rel_rep->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_DIAGNOSTIC_RELEASED) { pBuffer = &sc->fw_diag_buffer_list[ rel_rep->BufferType]; pBuffer->valid_data = TRUE; pBuffer->owned_by_firmware = FALSE; pBuffer->immediate = FALSE; } } else mps_dispatch_event(sc, baddr, (MPI2_EVENT_NOTIFICATION_REPLY *) reply); } else { + /* + * Ignore commands not in INQUEUE state + * since they've already been completed + * via another path. + */ cm = &sc->commands[ le16toh(desc->AddressReply.SMID)]; - if (cm->cm_state != MPS_CM_STATE_TIMEDOUT) + if (cm->cm_state == MPS_CM_STATE_INQUEUE) { cm->cm_state = MPS_CM_STATE_BUSY; - cm->cm_reply = reply; - cm->cm_reply_data = le32toh( - desc->AddressReply.ReplyFrameAddress); + cm->cm_reply = reply; + cm->cm_reply_data = le32toh( + desc->AddressReply.ReplyFrameAddress); + } else { + mps_dprint(sc, MPS_RECOVERY, + "Bad state for ADDRESS_REPLY status," + " ignoring state %d cm %p\n", + cm->cm_state, cm); + } } break; } case MPI2_RPY_DESCRIPT_FLAGS_TARGETASSIST_SUCCESS: case MPI2_RPY_DESCRIPT_FLAGS_TARGET_COMMAND_BUFFER: case MPI2_RPY_DESCRIPT_FLAGS_RAID_ACCELERATOR_SUCCESS: default: /* Unhandled */ mps_dprint(sc, MPS_ERROR, "Unhandled reply 0x%x\n", desc->Default.ReplyFlags); cm = NULL; break; } if (cm != NULL) { // Print Error reply frame if (cm->cm_reply) mps_display_reply_info(sc,cm->cm_reply); mps_complete_command(sc, cm); } } if (pq != sc->replypostindex) { mps_dprint(sc, MPS_TRACE, "%s sc %p writing postindex %d\n", __func__, sc, sc->replypostindex); mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, sc->replypostindex); } return; } static void mps_dispatch_event(struct mps_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *reply) { struct mps_event_handle *eh; int event, handled = 0; event = le16toh(reply->Event); TAILQ_FOREACH(eh, &sc->event_list, eh_list) { if (isset(eh->mask, event)) { eh->callback(sc, data, reply); handled++; } } if (handled == 0) mps_dprint(sc, MPS_EVENT, "Unhandled event 0x%x\n", le16toh(event)); /* * This is the only place that the event/reply should be freed. * Anything wanting to hold onto the event data should have * already copied it into their own storage. */ mps_free_reply(sc, data); } static void mps_reregister_events_complete(struct mps_softc *sc, struct mps_command *cm) { mps_dprint(sc, MPS_TRACE, "%s\n", __func__); if (cm->cm_reply) MPS_DPRINT_EVENT(sc, generic, (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply); mps_free_command(sc, cm); /* next, send a port enable */ mpssas_startup(sc); } /* * For both register_events and update_events, the caller supplies a bitmap * of events that it _wants_. These functions then turn that into a bitmask * suitable for the controller. */ int mps_register_events(struct mps_softc *sc, u32 *mask, mps_evt_callback_t *cb, void *data, struct mps_event_handle **handle) { struct mps_event_handle *eh; int error = 0; eh = malloc(sizeof(struct mps_event_handle), M_MPT2, M_WAITOK|M_ZERO); if(!eh) { mps_dprint(sc, MPS_ERROR, "Cannot allocate event memory\n"); return (ENOMEM); } eh->callback = cb; eh->data = data; TAILQ_INSERT_TAIL(&sc->event_list, eh, eh_list); if (mask != NULL) error = mps_update_events(sc, eh, mask); *handle = eh; return (error); } int mps_update_events(struct mps_softc *sc, struct mps_event_handle *handle, u32 *mask) { MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; MPI2_EVENT_NOTIFICATION_REPLY *reply = NULL; struct mps_command *cm; int error, i; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); if ((mask != NULL) && (handle != NULL)) bcopy(mask, &handle->mask[0], sizeof(u32) * MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) sc->event_mask[i] = -1; for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) sc->event_mask[i] &= ~handle->mask[i]; if ((cm = mps_alloc_command(sc)) == NULL) return (EBUSY); evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; evtreq->MsgFlags = 0; evtreq->SASBroadcastPrimitiveMasks = 0; #ifdef MPS_DEBUG_ALL_EVENTS { u_char fullmask[16]; memset(fullmask, 0x00, 16); bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) * MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); } #else for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) evtreq->EventMasks[i] = htole32(sc->event_mask[i]); #endif cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = NULL; error = mps_wait_command(sc, &cm, 60, 0); if (cm != NULL) reply = (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply; if ((reply == NULL) || (reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) error = ENXIO; if (reply) MPS_DPRINT_EVENT(sc, generic, reply); mps_dprint(sc, MPS_TRACE, "%s finished error %d\n", __func__, error); if (cm != NULL) mps_free_command(sc, cm); return (error); } static int mps_reregister_events(struct mps_softc *sc) { MPI2_EVENT_NOTIFICATION_REQUEST *evtreq; struct mps_command *cm; struct mps_event_handle *eh; int error, i; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); /* first, reregister events */ for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) sc->event_mask[i] = -1; TAILQ_FOREACH(eh, &sc->event_list, eh_list) { for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) sc->event_mask[i] &= ~eh->mask[i]; } if ((cm = mps_alloc_command(sc)) == NULL) return (EBUSY); evtreq = (MPI2_EVENT_NOTIFICATION_REQUEST *)cm->cm_req; evtreq->Function = MPI2_FUNCTION_EVENT_NOTIFICATION; evtreq->MsgFlags = 0; evtreq->SASBroadcastPrimitiveMasks = 0; #ifdef MPS_DEBUG_ALL_EVENTS { u_char fullmask[16]; memset(fullmask, 0x00, 16); bcopy(fullmask, &evtreq->EventMasks[0], sizeof(u32) * MPI2_EVENT_NOTIFY_EVENTMASK_WORDS); } #else for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) evtreq->EventMasks[i] = htole32(sc->event_mask[i]); #endif cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = NULL; cm->cm_complete = mps_reregister_events_complete; error = mps_map_command(sc, cm); mps_dprint(sc, MPS_TRACE, "%s finished with error %d\n", __func__, error); return (error); } void mps_deregister_events(struct mps_softc *sc, struct mps_event_handle *handle) { TAILQ_REMOVE(&sc->event_list, handle, eh_list); free(handle, M_MPT2); } /* * Add a chain element as the next SGE for the specified command. * Reset cm_sge and cm_sgesize to indicate all the available space. */ static int mps_add_chain(struct mps_command *cm) { MPI2_SGE_CHAIN32 *sgc; struct mps_chain *chain; u_int space; if (cm->cm_sglsize < MPS_SGC_SIZE) panic("MPS: Need SGE Error Code\n"); chain = mps_alloc_chain(cm->cm_sc); if (chain == NULL) return (ENOBUFS); space = cm->cm_sc->reqframesz; /* * Note: a double-linked list is used to make it easier to * walk for debugging. */ TAILQ_INSERT_TAIL(&cm->cm_chain_list, chain, chain_link); sgc = (MPI2_SGE_CHAIN32 *)&cm->cm_sge->MpiChain; sgc->Length = htole16(space); sgc->NextChainOffset = 0; /* TODO Looks like bug in Setting sgc->Flags. * sgc->Flags = ( MPI2_SGE_FLAGS_CHAIN_ELEMENT | MPI2_SGE_FLAGS_64_BIT_ADDRESSING | * MPI2_SGE_FLAGS_SYSTEM_ADDRESS) << MPI2_SGE_FLAGS_SHIFT * This is fine.. because we are not using simple element. In case of * MPI2_SGE_CHAIN32, we have separate Length and Flags feild. */ sgc->Flags = MPI2_SGE_FLAGS_CHAIN_ELEMENT; sgc->Address = htole32(chain->chain_busaddr); cm->cm_sge = (MPI2_SGE_IO_UNION *)&chain->chain->MpiSimple; cm->cm_sglsize = space; return (0); } /* * Add one scatter-gather element (chain, simple, transaction context) * to the scatter-gather list for a command. Maintain cm_sglsize and * cm_sge as the remaining size and pointer to the next SGE to fill * in, respectively. */ int mps_push_sge(struct mps_command *cm, void *sgep, size_t len, int segsleft) { MPI2_SGE_TRANSACTION_UNION *tc = sgep; MPI2_SGE_SIMPLE64 *sge = sgep; int error, type; uint32_t saved_buf_len, saved_address_low, saved_address_high; type = (tc->Flags & MPI2_SGE_FLAGS_ELEMENT_MASK); #ifdef INVARIANTS switch (type) { case MPI2_SGE_FLAGS_TRANSACTION_ELEMENT: { if (len != tc->DetailsLength + 4) panic("TC %p length %u or %zu?", tc, tc->DetailsLength + 4, len); } break; case MPI2_SGE_FLAGS_CHAIN_ELEMENT: /* Driver only uses 32-bit chain elements */ if (len != MPS_SGC_SIZE) panic("CHAIN %p length %u or %zu?", sgep, MPS_SGC_SIZE, len); break; case MPI2_SGE_FLAGS_SIMPLE_ELEMENT: /* Driver only uses 64-bit SGE simple elements */ if (len != MPS_SGE64_SIZE) panic("SGE simple %p length %u or %zu?", sge, MPS_SGE64_SIZE, len); if (((le32toh(sge->FlagsLength) >> MPI2_SGE_FLAGS_SHIFT) & MPI2_SGE_FLAGS_ADDRESS_SIZE) == 0) panic("SGE simple %p not marked 64-bit?", sge); break; default: panic("Unexpected SGE %p, flags %02x", tc, tc->Flags); } #endif /* * case 1: 1 more segment, enough room for it * case 2: 2 more segments, enough room for both * case 3: >=2 more segments, only enough room for 1 and a chain * case 4: >=1 more segment, enough room for only a chain * case 5: >=1 more segment, no room for anything (error) */ /* * There should be room for at least a chain element, or this * code is buggy. Case (5). */ if (cm->cm_sglsize < MPS_SGC_SIZE) panic("MPS: Need SGE Error Code\n"); if (segsleft >= 1 && cm->cm_sglsize < len + MPS_SGC_SIZE) { /* * 1 or more segment, enough room for only a chain. * Hope the previous element wasn't a Simple entry * that needed to be marked with * MPI2_SGE_FLAGS_LAST_ELEMENT. Case (4). */ if ((error = mps_add_chain(cm)) != 0) return (error); } if (segsleft >= 2 && cm->cm_sglsize < len + MPS_SGC_SIZE + MPS_SGE64_SIZE) { /* * There are 2 or more segments left to add, and only * enough room for 1 and a chain. Case (3). * * Mark as last element in this chain if necessary. */ if (type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) { sge->FlagsLength |= htole32( MPI2_SGE_FLAGS_LAST_ELEMENT << MPI2_SGE_FLAGS_SHIFT); } /* * Add the item then a chain. Do the chain now, * rather than on the next iteration, to simplify * understanding the code. */ cm->cm_sglsize -= len; bcopy(sgep, cm->cm_sge, len); cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); return (mps_add_chain(cm)); } #ifdef INVARIANTS /* Case 1: 1 more segment, enough room for it. */ if (segsleft == 1 && cm->cm_sglsize < len) panic("1 seg left and no room? %u versus %zu", cm->cm_sglsize, len); /* Case 2: 2 more segments, enough room for both */ if (segsleft == 2 && cm->cm_sglsize < len + MPS_SGE64_SIZE) panic("2 segs left and no room? %u versus %zu", cm->cm_sglsize, len); #endif if (segsleft == 1 && type == MPI2_SGE_FLAGS_SIMPLE_ELEMENT) { /* * If this is a bi-directional request, need to account for that * here. Save the pre-filled sge values. These will be used * either for the 2nd SGL or for a single direction SGL. If * cm_out_len is non-zero, this is a bi-directional request, so * fill in the OUT SGL first, then the IN SGL, otherwise just * fill in the IN SGL. Note that at this time, when filling in * 2 SGL's for a bi-directional request, they both use the same * DMA buffer (same cm command). */ saved_buf_len = le32toh(sge->FlagsLength) & 0x00FFFFFF; saved_address_low = sge->Address.Low; saved_address_high = sge->Address.High; if (cm->cm_out_len) { sge->FlagsLength = htole32(cm->cm_out_len | ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_HOST_TO_IOC | MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << MPI2_SGE_FLAGS_SHIFT)); cm->cm_sglsize -= len; bcopy(sgep, cm->cm_sge, len); cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); } saved_buf_len |= ((uint32_t)(MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_LAST_ELEMENT | MPI2_SGE_FLAGS_END_OF_LIST | MPI2_SGE_FLAGS_64_BIT_ADDRESSING) << MPI2_SGE_FLAGS_SHIFT); if (cm->cm_flags & MPS_CM_FLAGS_DATAIN) { saved_buf_len |= ((uint32_t)(MPI2_SGE_FLAGS_IOC_TO_HOST) << MPI2_SGE_FLAGS_SHIFT); } else { saved_buf_len |= ((uint32_t)(MPI2_SGE_FLAGS_HOST_TO_IOC) << MPI2_SGE_FLAGS_SHIFT); } sge->FlagsLength = htole32(saved_buf_len); sge->Address.Low = saved_address_low; sge->Address.High = saved_address_high; } cm->cm_sglsize -= len; bcopy(sgep, cm->cm_sge, len); cm->cm_sge = (MPI2_SGE_IO_UNION *)((uintptr_t)cm->cm_sge + len); return (0); } /* * Add one dma segment to the scatter-gather list for a command. */ int mps_add_dmaseg(struct mps_command *cm, vm_paddr_t pa, size_t len, u_int flags, int segsleft) { MPI2_SGE_SIMPLE64 sge; /* * This driver always uses 64-bit address elements for simplicity. */ bzero(&sge, sizeof(sge)); flags |= MPI2_SGE_FLAGS_SIMPLE_ELEMENT | MPI2_SGE_FLAGS_64_BIT_ADDRESSING; sge.FlagsLength = htole32(len | (flags << MPI2_SGE_FLAGS_SHIFT)); mps_from_u64(pa, &sge.Address); return (mps_push_sge(cm, &sge, sizeof sge, segsleft)); } static void mps_data_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct mps_softc *sc; struct mps_command *cm; u_int i, dir, sflags; cm = (struct mps_command *)arg; sc = cm->cm_sc; /* * In this case, just print out a warning and let the chip tell the * user they did the wrong thing. */ if ((cm->cm_max_segs != 0) && (nsegs > cm->cm_max_segs)) { mps_dprint(sc, MPS_ERROR, "%s: warning: busdma returned %d segments, " "more than the %d allowed\n", __func__, nsegs, cm->cm_max_segs); } /* * Set up DMA direction flags. Bi-directional requests are also handled * here. In that case, both direction flags will be set. */ sflags = 0; if (cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) { /* * We have to add a special case for SMP passthrough, there * is no easy way to generically handle it. The first * S/G element is used for the command (therefore the * direction bit needs to be set). The second one is used * for the reply. We'll leave it to the caller to make * sure we only have two buffers. */ /* * Even though the busdma man page says it doesn't make * sense to have both direction flags, it does in this case. * We have one s/g element being accessed in each direction. */ dir = BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD; /* * Set the direction flag on the first buffer in the SMP * passthrough request. We'll clear it for the second one. */ sflags |= MPI2_SGE_FLAGS_DIRECTION | MPI2_SGE_FLAGS_END_OF_BUFFER; } else if (cm->cm_flags & MPS_CM_FLAGS_DATAOUT) { sflags |= MPI2_SGE_FLAGS_HOST_TO_IOC; dir = BUS_DMASYNC_PREWRITE; } else dir = BUS_DMASYNC_PREREAD; for (i = 0; i < nsegs; i++) { if ((cm->cm_flags & MPS_CM_FLAGS_SMP_PASS) && (i != 0)) { sflags &= ~MPI2_SGE_FLAGS_DIRECTION; } error = mps_add_dmaseg(cm, segs[i].ds_addr, segs[i].ds_len, sflags, nsegs - i); if (error != 0) { /* Resource shortage, roll back! */ if (ratecheck(&sc->lastfail, &mps_chainfail_interval)) mps_dprint(sc, MPS_INFO, "Out of chain frames, " "consider increasing hw.mps.max_chains.\n"); cm->cm_flags |= MPS_CM_FLAGS_CHAIN_FAILED; mps_complete_command(sc, cm); return; } } bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir); mps_enqueue_request(sc, cm); return; } static void mps_data_cb2(void *arg, bus_dma_segment_t *segs, int nsegs, bus_size_t mapsize, int error) { mps_data_cb(arg, segs, nsegs, error); } /* * This is the routine to enqueue commands ansynchronously. * Note that the only error path here is from bus_dmamap_load(), which can * return EINPROGRESS if it is waiting for resources. Other than this, it's * assumed that if you have a command in-hand, then you have enough credits * to use it. */ int mps_map_command(struct mps_softc *sc, struct mps_command *cm) { int error = 0; if (cm->cm_flags & MPS_CM_FLAGS_USE_UIO) { error = bus_dmamap_load_uio(sc->buffer_dmat, cm->cm_dmamap, &cm->cm_uio, mps_data_cb2, cm, 0); } else if (cm->cm_flags & MPS_CM_FLAGS_USE_CCB) { error = bus_dmamap_load_ccb(sc->buffer_dmat, cm->cm_dmamap, cm->cm_data, mps_data_cb, cm, 0); } else if ((cm->cm_data != NULL) && (cm->cm_length != 0)) { error = bus_dmamap_load(sc->buffer_dmat, cm->cm_dmamap, cm->cm_data, cm->cm_length, mps_data_cb, cm, 0); } else { /* Add a zero-length element as needed */ if (cm->cm_sge != NULL) mps_add_dmaseg(cm, 0, 0, 0, 1); mps_enqueue_request(sc, cm); } return (error); } /* * This is the routine to enqueue commands synchronously. An error of * EINPROGRESS from mps_map_command() is ignored since the command will * be executed and enqueued automatically. Other errors come from msleep(). */ int mps_wait_command(struct mps_softc *sc, struct mps_command **cmp, int timeout, int sleep_flag) { int error, rc; struct timeval cur_time, start_time; struct mps_command *cm = *cmp; if (sc->mps_flags & MPS_FLAGS_DIAGRESET) return EBUSY; cm->cm_complete = NULL; cm->cm_flags |= MPS_CM_FLAGS_POLLED; error = mps_map_command(sc, cm); if ((error != 0) && (error != EINPROGRESS)) return (error); /* * Check for context and wait for 50 mSec at a time until time has * expired or the command has finished. If msleep can't be used, need * to poll. */ if (curthread->td_no_sleeping != 0) sleep_flag = NO_SLEEP; getmicrouptime(&start_time); if (mtx_owned(&sc->mps_mtx) && sleep_flag == CAN_SLEEP) { cm->cm_flags |= MPS_CM_FLAGS_WAKEUP; error = msleep(cm, &sc->mps_mtx, 0, "mpswait", timeout*hz); if (error == EWOULDBLOCK) { /* * Record the actual elapsed time in the case of a * timeout for the message below. */ getmicrouptime(&cur_time); timevalsub(&cur_time, &start_time); } } else { while ((cm->cm_flags & MPS_CM_FLAGS_COMPLETE) == 0) { mps_intr_locked(sc); if (sleep_flag == CAN_SLEEP) pause("mpswait", hz/20); else DELAY(50000); getmicrouptime(&cur_time); timevalsub(&cur_time, &start_time); if (cur_time.tv_sec > timeout) { error = EWOULDBLOCK; break; } } } if (error == EWOULDBLOCK) { if (cm->cm_timeout_handler == NULL) { mps_dprint(sc, MPS_FAULT, "Calling Reinit from %s, timeout=%d," " elapsed=%jd\n", __func__, timeout, (intmax_t)cur_time.tv_sec); rc = mps_reinit(sc); mps_dprint(sc, MPS_FAULT, "Reinit %s\n", (rc == 0) ? "success" : "failed"); } else cm->cm_timeout_handler(sc, cm); if (sc->mps_flags & MPS_FLAGS_REALLOCATED) { /* * Tell the caller that we freed the command in a * reinit. */ *cmp = NULL; } error = ETIMEDOUT; } return (error); } /* * The MPT driver had a verbose interface for config pages. In this driver, * reduce it to much simpler terms, similar to the Linux driver. */ int mps_read_config_page(struct mps_softc *sc, struct mps_config_params *params) { MPI2_CONFIG_REQUEST *req; struct mps_command *cm; int error; if (sc->mps_flags & MPS_FLAGS_BUSY) { return (EBUSY); } cm = mps_alloc_command(sc); if (cm == NULL) { return (EBUSY); } req = (MPI2_CONFIG_REQUEST *)cm->cm_req; req->Function = MPI2_FUNCTION_CONFIG; req->Action = params->action; req->SGLFlags = 0; req->ChainOffset = 0; req->PageAddress = params->page_address; if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { MPI2_CONFIG_EXTENDED_PAGE_HEADER *hdr; hdr = ¶ms->hdr.Ext; req->ExtPageType = hdr->ExtPageType; req->ExtPageLength = hdr->ExtPageLength; req->Header.PageType = MPI2_CONFIG_PAGETYPE_EXTENDED; req->Header.PageLength = 0; /* Must be set to zero */ req->Header.PageNumber = hdr->PageNumber; req->Header.PageVersion = hdr->PageVersion; } else { MPI2_CONFIG_PAGE_HEADER *hdr; hdr = ¶ms->hdr.Struct; req->Header.PageType = hdr->PageType; req->Header.PageNumber = hdr->PageNumber; req->Header.PageLength = hdr->PageLength; req->Header.PageVersion = hdr->PageVersion; } cm->cm_data = params->buffer; cm->cm_length = params->length; if (cm->cm_data != NULL) { cm->cm_sge = &req->PageBufferSGE; cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION); cm->cm_flags = MPS_CM_FLAGS_SGE_SIMPLE | MPS_CM_FLAGS_DATAIN; } else cm->cm_sge = NULL; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete_data = params; if (params->callback != NULL) { cm->cm_complete = mps_config_complete; return (mps_map_command(sc, cm)); } else { error = mps_wait_command(sc, &cm, 0, CAN_SLEEP); if (error) { mps_dprint(sc, MPS_FAULT, "Error %d reading config page\n", error); if (cm != NULL) mps_free_command(sc, cm); return (error); } mps_config_complete(sc, cm); } return (0); } int mps_write_config_page(struct mps_softc *sc, struct mps_config_params *params) { return (EINVAL); } static void mps_config_complete(struct mps_softc *sc, struct mps_command *cm) { MPI2_CONFIG_REPLY *reply; struct mps_config_params *params; MPS_FUNCTRACE(sc); params = cm->cm_complete_data; if (cm->cm_data != NULL) { bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); } /* * XXX KDM need to do more error recovery? This results in the * device in question not getting probed. */ if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { params->status = MPI2_IOCSTATUS_BUSY; goto done; } reply = (MPI2_CONFIG_REPLY *)cm->cm_reply; if (reply == NULL) { params->status = MPI2_IOCSTATUS_BUSY; goto done; } params->status = reply->IOCStatus; if (params->hdr.Struct.PageType == MPI2_CONFIG_PAGETYPE_EXTENDED) { params->hdr.Ext.ExtPageType = reply->ExtPageType; params->hdr.Ext.ExtPageLength = reply->ExtPageLength; params->hdr.Ext.PageType = reply->Header.PageType; params->hdr.Ext.PageNumber = reply->Header.PageNumber; params->hdr.Ext.PageVersion = reply->Header.PageVersion; } else { params->hdr.Struct.PageType = reply->Header.PageType; params->hdr.Struct.PageNumber = reply->Header.PageNumber; params->hdr.Struct.PageLength = reply->Header.PageLength; params->hdr.Struct.PageVersion = reply->Header.PageVersion; } done: mps_free_command(sc, cm); if (params->callback != NULL) params->callback(sc, params); return; } Index: head/sys/dev/mps/mps_sas.c =================================================================== --- head/sys/dev/mps/mps_sas.c (revision 349848) +++ head/sys/dev/mps/mps_sas.c (revision 349849) @@ -1,3639 +1,3641 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2015 Avago Technologies * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT2 */ /* TODO Move headers to mpsvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if __FreeBSD_version >= 900026 #include #endif #include #include #include #include #include #include #include #include #include #include #include #define MPSSAS_DISCOVERY_TIMEOUT 20 #define MPSSAS_MAX_DISCOVERY_TIMEOUTS 10 /* 200 seconds */ /* * static array to check SCSI OpCode for EEDP protection bits */ #define PRO_R MPI2_SCSIIO_EEDPFLAGS_CHECK_REMOVE_OP #define PRO_W MPI2_SCSIIO_EEDPFLAGS_INSERT_OP #define PRO_V MPI2_SCSIIO_EEDPFLAGS_INSERT_OP static uint8_t op_code_prot[256] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, PRO_W, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, PRO_R, 0, PRO_W, 0, 0, 0, PRO_W, PRO_V, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; MALLOC_DEFINE(M_MPSSAS, "MPSSAS", "MPS SAS memory"); static void mpssas_remove_device(struct mps_softc *, struct mps_command *); static void mpssas_remove_complete(struct mps_softc *, struct mps_command *); static void mpssas_action(struct cam_sim *sim, union ccb *ccb); static void mpssas_poll(struct cam_sim *sim); static int mpssas_send_abort(struct mps_softc *sc, struct mps_command *tm, struct mps_command *cm); static void mpssas_scsiio_timeout(void *data); static void mpssas_abort_complete(struct mps_softc *sc, struct mps_command *cm); static void mpssas_direct_drive_io(struct mpssas_softc *sassc, struct mps_command *cm, union ccb *ccb); static void mpssas_action_scsiio(struct mpssas_softc *, union ccb *); static void mpssas_scsiio_complete(struct mps_softc *, struct mps_command *); static void mpssas_action_resetdev(struct mpssas_softc *, union ccb *); #if __FreeBSD_version >= 900026 static void mpssas_smpio_complete(struct mps_softc *sc, struct mps_command *cm); static void mpssas_send_smpcmd(struct mpssas_softc *sassc, union ccb *ccb, uint64_t sasaddr); static void mpssas_action_smpio(struct mpssas_softc *sassc, union ccb *ccb); #endif //FreeBSD_version >= 900026 static void mpssas_resetdev_complete(struct mps_softc *, struct mps_command *); static void mpssas_async(void *callback_arg, uint32_t code, struct cam_path *path, void *arg); #if (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) static void mpssas_check_eedp(struct mps_softc *sc, struct cam_path *path, struct ccb_getdev *cgd); static void mpssas_read_cap_done(struct cam_periph *periph, union ccb *done_ccb); #endif static int mpssas_send_portenable(struct mps_softc *sc); static void mpssas_portenable_complete(struct mps_softc *sc, struct mps_command *cm); struct mpssas_target * mpssas_find_target_by_handle(struct mpssas_softc *sassc, int start, uint16_t handle) { struct mpssas_target *target; int i; for (i = start; i < sassc->maxtargets; i++) { target = &sassc->targets[i]; if (target->handle == handle) return (target); } return (NULL); } /* we need to freeze the simq during attach and diag reset, to avoid failing * commands before device handles have been found by discovery. Since * discovery involves reading config pages and possibly sending commands, * discovery actions may continue even after we receive the end of discovery * event, so refcount discovery actions instead of assuming we can unfreeze * the simq when we get the event. */ void mpssas_startup_increment(struct mpssas_softc *sassc) { MPS_FUNCTRACE(sassc->sc); if ((sassc->flags & MPSSAS_IN_STARTUP) != 0) { if (sassc->startup_refcount++ == 0) { /* just starting, freeze the simq */ mps_dprint(sassc->sc, MPS_INIT, "%s freezing simq\n", __func__); #if __FreeBSD_version >= 1000039 xpt_hold_boot(); #endif xpt_freeze_simq(sassc->sim, 1); } mps_dprint(sassc->sc, MPS_INIT, "%s refcount %u\n", __func__, sassc->startup_refcount); } } void mpssas_release_simq_reinit(struct mpssas_softc *sassc) { if (sassc->flags & MPSSAS_QUEUE_FROZEN) { sassc->flags &= ~MPSSAS_QUEUE_FROZEN; xpt_release_simq(sassc->sim, 1); mps_dprint(sassc->sc, MPS_INFO, "Unfreezing SIM queue\n"); } } void mpssas_startup_decrement(struct mpssas_softc *sassc) { MPS_FUNCTRACE(sassc->sc); if ((sassc->flags & MPSSAS_IN_STARTUP) != 0) { if (--sassc->startup_refcount == 0) { /* finished all discovery-related actions, release * the simq and rescan for the latest topology. */ mps_dprint(sassc->sc, MPS_INIT, "%s releasing simq\n", __func__); sassc->flags &= ~MPSSAS_IN_STARTUP; xpt_release_simq(sassc->sim, 1); #if __FreeBSD_version >= 1000039 xpt_release_boot(); #else mpssas_rescan_target(sassc->sc, NULL); #endif } mps_dprint(sassc->sc, MPS_INIT, "%s refcount %u\n", __func__, sassc->startup_refcount); } } /* * The firmware requires us to stop sending commands when we're doing task * management. * XXX The logic for serializing the device has been made lazy and moved to * mpssas_prepare_for_tm(). */ struct mps_command * mpssas_alloc_tm(struct mps_softc *sc) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mps_command *tm; tm = mps_alloc_high_priority_command(sc); if (tm == NULL) return (NULL); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->Function = MPI2_FUNCTION_SCSI_TASK_MGMT; return tm; } void mpssas_free_tm(struct mps_softc *sc, struct mps_command *tm) { int target_id = 0xFFFFFFFF; if (tm == NULL) return; /* * For TM's the devq is frozen for the device. Unfreeze it here and * free the resources used for freezing the devq. Must clear the * INRESET flag as well or scsi I/O will not work. */ if (tm->cm_targ != NULL) { tm->cm_targ->flags &= ~MPSSAS_TARGET_INRESET; target_id = tm->cm_targ->tid; } if (tm->cm_ccb) { mps_dprint(sc, MPS_INFO, "Unfreezing devq for target ID %d\n", target_id); xpt_release_devq(tm->cm_ccb->ccb_h.path, 1, TRUE); xpt_free_path(tm->cm_ccb->ccb_h.path); xpt_free_ccb(tm->cm_ccb); } mps_free_high_priority_command(sc, tm); } void mpssas_rescan_target(struct mps_softc *sc, struct mpssas_target *targ) { struct mpssas_softc *sassc = sc->sassc; path_id_t pathid; target_id_t targetid; union ccb *ccb; MPS_FUNCTRACE(sc); pathid = cam_sim_path(sassc->sim); if (targ == NULL) targetid = CAM_TARGET_WILDCARD; else targetid = targ - sassc->targets; /* * Allocate a CCB and schedule a rescan. */ ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mps_dprint(sc, MPS_ERROR, "unable to alloc CCB for rescan\n"); return; } if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { mps_dprint(sc, MPS_ERROR, "unable to create path for rescan\n"); xpt_free_ccb(ccb); return; } if (targetid == CAM_TARGET_WILDCARD) ccb->ccb_h.func_code = XPT_SCAN_BUS; else ccb->ccb_h.func_code = XPT_SCAN_TGT; mps_dprint(sc, MPS_TRACE, "%s targetid %u\n", __func__, targetid); xpt_rescan(ccb); } static void mpssas_log_command(struct mps_command *cm, u_int level, const char *fmt, ...) { struct sbuf sb; va_list ap; char str[192]; char path_str[64]; if (cm == NULL) return; /* No need to be in here if debugging isn't enabled */ if ((cm->cm_sc->mps_debug & level) == 0) return; sbuf_new(&sb, str, sizeof(str), 0); va_start(ap, fmt); if (cm->cm_ccb != NULL) { xpt_path_string(cm->cm_ccb->csio.ccb_h.path, path_str, sizeof(path_str)); sbuf_cat(&sb, path_str); if (cm->cm_ccb->ccb_h.func_code == XPT_SCSI_IO) { scsi_command_string(&cm->cm_ccb->csio, &sb); sbuf_printf(&sb, "length %d ", cm->cm_ccb->csio.dxfer_len); } } else { sbuf_printf(&sb, "(noperiph:%s%d:%u:%u:%u): ", cam_sim_name(cm->cm_sc->sassc->sim), cam_sim_unit(cm->cm_sc->sassc->sim), cam_sim_bus(cm->cm_sc->sassc->sim), cm->cm_targ ? cm->cm_targ->tid : 0xFFFFFFFF, cm->cm_lun); } sbuf_printf(&sb, "SMID %u ", cm->cm_desc.Default.SMID); sbuf_vprintf(&sb, fmt, ap); sbuf_finish(&sb); mps_print_field(cm->cm_sc, "%s", sbuf_data(&sb)); va_end(ap); } static void mpssas_remove_volume(struct mps_softc *sc, struct mps_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; struct mpssas_target *targ; uint16_t handle; MPS_FUNCTRACE(sc); reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; targ = tm->cm_targ; if (reply == NULL) { /* XXX retry the remove after the diag reset completes? */ mps_dprint(sc, MPS_FAULT, "%s NULL reply resetting device 0x%04x\n", __func__, handle); mpssas_free_tm(sc, tm); return; } if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mps_dprint(sc, MPS_ERROR, "IOCStatus = 0x%x while resetting device 0x%x\n", le16toh(reply->IOCStatus), handle); } mps_dprint(sc, MPS_XINFO, "Reset aborted %u commands\n", reply->TerminationCount); mps_free_reply(sc, tm->cm_reply_data); tm->cm_reply = NULL; /* Ensures the reply won't get re-freed */ mps_dprint(sc, MPS_XINFO, "clearing target %u handle 0x%04x\n", targ->tid, handle); /* * Don't clear target if remove fails because things will get confusing. * Leave the devname and sasaddr intact so that we know to avoid reusing * this target id if possible, and so we can assign the same target id * to this device if it comes back in the future. */ if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SUCCESS) { targ = tm->cm_targ; targ->handle = 0x0; targ->encl_handle = 0x0; targ->encl_slot = 0x0; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; targ->devinfo = 0x0; targ->flags = 0x0; } mpssas_free_tm(sc, tm); } /* * No Need to call "MPI2_SAS_OP_REMOVE_DEVICE" For Volume removal. * Otherwise Volume Delete is same as Bare Drive Removal. */ void mpssas_prepare_volume_remove(struct mpssas_softc *sassc, uint16_t handle) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mps_softc *sc; struct mps_command *tm; struct mpssas_target *targ = NULL; MPS_FUNCTRACE(sassc->sc); sc = sassc->sc; #ifdef WD_SUPPORT /* * If this is a WD controller, determine if the disk should be exposed * to the OS or not. If disk should be exposed, return from this * function without doing anything. */ if (sc->WD_available && (sc->WD_hide_expose == MPS_WD_EXPOSE_ALWAYS)) { return; } #endif //WD_SUPPORT targ = mpssas_find_target_by_handle(sassc, 0, handle); if (targ == NULL) { /* FIXME: what is the action? */ /* We don't know about this device? */ mps_dprint(sc, MPS_ERROR, "%s %d : invalid handle 0x%x \n", __func__,__LINE__, handle); return; } targ->flags |= MPSSAS_TARGET_INREMOVAL; tm = mpssas_alloc_tm(sc); if (tm == NULL) { mps_dprint(sc, MPS_ERROR, "%s: command alloc failure\n", __func__); return; } mpssas_rescan_target(sc, targ); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = targ->handle; req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; tm->cm_targ = targ; tm->cm_data = NULL; tm->cm_complete = mpssas_remove_volume; tm->cm_complete_data = (void *)(uintptr_t)handle; mps_dprint(sc, MPS_INFO, "%s: Sending reset for target ID %d\n", __func__, targ->tid); mpssas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD); mps_map_command(sc, tm); } /* * The MPT2 firmware performs debounce on the link to avoid transient link * errors and false removals. When it does decide that link has been lost * and a device need to go away, it expects that the host will perform a * target reset and then an op remove. The reset has the side-effect of * aborting any outstanding requests for the device, which is required for * the op-remove to succeed. It's not clear if the host should check for * the device coming back alive after the reset. */ void mpssas_prepare_remove(struct mpssas_softc *sassc, uint16_t handle) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mps_softc *sc; struct mps_command *cm; struct mpssas_target *targ = NULL; MPS_FUNCTRACE(sassc->sc); sc = sassc->sc; targ = mpssas_find_target_by_handle(sassc, 0, handle); if (targ == NULL) { /* FIXME: what is the action? */ /* We don't know about this device? */ mps_dprint(sc, MPS_ERROR, "%s : invalid handle 0x%x \n", __func__, handle); return; } targ->flags |= MPSSAS_TARGET_INREMOVAL; cm = mpssas_alloc_tm(sc); if (cm == NULL) { mps_dprint(sc, MPS_ERROR, "%s: command alloc failure\n", __func__); return; } mpssas_rescan_target(sc, targ); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)cm->cm_req; memset(req, 0, sizeof(*req)); req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; cm->cm_targ = targ; cm->cm_data = NULL; cm->cm_complete = mpssas_remove_device; cm->cm_complete_data = (void *)(uintptr_t)handle; mps_dprint(sc, MPS_INFO, "%s: Sending reset for target ID %d\n", __func__, targ->tid); mpssas_prepare_for_tm(sc, cm, targ, CAM_LUN_WILDCARD); mps_map_command(sc, cm); } static void mpssas_remove_device(struct mps_softc *sc, struct mps_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SAS_IOUNIT_CONTROL_REQUEST *req; struct mpssas_target *targ; struct mps_command *next_cm; uint16_t handle; MPS_FUNCTRACE(sc); reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_ERROR, "%s: cm_flags = %#x for remove of handle %#04x! " "This should not happen!\n", __func__, tm->cm_flags, handle); } if (reply == NULL) { /* XXX retry the remove after the diag reset completes? */ mps_dprint(sc, MPS_FAULT, "%s NULL reply resetting device 0x%04x\n", __func__, handle); mpssas_free_tm(sc, tm); return; } if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mps_dprint(sc, MPS_ERROR, "IOCStatus = 0x%x while resetting device 0x%x\n", le16toh(reply->IOCStatus), handle); } mps_dprint(sc, MPS_XINFO, "Reset aborted %u commands\n", le32toh(reply->TerminationCount)); mps_free_reply(sc, tm->cm_reply_data); tm->cm_reply = NULL; /* Ensures the reply won't get re-freed */ /* Reuse the existing command */ req = (MPI2_SAS_IOUNIT_CONTROL_REQUEST *)tm->cm_req; memset(req, 0, sizeof(*req)); req->Function = MPI2_FUNCTION_SAS_IO_UNIT_CONTROL; req->Operation = MPI2_SAS_OP_REMOVE_DEVICE; req->DevHandle = htole16(handle); tm->cm_data = NULL; tm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; tm->cm_complete = mpssas_remove_complete; tm->cm_complete_data = (void *)(uintptr_t)handle; mps_map_command(sc, tm); mps_dprint(sc, MPS_XINFO, "clearing target %u handle 0x%04x\n", targ->tid, handle); TAILQ_FOREACH_SAFE(tm, &targ->commands, cm_link, next_cm) { union ccb *ccb; mps_dprint(sc, MPS_XINFO, "Completing missed command %p\n", tm); ccb = tm->cm_complete_data; mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); mpssas_scsiio_complete(sc, tm); } } static void mpssas_remove_complete(struct mps_softc *sc, struct mps_command *tm) { MPI2_SAS_IOUNIT_CONTROL_REPLY *reply; uint16_t handle; struct mpssas_target *targ; struct mpssas_lun *lun; MPS_FUNCTRACE(sc); reply = (MPI2_SAS_IOUNIT_CONTROL_REPLY *)tm->cm_reply; handle = (uint16_t)(uintptr_t)tm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_XINFO, "%s: cm_flags = %#x for remove of handle %#04x! " "This should not happen!\n", __func__, tm->cm_flags, handle); mpssas_free_tm(sc, tm); return; } if (reply == NULL) { /* most likely a chip reset */ mps_dprint(sc, MPS_FAULT, "%s NULL reply removing device 0x%04x\n", __func__, handle); mpssas_free_tm(sc, tm); return; } mps_dprint(sc, MPS_XINFO, "%s on handle 0x%04x, IOCStatus= 0x%x\n", __func__, handle, le16toh(reply->IOCStatus)); /* * Don't clear target if remove fails because things will get confusing. * Leave the devname and sasaddr intact so that we know to avoid reusing * this target id if possible, and so we can assign the same target id * to this device if it comes back in the future. */ if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SUCCESS) { targ = tm->cm_targ; targ->handle = 0x0; targ->encl_handle = 0x0; targ->encl_slot = 0x0; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; targ->devinfo = 0x0; targ->flags = 0x0; while(!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPT2); } } mpssas_free_tm(sc, tm); } static int mpssas_register_events(struct mps_softc *sc) { u32 events[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS]; bzero(events, 16); setbit(events, MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_SAS_DISCOVERY); setbit(events, MPI2_EVENT_SAS_BROADCAST_PRIMITIVE); setbit(events, MPI2_EVENT_SAS_INIT_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_SAS_INIT_TABLE_OVERFLOW); setbit(events, MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST); setbit(events, MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE); setbit(events, MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST); setbit(events, MPI2_EVENT_IR_VOLUME); setbit(events, MPI2_EVENT_IR_PHYSICAL_DISK); setbit(events, MPI2_EVENT_IR_OPERATION_STATUS); setbit(events, MPI2_EVENT_LOG_ENTRY_ADDED); mps_register_events(sc, events, mpssas_evt_handler, NULL, &sc->sassc->mpssas_eh); return (0); } int mps_attach_sas(struct mps_softc *sc) { struct mpssas_softc *sassc; cam_status status; int unit, error = 0, reqs; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_INIT, "%s entered\n", __func__); sassc = malloc(sizeof(struct mpssas_softc), M_MPT2, M_WAITOK|M_ZERO); if(!sassc) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Cannot allocate SAS controller memory\n"); return (ENOMEM); } /* * XXX MaxTargets could change during a reinit. Since we don't * resize the targets[] array during such an event, cache the value * of MaxTargets here so that we don't get into trouble later. This * should move into the reinit logic. */ sassc->maxtargets = sc->facts->MaxTargets + sc->facts->MaxVolumes; sassc->targets = malloc(sizeof(struct mpssas_target) * sassc->maxtargets, M_MPT2, M_WAITOK|M_ZERO); if(!sassc->targets) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Cannot allocate SAS target memory\n"); free(sassc, M_MPT2); return (ENOMEM); } sc->sassc = sassc; sassc->sc = sc; reqs = sc->num_reqs - sc->num_prireqs - 1; if ((sassc->devq = cam_simq_alloc(reqs)) == NULL) { mps_dprint(sc, MPS_ERROR, "Cannot allocate SIMQ\n"); error = ENOMEM; goto out; } unit = device_get_unit(sc->mps_dev); sassc->sim = cam_sim_alloc(mpssas_action, mpssas_poll, "mps", sassc, unit, &sc->mps_mtx, reqs, reqs, sassc->devq); if (sassc->sim == NULL) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Cannot allocate SIM\n"); error = EINVAL; goto out; } TAILQ_INIT(&sassc->ev_queue); /* Initialize taskqueue for Event Handling */ TASK_INIT(&sassc->ev_task, 0, mpssas_firmware_event_work, sc); sassc->ev_tq = taskqueue_create("mps_taskq", M_NOWAIT | M_ZERO, taskqueue_thread_enqueue, &sassc->ev_tq); taskqueue_start_threads(&sassc->ev_tq, 1, PRIBIO, "%s taskq", device_get_nameunit(sc->mps_dev)); mps_lock(sc); /* * XXX There should be a bus for every port on the adapter, but since * we're just going to fake the topology for now, we'll pretend that * everything is just a target on a single bus. */ if ((error = xpt_bus_register(sassc->sim, sc->mps_dev, 0)) != 0) { mps_dprint(sc, MPS_INIT|MPS_ERROR, "Error %d registering SCSI bus\n", error); mps_unlock(sc); goto out; } /* * Assume that discovery events will start right away. * * Hold off boot until discovery is complete. */ sassc->flags |= MPSSAS_IN_STARTUP | MPSSAS_IN_DISCOVERY; sc->sassc->startup_refcount = 0; mpssas_startup_increment(sassc); callout_init(&sassc->discovery_callout, 1 /*mpsafe*/); /* * Register for async events so we can determine the EEDP * capabilities of devices. */ status = xpt_create_path(&sassc->path, /*periph*/NULL, cam_sim_path(sc->sassc->sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); if (status != CAM_REQ_CMP) { mps_dprint(sc, MPS_ERROR|MPS_INIT, "Error %#x creating sim path\n", status); sassc->path = NULL; } else { int event; #if (__FreeBSD_version >= 1000006) || \ ((__FreeBSD_version >= 901503) && (__FreeBSD_version < 1000000)) event = AC_ADVINFO_CHANGED; #else event = AC_FOUND_DEVICE; #endif status = xpt_register_async(event, mpssas_async, sc, sassc->path); if (status != CAM_REQ_CMP) { mps_dprint(sc, MPS_ERROR, "Error %#x registering async handler for " "AC_ADVINFO_CHANGED events\n", status); xpt_free_path(sassc->path); sassc->path = NULL; } } if (status != CAM_REQ_CMP) { /* * EEDP use is the exception, not the rule. * Warn the user, but do not fail to attach. */ mps_printf(sc, "EEDP capabilities disabled.\n"); } mps_unlock(sc); mpssas_register_events(sc); out: if (error) mps_detach_sas(sc); mps_dprint(sc, MPS_INIT, "%s exit error= %d\n", __func__, error); return (error); } int mps_detach_sas(struct mps_softc *sc) { struct mpssas_softc *sassc; struct mpssas_lun *lun, *lun_tmp; struct mpssas_target *targ; int i; MPS_FUNCTRACE(sc); if (sc->sassc == NULL) return (0); sassc = sc->sassc; mps_deregister_events(sc, sassc->mpssas_eh); /* * Drain and free the event handling taskqueue with the lock * unheld so that any parallel processing tasks drain properly * without deadlocking. */ if (sassc->ev_tq != NULL) taskqueue_free(sassc->ev_tq); /* Make sure CAM doesn't wedge if we had to bail out early. */ mps_lock(sc); while (sassc->startup_refcount != 0) mpssas_startup_decrement(sassc); /* Deregister our async handler */ if (sassc->path != NULL) { xpt_register_async(0, mpssas_async, sc, sassc->path); xpt_free_path(sassc->path); sassc->path = NULL; } if (sassc->flags & MPSSAS_IN_STARTUP) xpt_release_simq(sassc->sim, 1); if (sassc->sim != NULL) { xpt_bus_deregister(cam_sim_path(sassc->sim)); cam_sim_free(sassc->sim, FALSE); } mps_unlock(sc); if (sassc->devq != NULL) cam_simq_free(sassc->devq); for(i=0; i< sassc->maxtargets ;i++) { targ = &sassc->targets[i]; SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) { free(lun, M_MPT2); } } free(sassc->targets, M_MPT2); free(sassc, M_MPT2); sc->sassc = NULL; return (0); } void mpssas_discovery_end(struct mpssas_softc *sassc) { struct mps_softc *sc = sassc->sc; MPS_FUNCTRACE(sc); if (sassc->flags & MPSSAS_DISCOVERY_TIMEOUT_PENDING) callout_stop(&sassc->discovery_callout); /* * After discovery has completed, check the mapping table for any * missing devices and update their missing counts. Only do this once * whenever the driver is initialized so that missing counts aren't * updated unnecessarily. Note that just because discovery has * completed doesn't mean that events have been processed yet. The * check_devices function is a callout timer that checks if ALL devices * are missing. If so, it will wait a little longer for events to * complete and keep resetting itself until some device in the mapping * table is not missing, meaning that event processing has started. */ if (sc->track_mapping_events) { mps_dprint(sc, MPS_XINFO | MPS_MAPPING, "Discovery has " "completed. Check for missing devices in the mapping " "table.\n"); callout_reset(&sc->device_check_callout, MPS_MISSING_CHECK_DELAY * hz, mps_mapping_check_devices, sc); } } static void mpssas_action(struct cam_sim *sim, union ccb *ccb) { struct mpssas_softc *sassc; sassc = cam_sim_softc(sim); MPS_FUNCTRACE(sassc->sc); mps_dprint(sassc->sc, MPS_TRACE, "ccb func_code 0x%x\n", ccb->ccb_h.func_code); mtx_assert(&sassc->sc->mps_mtx, MA_OWNED); switch (ccb->ccb_h.func_code) { case XPT_PATH_INQ: { struct ccb_pathinq *cpi = &ccb->cpi; struct mps_softc *sc = sassc->sc; cpi->version_num = 1; cpi->hba_inquiry = PI_SDTR_ABLE|PI_TAG_ABLE|PI_WIDE_16; cpi->target_sprt = 0; #if __FreeBSD_version >= 1000039 cpi->hba_misc = PIM_NOBUSRESET | PIM_UNMAPPED | PIM_NOSCAN; #else cpi->hba_misc = PIM_NOBUSRESET | PIM_UNMAPPED; #endif cpi->hba_eng_cnt = 0; cpi->max_target = sassc->maxtargets - 1; cpi->max_lun = 255; /* * initiator_id is set here to an ID outside the set of valid * target IDs (including volumes). */ cpi->initiator_id = sassc->maxtargets; strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN); strlcpy(cpi->hba_vid, "Avago Tech", HBA_IDLEN); strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN); cpi->unit_number = cam_sim_unit(sim); cpi->bus_id = cam_sim_bus(sim); cpi->base_transfer_speed = 150000; cpi->transport = XPORT_SAS; cpi->transport_version = 0; cpi->protocol = PROTO_SCSI; cpi->protocol_version = SCSI_REV_SPC; cpi->maxio = sc->maxio; mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); break; } case XPT_GET_TRAN_SETTINGS: { struct ccb_trans_settings *cts; struct ccb_trans_settings_sas *sas; struct ccb_trans_settings_scsi *scsi; struct mpssas_target *targ; cts = &ccb->cts; sas = &cts->xport_specific.sas; scsi = &cts->proto_specific.scsi; KASSERT(cts->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_GET_TRANS_SETTINGS\n", cts->ccb_h.target_id)); targ = &sassc->targets[cts->ccb_h.target_id]; if (targ->handle == 0x0) { mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); break; } cts->protocol_version = SCSI_REV_SPC2; cts->transport = XPORT_SAS; cts->transport_version = 0; sas->valid = CTS_SAS_VALID_SPEED; switch (targ->linkrate) { case 0x08: sas->bitrate = 150000; break; case 0x09: sas->bitrate = 300000; break; case 0x0a: sas->bitrate = 600000; break; default: sas->valid = 0; } cts->protocol = PROTO_SCSI; scsi->valid = CTS_SCSI_VALID_TQ; scsi->flags = CTS_SCSI_FLAGS_TAG_ENB; mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); break; } case XPT_CALC_GEOMETRY: cam_calc_geometry(&ccb->ccg, /*extended*/1); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); break; case XPT_RESET_DEV: mps_dprint(sassc->sc, MPS_XINFO, "mpssas_action XPT_RESET_DEV\n"); mpssas_action_resetdev(sassc, ccb); return; case XPT_RESET_BUS: case XPT_ABORT: case XPT_TERM_IO: mps_dprint(sassc->sc, MPS_XINFO, "mpssas_action faking success for abort or reset\n"); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); break; case XPT_SCSI_IO: mpssas_action_scsiio(sassc, ccb); return; #if __FreeBSD_version >= 900026 case XPT_SMP_IO: mpssas_action_smpio(sassc, ccb); return; #endif default: mpssas_set_ccbstatus(ccb, CAM_FUNC_NOTAVAIL); break; } xpt_done(ccb); } static void mpssas_announce_reset(struct mps_softc *sc, uint32_t ac_code, target_id_t target_id, lun_id_t lun_id) { path_id_t path_id = cam_sim_path(sc->sassc->sim); struct cam_path *path; mps_dprint(sc, MPS_XINFO, "%s code %x target %d lun %jx\n", __func__, ac_code, target_id, (uintmax_t)lun_id); if (xpt_create_path(&path, NULL, path_id, target_id, lun_id) != CAM_REQ_CMP) { mps_dprint(sc, MPS_ERROR, "unable to create path for reset " "notification\n"); return; } xpt_async(ac_code, path, NULL); xpt_free_path(path); } static void mpssas_complete_all_commands(struct mps_softc *sc) { struct mps_command *cm; int i; int completed; MPS_FUNCTRACE(sc); mtx_assert(&sc->mps_mtx, MA_OWNED); /* complete all commands with a NULL reply */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_state == MPS_CM_STATE_FREE) continue; cm->cm_state = MPS_CM_STATE_BUSY; cm->cm_reply = NULL; completed = 0; if (cm->cm_flags & MPS_CM_FLAGS_SATA_ID_TIMEOUT) { MPASS(cm->cm_data); free(cm->cm_data, M_MPT2); cm->cm_data = NULL; } if (cm->cm_flags & MPS_CM_FLAGS_POLLED) cm->cm_flags |= MPS_CM_FLAGS_COMPLETE; if (cm->cm_complete != NULL) { mpssas_log_command(cm, MPS_RECOVERY, "completing cm %p state %x ccb %p for diag reset\n", cm, cm->cm_state, cm->cm_ccb); cm->cm_complete(sc, cm); completed = 1; } else if (cm->cm_flags & MPS_CM_FLAGS_WAKEUP) { mpssas_log_command(cm, MPS_RECOVERY, "waking up cm %p state %x ccb %p for diag reset\n", cm, cm->cm_state, cm->cm_ccb); wakeup(cm); completed = 1; } if ((completed == 0) && (cm->cm_state != MPS_CM_STATE_FREE)) { /* this should never happen, but if it does, log */ mpssas_log_command(cm, MPS_RECOVERY, "cm %p state %x flags 0x%x ccb %p during diag " "reset\n", cm, cm->cm_state, cm->cm_flags, cm->cm_ccb); } } sc->io_cmds_active = 0; } void mpssas_handle_reinit(struct mps_softc *sc) { int i; /* Go back into startup mode and freeze the simq, so that CAM * doesn't send any commands until after we've rediscovered all * targets and found the proper device handles for them. * * After the reset, portenable will trigger discovery, and after all * discovery-related activities have finished, the simq will be * released. */ mps_dprint(sc, MPS_INIT, "%s startup\n", __func__); sc->sassc->flags |= MPSSAS_IN_STARTUP; sc->sassc->flags |= MPSSAS_IN_DISCOVERY; mpssas_startup_increment(sc->sassc); /* notify CAM of a bus reset */ mpssas_announce_reset(sc, AC_BUS_RESET, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); /* complete and cleanup after all outstanding commands */ mpssas_complete_all_commands(sc); mps_dprint(sc, MPS_INIT, "%s startup %u after command completion\n", __func__, sc->sassc->startup_refcount); /* zero all the target handles, since they may change after the * reset, and we have to rediscover all the targets and use the new * handles. */ for (i = 0; i < sc->sassc->maxtargets; i++) { if (sc->sassc->targets[i].outstanding != 0) mps_dprint(sc, MPS_INIT, "target %u outstanding %u\n", i, sc->sassc->targets[i].outstanding); sc->sassc->targets[i].handle = 0x0; sc->sassc->targets[i].exp_dev_handle = 0x0; sc->sassc->targets[i].outstanding = 0; sc->sassc->targets[i].flags = MPSSAS_TARGET_INDIAGRESET; } } static void mpssas_tm_timeout(void *data) { struct mps_command *tm = data; struct mps_softc *sc = tm->cm_sc; mtx_assert(&sc->mps_mtx, MA_OWNED); mpssas_log_command(tm, MPS_INFO|MPS_RECOVERY, "task mgmt %p timed out\n", tm); KASSERT(tm->cm_state == MPS_CM_STATE_INQUEUE, ("command not inqueue\n")); tm->cm_state = MPS_CM_STATE_BUSY; mps_reinit(sc); } static void mpssas_logical_unit_reset_complete(struct mps_softc *sc, struct mps_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; unsigned int cm_count = 0; struct mps_command *cm; struct mpssas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. * XXXSL So should it be an assertion? */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_RECOVERY|MPS_ERROR, "%s: cm_flags = %#x for LUN reset! " "This should not happen!\n", __func__, tm->cm_flags); mpssas_free_tm(sc, tm); return; } if (reply == NULL) { mps_dprint(sc, MPS_RECOVERY, "NULL reset reply for tm %p\n", tm); if ((sc->mps_flags & MPS_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mps_dprint(sc, MPS_RECOVERY, "Hardware undergoing " "reset, ignoring NULL LUN reset reply\n"); targ->tm = NULL; mpssas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "NULL reply on " "LUN reset attempt, resetting controller\n"); mps_reinit(sc); } return; } mps_dprint(sc, MPS_RECOVERY, "logical unit reset status 0x%x code 0x%x count %u\n", le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); /* * See if there are any outstanding commands for this LUN. * This could be made more efficient by using a per-LU data * structure of some sort. */ TAILQ_FOREACH(cm, &targ->commands, cm_link) { if (cm->cm_lun == tm->cm_lun) cm_count++; } if (cm_count == 0) { mps_dprint(sc, MPS_RECOVERY|MPS_INFO, "Finished recovery after LUN reset for target %u\n", targ->tid); mpssas_announce_reset(sc, AC_SENT_BDR, targ->tid, tm->cm_lun); /* * We've finished recovery for this logical unit. check and * see if some other logical unit has a timedout command * that needs to be processed. */ cm = TAILQ_FIRST(&targ->timedout_commands); if (cm) { mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "More commands to abort for target %u\n", targ->tid); mpssas_send_abort(sc, tm, cm); } else { targ->tm = NULL; mpssas_free_tm(sc, tm); } } else { /* * If we still have commands for this LUN, the reset * effectively failed, regardless of the status reported. * Escalate to a target reset. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "logical unit reset complete for target %u, but still " "have %u command(s), sending target reset\n", targ->tid, cm_count); mpssas_send_reset(sc, tm, MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET); } } static void mpssas_target_reset_complete(struct mps_softc *sc, struct mps_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpssas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_ERROR,"%s: cm_flags = %#x for target reset! " "This should not happen!\n", __func__, tm->cm_flags); mpssas_free_tm(sc, tm); return; } if (reply == NULL) { mps_dprint(sc, MPS_RECOVERY, "NULL target reset reply for tm %pi TaskMID %u\n", tm, le16toh(req->TaskMID)); if ((sc->mps_flags & MPS_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mps_dprint(sc, MPS_RECOVERY, "Hardware undergoing " "reset, ignoring NULL target reset reply\n"); targ->tm = NULL; mpssas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "NULL reply on " "target reset attempt, resetting controller\n"); mps_reinit(sc); } return; } mps_dprint(sc, MPS_RECOVERY, "target reset status 0x%x code 0x%x count %u\n", le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); if (targ->outstanding == 0) { /* we've finished recovery for this target and all * of its logical units. */ mps_dprint(sc, MPS_RECOVERY|MPS_INFO, "Finished reset recovery for target %u\n", targ->tid); mpssas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid, CAM_LUN_WILDCARD); targ->tm = NULL; mpssas_free_tm(sc, tm); } else { /* * After a target reset, if this target still has * outstanding commands, the reset effectively failed, * regardless of the status reported. escalate. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "Target reset complete for target %u, but still have %u " "command(s), resetting controller\n", targ->tid, targ->outstanding); mps_reinit(sc); } } #define MPS_RESET_TIMEOUT 30 int mpssas_send_reset(struct mps_softc *sc, struct mps_command *tm, uint8_t type) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpssas_target *target; int err; target = tm->cm_targ; if (target->handle == 0) { mps_dprint(sc, MPS_ERROR,"%s null devhandle for target_id %d\n", __func__, target->tid); return -1; } req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(target->handle); req->TaskType = type; if (type == MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET) { /* XXX Need to handle invalid LUNs */ MPS_SET_LUN(req->LUN, tm->cm_lun); tm->cm_targ->logical_unit_resets++; mps_dprint(sc, MPS_RECOVERY|MPS_INFO, "Sending logical unit reset to target %u lun %d\n", target->tid, tm->cm_lun); tm->cm_complete = mpssas_logical_unit_reset_complete; mpssas_prepare_for_tm(sc, tm, target, tm->cm_lun); } else if (type == MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET) { /* * Target reset method = * SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; tm->cm_targ->target_resets++; mps_dprint(sc, MPS_RECOVERY|MPS_INFO, "Sending target reset to target %u\n", target->tid); tm->cm_complete = mpssas_target_reset_complete; mpssas_prepare_for_tm(sc, tm, target, CAM_LUN_WILDCARD); } else { mps_dprint(sc, MPS_ERROR, "unexpected reset type 0x%x\n", type); return -1; } tm->cm_data = NULL; tm->cm_complete_data = (void *)tm; callout_reset(&tm->cm_callout, MPS_RESET_TIMEOUT * hz, mpssas_tm_timeout, tm); err = mps_map_command(sc, tm); if (err) mps_dprint(sc, MPS_ERROR|MPS_RECOVERY, "error %d sending reset type %u\n", err, type); return err; } static void mpssas_abort_complete(struct mps_softc *sc, struct mps_command *tm) { struct mps_command *cm; MPI2_SCSI_TASK_MANAGE_REPLY *reply; MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpssas_target *targ; callout_stop(&tm->cm_callout); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; reply = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; targ = tm->cm_targ; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_RECOVERY, "cm_flags = %#x for abort %p TaskMID %u!\n", tm->cm_flags, tm, le16toh(req->TaskMID)); mpssas_free_tm(sc, tm); return; } if (reply == NULL) { mps_dprint(sc, MPS_RECOVERY, "NULL abort reply for tm %p TaskMID %u\n", tm, le16toh(req->TaskMID)); if ((sc->mps_flags & MPS_FLAGS_DIAGRESET) != 0) { /* this completion was due to a reset, just cleanup */ mps_dprint(sc, MPS_RECOVERY, "Hardware undergoing " "reset, ignoring NULL abort reply\n"); targ->tm = NULL; mpssas_free_tm(sc, tm); } else { /* we should have gotten a reply. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "NULL reply on " "abort attempt, resetting controller\n"); mps_reinit(sc); } return; } mps_dprint(sc, MPS_RECOVERY, "abort TaskMID %u status 0x%x code 0x%x count %u\n", le16toh(req->TaskMID), le16toh(reply->IOCStatus), le32toh(reply->ResponseCode), le32toh(reply->TerminationCount)); cm = TAILQ_FIRST(&tm->cm_targ->timedout_commands); if (cm == NULL) { /* * If there are no more timedout commands, we're done with * error recovery for this target. */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "Finished abort recovery for target %u\n", targ->tid); targ->tm = NULL; mpssas_free_tm(sc, tm); } else if (le16toh(req->TaskMID) != cm->cm_desc.Default.SMID) { /* abort success, but we have more timedout commands to abort */ mps_dprint(sc, MPS_INFO|MPS_RECOVERY, "Continuing abort recovery for target %u\n", targ->tid); mpssas_send_abort(sc, tm, cm); } else { /* we didn't get a command completion, so the abort * failed as far as we're concerned. escalate. */ mps_dprint(sc, MPS_RECOVERY, "Abort failed for target %u, sending logical unit reset\n", targ->tid); mpssas_send_reset(sc, tm, MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET); } } #define MPS_ABORT_TIMEOUT 5 static int mpssas_send_abort(struct mps_softc *sc, struct mps_command *tm, struct mps_command *cm) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mpssas_target *targ; int err; targ = cm->cm_targ; if (targ->handle == 0) { mps_dprint(sc, MPS_ERROR|MPS_RECOVERY, "%s null devhandle for target_id %d\n", __func__, cm->cm_ccb->ccb_h.target_id); return -1; } mpssas_log_command(cm, MPS_RECOVERY|MPS_INFO, "Aborting command %p\n", cm); req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_ABORT_TASK; /* XXX Need to handle invalid LUNs */ MPS_SET_LUN(req->LUN, cm->cm_ccb->ccb_h.target_lun); req->TaskMID = htole16(cm->cm_desc.Default.SMID); tm->cm_data = NULL; tm->cm_complete = mpssas_abort_complete; tm->cm_complete_data = (void *)tm; tm->cm_targ = cm->cm_targ; tm->cm_lun = cm->cm_lun; callout_reset(&tm->cm_callout, MPS_ABORT_TIMEOUT * hz, mpssas_tm_timeout, tm); targ->aborts++; mpssas_prepare_for_tm(sc, tm, targ, tm->cm_lun); err = mps_map_command(sc, tm); if (err) mps_dprint(sc, MPS_ERROR|MPS_RECOVERY, "error %d sending abort for cm %p SMID %u\n", err, cm, req->TaskMID); return err; } static void mpssas_scsiio_timeout(void *data) { sbintime_t elapsed, now; union ccb *ccb; struct mps_softc *sc; struct mps_command *cm; struct mpssas_target *targ; cm = (struct mps_command *)data; sc = cm->cm_sc; ccb = cm->cm_ccb; now = sbinuptime(); MPS_FUNCTRACE(sc); mtx_assert(&sc->mps_mtx, MA_OWNED); mps_dprint(sc, MPS_XINFO|MPS_RECOVERY, "Timeout checking cm %p\n", sc); /* * Run the interrupt handler to make sure it's not pending. This * isn't perfect because the command could have already completed * and been re-used, though this is unlikely. */ mps_intr_locked(sc); - if (cm->cm_state != MPS_CM_STATE_INQUEUE) { + if (cm->cm_flags & MPS_CM_FLAGS_ON_RECOVERY) { mpssas_log_command(cm, MPS_XINFO, "SCSI command %p almost timed out\n", cm); return; } if (cm->cm_ccb == NULL) { mps_dprint(sc, MPS_ERROR, "command timeout with NULL ccb\n"); return; } targ = cm->cm_targ; targ->timeouts++; elapsed = now - ccb->ccb_h.qos.sim_data; mpssas_log_command(cm, MPS_INFO|MPS_RECOVERY, "Command timeout on target %u(0x%04x) %d set, %d.%d elapsed\n", targ->tid, targ->handle, ccb->ccb_h.timeout, sbintime_getsec(elapsed), elapsed & 0xffffffff); /* XXX first, check the firmware state, to see if it's still * operational. if not, do a diag reset. */ mpssas_set_ccbstatus(cm->cm_ccb, CAM_CMD_TIMEOUT); - cm->cm_state = MPS_CM_STATE_TIMEDOUT; + cm->cm_flags |= MPS_CM_FLAGS_ON_RECOVERY | MPS_CM_FLAGS_TIMEDOUT; TAILQ_INSERT_TAIL(&targ->timedout_commands, cm, cm_recovery); if (targ->tm != NULL) { /* target already in recovery, just queue up another * timedout command to be processed later. */ mps_dprint(sc, MPS_RECOVERY, "queued timedout cm %p for processing by tm %p\n", cm, targ->tm); } else if ((targ->tm = mpssas_alloc_tm(sc)) != NULL) { mps_dprint(sc, MPS_RECOVERY|MPS_INFO, "Sending abort to target %u for SMID %d\n", targ->tid, cm->cm_desc.Default.SMID); mps_dprint(sc, MPS_RECOVERY, "timedout cm %p allocated tm %p\n", cm, targ->tm); /* start recovery by aborting the first timedout command */ mpssas_send_abort(sc, targ->tm, cm); } else { /* XXX queue this target up for recovery once a TM becomes * available. The firmware only has a limited number of * HighPriority credits for the high priority requests used * for task management, and we ran out. * * Isilon: don't worry about this for now, since we have * more credits than disks in an enclosure, and limit * ourselves to one TM per target for recovery. */ mps_dprint(sc, MPS_ERROR|MPS_RECOVERY, "timedout cm %p failed to allocate a tm\n", cm); } } static void mpssas_action_scsiio(struct mpssas_softc *sassc, union ccb *ccb) { MPI2_SCSI_IO_REQUEST *req; struct ccb_scsiio *csio; struct mps_softc *sc; struct mpssas_target *targ; struct mpssas_lun *lun; struct mps_command *cm; uint8_t i, lba_byte, *ref_tag_addr; uint16_t eedp_flags; uint32_t mpi_control; sc = sassc->sc; MPS_FUNCTRACE(sc); mtx_assert(&sc->mps_mtx, MA_OWNED); csio = &ccb->csio; KASSERT(csio->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_SCSI_IO\n", csio->ccb_h.target_id)); targ = &sassc->targets[csio->ccb_h.target_id]; mps_dprint(sc, MPS_TRACE, "ccb %p target flag %x\n", ccb, targ->flags); if (targ->handle == 0x0) { mps_dprint(sc, MPS_ERROR, "%s NULL handle for target %u\n", __func__, csio->ccb_h.target_id); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } if (targ->flags & MPS_TARGET_FLAGS_RAID_COMPONENT) { mps_dprint(sc, MPS_ERROR, "%s Raid component no SCSI IO " "supported %u\n", __func__, csio->ccb_h.target_id); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } /* * Sometimes, it is possible to get a command that is not "In * Progress" and was actually aborted by the upper layer. Check for * this here and complete the command without error. */ if (mpssas_get_ccbstatus(ccb) != CAM_REQ_INPROG) { mps_dprint(sc, MPS_TRACE, "%s Command is not in progress for " "target %u\n", __func__, csio->ccb_h.target_id); xpt_done(ccb); return; } /* * If devinfo is 0 this will be a volume. In that case don't tell CAM * that the volume has timed out. We want volumes to be enumerated * until they are deleted/removed, not just failed. */ if (targ->flags & MPSSAS_TARGET_INREMOVAL) { if (targ->devinfo == 0) mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); else mpssas_set_ccbstatus(ccb, CAM_SEL_TIMEOUT); xpt_done(ccb); return; } if ((sc->mps_flags & MPS_FLAGS_SHUTDOWN) != 0) { mps_dprint(sc, MPS_INFO, "%s shutting down\n", __func__); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); xpt_done(ccb); return; } /* * If target has a reset in progress, freeze the devq and return. The * devq will be released when the TM reset is finished. */ if (targ->flags & MPSSAS_TARGET_INRESET) { ccb->ccb_h.status = CAM_BUSY | CAM_DEV_QFRZN; mps_dprint(sc, MPS_INFO, "%s: Freezing devq for target ID %d\n", __func__, targ->tid); xpt_freeze_devq(ccb->ccb_h.path, 1); xpt_done(ccb); return; } cm = mps_alloc_command(sc); if (cm == NULL || (sc->mps_flags & MPS_FLAGS_DIAGRESET)) { if (cm != NULL) { mps_free_command(sc, cm); } if ((sassc->flags & MPSSAS_QUEUE_FROZEN) == 0) { xpt_freeze_simq(sassc->sim, 1); sassc->flags |= MPSSAS_QUEUE_FROZEN; } ccb->ccb_h.status &= ~CAM_SIM_QUEUED; ccb->ccb_h.status |= CAM_REQUEUE_REQ; xpt_done(ccb); return; } req = (MPI2_SCSI_IO_REQUEST *)cm->cm_req; bzero(req, sizeof(*req)); req->DevHandle = htole16(targ->handle); req->Function = MPI2_FUNCTION_SCSI_IO_REQUEST; req->MsgFlags = 0; req->SenseBufferLowAddress = htole32(cm->cm_sense_busaddr); req->SenseBufferLength = MPS_SENSE_LEN; req->SGLFlags = 0; req->ChainOffset = 0; req->SGLOffset0 = 24; /* 32bit word offset to the SGL */ req->SGLOffset1= 0; req->SGLOffset2= 0; req->SGLOffset3= 0; req->SkipCount = 0; req->DataLength = htole32(csio->dxfer_len); req->BidirectionalDataLength = 0; req->IoFlags = htole16(csio->cdb_len); req->EEDPFlags = 0; /* Note: BiDirectional transfers are not supported */ switch (csio->ccb_h.flags & CAM_DIR_MASK) { case CAM_DIR_IN: mpi_control = MPI2_SCSIIO_CONTROL_READ; cm->cm_flags |= MPS_CM_FLAGS_DATAIN; break; case CAM_DIR_OUT: mpi_control = MPI2_SCSIIO_CONTROL_WRITE; cm->cm_flags |= MPS_CM_FLAGS_DATAOUT; break; case CAM_DIR_NONE: default: mpi_control = MPI2_SCSIIO_CONTROL_NODATATRANSFER; break; } if (csio->cdb_len == 32) mpi_control |= 4 << MPI2_SCSIIO_CONTROL_ADDCDBLEN_SHIFT; /* * It looks like the hardware doesn't require an explicit tag * number for each transaction. SAM Task Management not supported * at the moment. */ switch (csio->tag_action) { case MSG_HEAD_OF_Q_TAG: mpi_control |= MPI2_SCSIIO_CONTROL_HEADOFQ; break; case MSG_ORDERED_Q_TAG: mpi_control |= MPI2_SCSIIO_CONTROL_ORDEREDQ; break; case MSG_ACA_TASK: mpi_control |= MPI2_SCSIIO_CONTROL_ACAQ; break; case CAM_TAG_ACTION_NONE: case MSG_SIMPLE_Q_TAG: default: mpi_control |= MPI2_SCSIIO_CONTROL_SIMPLEQ; break; } mpi_control |= sc->mapping_table[csio->ccb_h.target_id].TLR_bits; req->Control = htole32(mpi_control); if (MPS_SET_LUN(req->LUN, csio->ccb_h.target_lun) != 0) { mps_free_command(sc, cm); mpssas_set_ccbstatus(ccb, CAM_LUN_INVALID); xpt_done(ccb); return; } if (csio->ccb_h.flags & CAM_CDB_POINTER) bcopy(csio->cdb_io.cdb_ptr, &req->CDB.CDB32[0], csio->cdb_len); else bcopy(csio->cdb_io.cdb_bytes, &req->CDB.CDB32[0],csio->cdb_len); req->IoFlags = htole16(csio->cdb_len); /* * Check if EEDP is supported and enabled. If it is then check if the * SCSI opcode could be using EEDP. If so, make sure the LUN exists and * is formatted for EEDP support. If all of this is true, set CDB up * for EEDP transfer. */ eedp_flags = op_code_prot[req->CDB.CDB32[0]]; if (sc->eedp_enabled && eedp_flags) { SLIST_FOREACH(lun, &targ->luns, lun_link) { if (lun->lun_id == csio->ccb_h.target_lun) { break; } } if ((lun != NULL) && (lun->eedp_formatted)) { req->EEDPBlockSize = htole16(lun->eedp_block_size); eedp_flags |= (MPI2_SCSIIO_EEDPFLAGS_INC_PRI_REFTAG | MPI2_SCSIIO_EEDPFLAGS_CHECK_REFTAG | MPI2_SCSIIO_EEDPFLAGS_CHECK_GUARD); req->EEDPFlags = htole16(eedp_flags); /* * If CDB less than 32, fill in Primary Ref Tag with * low 4 bytes of LBA. If CDB is 32, tag stuff is * already there. Also, set protection bit. FreeBSD * currently does not support CDBs bigger than 16, but * the code doesn't hurt, and will be here for the * future. */ if (csio->cdb_len != 32) { lba_byte = (csio->cdb_len == 16) ? 6 : 2; ref_tag_addr = (uint8_t *)&req->CDB.EEDP32. PrimaryReferenceTag; for (i = 0; i < 4; i++) { *ref_tag_addr = req->CDB.CDB32[lba_byte + i]; ref_tag_addr++; } req->CDB.EEDP32.PrimaryReferenceTag = htole32(req->CDB.EEDP32.PrimaryReferenceTag); req->CDB.EEDP32.PrimaryApplicationTagMask = 0xFFFF; req->CDB.CDB32[1] = (req->CDB.CDB32[1] & 0x1F) | 0x20; } else { eedp_flags |= MPI2_SCSIIO_EEDPFLAGS_INC_PRI_APPTAG; req->EEDPFlags = htole16(eedp_flags); req->CDB.CDB32[10] = (req->CDB.CDB32[10] & 0x1F) | 0x20; } } } cm->cm_length = csio->dxfer_len; if (cm->cm_length != 0) { cm->cm_data = ccb; cm->cm_flags |= MPS_CM_FLAGS_USE_CCB; } else { cm->cm_data = NULL; } cm->cm_sge = &req->SGL; cm->cm_sglsize = (32 - 24) * 4; cm->cm_desc.SCSIIO.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO; cm->cm_desc.SCSIIO.DevHandle = htole16(targ->handle); cm->cm_complete = mpssas_scsiio_complete; cm->cm_complete_data = ccb; cm->cm_targ = targ; cm->cm_lun = csio->ccb_h.target_lun; cm->cm_ccb = ccb; /* * If HBA is a WD and the command is not for a retry, try to build a * direct I/O message. If failed, or the command is for a retry, send * the I/O to the IR volume itself. */ if (sc->WD_valid_config) { if (ccb->ccb_h.sim_priv.entries[0].field == MPS_WD_RETRY) { mpssas_direct_drive_io(sassc, cm, ccb); } else { mpssas_set_ccbstatus(ccb, CAM_REQ_INPROG); } } #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) if (csio->bio != NULL) biotrack(csio->bio, __func__); #endif csio->ccb_h.qos.sim_data = sbinuptime(); callout_reset_sbt(&cm->cm_callout, SBT_1MS * ccb->ccb_h.timeout, 0, mpssas_scsiio_timeout, cm, 0); targ->issued++; targ->outstanding++; TAILQ_INSERT_TAIL(&targ->commands, cm, cm_link); ccb->ccb_h.status |= CAM_SIM_QUEUED; mpssas_log_command(cm, MPS_XINFO, "%s cm %p ccb %p outstanding %u\n", __func__, cm, ccb, targ->outstanding); mps_map_command(sc, cm); return; } /** * mps_sc_failed_io_info - translated non-succesfull SCSI_IO request */ static void mps_sc_failed_io_info(struct mps_softc *sc, struct ccb_scsiio *csio, Mpi2SCSIIOReply_t *mpi_reply) { u32 response_info; u8 *response_bytes; u16 ioc_status = le16toh(mpi_reply->IOCStatus) & MPI2_IOCSTATUS_MASK; u8 scsi_state = mpi_reply->SCSIState; u8 scsi_status = mpi_reply->SCSIStatus; u32 log_info = le32toh(mpi_reply->IOCLogInfo); const char *desc_ioc_state, *desc_scsi_status; if (log_info == 0x31170000) return; desc_ioc_state = mps_describe_table(mps_iocstatus_string, ioc_status); desc_scsi_status = mps_describe_table(mps_scsi_status_string, scsi_status); mps_dprint(sc, MPS_XINFO, "\thandle(0x%04x), ioc_status(%s)(0x%04x)\n", le16toh(mpi_reply->DevHandle), desc_ioc_state, ioc_status); /* *We can add more detail about underflow data here * TO-DO */ mps_dprint(sc, MPS_XINFO, "\tscsi_status(%s)(0x%02x), " "scsi_state %b\n", desc_scsi_status, scsi_status, scsi_state, "\20" "\1AutosenseValid" "\2AutosenseFailed" "\3NoScsiStatus" "\4Terminated" "\5Response InfoValid"); if (sc->mps_debug & MPS_XINFO && scsi_state & MPI2_SCSI_STATE_AUTOSENSE_VALID) { mps_dprint(sc, MPS_XINFO, "-> Sense Buffer Data : Start :\n"); scsi_sense_print(csio); mps_dprint(sc, MPS_XINFO, "-> Sense Buffer Data : End :\n"); } if (scsi_state & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) { response_info = le32toh(mpi_reply->ResponseInfo); response_bytes = (u8 *)&response_info; mps_dprint(sc, MPS_XINFO, "response code(0x%1x): %s\n", response_bytes[0], mps_describe_table(mps_scsi_taskmgmt_string, response_bytes[0])); } } static void mpssas_scsiio_complete(struct mps_softc *sc, struct mps_command *cm) { MPI2_SCSI_IO_REPLY *rep; union ccb *ccb; struct ccb_scsiio *csio; struct mpssas_softc *sassc; struct scsi_vpd_supported_page_list *vpd_list = NULL; u8 *TLR_bits, TLR_on; int dir = 0, i; u16 alloc_len; struct mpssas_target *target; target_id_t target_id; MPS_FUNCTRACE(sc); mps_dprint(sc, MPS_TRACE, "cm %p SMID %u ccb %p reply %p outstanding %u\n", cm, cm->cm_desc.Default.SMID, cm->cm_ccb, cm->cm_reply, cm->cm_targ->outstanding); callout_stop(&cm->cm_callout); mtx_assert(&sc->mps_mtx, MA_OWNED); sassc = sc->sassc; ccb = cm->cm_complete_data; csio = &ccb->csio; target_id = csio->ccb_h.target_id; rep = (MPI2_SCSI_IO_REPLY *)cm->cm_reply; /* * XXX KDM if the chain allocation fails, does it matter if we do * the sync and unload here? It is simpler to do it in every case, * assuming it doesn't cause problems. */ if (cm->cm_data != NULL) { if (cm->cm_flags & MPS_CM_FLAGS_DATAIN) dir = BUS_DMASYNC_POSTREAD; else if (cm->cm_flags & MPS_CM_FLAGS_DATAOUT) dir = BUS_DMASYNC_POSTWRITE; bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, dir); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); } cm->cm_targ->completed++; cm->cm_targ->outstanding--; TAILQ_REMOVE(&cm->cm_targ->commands, cm, cm_link); ccb->ccb_h.status &= ~(CAM_STATUS_MASK | CAM_SIM_QUEUED); #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) if (ccb->csio.bio != NULL) biotrack(ccb->csio.bio, __func__); #endif - if (cm->cm_state == MPS_CM_STATE_TIMEDOUT) { + if (cm->cm_flags & MPS_CM_FLAGS_ON_RECOVERY) { TAILQ_REMOVE(&cm->cm_targ->timedout_commands, cm, cm_recovery); - cm->cm_state = MPS_CM_STATE_BUSY; + KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, + ("Not busy for CM_FLAGS_TIMEDOUT: %d\n", cm->cm_state)); + cm->cm_flags &= ~MPS_CM_FLAGS_ON_RECOVERY; if (cm->cm_reply != NULL) mpssas_log_command(cm, MPS_RECOVERY, "completed timedout cm %p ccb %p during recovery " "ioc %x scsi %x state %x xfer %u\n", cm, cm->cm_ccb, le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); else mpssas_log_command(cm, MPS_RECOVERY, "completed timedout cm %p ccb %p during recovery\n", cm, cm->cm_ccb); } else if (cm->cm_targ->tm != NULL) { if (cm->cm_reply != NULL) mpssas_log_command(cm, MPS_RECOVERY, "completed cm %p ccb %p during recovery " "ioc %x scsi %x state %x xfer %u\n", cm, cm->cm_ccb, le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); else mpssas_log_command(cm, MPS_RECOVERY, "completed cm %p ccb %p during recovery\n", cm, cm->cm_ccb); } else if ((sc->mps_flags & MPS_FLAGS_DIAGRESET) != 0) { mpssas_log_command(cm, MPS_RECOVERY, "reset completed cm %p ccb %p\n", cm, cm->cm_ccb); } if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { /* * We ran into an error after we tried to map the command, * so we're getting a callback without queueing the command * to the hardware. So we set the status here, and it will * be retained below. We'll go through the "fast path", * because there can be no reply when we haven't actually * gone out to the hardware. */ mpssas_set_ccbstatus(ccb, CAM_REQUEUE_REQ); /* * Currently the only error included in the mask is * MPS_CM_FLAGS_CHAIN_FAILED, which means we're out of * chain frames. We need to freeze the queue until we get * a command that completed without this error, which will * hopefully have some chain frames attached that we can * use. If we wanted to get smarter about it, we would * only unfreeze the queue in this condition when we're * sure that we're getting some chain frames back. That's * probably unnecessary. */ if ((sassc->flags & MPSSAS_QUEUE_FROZEN) == 0) { xpt_freeze_simq(sassc->sim, 1); sassc->flags |= MPSSAS_QUEUE_FROZEN; mps_dprint(sc, MPS_XINFO, "Error sending command, " "freezing SIM queue\n"); } } /* * If this is a Start Stop Unit command and it was issued by the driver * during shutdown, decrement the refcount to account for all of the * commands that were sent. All SSU commands should be completed before * shutdown completes, meaning SSU_refcount will be 0 after SSU_started * is TRUE. */ if (sc->SSU_started && (csio->cdb_io.cdb_bytes[0] == START_STOP_UNIT)) { mps_dprint(sc, MPS_INFO, "Decrementing SSU count.\n"); sc->SSU_refcount--; } /* Take the fast path to completion */ if (cm->cm_reply == NULL) { if (mpssas_get_ccbstatus(ccb) == CAM_REQ_INPROG) { if ((sc->mps_flags & MPS_FLAGS_DIAGRESET) != 0) mpssas_set_ccbstatus(ccb, CAM_SCSI_BUS_RESET); else { mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); ccb->csio.scsi_status = SCSI_STATUS_OK; } if (sassc->flags & MPSSAS_QUEUE_FROZEN) { ccb->ccb_h.status |= CAM_RELEASE_SIMQ; sassc->flags &= ~MPSSAS_QUEUE_FROZEN; mps_dprint(sc, MPS_XINFO, "Unfreezing SIM queue\n"); } } /* * There are two scenarios where the status won't be * CAM_REQ_CMP. The first is if MPS_CM_FLAGS_ERROR_MASK is * set, the second is in the MPS_FLAGS_DIAGRESET above. */ if (mpssas_get_ccbstatus(ccb) != CAM_REQ_CMP) { /* * Freeze the dev queue so that commands are * executed in the correct order after error * recovery. */ ccb->ccb_h.status |= CAM_DEV_QFRZN; xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1); } mps_free_command(sc, cm); xpt_done(ccb); return; } mpssas_log_command(cm, MPS_XINFO, "ioc %x scsi %x state %x xfer %u\n", le16toh(rep->IOCStatus), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); /* * If this is a Direct Drive I/O, reissue the I/O to the original IR * Volume if an error occurred (normal I/O retry). Use the original * CCB, but set a flag that this will be a retry so that it's sent to * the original volume. Free the command but reuse the CCB. */ if (cm->cm_flags & MPS_CM_FLAGS_DD_IO) { mps_free_command(sc, cm); ccb->ccb_h.sim_priv.entries[0].field = MPS_WD_RETRY; mpssas_action_scsiio(sassc, ccb); return; } else ccb->ccb_h.sim_priv.entries[0].field = 0; switch (le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) { case MPI2_IOCSTATUS_SCSI_DATA_UNDERRUN: csio->resid = cm->cm_length - le32toh(rep->TransferCount); /* FALLTHROUGH */ case MPI2_IOCSTATUS_SUCCESS: case MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR: if ((le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK) == MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR) mpssas_log_command(cm, MPS_XINFO, "recovered error\n"); /* Completion failed at the transport level. */ if (rep->SCSIState & (MPI2_SCSI_STATE_NO_SCSI_STATUS | MPI2_SCSI_STATE_TERMINATED)) { mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); break; } /* In a modern packetized environment, an autosense failure * implies that there's not much else that can be done to * recover the command. */ if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_FAILED) { mpssas_set_ccbstatus(ccb, CAM_AUTOSENSE_FAIL); break; } /* * CAM doesn't care about SAS Response Info data, but if this is * the state check if TLR should be done. If not, clear the * TLR_bits for the target. */ if ((rep->SCSIState & MPI2_SCSI_STATE_RESPONSE_INFO_VALID) && ((le32toh(rep->ResponseInfo) & MPI2_SCSI_RI_MASK_REASONCODE) == MPS_SCSI_RI_INVALID_FRAME)) { sc->mapping_table[target_id].TLR_bits = (u8)MPI2_SCSIIO_CONTROL_NO_TLR; } /* * Intentionally override the normal SCSI status reporting * for these two cases. These are likely to happen in a * multi-initiator environment, and we want to make sure that * CAM retries these commands rather than fail them. */ if ((rep->SCSIStatus == MPI2_SCSI_STATUS_COMMAND_TERMINATED) || (rep->SCSIStatus == MPI2_SCSI_STATUS_TASK_ABORTED)) { mpssas_set_ccbstatus(ccb, CAM_REQ_ABORTED); break; } /* Handle normal status and sense */ csio->scsi_status = rep->SCSIStatus; if (rep->SCSIStatus == MPI2_SCSI_STATUS_GOOD) mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); else mpssas_set_ccbstatus(ccb, CAM_SCSI_STATUS_ERROR); if (rep->SCSIState & MPI2_SCSI_STATE_AUTOSENSE_VALID) { int sense_len, returned_sense_len; returned_sense_len = min(le32toh(rep->SenseCount), sizeof(struct scsi_sense_data)); if (returned_sense_len < ccb->csio.sense_len) ccb->csio.sense_resid = ccb->csio.sense_len - returned_sense_len; else ccb->csio.sense_resid = 0; sense_len = min(returned_sense_len, ccb->csio.sense_len - ccb->csio.sense_resid); bzero(&ccb->csio.sense_data, sizeof(ccb->csio.sense_data)); bcopy(cm->cm_sense, &ccb->csio.sense_data, sense_len); ccb->ccb_h.status |= CAM_AUTOSNS_VALID; } /* * Check if this is an INQUIRY command. If it's a VPD inquiry, * and it's page code 0 (Supported Page List), and there is * inquiry data, and this is for a sequential access device, and * the device is an SSP target, and TLR is supported by the * controller, turn the TLR_bits value ON if page 0x90 is * supported. */ if ((csio->cdb_io.cdb_bytes[0] == INQUIRY) && (csio->cdb_io.cdb_bytes[1] & SI_EVPD) && (csio->cdb_io.cdb_bytes[2] == SVPD_SUPPORTED_PAGE_LIST) && ((csio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_VADDR) && (csio->data_ptr != NULL) && ((csio->data_ptr[0] & 0x1f) == T_SEQUENTIAL) && (sc->control_TLR) && (sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_SSP_TARGET)) { vpd_list = (struct scsi_vpd_supported_page_list *) csio->data_ptr; TLR_bits = &sc->mapping_table[target_id].TLR_bits; *TLR_bits = (u8)MPI2_SCSIIO_CONTROL_NO_TLR; TLR_on = (u8)MPI2_SCSIIO_CONTROL_TLR_ON; alloc_len = ((u16)csio->cdb_io.cdb_bytes[3] << 8) + csio->cdb_io.cdb_bytes[4]; alloc_len -= csio->resid; for (i = 0; i < MIN(vpd_list->length, alloc_len); i++) { if (vpd_list->list[i] == 0x90) { *TLR_bits = TLR_on; break; } } } /* * If this is a SATA direct-access end device, mark it so that * a SCSI StartStopUnit command will be sent to it when the * driver is being shutdown. */ if ((csio->cdb_io.cdb_bytes[0] == INQUIRY) && ((csio->data_ptr[0] & 0x1f) == T_DIRECT) && (sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_SATA_DEVICE) && ((sc->mapping_table[target_id].device_info & MPI2_SAS_DEVICE_INFO_MASK_DEVICE_TYPE) == MPI2_SAS_DEVICE_INFO_END_DEVICE)) { target = &sassc->targets[target_id]; target->supports_SSU = TRUE; mps_dprint(sc, MPS_XINFO, "Target %d supports SSU\n", target_id); } break; case MPI2_IOCSTATUS_SCSI_INVALID_DEVHANDLE: case MPI2_IOCSTATUS_SCSI_DEVICE_NOT_THERE: /* * If devinfo is 0 this will be a volume. In that case don't * tell CAM that the volume is not there. We want volumes to * be enumerated until they are deleted/removed, not just * failed. */ if (cm->cm_targ->devinfo == 0) mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); else mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); break; case MPI2_IOCSTATUS_INVALID_SGL: mps_print_scsiio_cmd(sc, cm); mpssas_set_ccbstatus(ccb, CAM_UNREC_HBA_ERROR); break; case MPI2_IOCSTATUS_SCSI_TASK_TERMINATED: /* * This is one of the responses that comes back when an I/O * has been aborted. If it is because of a timeout that we * initiated, just set the status to CAM_CMD_TIMEOUT. * Otherwise set it to CAM_REQ_ABORTED. The effect on the * command is the same (it gets retried, subject to the * retry counter), the only difference is what gets printed * on the console. */ - if (cm->cm_state == MPS_CM_STATE_TIMEDOUT) + if (cm->cm_flags & MPS_CM_FLAGS_TIMEDOUT) mpssas_set_ccbstatus(ccb, CAM_CMD_TIMEOUT); else mpssas_set_ccbstatus(ccb, CAM_REQ_ABORTED); break; case MPI2_IOCSTATUS_SCSI_DATA_OVERRUN: /* resid is ignored for this condition */ csio->resid = 0; mpssas_set_ccbstatus(ccb, CAM_DATA_RUN_ERR); break; case MPI2_IOCSTATUS_SCSI_IOC_TERMINATED: case MPI2_IOCSTATUS_SCSI_EXT_TERMINATED: /* * These can sometimes be transient transport-related * errors, and sometimes persistent drive-related errors. * We used to retry these without decrementing the retry * count by returning CAM_REQUEUE_REQ. Unfortunately, if * we hit a persistent drive problem that returns one of * these error codes, we would retry indefinitely. So, * return CAM_REQ_CMP_ERROR so that we decrement the retry * count and avoid infinite retries. We're taking the * potential risk of flagging false failures in the event * of a topology-related error (e.g. a SAS expander problem * causes a command addressed to a drive to fail), but * avoiding getting into an infinite retry loop. */ mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); mps_dprint(sc, MPS_INFO, "Controller reported %s tgt %u SMID %u loginfo %x\n", mps_describe_table(mps_iocstatus_string, le16toh(rep->IOCStatus) & MPI2_IOCSTATUS_MASK), target_id, cm->cm_desc.Default.SMID, le32toh(rep->IOCLogInfo)); mps_dprint(sc, MPS_XINFO, "SCSIStatus %x SCSIState %x xfercount %u\n", rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); break; case MPI2_IOCSTATUS_INVALID_FUNCTION: case MPI2_IOCSTATUS_INTERNAL_ERROR: case MPI2_IOCSTATUS_INVALID_VPID: case MPI2_IOCSTATUS_INVALID_FIELD: case MPI2_IOCSTATUS_INVALID_STATE: case MPI2_IOCSTATUS_OP_STATE_NOT_SUPPORTED: case MPI2_IOCSTATUS_SCSI_IO_DATA_ERROR: case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR: case MPI2_IOCSTATUS_SCSI_RESIDUAL_MISMATCH: case MPI2_IOCSTATUS_SCSI_TASK_MGMT_FAILED: default: mpssas_log_command(cm, MPS_XINFO, "completed ioc %x loginfo %x scsi %x state %x xfer %u\n", le16toh(rep->IOCStatus), le32toh(rep->IOCLogInfo), rep->SCSIStatus, rep->SCSIState, le32toh(rep->TransferCount)); csio->resid = cm->cm_length; mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); break; } mps_sc_failed_io_info(sc,csio,rep); if (sassc->flags & MPSSAS_QUEUE_FROZEN) { ccb->ccb_h.status |= CAM_RELEASE_SIMQ; sassc->flags &= ~MPSSAS_QUEUE_FROZEN; mps_dprint(sc, MPS_XINFO, "Command completed, " "unfreezing SIM queue\n"); } if (mpssas_get_ccbstatus(ccb) != CAM_REQ_CMP) { ccb->ccb_h.status |= CAM_DEV_QFRZN; xpt_freeze_devq(ccb->ccb_h.path, /*count*/ 1); } mps_free_command(sc, cm); xpt_done(ccb); } /* All Request reached here are Endian safe */ static void mpssas_direct_drive_io(struct mpssas_softc *sassc, struct mps_command *cm, union ccb *ccb) { pMpi2SCSIIORequest_t pIO_req; struct mps_softc *sc = sassc->sc; uint64_t virtLBA; uint32_t physLBA, stripe_offset, stripe_unit; uint32_t io_size, column; uint8_t *ptrLBA, lba_idx, physLBA_byte, *CDB; /* * If this is a valid SCSI command (Read6, Read10, Read16, Write6, * Write10, or Write16), build a direct I/O message. Otherwise, the I/O * will be sent to the IR volume itself. Since Read6 and Write6 are a * bit different than the 10/16 CDBs, handle them separately. */ pIO_req = (pMpi2SCSIIORequest_t)cm->cm_req; CDB = pIO_req->CDB.CDB32; /* * Handle 6 byte CDBs. */ if ((pIO_req->DevHandle == sc->DD_dev_handle) && ((CDB[0] == READ_6) || (CDB[0] == WRITE_6))) { /* * Get the transfer size in blocks. */ io_size = (cm->cm_length >> sc->DD_block_exponent); /* * Get virtual LBA given in the CDB. */ virtLBA = ((uint64_t)(CDB[1] & 0x1F) << 16) | ((uint64_t)CDB[2] << 8) | (uint64_t)CDB[3]; /* * Check that LBA range for I/O does not exceed volume's * MaxLBA. */ if ((virtLBA + (uint64_t)io_size - 1) <= sc->DD_max_lba) { /* * Check if the I/O crosses a stripe boundary. If not, * translate the virtual LBA to a physical LBA and set * the DevHandle for the PhysDisk to be used. If it * does cross a boundary, do normal I/O. To get the * right DevHandle to use, get the map number for the * column, then use that map number to look up the * DevHandle of the PhysDisk. */ stripe_offset = (uint32_t)virtLBA & (sc->DD_stripe_size - 1); if ((stripe_offset + io_size) <= sc->DD_stripe_size) { physLBA = (uint32_t)virtLBA >> sc->DD_stripe_exponent; stripe_unit = physLBA / sc->DD_num_phys_disks; column = physLBA % sc->DD_num_phys_disks; pIO_req->DevHandle = htole16(sc->DD_column_map[column].dev_handle); /* ???? Is this endian safe*/ cm->cm_desc.SCSIIO.DevHandle = pIO_req->DevHandle; physLBA = (stripe_unit << sc->DD_stripe_exponent) + stripe_offset; ptrLBA = &pIO_req->CDB.CDB32[1]; physLBA_byte = (uint8_t)(physLBA >> 16); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[2]; physLBA_byte = (uint8_t)(physLBA >> 8); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[3]; physLBA_byte = (uint8_t)physLBA; *ptrLBA = physLBA_byte; /* * Set flag that Direct Drive I/O is * being done. */ cm->cm_flags |= MPS_CM_FLAGS_DD_IO; } } return; } /* * Handle 10, 12 or 16 byte CDBs. */ if ((pIO_req->DevHandle == sc->DD_dev_handle) && ((CDB[0] == READ_10) || (CDB[0] == WRITE_10) || (CDB[0] == READ_16) || (CDB[0] == WRITE_16) || (CDB[0] == READ_12) || (CDB[0] == WRITE_12))) { /* * For 16-byte CDB's, verify that the upper 4 bytes of the CDB * are 0. If not, this is accessing beyond 2TB so handle it in * the else section. 10-byte and 12-byte CDB's are OK. * FreeBSD sends very rare 12 byte READ/WRITE, but driver is * ready to accept 12byte CDB for Direct IOs. */ if ((CDB[0] == READ_10 || CDB[0] == WRITE_10) || (CDB[0] == READ_12 || CDB[0] == WRITE_12) || !(CDB[2] | CDB[3] | CDB[4] | CDB[5])) { /* * Get the transfer size in blocks. */ io_size = (cm->cm_length >> sc->DD_block_exponent); /* * Get virtual LBA. Point to correct lower 4 bytes of * LBA in the CDB depending on command. */ lba_idx = ((CDB[0] == READ_12) || (CDB[0] == WRITE_12) || (CDB[0] == READ_10) || (CDB[0] == WRITE_10))? 2 : 6; virtLBA = ((uint64_t)CDB[lba_idx] << 24) | ((uint64_t)CDB[lba_idx + 1] << 16) | ((uint64_t)CDB[lba_idx + 2] << 8) | (uint64_t)CDB[lba_idx + 3]; /* * Check that LBA range for I/O does not exceed volume's * MaxLBA. */ if ((virtLBA + (uint64_t)io_size - 1) <= sc->DD_max_lba) { /* * Check if the I/O crosses a stripe boundary. * If not, translate the virtual LBA to a * physical LBA and set the DevHandle for the * PhysDisk to be used. If it does cross a * boundary, do normal I/O. To get the right * DevHandle to use, get the map number for the * column, then use that map number to look up * the DevHandle of the PhysDisk. */ stripe_offset = (uint32_t)virtLBA & (sc->DD_stripe_size - 1); if ((stripe_offset + io_size) <= sc->DD_stripe_size) { physLBA = (uint32_t)virtLBA >> sc->DD_stripe_exponent; stripe_unit = physLBA / sc->DD_num_phys_disks; column = physLBA % sc->DD_num_phys_disks; pIO_req->DevHandle = htole16(sc->DD_column_map[column]. dev_handle); cm->cm_desc.SCSIIO.DevHandle = pIO_req->DevHandle; physLBA = (stripe_unit << sc->DD_stripe_exponent) + stripe_offset; ptrLBA = &pIO_req->CDB.CDB32[lba_idx]; physLBA_byte = (uint8_t)(physLBA >> 24); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[lba_idx + 1]; physLBA_byte = (uint8_t)(physLBA >> 16); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[lba_idx + 2]; physLBA_byte = (uint8_t)(physLBA >> 8); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[lba_idx + 3]; physLBA_byte = (uint8_t)physLBA; *ptrLBA = physLBA_byte; /* * Set flag that Direct Drive I/O is * being done. */ cm->cm_flags |= MPS_CM_FLAGS_DD_IO; } } } else { /* * 16-byte CDB and the upper 4 bytes of the CDB are not * 0. Get the transfer size in blocks. */ io_size = (cm->cm_length >> sc->DD_block_exponent); /* * Get virtual LBA. */ virtLBA = ((uint64_t)CDB[2] << 54) | ((uint64_t)CDB[3] << 48) | ((uint64_t)CDB[4] << 40) | ((uint64_t)CDB[5] << 32) | ((uint64_t)CDB[6] << 24) | ((uint64_t)CDB[7] << 16) | ((uint64_t)CDB[8] << 8) | (uint64_t)CDB[9]; /* * Check that LBA range for I/O does not exceed volume's * MaxLBA. */ if ((virtLBA + (uint64_t)io_size - 1) <= sc->DD_max_lba) { /* * Check if the I/O crosses a stripe boundary. * If not, translate the virtual LBA to a * physical LBA and set the DevHandle for the * PhysDisk to be used. If it does cross a * boundary, do normal I/O. To get the right * DevHandle to use, get the map number for the * column, then use that map number to look up * the DevHandle of the PhysDisk. */ stripe_offset = (uint32_t)virtLBA & (sc->DD_stripe_size - 1); if ((stripe_offset + io_size) <= sc->DD_stripe_size) { physLBA = (uint32_t)(virtLBA >> sc->DD_stripe_exponent); stripe_unit = physLBA / sc->DD_num_phys_disks; column = physLBA % sc->DD_num_phys_disks; pIO_req->DevHandle = htole16(sc->DD_column_map[column]. dev_handle); cm->cm_desc.SCSIIO.DevHandle = pIO_req->DevHandle; physLBA = (stripe_unit << sc->DD_stripe_exponent) + stripe_offset; /* * Set upper 4 bytes of LBA to 0. We * assume that the phys disks are less * than 2 TB's in size. Then, set the * lower 4 bytes. */ pIO_req->CDB.CDB32[2] = 0; pIO_req->CDB.CDB32[3] = 0; pIO_req->CDB.CDB32[4] = 0; pIO_req->CDB.CDB32[5] = 0; ptrLBA = &pIO_req->CDB.CDB32[6]; physLBA_byte = (uint8_t)(physLBA >> 24); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[7]; physLBA_byte = (uint8_t)(physLBA >> 16); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[8]; physLBA_byte = (uint8_t)(physLBA >> 8); *ptrLBA = physLBA_byte; ptrLBA = &pIO_req->CDB.CDB32[9]; physLBA_byte = (uint8_t)physLBA; *ptrLBA = physLBA_byte; /* * Set flag that Direct Drive I/O is * being done. */ cm->cm_flags |= MPS_CM_FLAGS_DD_IO; } } } } } #if __FreeBSD_version >= 900026 static void mpssas_smpio_complete(struct mps_softc *sc, struct mps_command *cm) { MPI2_SMP_PASSTHROUGH_REPLY *rpl; MPI2_SMP_PASSTHROUGH_REQUEST *req; uint64_t sasaddr; union ccb *ccb; ccb = cm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and SMP * commands require two S/G elements only. That should be handled * in the standard request size. */ if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_ERROR,"%s: cm_flags = %#x on SMP request!\n", __func__, cm->cm_flags); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } rpl = (MPI2_SMP_PASSTHROUGH_REPLY *)cm->cm_reply; if (rpl == NULL) { mps_dprint(sc, MPS_ERROR, "%s: NULL cm_reply!\n", __func__); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req; sasaddr = le32toh(req->SASAddress.Low); sasaddr |= ((uint64_t)(le32toh(req->SASAddress.High))) << 32; if ((le16toh(rpl->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS || rpl->SASStatus != MPI2_SASSTATUS_SUCCESS) { mps_dprint(sc, MPS_XINFO, "%s: IOCStatus %04x SASStatus %02x\n", __func__, le16toh(rpl->IOCStatus), rpl->SASStatus); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } mps_dprint(sc, MPS_XINFO, "%s: SMP request to SAS address " "%#jx completed successfully\n", __func__, (uintmax_t)sasaddr); if (ccb->smpio.smp_response[2] == SMP_FR_ACCEPTED) mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); else mpssas_set_ccbstatus(ccb, CAM_SMP_STATUS_ERROR); bailout: /* * We sync in both directions because we had DMAs in the S/G list * in both directions. */ bus_dmamap_sync(sc->buffer_dmat, cm->cm_dmamap, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->buffer_dmat, cm->cm_dmamap); mps_free_command(sc, cm); xpt_done(ccb); } static void mpssas_send_smpcmd(struct mpssas_softc *sassc, union ccb *ccb, uint64_t sasaddr) { struct mps_command *cm; uint8_t *request, *response; MPI2_SMP_PASSTHROUGH_REQUEST *req; struct mps_softc *sc; int error; sc = sassc->sc; error = 0; /* * XXX We don't yet support physical addresses here. */ switch ((ccb->ccb_h.flags & CAM_DATA_MASK)) { case CAM_DATA_PADDR: case CAM_DATA_SG_PADDR: mps_dprint(sc, MPS_ERROR, "%s: physical addresses not supported\n", __func__); mpssas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; case CAM_DATA_SG: /* * The chip does not support more than one buffer for the * request or response. */ if ((ccb->smpio.smp_request_sglist_cnt > 1) || (ccb->smpio.smp_response_sglist_cnt > 1)) { mps_dprint(sc, MPS_ERROR, "%s: multiple request or response " "buffer segments not supported for SMP\n", __func__); mpssas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } /* * The CAM_SCATTER_VALID flag was originally implemented * for the XPT_SCSI_IO CCB, which only has one data pointer. * We have two. So, just take that flag to mean that we * might have S/G lists, and look at the S/G segment count * to figure out whether that is the case for each individual * buffer. */ if (ccb->smpio.smp_request_sglist_cnt != 0) { bus_dma_segment_t *req_sg; req_sg = (bus_dma_segment_t *)ccb->smpio.smp_request; request = (uint8_t *)(uintptr_t)req_sg[0].ds_addr; } else request = ccb->smpio.smp_request; if (ccb->smpio.smp_response_sglist_cnt != 0) { bus_dma_segment_t *rsp_sg; rsp_sg = (bus_dma_segment_t *)ccb->smpio.smp_response; response = (uint8_t *)(uintptr_t)rsp_sg[0].ds_addr; } else response = ccb->smpio.smp_response; break; case CAM_DATA_VADDR: request = ccb->smpio.smp_request; response = ccb->smpio.smp_response; break; default: mpssas_set_ccbstatus(ccb, CAM_REQ_INVALID); xpt_done(ccb); return; } cm = mps_alloc_command(sc); if (cm == NULL) { mps_dprint(sc, MPS_ERROR, "%s: cannot allocate command\n", __func__); mpssas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } req = (MPI2_SMP_PASSTHROUGH_REQUEST *)cm->cm_req; bzero(req, sizeof(*req)); req->Function = MPI2_FUNCTION_SMP_PASSTHROUGH; /* Allow the chip to use any route to this SAS address. */ req->PhysicalPort = 0xff; req->RequestDataLength = htole16(ccb->smpio.smp_request_len); req->SGLFlags = MPI2_SGLFLAGS_SYSTEM_ADDRESS_SPACE | MPI2_SGLFLAGS_SGL_TYPE_MPI; mps_dprint(sc, MPS_XINFO, "%s: sending SMP request to SAS " "address %#jx\n", __func__, (uintmax_t)sasaddr); mpi_init_sge(cm, req, &req->SGL); /* * Set up a uio to pass into mps_map_command(). This allows us to * do one map command, and one busdma call in there. */ cm->cm_uio.uio_iov = cm->cm_iovec; cm->cm_uio.uio_iovcnt = 2; cm->cm_uio.uio_segflg = UIO_SYSSPACE; /* * The read/write flag isn't used by busdma, but set it just in * case. This isn't exactly accurate, either, since we're going in * both directions. */ cm->cm_uio.uio_rw = UIO_WRITE; cm->cm_iovec[0].iov_base = request; cm->cm_iovec[0].iov_len = le16toh(req->RequestDataLength); cm->cm_iovec[1].iov_base = response; cm->cm_iovec[1].iov_len = ccb->smpio.smp_response_len; cm->cm_uio.uio_resid = cm->cm_iovec[0].iov_len + cm->cm_iovec[1].iov_len; /* * Trigger a warning message in mps_data_cb() for the user if we * wind up exceeding two S/G segments. The chip expects one * segment for the request and another for the response. */ cm->cm_max_segs = 2; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete = mpssas_smpio_complete; cm->cm_complete_data = ccb; /* * Tell the mapping code that we're using a uio, and that this is * an SMP passthrough request. There is a little special-case * logic there (in mps_data_cb()) to handle the bidirectional * transfer. */ cm->cm_flags |= MPS_CM_FLAGS_USE_UIO | MPS_CM_FLAGS_SMP_PASS | MPS_CM_FLAGS_DATAIN | MPS_CM_FLAGS_DATAOUT; /* The chip data format is little endian. */ req->SASAddress.High = htole32(sasaddr >> 32); req->SASAddress.Low = htole32(sasaddr); /* * XXX Note that we don't have a timeout/abort mechanism here. * From the manual, it looks like task management requests only * work for SCSI IO and SATA passthrough requests. We may need to * have a mechanism to retry requests in the event of a chip reset * at least. Hopefully the chip will insure that any errors short * of that are relayed back to the driver. */ error = mps_map_command(sc, cm); if ((error != 0) && (error != EINPROGRESS)) { mps_dprint(sc, MPS_ERROR, "%s: error %d returned from mps_map_command()\n", __func__, error); goto bailout_error; } return; bailout_error: mps_free_command(sc, cm); mpssas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } static void mpssas_action_smpio(struct mpssas_softc *sassc, union ccb *ccb) { struct mps_softc *sc; struct mpssas_target *targ; uint64_t sasaddr = 0; sc = sassc->sc; /* * Make sure the target exists. */ KASSERT(ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_SMP_IO\n", ccb->ccb_h.target_id)); targ = &sassc->targets[ccb->ccb_h.target_id]; if (targ->handle == 0x0) { mps_dprint(sc, MPS_ERROR, "%s: target %d does not exist!\n", __func__, ccb->ccb_h.target_id); mpssas_set_ccbstatus(ccb, CAM_SEL_TIMEOUT); xpt_done(ccb); return; } /* * If this device has an embedded SMP target, we'll talk to it * directly. * figure out what the expander's address is. */ if ((targ->devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) != 0) sasaddr = targ->sasaddr; /* * If we don't have a SAS address for the expander yet, try * grabbing it from the page 0x83 information cached in the * transport layer for this target. LSI expanders report the * expander SAS address as the port-associated SAS address in * Inquiry VPD page 0x83. Maxim expanders don't report it in page * 0x83. * * XXX KDM disable this for now, but leave it commented out so that * it is obvious that this is another possible way to get the SAS * address. * * The parent handle method below is a little more reliable, and * the other benefit is that it works for devices other than SES * devices. So you can send a SMP request to a da(4) device and it * will get routed to the expander that device is attached to. * (Assuming the da(4) device doesn't contain an SMP target...) */ #if 0 if (sasaddr == 0) sasaddr = xpt_path_sas_addr(ccb->ccb_h.path); #endif /* * If we still don't have a SAS address for the expander, look for * the parent device of this device, which is probably the expander. */ if (sasaddr == 0) { #ifdef OLD_MPS_PROBE struct mpssas_target *parent_target; #endif if (targ->parent_handle == 0x0) { mps_dprint(sc, MPS_ERROR, "%s: handle %d does not have a valid " "parent handle!\n", __func__, targ->handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } #ifdef OLD_MPS_PROBE parent_target = mpssas_find_target_by_handle(sassc, 0, targ->parent_handle); if (parent_target == NULL) { mps_dprint(sc, MPS_ERROR, "%s: handle %d does not have a valid " "parent target!\n", __func__, targ->handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } if ((parent_target->devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) { mps_dprint(sc, MPS_ERROR, "%s: handle %d parent %d does not " "have an SMP target!\n", __func__, targ->handle, parent_target->handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } sasaddr = parent_target->sasaddr; #else /* OLD_MPS_PROBE */ if ((targ->parent_devinfo & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) { mps_dprint(sc, MPS_ERROR, "%s: handle %d parent %d does not " "have an SMP target!\n", __func__, targ->handle, targ->parent_handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } if (targ->parent_sasaddr == 0x0) { mps_dprint(sc, MPS_ERROR, "%s: handle %d parent handle %d does " "not have a valid SAS address!\n", __func__, targ->handle, targ->parent_handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } sasaddr = targ->parent_sasaddr; #endif /* OLD_MPS_PROBE */ } if (sasaddr == 0) { mps_dprint(sc, MPS_INFO, "%s: unable to find SAS address for handle %d\n", __func__, targ->handle); mpssas_set_ccbstatus(ccb, CAM_DEV_NOT_THERE); goto bailout; } mpssas_send_smpcmd(sassc, ccb, sasaddr); return; bailout: xpt_done(ccb); } #endif //__FreeBSD_version >= 900026 static void mpssas_action_resetdev(struct mpssas_softc *sassc, union ccb *ccb) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; struct mps_softc *sc; struct mps_command *tm; struct mpssas_target *targ; MPS_FUNCTRACE(sassc->sc); mtx_assert(&sassc->sc->mps_mtx, MA_OWNED); KASSERT(ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in XPT_RESET_DEV\n", ccb->ccb_h.target_id)); sc = sassc->sc; tm = mpssas_alloc_tm(sc); if (tm == NULL) { mps_dprint(sc, MPS_ERROR, "command alloc failure in mpssas_action_resetdev\n"); mpssas_set_ccbstatus(ccb, CAM_RESRC_UNAVAIL); xpt_done(ccb); return; } targ = &sassc->targets[ccb->ccb_h.target_id]; req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; req->DevHandle = htole16(targ->handle); req->TaskType = MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET; /* SAS Hard Link Reset / SATA Link Reset */ req->MsgFlags = MPI2_SCSITASKMGMT_MSGFLAGS_LINK_RESET; tm->cm_data = NULL; tm->cm_complete = mpssas_resetdev_complete; tm->cm_complete_data = ccb; tm->cm_targ = targ; mpssas_prepare_for_tm(sc, tm, targ, CAM_LUN_WILDCARD); mps_map_command(sc, tm); } static void mpssas_resetdev_complete(struct mps_softc *sc, struct mps_command *tm) { MPI2_SCSI_TASK_MANAGE_REPLY *resp; union ccb *ccb; MPS_FUNCTRACE(sc); mtx_assert(&sc->mps_mtx, MA_OWNED); resp = (MPI2_SCSI_TASK_MANAGE_REPLY *)tm->cm_reply; ccb = tm->cm_complete_data; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * task management commands don't have S/G lists. */ if ((tm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { MPI2_SCSI_TASK_MANAGE_REQUEST *req; req = (MPI2_SCSI_TASK_MANAGE_REQUEST *)tm->cm_req; mps_dprint(sc, MPS_ERROR, "%s: cm_flags = %#x for reset of handle %#04x! " "This should not happen!\n", __func__, tm->cm_flags, req->DevHandle); mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); goto bailout; } mps_dprint(sc, MPS_XINFO, "%s: IOCStatus = 0x%x ResponseCode = 0x%x\n", __func__, le16toh(resp->IOCStatus), le32toh(resp->ResponseCode)); if (le32toh(resp->ResponseCode) == MPI2_SCSITASKMGMT_RSP_TM_COMPLETE) { mpssas_set_ccbstatus(ccb, CAM_REQ_CMP); mpssas_announce_reset(sc, AC_SENT_BDR, tm->cm_targ->tid, CAM_LUN_WILDCARD); } else mpssas_set_ccbstatus(ccb, CAM_REQ_CMP_ERR); bailout: mpssas_free_tm(sc, tm); xpt_done(ccb); } static void mpssas_poll(struct cam_sim *sim) { struct mpssas_softc *sassc; sassc = cam_sim_softc(sim); if (sassc->sc->mps_debug & MPS_TRACE) { /* frequent debug messages during a panic just slow * everything down too much. */ mps_printf(sassc->sc, "%s clearing MPS_TRACE\n", __func__); sassc->sc->mps_debug &= ~MPS_TRACE; } mps_intr_locked(sassc->sc); } static void mpssas_async(void *callback_arg, uint32_t code, struct cam_path *path, void *arg) { struct mps_softc *sc; sc = (struct mps_softc *)callback_arg; switch (code) { #if (__FreeBSD_version >= 1000006) || \ ((__FreeBSD_version >= 901503) && (__FreeBSD_version < 1000000)) case AC_ADVINFO_CHANGED: { struct mpssas_target *target; struct mpssas_softc *sassc; struct scsi_read_capacity_data_long rcap_buf; struct ccb_dev_advinfo cdai; struct mpssas_lun *lun; lun_id_t lunid; int found_lun; uintptr_t buftype; buftype = (uintptr_t)arg; found_lun = 0; sassc = sc->sassc; /* * We're only interested in read capacity data changes. */ if (buftype != CDAI_TYPE_RCAPLONG) break; /* * We should have a handle for this, but check to make sure. */ KASSERT(xpt_path_target_id(path) < sassc->maxtargets, ("Target %d out of bounds in mpssas_async\n", xpt_path_target_id(path))); target = &sassc->targets[xpt_path_target_id(path)]; if (target->handle == 0) break; lunid = xpt_path_lun_id(path); SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id == lunid) { found_lun = 1; break; } } if (found_lun == 0) { lun = malloc(sizeof(struct mpssas_lun), M_MPT2, M_NOWAIT | M_ZERO); if (lun == NULL) { mps_dprint(sc, MPS_ERROR, "Unable to alloc " "LUN for EEDP support.\n"); break; } lun->lun_id = lunid; SLIST_INSERT_HEAD(&target->luns, lun, lun_link); } bzero(&rcap_buf, sizeof(rcap_buf)); xpt_setup_ccb(&cdai.ccb_h, path, CAM_PRIORITY_NORMAL); cdai.ccb_h.func_code = XPT_DEV_ADVINFO; cdai.ccb_h.flags = CAM_DIR_IN; cdai.buftype = CDAI_TYPE_RCAPLONG; #if (__FreeBSD_version >= 1100061) || \ ((__FreeBSD_version >= 1001510) && (__FreeBSD_version < 1100000)) cdai.flags = CDAI_FLAG_NONE; #else cdai.flags = 0; #endif cdai.bufsiz = sizeof(rcap_buf); cdai.buf = (uint8_t *)&rcap_buf; xpt_action((union ccb *)&cdai); if ((cdai.ccb_h.status & CAM_DEV_QFRZN) != 0) cam_release_devq(cdai.ccb_h.path, 0, 0, 0, FALSE); if ((mpssas_get_ccbstatus((union ccb *)&cdai) == CAM_REQ_CMP) && (rcap_buf.prot & SRC16_PROT_EN)) { switch (rcap_buf.prot & SRC16_P_TYPE) { case SRC16_PTYPE_1: case SRC16_PTYPE_3: lun->eedp_formatted = TRUE; lun->eedp_block_size = scsi_4btoul(rcap_buf.length); break; case SRC16_PTYPE_2: default: lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; break; } } else { lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; } break; } #else case AC_FOUND_DEVICE: { struct ccb_getdev *cgd; cgd = arg; mpssas_check_eedp(sc, path, cgd); break; } #endif default: break; } } #if (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) static void mpssas_check_eedp(struct mps_softc *sc, struct cam_path *path, struct ccb_getdev *cgd) { struct mpssas_softc *sassc = sc->sassc; struct ccb_scsiio *csio; struct scsi_read_capacity_16 *scsi_cmd; struct scsi_read_capacity_eedp *rcap_buf; path_id_t pathid; target_id_t targetid; lun_id_t lunid; union ccb *ccb; struct cam_path *local_path; struct mpssas_target *target; struct mpssas_lun *lun; uint8_t found_lun; char path_str[64]; sassc = sc->sassc; pathid = cam_sim_path(sassc->sim); targetid = xpt_path_target_id(path); lunid = xpt_path_lun_id(path); KASSERT(targetid < sassc->maxtargets, ("Target %d out of bounds in mpssas_check_eedp\n", targetid)); target = &sassc->targets[targetid]; if (target->handle == 0x0) return; /* * Determine if the device is EEDP capable. * * If this flag is set in the inquiry data, * the device supports protection information, * and must support the 16 byte read * capacity command, otherwise continue without * sending read cap 16 */ if ((cgd->inq_data.spc3_flags & SPC3_SID_PROTECT) == 0) return; /* * Issue a READ CAPACITY 16 command. This info * is used to determine if the LUN is formatted * for EEDP support. */ ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mps_dprint(sc, MPS_ERROR, "Unable to alloc CCB " "for EEDP support.\n"); return; } if (xpt_create_path(&local_path, xpt_periph, pathid, targetid, lunid) != CAM_REQ_CMP) { mps_dprint(sc, MPS_ERROR, "Unable to create " "path for EEDP support\n"); xpt_free_ccb(ccb); return; } /* * If LUN is already in list, don't create a new * one. */ found_lun = FALSE; SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id == lunid) { found_lun = TRUE; break; } } if (!found_lun) { lun = malloc(sizeof(struct mpssas_lun), M_MPT2, M_NOWAIT | M_ZERO); if (lun == NULL) { mps_dprint(sc, MPS_ERROR, "Unable to alloc LUN for EEDP support.\n"); xpt_free_path(local_path); xpt_free_ccb(ccb); return; } lun->lun_id = lunid; SLIST_INSERT_HEAD(&target->luns, lun, lun_link); } xpt_path_string(local_path, path_str, sizeof(path_str)); mps_dprint(sc, MPS_INFO, "Sending read cap: path %s handle %d\n", path_str, target->handle); /* * Issue a READ CAPACITY 16 command for the LUN. * The mpssas_read_cap_done function will load * the read cap info into the LUN struct. */ rcap_buf = malloc(sizeof(struct scsi_read_capacity_eedp), M_MPT2, M_NOWAIT | M_ZERO); if (rcap_buf == NULL) { mps_dprint(sc, MPS_FAULT, "Unable to alloc read capacity buffer for EEDP support.\n"); xpt_free_path(ccb->ccb_h.path); xpt_free_ccb(ccb); return; } xpt_setup_ccb(&ccb->ccb_h, local_path, CAM_PRIORITY_XPT); csio = &ccb->csio; csio->ccb_h.func_code = XPT_SCSI_IO; csio->ccb_h.flags = CAM_DIR_IN; csio->ccb_h.retry_count = 4; csio->ccb_h.cbfcnp = mpssas_read_cap_done; csio->ccb_h.timeout = 60000; csio->data_ptr = (uint8_t *)rcap_buf; csio->dxfer_len = sizeof(struct scsi_read_capacity_eedp); csio->sense_len = MPS_SENSE_LEN; csio->cdb_len = sizeof(*scsi_cmd); csio->tag_action = MSG_SIMPLE_Q_TAG; scsi_cmd = (struct scsi_read_capacity_16 *)&csio->cdb_io.cdb_bytes; bzero(scsi_cmd, sizeof(*scsi_cmd)); scsi_cmd->opcode = 0x9E; scsi_cmd->service_action = SRC16_SERVICE_ACTION; ((uint8_t *)scsi_cmd)[13] = sizeof(struct scsi_read_capacity_eedp); ccb->ccb_h.ppriv_ptr1 = sassc; xpt_action(ccb); } static void mpssas_read_cap_done(struct cam_periph *periph, union ccb *done_ccb) { struct mpssas_softc *sassc; struct mpssas_target *target; struct mpssas_lun *lun; struct scsi_read_capacity_eedp *rcap_buf; if (done_ccb == NULL) return; /* Driver need to release devq, it Scsi command is * generated by driver internally. * Currently there is a single place where driver * calls scsi command internally. In future if driver * calls more scsi command internally, it needs to release * devq internally, since those command will not go back to * cam_periph. */ if ((done_ccb->ccb_h.status & CAM_DEV_QFRZN) ) { done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN; xpt_release_devq(done_ccb->ccb_h.path, /*count*/ 1, /*run_queue*/TRUE); } rcap_buf = (struct scsi_read_capacity_eedp *)done_ccb->csio.data_ptr; /* * Get the LUN ID for the path and look it up in the LUN list for the * target. */ sassc = (struct mpssas_softc *)done_ccb->ccb_h.ppriv_ptr1; KASSERT(done_ccb->ccb_h.target_id < sassc->maxtargets, ("Target %d out of bounds in mpssas_read_cap_done\n", done_ccb->ccb_h.target_id)); target = &sassc->targets[done_ccb->ccb_h.target_id]; SLIST_FOREACH(lun, &target->luns, lun_link) { if (lun->lun_id != done_ccb->ccb_h.target_lun) continue; /* * Got the LUN in the target's LUN list. Fill it in * with EEDP info. If the READ CAP 16 command had some * SCSI error (common if command is not supported), mark * the lun as not supporting EEDP and set the block size * to 0. */ if ((mpssas_get_ccbstatus(done_ccb) != CAM_REQ_CMP) || (done_ccb->csio.scsi_status != SCSI_STATUS_OK)) { lun->eedp_formatted = FALSE; lun->eedp_block_size = 0; break; } if (rcap_buf->protect & 0x01) { mps_dprint(sassc->sc, MPS_INFO, "LUN %d for " "target ID %d is formatted for EEDP " "support.\n", done_ccb->ccb_h.target_lun, done_ccb->ccb_h.target_id); lun->eedp_formatted = TRUE; lun->eedp_block_size = scsi_4btoul(rcap_buf->length); } break; } // Finished with this CCB and path. free(rcap_buf, M_MPT2); xpt_free_path(done_ccb->ccb_h.path); xpt_free_ccb(done_ccb); } #endif /* (__FreeBSD_version < 901503) || \ ((__FreeBSD_version >= 1000000) && (__FreeBSD_version < 1000006)) */ /* * Set the INRESET flag for this target so that no I/O will be sent to * the target until the reset has completed. If an I/O request does * happen, the devq will be frozen. The CCB holds the path which is * used to release the devq. The devq is released and the CCB is freed * when the TM completes. */ void mpssas_prepare_for_tm(struct mps_softc *sc, struct mps_command *tm, struct mpssas_target *target, lun_id_t lun_id) { union ccb *ccb; path_id_t path_id; ccb = xpt_alloc_ccb_nowait(); if (ccb) { path_id = cam_sim_path(sc->sassc->sim); if (xpt_create_path(&ccb->ccb_h.path, xpt_periph, path_id, target->tid, lun_id) != CAM_REQ_CMP) { xpt_free_ccb(ccb); } else { tm->cm_ccb = ccb; tm->cm_targ = target; target->flags |= MPSSAS_TARGET_INRESET; } } } int mpssas_startup(struct mps_softc *sc) { /* * Send the port enable message and set the wait_for_port_enable flag. * This flag helps to keep the simq frozen until all discovery events * are processed. */ sc->wait_for_port_enable = 1; mpssas_send_portenable(sc); return (0); } static int mpssas_send_portenable(struct mps_softc *sc) { MPI2_PORT_ENABLE_REQUEST *request; struct mps_command *cm; MPS_FUNCTRACE(sc); if ((cm = mps_alloc_command(sc)) == NULL) return (EBUSY); request = (MPI2_PORT_ENABLE_REQUEST *)cm->cm_req; request->Function = MPI2_FUNCTION_PORT_ENABLE; request->MsgFlags = 0; request->VP_ID = 0; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_complete = mpssas_portenable_complete; cm->cm_data = NULL; cm->cm_sge = NULL; mps_map_command(sc, cm); mps_dprint(sc, MPS_XINFO, "mps_send_portenable finished cm %p req %p complete %p\n", cm, cm->cm_req, cm->cm_complete); return (0); } static void mpssas_portenable_complete(struct mps_softc *sc, struct mps_command *cm) { MPI2_PORT_ENABLE_REPLY *reply; struct mpssas_softc *sassc; MPS_FUNCTRACE(sc); sassc = sc->sassc; /* * Currently there should be no way we can hit this case. It only * happens when we have a failure to allocate chain frames, and * port enable commands don't have S/G lists. */ if ((cm->cm_flags & MPS_CM_FLAGS_ERROR_MASK) != 0) { mps_dprint(sc, MPS_ERROR, "%s: cm_flags = %#x for port enable! " "This should not happen!\n", __func__, cm->cm_flags); } reply = (MPI2_PORT_ENABLE_REPLY *)cm->cm_reply; if (reply == NULL) mps_dprint(sc, MPS_FAULT, "Portenable NULL reply\n"); else if (le16toh(reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) mps_dprint(sc, MPS_FAULT, "Portenable failed\n"); mps_free_command(sc, cm); /* * Get WarpDrive info after discovery is complete but before the scan * starts. At this point, all devices are ready to be exposed to the * OS. If devices should be hidden instead, take them out of the * 'targets' array before the scan. The devinfo for a disk will have * some info and a volume's will be 0. Use that to remove disks. */ mps_wd_config_pages(sc); /* * Done waiting for port enable to complete. Decrement the refcount. * If refcount is 0, discovery is complete and a rescan of the bus can * take place. Since the simq was explicitly frozen before port * enable, it must be explicitly released here to keep the * freeze/release count in sync. */ sc->wait_for_port_enable = 0; sc->port_enable_complete = 1; wakeup(&sc->port_enable_complete); mpssas_startup_decrement(sassc); } int mpssas_check_id(struct mpssas_softc *sassc, int id) { struct mps_softc *sc = sassc->sc; char *ids; char *name; ids = &sc->exclude_ids[0]; while((name = strsep(&ids, ",")) != NULL) { if (name[0] == '\0') continue; if (strtol(name, NULL, 0) == (long)id) return (1); } return (0); } void mpssas_realloc_targets(struct mps_softc *sc, int maxtargets) { struct mpssas_softc *sassc; struct mpssas_lun *lun, *lun_tmp; struct mpssas_target *targ; int i; sassc = sc->sassc; /* * The number of targets is based on IOC Facts, so free all of * the allocated LUNs for each target and then the target buffer * itself. */ for (i=0; i< maxtargets; i++) { targ = &sassc->targets[i]; SLIST_FOREACH_SAFE(lun, &targ->luns, lun_link, lun_tmp) { free(lun, M_MPT2); } } free(sassc->targets, M_MPT2); sassc->targets = malloc(sizeof(struct mpssas_target) * maxtargets, M_MPT2, M_WAITOK|M_ZERO); if (!sassc->targets) { panic("%s failed to alloc targets with error %d\n", __func__, ENOMEM); } } Index: head/sys/dev/mps/mps_sas_lsi.c =================================================================== --- head/sys/dev/mps/mps_sas_lsi.c (revision 349848) +++ head/sys/dev/mps/mps_sas_lsi.c (revision 349849) @@ -1,1329 +1,1331 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2015 Avago Technologies * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD */ #include __FBSDID("$FreeBSD$"); /* Communications core for Avago Technologies (LSI) MPT2 */ /* TODO Move headers to mpsvar */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* For Hashed SAS Address creation for SATA Drives */ #define MPT2SAS_SN_LEN 20 #define MPT2SAS_MN_LEN 40 struct mps_fw_event_work { u16 event; void *event_data; TAILQ_ENTRY(mps_fw_event_work) ev_link; }; union _sata_sas_address { u8 wwid[8]; struct { u32 high; u32 low; } word; }; /* * define the IDENTIFY DEVICE structure */ struct _ata_identify_device_data { u16 reserved1[10]; /* 0-9 */ u16 serial_number[10]; /* 10-19 */ u16 reserved2[7]; /* 20-26 */ u16 model_number[20]; /* 27-46*/ u16 reserved3[170]; /* 47-216 */ u16 rotational_speed; /* 217 */ u16 reserved4[38]; /* 218-255 */ }; static u32 event_count; static void mpssas_fw_work(struct mps_softc *sc, struct mps_fw_event_work *fw_event); static void mpssas_fw_event_free(struct mps_softc *, struct mps_fw_event_work *); static int mpssas_add_device(struct mps_softc *sc, u16 handle, u8 linkrate); static int mpssas_get_sata_identify(struct mps_softc *sc, u16 handle, Mpi2SataPassthroughReply_t *mpi_reply, char *id_buffer, int sz, u32 devinfo); static void mpssas_ata_id_timeout(struct mps_softc *, struct mps_command *); int mpssas_get_sas_address_for_sata_disk(struct mps_softc *sc, u64 *sas_address, u16 handle, u32 device_info, u8 *is_SATA_SSD); static int mpssas_volume_add(struct mps_softc *sc, u16 handle); static void mpssas_SSU_to_SATA_devices(struct mps_softc *sc, int howto); static void mpssas_stop_unit_done(struct cam_periph *periph, union ccb *done_ccb); void mpssas_evt_handler(struct mps_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event) { struct mps_fw_event_work *fw_event; u16 sz; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); MPS_DPRINT_EVENT(sc, sas, event); mpssas_record_event(sc, event); fw_event = malloc(sizeof(struct mps_fw_event_work), M_MPT2, M_ZERO|M_NOWAIT); if (!fw_event) { printf("%s: allocate failed for fw_event\n", __func__); return; } sz = le16toh(event->EventDataLength) * 4; fw_event->event_data = malloc(sz, M_MPT2, M_ZERO|M_NOWAIT); if (!fw_event->event_data) { printf("%s: allocate failed for event_data\n", __func__); free(fw_event, M_MPT2); return; } bcopy(event->EventData, fw_event->event_data, sz); fw_event->event = event->Event; if ((event->Event == MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE || event->Event == MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST) && sc->track_mapping_events) sc->pending_map_events++; /* * When wait_for_port_enable flag is set, make sure that all the events * are processed. Increment the startup_refcount and decrement it after * events are processed. */ if ((event->Event == MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST || event->Event == MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST) && sc->wait_for_port_enable) mpssas_startup_increment(sc->sassc); TAILQ_INSERT_TAIL(&sc->sassc->ev_queue, fw_event, ev_link); taskqueue_enqueue(sc->sassc->ev_tq, &sc->sassc->ev_task); } static void mpssas_fw_event_free(struct mps_softc *sc, struct mps_fw_event_work *fw_event) { free(fw_event->event_data, M_MPT2); free(fw_event, M_MPT2); } /** * _mps_fw_work - delayed task for processing firmware events * @sc: per adapter object * @fw_event: The fw_event_work object * Context: user. * * Return nothing. */ static void mpssas_fw_work(struct mps_softc *sc, struct mps_fw_event_work *fw_event) { struct mpssas_softc *sassc; sassc = sc->sassc; mps_dprint(sc, MPS_EVENT, "(%d)->(%s) Working on Event: [%x]\n", event_count++,__func__,fw_event->event); switch (fw_event->event) { case MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST: { MPI2_EVENT_DATA_SAS_TOPOLOGY_CHANGE_LIST *data; MPI2_EVENT_SAS_TOPO_PHY_ENTRY *phy; int i; data = (MPI2_EVENT_DATA_SAS_TOPOLOGY_CHANGE_LIST *) fw_event->event_data; mps_mapping_topology_change_event(sc, fw_event->event_data); for (i = 0; i < data->NumEntries; i++) { phy = &data->PHY[i]; switch (phy->PhyStatus & MPI2_EVENT_SAS_TOPO_RC_MASK) { case MPI2_EVENT_SAS_TOPO_RC_TARG_ADDED: if (mpssas_add_device(sc, le16toh(phy->AttachedDevHandle), phy->LinkRate)){ mps_dprint(sc, MPS_ERROR, "%s: " "failed to add device with handle " "0x%x\n", __func__, le16toh(phy->AttachedDevHandle)); mpssas_prepare_remove(sassc, le16toh( phy->AttachedDevHandle)); } break; case MPI2_EVENT_SAS_TOPO_RC_TARG_NOT_RESPONDING: mpssas_prepare_remove(sassc,le16toh( phy->AttachedDevHandle)); break; case MPI2_EVENT_SAS_TOPO_RC_PHY_CHANGED: case MPI2_EVENT_SAS_TOPO_RC_NO_CHANGE: case MPI2_EVENT_SAS_TOPO_RC_DELAY_NOT_RESPONDING: default: break; } } /* * refcount was incremented for this event in * mpssas_evt_handler. Decrement it here because the event has * been processed. */ mpssas_startup_decrement(sassc); break; } case MPI2_EVENT_SAS_DISCOVERY: { MPI2_EVENT_DATA_SAS_DISCOVERY *data; data = (MPI2_EVENT_DATA_SAS_DISCOVERY *)fw_event->event_data; if (data->ReasonCode & MPI2_EVENT_SAS_DISC_RC_STARTED) mps_dprint(sc, MPS_TRACE,"SAS discovery start event\n"); if (data->ReasonCode & MPI2_EVENT_SAS_DISC_RC_COMPLETED) { mps_dprint(sc, MPS_TRACE,"SAS discovery stop event\n"); sassc->flags &= ~MPSSAS_IN_DISCOVERY; mpssas_discovery_end(sassc); } break; } case MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE: { Mpi2EventDataSasEnclDevStatusChange_t *data; data = (Mpi2EventDataSasEnclDevStatusChange_t *) fw_event->event_data; mps_mapping_enclosure_dev_status_change_event(sc, fw_event->event_data); break; } case MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST: { Mpi2EventIrConfigElement_t *element; int i; u8 foreign_config; Mpi2EventDataIrConfigChangeList_t *event_data; struct mpssas_target *targ; unsigned int id; event_data = fw_event->event_data; foreign_config = (le32toh(event_data->Flags) & MPI2_EVENT_IR_CHANGE_FLAGS_FOREIGN_CONFIG) ? 1 : 0; element = (Mpi2EventIrConfigElement_t *)&event_data->ConfigElement[0]; id = mps_mapping_get_raid_tid_from_handle(sc, element->VolDevHandle); mps_mapping_ir_config_change_event(sc, event_data); for (i = 0; i < event_data->NumElements; i++, element++) { switch (element->ReasonCode) { case MPI2_EVENT_IR_CHANGE_RC_VOLUME_CREATED: case MPI2_EVENT_IR_CHANGE_RC_ADDED: if (!foreign_config) { if (mpssas_volume_add(sc, le16toh(element->VolDevHandle))){ printf("%s: failed to add RAID " "volume with handle 0x%x\n", __func__, le16toh(element-> VolDevHandle)); } } break; case MPI2_EVENT_IR_CHANGE_RC_VOLUME_DELETED: case MPI2_EVENT_IR_CHANGE_RC_REMOVED: /* * Rescan after volume is deleted or removed. */ if (!foreign_config) { if (id == MPS_MAP_BAD_ID) { printf("%s: could not get ID " "for volume with handle " "0x%04x\n", __func__, le16toh(element->VolDevHandle)); break; } targ = &sassc->targets[id]; targ->handle = 0x0; targ->encl_slot = 0x0; targ->encl_handle = 0x0; targ->exp_dev_handle = 0x0; targ->phy_num = 0x0; targ->linkrate = 0x0; mpssas_rescan_target(sc, targ); printf("RAID target id 0x%x removed\n", targ->tid); } break; case MPI2_EVENT_IR_CHANGE_RC_PD_CREATED: case MPI2_EVENT_IR_CHANGE_RC_HIDE: /* * Phys Disk of a volume has been created. Hide * it from the OS. */ targ = mpssas_find_target_by_handle(sassc, 0, element->PhysDiskDevHandle); if (targ == NULL) break; /* * Set raid component flags only if it is not * WD. OR WrapDrive with * WD_HIDE_ALWAYS/WD_HIDE_IF_VOLUME is set in * NVRAM */ if((!sc->WD_available) || ((sc->WD_available && (sc->WD_hide_expose == MPS_WD_HIDE_ALWAYS)) || (sc->WD_valid_config && (sc->WD_hide_expose == MPS_WD_HIDE_IF_VOLUME)))) { targ->flags |= MPS_TARGET_FLAGS_RAID_COMPONENT; } mpssas_rescan_target(sc, targ); break; case MPI2_EVENT_IR_CHANGE_RC_PD_DELETED: /* * Phys Disk of a volume has been deleted. * Expose it to the OS. */ if (mpssas_add_device(sc, le16toh(element->PhysDiskDevHandle), 0)){ printf("%s: failed to add device with " "handle 0x%x\n", __func__, le16toh(element->PhysDiskDevHandle)); mpssas_prepare_remove(sassc, le16toh(element-> PhysDiskDevHandle)); } break; } } /* * refcount was incremented for this event in * mpssas_evt_handler. Decrement it here because the event has * been processed. */ mpssas_startup_decrement(sassc); break; } case MPI2_EVENT_IR_VOLUME: { Mpi2EventDataIrVolume_t *event_data = fw_event->event_data; /* * Informational only. */ mps_dprint(sc, MPS_EVENT, "Received IR Volume event:\n"); switch (event_data->ReasonCode) { case MPI2_EVENT_IR_VOLUME_RC_SETTINGS_CHANGED: mps_dprint(sc, MPS_EVENT, " Volume Settings " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); break; case MPI2_EVENT_IR_VOLUME_RC_STATUS_FLAGS_CHANGED: mps_dprint(sc, MPS_EVENT, " Volume Status " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); break; case MPI2_EVENT_IR_VOLUME_RC_STATE_CHANGED: mps_dprint(sc, MPS_EVENT, " Volume State " "changed from 0x%x to 0x%x for Volome with " "handle 0x%x", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), le16toh(event_data->VolDevHandle)); u32 state; struct mpssas_target *targ; state = le32toh(event_data->NewValue); switch (state) { case MPI2_RAID_VOL_STATE_MISSING: case MPI2_RAID_VOL_STATE_FAILED: mpssas_prepare_volume_remove(sassc, event_data-> VolDevHandle); break; case MPI2_RAID_VOL_STATE_ONLINE: case MPI2_RAID_VOL_STATE_DEGRADED: case MPI2_RAID_VOL_STATE_OPTIMAL: targ = mpssas_find_target_by_handle(sassc, 0, event_data->VolDevHandle); if (targ) { printf("%s %d: Volume handle 0x%x is already added \n", __func__, __LINE__ , event_data->VolDevHandle); break; } if (mpssas_volume_add(sc, le16toh(event_data->VolDevHandle))) { printf("%s: failed to add RAID " "volume with handle 0x%x\n", __func__, le16toh(event_data-> VolDevHandle)); } break; default: break; } break; default: break; } break; } case MPI2_EVENT_IR_PHYSICAL_DISK: { Mpi2EventDataIrPhysicalDisk_t *event_data = fw_event->event_data; struct mpssas_target *targ; /* * Informational only. */ mps_dprint(sc, MPS_EVENT, "Received IR Phys Disk event:\n"); switch (event_data->ReasonCode) { case MPI2_EVENT_IR_PHYSDISK_RC_SETTINGS_CHANGED: mps_dprint(sc, MPS_EVENT, " Phys Disk Settings " "changed from 0x%x to 0x%x for Phys Disk Number " "%d and handle 0x%x at Enclosure handle 0x%x, Slot " "%d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); break; case MPI2_EVENT_IR_PHYSDISK_RC_STATUS_FLAGS_CHANGED: mps_dprint(sc, MPS_EVENT, " Phys Disk Status changed " "from 0x%x to 0x%x for Phys Disk Number %d and " "handle 0x%x at Enclosure handle 0x%x, Slot %d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); break; case MPI2_EVENT_IR_PHYSDISK_RC_STATE_CHANGED: mps_dprint(sc, MPS_EVENT, " Phys Disk State changed " "from 0x%x to 0x%x for Phys Disk Number %d and " "handle 0x%x at Enclosure handle 0x%x, Slot %d", le32toh(event_data->PreviousValue), le32toh(event_data->NewValue), event_data->PhysDiskNum, le16toh(event_data->PhysDiskDevHandle), le16toh(event_data->EnclosureHandle), le16toh(event_data->Slot)); switch (event_data->NewValue) { case MPI2_RAID_PD_STATE_ONLINE: case MPI2_RAID_PD_STATE_DEGRADED: case MPI2_RAID_PD_STATE_REBUILDING: case MPI2_RAID_PD_STATE_OPTIMAL: case MPI2_RAID_PD_STATE_HOT_SPARE: targ = mpssas_find_target_by_handle(sassc, 0, event_data->PhysDiskDevHandle); if (targ) { if(!sc->WD_available) { targ->flags |= MPS_TARGET_FLAGS_RAID_COMPONENT; printf("%s %d: Found Target for handle 0x%x. \n", __func__, __LINE__ , event_data->PhysDiskDevHandle); } else if ((sc->WD_available && (sc->WD_hide_expose == MPS_WD_HIDE_ALWAYS)) || (sc->WD_valid_config && (sc->WD_hide_expose == MPS_WD_HIDE_IF_VOLUME))) { targ->flags |= MPS_TARGET_FLAGS_RAID_COMPONENT; printf("%s %d: WD: Found Target for handle 0x%x. \n", __func__, __LINE__ , event_data->PhysDiskDevHandle); } } break; case MPI2_RAID_PD_STATE_OFFLINE: case MPI2_RAID_PD_STATE_NOT_CONFIGURED: case MPI2_RAID_PD_STATE_NOT_COMPATIBLE: default: targ = mpssas_find_target_by_handle(sassc, 0, event_data->PhysDiskDevHandle); if (targ) { targ->flags |= ~MPS_TARGET_FLAGS_RAID_COMPONENT; printf("%s %d: Found Target for handle 0x%x. \n", __func__, __LINE__ , event_data->PhysDiskDevHandle); } break; } default: break; } break; } case MPI2_EVENT_IR_OPERATION_STATUS: { Mpi2EventDataIrOperationStatus_t *event_data = fw_event->event_data; /* * Informational only. */ mps_dprint(sc, MPS_EVENT, "Received IR Op Status event:\n"); mps_dprint(sc, MPS_EVENT, " RAID Operation of %d is %d " "percent complete for Volume with handle 0x%x", event_data->RAIDOperation, event_data->PercentComplete, le16toh(event_data->VolDevHandle)); break; } case MPI2_EVENT_LOG_ENTRY_ADDED: { pMpi2EventDataLogEntryAdded_t logEntry; uint16_t logQualifier; uint8_t logCode; logEntry = (pMpi2EventDataLogEntryAdded_t)fw_event->event_data; logQualifier = logEntry->LogEntryQualifier; if (logQualifier == MPI2_WD_LOG_ENTRY) { logCode = logEntry->LogData[0]; switch (logCode) { case MPI2_WD_SSD_THROTTLING: printf("WarpDrive Warning: IO Throttling has " "occurred in the WarpDrive subsystem. " "Check WarpDrive documentation for " "additional details\n"); break; case MPI2_WD_DRIVE_LIFE_WARN: printf("WarpDrive Warning: Program/Erase " "Cycles for the WarpDrive subsystem in " "degraded range. Check WarpDrive " "documentation for additional details\n"); break; case MPI2_WD_DRIVE_LIFE_DEAD: printf("WarpDrive Fatal Error: There are no " "Program/Erase Cycles for the WarpDrive " "subsystem. The storage device will be in " "read-only mode. Check WarpDrive " "documentation for additional details\n"); break; case MPI2_WD_RAIL_MON_FAIL: printf("WarpDrive Fatal Error: The Backup Rail " "Monitor has failed on the WarpDrive " "subsystem. Check WarpDrive documentation " "for additional details\n"); break; default: break; } } break; } case MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE: case MPI2_EVENT_SAS_BROADCAST_PRIMITIVE: default: mps_dprint(sc, MPS_TRACE,"Unhandled event 0x%0X\n", fw_event->event); break; } mps_dprint(sc, MPS_EVENT, "(%d)->(%s) Event Free: [%x]\n",event_count,__func__, fw_event->event); mpssas_fw_event_free(sc, fw_event); } void mpssas_firmware_event_work(void *arg, int pending) { struct mps_fw_event_work *fw_event; struct mps_softc *sc; sc = (struct mps_softc *)arg; mps_lock(sc); while ((fw_event = TAILQ_FIRST(&sc->sassc->ev_queue)) != NULL) { TAILQ_REMOVE(&sc->sassc->ev_queue, fw_event, ev_link); mpssas_fw_work(sc, fw_event); } mps_unlock(sc); } static int mpssas_add_device(struct mps_softc *sc, u16 handle, u8 linkrate){ char devstring[80]; struct mpssas_softc *sassc; struct mpssas_target *targ; Mpi2ConfigReply_t mpi_reply; Mpi2SasDevicePage0_t config_page; uint64_t sas_address; uint64_t parent_sas_address = 0; u32 device_info, parent_devinfo = 0; unsigned int id; int ret = 1, error = 0, i; struct mpssas_lun *lun; u8 is_SATA_SSD = 0; struct mps_command *cm; sassc = sc->sassc; mpssas_startup_increment(sassc); if (mps_config_get_sas_device_pg0(sc, &mpi_reply, &config_page, MPI2_SAS_DEVICE_PGAD_FORM_HANDLE, handle) != 0) { mps_dprint(sc, MPS_INFO|MPS_MAPPING|MPS_FAULT, "Error reading SAS device %#x page0, iocstatus= 0x%x\n", handle, mpi_reply.IOCStatus); error = ENXIO; goto out; } device_info = le32toh(config_page.DeviceInfo); if (((device_info & MPI2_SAS_DEVICE_INFO_SMP_TARGET) == 0) && (le16toh(config_page.ParentDevHandle) != 0)) { Mpi2ConfigReply_t tmp_mpi_reply; Mpi2SasDevicePage0_t parent_config_page; if (mps_config_get_sas_device_pg0(sc, &tmp_mpi_reply, &parent_config_page, MPI2_SAS_DEVICE_PGAD_FORM_HANDLE, le16toh(config_page.ParentDevHandle)) != 0) { mps_dprint(sc, MPS_MAPPING|MPS_FAULT, "Error reading parent SAS device %#x page0, " "iocstatus= 0x%x\n", le16toh(config_page.ParentDevHandle), tmp_mpi_reply.IOCStatus); } else { parent_sas_address = parent_config_page.SASAddress.High; parent_sas_address = (parent_sas_address << 32) | parent_config_page.SASAddress.Low; parent_devinfo = le32toh(parent_config_page.DeviceInfo); } } /* TODO Check proper endianness */ sas_address = config_page.SASAddress.High; sas_address = (sas_address << 32) | config_page.SASAddress.Low; mps_dprint(sc, MPS_MAPPING, "Handle 0x%04x SAS Address from SAS device " "page0 = %jx\n", handle, sas_address); /* * Always get SATA Identify information because this is used to * determine if Start/Stop Unit should be sent to the drive when the * system is shutdown. */ if (device_info & MPI2_SAS_DEVICE_INFO_SATA_DEVICE) { ret = mpssas_get_sas_address_for_sata_disk(sc, &sas_address, handle, device_info, &is_SATA_SSD); if (ret) { mps_dprint(sc, MPS_MAPPING|MPS_ERROR, "%s: failed to get disk type (SSD or HDD) for SATA " "device with handle 0x%04x\n", __func__, handle); } else { mps_dprint(sc, MPS_MAPPING, "Handle 0x%04x SAS Address " "from SATA device = %jx\n", handle, sas_address); } } /* * use_phynum: * 1 - use the PhyNum field as a fallback to the mapping logic * 0 - never use the PhyNum field * -1 - only use the PhyNum field * * Note that using the Phy number to map a device can cause device adds * to fail if multiple enclosures/expanders are in the topology. For * example, if two devices are in the same slot number in two different * enclosures within the topology, only one of those devices will be * added. PhyNum mapping should not be used if multiple enclosures are * in the topology. */ id = MPS_MAP_BAD_ID; if (sc->use_phynum != -1) id = mps_mapping_get_tid(sc, sas_address, handle); if (id == MPS_MAP_BAD_ID) { if ((sc->use_phynum == 0) || ((id = config_page.PhyNum) > sassc->maxtargets)) { mps_dprint(sc, MPS_INFO, "failure at %s:%d/%s()! " "Could not get ID for device with handle 0x%04x\n", __FILE__, __LINE__, __func__, handle); error = ENXIO; goto out; } } mps_dprint(sc, MPS_MAPPING, "%s: Target ID for added device is %d.\n", __func__, id); /* * Only do the ID check and reuse check if the target is not from a * RAID Component. For Physical Disks of a Volume, the ID will be reused * when a volume is deleted because the mapping entry for the PD will * still be in the mapping table. The ID check should not be done here * either since this PD is already being used. */ targ = &sassc->targets[id]; if (!(targ->flags & MPS_TARGET_FLAGS_RAID_COMPONENT)) { if (mpssas_check_id(sassc, id) != 0) { mps_dprint(sc, MPS_MAPPING|MPS_INFO, "Excluding target id %d\n", id); error = ENXIO; goto out; } if (targ->handle != 0x0) { mps_dprint(sc, MPS_MAPPING, "Attempting to reuse " "target id %d handle 0x%04x\n", id, targ->handle); error = ENXIO; goto out; } } targ->devinfo = device_info; targ->devname = le32toh(config_page.DeviceName.High); targ->devname = (targ->devname << 32) | le32toh(config_page.DeviceName.Low); targ->encl_handle = le16toh(config_page.EnclosureHandle); targ->encl_slot = le16toh(config_page.Slot); targ->handle = handle; targ->parent_handle = le16toh(config_page.ParentDevHandle); targ->sasaddr = mps_to_u64(&config_page.SASAddress); targ->parent_sasaddr = le64toh(parent_sas_address); targ->parent_devinfo = parent_devinfo; targ->tid = id; targ->linkrate = (linkrate>>4); targ->flags = 0; if (is_SATA_SSD) { targ->flags = MPS_TARGET_IS_SATA_SSD; } TAILQ_INIT(&targ->commands); TAILQ_INIT(&targ->timedout_commands); while(!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPT2); } SLIST_INIT(&targ->luns); mps_describe_devinfo(targ->devinfo, devstring, 80); mps_dprint(sc, MPS_MAPPING, "Found device <%s> <%s> <0x%04x> <%d/%d>\n", devstring, mps_describe_table(mps_linkrate_names, targ->linkrate), targ->handle, targ->encl_handle, targ->encl_slot); #if __FreeBSD_version < 1000039 if ((sassc->flags & MPSSAS_IN_STARTUP) == 0) #endif mpssas_rescan_target(sc, targ); mps_dprint(sc, MPS_MAPPING, "Target id 0x%x added\n", targ->tid); /* * Check all commands to see if the SATA_ID_TIMEOUT flag has been set. * If so, send a Target Reset TM to the target that was just created. * An Abort Task TM should be used instead of a Target Reset, but that * would be much more difficult because targets have not been fully * discovered yet, and LUN's haven't been setup. So, just reset the * target instead of the LUN. */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_flags & MPS_CM_FLAGS_SATA_ID_TIMEOUT) { targ->timeouts++; - cm->cm_state = MPS_CM_STATE_TIMEDOUT; + cm->cm_flags |= MPS_CM_FLAGS_TIMEDOUT; if ((targ->tm = mpssas_alloc_tm(sc)) != NULL) { mps_dprint(sc, MPS_INFO, "%s: sending Target " "Reset for stuck SATA identify command " "(cm = %p)\n", __func__, cm); targ->tm->cm_targ = targ; mpssas_send_reset(sc, targ->tm, MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET); } else { mps_dprint(sc, MPS_ERROR, "Failed to allocate " "tm for Target Reset after SATA ID command " "timed out (cm %p)\n", cm); } /* * No need to check for more since the target is * already being reset. */ break; } } out: /* * Free the commands that may not have been freed from the SATA ID call */ for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; if (cm->cm_flags & MPS_CM_FLAGS_SATA_ID_TIMEOUT) { free(cm->cm_data, M_MPT2); mps_free_command(sc, cm); } } mpssas_startup_decrement(sassc); return (error); } int mpssas_get_sas_address_for_sata_disk(struct mps_softc *sc, u64 *sas_address, u16 handle, u32 device_info, u8 *is_SATA_SSD) { Mpi2SataPassthroughReply_t mpi_reply; int i, rc, try_count; u32 *bufferptr; union _sata_sas_address hash_address; struct _ata_identify_device_data ata_identify; u8 buffer[MPT2SAS_MN_LEN + MPT2SAS_SN_LEN]; u32 ioc_status; u8 sas_status; memset(&ata_identify, 0, sizeof(ata_identify)); try_count = 0; do { rc = mpssas_get_sata_identify(sc, handle, &mpi_reply, (char *)&ata_identify, sizeof(ata_identify), device_info); try_count++; ioc_status = le16toh(mpi_reply.IOCStatus) & MPI2_IOCSTATUS_MASK; sas_status = mpi_reply.SASStatus; switch (ioc_status) { case MPI2_IOCSTATUS_SUCCESS: break; case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR: /* No sense sleeping. this error won't get better */ break; default: if (sc->spinup_wait_time > 0) { mps_dprint(sc, MPS_INFO, "Sleeping %d seconds " "after SATA ID error to wait for spinup\n", sc->spinup_wait_time); msleep(&sc->msleep_fake_chan, &sc->mps_mtx, 0, "mpsid", sc->spinup_wait_time * hz); } } } while (((rc && (rc != EWOULDBLOCK)) || (ioc_status && (ioc_status != MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR)) || sas_status) && (try_count < 5)); if (rc == 0 && !ioc_status && !sas_status) { mps_dprint(sc, MPS_MAPPING, "%s: got SATA identify " "successfully for handle = 0x%x with try_count = %d\n", __func__, handle, try_count); } else { mps_dprint(sc, MPS_MAPPING, "%s: handle = 0x%x failed\n", __func__, handle); return -1; } /* Copy & byteswap the 40 byte model number to a buffer */ for (i = 0; i < MPT2SAS_MN_LEN; i += 2) { buffer[i] = ((u8 *)ata_identify.model_number)[i + 1]; buffer[i + 1] = ((u8 *)ata_identify.model_number)[i]; } /* Copy & byteswap the 20 byte serial number to a buffer */ for (i = 0; i < MPT2SAS_SN_LEN; i += 2) { buffer[MPT2SAS_MN_LEN + i] = ((u8 *)ata_identify.serial_number)[i + 1]; buffer[MPT2SAS_MN_LEN + i + 1] = ((u8 *)ata_identify.serial_number)[i]; } bufferptr = (u32 *)buffer; /* There are 60 bytes to hash down to 8. 60 isn't divisible by 8, * so loop through the first 56 bytes (7*8), * and then add in the last dword. */ hash_address.word.low = 0; hash_address.word.high = 0; for (i = 0; (i < ((MPT2SAS_MN_LEN+MPT2SAS_SN_LEN)/8)); i++) { hash_address.word.low += *bufferptr; bufferptr++; hash_address.word.high += *bufferptr; bufferptr++; } /* Add the last dword */ hash_address.word.low += *bufferptr; /* Make sure the hash doesn't start with 5, because it could clash * with a SAS address. Change 5 to a D. */ if ((hash_address.word.high & 0x000000F0) == (0x00000050)) hash_address.word.high |= 0x00000080; *sas_address = (u64)hash_address.wwid[0] << 56 | (u64)hash_address.wwid[1] << 48 | (u64)hash_address.wwid[2] << 40 | (u64)hash_address.wwid[3] << 32 | (u64)hash_address.wwid[4] << 24 | (u64)hash_address.wwid[5] << 16 | (u64)hash_address.wwid[6] << 8 | (u64)hash_address.wwid[7]; if (ata_identify.rotational_speed == 1) { *is_SATA_SSD = 1; } return 0; } static int mpssas_get_sata_identify(struct mps_softc *sc, u16 handle, Mpi2SataPassthroughReply_t *mpi_reply, char *id_buffer, int sz, u32 devinfo) { Mpi2SataPassthroughRequest_t *mpi_request; Mpi2SataPassthroughReply_t *reply = NULL; struct mps_command *cm; char *buffer; int error = 0; buffer = malloc( sz, M_MPT2, M_NOWAIT | M_ZERO); if (!buffer) return ENOMEM; if ((cm = mps_alloc_command(sc)) == NULL) { free(buffer, M_MPT2); return (EBUSY); } mpi_request = (MPI2_SATA_PASSTHROUGH_REQUEST *)cm->cm_req; bzero(mpi_request,sizeof(MPI2_SATA_PASSTHROUGH_REQUEST)); mpi_request->Function = MPI2_FUNCTION_SATA_PASSTHROUGH; mpi_request->VF_ID = 0; mpi_request->DevHandle = htole16(handle); mpi_request->PassthroughFlags = (MPI2_SATA_PT_REQ_PT_FLAGS_PIO | MPI2_SATA_PT_REQ_PT_FLAGS_READ); mpi_request->DataLength = htole32(sz); mpi_request->CommandFIS[0] = 0x27; mpi_request->CommandFIS[1] = 0x80; mpi_request->CommandFIS[2] = (devinfo & MPI2_SAS_DEVICE_INFO_ATAPI_DEVICE) ? 0xA1 : 0xEC; cm->cm_sge = &mpi_request->SGL; cm->cm_sglsize = sizeof(MPI2_SGE_IO_UNION); cm->cm_flags = MPS_CM_FLAGS_SGE_SIMPLE | MPS_CM_FLAGS_DATAIN; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = buffer; cm->cm_length = htole32(sz); /* * Use a custom handler to avoid reinit'ing the controller on timeout. * This fixes a problem where the FW does not send a reply sometimes * when a bad disk is in the topology. So, this is used to timeout the * command so that processing can continue normally. */ cm->cm_timeout_handler = mpssas_ata_id_timeout; error = mps_wait_command(sc, &cm, MPS_ATA_ID_TIMEOUT, CAN_SLEEP); /* mpssas_ata_id_timeout does not reset controller */ KASSERT(cm != NULL, ("%s: surprise command freed", __func__)); reply = (Mpi2SataPassthroughReply_t *)cm->cm_reply; if (error || (reply == NULL)) { /* FIXME */ /* * If the request returns an error then we need to do a diag * reset */ mps_dprint(sc, MPS_INFO|MPS_FAULT|MPS_MAPPING, "Request for SATA PASSTHROUGH page completed with error %d\n", error); error = ENXIO; goto out; } bcopy(buffer, id_buffer, sz); bcopy(reply, mpi_reply, sizeof(Mpi2SataPassthroughReply_t)); if ((le16toh(reply->IOCStatus) & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mps_dprint(sc, MPS_INFO|MPS_MAPPING|MPS_FAULT, "Error reading device %#x SATA PASSTHRU; iocstatus= 0x%x\n", handle, reply->IOCStatus); error = ENXIO; goto out; } out: /* * If the SATA_ID_TIMEOUT flag has been set for this command, don't free * it. The command and buffer will be freed after sending an Abort * Task TM. */ if ((cm->cm_flags & MPS_CM_FLAGS_SATA_ID_TIMEOUT) == 0) { mps_free_command(sc, cm); free(buffer, M_MPT2); } return (error); } static void mpssas_ata_id_timeout(struct mps_softc *sc, struct mps_command *cm) { mps_dprint(sc, MPS_INFO, "%s ATA ID command timeout cm %p sc %p\n", __func__, cm, sc); /* * The Abort Task cannot be sent from here because the driver has not * completed setting up targets. Instead, the command is flagged so - * that special handling will be used to send the abort. + * that special handling will be used to send the abort. Now that + * this command has timed out, it's no longer in the queue. */ cm->cm_flags |= MPS_CM_FLAGS_SATA_ID_TIMEOUT; + cm->cm_state = MPS_CM_STATE_BUSY; } static int mpssas_volume_add(struct mps_softc *sc, u16 handle) { struct mpssas_softc *sassc; struct mpssas_target *targ; u64 wwid; unsigned int id; int error = 0; struct mpssas_lun *lun; sassc = sc->sassc; mpssas_startup_increment(sassc); /* wwid is endian safe */ mps_config_get_volume_wwid(sc, handle, &wwid); if (!wwid) { printf("%s: invalid WWID; cannot add volume to mapping table\n", __func__); error = ENXIO; goto out; } id = mps_mapping_get_raid_tid(sc, wwid, handle); if (id == MPS_MAP_BAD_ID) { printf("%s: could not get ID for volume with handle 0x%04x and " "WWID 0x%016llx\n", __func__, handle, (unsigned long long)wwid); error = ENXIO; goto out; } targ = &sassc->targets[id]; targ->tid = id; targ->handle = handle; targ->devname = wwid; TAILQ_INIT(&targ->commands); TAILQ_INIT(&targ->timedout_commands); while(!SLIST_EMPTY(&targ->luns)) { lun = SLIST_FIRST(&targ->luns); SLIST_REMOVE_HEAD(&targ->luns, lun_link); free(lun, M_MPT2); } SLIST_INIT(&targ->luns); #if __FreeBSD_version < 1000039 if ((sassc->flags & MPSSAS_IN_STARTUP) == 0) #endif mpssas_rescan_target(sc, targ); mps_dprint(sc, MPS_MAPPING, "RAID target id %d added (WWID = 0x%jx)\n", targ->tid, wwid); out: mpssas_startup_decrement(sassc); return (error); } /** * mpssas_SSU_to_SATA_devices * @sc: per adapter object * @howto: mast of RB_* bits for how we're rebooting * * Looks through the target list and issues a StartStopUnit SCSI command to each * SATA direct-access device. This helps to ensure that data corruption is * avoided when the system is being shut down. This must be called after the IR * System Shutdown RAID Action is sent if in IR mode. * * Return nothing. */ static void mpssas_SSU_to_SATA_devices(struct mps_softc *sc, int howto) { struct mpssas_softc *sassc = sc->sassc; union ccb *ccb; path_id_t pathid = cam_sim_path(sassc->sim); target_id_t targetid; struct mpssas_target *target; char path_str[64]; int timeout; /* * For each target, issue a StartStopUnit command to stop the device. */ sc->SSU_started = TRUE; sc->SSU_refcount = 0; for (targetid = 0; targetid < sc->max_devices; targetid++) { target = &sassc->targets[targetid]; if (target->handle == 0x0) { continue; } ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { mps_dprint(sc, MPS_FAULT, "Unable to alloc CCB to stop " "unit.\n"); return; } /* * The stop_at_shutdown flag will be set if this device is * a SATA direct-access end device. */ if (target->stop_at_shutdown) { if (xpt_create_path(&ccb->ccb_h.path, xpt_periph, pathid, targetid, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { mps_dprint(sc, MPS_FAULT, "Unable to create " "LUN path to stop unit.\n"); xpt_free_ccb(ccb); return; } xpt_path_string(ccb->ccb_h.path, path_str, sizeof(path_str)); mps_dprint(sc, MPS_INFO, "Sending StopUnit: path %s " "handle %d\n", path_str, target->handle); /* * Issue a START STOP UNIT command for the target. * Increment the SSU counter to be used to count the * number of required replies. */ mps_dprint(sc, MPS_INFO, "Incrementing SSU count\n"); sc->SSU_refcount++; ccb->ccb_h.target_id = xpt_path_target_id(ccb->ccb_h.path); ccb->ccb_h.ppriv_ptr1 = sassc; scsi_start_stop(&ccb->csio, /*retries*/0, mpssas_stop_unit_done, MSG_SIMPLE_Q_TAG, /*start*/FALSE, /*load/eject*/0, /*immediate*/FALSE, MPS_SENSE_LEN, /*timeout*/10000); xpt_action(ccb); } } /* * Timeout after 60 seconds by default or 10 seconds if howto has * RB_NOSYNC set which indicates we're likely handling a panic. */ timeout = 600; if (howto & RB_NOSYNC) timeout = 100; /* * Wait until all of the SSU commands have completed or timeout has * expired. Pause for 100ms each time through. If any command * times out, the target will be reset in the SCSI command timeout * routine. */ while (sc->SSU_refcount > 0) { pause("mpswait", hz/10); if (SCHEDULER_STOPPED()) xpt_sim_poll(sassc->sim); if (--timeout == 0) { mps_dprint(sc, MPS_FAULT, "Time has expired waiting " "for SSU commands to complete.\n"); break; } } } static void mpssas_stop_unit_done(struct cam_periph *periph, union ccb *done_ccb) { struct mpssas_softc *sassc; char path_str[64]; if (done_ccb == NULL) return; sassc = (struct mpssas_softc *)done_ccb->ccb_h.ppriv_ptr1; xpt_path_string(done_ccb->ccb_h.path, path_str, sizeof(path_str)); mps_dprint(sassc->sc, MPS_INFO, "Completing stop unit for %s\n", path_str); /* * Nothing more to do except free the CCB and path. If the command * timed out, an abort reset, then target reset will be issued during * the SCSI Command process. */ xpt_free_path(done_ccb->ccb_h.path); xpt_free_ccb(done_ccb); } /** * mpssas_ir_shutdown - IR shutdown notification * @sc: per adapter object * @howto: mast of RB_* bits for how we're rebooting * * Sending RAID Action to alert the Integrated RAID subsystem of the IOC that * the host system is shutting down. * * Return nothing. */ void mpssas_ir_shutdown(struct mps_softc *sc, int howto) { u16 volume_mapping_flags; u16 ioc_pg8_flags = le16toh(sc->ioc_pg8.Flags); struct dev_mapping_table *mt_entry; u32 start_idx, end_idx; unsigned int id, found_volume = 0; struct mps_command *cm; Mpi2RaidActionRequest_t *action; target_id_t targetid; struct mpssas_target *target; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); /* is IR firmware build loaded? */ if (!sc->ir_firmware) goto out; /* are there any volumes? Look at IR target IDs. */ // TODO-later, this should be looked up in the RAID config structure // when it is implemented. volume_mapping_flags = le16toh(sc->ioc_pg8.IRVolumeMappingFlags) & MPI2_IOCPAGE8_IRFLAGS_MASK_VOLUME_MAPPING_MODE; if (volume_mapping_flags == MPI2_IOCPAGE8_IRFLAGS_LOW_VOLUME_MAPPING) { start_idx = 0; if (ioc_pg8_flags & MPI2_IOCPAGE8_FLAGS_RESERVED_TARGETID_0) start_idx = 1; } else start_idx = sc->max_devices - sc->max_volumes; end_idx = start_idx + sc->max_volumes - 1; for (id = start_idx; id < end_idx; id++) { mt_entry = &sc->mapping_table[id]; if ((mt_entry->physical_id != 0) && (mt_entry->missing_count == 0)) { found_volume = 1; break; } } if (!found_volume) goto out; if ((cm = mps_alloc_command(sc)) == NULL) { printf("%s: command alloc failed\n", __func__); goto out; } action = (MPI2_RAID_ACTION_REQUEST *)cm->cm_req; action->Function = MPI2_FUNCTION_RAID_ACTION; action->Action = MPI2_RAID_ACTION_SYSTEM_SHUTDOWN_INITIATED; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; mps_lock(sc); mps_wait_command(sc, &cm, 5, CAN_SLEEP); mps_unlock(sc); /* * Don't check for reply, just leave. */ if (cm) mps_free_command(sc, cm); out: /* * All of the targets must have the correct value set for * 'stop_at_shutdown' for the current 'enable_ssu' sysctl variable. * * The possible values for the 'enable_ssu' variable are: * 0: disable to SSD and HDD * 1: disable only to HDD (default) * 2: disable only to SSD * 3: enable to SSD and HDD * anything else will default to 1. */ for (targetid = 0; targetid < sc->max_devices; targetid++) { target = &sc->sassc->targets[targetid]; if (target->handle == 0x0) { continue; } if (target->supports_SSU) { switch (sc->enable_ssu) { case MPS_SSU_DISABLE_SSD_DISABLE_HDD: target->stop_at_shutdown = FALSE; break; case MPS_SSU_DISABLE_SSD_ENABLE_HDD: target->stop_at_shutdown = TRUE; if (target->flags & MPS_TARGET_IS_SATA_SSD) { target->stop_at_shutdown = FALSE; } break; case MPS_SSU_ENABLE_SSD_ENABLE_HDD: target->stop_at_shutdown = TRUE; break; case MPS_SSU_ENABLE_SSD_DISABLE_HDD: default: target->stop_at_shutdown = TRUE; if ((target->flags & MPS_TARGET_IS_SATA_SSD) == 0) { target->stop_at_shutdown = FALSE; } break; } } } mpssas_SSU_to_SATA_devices(sc, howto); } Index: head/sys/dev/mps/mpsvar.h =================================================================== --- head/sys/dev/mps/mpsvar.h (revision 349848) +++ head/sys/dev/mps/mpsvar.h (revision 349849) @@ -1,862 +1,865 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2011-2015 LSI Corp. * Copyright (c) 2013-2015 Avago Technologies * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Avago Technologies (LSI) MPT-Fusion Host Adapter FreeBSD * * $FreeBSD$ */ #ifndef _MPSVAR_H #define _MPSVAR_H #include #include #define MPS_DRIVER_VERSION "21.02.00.00-fbsd" #define MPS_DB_MAX_WAIT 2500 #define MPS_REQ_FRAMES 2048 #define MPS_PRI_REQ_FRAMES 128 #define MPS_EVT_REPLY_FRAMES 32 #define MPS_REPLY_FRAMES MPS_REQ_FRAMES #define MPS_CHAIN_FRAMES 16384 #define MPS_MAXIO_PAGES (-1) #define MPS_SENSE_LEN SSD_FULL_SIZE #define MPS_MSI_MAX 1 #define MPS_MSIX_MAX 16 #define MPS_SGE64_SIZE 12 #define MPS_SGE32_SIZE 8 #define MPS_SGC_SIZE 8 #define CAN_SLEEP 1 #define NO_SLEEP 0 #define MPS_PERIODIC_DELAY 1 /* 1 second heartbeat/watchdog check */ #define MPS_ATA_ID_TIMEOUT 5 /* 5 second timeout for SATA ID cmd */ #define MPS_MISSING_CHECK_DELAY 10 /* 10 seconds between missing check */ #define MPS_SCSI_RI_INVALID_FRAME (0x00000002) #define DEFAULT_SPINUP_WAIT 3 /* seconds to wait for spinup */ #include /* * host mapping related macro definitions */ #define MPS_MAPTABLE_BAD_IDX 0xFFFFFFFF #define MPS_DPM_BAD_IDX 0xFFFF #define MPS_ENCTABLE_BAD_IDX 0xFF #define MPS_MAX_MISSING_COUNT 0x0F #define MPS_DEV_RESERVED 0x20000000 #define MPS_MAP_IN_USE 0x10000000 #define MPS_MAP_BAD_ID 0xFFFFFFFF /* * WarpDrive controller */ #define MPS_CHIP_WD_DEVICE_ID 0x007E #define MPS_WD_LSI_OEM 0x80 #define MPS_WD_HIDE_EXPOSE_MASK 0x03 #define MPS_WD_HIDE_ALWAYS 0x00 #define MPS_WD_EXPOSE_ALWAYS 0x01 #define MPS_WD_HIDE_IF_VOLUME 0x02 #define MPS_WD_RETRY 0x01 #define MPS_MAN_PAGE10_SIZE 0x5C /* Hardcode for now */ #define MPS_MAX_DISKS_IN_VOL 10 /* * WarpDrive Event Logging */ #define MPI2_WD_LOG_ENTRY 0x8002 #define MPI2_WD_SSD_THROTTLING 0x0041 #define MPI2_WD_DRIVE_LIFE_WARN 0x0043 #define MPI2_WD_DRIVE_LIFE_DEAD 0x0044 #define MPI2_WD_RAIL_MON_FAIL 0x004D typedef uint8_t u8; typedef uint16_t u16; typedef uint32_t u32; typedef uint64_t u64; /** * struct dev_mapping_table - device mapping information * @physical_id: SAS address for drives or WWID for RAID volumes * @device_info: bitfield provides detailed info about the device * @phy_bits: bitfields indicating controller phys * @dpm_entry_num: index of this device in device persistent map table * @dev_handle: device handle for the device pointed by this entry * @id: target id * @missing_count: number of times the device not detected by driver * @hide_flag: Hide this physical disk/not (foreign configuration) * @init_complete: Whether the start of the day checks completed or not */ struct dev_mapping_table { u64 physical_id; u32 device_info; u32 phy_bits; u16 dpm_entry_num; u16 dev_handle; u16 reserved1; u16 id; u8 missing_count; u8 init_complete; u8 TLR_bits; u8 reserved2; }; /** * struct enc_mapping_table - mapping information about an enclosure * @enclosure_id: Logical ID of this enclosure * @start_index: index to the entry in dev_mapping_table * @phy_bits: bitfields indicating controller phys * @dpm_entry_num: index of this enclosure in device persistent map table * @enc_handle: device handle for the enclosure pointed by this entry * @num_slots: number of slots in the enclosure * @start_slot: Starting slot id * @missing_count: number of times the device not detected by driver * @removal_flag: used to mark the device for removal * @skip_search: used as a flag to include/exclude enclosure for search * @init_complete: Whether the start of the day checks completed or not */ struct enc_mapping_table { u64 enclosure_id; u32 start_index; u32 phy_bits; u16 dpm_entry_num; u16 enc_handle; u16 num_slots; u16 start_slot; u8 missing_count; u8 removal_flag; u8 skip_search; u8 init_complete; }; /** * struct map_removal_table - entries to be removed from mapping table * @dpm_entry_num: index of this device in device persistent map table * @dev_handle: device handle for the device pointed by this entry */ struct map_removal_table{ u16 dpm_entry_num; u16 dev_handle; }; typedef struct mps_fw_diagnostic_buffer { size_t size; uint8_t extended_type; uint8_t buffer_type; uint8_t force_release; uint32_t product_specific[23]; uint8_t immediate; uint8_t enabled; uint8_t valid_data; uint8_t owned_by_firmware; uint32_t unique_id; } mps_fw_diagnostic_buffer_t; struct mps_softc; struct mps_command; struct mpssas_softc; union ccb; struct mpssas_target; struct mps_column_map; MALLOC_DECLARE(M_MPT2); typedef void mps_evt_callback_t(struct mps_softc *, uintptr_t, MPI2_EVENT_NOTIFICATION_REPLY *reply); typedef void mps_command_callback_t(struct mps_softc *, struct mps_command *cm); struct mps_chain { TAILQ_ENTRY(mps_chain) chain_link; MPI2_SGE_IO_UNION *chain; uint32_t chain_busaddr; }; /* * This needs to be at least 2 to support SMP passthrough. */ #define MPS_IOVEC_COUNT 2 struct mps_command { TAILQ_ENTRY(mps_command) cm_link; TAILQ_ENTRY(mps_command) cm_recovery; struct mps_softc *cm_sc; union ccb *cm_ccb; void *cm_data; u_int cm_length; u_int cm_out_len; struct uio cm_uio; struct iovec cm_iovec[MPS_IOVEC_COUNT]; u_int cm_max_segs; u_int cm_sglsize; MPI2_SGE_IO_UNION *cm_sge; uint8_t *cm_req; uint8_t *cm_reply; uint32_t cm_reply_data; mps_command_callback_t *cm_complete; void *cm_complete_data; struct mpssas_target *cm_targ; MPI2_REQUEST_DESCRIPTOR_UNION cm_desc; u_int cm_lun; u_int cm_flags; #define MPS_CM_FLAGS_POLLED (1 << 0) #define MPS_CM_FLAGS_COMPLETE (1 << 1) #define MPS_CM_FLAGS_SGE_SIMPLE (1 << 2) #define MPS_CM_FLAGS_DATAOUT (1 << 3) #define MPS_CM_FLAGS_DATAIN (1 << 4) #define MPS_CM_FLAGS_WAKEUP (1 << 5) #define MPS_CM_FLAGS_DD_IO (1 << 6) #define MPS_CM_FLAGS_USE_UIO (1 << 7) #define MPS_CM_FLAGS_SMP_PASS (1 << 8) #define MPS_CM_FLAGS_CHAIN_FAILED (1 << 9) #define MPS_CM_FLAGS_ERROR_MASK MPS_CM_FLAGS_CHAIN_FAILED #define MPS_CM_FLAGS_USE_CCB (1 << 10) #define MPS_CM_FLAGS_SATA_ID_TIMEOUT (1 << 11) +#define MPS_CM_FLAGS_ON_RECOVERY (1 << 12) +#define MPS_CM_FLAGS_TIMEDOUT (1 << 13) u_int cm_state; #define MPS_CM_STATE_FREE 0 #define MPS_CM_STATE_BUSY 1 -#define MPS_CM_STATE_TIMEDOUT 2 -#define MPS_CM_STATE_INQUEUE 3 +#define MPS_CM_STATE_INQUEUE 2 bus_dmamap_t cm_dmamap; struct scsi_sense_data *cm_sense; TAILQ_HEAD(, mps_chain) cm_chain_list; uint32_t cm_req_busaddr; uint32_t cm_sense_busaddr; struct callout cm_callout; mps_command_callback_t *cm_timeout_handler; }; struct mps_column_map { uint16_t dev_handle; uint8_t phys_disk_num; }; struct mps_event_handle { TAILQ_ENTRY(mps_event_handle) eh_list; mps_evt_callback_t *callback; void *data; u32 mask[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS]; }; struct mps_busdma_context { int completed; int abandoned; int error; bus_addr_t *addr; struct mps_softc *softc; bus_dmamap_t buffer_dmamap; bus_dma_tag_t buffer_dmat; }; struct mps_queue { struct mps_softc *sc; int qnum; MPI2_REPLY_DESCRIPTORS_UNION *post_queue; int replypostindex; #ifdef notyet ck_ring_buffer_t *ringmem; ck_ring_buffer_t *chainmem; ck_ring_t req_ring; ck_ring_t chain_ring; #endif bus_dma_tag_t buffer_dmat; int io_cmds_highwater; int chain_free_lowwater; int chain_alloc_fail; struct resource *irq; void *intrhand; int irq_rid; }; struct mps_softc { device_t mps_dev; struct cdev *mps_cdev; u_int mps_flags; #define MPS_FLAGS_INTX (1 << 0) #define MPS_FLAGS_MSI (1 << 1) #define MPS_FLAGS_BUSY (1 << 2) #define MPS_FLAGS_SHUTDOWN (1 << 3) #define MPS_FLAGS_DIAGRESET (1 << 4) #define MPS_FLAGS_ATTACH_DONE (1 << 5) #define MPS_FLAGS_WD_AVAILABLE (1 << 6) #define MPS_FLAGS_REALLOCATED (1 << 7) u_int mps_debug; u_int msi_msgs; u_int reqframesz; u_int replyframesz; int tm_cmds_active; int io_cmds_active; int io_cmds_highwater; int chain_free; int max_chains; int max_io_pages; u_int maxio; int chain_free_lowwater; u_int enable_ssu; int spinup_wait_time; int use_phynum; uint64_t chain_alloc_fail; struct sysctl_ctx_list sysctl_ctx; struct sysctl_oid *sysctl_tree; char fw_version[16]; struct mps_command *commands; struct mps_chain *chains; struct callout periodic; struct callout device_check_callout; struct mps_queue *queues; struct mpssas_softc *sassc; TAILQ_HEAD(, mps_command) req_list; TAILQ_HEAD(, mps_command) high_priority_req_list; TAILQ_HEAD(, mps_chain) chain_list; TAILQ_HEAD(, mps_command) tm_list; int replypostindex; int replyfreeindex; struct resource *mps_regs_resource; bus_space_handle_t mps_bhandle; bus_space_tag_t mps_btag; int mps_regs_rid; bus_dma_tag_t mps_parent_dmat; bus_dma_tag_t buffer_dmat; MPI2_IOC_FACTS_REPLY *facts; int num_reqs; int num_prireqs; int num_replies; int num_chains; int fqdepth; /* Free queue */ int pqdepth; /* Post queue */ u32 event_mask[MPI2_EVENT_NOTIFY_EVENTMASK_WORDS]; TAILQ_HEAD(, mps_event_handle) event_list; struct mps_event_handle *mps_log_eh; struct mtx mps_mtx; struct intr_config_hook mps_ich; uint8_t *req_frames; bus_addr_t req_busaddr; bus_dma_tag_t req_dmat; bus_dmamap_t req_map; uint8_t *reply_frames; bus_addr_t reply_busaddr; bus_dma_tag_t reply_dmat; bus_dmamap_t reply_map; struct scsi_sense_data *sense_frames; bus_addr_t sense_busaddr; bus_dma_tag_t sense_dmat; bus_dmamap_t sense_map; uint8_t *chain_frames; bus_dma_tag_t chain_dmat; bus_dmamap_t chain_map; MPI2_REPLY_DESCRIPTORS_UNION *post_queue; bus_addr_t post_busaddr; uint32_t *free_queue; bus_addr_t free_busaddr; bus_dma_tag_t queues_dmat; bus_dmamap_t queues_map; uint8_t *fw_diag_buffer; bus_addr_t fw_diag_busaddr; bus_dma_tag_t fw_diag_dmat; bus_dmamap_t fw_diag_map; uint8_t ir_firmware; /* static config pages */ Mpi2IOCPage8_t ioc_pg8; /* host mapping support */ struct dev_mapping_table *mapping_table; struct enc_mapping_table *enclosure_table; struct map_removal_table *removal_table; uint8_t *dpm_entry_used; uint8_t *dpm_flush_entry; Mpi2DriverMappingPage0_t *dpm_pg0; uint16_t max_devices; uint16_t max_enclosures; uint16_t max_expanders; uint8_t max_volumes; uint8_t num_enc_table_entries; uint8_t num_rsvd_entries; uint16_t max_dpm_entries; uint8_t is_dpm_enable; uint8_t track_mapping_events; uint32_t pending_map_events; /* FW diag Buffer List */ mps_fw_diagnostic_buffer_t fw_diag_buffer_list[MPI2_DIAG_BUF_TYPE_COUNT]; /* Event Recording IOCTL support */ uint32_t events_to_record[4]; mps_event_entry_t recorded_events[MPS_EVENT_QUEUE_SIZE]; uint8_t event_index; uint32_t event_number; /* EEDP and TLR support */ uint8_t eedp_enabled; uint8_t control_TLR; /* Shutdown Event Handler */ eventhandler_tag shutdown_eh; /* To track topo events during reset */ #define MPS_DIAG_RESET_TIMEOUT 300000 uint8_t wait_for_port_enable; uint8_t port_enable_complete; uint8_t msleep_fake_chan; /* WD controller */ uint8_t WD_available; uint8_t WD_valid_config; uint8_t WD_hide_expose; /* Direct Drive for WarpDrive */ uint8_t DD_num_phys_disks; uint16_t DD_dev_handle; uint32_t DD_stripe_size; uint32_t DD_stripe_exponent; uint32_t DD_block_size; uint16_t DD_block_exponent; uint64_t DD_max_lba; struct mps_column_map DD_column_map[MPS_MAX_DISKS_IN_VOL]; /* StartStopUnit command handling at shutdown */ uint32_t SSU_refcount; uint8_t SSU_started; /* Configuration tunables */ u_int disable_msix; u_int disable_msi; u_int max_msix; u_int max_reqframes; u_int max_prireqframes; u_int max_replyframes; u_int max_evtframes; char exclude_ids[80]; struct timeval lastfail; }; struct mps_config_params { MPI2_CONFIG_EXT_PAGE_HEADER_UNION hdr; u_int action; u_int page_address; /* Attributes, not a phys address */ u_int status; void *buffer; u_int length; int timeout; void (*callback)(struct mps_softc *, struct mps_config_params *); void *cbdata; }; struct scsi_read_capacity_eedp { uint8_t addr[8]; uint8_t length[4]; uint8_t protect; }; static __inline uint32_t mps_regread(struct mps_softc *sc, uint32_t offset) { return (bus_space_read_4(sc->mps_btag, sc->mps_bhandle, offset)); } static __inline void mps_regwrite(struct mps_softc *sc, uint32_t offset, uint32_t val) { bus_space_write_4(sc->mps_btag, sc->mps_bhandle, offset, val); } /* free_queue must have Little Endian address * TODO- cm_reply_data is unwanted. We can remove it. * */ static __inline void mps_free_reply(struct mps_softc *sc, uint32_t busaddr) { if (++sc->replyfreeindex >= sc->fqdepth) sc->replyfreeindex = 0; sc->free_queue[sc->replyfreeindex] = htole32(busaddr); mps_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); } static __inline struct mps_chain * mps_alloc_chain(struct mps_softc *sc) { struct mps_chain *chain; if ((chain = TAILQ_FIRST(&sc->chain_list)) != NULL) { TAILQ_REMOVE(&sc->chain_list, chain, chain_link); sc->chain_free--; if (sc->chain_free < sc->chain_free_lowwater) sc->chain_free_lowwater = sc->chain_free; } else sc->chain_alloc_fail++; return (chain); } static __inline void mps_free_chain(struct mps_softc *sc, struct mps_chain *chain) { sc->chain_free++; TAILQ_INSERT_TAIL(&sc->chain_list, chain, chain_link); } static __inline void mps_free_command(struct mps_softc *sc, struct mps_command *cm) { struct mps_chain *chain, *chain_temp; - KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, ("state not busy\n")); + KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, + ("state not busy: %d\n", cm->cm_state)); if (cm->cm_reply != NULL) mps_free_reply(sc, cm->cm_reply_data); cm->cm_reply = NULL; cm->cm_flags = 0; cm->cm_complete = NULL; cm->cm_complete_data = NULL; cm->cm_ccb = NULL; cm->cm_targ = NULL; cm->cm_max_segs = 0; cm->cm_lun = 0; cm->cm_state = MPS_CM_STATE_FREE; cm->cm_data = NULL; cm->cm_length = 0; cm->cm_out_len = 0; cm->cm_sglsize = 0; cm->cm_sge = NULL; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain_temp) { TAILQ_REMOVE(&cm->cm_chain_list, chain, chain_link); mps_free_chain(sc, chain); } TAILQ_INSERT_TAIL(&sc->req_list, cm, cm_link); } static __inline struct mps_command * mps_alloc_command(struct mps_softc *sc) { struct mps_command *cm; cm = TAILQ_FIRST(&sc->req_list); if (cm == NULL) return (NULL); KASSERT(cm->cm_state == MPS_CM_STATE_FREE, - ("mps: Allocating busy command\n")); + ("mps: Allocating busy command: %d\n", cm->cm_state)); TAILQ_REMOVE(&sc->req_list, cm, cm_link); cm->cm_state = MPS_CM_STATE_BUSY; cm->cm_timeout_handler = NULL; return (cm); } static __inline void mps_free_high_priority_command(struct mps_softc *sc, struct mps_command *cm) { struct mps_chain *chain, *chain_temp; - KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, ("state not busy\n")); + KASSERT(cm->cm_state == MPS_CM_STATE_BUSY, + ("state not busy: %d\n", cm->cm_state)); if (cm->cm_reply != NULL) mps_free_reply(sc, cm->cm_reply_data); cm->cm_reply = NULL; cm->cm_flags = 0; cm->cm_complete = NULL; cm->cm_complete_data = NULL; cm->cm_ccb = NULL; cm->cm_targ = NULL; cm->cm_lun = 0; cm->cm_state = MPS_CM_STATE_FREE; TAILQ_FOREACH_SAFE(chain, &cm->cm_chain_list, chain_link, chain_temp) { TAILQ_REMOVE(&cm->cm_chain_list, chain, chain_link); mps_free_chain(sc, chain); } TAILQ_INSERT_TAIL(&sc->high_priority_req_list, cm, cm_link); } static __inline struct mps_command * mps_alloc_high_priority_command(struct mps_softc *sc) { struct mps_command *cm; cm = TAILQ_FIRST(&sc->high_priority_req_list); if (cm == NULL) return (NULL); KASSERT(cm->cm_state == MPS_CM_STATE_FREE, - ("mps: Allocating busy command\n")); + ("mps: Allocating high priority busy command: %d\n", cm->cm_state)); TAILQ_REMOVE(&sc->high_priority_req_list, cm, cm_link); cm->cm_state = MPS_CM_STATE_BUSY; cm->cm_timeout_handler = NULL; cm->cm_desc.HighPriority.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_HIGH_PRIORITY; return (cm); } static __inline void mps_lock(struct mps_softc *sc) { mtx_lock(&sc->mps_mtx); } static __inline void mps_unlock(struct mps_softc *sc) { mtx_unlock(&sc->mps_mtx); } #define MPS_INFO (1 << 0) /* Basic info */ #define MPS_FAULT (1 << 1) /* Hardware faults */ #define MPS_EVENT (1 << 2) /* Event data from the controller */ #define MPS_LOG (1 << 3) /* Log data from the controller */ #define MPS_RECOVERY (1 << 4) /* Command error recovery tracing */ #define MPS_ERROR (1 << 5) /* Parameter errors, programming bugs */ #define MPS_INIT (1 << 6) /* Things related to system init */ #define MPS_XINFO (1 << 7) /* More detailed/noisy info */ #define MPS_USER (1 << 8) /* Trace user-generated commands */ #define MPS_MAPPING (1 << 9) /* Trace device mappings */ #define MPS_TRACE (1 << 10) /* Function-by-function trace */ #define MPS_SSU_DISABLE_SSD_DISABLE_HDD 0 #define MPS_SSU_ENABLE_SSD_DISABLE_HDD 1 #define MPS_SSU_DISABLE_SSD_ENABLE_HDD 2 #define MPS_SSU_ENABLE_SSD_ENABLE_HDD 3 #define mps_printf(sc, args...) \ device_printf((sc)->mps_dev, ##args) #define mps_print_field(sc, msg, args...) \ printf("\t" msg, ##args) #define mps_vprintf(sc, args...) \ do { \ if (bootverbose) \ mps_printf(sc, ##args); \ } while (0) #define mps_dprint(sc, level, msg, args...) \ do { \ if ((sc)->mps_debug & (level)) \ device_printf((sc)->mps_dev, msg, ##args); \ } while (0) #define MPS_PRINTFIELD_START(sc, tag...) \ mps_printf((sc), ##tag); \ mps_print_field((sc), ":\n") #define MPS_PRINTFIELD_END(sc, tag) \ mps_printf((sc), tag "\n") #define MPS_PRINTFIELD(sc, facts, attr, fmt) \ mps_print_field((sc), #attr ": " #fmt "\n", (facts)->attr) #define MPS_FUNCTRACE(sc) \ mps_dprint((sc), MPS_TRACE, "%s\n", __func__) #define CAN_SLEEP 1 #define NO_SLEEP 0 static __inline void mps_from_u64(uint64_t data, U64 *mps) { (mps)->High = htole32((uint32_t)((data) >> 32)); (mps)->Low = htole32((uint32_t)((data) & 0xffffffff)); } static __inline uint64_t mps_to_u64(U64 *data) { return (((uint64_t)le32toh(data->High) << 32) | le32toh(data->Low)); } static __inline void mps_mask_intr(struct mps_softc *sc) { uint32_t mask; mask = mps_regread(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET); mask |= MPI2_HIM_REPLY_INT_MASK; mps_regwrite(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET, mask); } static __inline void mps_unmask_intr(struct mps_softc *sc) { uint32_t mask; mask = mps_regread(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET); mask &= ~MPI2_HIM_REPLY_INT_MASK; mps_regwrite(sc, MPI2_HOST_INTERRUPT_MASK_OFFSET, mask); } int mps_pci_setup_interrupts(struct mps_softc *sc); void mps_pci_free_interrupts(struct mps_softc *sc); int mps_pci_restore(struct mps_softc *sc); void mps_get_tunables(struct mps_softc *sc); int mps_attach(struct mps_softc *sc); int mps_free(struct mps_softc *sc); void mps_intr(void *); void mps_intr_msi(void *); void mps_intr_locked(void *); int mps_register_events(struct mps_softc *, u32 *, mps_evt_callback_t *, void *, struct mps_event_handle **); int mps_restart(struct mps_softc *); int mps_update_events(struct mps_softc *, struct mps_event_handle *, u32 *); void mps_deregister_events(struct mps_softc *, struct mps_event_handle *); int mps_push_sge(struct mps_command *, void *, size_t, int); int mps_add_dmaseg(struct mps_command *, vm_paddr_t, size_t, u_int, int); int mps_attach_sas(struct mps_softc *sc); int mps_detach_sas(struct mps_softc *sc); int mps_read_config_page(struct mps_softc *, struct mps_config_params *); int mps_write_config_page(struct mps_softc *, struct mps_config_params *); void mps_memaddr_cb(void *, bus_dma_segment_t *, int , int ); void mps_memaddr_wait_cb(void *, bus_dma_segment_t *, int , int ); void mpi_init_sge(struct mps_command *cm, void *req, void *sge); int mps_attach_user(struct mps_softc *); void mps_detach_user(struct mps_softc *); void mpssas_record_event(struct mps_softc *sc, MPI2_EVENT_NOTIFICATION_REPLY *event_reply); int mps_map_command(struct mps_softc *sc, struct mps_command *cm); int mps_wait_command(struct mps_softc *sc, struct mps_command **cm, int timeout, int sleep_flag); int mps_config_get_bios_pg3(struct mps_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2BiosPage3_t *config_page); int mps_config_get_raid_volume_pg0(struct mps_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidVolPage0_t *config_page, u32 page_address); int mps_config_get_ioc_pg8(struct mps_softc *sc, Mpi2ConfigReply_t *, Mpi2IOCPage8_t *); int mps_config_get_man_pg10(struct mps_softc *sc, Mpi2ConfigReply_t *mpi_reply); int mps_config_get_sas_device_pg0(struct mps_softc *, Mpi2ConfigReply_t *, Mpi2SasDevicePage0_t *, u32 , u16 ); int mps_config_get_dpm_pg0(struct mps_softc *, Mpi2ConfigReply_t *, Mpi2DriverMappingPage0_t *, u16 ); int mps_config_get_raid_volume_pg1(struct mps_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidVolPage1_t *config_page, u32 form, u16 handle); int mps_config_get_volume_wwid(struct mps_softc *sc, u16 volume_handle, u64 *wwid); int mps_config_get_raid_pd_pg0(struct mps_softc *sc, Mpi2ConfigReply_t *mpi_reply, Mpi2RaidPhysDiskPage0_t *config_page, u32 page_address); void mpssas_ir_shutdown(struct mps_softc *sc, int howto); int mps_reinit(struct mps_softc *sc); void mpssas_handle_reinit(struct mps_softc *sc); void mps_base_static_config_pages(struct mps_softc *sc); void mps_wd_config_pages(struct mps_softc *sc); int mps_mapping_initialize(struct mps_softc *); void mps_mapping_topology_change_event(struct mps_softc *, Mpi2EventDataSasTopologyChangeList_t *); void mps_mapping_free_memory(struct mps_softc *sc); int mps_config_set_dpm_pg0(struct mps_softc *, Mpi2ConfigReply_t *, Mpi2DriverMappingPage0_t *, u16 ); void mps_mapping_exit(struct mps_softc *); void mps_mapping_check_devices(void *); int mps_mapping_allocate_memory(struct mps_softc *sc); unsigned int mps_mapping_get_tid(struct mps_softc *, uint64_t , u16); unsigned int mps_mapping_get_tid_from_handle(struct mps_softc *sc, u16 handle); unsigned int mps_mapping_get_raid_tid(struct mps_softc *sc, u64 wwid, u16 volHandle); unsigned int mps_mapping_get_raid_tid_from_handle(struct mps_softc *sc, u16 volHandle); void mps_mapping_enclosure_dev_status_change_event(struct mps_softc *, Mpi2EventDataSasEnclDevStatusChange_t *event_data); void mps_mapping_ir_config_change_event(struct mps_softc *sc, Mpi2EventDataIrConfigChangeList_t *event_data); int mps_mapping_dump(SYSCTL_HANDLER_ARGS); int mps_mapping_encl_dump(SYSCTL_HANDLER_ARGS); void mpssas_evt_handler(struct mps_softc *sc, uintptr_t data, MPI2_EVENT_NOTIFICATION_REPLY *event); void mpssas_prepare_remove(struct mpssas_softc *sassc, uint16_t handle); void mpssas_prepare_volume_remove(struct mpssas_softc *sassc, uint16_t handle); int mpssas_startup(struct mps_softc *sc); struct mpssas_target * mpssas_find_target_by_handle(struct mpssas_softc *, int, uint16_t); void mpssas_realloc_targets(struct mps_softc *sc, int maxtargets); struct mps_command * mpssas_alloc_tm(struct mps_softc *sc); void mpssas_free_tm(struct mps_softc *sc, struct mps_command *tm); void mpssas_release_simq_reinit(struct mpssas_softc *sassc); int mpssas_send_reset(struct mps_softc *sc, struct mps_command *tm, uint8_t type); SYSCTL_DECL(_hw_mps); /* Compatibility shims for different OS versions */ #if __FreeBSD_version >= 800001 #define mps_kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) \ kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) #define mps_kproc_exit(arg) kproc_exit(arg) #else #define mps_kproc_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) \ kthread_create(func, farg, proc_ptr, flags, stackpgs, fmtstr, arg) #define mps_kproc_exit(arg) kthread_exit(arg) #endif #if defined(CAM_PRIORITY_XPT) #define MPS_PRIORITY_XPT CAM_PRIORITY_XPT #else #define MPS_PRIORITY_XPT 5 #endif #if __FreeBSD_version < 800107 // Prior to FreeBSD-8.0 scp3_flags was not defined. #define spc3_flags reserved #define SPC3_SID_PROTECT 0x01 #define SPC3_SID_3PC 0x08 #define SPC3_SID_TPGS_MASK 0x30 #define SPC3_SID_TPGS_IMPLICIT 0x10 #define SPC3_SID_TPGS_EXPLICIT 0x20 #define SPC3_SID_ACC 0x40 #define SPC3_SID_SCCS 0x80 #define CAM_PRIORITY_NORMAL CAM_PRIORITY_NONE #endif #endif