diff --git a/sys/dev/mps/mps.c b/sys/dev/mps/mps.c index 7a3e4f7637d2..5e41b0ab2183 100644 --- a/sys/dev/mps/mps.c +++ b/sys/dev/mps/mps.c @@ -1,2486 +1,2486 @@ /*- * Copyright (c) 2009 Yahoo! Inc. * Copyright (c) 2012 LSI Corp. * 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. * * LSI MPT-Fusion Host Adapter FreeBSD * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); /* Communications core for 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 static int mps_diag_reset(struct mps_softc *sc, int sleep_flag); static int mps_init_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 void mps_startup(void *arg); static int mps_send_iocinit(struct mps_softc *sc); static int mps_attach_log(struct mps_softc *sc); static __inline void mps_complete_command(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_wait_db_ack(struct mps_softc *sc, int timeout, int sleep_flag); SYSCTL_NODE(_hw, OID_AUTO, mps, CTLFLAG_RD, 0, "MPS Driver Parameters"); MALLOC_DEFINE(M_MPT2, "mps", "mpt2 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 support unint64_t to be passed as argument. * Otherwise it will through 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; /* * 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; mps_dprint(sc, MPS_TRACE, "%s\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_pflags & TDP_NOSLEEPING) + */ + if (curthread->td_no_sleeping != 0) sleep_flag = NO_SLEEP; /* 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) return (error); /* Send the actual reset. XXX need to refresh the reg? */ mps_regwrite(sc, MPI2_HOST_DIAGNOSTIC_OFFSET, reg | MPI2_DIAG_RESET_ADAPTER); /* Wait up to 300 seconds in 50ms intervals */ error = ETIMEDOUT; for (i = 0; i < 60000; i++) { /* wait 50 msec */ 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); reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); if ((reg & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_RESET) { error = 0; break; } } if (error) return (error); mps_regwrite(sc, MPI2_WRITE_SEQUENCE_OFFSET, 0x0); return (0); } static int mps_message_unit_reset(struct mps_softc *sc, int sleep_flag) { mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 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_FAULT, "Doorbell handshake failed : <%s>\n", __func__); return (ETIMEDOUT); } return (0); } static int mps_transition_ready(struct mps_softc *sc) { uint32_t reg, state; int error, tries = 0; int sleep_flags; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); /* If we are in attach call, do not sleep */ sleep_flags = (sc->mps_flags & MPS_FLAGS_ATTACH_DONE) ? CAN_SLEEP:NO_SLEEP; error = 0; while (tries++ < 5) { reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_dprint(sc, MPS_INFO, "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_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)) { device_printf(sc->mps_dev, "IOC is under the control " "of another peer host, aborting initialization.\n"); return (ENXIO); } 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_INFO, "IOC in fault state 0x%x\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_FAULT, "IOC in unexpected reset state\n"); } else { mps_dprint(sc, MPS_FAULT, "IOC in unknown state 0x%x\n", state); error = EINVAL; break; } /* Wait 50ms for things to settle down. */ DELAY(50000); } if (error) device_printf(sc->mps_dev, "Cannot transition IOC to ready\n"); return (error); } static int mps_transition_operational(struct mps_softc *sc) { uint32_t reg, state; int error; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); error = 0; reg = mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_dprint(sc, MPS_INFO, "Doorbell= 0x%x\n", reg); state = reg & MPI2_IOC_STATE_MASK; if (state != MPI2_IOC_STATE_READY) { if ((error = mps_transition_ready(sc)) != 0) { mps_dprint(sc, MPS_FAULT, "%s failed to transition ready\n", __func__); return (error); } } error = mps_send_iocinit(sc); return (error); } /* * XXX Some of this should probably move to mps.c * * 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; uint32_t db; mps_printf(sc, "%s sc %p\n", __func__, sc); mtx_assert(&sc->mps_mtx, MA_OWNED); if (sc->mps_flags & MPS_FLAGS_DIAGRESET) { mps_printf(sc, "%s reset already in progress\n", __func__); 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; mps_printf(sc, "%s mask interrupts\n", __func__); mps_mask_intr(sc); error = mps_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 */ mps_pci_restore(sc); /* Give the I/O subsystem special priority to get itself prepared */ mpssas_handle_reinit(sc); /* reinitialize queues after the reset */ bzero(sc->free_queue, sc->fqdepth * 4); mps_init_queues(sc); /* get the chip out of the reset state */ error = mps_transition_operational(sc); if (error != 0) panic("%s transition operational failed with error %d\n", __func__, error); /* Reinitialize the reply queue. This is delicate because this * function is typically invoked by task mgmt completion callbacks, * which are called by the interrupt thread. We need to make sure * the interrupt handler loop will exit when we return to it, and * that it will recognize the indexes we've changed. */ sc->replypostindex = 0; mps_regwrite(sc, MPI2_REPLY_FREE_HOST_INDEX_OFFSET, sc->replyfreeindex); mps_regwrite(sc, MPI2_REPLY_POST_HOST_INDEX_OFFSET, sc->replypostindex); db = mps_regread(sc, MPI2_DOORBELL_OFFSET); mps_printf(sc, "%s doorbell 0x%08x\n", __func__, db); mps_printf(sc, "%s unmask interrupts post %u free %u\n", __func__, sc->replypostindex, sc->replyfreeindex); mps_unmask_intr(sc); mps_printf(sc, "%s restarting post %u free %u\n", __func__, sc->replypostindex, sc->replyfreeindex); /* 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_printf(sc, "%s finished sc %p post %u free %u\n", __func__, sc, sc->replypostindex, sc->replyfreeindex); sc->mps_flags &= ~MPS_FLAGS_DIAGRESET; 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_INFO, "%s: successfull 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_pflags & TDP_NOSLEEPING) + + 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_FAULT, "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_dprint(sc, MPS_TRACE, "%s SMID %u cm %p ccb %p\n", __func__, 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); /* 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_dprint(sc, MPS_TRACE, "%s\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); return (error); } static int mps_get_portfacts(struct mps_softc *sc, MPI2_PORT_FACTS_REPLY *facts, int port) { MPI2_PORT_FACTS_REQUEST *request; MPI2_PORT_FACTS_REPLY *reply; struct mps_command *cm; int error; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); if ((cm = mps_alloc_command(sc)) == NULL) return (EBUSY); request = (MPI2_PORT_FACTS_REQUEST *)cm->cm_req; request->Function = MPI2_FUNCTION_PORT_FACTS; request->PortNumber = port; cm->cm_desc.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE; cm->cm_data = NULL; error = mps_request_polled(sc, cm); reply = (MPI2_PORT_FACTS_REPLY *)cm->cm_reply; if (reply == NULL) { mps_printf(sc, "%s NULL reply\n", __func__); goto done; } if ((reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) { mps_printf(sc, "%s error %d iocstatus 0x%x iocloginfo 0x%x type 0x%x\n", __func__, error, reply->IOCStatus, reply->IOCLogInfo, reply->PortType); error = ENXIO; } bcopy(reply, facts, sizeof(MPI2_PORT_FACTS_REPLY)); done: mps_free_command(sc, cm); 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; mps_dprint(sc, MPS_TRACE, "%s\n", __func__); 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(sc->facts->IOCRequestFrameSize); 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); init.TimeStamp.High = 0; init.TimeStamp.Low = htole32((uint32_t)time_uptime); 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_INFO, "IOCInit status= 0x%x\n", reply.IOCStatus); 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; } static int mps_alloc_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)) { device_printf(sc->mps_dev, "Cannot allocate queues DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->queues_dmat, (void **)&queues, BUS_DMA_NOWAIT, &sc->queues_map)) { device_printf(sc->mps_dev, "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; return (0); } static int mps_alloc_replies(struct mps_softc *sc) { int rsize, num_replies; /* * 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->facts->ReplyFrameSize * num_replies * 4; 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)) { device_printf(sc->mps_dev, "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)) { device_printf(sc->mps_dev, "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); return (0); } static int mps_alloc_requests(struct mps_softc *sc) { struct mps_command *cm; struct mps_chain *chain; int i, rsize, nsegs; rsize = sc->facts->IOCRequestFrameSize * sc->num_reqs * 4; 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)) { device_printf(sc->mps_dev, "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)) { device_printf(sc->mps_dev, "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); rsize = sc->facts->IOCRequestFrameSize * sc->max_chains * 4; 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->chain_dmat)) { device_printf(sc->mps_dev, "Cannot allocate chain DMA tag\n"); return (ENOMEM); } if (bus_dmamem_alloc(sc->chain_dmat, (void **)&sc->chain_frames, BUS_DMA_NOWAIT, &sc->chain_map)) { device_printf(sc->mps_dev, "Cannot allocate chain memory\n"); return (ENOMEM); } bzero(sc->chain_frames, rsize); bus_dmamap_load(sc->chain_dmat, sc->chain_map, sc->chain_frames, rsize, mps_memaddr_cb, &sc->chain_busaddr, 0); 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)) { device_printf(sc->mps_dev, "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)) { device_printf(sc->mps_dev, "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); sc->chains = malloc(sizeof(struct mps_chain) * sc->max_chains, M_MPT2, M_WAITOK | M_ZERO); if(!sc->chains) { device_printf(sc->mps_dev, "Cannot allocate chains memory %s %d\n", __func__, __LINE__); return (ENOMEM); } for (i = 0; i < sc->max_chains; i++) { chain = &sc->chains[i]; chain->chain = (MPI2_SGE_IO_UNION *)(sc->chain_frames + i * sc->facts->IOCRequestFrameSize * 4); chain->chain_busaddr = sc->chain_busaddr + i * sc->facts->IOCRequestFrameSize * 4; mps_free_chain(sc, chain); sc->chain_free_lowwater++; } /* XXX Need to pick a more precise value */ nsegs = (MAXPHYS / 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)) { device_printf(sc->mps_dev, "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) { device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n", __func__, __LINE__); return (ENOMEM); } for (i = 1; i < sc->num_reqs; i++) { cm = &sc->commands[i]; cm->cm_req = sc->req_frames + i * sc->facts->IOCRequestFrameSize * 4; cm->cm_req_busaddr = sc->req_busaddr + i * sc->facts->IOCRequestFrameSize * 4; 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; 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->facts->HighPriorityCredit) 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->facts->ReplyFrameSize * 4); 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. */ static void mps_get_tunables(struct mps_softc *sc) { char tmpstr[80]; /* XXX default to some debugging for now */ sc->mps_debug = MPS_FAULT; sc->disable_msix = 0; sc->disable_msi = 0; sc->max_chains = MPS_CHAIN_FRAMES; /* * Grab the global variables. */ TUNABLE_INT_FETCH("hw.mps.debug_level", &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_chains", &sc->max_chains); /* Grab the unit-instance variables */ snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.debug_level", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &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_chains", device_get_unit(sc->mps_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->max_chains); } 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_INT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "debug_level", CTLFLAG_RW, &sc->mps_debug, 0, "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_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"); #if __FreeBSD_version >= 900030 SYSCTL_ADD_UQUAD(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO, "chain_alloc_fail", CTLFLAG_RD, &sc->chain_alloc_fail, "chain allocation failures"); #endif //FreeBSD_version >= 900030 } int mps_attach(struct mps_softc *sc) { int i, error; mps_get_tunables(sc); mps_dprint(sc, MPS_TRACE, "%s\n", __func__); mtx_init(&sc->mps_mtx, "MPT2SAS lock", NULL, MTX_DEF); callout_init_mtx(&sc->periodic, &sc->mps_mtx, 0); TAILQ_INIT(&sc->event_list); if ((error = mps_transition_ready(sc)) != 0) { mps_printf(sc, "%s failed to transition ready\n", __func__); return (error); } sc->facts = malloc(sizeof(MPI2_IOC_FACTS_REPLY), M_MPT2, M_ZERO|M_NOWAIT); if(!sc->facts) { device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n", __func__, __LINE__); return (ENOMEM); } if ((error = mps_get_iocfacts(sc, sc->facts)) != 0) return (error); mps_print_iocfacts(sc, 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_printf(sc, "Firmware: %s, Driver: %s\n", sc->fw_version, MPS_DRIVER_VERSION); mps_printf(sc, "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. */ if ((sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY) == 0) { mps_diag_reset(sc, NO_SLEEP); if ((error = mps_transition_ready(sc)) != 0) return (error); } /* * Set flag if IR Firmware is loaded. */ if (sc->facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID) sc->ir_firmware = 1; /* * 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; /* * Size the queues. Since the reply queues always need one free entry, * we'll just deduct one reply message here. */ sc->num_reqs = MIN(MPS_REQ_FRAMES, sc->facts->RequestCredit); sc->num_replies = MIN(MPS_REPLY_FRAMES + MPS_EVT_REPLY_FRAMES, sc->facts->MaxReplyDescriptorPostQueueDepth) - 1; TAILQ_INIT(&sc->req_list); TAILQ_INIT(&sc->high_priority_req_list); TAILQ_INIT(&sc->chain_list); TAILQ_INIT(&sc->tm_list); if (((error = mps_alloc_queues(sc)) != 0) || ((error = mps_alloc_replies(sc)) != 0) || ((error = mps_alloc_requests(sc)) != 0)) { mps_printf(sc, "%s failed to alloc\n", __func__); mps_free(sc); return (error); } if (((error = mps_init_queues(sc)) != 0) || ((error = mps_transition_operational(sc)) != 0)) { mps_printf(sc, "%s failed to transition operational\n", __func__); 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); sc->pfacts = malloc(sizeof(MPI2_PORT_FACTS_REPLY) * sc->facts->NumberOfPorts, M_MPT2, M_ZERO|M_WAITOK); if(!sc->pfacts) { device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n", __func__, __LINE__); return (ENOMEM); } for (i = 0; i < sc->facts->NumberOfPorts; i++) { if ((error = mps_get_portfacts(sc, &sc->pfacts[i], i)) != 0) { mps_printf(sc, "%s failed to get portfacts for port %d\n", __func__, i); mps_free(sc); return (error); } mps_print_portfacts(sc, &sc->pfacts[i]); } /* Attach the subsystems so they can prepare their event masks. */ /* XXX Should be dynamic so that IM/IR and user modules can attach */ if (((error = mps_attach_log(sc)) != 0) || ((error = mps_attach_sas(sc)) != 0) || ((error = mps_attach_user(sc)) != 0)) { mps_printf(sc, "%s failed to attach all subsystems: error %d\n", __func__, error); mps_free(sc); return (error); } if ((error = mps_pci_setup_interrupts(sc)) != 0) { mps_printf(sc, "%s failed to setup interrupts\n", __func__); mps_free(sc); return (error); } /* * The static page function currently read is ioc page8. Others can be * added in future. */ mps_base_static_config_pages(sc); /* 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_FAULT, "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_FAULT, "shutdown event registration " "failed\n"); mps_setup_sysctl(sc); sc->mps_flags |= MPS_FLAGS_ATTACH_DONE; return (error); } /* Run through any late-start handlers. */ static void mps_startup(void *arg) { struct mps_softc *sc; sc = (struct mps_softc *)arg; mps_lock(sc); mps_unmask_intr(sc); /* initialize device mapping tables */ mps_mapping_initialize(sc); mpssas_startup(sc); mps_unlock(sc); } /* 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) { device_printf(sc->mps_dev, "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_print_event(sc, event); switch (event->Event) { case MPI2_EVENT_LOG_DATA: device_printf(sc->mps_dev, "MPI2_EVENT_LOG_DATA:\n"); 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_INFO, "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) { struct mps_command *cm; int i, error; /* 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); if (((error = mps_detach_log(sc)) != 0) || ((error = mps_detach_sas(sc)) != 0)) 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); if (sc->pfacts != NULL) free(sc->pfacts, M_MPT2); if (sc->post_busaddr != 0) bus_dmamap_unload(sc->queues_dmat, sc->queues_map); if (sc->post_queue != NULL) bus_dmamem_free(sc->queues_dmat, sc->post_queue, sc->queues_map); if (sc->queues_dmat != NULL) bus_dma_tag_destroy(sc->queues_dmat); if (sc->chain_busaddr != 0) bus_dmamap_unload(sc->chain_dmat, sc->chain_map); if (sc->chain_frames != NULL) 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); if (sc->sysctl_tree != NULL) sysctl_ctx_free(&sc->sysctl_ctx); mps_mapping_free_memory(sc); /* Deregister the shutdown function */ if (sc->shutdown_eh != NULL) EVENTHANDLER_DEREGISTER(shutdown_final, sc->shutdown_eh); mtx_destroy(&sc->mps_mtx); return (0); } static __inline void mps_complete_command(struct mps_command *cm) { if (cm->cm_flags & MPS_CM_FLAGS_POLLED) cm->cm_flags |= MPS_CM_FLAGS_COMPLETE; if (cm->cm_complete != NULL) { mps_dprint(cm->cm_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(cm->cm_sc, cm); } if (cm->cm_flags & MPS_CM_FLAGS_WAKEUP) { mps_dprint(cm->cm_sc, MPS_TRACE, "%s: waking up %p\n", __func__, cm); wakeup(cm); } if (cm->cm_sc->io_cmds_active != 0) { cm->cm_sc->io_cmds_active--; } else { mps_dprint(cm->cm_sc, MPS_INFO, "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_INFO, "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; struct mps_softc *sc; struct mps_command *cm = NULL; uint8_t flags; u_int pq; MPI2_DIAG_RELEASE_REPLY *rel_rep; mps_fw_diagnostic_buffer_t *pBuffer; 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]; 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)]; 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->facts->ReplyFrameSize * 4)))) { 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->facts->ReplyFrameSize * 4); printf("%s: baddr %#x,\n", __func__, baddr); /* LSI-TODO. See Linux Code. Need Gracefull 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_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 { cm = &sc->commands[le16toh(desc->AddressReply.SMID)]; cm->cm_reply = reply; cm->cm_reply_data = le32toh(desc->AddressReply.ReplyFrameAddress); } 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 */ device_printf(sc->mps_dev, "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(cm); } desc->Words.Low = 0xffffffff; desc->Words.High = 0xffffffff; } 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) device_printf(sc->mps_dev, "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_print_event(sc, (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) { device_printf(sc->mps_dev, "Cannot allocate memory %s %d\n", __func__, __LINE__); 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; 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_request_polled(sc, cm); reply = (MPI2_EVENT_NOTIFICATION_REPLY *)cm->cm_reply; if ((reply == NULL) || (reply->IOCStatus & MPI2_IOCSTATUS_MASK) != MPI2_IOCSTATUS_SUCCESS) error = ENXIO; mps_print_event(sc, reply); mps_dprint(sc, MPS_TRACE, "%s finished error %d\n", __func__, error); 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; 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 = (int)cm->cm_sc->facts->IOCRequestFrameSize * 4; /* * 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 seperate 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 >= 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)); } 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); } #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_printf(sc, "%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! */ mps_dprint(sc, MPS_INFO, "out of chain frames\n"); cm->cm_flags |= MPS_CM_FLAGS_CHAIN_FAILED; mps_complete_command(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 *cm, int timeout) { int error, rc; mtx_assert(&sc->mps_mtx, MA_OWNED); if(sc->mps_flags & MPS_FLAGS_DIAGRESET) return EBUSY; cm->cm_complete = NULL; cm->cm_flags |= MPS_CM_FLAGS_WAKEUP; error = mps_map_command(sc, cm); if ((error != 0) && (error != EINPROGRESS)) return (error); error = msleep(cm, &sc->mps_mtx, 0, "mpswait", timeout*hz); if (error == EWOULDBLOCK) { mps_dprint(sc, MPS_FAULT, "Calling Reinit from %s\n", __func__); rc = mps_reinit(sc); mps_dprint(sc, MPS_FAULT, "Reinit %s\n", (rc == 0) ? "success" : "failed"); error = ETIMEDOUT; } return (error); } /* * This is the routine to enqueue a command synchonously and poll for * completion. Its use should be rare. */ int mps_request_polled(struct mps_softc *sc, struct mps_command *cm) { int error, timeout = 0, rc; error = 0; cm->cm_flags |= MPS_CM_FLAGS_POLLED; cm->cm_complete = NULL; mps_map_command(sc, cm); while ((cm->cm_flags & MPS_CM_FLAGS_COMPLETE) == 0) { mps_intr_locked(sc); DELAY(50 * 1000); if (timeout++ > 1000) { mps_dprint(sc, MPS_FAULT, "polling failed\n"); error = ETIMEDOUT; break; } } if (error) { mps_dprint(sc, MPS_FAULT, "Calling Reinit from %s\n", __func__); rc = mps_reinit(sc); mps_dprint(sc, MPS_FAULT, "Reinit %s\n", (rc == 0) ? "success" : "failed"); } return (error); } /* * The MPT driver had a verbose interface for config pages. In this driver, * reduce it to much simplier 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.Ext.ExtPageType != 0) { 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; 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; 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); if (error) { mps_dprint(sc, MPS_FAULT, "Error %d reading config page\n", error); 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; 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.Ext.ExtPageType != 0) { params->hdr.Ext.ExtPageType = reply->ExtPageType; params->hdr.Ext.ExtPageLength = reply->ExtPageLength; } 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; } diff --git a/sys/kern/subr_sleepqueue.c b/sys/kern/subr_sleepqueue.c index b6bd8fcc8842..f1875446ab8f 100644 --- a/sys/kern/subr_sleepqueue.c +++ b/sys/kern/subr_sleepqueue.c @@ -1,1238 +1,1238 @@ /*- * Copyright (c) 2004 John Baldwin * 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. * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * 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. */ /* * Implementation of sleep queues used to hold queue of threads blocked on * a wait channel. Sleep queues different from turnstiles in that wait * channels are not owned by anyone, so there is no priority propagation. * Sleep queues can also provide a timeout and can also be interrupted by * signals. That said, there are several similarities between the turnstile * and sleep queue implementations. (Note: turnstiles were implemented * first.) For example, both use a hash table of the same size where each * bucket is referred to as a "chain" that contains both a spin lock and * a linked list of queues. An individual queue is located by using a hash * to pick a chain, locking the chain, and then walking the chain searching * for the queue. This means that a wait channel object does not need to * embed it's queue head just as locks do not embed their turnstile queue * head. Threads also carry around a sleep queue that they lend to the * wait channel when blocking. Just as in turnstiles, the queue includes * a free list of the sleep queues of other threads blocked on the same * wait channel in the case of multiple waiters. * * Some additional functionality provided by sleep queues include the * ability to set a timeout. The timeout is managed using a per-thread * callout that resumes a thread if it is asleep. A thread may also * catch signals while it is asleep (aka an interruptible sleep). The * signal code uses sleepq_abort() to interrupt a sleeping thread. Finally, * sleep queues also provide some extra assertions. One is not allowed to * mix the sleep/wakeup and cv APIs for a given wait channel. Also, one * must consistently use the same lock to synchronize with a wait channel, * though this check is currently only a warning for sleep/wakeup due to * pre-existing abuse of that API. The same lock must also be held when * awakening threads, though that is currently only enforced for condition * variables. */ #include __FBSDID("$FreeBSD$"); #include "opt_sleepqueue_profiling.h" #include "opt_ddb.h" #include "opt_kdtrace.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif /* * Constants for the hash table of sleep queue chains. These constants are * the same ones that 4BSD (and possibly earlier versions of BSD) used. * Basically, we ignore the lower 8 bits of the address since most wait * channel pointers are aligned and only look at the next 7 bits for the * hash. SC_TABLESIZE must be a power of two for SC_MASK to work properly. */ #define SC_TABLESIZE 128 /* Must be power of 2. */ #define SC_MASK (SC_TABLESIZE - 1) #define SC_SHIFT 8 #define SC_HASH(wc) (((uintptr_t)(wc) >> SC_SHIFT) & SC_MASK) #define SC_LOOKUP(wc) &sleepq_chains[SC_HASH(wc)] #define NR_SLEEPQS 2 /* * There two different lists of sleep queues. Both lists are connected * via the sq_hash entries. The first list is the sleep queue chain list * that a sleep queue is on when it is attached to a wait channel. The * second list is the free list hung off of a sleep queue that is attached * to a wait channel. * * Each sleep queue also contains the wait channel it is attached to, the * list of threads blocked on that wait channel, flags specific to the * wait channel, and the lock used to synchronize with a wait channel. * The flags are used to catch mismatches between the various consumers * of the sleep queue API (e.g. sleep/wakeup and condition variables). * The lock pointer is only used when invariants are enabled for various * debugging checks. * * Locking key: * c - sleep queue chain lock */ struct sleepqueue { TAILQ_HEAD(, thread) sq_blocked[NR_SLEEPQS]; /* (c) Blocked threads. */ u_int sq_blockedcnt[NR_SLEEPQS]; /* (c) N. of blocked threads. */ LIST_ENTRY(sleepqueue) sq_hash; /* (c) Chain and free list. */ LIST_HEAD(, sleepqueue) sq_free; /* (c) Free queues. */ void *sq_wchan; /* (c) Wait channel. */ int sq_type; /* (c) Queue type. */ #ifdef INVARIANTS struct lock_object *sq_lock; /* (c) Associated lock. */ #endif }; struct sleepqueue_chain { LIST_HEAD(, sleepqueue) sc_queues; /* List of sleep queues. */ struct mtx sc_lock; /* Spin lock for this chain. */ #ifdef SLEEPQUEUE_PROFILING u_int sc_depth; /* Length of sc_queues. */ u_int sc_max_depth; /* Max length of sc_queues. */ #endif }; #ifdef SLEEPQUEUE_PROFILING u_int sleepq_max_depth; static SYSCTL_NODE(_debug, OID_AUTO, sleepq, CTLFLAG_RD, 0, "sleepq profiling"); static SYSCTL_NODE(_debug_sleepq, OID_AUTO, chains, CTLFLAG_RD, 0, "sleepq chain stats"); SYSCTL_UINT(_debug_sleepq, OID_AUTO, max_depth, CTLFLAG_RD, &sleepq_max_depth, 0, "maxmimum depth achieved of a single chain"); static void sleepq_profile(const char *wmesg); static int prof_enabled; #endif static struct sleepqueue_chain sleepq_chains[SC_TABLESIZE]; static uma_zone_t sleepq_zone; /* * Prototypes for non-exported routines. */ static int sleepq_catch_signals(void *wchan, int pri); static int sleepq_check_signals(void); static int sleepq_check_timeout(void); #ifdef INVARIANTS static void sleepq_dtor(void *mem, int size, void *arg); #endif static int sleepq_init(void *mem, int size, int flags); static int sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri); static void sleepq_switch(void *wchan, int pri); static void sleepq_timeout(void *arg); SDT_PROBE_DECLARE(sched, , , sleep); SDT_PROBE_DECLARE(sched, , , wakeup); /* * Early initialization of sleep queues that is called from the sleepinit() * SYSINIT. */ void init_sleepqueues(void) { #ifdef SLEEPQUEUE_PROFILING struct sysctl_oid *chain_oid; char chain_name[10]; #endif int i; for (i = 0; i < SC_TABLESIZE; i++) { LIST_INIT(&sleepq_chains[i].sc_queues); mtx_init(&sleepq_chains[i].sc_lock, "sleepq chain", NULL, MTX_SPIN | MTX_RECURSE); #ifdef SLEEPQUEUE_PROFILING snprintf(chain_name, sizeof(chain_name), "%d", i); chain_oid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(_debug_sleepq_chains), OID_AUTO, chain_name, CTLFLAG_RD, NULL, "sleepq chain stats"); SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO, "depth", CTLFLAG_RD, &sleepq_chains[i].sc_depth, 0, NULL); SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO, "max_depth", CTLFLAG_RD, &sleepq_chains[i].sc_max_depth, 0, NULL); #endif } sleepq_zone = uma_zcreate("SLEEPQUEUE", sizeof(struct sleepqueue), #ifdef INVARIANTS NULL, sleepq_dtor, sleepq_init, NULL, UMA_ALIGN_CACHE, 0); #else NULL, NULL, sleepq_init, NULL, UMA_ALIGN_CACHE, 0); #endif thread0.td_sleepqueue = sleepq_alloc(); } /* * Get a sleep queue for a new thread. */ struct sleepqueue * sleepq_alloc(void) { return (uma_zalloc(sleepq_zone, M_WAITOK)); } /* * Free a sleep queue when a thread is destroyed. */ void sleepq_free(struct sleepqueue *sq) { uma_zfree(sleepq_zone, sq); } /* * Lock the sleep queue chain associated with the specified wait channel. */ void sleepq_lock(void *wchan) { struct sleepqueue_chain *sc; sc = SC_LOOKUP(wchan); mtx_lock_spin(&sc->sc_lock); } /* * Look up the sleep queue associated with a given wait channel in the hash * table locking the associated sleep queue chain. If no queue is found in * the table, NULL is returned. */ struct sleepqueue * sleepq_lookup(void *wchan) { struct sleepqueue_chain *sc; struct sleepqueue *sq; KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); LIST_FOREACH(sq, &sc->sc_queues, sq_hash) if (sq->sq_wchan == wchan) return (sq); return (NULL); } /* * Unlock the sleep queue chain associated with a given wait channel. */ void sleepq_release(void *wchan) { struct sleepqueue_chain *sc; sc = SC_LOOKUP(wchan); mtx_unlock_spin(&sc->sc_lock); } /* * Places the current thread on the sleep queue for the specified wait * channel. If INVARIANTS is enabled, then it associates the passed in * lock with the sleepq to make sure it is held when that sleep queue is * woken up. */ void sleepq_add(void *wchan, struct lock_object *lock, const char *wmesg, int flags, int queue) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(td->td_sleepqueue != NULL); MPASS(wchan != NULL); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); /* If this thread is not allowed to sleep, die a horrible death. */ - KASSERT(!(td->td_pflags & TDP_NOSLEEPING), - ("%s: td %p to sleep on wchan %p with TDP_NOSLEEPING on", + KASSERT(td->td_no_sleeping == 0, + ("%s: td %p to sleep on wchan %p with sleeping prohibited", __func__, td, wchan)); /* Look up the sleep queue associated with the wait channel 'wchan'. */ sq = sleepq_lookup(wchan); /* * If the wait channel does not already have a sleep queue, use * this thread's sleep queue. Otherwise, insert the current thread * into the sleep queue already in use by this wait channel. */ if (sq == NULL) { #ifdef INVARIANTS int i; sq = td->td_sleepqueue; for (i = 0; i < NR_SLEEPQS; i++) { KASSERT(TAILQ_EMPTY(&sq->sq_blocked[i]), ("thread's sleep queue %d is not empty", i)); KASSERT(sq->sq_blockedcnt[i] == 0, ("thread's sleep queue %d count mismatches", i)); } KASSERT(LIST_EMPTY(&sq->sq_free), ("thread's sleep queue has a non-empty free list")); KASSERT(sq->sq_wchan == NULL, ("stale sq_wchan pointer")); sq->sq_lock = lock; #endif #ifdef SLEEPQUEUE_PROFILING sc->sc_depth++; if (sc->sc_depth > sc->sc_max_depth) { sc->sc_max_depth = sc->sc_depth; if (sc->sc_max_depth > sleepq_max_depth) sleepq_max_depth = sc->sc_max_depth; } #endif sq = td->td_sleepqueue; LIST_INSERT_HEAD(&sc->sc_queues, sq, sq_hash); sq->sq_wchan = wchan; sq->sq_type = flags & SLEEPQ_TYPE; } else { MPASS(wchan == sq->sq_wchan); MPASS(lock == sq->sq_lock); MPASS((flags & SLEEPQ_TYPE) == sq->sq_type); LIST_INSERT_HEAD(&sq->sq_free, td->td_sleepqueue, sq_hash); } thread_lock(td); TAILQ_INSERT_TAIL(&sq->sq_blocked[queue], td, td_slpq); sq->sq_blockedcnt[queue]++; td->td_sleepqueue = NULL; td->td_sqqueue = queue; td->td_wchan = wchan; td->td_wmesg = wmesg; if (flags & SLEEPQ_INTERRUPTIBLE) { td->td_flags |= TDF_SINTR; td->td_flags &= ~TDF_SLEEPABORT; } thread_unlock(td); } /* * Sets a timeout that will remove the current thread from the specified * sleep queue after timo ticks if the thread has not already been awakened. */ void sleepq_set_timeout(void *wchan, int timo) { struct sleepqueue_chain *sc; struct thread *td; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(TD_ON_SLEEPQ(td)); MPASS(td->td_sleepqueue == NULL); MPASS(wchan != NULL); callout_reset_curcpu(&td->td_slpcallout, timo, sleepq_timeout, td); } /* * Return the number of actual sleepers for the specified queue. */ u_int sleepq_sleepcnt(void *wchan, int queue) { struct sleepqueue *sq; KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); if (sq == NULL) return (0); return (sq->sq_blockedcnt[queue]); } /* * Marks the pending sleep of the current thread as interruptible and * makes an initial check for pending signals before putting a thread * to sleep. Enters and exits with the thread lock held. Thread lock * may have transitioned from the sleepq lock to a run lock. */ static int sleepq_catch_signals(void *wchan, int pri) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; struct proc *p; struct sigacts *ps; int sig, ret, stop_allowed; td = curthread; p = curproc; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); MPASS(wchan != NULL); if ((td->td_pflags & TDP_WAKEUP) != 0) { td->td_pflags &= ~TDP_WAKEUP; ret = EINTR; thread_lock(td); goto out; } /* * See if there are any pending signals for this thread. If not * we can switch immediately. Otherwise do the signal processing * directly. */ thread_lock(td); if ((td->td_flags & (TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK)) == 0) { sleepq_switch(wchan, pri); return (0); } stop_allowed = (td->td_flags & TDF_SBDRY) ? SIG_STOP_NOT_ALLOWED : SIG_STOP_ALLOWED; thread_unlock(td); mtx_unlock_spin(&sc->sc_lock); CTR3(KTR_PROC, "sleepq catching signals: thread %p (pid %ld, %s)", (void *)td, (long)p->p_pid, td->td_name); PROC_LOCK(p); ps = p->p_sigacts; mtx_lock(&ps->ps_mtx); sig = cursig(td, stop_allowed); if (sig == 0) { mtx_unlock(&ps->ps_mtx); ret = thread_suspend_check(1); MPASS(ret == 0 || ret == EINTR || ret == ERESTART); } else { if (SIGISMEMBER(ps->ps_sigintr, sig)) ret = EINTR; else ret = ERESTART; mtx_unlock(&ps->ps_mtx); } /* * Lock the per-process spinlock prior to dropping the PROC_LOCK * to avoid a signal delivery race. PROC_LOCK, PROC_SLOCK, and * thread_lock() are currently held in tdsendsignal(). */ PROC_SLOCK(p); mtx_lock_spin(&sc->sc_lock); PROC_UNLOCK(p); thread_lock(td); PROC_SUNLOCK(p); if (ret == 0) { sleepq_switch(wchan, pri); return (0); } out: /* * There were pending signals and this thread is still * on the sleep queue, remove it from the sleep queue. */ if (TD_ON_SLEEPQ(td)) { sq = sleepq_lookup(wchan); if (sleepq_resume_thread(sq, td, 0)) { #ifdef INVARIANTS /* * This thread hasn't gone to sleep yet, so it * should not be swapped out. */ panic("not waking up swapper"); #endif } } mtx_unlock_spin(&sc->sc_lock); MPASS(td->td_lock != &sc->sc_lock); return (ret); } /* * Switches to another thread if we are still asleep on a sleep queue. * Returns with thread lock. */ static void sleepq_switch(void *wchan, int pri) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; td = curthread; sc = SC_LOOKUP(wchan); mtx_assert(&sc->sc_lock, MA_OWNED); THREAD_LOCK_ASSERT(td, MA_OWNED); /* * If we have a sleep queue, then we've already been woken up, so * just return. */ if (td->td_sleepqueue != NULL) { mtx_unlock_spin(&sc->sc_lock); return; } /* * If TDF_TIMEOUT is set, then our sleep has been timed out * already but we are still on the sleep queue, so dequeue the * thread and return. */ if (td->td_flags & TDF_TIMEOUT) { MPASS(TD_ON_SLEEPQ(td)); sq = sleepq_lookup(wchan); if (sleepq_resume_thread(sq, td, 0)) { #ifdef INVARIANTS /* * This thread hasn't gone to sleep yet, so it * should not be swapped out. */ panic("not waking up swapper"); #endif } mtx_unlock_spin(&sc->sc_lock); return; } #ifdef SLEEPQUEUE_PROFILING if (prof_enabled) sleepq_profile(td->td_wmesg); #endif MPASS(td->td_sleepqueue == NULL); sched_sleep(td, pri); thread_lock_set(td, &sc->sc_lock); SDT_PROBE0(sched, , , sleep); TD_SET_SLEEPING(td); mi_switch(SW_VOL | SWT_SLEEPQ, NULL); KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING")); CTR3(KTR_PROC, "sleepq resume: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); } /* * Check to see if we timed out. */ static int sleepq_check_timeout(void) { struct thread *td; td = curthread; THREAD_LOCK_ASSERT(td, MA_OWNED); /* * If TDF_TIMEOUT is set, we timed out. */ if (td->td_flags & TDF_TIMEOUT) { td->td_flags &= ~TDF_TIMEOUT; return (EWOULDBLOCK); } /* * If TDF_TIMOFAIL is set, the timeout ran after we had * already been woken up. */ if (td->td_flags & TDF_TIMOFAIL) td->td_flags &= ~TDF_TIMOFAIL; /* * If callout_stop() fails, then the timeout is running on * another CPU, so synchronize with it to avoid having it * accidentally wake up a subsequent sleep. */ else if (callout_stop(&td->td_slpcallout) == 0) { td->td_flags |= TDF_TIMEOUT; TD_SET_SLEEPING(td); mi_switch(SW_INVOL | SWT_SLEEPQTIMO, NULL); } return (0); } /* * Check to see if we were awoken by a signal. */ static int sleepq_check_signals(void) { struct thread *td; td = curthread; THREAD_LOCK_ASSERT(td, MA_OWNED); /* We are no longer in an interruptible sleep. */ if (td->td_flags & TDF_SINTR) td->td_flags &= ~TDF_SINTR; if (td->td_flags & TDF_SLEEPABORT) { td->td_flags &= ~TDF_SLEEPABORT; return (td->td_intrval); } return (0); } /* * Block the current thread until it is awakened from its sleep queue. */ void sleepq_wait(void *wchan, int pri) { struct thread *td; td = curthread; MPASS(!(td->td_flags & TDF_SINTR)); thread_lock(td); sleepq_switch(wchan, pri); thread_unlock(td); } /* * Block the current thread until it is awakened from its sleep queue * or it is interrupted by a signal. */ int sleepq_wait_sig(void *wchan, int pri) { int rcatch; int rval; rcatch = sleepq_catch_signals(wchan, pri); rval = sleepq_check_signals(); thread_unlock(curthread); if (rcatch) return (rcatch); return (rval); } /* * Block the current thread until it is awakened from its sleep queue * or it times out while waiting. */ int sleepq_timedwait(void *wchan, int pri) { struct thread *td; int rval; td = curthread; MPASS(!(td->td_flags & TDF_SINTR)); thread_lock(td); sleepq_switch(wchan, pri); rval = sleepq_check_timeout(); thread_unlock(td); return (rval); } /* * Block the current thread until it is awakened from its sleep queue, * it is interrupted by a signal, or it times out waiting to be awakened. */ int sleepq_timedwait_sig(void *wchan, int pri) { int rcatch, rvalt, rvals; rcatch = sleepq_catch_signals(wchan, pri); rvalt = sleepq_check_timeout(); rvals = sleepq_check_signals(); thread_unlock(curthread); if (rcatch) return (rcatch); if (rvals) return (rvals); return (rvalt); } /* * Returns the type of sleepqueue given a waitchannel. */ int sleepq_type(void *wchan) { struct sleepqueue *sq; int type; MPASS(wchan != NULL); sleepq_lock(wchan); sq = sleepq_lookup(wchan); if (sq == NULL) { sleepq_release(wchan); return (-1); } type = sq->sq_type; sleepq_release(wchan); return (type); } /* * Removes a thread from a sleep queue and makes it * runnable. */ static int sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri) { struct sleepqueue_chain *sc; MPASS(td != NULL); MPASS(sq->sq_wchan != NULL); MPASS(td->td_wchan == sq->sq_wchan); MPASS(td->td_sqqueue < NR_SLEEPQS && td->td_sqqueue >= 0); THREAD_LOCK_ASSERT(td, MA_OWNED); sc = SC_LOOKUP(sq->sq_wchan); mtx_assert(&sc->sc_lock, MA_OWNED); SDT_PROBE2(sched, , , wakeup, td, td->td_proc); /* Remove the thread from the queue. */ sq->sq_blockedcnt[td->td_sqqueue]--; TAILQ_REMOVE(&sq->sq_blocked[td->td_sqqueue], td, td_slpq); /* * Get a sleep queue for this thread. If this is the last waiter, * use the queue itself and take it out of the chain, otherwise, * remove a queue from the free list. */ if (LIST_EMPTY(&sq->sq_free)) { td->td_sleepqueue = sq; #ifdef INVARIANTS sq->sq_wchan = NULL; #endif #ifdef SLEEPQUEUE_PROFILING sc->sc_depth--; #endif } else td->td_sleepqueue = LIST_FIRST(&sq->sq_free); LIST_REMOVE(td->td_sleepqueue, sq_hash); td->td_wmesg = NULL; td->td_wchan = NULL; td->td_flags &= ~TDF_SINTR; CTR3(KTR_PROC, "sleepq_wakeup: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, td->td_name); /* Adjust priority if requested. */ MPASS(pri == 0 || (pri >= PRI_MIN && pri <= PRI_MAX)); if (pri != 0 && td->td_priority > pri && PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) sched_prio(td, pri); /* * Note that thread td might not be sleeping if it is running * sleepq_catch_signals() on another CPU or is blocked on its * proc lock to check signals. There's no need to mark the * thread runnable in that case. */ if (TD_IS_SLEEPING(td)) { TD_CLR_SLEEPING(td); return (setrunnable(td)); } return (0); } #ifdef INVARIANTS /* * UMA zone item deallocator. */ static void sleepq_dtor(void *mem, int size, void *arg) { struct sleepqueue *sq; int i; sq = mem; for (i = 0; i < NR_SLEEPQS; i++) { MPASS(TAILQ_EMPTY(&sq->sq_blocked[i])); MPASS(sq->sq_blockedcnt[i] == 0); } } #endif /* * UMA zone item initializer. */ static int sleepq_init(void *mem, int size, int flags) { struct sleepqueue *sq; int i; bzero(mem, size); sq = mem; for (i = 0; i < NR_SLEEPQS; i++) { TAILQ_INIT(&sq->sq_blocked[i]); sq->sq_blockedcnt[i] = 0; } LIST_INIT(&sq->sq_free); return (0); } /* * Find the highest priority thread sleeping on a wait channel and resume it. */ int sleepq_signal(void *wchan, int flags, int pri, int queue) { struct sleepqueue *sq; struct thread *td, *besttd; int wakeup_swapper; CTR2(KTR_PROC, "sleepq_signal(%p, %d)", wchan, flags); KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); if (sq == NULL) return (0); KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE), ("%s: mismatch between sleep/wakeup and cv_*", __func__)); /* * Find the highest priority thread on the queue. If there is a * tie, use the thread that first appears in the queue as it has * been sleeping the longest since threads are always added to * the tail of sleep queues. */ besttd = NULL; TAILQ_FOREACH(td, &sq->sq_blocked[queue], td_slpq) { if (besttd == NULL || td->td_priority < besttd->td_priority) besttd = td; } MPASS(besttd != NULL); thread_lock(besttd); wakeup_swapper = sleepq_resume_thread(sq, besttd, pri); thread_unlock(besttd); return (wakeup_swapper); } /* * Resume all threads sleeping on a specified wait channel. */ int sleepq_broadcast(void *wchan, int flags, int pri, int queue) { struct sleepqueue *sq; struct thread *td, *tdn; int wakeup_swapper; CTR2(KTR_PROC, "sleepq_broadcast(%p, %d)", wchan, flags); KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__)); MPASS((queue >= 0) && (queue < NR_SLEEPQS)); sq = sleepq_lookup(wchan); if (sq == NULL) return (0); KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE), ("%s: mismatch between sleep/wakeup and cv_*", __func__)); /* Resume all blocked threads on the sleep queue. */ wakeup_swapper = 0; TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq, tdn) { thread_lock(td); if (sleepq_resume_thread(sq, td, pri)) wakeup_swapper = 1; thread_unlock(td); } return (wakeup_swapper); } /* * Time sleeping threads out. When the timeout expires, the thread is * removed from the sleep queue and made runnable if it is still asleep. */ static void sleepq_timeout(void *arg) { struct sleepqueue_chain *sc; struct sleepqueue *sq; struct thread *td; void *wchan; int wakeup_swapper; td = arg; wakeup_swapper = 0; CTR3(KTR_PROC, "sleepq_timeout: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); /* * First, see if the thread is asleep and get the wait channel if * it is. */ thread_lock(td); if (TD_IS_SLEEPING(td) && TD_ON_SLEEPQ(td)) { wchan = td->td_wchan; sc = SC_LOOKUP(wchan); THREAD_LOCKPTR_ASSERT(td, &sc->sc_lock); sq = sleepq_lookup(wchan); MPASS(sq != NULL); td->td_flags |= TDF_TIMEOUT; wakeup_swapper = sleepq_resume_thread(sq, td, 0); thread_unlock(td); if (wakeup_swapper) kick_proc0(); return; } /* * If the thread is on the SLEEPQ but isn't sleeping yet, it * can either be on another CPU in between sleepq_add() and * one of the sleepq_*wait*() routines or it can be in * sleepq_catch_signals(). */ if (TD_ON_SLEEPQ(td)) { td->td_flags |= TDF_TIMEOUT; thread_unlock(td); return; } /* * Now check for the edge cases. First, if TDF_TIMEOUT is set, * then the other thread has already yielded to us, so clear * the flag and resume it. If TDF_TIMEOUT is not set, then the * we know that the other thread is not on a sleep queue, but it * hasn't resumed execution yet. In that case, set TDF_TIMOFAIL * to let it know that the timeout has already run and doesn't * need to be canceled. */ if (td->td_flags & TDF_TIMEOUT) { MPASS(TD_IS_SLEEPING(td)); td->td_flags &= ~TDF_TIMEOUT; TD_CLR_SLEEPING(td); wakeup_swapper = setrunnable(td); } else td->td_flags |= TDF_TIMOFAIL; thread_unlock(td); if (wakeup_swapper) kick_proc0(); } /* * Resumes a specific thread from the sleep queue associated with a specific * wait channel if it is on that queue. */ void sleepq_remove(struct thread *td, void *wchan) { struct sleepqueue *sq; int wakeup_swapper; /* * Look up the sleep queue for this wait channel, then re-check * that the thread is asleep on that channel, if it is not, then * bail. */ MPASS(wchan != NULL); sleepq_lock(wchan); sq = sleepq_lookup(wchan); /* * We can not lock the thread here as it may be sleeping on a * different sleepq. However, holding the sleepq lock for this * wchan can guarantee that we do not miss a wakeup for this * channel. The asserts below will catch any false positives. */ if (!TD_ON_SLEEPQ(td) || td->td_wchan != wchan) { sleepq_release(wchan); return; } /* Thread is asleep on sleep queue sq, so wake it up. */ thread_lock(td); MPASS(sq != NULL); MPASS(td->td_wchan == wchan); wakeup_swapper = sleepq_resume_thread(sq, td, 0); thread_unlock(td); sleepq_release(wchan); if (wakeup_swapper) kick_proc0(); } /* * Abort a thread as if an interrupt had occurred. Only abort * interruptible waits (unfortunately it isn't safe to abort others). */ int sleepq_abort(struct thread *td, int intrval) { struct sleepqueue *sq; void *wchan; THREAD_LOCK_ASSERT(td, MA_OWNED); MPASS(TD_ON_SLEEPQ(td)); MPASS(td->td_flags & TDF_SINTR); MPASS(intrval == EINTR || intrval == ERESTART); /* * If the TDF_TIMEOUT flag is set, just leave. A * timeout is scheduled anyhow. */ if (td->td_flags & TDF_TIMEOUT) return (0); CTR3(KTR_PROC, "sleepq_abort: thread %p (pid %ld, %s)", (void *)td, (long)td->td_proc->p_pid, (void *)td->td_name); td->td_intrval = intrval; td->td_flags |= TDF_SLEEPABORT; /* * If the thread has not slept yet it will find the signal in * sleepq_catch_signals() and call sleepq_resume_thread. Otherwise * we have to do it here. */ if (!TD_IS_SLEEPING(td)) return (0); wchan = td->td_wchan; MPASS(wchan != NULL); sq = sleepq_lookup(wchan); MPASS(sq != NULL); /* Thread is asleep on sleep queue sq, so wake it up. */ return (sleepq_resume_thread(sq, td, 0)); } #ifdef SLEEPQUEUE_PROFILING #define SLEEPQ_PROF_LOCATIONS 1024 #define SLEEPQ_SBUFSIZE 512 struct sleepq_prof { LIST_ENTRY(sleepq_prof) sp_link; const char *sp_wmesg; long sp_count; }; LIST_HEAD(sqphead, sleepq_prof); struct sqphead sleepq_prof_free; struct sqphead sleepq_hash[SC_TABLESIZE]; static struct sleepq_prof sleepq_profent[SLEEPQ_PROF_LOCATIONS]; static struct mtx sleepq_prof_lock; MTX_SYSINIT(sleepq_prof_lock, &sleepq_prof_lock, "sleepq_prof", MTX_SPIN); static void sleepq_profile(const char *wmesg) { struct sleepq_prof *sp; mtx_lock_spin(&sleepq_prof_lock); if (prof_enabled == 0) goto unlock; LIST_FOREACH(sp, &sleepq_hash[SC_HASH(wmesg)], sp_link) if (sp->sp_wmesg == wmesg) goto done; sp = LIST_FIRST(&sleepq_prof_free); if (sp == NULL) goto unlock; sp->sp_wmesg = wmesg; LIST_REMOVE(sp, sp_link); LIST_INSERT_HEAD(&sleepq_hash[SC_HASH(wmesg)], sp, sp_link); done: sp->sp_count++; unlock: mtx_unlock_spin(&sleepq_prof_lock); return; } static void sleepq_prof_reset(void) { struct sleepq_prof *sp; int enabled; int i; mtx_lock_spin(&sleepq_prof_lock); enabled = prof_enabled; prof_enabled = 0; for (i = 0; i < SC_TABLESIZE; i++) LIST_INIT(&sleepq_hash[i]); LIST_INIT(&sleepq_prof_free); for (i = 0; i < SLEEPQ_PROF_LOCATIONS; i++) { sp = &sleepq_profent[i]; sp->sp_wmesg = NULL; sp->sp_count = 0; LIST_INSERT_HEAD(&sleepq_prof_free, sp, sp_link); } prof_enabled = enabled; mtx_unlock_spin(&sleepq_prof_lock); } static int enable_sleepq_prof(SYSCTL_HANDLER_ARGS) { int error, v; v = prof_enabled; error = sysctl_handle_int(oidp, &v, v, req); if (error) return (error); if (req->newptr == NULL) return (error); if (v == prof_enabled) return (0); if (v == 1) sleepq_prof_reset(); mtx_lock_spin(&sleepq_prof_lock); prof_enabled = !!v; mtx_unlock_spin(&sleepq_prof_lock); return (0); } static int reset_sleepq_prof_stats(SYSCTL_HANDLER_ARGS) { int error, v; v = 0; error = sysctl_handle_int(oidp, &v, 0, req); if (error) return (error); if (req->newptr == NULL) return (error); if (v == 0) return (0); sleepq_prof_reset(); return (0); } static int dump_sleepq_prof_stats(SYSCTL_HANDLER_ARGS) { struct sleepq_prof *sp; struct sbuf *sb; int enabled; int error; int i; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sb = sbuf_new_for_sysctl(NULL, NULL, SLEEPQ_SBUFSIZE, req); sbuf_printf(sb, "\nwmesg\tcount\n"); enabled = prof_enabled; mtx_lock_spin(&sleepq_prof_lock); prof_enabled = 0; mtx_unlock_spin(&sleepq_prof_lock); for (i = 0; i < SC_TABLESIZE; i++) { LIST_FOREACH(sp, &sleepq_hash[i], sp_link) { sbuf_printf(sb, "%s\t%ld\n", sp->sp_wmesg, sp->sp_count); } } mtx_lock_spin(&sleepq_prof_lock); prof_enabled = enabled; mtx_unlock_spin(&sleepq_prof_lock); error = sbuf_finish(sb); sbuf_delete(sb); return (error); } SYSCTL_PROC(_debug_sleepq, OID_AUTO, stats, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, dump_sleepq_prof_stats, "A", "Sleepqueue profiling statistics"); SYSCTL_PROC(_debug_sleepq, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_RW, NULL, 0, reset_sleepq_prof_stats, "I", "Reset sleepqueue profiling statistics"); SYSCTL_PROC(_debug_sleepq, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW, NULL, 0, enable_sleepq_prof, "I", "Enable sleepqueue profiling"); #endif #ifdef DDB DB_SHOW_COMMAND(sleepq, db_show_sleepqueue) { struct sleepqueue_chain *sc; struct sleepqueue *sq; #ifdef INVARIANTS struct lock_object *lock; #endif struct thread *td; void *wchan; int i; if (!have_addr) return; /* * First, see if there is an active sleep queue for the wait channel * indicated by the address. */ wchan = (void *)addr; sc = SC_LOOKUP(wchan); LIST_FOREACH(sq, &sc->sc_queues, sq_hash) if (sq->sq_wchan == wchan) goto found; /* * Second, see if there is an active sleep queue at the address * indicated. */ for (i = 0; i < SC_TABLESIZE; i++) LIST_FOREACH(sq, &sleepq_chains[i].sc_queues, sq_hash) { if (sq == (struct sleepqueue *)addr) goto found; } db_printf("Unable to locate a sleep queue via %p\n", (void *)addr); return; found: db_printf("Wait channel: %p\n", sq->sq_wchan); db_printf("Queue type: %d\n", sq->sq_type); #ifdef INVARIANTS if (sq->sq_lock) { lock = sq->sq_lock; db_printf("Associated Interlock: %p - (%s) %s\n", lock, LOCK_CLASS(lock)->lc_name, lock->lo_name); } #endif db_printf("Blocked threads:\n"); for (i = 0; i < NR_SLEEPQS; i++) { db_printf("\nQueue[%d]:\n", i); if (TAILQ_EMPTY(&sq->sq_blocked[i])) db_printf("\tempty\n"); else TAILQ_FOREACH(td, &sq->sq_blocked[0], td_slpq) { db_printf("\t%p (tid %d, pid %d, \"%s\")\n", td, td->td_tid, td->td_proc->p_pid, td->td_name); } db_printf("(expected: %u)\n", sq->sq_blockedcnt[i]); } } /* Alias 'show sleepqueue' to 'show sleepq'. */ DB_SHOW_ALIAS(sleepqueue, db_show_sleepqueue); #endif diff --git a/sys/kern/subr_trap.c b/sys/kern/subr_trap.c index bd06f201d117..1f24e8848cc1 100644 --- a/sys/kern/subr_trap.c +++ b/sys/kern/subr_trap.c @@ -1,310 +1,310 @@ /*- * Copyright (C) 1994, David Greenman * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007 The FreeBSD Foundation * * This code is derived from software contributed to Berkeley by * the University of Utah, and William Jolitz. * * Portions of this software were developed by A. Joseph Koshy under * sponsorship from the FreeBSD Foundation and Google, Inc. * * 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. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include "opt_ktrace.h" #include "opt_kdtrace.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #include #endif #include #include #ifdef VIMAGE #include #endif #ifdef XEN #include #include #include #endif #ifdef HWPMC_HOOKS #include #endif #include /* * Define the code needed before returning to user mode, for trap and * syscall. */ void userret(struct thread *td, struct trapframe *frame) { struct proc *p = td->td_proc; #ifdef RACCT int sig; #endif CTR3(KTR_SYSC, "userret: thread %p (pid %d, %s)", td, p->p_pid, td->td_name); KASSERT((p->p_flag & P_WEXIT) == 0, ("Exiting process returns to usermode")); #if 0 #ifdef DIAGNOSTIC /* Check that we called signotify() enough. */ PROC_LOCK(p); thread_lock(td); if (SIGPENDING(td) && ((td->td_flags & TDF_NEEDSIGCHK) == 0 || (td->td_flags & TDF_ASTPENDING) == 0)) printf("failed to set signal flags properly for ast()\n"); thread_unlock(td); PROC_UNLOCK(p); #endif #endif #ifdef KTRACE KTRUSERRET(td); #endif /* * If this thread tickled GEOM, we need to wait for the giggling to * stop before we return to userland */ if (td->td_pflags & TDP_GEOM) g_waitidle(); /* * Charge system time if profiling. */ if (p->p_flag & P_PROFIL) addupc_task(td, TRAPF_PC(frame), td->td_pticks * psratio); /* * Let the scheduler adjust our priority etc. */ sched_userret(td); #ifdef XEN PT_UPDATES_FLUSH(); #endif /* * Check for misbehavior. * * In case there is a callchain tracing ongoing because of * hwpmc(4), skip the scheduler pinning check. * hwpmc(4) subsystem, infact, will collect callchain informations * at ast() checkpoint, which is past userret(). */ WITNESS_WARN(WARN_PANIC, NULL, "userret: returning"); KASSERT(td->td_critnest == 0, ("userret: Returning in a critical section")); KASSERT(td->td_locks == 0, ("userret: Returning with %d locks held", td->td_locks)); KASSERT((td->td_pflags & TDP_NOFAULTING) == 0, ("userret: Returning with pagefaults disabled")); - KASSERT((td->td_pflags & TDP_NOSLEEPING) == 0, + KASSERT(td->td_no_sleeping == 0, ("userret: Returning with sleep disabled")); KASSERT(td->td_pinned == 0 || (td->td_pflags & TDP_CALLCHAIN) != 0, ("userret: Returning with with pinned thread")); KASSERT(td->td_vp_reserv == 0, ("userret: Returning while holding vnode reservation")); KASSERT((td->td_flags & TDF_SBDRY) == 0, ("userret: Returning with stop signals deferred")); #ifdef VIMAGE /* Unfortunately td_vnet_lpush needs VNET_DEBUG. */ VNET_ASSERT(curvnet == NULL, ("%s: Returning on td %p (pid %d, %s) with vnet %p set in %s", __func__, td, p->p_pid, td->td_name, curvnet, (td->td_vnet_lpush != NULL) ? td->td_vnet_lpush : "N/A")); #endif #ifdef RACCT PROC_LOCK(p); while (p->p_throttled == 1) { sig = msleep(p->p_racct, &p->p_mtx, PCATCH | PBDRY, "racct", hz); if ((sig == EINTR) || (sig == ERESTART)) break; } PROC_UNLOCK(p); #endif } /* * Process an asynchronous software trap. * This is relatively easy. * This function will return with preemption disabled. */ void ast(struct trapframe *framep) { struct thread *td; struct proc *p; int flags; int sig; td = curthread; p = td->td_proc; CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid, p->p_comm); KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode")); WITNESS_WARN(WARN_PANIC, NULL, "Returning to user mode"); mtx_assert(&Giant, MA_NOTOWNED); THREAD_LOCK_ASSERT(td, MA_NOTOWNED); td->td_frame = framep; td->td_pticks = 0; /* * This updates the td_flag's for the checks below in one * "atomic" operation with turning off the astpending flag. * If another AST is triggered while we are handling the * AST's saved in flags, the astpending flag will be set and * ast() will be called again. */ thread_lock(td); flags = td->td_flags; td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK | TDF_NEEDRESCHED | TDF_ALRMPEND | TDF_PROFPEND | TDF_MACPEND); thread_unlock(td); PCPU_INC(cnt.v_trap); if (td->td_ucred != p->p_ucred) cred_update_thread(td); if (td->td_pflags & TDP_OWEUPC && p->p_flag & P_PROFIL) { addupc_task(td, td->td_profil_addr, td->td_profil_ticks); td->td_profil_ticks = 0; td->td_pflags &= ~TDP_OWEUPC; } #ifdef HWPMC_HOOKS /* Handle Software PMC callchain capture. */ if (PMC_IS_PENDING_CALLCHAIN(td)) PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_USER_CALLCHAIN_SOFT, (void *) framep); #endif if (flags & TDF_ALRMPEND) { PROC_LOCK(p); kern_psignal(p, SIGVTALRM); PROC_UNLOCK(p); } if (flags & TDF_PROFPEND) { PROC_LOCK(p); kern_psignal(p, SIGPROF); PROC_UNLOCK(p); } #ifdef MAC if (flags & TDF_MACPEND) mac_thread_userret(td); #endif if (flags & TDF_NEEDRESCHED) { #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(1, 1, __func__); #endif thread_lock(td); sched_prio(td, td->td_user_pri); mi_switch(SW_INVOL | SWT_NEEDRESCHED, NULL); thread_unlock(td); #ifdef KTRACE if (KTRPOINT(td, KTR_CSW)) ktrcsw(0, 1, __func__); #endif } /* * Check for signals. Unlocked reads of p_pendingcnt or * p_siglist might cause process-directed signal to be handled * later. */ if (flags & TDF_NEEDSIGCHK || p->p_pendingcnt > 0 || !SIGISEMPTY(p->p_siglist)) { PROC_LOCK(p); mtx_lock(&p->p_sigacts->ps_mtx); while ((sig = cursig(td, SIG_STOP_ALLOWED)) != 0) postsig(sig); mtx_unlock(&p->p_sigacts->ps_mtx); PROC_UNLOCK(p); } /* * We need to check to see if we have to exit or wait due to a * single threading requirement or some other STOP condition. */ if (flags & TDF_NEEDSUSPCHK) { PROC_LOCK(p); thread_suspend_check(0); PROC_UNLOCK(p); } if (td->td_pflags & TDP_OLDMASK) { td->td_pflags &= ~TDP_OLDMASK; kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0); } userret(td, framep); } const char * syscallname(struct proc *p, u_int code) { static const char unknown[] = "unknown"; struct sysentvec *sv; sv = p->p_sysent; if (sv->sv_syscallnames == NULL || code >= sv->sv_size) return (unknown); return (sv->sv_syscallnames[code]); } diff --git a/sys/sys/proc.h b/sys/sys/proc.h index 46f5820e745b..f6cf2e4d6658 100644 --- a/sys/sys/proc.h +++ b/sys/sys/proc.h @@ -1,978 +1,971 @@ /*- * Copyright (c) 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * 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. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)proc.h 8.15 (Berkeley) 5/19/95 * $FreeBSD$ */ #ifndef _SYS_PROC_H_ #define _SYS_PROC_H_ #include /* For struct callout. */ #include /* For struct klist. */ #include #ifndef _KERNEL #include #endif #include #include #include #include #include #include #include /* XXX. */ #include #include #include #include #include #ifndef _KERNEL #include /* For structs itimerval, timeval. */ #else #include #endif #include #include #include /* Machine-dependent proc substruct. */ /* * One structure allocated per session. * * List of locks * (m) locked by s_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct session { u_int s_count; /* Ref cnt; pgrps in session - atomic. */ struct proc *s_leader; /* (m + e) Session leader. */ struct vnode *s_ttyvp; /* (m) Vnode of controlling tty. */ struct cdev_priv *s_ttydp; /* (m) Device of controlling tty. */ struct tty *s_ttyp; /* (e) Controlling tty. */ pid_t s_sid; /* (c) Session ID. */ /* (m) Setlogin() name: */ char s_login[roundup(MAXLOGNAME, sizeof(long))]; struct mtx s_mtx; /* Mutex to protect members. */ }; /* * One structure allocated per process group. * * List of locks * (m) locked by pg_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct pgrp { LIST_ENTRY(pgrp) pg_hash; /* (e) Hash chain. */ LIST_HEAD(, proc) pg_members; /* (m + e) Pointer to pgrp members. */ struct session *pg_session; /* (c) Pointer to session. */ struct sigiolst pg_sigiolst; /* (m) List of sigio sources. */ pid_t pg_id; /* (c) Process group id. */ int pg_jobc; /* (m) Job control process count. */ struct mtx pg_mtx; /* Mutex to protect members */ }; /* * pargs, used to hold a copy of the command line, if it had a sane length. */ struct pargs { u_int ar_ref; /* Reference count. */ u_int ar_length; /* Length. */ u_char ar_args[1]; /* Arguments. */ }; /*- * Description of a process. * * This structure contains the information needed to manage a thread of * control, known in UN*X as a process; it has references to substructures * containing descriptions of things that the process uses, but may share * with related processes. The process structure and the substructures * are always addressable except for those marked "(CPU)" below, * which might be addressable only on a processor on which the process * is running. * * Below is a key of locks used to protect each member of struct proc. The * lock is indicated by a reference to a specific character in parens in the * associated comment. * * - not yet protected * a - only touched by curproc or parent during fork/wait * b - created at fork, never changes * (exception aiods switch vmspaces, but they are also * marked 'P_SYSTEM' so hopefully it will be left alone) * c - locked by proc mtx * d - locked by allproc_lock lock * e - locked by proctree_lock lock * f - session mtx * g - process group mtx * h - callout_lock mtx * i - by curproc or the master session mtx * j - locked by proc slock * k - only accessed by curthread * k*- only accessed by curthread and from an interrupt * l - the attaching proc or attaching proc parent * m - Giant * n - not locked, lazy * o - ktrace lock * q - td_contested lock * r - p_peers lock * t - thread lock * x - created at fork, only changes during single threading in exec * y - created at first aio, doesn't change until exit or exec at which * point we are single-threaded and only curthread changes it * z - zombie threads lock * * If the locking key specifies two identifiers (for example, p_pptr) then * either lock is sufficient for read access, but both locks must be held * for write access. */ struct cpuset; struct kaioinfo; struct kaudit_record; struct kdtrace_proc; struct kdtrace_thread; struct mqueue_notifier; struct nlminfo; struct p_sched; struct proc; struct procdesc; struct racct; struct sbuf; struct sleepqueue; struct td_sched; struct thread; struct trapframe; struct turnstile; /* * XXX: Does this belong in resource.h or resourcevar.h instead? * Resource usage extension. The times in rusage structs in the kernel are * never up to date. The actual times are kept as runtimes and tick counts * (with control info in the "previous" times), and are converted when * userland asks for rusage info. Backwards compatibility prevents putting * this directly in the user-visible rusage struct. * * Locking for p_rux: (cj) means (j) for p_rux and (c) for p_crux. * Locking for td_rux: (t) for all fields. */ struct rusage_ext { uint64_t rux_runtime; /* (cj) Real time. */ uint64_t rux_uticks; /* (cj) Statclock hits in user mode. */ uint64_t rux_sticks; /* (cj) Statclock hits in sys mode. */ uint64_t rux_iticks; /* (cj) Statclock hits in intr mode. */ uint64_t rux_uu; /* (c) Previous user time in usec. */ uint64_t rux_su; /* (c) Previous sys time in usec. */ uint64_t rux_tu; /* (c) Previous total time in usec. */ }; /* * Kernel runnable context (thread). * This is what is put to sleep and reactivated. * Thread context. Processes may have multiple threads. */ struct thread { struct mtx *volatile td_lock; /* replaces sched lock */ struct proc *td_proc; /* (*) Associated process. */ TAILQ_ENTRY(thread) td_plist; /* (*) All threads in this proc. */ TAILQ_ENTRY(thread) td_runq; /* (t) Run queue. */ TAILQ_ENTRY(thread) td_slpq; /* (t) Sleep queue. */ TAILQ_ENTRY(thread) td_lockq; /* (t) Lock queue. */ LIST_ENTRY(thread) td_hash; /* (d) Hash chain. */ struct cpuset *td_cpuset; /* (t) CPU affinity mask. */ struct seltd *td_sel; /* Select queue/channel. */ struct sleepqueue *td_sleepqueue; /* (k) Associated sleep queue. */ struct turnstile *td_turnstile; /* (k) Associated turnstile. */ struct rl_q_entry *td_rlqe; /* (k) Associated range lock entry. */ struct umtx_q *td_umtxq; /* (c?) Link for when we're blocked. */ lwpid_t td_tid; /* (b) Thread ID. */ sigqueue_t td_sigqueue; /* (c) Sigs arrived, not delivered. */ #define td_siglist td_sigqueue.sq_signals u_char td_lend_user_pri; /* (t) Lend user pri. */ /* Cleared during fork1() */ #define td_startzero td_flags int td_flags; /* (t) TDF_* flags. */ int td_inhibitors; /* (t) Why can not run. */ int td_pflags; /* (k) Private thread (TDP_*) flags. */ int td_dupfd; /* (k) Ret value from fdopen. XXX */ int td_sqqueue; /* (t) Sleepqueue queue blocked on. */ void *td_wchan; /* (t) Sleep address. */ const char *td_wmesg; /* (t) Reason for sleep. */ u_char td_lastcpu; /* (t) Last cpu we were on. */ u_char td_oncpu; /* (t) Which cpu we are on. */ volatile u_char td_owepreempt; /* (k*) Preempt on last critical_exit */ u_char td_tsqueue; /* (t) Turnstile queue blocked on. */ short td_locks; /* (k) Count of non-spin locks. */ short td_rw_rlocks; /* (k) Count of rwlock read locks. */ short td_lk_slocks; /* (k) Count of lockmgr shared locks. */ short td_stopsched; /* (k) Scheduler stopped. */ struct turnstile *td_blocked; /* (t) Lock thread is blocked on. */ const char *td_lockname; /* (t) Name of lock blocked on. */ LIST_HEAD(, turnstile) td_contested; /* (q) Contested locks. */ struct lock_list_entry *td_sleeplocks; /* (k) Held sleep locks. */ int td_intr_nesting_level; /* (k) Interrupt recursion. */ int td_pinned; /* (k) Temporary cpu pin count. */ struct ucred *td_ucred; /* (k) Reference to credentials. */ u_int td_estcpu; /* (t) estimated cpu utilization */ int td_slptick; /* (t) Time at sleep. */ int td_blktick; /* (t) Time spent blocked. */ int td_swvoltick; /* (t) Time at last SW_VOL switch. */ u_int td_cow; /* (*) Number of copy-on-write faults */ struct rusage td_ru; /* (t) rusage information. */ struct rusage_ext td_rux; /* (t) Internal rusage information. */ uint64_t td_incruntime; /* (t) Cpu ticks to transfer to proc. */ uint64_t td_runtime; /* (t) How many cpu ticks we've run. */ u_int td_pticks; /* (t) Statclock hits for profiling */ u_int td_sticks; /* (t) Statclock hits in system mode. */ u_int td_iticks; /* (t) Statclock hits in intr mode. */ u_int td_uticks; /* (t) Statclock hits in user mode. */ int td_intrval; /* (t) Return value for sleepq. */ sigset_t td_oldsigmask; /* (k) Saved mask from pre sigpause. */ volatile u_int td_generation; /* (k) For detection of preemption */ stack_t td_sigstk; /* (k) Stack ptr and on-stack flag. */ int td_xsig; /* (c) Signal for ptrace */ u_long td_profil_addr; /* (k) Temporary addr until AST. */ u_int td_profil_ticks; /* (k) Temporary ticks until AST. */ char td_name[MAXCOMLEN + 1]; /* (*) Thread name. */ struct file *td_fpop; /* (k) file referencing cdev under op */ int td_dbgflags; /* (c) Userland debugger flags */ struct ksiginfo td_dbgksi; /* (c) ksi reflected to debugger. */ int td_ng_outbound; /* (k) Thread entered ng from above. */ struct osd td_osd; /* (k) Object specific data. */ struct vm_map_entry *td_map_def_user; /* (k) Deferred entries. */ pid_t td_dbg_forked; /* (c) Child pid for debugger. */ u_int td_vp_reserv; /* (k) Count of reserved vnodes. */ + int td_no_sleeping; /* (k) Sleeping disabled count. */ #define td_endzero td_sigmask /* Copied during fork1() or create_thread(). */ #define td_startcopy td_endzero sigset_t td_sigmask; /* (c) Current signal mask. */ u_char td_rqindex; /* (t) Run queue index. */ u_char td_base_pri; /* (t) Thread base kernel priority. */ u_char td_priority; /* (t) Thread active priority. */ u_char td_pri_class; /* (t) Scheduling class. */ u_char td_user_pri; /* (t) User pri from estcpu and nice. */ u_char td_base_user_pri; /* (t) Base user pri */ #define td_endcopy td_pcb /* * Fields that must be manually set in fork1() or create_thread() * or already have been set in the allocator, constructor, etc. */ struct pcb *td_pcb; /* (k) Kernel VA of pcb and kstack. */ enum { TDS_INACTIVE = 0x0, TDS_INHIBITED, TDS_CAN_RUN, TDS_RUNQ, TDS_RUNNING } td_state; /* (t) thread state */ register_t td_retval[2]; /* (k) Syscall aux returns. */ struct callout td_slpcallout; /* (h) Callout for sleep. */ struct trapframe *td_frame; /* (k) */ struct vm_object *td_kstack_obj;/* (a) Kstack object. */ vm_offset_t td_kstack; /* (a) Kernel VA of kstack. */ int td_kstack_pages; /* (a) Size of the kstack. */ volatile u_int td_critnest; /* (k*) Critical section nest level. */ struct mdthread td_md; /* (k) Any machine-dependent fields. */ struct td_sched *td_sched; /* (*) Scheduler-specific data. */ struct kaudit_record *td_ar; /* (k) Active audit record, if any. */ struct lpohead td_lprof[2]; /* (a) lock profiling objects. */ struct kdtrace_thread *td_dtrace; /* (*) DTrace-specific data. */ int td_errno; /* Error returned by last syscall. */ struct vnet *td_vnet; /* (k) Effective vnet. */ const char *td_vnet_lpush; /* (k) Debugging vnet push / pop. */ struct trapframe *td_intr_frame;/* (k) Frame of the current irq */ struct proc *td_rfppwait_p; /* (k) The vforked child */ struct vm_page **td_ma; /* (k) uio pages held */ int td_ma_cnt; /* (k) size of *td_ma */ }; struct mtx *thread_lock_block(struct thread *); void thread_lock_unblock(struct thread *, struct mtx *); void thread_lock_set(struct thread *, struct mtx *); #define THREAD_LOCK_ASSERT(td, type) \ do { \ struct mtx *__m = (td)->td_lock; \ if (__m != &blocked_lock) \ mtx_assert(__m, (type)); \ } while (0) #ifdef INVARIANTS #define THREAD_LOCKPTR_ASSERT(td, lock) \ do { \ struct mtx *__m = (td)->td_lock; \ KASSERT((__m == &blocked_lock || __m == (lock)), \ ("Thread %p lock %p does not match %p", td, __m, (lock))); \ } while (0) #else #define THREAD_LOCKPTR_ASSERT(td, lock) #endif #define CRITICAL_ASSERT(td) \ KASSERT((td)->td_critnest >= 1, ("Not in critical section")); /* * Flags kept in td_flags: * To change these you MUST have the scheduler lock. */ #define TDF_BORROWING 0x00000001 /* Thread is borrowing pri from another. */ #define TDF_INPANIC 0x00000002 /* Caused a panic, let it drive crashdump. */ #define TDF_INMEM 0x00000004 /* Thread's stack is in memory. */ #define TDF_SINTR 0x00000008 /* Sleep is interruptible. */ #define TDF_TIMEOUT 0x00000010 /* Timing out during sleep. */ #define TDF_IDLETD 0x00000020 /* This is a per-CPU idle thread. */ #define TDF_CANSWAP 0x00000040 /* Thread can be swapped. */ #define TDF_SLEEPABORT 0x00000080 /* sleepq_abort was called. */ #define TDF_KTH_SUSP 0x00000100 /* kthread is suspended */ #define TDF_UNUSED09 0x00000200 /* --available-- */ #define TDF_BOUNDARY 0x00000400 /* Thread suspended at user boundary */ #define TDF_ASTPENDING 0x00000800 /* Thread has some asynchronous events. */ #define TDF_TIMOFAIL 0x00001000 /* Timeout from sleep after we were awake. */ #define TDF_SBDRY 0x00002000 /* Stop only on usermode boundary. */ #define TDF_UPIBLOCKED 0x00004000 /* Thread blocked on user PI mutex. */ #define TDF_NEEDSUSPCHK 0x00008000 /* Thread may need to suspend. */ #define TDF_NEEDRESCHED 0x00010000 /* Thread needs to yield. */ #define TDF_NEEDSIGCHK 0x00020000 /* Thread may need signal delivery. */ #define TDF_NOLOAD 0x00040000 /* Ignore during load avg calculations. */ #define TDF_UNUSED19 0x00080000 /* --available-- */ #define TDF_THRWAKEUP 0x00100000 /* Libthr thread must not suspend itself. */ #define TDF_UNUSED21 0x00200000 /* --available-- */ #define TDF_SWAPINREQ 0x00400000 /* Swapin request due to wakeup. */ #define TDF_UNUSED23 0x00800000 /* --available-- */ #define TDF_SCHED0 0x01000000 /* Reserved for scheduler private use */ #define TDF_SCHED1 0x02000000 /* Reserved for scheduler private use */ #define TDF_SCHED2 0x04000000 /* Reserved for scheduler private use */ #define TDF_SCHED3 0x08000000 /* Reserved for scheduler private use */ #define TDF_ALRMPEND 0x10000000 /* Pending SIGVTALRM needs to be posted. */ #define TDF_PROFPEND 0x20000000 /* Pending SIGPROF needs to be posted. */ #define TDF_MACPEND 0x40000000 /* AST-based MAC event pending. */ /* Userland debug flags */ #define TDB_SUSPEND 0x00000001 /* Thread is suspended by debugger */ #define TDB_XSIG 0x00000002 /* Thread is exchanging signal under trace */ #define TDB_USERWR 0x00000004 /* Debugger modified memory or registers */ #define TDB_SCE 0x00000008 /* Thread performs syscall enter */ #define TDB_SCX 0x00000010 /* Thread performs syscall exit */ #define TDB_EXEC 0x00000020 /* TDB_SCX from exec(2) family */ #define TDB_FORK 0x00000040 /* TDB_SCX from fork(2) that created new process */ #define TDB_STOPATFORK 0x00000080 /* Stop at the return from fork (child only) */ #define TDB_CHILD 0x00000100 /* New child indicator for ptrace() */ /* * "Private" flags kept in td_pflags: * These are only written by curthread and thus need no locking. */ #define TDP_OLDMASK 0x00000001 /* Need to restore mask after suspend. */ #define TDP_INKTR 0x00000002 /* Thread is currently in KTR code. */ #define TDP_INKTRACE 0x00000004 /* Thread is currently in KTRACE code. */ #define TDP_BUFNEED 0x00000008 /* Do not recurse into the buf flush */ #define TDP_COWINPROGRESS 0x00000010 /* Snapshot copy-on-write in progress. */ #define TDP_ALTSTACK 0x00000020 /* Have alternate signal stack. */ #define TDP_DEADLKTREAT 0x00000040 /* Lock aquisition - deadlock treatment. */ #define TDP_NOFAULTING 0x00000080 /* Do not handle page faults. */ -#define TDP_NOSLEEPING 0x00000100 /* Thread is not allowed to sleep on a sq. */ +#define TDP_UNUSED9 0x00000100 /* --available-- */ #define TDP_OWEUPC 0x00000200 /* Call addupc() at next AST. */ #define TDP_ITHREAD 0x00000400 /* Thread is an interrupt thread. */ #define TDP_SYNCIO 0x00000800 /* Local override, disable async i/o. */ #define TDP_SCHED1 0x00001000 /* Reserved for scheduler private use */ #define TDP_SCHED2 0x00002000 /* Reserved for scheduler private use */ #define TDP_SCHED3 0x00004000 /* Reserved for scheduler private use */ #define TDP_SCHED4 0x00008000 /* Reserved for scheduler private use */ #define TDP_GEOM 0x00010000 /* Settle GEOM before finishing syscall */ #define TDP_SOFTDEP 0x00020000 /* Stuck processing softdep worklist */ #define TDP_NORUNNINGBUF 0x00040000 /* Ignore runningbufspace check */ #define TDP_WAKEUP 0x00080000 /* Don't sleep in umtx cond_wait */ #define TDP_INBDFLUSH 0x00100000 /* Already in BO_BDFLUSH, do not recurse */ #define TDP_KTHREAD 0x00200000 /* This is an official kernel thread */ #define TDP_CALLCHAIN 0x00400000 /* Capture thread's callchain */ #define TDP_IGNSUSP 0x00800000 /* Permission to ignore the MNTK_SUSPEND* */ #define TDP_AUDITREC 0x01000000 /* Audit record pending on thread */ #define TDP_RFPPWAIT 0x02000000 /* Handle RFPPWAIT on syscall exit */ #define TDP_RESETSPUR 0x04000000 /* Reset spurious page fault history. */ #define TDP_NERRNO 0x08000000 /* Last errno is already in td_errno */ #define TDP_UIOHELD 0x10000000 /* Current uio has pages held in td_ma */ /* * Reasons that the current thread can not be run yet. * More than one may apply. */ #define TDI_SUSPENDED 0x0001 /* On suspension queue. */ #define TDI_SLEEPING 0x0002 /* Actually asleep! (tricky). */ #define TDI_SWAPPED 0x0004 /* Stack not in mem. Bad juju if run. */ #define TDI_LOCK 0x0008 /* Stopped on a lock. */ #define TDI_IWAIT 0x0010 /* Awaiting interrupt. */ #define TD_IS_SLEEPING(td) ((td)->td_inhibitors & TDI_SLEEPING) #define TD_ON_SLEEPQ(td) ((td)->td_wchan != NULL) #define TD_IS_SUSPENDED(td) ((td)->td_inhibitors & TDI_SUSPENDED) #define TD_IS_SWAPPED(td) ((td)->td_inhibitors & TDI_SWAPPED) #define TD_ON_LOCK(td) ((td)->td_inhibitors & TDI_LOCK) #define TD_AWAITING_INTR(td) ((td)->td_inhibitors & TDI_IWAIT) #define TD_IS_RUNNING(td) ((td)->td_state == TDS_RUNNING) #define TD_ON_RUNQ(td) ((td)->td_state == TDS_RUNQ) #define TD_CAN_RUN(td) ((td)->td_state == TDS_CAN_RUN) #define TD_IS_INHIBITED(td) ((td)->td_state == TDS_INHIBITED) #define TD_ON_UPILOCK(td) ((td)->td_flags & TDF_UPIBLOCKED) #define TD_IS_IDLETHREAD(td) ((td)->td_flags & TDF_IDLETD) #define TD_SET_INHIB(td, inhib) do { \ (td)->td_state = TDS_INHIBITED; \ (td)->td_inhibitors |= (inhib); \ } while (0) #define TD_CLR_INHIB(td, inhib) do { \ if (((td)->td_inhibitors & (inhib)) && \ (((td)->td_inhibitors &= ~(inhib)) == 0)) \ (td)->td_state = TDS_CAN_RUN; \ } while (0) #define TD_SET_SLEEPING(td) TD_SET_INHIB((td), TDI_SLEEPING) #define TD_SET_SWAPPED(td) TD_SET_INHIB((td), TDI_SWAPPED) #define TD_SET_LOCK(td) TD_SET_INHIB((td), TDI_LOCK) #define TD_SET_SUSPENDED(td) TD_SET_INHIB((td), TDI_SUSPENDED) #define TD_SET_IWAIT(td) TD_SET_INHIB((td), TDI_IWAIT) #define TD_SET_EXITING(td) TD_SET_INHIB((td), TDI_EXITING) #define TD_CLR_SLEEPING(td) TD_CLR_INHIB((td), TDI_SLEEPING) #define TD_CLR_SWAPPED(td) TD_CLR_INHIB((td), TDI_SWAPPED) #define TD_CLR_LOCK(td) TD_CLR_INHIB((td), TDI_LOCK) #define TD_CLR_SUSPENDED(td) TD_CLR_INHIB((td), TDI_SUSPENDED) #define TD_CLR_IWAIT(td) TD_CLR_INHIB((td), TDI_IWAIT) #define TD_SET_RUNNING(td) (td)->td_state = TDS_RUNNING #define TD_SET_RUNQ(td) (td)->td_state = TDS_RUNQ #define TD_SET_CAN_RUN(td) (td)->td_state = TDS_CAN_RUN /* * Process structure. */ struct proc { LIST_ENTRY(proc) p_list; /* (d) List of all processes. */ TAILQ_HEAD(, thread) p_threads; /* (c) all threads. */ struct mtx p_slock; /* process spin lock */ struct ucred *p_ucred; /* (c) Process owner's identity. */ struct filedesc *p_fd; /* (b) Open files. */ struct filedesc_to_leader *p_fdtol; /* (b) Tracking node */ struct pstats *p_stats; /* (b) Accounting/statistics (CPU). */ struct plimit *p_limit; /* (c) Process limits. */ struct callout p_limco; /* (c) Limit callout handle */ struct sigacts *p_sigacts; /* (x) Signal actions, state (CPU). */ /* * The following don't make too much sense. * See the td_ or ke_ versions of the same flags. */ int p_flag; /* (c) P_* flags. */ enum { PRS_NEW = 0, /* In creation */ PRS_NORMAL, /* threads can be run. */ PRS_ZOMBIE } p_state; /* (j/c) Process status. */ pid_t p_pid; /* (b) Process identifier. */ LIST_ENTRY(proc) p_hash; /* (d) Hash chain. */ LIST_ENTRY(proc) p_pglist; /* (g + e) List of processes in pgrp. */ struct proc *p_pptr; /* (c + e) Pointer to parent process. */ LIST_ENTRY(proc) p_sibling; /* (e) List of sibling processes. */ LIST_HEAD(, proc) p_children; /* (e) Pointer to list of children. */ struct mtx p_mtx; /* (n) Lock for this struct. */ struct ksiginfo *p_ksi; /* Locked by parent proc lock */ sigqueue_t p_sigqueue; /* (c) Sigs not delivered to a td. */ #define p_siglist p_sigqueue.sq_signals /* The following fields are all zeroed upon creation in fork. */ #define p_startzero p_oppid pid_t p_oppid; /* (c + e) Save ppid in ptrace. XXX */ struct vmspace *p_vmspace; /* (b) Address space. */ u_int p_swtick; /* (c) Tick when swapped in or out. */ struct itimerval p_realtimer; /* (c) Alarm timer. */ struct rusage p_ru; /* (a) Exit information. */ struct rusage_ext p_rux; /* (cj) Internal resource usage. */ struct rusage_ext p_crux; /* (c) Internal child resource usage. */ int p_profthreads; /* (c) Num threads in addupc_task. */ volatile int p_exitthreads; /* (j) Number of threads exiting */ int p_traceflag; /* (o) Kernel trace points. */ struct vnode *p_tracevp; /* (c + o) Trace to vnode. */ struct ucred *p_tracecred; /* (o) Credentials to trace with. */ struct vnode *p_textvp; /* (b) Vnode of executable. */ u_int p_lock; /* (c) Proclock (prevent swap) count. */ struct sigiolst p_sigiolst; /* (c) List of sigio sources. */ int p_sigparent; /* (c) Signal to parent on exit. */ int p_sig; /* (n) For core dump/debugger XXX. */ u_long p_code; /* (n) For core dump/debugger XXX. */ u_int p_stops; /* (c) Stop event bitmask. */ u_int p_stype; /* (c) Stop event type. */ char p_step; /* (c) Process is stopped. */ u_char p_pfsflags; /* (c) Procfs flags. */ struct nlminfo *p_nlminfo; /* (?) Only used by/for lockd. */ struct kaioinfo *p_aioinfo; /* (y) ASYNC I/O info. */ struct thread *p_singlethread;/* (c + j) If single threading this is it */ int p_suspcount; /* (j) Num threads in suspended mode. */ struct thread *p_xthread; /* (c) Trap thread */ int p_boundary_count;/* (j) Num threads at user boundary */ int p_pendingcnt; /* how many signals are pending */ struct itimers *p_itimers; /* (c) POSIX interval timers. */ struct procdesc *p_procdesc; /* (e) Process descriptor, if any. */ /* End area that is zeroed on creation. */ #define p_endzero p_magic /* The following fields are all copied upon creation in fork. */ #define p_startcopy p_endzero u_int p_magic; /* (b) Magic number. */ int p_osrel; /* (x) osreldate for the binary (from ELF note, if any) */ char p_comm[MAXCOMLEN + 1]; /* (b) Process name. */ struct pgrp *p_pgrp; /* (c + e) Pointer to process group. */ struct sysentvec *p_sysent; /* (b) Syscall dispatch info. */ struct pargs *p_args; /* (c) Process arguments. */ rlim_t p_cpulimit; /* (c) Current CPU limit in seconds. */ signed char p_nice; /* (c) Process "nice" value. */ int p_fibnum; /* in this routing domain XXX MRT */ /* End area that is copied on creation. */ #define p_endcopy p_xstat u_short p_xstat; /* (c) Exit status; also stop sig. */ struct knlist p_klist; /* (c) Knotes attached to this proc. */ int p_numthreads; /* (c) Number of threads. */ struct mdproc p_md; /* Any machine-dependent fields. */ struct callout p_itcallout; /* (h + c) Interval timer callout. */ u_short p_acflag; /* (c) Accounting flags. */ struct proc *p_peers; /* (r) */ struct proc *p_leader; /* (b) */ void *p_emuldata; /* (c) Emulator state data. */ struct label *p_label; /* (*) Proc (not subject) MAC label. */ struct p_sched *p_sched; /* (*) Scheduler-specific data. */ STAILQ_HEAD(, ktr_request) p_ktr; /* (o) KTR event queue. */ LIST_HEAD(, mqueue_notifier) p_mqnotifier; /* (c) mqueue notifiers.*/ struct kdtrace_proc *p_dtrace; /* (*) DTrace-specific data. */ struct cv p_pwait; /* (*) wait cv for exit/exec. */ struct cv p_dbgwait; /* (*) wait cv for debugger attach after fork. */ uint64_t p_prev_runtime; /* (c) Resource usage accounting. */ struct racct *p_racct; /* (b) Resource accounting. */ u_char p_throttled; /* (c) Flag for racct pcpu throttling */ /* * An orphan is the child that has beed re-parented to the * debugger as a result of attaching to it. Need to keep * track of them for parent to be able to collect the exit * status of what used to be children. */ LIST_ENTRY(proc) p_orphan; /* (e) List of orphan processes. */ LIST_HEAD(, proc) p_orphans; /* (e) Pointer to list of orphans. */ }; #define p_session p_pgrp->pg_session #define p_pgid p_pgrp->pg_id #define NOCPU 0xff /* For when we aren't on a CPU. */ #define PROC_SLOCK(p) mtx_lock_spin(&(p)->p_slock) #define PROC_SUNLOCK(p) mtx_unlock_spin(&(p)->p_slock) #define PROC_SLOCK_ASSERT(p, type) mtx_assert(&(p)->p_slock, (type)) /* These flags are kept in p_flag. */ #define P_ADVLOCK 0x00001 /* Process may hold a POSIX advisory lock. */ #define P_CONTROLT 0x00002 /* Has a controlling terminal. */ #define P_KTHREAD 0x00004 /* Kernel thread (*). */ #define P_FOLLOWFORK 0x00008 /* Attach parent debugger to children. */ #define P_PPWAIT 0x00010 /* Parent is waiting for child to exec/exit. */ #define P_PROFIL 0x00020 /* Has started profiling. */ #define P_STOPPROF 0x00040 /* Has thread requesting to stop profiling. */ #define P_HADTHREADS 0x00080 /* Has had threads (no cleanup shortcuts) */ #define P_SUGID 0x00100 /* Had set id privileges since last exec. */ #define P_SYSTEM 0x00200 /* System proc: no sigs, stats or swapping. */ #define P_SINGLE_EXIT 0x00400 /* Threads suspending should exit, not wait. */ #define P_TRACED 0x00800 /* Debugged process being traced. */ #define P_WAITED 0x01000 /* Someone is waiting for us. */ #define P_WEXIT 0x02000 /* Working on exiting. */ #define P_EXEC 0x04000 /* Process called exec. */ #define P_WKILLED 0x08000 /* Killed, go to kernel/user boundary ASAP. */ #define P_CONTINUED 0x10000 /* Proc has continued from a stopped state. */ #define P_STOPPED_SIG 0x20000 /* Stopped due to SIGSTOP/SIGTSTP. */ #define P_STOPPED_TRACE 0x40000 /* Stopped because of tracing. */ #define P_STOPPED_SINGLE 0x80000 /* Only 1 thread can continue (not to user). */ #define P_PROTECTED 0x100000 /* Do not kill on memory overcommit. */ #define P_SIGEVENT 0x200000 /* Process pending signals changed. */ #define P_SINGLE_BOUNDARY 0x400000 /* Threads should suspend at user boundary. */ #define P_HWPMC 0x800000 /* Process is using HWPMCs */ #define P_JAILED 0x1000000 /* Process is in jail. */ #define P_ORPHAN 0x2000000 /* Orphaned. */ #define P_INEXEC 0x4000000 /* Process is in execve(). */ #define P_STATCHILD 0x8000000 /* Child process stopped or exited. */ #define P_INMEM 0x10000000 /* Loaded into memory. */ #define P_SWAPPINGOUT 0x20000000 /* Process is being swapped out. */ #define P_SWAPPINGIN 0x40000000 /* Process is being swapped in. */ #define P_PPTRACE 0x80000000 /* PT_TRACEME by vforked child. */ #define P_STOPPED (P_STOPPED_SIG|P_STOPPED_SINGLE|P_STOPPED_TRACE) #define P_SHOULDSTOP(p) ((p)->p_flag & P_STOPPED) #define P_KILLED(p) ((p)->p_flag & P_WKILLED) /* * These were process status values (p_stat), now they are only used in * legacy conversion code. */ #define SIDL 1 /* Process being created by fork. */ #define SRUN 2 /* Currently runnable. */ #define SSLEEP 3 /* Sleeping on an address. */ #define SSTOP 4 /* Process debugging or suspension. */ #define SZOMB 5 /* Awaiting collection by parent. */ #define SWAIT 6 /* Waiting for interrupt. */ #define SLOCK 7 /* Blocked on a lock. */ #define P_MAGIC 0xbeefface #ifdef _KERNEL /* Types and flags for mi_switch(). */ #define SW_TYPE_MASK 0xff /* First 8 bits are switch type */ #define SWT_NONE 0 /* Unspecified switch. */ #define SWT_PREEMPT 1 /* Switching due to preemption. */ #define SWT_OWEPREEMPT 2 /* Switching due to opepreempt. */ #define SWT_TURNSTILE 3 /* Turnstile contention. */ #define SWT_SLEEPQ 4 /* Sleepq wait. */ #define SWT_SLEEPQTIMO 5 /* Sleepq timeout wait. */ #define SWT_RELINQUISH 6 /* yield call. */ #define SWT_NEEDRESCHED 7 /* NEEDRESCHED was set. */ #define SWT_IDLE 8 /* Switching from the idle thread. */ #define SWT_IWAIT 9 /* Waiting for interrupts. */ #define SWT_SUSPEND 10 /* Thread suspended. */ #define SWT_REMOTEPREEMPT 11 /* Remote processor preempted. */ #define SWT_REMOTEWAKEIDLE 12 /* Remote processor preempted idle. */ #define SWT_COUNT 13 /* Number of switch types. */ /* Flags */ #define SW_VOL 0x0100 /* Voluntary switch. */ #define SW_INVOL 0x0200 /* Involuntary switch. */ #define SW_PREEMPT 0x0400 /* The invol switch is a preemption */ /* How values for thread_single(). */ #define SINGLE_NO_EXIT 0 #define SINGLE_EXIT 1 #define SINGLE_BOUNDARY 2 #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_PARGS); MALLOC_DECLARE(M_PGRP); MALLOC_DECLARE(M_SESSION); MALLOC_DECLARE(M_SUBPROC); #endif #define FOREACH_PROC_IN_SYSTEM(p) \ LIST_FOREACH((p), &allproc, p_list) #define FOREACH_THREAD_IN_PROC(p, td) \ TAILQ_FOREACH((td), &(p)->p_threads, td_plist) #define FIRST_THREAD_IN_PROC(p) TAILQ_FIRST(&(p)->p_threads) /* * We use process IDs <= pid_max <= PID_MAX; PID_MAX + 1 must also fit * in a pid_t, as it is used to represent "no process group". */ #define PID_MAX 99999 #define NO_PID 100000 extern pid_t pid_max; #define SESS_LEADER(p) ((p)->p_session->s_leader == (p)) #define STOPEVENT(p, e, v) do { \ if ((p)->p_stops & (e)) { \ PROC_LOCK(p); \ stopevent((p), (e), (v)); \ PROC_UNLOCK(p); \ } \ } while (0) #define _STOPEVENT(p, e, v) do { \ PROC_LOCK_ASSERT(p, MA_OWNED); \ WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &p->p_mtx.lock_object, \ "checking stopevent %d", (e)); \ if ((p)->p_stops & (e)) \ stopevent((p), (e), (v)); \ } while (0) /* Lock and unlock a process. */ #define PROC_LOCK(p) mtx_lock(&(p)->p_mtx) #define PROC_TRYLOCK(p) mtx_trylock(&(p)->p_mtx) #define PROC_UNLOCK(p) mtx_unlock(&(p)->p_mtx) #define PROC_LOCKED(p) mtx_owned(&(p)->p_mtx) #define PROC_LOCK_ASSERT(p, type) mtx_assert(&(p)->p_mtx, (type)) /* Lock and unlock a process group. */ #define PGRP_LOCK(pg) mtx_lock(&(pg)->pg_mtx) #define PGRP_UNLOCK(pg) mtx_unlock(&(pg)->pg_mtx) #define PGRP_LOCKED(pg) mtx_owned(&(pg)->pg_mtx) #define PGRP_LOCK_ASSERT(pg, type) mtx_assert(&(pg)->pg_mtx, (type)) #define PGRP_LOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_LOCK(pg); \ } while (0) #define PGRP_UNLOCK_PGSIGNAL(pg) do { \ if ((pg) != NULL) \ PGRP_UNLOCK(pg); \ } while (0) /* Lock and unlock a session. */ #define SESS_LOCK(s) mtx_lock(&(s)->s_mtx) #define SESS_UNLOCK(s) mtx_unlock(&(s)->s_mtx) #define SESS_LOCKED(s) mtx_owned(&(s)->s_mtx) #define SESS_LOCK_ASSERT(s, type) mtx_assert(&(s)->s_mtx, (type)) /* Hold process U-area in memory, normally for ptrace/procfs work. */ #define PHOLD(p) do { \ PROC_LOCK(p); \ _PHOLD(p); \ PROC_UNLOCK(p); \ } while (0) #define _PHOLD(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ KASSERT(!((p)->p_flag & P_WEXIT) || (p) == curproc, \ ("PHOLD of exiting process")); \ (p)->p_lock++; \ if (((p)->p_flag & P_INMEM) == 0) \ faultin((p)); \ } while (0) #define PROC_ASSERT_HELD(p) do { \ KASSERT((p)->p_lock > 0, ("process not held")); \ } while (0) #define PRELE(p) do { \ PROC_LOCK((p)); \ _PRELE((p)); \ PROC_UNLOCK((p)); \ } while (0) #define _PRELE(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ PROC_ASSERT_HELD(p); \ (--(p)->p_lock); \ if (((p)->p_flag & P_WEXIT) && (p)->p_lock == 0) \ wakeup(&(p)->p_lock); \ } while (0) #define PROC_ASSERT_NOT_HELD(p) do { \ KASSERT((p)->p_lock == 0, ("process held")); \ } while (0) /* Check whether a thread is safe to be swapped out. */ #define thread_safetoswapout(td) ((td)->td_flags & TDF_CANSWAP) /* Control whether or not it is safe for curthread to sleep. */ -#define THREAD_NO_SLEEPING() do { \ - KASSERT(!(curthread->td_pflags & TDP_NOSLEEPING), \ - ("nested no sleeping")); \ - curthread->td_pflags |= TDP_NOSLEEPING; \ -} while (0) +#define THREAD_NO_SLEEPING() ((curthread)->td_no_sleeping++) -#define THREAD_SLEEPING_OK() do { \ - KASSERT((curthread->td_pflags & TDP_NOSLEEPING), \ - ("nested sleeping ok")); \ - curthread->td_pflags &= ~TDP_NOSLEEPING; \ -} while (0) +#define THREAD_SLEEPING_OK() ((curthread)->td_no_sleeping--) #define PIDHASH(pid) (&pidhashtbl[(pid) & pidhash]) extern LIST_HEAD(pidhashhead, proc) *pidhashtbl; extern u_long pidhash; #define TIDHASH(tid) (&tidhashtbl[(tid) & tidhash]) extern LIST_HEAD(tidhashhead, thread) *tidhashtbl; extern u_long tidhash; extern struct rwlock tidhash_lock; #define PGRPHASH(pgid) (&pgrphashtbl[(pgid) & pgrphash]) extern LIST_HEAD(pgrphashhead, pgrp) *pgrphashtbl; extern u_long pgrphash; extern struct sx allproc_lock; extern struct sx proctree_lock; extern struct mtx ppeers_lock; extern struct proc proc0; /* Process slot for swapper. */ extern struct thread thread0; /* Primary thread in proc0. */ extern struct vmspace vmspace0; /* VM space for proc0. */ extern int hogticks; /* Limit on kernel cpu hogs. */ extern int lastpid; extern int nprocs, maxproc; /* Current and max number of procs. */ extern int maxprocperuid; /* Max procs per uid. */ extern u_long ps_arg_cache_limit; LIST_HEAD(proclist, proc); TAILQ_HEAD(procqueue, proc); TAILQ_HEAD(threadqueue, thread); extern struct proclist allproc; /* List of all processes. */ extern struct proclist zombproc; /* List of zombie processes. */ extern struct proc *initproc, *pageproc; /* Process slots for init, pager. */ extern struct uma_zone *proc_zone; struct proc *pfind(pid_t); /* Find process by id. */ struct proc *pfind_locked(pid_t pid); struct pgrp *pgfind(pid_t); /* Find process group by id. */ struct proc *zpfind(pid_t); /* Find zombie process by id. */ /* * pget() flags. */ #define PGET_HOLD 0x00001 /* Hold the process. */ #define PGET_CANSEE 0x00002 /* Check against p_cansee(). */ #define PGET_CANDEBUG 0x00004 /* Check against p_candebug(). */ #define PGET_ISCURRENT 0x00008 /* Check that the found process is current. */ #define PGET_NOTWEXIT 0x00010 /* Check that the process is not in P_WEXIT. */ #define PGET_NOTINEXEC 0x00020 /* Check that the process is not in P_INEXEC. */ #define PGET_NOTID 0x00040 /* Do not assume tid if pid > PID_MAX. */ #define PGET_WANTREAD (PGET_HOLD | PGET_CANDEBUG | PGET_NOTWEXIT) int pget(pid_t pid, int flags, struct proc **pp); void ast(struct trapframe *framep); struct thread *choosethread(void); int cr_cansignal(struct ucred *cred, struct proc *proc, int signum); int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp, struct session *sess); int enterthispgrp(struct proc *p, struct pgrp *pgrp); void faultin(struct proc *p); void fixjobc(struct proc *p, struct pgrp *pgrp, int entering); int fork1(struct thread *, int, int, struct proc **, int *, int); void fork_exit(void (*)(void *, struct trapframe *), void *, struct trapframe *); void fork_return(struct thread *, struct trapframe *); int inferior(struct proc *p); void kern_yield(int); void kick_proc0(void); int leavepgrp(struct proc *p); int maybe_preempt(struct thread *td); void maybe_yield(void); void mi_switch(int flags, struct thread *newtd); int p_candebug(struct thread *td, struct proc *p); int p_cansee(struct thread *td, struct proc *p); int p_cansched(struct thread *td, struct proc *p); int p_cansignal(struct thread *td, struct proc *p, int signum); int p_canwait(struct thread *td, struct proc *p); struct pargs *pargs_alloc(int len); void pargs_drop(struct pargs *pa); void pargs_hold(struct pargs *pa); int proc_getargv(struct thread *td, struct proc *p, struct sbuf *sb); int proc_getenvv(struct thread *td, struct proc *p, struct sbuf *sb); void procinit(void); void proc_linkup0(struct proc *p, struct thread *td); void proc_linkup(struct proc *p, struct thread *td); void proc_reap(struct thread *td, struct proc *p, int *status, int options); void proc_reparent(struct proc *child, struct proc *newparent); struct pstats *pstats_alloc(void); void pstats_fork(struct pstats *src, struct pstats *dst); void pstats_free(struct pstats *ps); int securelevel_ge(struct ucred *cr, int level); int securelevel_gt(struct ucred *cr, int level); void sess_hold(struct session *); void sess_release(struct session *); int setrunnable(struct thread *); void setsugid(struct proc *p); int should_yield(void); int sigonstack(size_t sp); void sleepinit(void); void stopevent(struct proc *, u_int, u_int); struct thread *tdfind(lwpid_t, pid_t); void threadinit(void); void tidhash_add(struct thread *); void tidhash_remove(struct thread *); void cpu_idle(int); int cpu_idle_wakeup(int); extern void (*cpu_idle_hook)(sbintime_t); /* Hook to machdep CPU idler. */ void cpu_switch(struct thread *, struct thread *, struct mtx *); void cpu_throw(struct thread *, struct thread *) __dead2; void unsleep(struct thread *); void userret(struct thread *, struct trapframe *); void cpu_exit(struct thread *); void exit1(struct thread *, int) __dead2; struct syscall_args; int cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa); void cpu_fork(struct thread *, struct proc *, struct thread *, int); void cpu_set_fork_handler(struct thread *, void (*)(void *), void *); void cpu_set_syscall_retval(struct thread *, int); void cpu_set_upcall(struct thread *td, struct thread *td0); void cpu_set_upcall_kse(struct thread *, void (*)(void *), void *, stack_t *); int cpu_set_user_tls(struct thread *, void *tls_base); void cpu_thread_alloc(struct thread *); void cpu_thread_clean(struct thread *); void cpu_thread_exit(struct thread *); void cpu_thread_free(struct thread *); void cpu_thread_swapin(struct thread *); void cpu_thread_swapout(struct thread *); struct thread *thread_alloc(int pages); int thread_alloc_stack(struct thread *, int pages); void thread_exit(void) __dead2; void thread_free(struct thread *td); void thread_link(struct thread *td, struct proc *p); void thread_reap(void); int thread_single(int how); void thread_single_end(void); void thread_stash(struct thread *td); void thread_stopped(struct proc *p); void childproc_stopped(struct proc *child, int reason); void childproc_continued(struct proc *child); void childproc_exited(struct proc *child); int thread_suspend_check(int how); void thread_suspend_switch(struct thread *); void thread_suspend_one(struct thread *td); void thread_unlink(struct thread *td); void thread_unsuspend(struct proc *p); int thread_unsuspend_one(struct thread *td); void thread_unthread(struct thread *td); void thread_wait(struct proc *p); struct thread *thread_find(struct proc *p, lwpid_t tid); static __inline int curthread_pflags_set(int flags) { struct thread *td; int save; td = curthread; save = ~flags | (td->td_pflags & flags); td->td_pflags |= flags; return (save); } static __inline void curthread_pflags_restore(int save) { curthread->td_pflags &= save; } #endif /* _KERNEL */ #endif /* !_SYS_PROC_H_ */ diff --git a/sys/sys/rmlock.h b/sys/sys/rmlock.h index 0ae20996427d..5a0fa8af3937 100644 --- a/sys/sys/rmlock.h +++ b/sys/sys/rmlock.h @@ -1,127 +1,127 @@ /*- * Copyright (c) 2007 Stephan Uphoff * 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. * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * 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. * * $FreeBSD$ */ #ifndef _SYS_RMLOCK_H_ #define _SYS_RMLOCK_H_ #include #include #include #include #ifdef _KERNEL /* - * Flags passed to rm_init(9). + * Flags passed to rm_init_flags(9). */ #define RM_NOWITNESS 0x00000001 #define RM_RECURSE 0x00000002 #define RM_SLEEPABLE 0x00000004 void rm_init(struct rmlock *rm, const char *name); void rm_init_flags(struct rmlock *rm, const char *name, int opts); void rm_destroy(struct rmlock *rm); int rm_wowned(const struct rmlock *rm); void rm_sysinit(void *arg); void rm_sysinit_flags(void *arg); void _rm_wlock_debug(struct rmlock *rm, const char *file, int line); void _rm_wunlock_debug(struct rmlock *rm, const char *file, int line); int _rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, int trylock, const char *file, int line); void _rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, const char *file, int line); void _rm_wlock(struct rmlock *rm); void _rm_wunlock(struct rmlock *rm); int _rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock); void _rm_runlock(struct rmlock *rm, struct rm_priotracker *tracker); /* * Public interface for lock operations. */ #ifndef LOCK_DEBUG #error LOCK_DEBUG not defined, include before #endif #if LOCK_DEBUG > 0 #define rm_wlock(rm) _rm_wlock_debug((rm), LOCK_FILE, LOCK_LINE) #define rm_wunlock(rm) _rm_wunlock_debug((rm), LOCK_FILE, LOCK_LINE) #define rm_rlock(rm,tracker) \ ((void)_rm_rlock_debug((rm),(tracker), 0, LOCK_FILE, LOCK_LINE )) #define rm_try_rlock(rm,tracker) \ _rm_rlock_debug((rm),(tracker), 1, LOCK_FILE, LOCK_LINE ) #define rm_runlock(rm,tracker) \ _rm_runlock_debug((rm), (tracker), LOCK_FILE, LOCK_LINE ) #else #define rm_wlock(rm) _rm_wlock((rm)) #define rm_wunlock(rm) _rm_wunlock((rm)) #define rm_rlock(rm,tracker) ((void)_rm_rlock((rm),(tracker), 0)) #define rm_try_rlock(rm,tracker) _rm_rlock((rm),(tracker), 1) #define rm_runlock(rm,tracker) _rm_runlock((rm), (tracker)) #endif struct rm_args { struct rmlock *ra_rm; const char *ra_desc; }; struct rm_args_flags { struct rmlock *ra_rm; const char *ra_desc; int ra_opts; }; #define RM_SYSINIT(name, rm, desc) \ static struct rm_args name##_args = { \ (rm), \ (desc), \ }; \ SYSINIT(name##_rm_sysinit, SI_SUB_LOCK, SI_ORDER_MIDDLE, \ rm_sysinit, &name##_args); \ SYSUNINIT(name##_rm_sysuninit, SI_SUB_LOCK, SI_ORDER_MIDDLE, \ rm_destroy, (rm)) #define RM_SYSINIT_FLAGS(name, rm, desc, opts) \ static struct rm_args name##_args = { \ (rm), \ (desc), \ (opts), \ }; \ SYSINIT(name##_rm_sysinit, SI_SUB_LOCK, SI_ORDER_MIDDLE, \ rm_sysinit_flags, &name##_args); \ SYSUNINIT(name##_rm_sysuninit, SI_SUB_LOCK, SI_ORDER_MIDDLE, \ rm_destroy, (rm)) #endif /* _KERNEL */ #endif /* !_SYS_RMLOCK_H_ */