Index: head/sys/dev/hyperv/netvsc/hv_net_vsc.c =================================================================== --- head/sys/dev/hyperv/netvsc/hv_net_vsc.c (revision 295307) +++ head/sys/dev/hyperv/netvsc/hv_net_vsc.c (revision 295308) @@ -1,1052 +1,1041 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2010-2012 Citrix Inc. * Copyright (c) 2012 NetApp Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, 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 ``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 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. * * $FreeBSD$ */ /** * HyperV vmbus network VSC (virtual services client) module * */ #include #include #include #include #include #include #include #include #include #include #include "hv_net_vsc.h" #include "hv_rndis.h" #include "hv_rndis_filter.h" MALLOC_DEFINE(M_NETVSC, "netvsc", "Hyper-V netvsc driver"); /* * Forward declarations */ static void hv_nv_on_channel_callback(void *context); static int hv_nv_init_send_buffer_with_net_vsp(struct hv_device *device); static int hv_nv_init_rx_buffer_with_net_vsp(struct hv_device *device); static int hv_nv_destroy_send_buffer(netvsc_dev *net_dev); static int hv_nv_destroy_rx_buffer(netvsc_dev *net_dev); static int hv_nv_connect_to_vsp(struct hv_device *device); static void hv_nv_on_send_completion(netvsc_dev *net_dev, struct hv_device *device, hv_vm_packet_descriptor *pkt); static void hv_nv_on_receive(netvsc_dev *net_dev, struct hv_device *device, hv_vm_packet_descriptor *pkt); /* * */ static inline netvsc_dev * hv_nv_alloc_net_device(struct hv_device *device) { netvsc_dev *net_dev; hn_softc_t *sc = device_get_softc(device->device); - net_dev = malloc(sizeof(netvsc_dev), M_NETVSC, M_NOWAIT | M_ZERO); - if (net_dev == NULL) { - return (NULL); - } + net_dev = malloc(sizeof(netvsc_dev), M_NETVSC, M_WAITOK | M_ZERO); net_dev->dev = device; net_dev->destroy = FALSE; sc->net_dev = net_dev; return (net_dev); } /* * */ static inline netvsc_dev * hv_nv_get_outbound_net_device(struct hv_device *device) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev;; if ((net_dev != NULL) && net_dev->destroy) { return (NULL); } return (net_dev); } /* * */ static inline netvsc_dev * hv_nv_get_inbound_net_device(struct hv_device *device) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev;; if (net_dev == NULL) { return (net_dev); } /* * When the device is being destroyed; we only * permit incoming packets if and only if there * are outstanding sends. */ if (net_dev->destroy && net_dev->num_outstanding_sends == 0) { return (NULL); } return (net_dev); } int hv_nv_get_next_send_section(netvsc_dev *net_dev) { unsigned long bitsmap_words = net_dev->bitsmap_words; unsigned long *bitsmap = net_dev->send_section_bitsmap; unsigned long idx; int ret = NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX; int i; for (i = 0; i < bitsmap_words; i++) { idx = ffsl(~bitsmap[i]); if (0 == idx) continue; idx--; KASSERT(i * BITS_PER_LONG + idx < net_dev->send_section_count, ("invalid i %d and idx %lu", i, idx)); if (atomic_testandset_long(&bitsmap[i], idx)) continue; ret = i * BITS_PER_LONG + idx; break; } return (ret); } /* * Net VSC initialize receive buffer with net VSP * * Net VSP: Network virtual services client, also known as the * Hyper-V extensible switch and the synthetic data path. */ static int hv_nv_init_rx_buffer_with_net_vsp(struct hv_device *device) { netvsc_dev *net_dev; nvsp_msg *init_pkt; int ret = 0; net_dev = hv_nv_get_outbound_net_device(device); if (!net_dev) { return (ENODEV); } net_dev->rx_buf = contigmalloc(net_dev->rx_buf_size, M_NETVSC, M_ZERO, 0UL, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); /* * Establish the GPADL handle for this buffer on this channel. * Note: This call uses the vmbus connection rather than the * channel to establish the gpadl handle. * GPADL: Guest physical address descriptor list. */ ret = hv_vmbus_channel_establish_gpadl( device->channel, net_dev->rx_buf, net_dev->rx_buf_size, &net_dev->rx_buf_gpadl_handle); if (ret != 0) { goto cleanup; } /* sema_wait(&ext->channel_init_sema); KYS CHECK */ /* Notify the NetVsp of the gpadl handle */ init_pkt = &net_dev->channel_init_packet; memset(init_pkt, 0, sizeof(nvsp_msg)); init_pkt->hdr.msg_type = nvsp_msg_1_type_send_rx_buf; init_pkt->msgs.vers_1_msgs.send_rx_buf.gpadl_handle = net_dev->rx_buf_gpadl_handle; init_pkt->msgs.vers_1_msgs.send_rx_buf.id = NETVSC_RECEIVE_BUFFER_ID; /* Send the gpadl notification request */ ret = hv_vmbus_channel_send_packet(device->channel, init_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)init_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) { goto cleanup; } sema_wait(&net_dev->channel_init_sema); /* Check the response */ if (init_pkt->msgs.vers_1_msgs.send_rx_buf_complete.status != nvsp_status_success) { ret = EINVAL; goto cleanup; } net_dev->rx_section_count = init_pkt->msgs.vers_1_msgs.send_rx_buf_complete.num_sections; net_dev->rx_sections = malloc(net_dev->rx_section_count * - sizeof(nvsp_1_rx_buf_section), M_NETVSC, M_NOWAIT); - if (net_dev->rx_sections == NULL) { - ret = EINVAL; - goto cleanup; - } + sizeof(nvsp_1_rx_buf_section), M_NETVSC, M_WAITOK); memcpy(net_dev->rx_sections, init_pkt->msgs.vers_1_msgs.send_rx_buf_complete.sections, net_dev->rx_section_count * sizeof(nvsp_1_rx_buf_section)); /* * For first release, there should only be 1 section that represents * the entire receive buffer */ if (net_dev->rx_section_count != 1 || net_dev->rx_sections->offset != 0) { ret = EINVAL; goto cleanup; } goto exit; cleanup: hv_nv_destroy_rx_buffer(net_dev); exit: return (ret); } /* * Net VSC initialize send buffer with net VSP */ static int hv_nv_init_send_buffer_with_net_vsp(struct hv_device *device) { netvsc_dev *net_dev; nvsp_msg *init_pkt; int ret = 0; net_dev = hv_nv_get_outbound_net_device(device); if (!net_dev) { return (ENODEV); } net_dev->send_buf = contigmalloc(net_dev->send_buf_size, M_NETVSC, M_ZERO, 0UL, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); if (net_dev->send_buf == NULL) { ret = ENOMEM; goto cleanup; } /* * Establish the gpadl handle for this buffer on this channel. * Note: This call uses the vmbus connection rather than the * channel to establish the gpadl handle. */ ret = hv_vmbus_channel_establish_gpadl(device->channel, net_dev->send_buf, net_dev->send_buf_size, &net_dev->send_buf_gpadl_handle); if (ret != 0) { goto cleanup; } /* Notify the NetVsp of the gpadl handle */ init_pkt = &net_dev->channel_init_packet; memset(init_pkt, 0, sizeof(nvsp_msg)); init_pkt->hdr.msg_type = nvsp_msg_1_type_send_send_buf; init_pkt->msgs.vers_1_msgs.send_rx_buf.gpadl_handle = net_dev->send_buf_gpadl_handle; init_pkt->msgs.vers_1_msgs.send_rx_buf.id = NETVSC_SEND_BUFFER_ID; /* Send the gpadl notification request */ ret = hv_vmbus_channel_send_packet(device->channel, init_pkt, sizeof(nvsp_msg), (uint64_t)init_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) { goto cleanup; } sema_wait(&net_dev->channel_init_sema); /* Check the response */ if (init_pkt->msgs.vers_1_msgs.send_send_buf_complete.status != nvsp_status_success) { ret = EINVAL; goto cleanup; } net_dev->send_section_size = init_pkt->msgs.vers_1_msgs.send_send_buf_complete.section_size; net_dev->send_section_count = net_dev->send_buf_size / net_dev->send_section_size; net_dev->bitsmap_words = howmany(net_dev->send_section_count, BITS_PER_LONG); net_dev->send_section_bitsmap = malloc(net_dev->bitsmap_words * sizeof(long), M_NETVSC, - M_NOWAIT | M_ZERO); - if (NULL == net_dev->send_section_bitsmap) { - ret = ENOMEM; - goto cleanup; - } + M_WAITOK | M_ZERO); goto exit; cleanup: hv_nv_destroy_send_buffer(net_dev); exit: return (ret); } /* * Net VSC destroy receive buffer */ static int hv_nv_destroy_rx_buffer(netvsc_dev *net_dev) { nvsp_msg *revoke_pkt; int ret = 0; /* * If we got a section count, it means we received a * send_rx_buf_complete msg * (ie sent nvsp_msg_1_type_send_rx_buf msg) therefore, * we need to send a revoke msg here */ if (net_dev->rx_section_count) { /* Send the revoke receive buffer */ revoke_pkt = &net_dev->revoke_packet; memset(revoke_pkt, 0, sizeof(nvsp_msg)); revoke_pkt->hdr.msg_type = nvsp_msg_1_type_revoke_rx_buf; revoke_pkt->msgs.vers_1_msgs.revoke_rx_buf.id = NETVSC_RECEIVE_BUFFER_ID; ret = hv_vmbus_channel_send_packet(net_dev->dev->channel, revoke_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)revoke_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0); /* * If we failed here, we might as well return and have a leak * rather than continue and a bugchk */ if (ret != 0) { return (ret); } } /* Tear down the gpadl on the vsp end */ if (net_dev->rx_buf_gpadl_handle) { ret = hv_vmbus_channel_teardown_gpdal(net_dev->dev->channel, net_dev->rx_buf_gpadl_handle); /* * If we failed here, we might as well return and have a leak * rather than continue and a bugchk */ if (ret != 0) { return (ret); } net_dev->rx_buf_gpadl_handle = 0; } if (net_dev->rx_buf) { /* Free up the receive buffer */ contigfree(net_dev->rx_buf, net_dev->rx_buf_size, M_NETVSC); net_dev->rx_buf = NULL; } if (net_dev->rx_sections) { free(net_dev->rx_sections, M_NETVSC); net_dev->rx_sections = NULL; net_dev->rx_section_count = 0; } return (ret); } /* * Net VSC destroy send buffer */ static int hv_nv_destroy_send_buffer(netvsc_dev *net_dev) { nvsp_msg *revoke_pkt; int ret = 0; /* * If we got a section count, it means we received a * send_rx_buf_complete msg * (ie sent nvsp_msg_1_type_send_rx_buf msg) therefore, * we need to send a revoke msg here */ if (net_dev->send_section_size) { /* Send the revoke send buffer */ revoke_pkt = &net_dev->revoke_packet; memset(revoke_pkt, 0, sizeof(nvsp_msg)); revoke_pkt->hdr.msg_type = nvsp_msg_1_type_revoke_send_buf; revoke_pkt->msgs.vers_1_msgs.revoke_send_buf.id = NETVSC_SEND_BUFFER_ID; ret = hv_vmbus_channel_send_packet(net_dev->dev->channel, revoke_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)revoke_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0); /* * If we failed here, we might as well return and have a leak * rather than continue and a bugchk */ if (ret != 0) { return (ret); } } /* Tear down the gpadl on the vsp end */ if (net_dev->send_buf_gpadl_handle) { ret = hv_vmbus_channel_teardown_gpdal(net_dev->dev->channel, net_dev->send_buf_gpadl_handle); /* * If we failed here, we might as well return and have a leak * rather than continue and a bugchk */ if (ret != 0) { return (ret); } net_dev->send_buf_gpadl_handle = 0; } if (net_dev->send_buf) { /* Free up the receive buffer */ contigfree(net_dev->send_buf, net_dev->send_buf_size, M_NETVSC); net_dev->send_buf = NULL; } if (net_dev->send_section_bitsmap) { free(net_dev->send_section_bitsmap, M_NETVSC); } return (ret); } /* * Attempt to negotiate the caller-specified NVSP version * * For NVSP v2, Server 2008 R2 does not set * init_pkt->msgs.init_msgs.init_compl.negotiated_prot_vers * to the negotiated version, so we cannot rely on that. */ static int hv_nv_negotiate_nvsp_protocol(struct hv_device *device, netvsc_dev *net_dev, uint32_t nvsp_ver) { nvsp_msg *init_pkt; int ret; init_pkt = &net_dev->channel_init_packet; memset(init_pkt, 0, sizeof(nvsp_msg)); init_pkt->hdr.msg_type = nvsp_msg_type_init; /* * Specify parameter as the only acceptable protocol version */ init_pkt->msgs.init_msgs.init.p1.protocol_version = nvsp_ver; init_pkt->msgs.init_msgs.init.protocol_version_2 = nvsp_ver; /* Send the init request */ ret = hv_vmbus_channel_send_packet(device->channel, init_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)init_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) return (-1); sema_wait(&net_dev->channel_init_sema); if (init_pkt->msgs.init_msgs.init_compl.status != nvsp_status_success) return (EINVAL); return (0); } /* * Send NDIS version 2 config packet containing MTU. * * Not valid for NDIS version 1. */ static int hv_nv_send_ndis_config(struct hv_device *device, uint32_t mtu) { netvsc_dev *net_dev; nvsp_msg *init_pkt; int ret; net_dev = hv_nv_get_outbound_net_device(device); if (!net_dev) return (-ENODEV); /* * Set up configuration packet, write MTU * Indicate we are capable of handling VLAN tags */ init_pkt = &net_dev->channel_init_packet; memset(init_pkt, 0, sizeof(nvsp_msg)); init_pkt->hdr.msg_type = nvsp_msg_2_type_send_ndis_config; init_pkt->msgs.vers_2_msgs.send_ndis_config.mtu = mtu; init_pkt-> msgs.vers_2_msgs.send_ndis_config.capabilities.u1.u2.ieee8021q = 1; /* Send the configuration packet */ ret = hv_vmbus_channel_send_packet(device->channel, init_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)init_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0); if (ret != 0) return (-EINVAL); return (0); } /* * Net VSC connect to VSP */ static int hv_nv_connect_to_vsp(struct hv_device *device) { netvsc_dev *net_dev; nvsp_msg *init_pkt; uint32_t ndis_version; uint32_t protocol_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2, NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 }; int i; int protocol_number = nitems(protocol_list); int ret = 0; device_t dev = device->device; hn_softc_t *sc = device_get_softc(dev); struct ifnet *ifp = sc->hn_ifp; net_dev = hv_nv_get_outbound_net_device(device); if (!net_dev) { return (ENODEV); } /* * Negotiate the NVSP version. Try the latest NVSP first. */ for (i = protocol_number - 1; i >= 0; i--) { if (hv_nv_negotiate_nvsp_protocol(device, net_dev, protocol_list[i]) == 0) { net_dev->nvsp_version = protocol_list[i]; if (bootverbose) device_printf(dev, "Netvsc: got version 0x%x\n", net_dev->nvsp_version); break; } } if (i < 0) { if (bootverbose) device_printf(dev, "failed to negotiate a valid " "protocol.\n"); return (EPROTO); } /* * Set the MTU if supported by this NVSP protocol version * This needs to be right after the NVSP init message per Haiyang */ if (net_dev->nvsp_version >= NVSP_PROTOCOL_VERSION_2) ret = hv_nv_send_ndis_config(device, ifp->if_mtu); /* * Send the NDIS version */ init_pkt = &net_dev->channel_init_packet; memset(init_pkt, 0, sizeof(nvsp_msg)); if (net_dev->nvsp_version <= NVSP_PROTOCOL_VERSION_4) { ndis_version = NDIS_VERSION_6_1; } else { ndis_version = NDIS_VERSION_6_30; } init_pkt->hdr.msg_type = nvsp_msg_1_type_send_ndis_vers; init_pkt->msgs.vers_1_msgs.send_ndis_vers.ndis_major_vers = (ndis_version & 0xFFFF0000) >> 16; init_pkt->msgs.vers_1_msgs.send_ndis_vers.ndis_minor_vers = ndis_version & 0xFFFF; /* Send the init request */ ret = hv_vmbus_channel_send_packet(device->channel, init_pkt, sizeof(nvsp_msg), (uint64_t)(uintptr_t)init_pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, 0); if (ret != 0) { goto cleanup; } /* * TODO: BUGBUG - We have to wait for the above msg since the netvsp * uses KMCL which acknowledges packet (completion packet) * since our Vmbus always set the * HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED flag */ /* sema_wait(&NetVscChannel->channel_init_sema); */ /* Post the big receive buffer to NetVSP */ if (net_dev->nvsp_version <= NVSP_PROTOCOL_VERSION_2) net_dev->rx_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY; else net_dev->rx_buf_size = NETVSC_RECEIVE_BUFFER_SIZE; net_dev->send_buf_size = NETVSC_SEND_BUFFER_SIZE; ret = hv_nv_init_rx_buffer_with_net_vsp(device); if (ret == 0) ret = hv_nv_init_send_buffer_with_net_vsp(device); cleanup: return (ret); } /* * Net VSC disconnect from VSP */ static void hv_nv_disconnect_from_vsp(netvsc_dev *net_dev) { hv_nv_destroy_rx_buffer(net_dev); hv_nv_destroy_send_buffer(net_dev); } /* * Net VSC on device add * * Callback when the device belonging to this driver is added */ netvsc_dev * hv_nv_on_device_add(struct hv_device *device, void *additional_info) { netvsc_dev *net_dev; int ret = 0; net_dev = hv_nv_alloc_net_device(device); if (!net_dev) goto cleanup; /* Initialize the NetVSC channel extension */ sema_init(&net_dev->channel_init_sema, 0, "netdev_sema"); /* * Open the channel */ ret = hv_vmbus_channel_open(device->channel, NETVSC_DEVICE_RING_BUFFER_SIZE, NETVSC_DEVICE_RING_BUFFER_SIZE, NULL, 0, hv_nv_on_channel_callback, device); if (ret != 0) goto cleanup; /* * Connect with the NetVsp */ ret = hv_nv_connect_to_vsp(device); if (ret != 0) goto close; return (net_dev); close: /* Now, we can close the channel safely */ hv_vmbus_channel_close(device->channel); cleanup: /* * Free the packet buffers on the netvsc device packet queue. * Release other resources. */ if (net_dev) { sema_destroy(&net_dev->channel_init_sema); free(net_dev, M_NETVSC); } return (NULL); } /* * Net VSC on device remove */ int hv_nv_on_device_remove(struct hv_device *device, boolean_t destroy_channel) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev;; /* Stop outbound traffic ie sends and receives completions */ mtx_lock(&device->channel->inbound_lock); net_dev->destroy = TRUE; mtx_unlock(&device->channel->inbound_lock); /* Wait for all send completions */ while (net_dev->num_outstanding_sends) { DELAY(100); } hv_nv_disconnect_from_vsp(net_dev); /* At this point, no one should be accessing net_dev except in here */ /* Now, we can close the channel safely */ if (!destroy_channel) { device->channel->state = HV_CHANNEL_CLOSING_NONDESTRUCTIVE_STATE; } hv_vmbus_channel_close(device->channel); sema_destroy(&net_dev->channel_init_sema); free(net_dev, M_NETVSC); return (0); } /* * Net VSC on send completion */ static void hv_nv_on_send_completion(netvsc_dev *net_dev, struct hv_device *device, hv_vm_packet_descriptor *pkt) { nvsp_msg *nvsp_msg_pkt; netvsc_packet *net_vsc_pkt; nvsp_msg_pkt = (nvsp_msg *)((unsigned long)pkt + (pkt->data_offset8 << 3)); if (nvsp_msg_pkt->hdr.msg_type == nvsp_msg_type_init_complete || nvsp_msg_pkt->hdr.msg_type == nvsp_msg_1_type_send_rx_buf_complete || nvsp_msg_pkt->hdr.msg_type == nvsp_msg_1_type_send_send_buf_complete) { /* Copy the response back */ memcpy(&net_dev->channel_init_packet, nvsp_msg_pkt, sizeof(nvsp_msg)); sema_post(&net_dev->channel_init_sema); } else if (nvsp_msg_pkt->hdr.msg_type == nvsp_msg_1_type_send_rndis_pkt_complete) { /* Get the send context */ net_vsc_pkt = (netvsc_packet *)(unsigned long)pkt->transaction_id; if (NULL != net_vsc_pkt) { if (net_vsc_pkt->send_buf_section_idx != NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) { u_long mask; int idx; idx = net_vsc_pkt->send_buf_section_idx / BITS_PER_LONG; KASSERT(idx < net_dev->bitsmap_words, ("invalid section index %u", net_vsc_pkt->send_buf_section_idx)); mask = 1UL << (net_vsc_pkt->send_buf_section_idx % BITS_PER_LONG); KASSERT(net_dev->send_section_bitsmap[idx] & mask, ("index bitmap 0x%lx, section index %u, " "bitmap idx %d, bitmask 0x%lx", net_dev->send_section_bitsmap[idx], net_vsc_pkt->send_buf_section_idx, idx, mask)); atomic_clear_long( &net_dev->send_section_bitsmap[idx], mask); } /* Notify the layer above us */ net_vsc_pkt->compl.send.on_send_completion( net_vsc_pkt->compl.send.send_completion_context); } atomic_subtract_int(&net_dev->num_outstanding_sends, 1); } } /* * Net VSC on send * Sends a packet on the specified Hyper-V device. * Returns 0 on success, non-zero on failure. */ int hv_nv_on_send(struct hv_device *device, netvsc_packet *pkt) { netvsc_dev *net_dev; nvsp_msg send_msg; int ret; net_dev = hv_nv_get_outbound_net_device(device); if (!net_dev) return (ENODEV); send_msg.hdr.msg_type = nvsp_msg_1_type_send_rndis_pkt; if (pkt->is_data_pkt) { /* 0 is RMC_DATA */ send_msg.msgs.vers_1_msgs.send_rndis_pkt.chan_type = 0; } else { /* 1 is RMC_CONTROL */ send_msg.msgs.vers_1_msgs.send_rndis_pkt.chan_type = 1; } send_msg.msgs.vers_1_msgs.send_rndis_pkt.send_buf_section_idx = pkt->send_buf_section_idx; send_msg.msgs.vers_1_msgs.send_rndis_pkt.send_buf_section_size = pkt->send_buf_section_size; if (pkt->page_buf_count) { ret = hv_vmbus_channel_send_packet_pagebuffer(device->channel, pkt->page_buffers, pkt->page_buf_count, &send_msg, sizeof(nvsp_msg), (uint64_t)(uintptr_t)pkt); } else { ret = hv_vmbus_channel_send_packet(device->channel, &send_msg, sizeof(nvsp_msg), (uint64_t)(uintptr_t)pkt, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); } /* Record outstanding send only if send_packet() succeeded */ if (ret == 0) atomic_add_int(&net_dev->num_outstanding_sends, 1); return (ret); } /* * Net VSC on receive * * In the FreeBSD Hyper-V virtual world, this function deals exclusively * with virtual addresses. */ static void hv_nv_on_receive(netvsc_dev *net_dev, struct hv_device *device, hv_vm_packet_descriptor *pkt) { hv_vm_transfer_page_packet_header *vm_xfer_page_pkt; nvsp_msg *nvsp_msg_pkt; netvsc_packet vsc_pkt; netvsc_packet *net_vsc_pkt = &vsc_pkt; device_t dev = device->device; int count = 0; int i = 0; int status = nvsp_status_success; /* * All inbound packets other than send completion should be * xfer page packet. */ if (pkt->type != HV_VMBUS_PACKET_TYPE_DATA_USING_TRANSFER_PAGES) { device_printf(dev, "packet type %d is invalid!\n", pkt->type); return; } nvsp_msg_pkt = (nvsp_msg *)((unsigned long)pkt + (pkt->data_offset8 << 3)); /* Make sure this is a valid nvsp packet */ if (nvsp_msg_pkt->hdr.msg_type != nvsp_msg_1_type_send_rndis_pkt) { device_printf(dev, "packet hdr type %d is invalid!\n", pkt->type); return; } vm_xfer_page_pkt = (hv_vm_transfer_page_packet_header *)pkt; if (vm_xfer_page_pkt->transfer_page_set_id != NETVSC_RECEIVE_BUFFER_ID) { device_printf(dev, "transfer_page_set_id %d is invalid!\n", vm_xfer_page_pkt->transfer_page_set_id); return; } count = vm_xfer_page_pkt->range_count; net_vsc_pkt->device = device; /* Each range represents 1 RNDIS pkt that contains 1 Ethernet frame */ for (i = 0; i < count; i++) { net_vsc_pkt->status = nvsp_status_success; net_vsc_pkt->data = (void *)((unsigned long)net_dev->rx_buf + vm_xfer_page_pkt->ranges[i].byte_offset); net_vsc_pkt->tot_data_buf_len = vm_xfer_page_pkt->ranges[i].byte_count; hv_rf_on_receive(net_dev, device, net_vsc_pkt); if (net_vsc_pkt->status != nvsp_status_success) { status = nvsp_status_failure; } } /* * Moved completion call back here so that all received * messages (not just data messages) will trigger a response * message back to the host. */ hv_nv_on_receive_completion(device, vm_xfer_page_pkt->d.transaction_id, status); hv_rf_receive_rollup(net_dev); } /* * Net VSC on receive completion * * Send a receive completion packet to RNDIS device (ie NetVsp) */ void hv_nv_on_receive_completion(struct hv_device *device, uint64_t tid, uint32_t status) { nvsp_msg rx_comp_msg; int retries = 0; int ret = 0; rx_comp_msg.hdr.msg_type = nvsp_msg_1_type_send_rndis_pkt_complete; /* Pass in the status */ rx_comp_msg.msgs.vers_1_msgs.send_rndis_pkt_complete.status = status; retry_send_cmplt: /* Send the completion */ ret = hv_vmbus_channel_send_packet(device->channel, &rx_comp_msg, sizeof(nvsp_msg), tid, HV_VMBUS_PACKET_TYPE_COMPLETION, 0); if (ret == 0) { /* success */ /* no-op */ } else if (ret == EAGAIN) { /* no more room... wait a bit and attempt to retry 3 times */ retries++; if (retries < 4) { DELAY(100); goto retry_send_cmplt; } } } /* * Net VSC on channel callback */ static void hv_nv_on_channel_callback(void *context) { struct hv_device *device = (struct hv_device *)context; netvsc_dev *net_dev; device_t dev = device->device; uint32_t bytes_rxed; uint64_t request_id; hv_vm_packet_descriptor *desc; uint8_t *buffer; int bufferlen = NETVSC_PACKET_SIZE; int ret = 0; net_dev = hv_nv_get_inbound_net_device(device); if (net_dev == NULL) return; buffer = net_dev->callback_buf; do { ret = hv_vmbus_channel_recv_packet_raw(device->channel, buffer, bufferlen, &bytes_rxed, &request_id); if (ret == 0) { if (bytes_rxed > 0) { desc = (hv_vm_packet_descriptor *)buffer; switch (desc->type) { case HV_VMBUS_PACKET_TYPE_COMPLETION: hv_nv_on_send_completion(net_dev, device, desc); break; case HV_VMBUS_PACKET_TYPE_DATA_USING_TRANSFER_PAGES: hv_nv_on_receive(net_dev, device, desc); break; default: device_printf(dev, "hv_cb recv unknow type %d " " packet\n", desc->type); break; } } else { break; } } else if (ret == ENOBUFS) { /* Handle large packet */ if (bufferlen > NETVSC_PACKET_SIZE) { free(buffer, M_NETVSC); buffer = NULL; } /* alloc new buffer */ buffer = malloc(bytes_rxed, M_NETVSC, M_NOWAIT); if (buffer == NULL) { device_printf(dev, "hv_cb malloc buffer failed, len=%u\n", bytes_rxed); bufferlen = 0; break; } bufferlen = bytes_rxed; } } while (1); if (bufferlen > NETVSC_PACKET_SIZE) free(buffer, M_NETVSC); hv_rf_channel_rollup(net_dev); } Index: head/sys/dev/hyperv/netvsc/hv_rndis_filter.c =================================================================== --- head/sys/dev/hyperv/netvsc/hv_rndis_filter.c (revision 295307) +++ head/sys/dev/hyperv/netvsc/hv_rndis_filter.c (revision 295308) @@ -1,994 +1,988 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2010-2012 Citrix Inc. * Copyright (c) 2012 NetApp Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, 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 ``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 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. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "hv_net_vsc.h" #include "hv_rndis.h" #include "hv_rndis_filter.h" /* * Forward declarations */ static int hv_rf_send_request(rndis_device *device, rndis_request *request, uint32_t message_type); static void hv_rf_receive_response(rndis_device *device, rndis_msg *response); static void hv_rf_receive_indicate_status(rndis_device *device, rndis_msg *response); static void hv_rf_receive_data(rndis_device *device, rndis_msg *message, netvsc_packet *pkt); static int hv_rf_query_device(rndis_device *device, uint32_t oid, void *result, uint32_t *result_size); static inline int hv_rf_query_device_mac(rndis_device *device); static inline int hv_rf_query_device_link_status(rndis_device *device); static int hv_rf_set_packet_filter(rndis_device *device, uint32_t new_filter); static int hv_rf_init_device(rndis_device *device); static int hv_rf_open_device(rndis_device *device); static int hv_rf_close_device(rndis_device *device); static void hv_rf_on_send_request_completion(void *context); static void hv_rf_on_send_request_halt_completion(void *context); int hv_rf_send_offload_request(struct hv_device *device, rndis_offload_params *offloads); /* * Set the Per-Packet-Info with the specified type */ void * hv_set_rppi_data(rndis_msg *rndis_mesg, uint32_t rppi_size, int pkt_type) { rndis_packet *rndis_pkt; rndis_per_packet_info *rppi; rndis_pkt = &rndis_mesg->msg.packet; rndis_pkt->data_offset += rppi_size; rppi = (rndis_per_packet_info *)((char *)rndis_pkt + rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_length); rppi->size = rppi_size; rppi->type = pkt_type; rppi->per_packet_info_offset = sizeof(rndis_per_packet_info); rndis_pkt->per_pkt_info_length += rppi_size; return (rppi); } /* * Get the Per-Packet-Info with the specified type * return NULL if not found. */ void * hv_get_ppi_data(rndis_packet *rpkt, uint32_t type) { rndis_per_packet_info *ppi; int len; if (rpkt->per_pkt_info_offset == 0) return (NULL); ppi = (rndis_per_packet_info *)((unsigned long)rpkt + rpkt->per_pkt_info_offset); len = rpkt->per_pkt_info_length; while (len > 0) { if (ppi->type == type) return (void *)((unsigned long)ppi + ppi->per_packet_info_offset); len -= ppi->size; ppi = (rndis_per_packet_info *)((unsigned long)ppi + ppi->size); } return (NULL); } /* * Allow module_param to work and override to switch to promiscuous mode. */ static inline rndis_device * hv_get_rndis_device(void) { rndis_device *device; - device = malloc(sizeof(rndis_device), M_NETVSC, M_NOWAIT | M_ZERO); - if (device == NULL) { - return (NULL); - } + device = malloc(sizeof(rndis_device), M_NETVSC, M_WAITOK | M_ZERO); - mtx_init(&device->req_lock, "HV-FRL", NULL, MTX_SPIN | MTX_RECURSE); + mtx_init(&device->req_lock, "HV-FRL", NULL, MTX_DEF); /* Same effect as STAILQ_HEAD_INITIALIZER() static initializer */ STAILQ_INIT(&device->myrequest_list); device->state = RNDIS_DEV_UNINITIALIZED; return (device); } /* * */ static inline void hv_put_rndis_device(rndis_device *device) { mtx_destroy(&device->req_lock); free(device, M_NETVSC); } /* * */ static inline rndis_request * hv_rndis_request(rndis_device *device, uint32_t message_type, uint32_t message_length) { rndis_request *request; rndis_msg *rndis_mesg; rndis_set_request *set; - request = malloc(sizeof(rndis_request), M_NETVSC, M_NOWAIT | M_ZERO); - if (request == NULL) { - return (NULL); - } + request = malloc(sizeof(rndis_request), M_NETVSC, M_WAITOK | M_ZERO); sema_init(&request->wait_sema, 0, "rndis sema"); rndis_mesg = &request->request_msg; rndis_mesg->ndis_msg_type = message_type; rndis_mesg->msg_len = message_length; /* * Set the request id. This field is always after the rndis header * for request/response packet types so we just use the set_request * as a template. */ set = &rndis_mesg->msg.set_request; set->request_id = atomic_fetchadd_int(&device->new_request_id, 1); /* Increment to get the new value (call above returns old value) */ set->request_id += 1; /* Add to the request list */ - mtx_lock_spin(&device->req_lock); + mtx_lock(&device->req_lock); STAILQ_INSERT_TAIL(&device->myrequest_list, request, mylist_entry); - mtx_unlock_spin(&device->req_lock); + mtx_unlock(&device->req_lock); return (request); } /* * */ static inline void hv_put_rndis_request(rndis_device *device, rndis_request *request) { - mtx_lock_spin(&device->req_lock); + mtx_lock(&device->req_lock); /* Fixme: Has O(n) performance */ /* * XXXKYS: Use Doubly linked lists. */ STAILQ_REMOVE(&device->myrequest_list, request, rndis_request_, mylist_entry); - mtx_unlock_spin(&device->req_lock); + mtx_unlock(&device->req_lock); sema_destroy(&request->wait_sema); free(request, M_NETVSC); } /* * */ static int hv_rf_send_request(rndis_device *device, rndis_request *request, uint32_t message_type) { int ret; netvsc_packet *packet; /* Set up the packet to send it */ packet = &request->pkt; packet->is_data_pkt = FALSE; packet->tot_data_buf_len = request->request_msg.msg_len; packet->page_buf_count = 1; packet->page_buffers[0].pfn = hv_get_phys_addr(&request->request_msg) >> PAGE_SHIFT; packet->page_buffers[0].length = request->request_msg.msg_len; packet->page_buffers[0].offset = (unsigned long)&request->request_msg & (PAGE_SIZE - 1); packet->compl.send.send_completion_context = request; /* packet */ if (message_type != REMOTE_NDIS_HALT_MSG) { packet->compl.send.on_send_completion = hv_rf_on_send_request_completion; } else { packet->compl.send.on_send_completion = hv_rf_on_send_request_halt_completion; } packet->compl.send.send_completion_tid = (unsigned long)device; packet->send_buf_section_idx = NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX; packet->send_buf_section_size = 0; ret = hv_nv_on_send(device->net_dev->dev, packet); return (ret); } /* * RNDIS filter receive response */ static void hv_rf_receive_response(rndis_device *device, rndis_msg *response) { rndis_request *request = NULL; rndis_request *next_request; boolean_t found = FALSE; - mtx_lock_spin(&device->req_lock); + mtx_lock(&device->req_lock); request = STAILQ_FIRST(&device->myrequest_list); while (request != NULL) { /* * All request/response message contains request_id as the * first field */ if (request->request_msg.msg.init_request.request_id == response->msg.init_complete.request_id) { found = TRUE; break; } next_request = STAILQ_NEXT(request, mylist_entry); request = next_request; } - mtx_unlock_spin(&device->req_lock); + mtx_unlock(&device->req_lock); if (found) { if (response->msg_len <= sizeof(rndis_msg)) { memcpy(&request->response_msg, response, response->msg_len); } else { if (response->ndis_msg_type == REMOTE_NDIS_RESET_CMPLT) { /* Does not have a request id field */ request->response_msg.msg.reset_complete.status = STATUS_BUFFER_OVERFLOW; } else { request->response_msg.msg.init_complete.status = STATUS_BUFFER_OVERFLOW; } } sema_post(&request->wait_sema); } } int hv_rf_send_offload_request(struct hv_device *device, rndis_offload_params *offloads) { rndis_request *request; rndis_set_request *set; rndis_offload_params *offload_req; rndis_set_complete *set_complete; rndis_device *rndis_dev; hn_softc_t *sc = device_get_softc(device->device); device_t dev = device->device; netvsc_dev *net_dev = sc->net_dev; uint32_t vsp_version = net_dev->nvsp_version; uint32_t extlen = sizeof(rndis_offload_params); int ret; if (vsp_version <= NVSP_PROTOCOL_VERSION_4) { extlen = VERSION_4_OFFLOAD_SIZE; /* On NVSP_PROTOCOL_VERSION_4 and below, we do not support * UDP checksum offload. */ offloads->udp_ipv4_csum = 0; offloads->udp_ipv6_csum = 0; } rndis_dev = net_dev->extension; request = hv_rndis_request(rndis_dev, REMOTE_NDIS_SET_MSG, RNDIS_MESSAGE_SIZE(rndis_set_request) + extlen); if (!request) return (ENOMEM); set = &request->request_msg.msg.set_request; set->oid = RNDIS_OID_TCP_OFFLOAD_PARAMETERS; set->info_buffer_length = extlen; set->info_buffer_offset = sizeof(rndis_set_request); set->device_vc_handle = 0; offload_req = (rndis_offload_params *)((unsigned long)set + set->info_buffer_offset); *offload_req = *offloads; offload_req->header.type = RNDIS_OBJECT_TYPE_DEFAULT; offload_req->header.revision = RNDIS_OFFLOAD_PARAMETERS_REVISION_3; offload_req->header.size = extlen; ret = hv_rf_send_request(rndis_dev, request, REMOTE_NDIS_SET_MSG); if (ret != 0) { device_printf(dev, "hv send offload request failed, ret=%d!\n", ret); goto cleanup; } ret = sema_timedwait(&request->wait_sema, 500); if (ret != 0) { device_printf(dev, "hv send offload request timeout\n"); goto cleanup; } set_complete = &request->response_msg.msg.set_complete; if (set_complete->status == RNDIS_STATUS_SUCCESS) { device_printf(dev, "hv send offload request succeeded\n"); ret = 0; } else { if (set_complete->status == STATUS_NOT_SUPPORTED) { device_printf(dev, "HV Not support offload\n"); ret = 0; } else { ret = set_complete->status; } } cleanup: if (request) hv_put_rndis_request(rndis_dev, request); return (ret); } /* * RNDIS filter receive indicate status */ static void hv_rf_receive_indicate_status(rndis_device *device, rndis_msg *response) { rndis_indicate_status *indicate = &response->msg.indicate_status; switch(indicate->status) { case RNDIS_STATUS_MEDIA_CONNECT: netvsc_linkstatus_callback(device->net_dev->dev, 1); break; case RNDIS_STATUS_MEDIA_DISCONNECT: netvsc_linkstatus_callback(device->net_dev->dev, 0); break; default: /* TODO: */ device_printf(device->net_dev->dev->device, "unknown status %d received\n", indicate->status); break; } } /* * RNDIS filter receive data */ static void hv_rf_receive_data(rndis_device *device, rndis_msg *message, netvsc_packet *pkt) { rndis_packet *rndis_pkt; ndis_8021q_info *rppi_vlan_info; uint32_t data_offset; rndis_tcp_ip_csum_info *csum_info = NULL; device_t dev = device->net_dev->dev->device; rndis_pkt = &message->msg.packet; /* * Fixme: Handle multiple rndis pkt msgs that may be enclosed in this * netvsc packet (ie tot_data_buf_len != message_length) */ /* Remove rndis header, then pass data packet up the stack */ data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset; pkt->tot_data_buf_len -= data_offset; if (pkt->tot_data_buf_len < rndis_pkt->data_length) { pkt->status = nvsp_status_failure; device_printf(dev, "total length %u is less than data length %u\n", pkt->tot_data_buf_len, rndis_pkt->data_length); return; } pkt->tot_data_buf_len = rndis_pkt->data_length; pkt->data = (void *)((unsigned long)pkt->data + data_offset); rppi_vlan_info = hv_get_ppi_data(rndis_pkt, ieee_8021q_info); if (rppi_vlan_info) { pkt->vlan_tci = rppi_vlan_info->u1.s1.vlan_id; } else { pkt->vlan_tci = 0; } csum_info = hv_get_ppi_data(rndis_pkt, tcpip_chksum_info); netvsc_recv(device->net_dev->dev, pkt, csum_info); } /* * RNDIS filter on receive */ int hv_rf_on_receive(netvsc_dev *net_dev, struct hv_device *device, netvsc_packet *pkt) { rndis_device *rndis_dev; rndis_msg *rndis_hdr; /* Make sure the rndis device state is initialized */ if (net_dev->extension == NULL) { pkt->status = nvsp_status_failure; return (ENODEV); } rndis_dev = (rndis_device *)net_dev->extension; if (rndis_dev->state == RNDIS_DEV_UNINITIALIZED) { pkt->status = nvsp_status_failure; return (EINVAL); } rndis_hdr = pkt->data; switch (rndis_hdr->ndis_msg_type) { /* data message */ case REMOTE_NDIS_PACKET_MSG: hv_rf_receive_data(rndis_dev, rndis_hdr, pkt); break; /* completion messages */ case REMOTE_NDIS_INITIALIZE_CMPLT: case REMOTE_NDIS_QUERY_CMPLT: case REMOTE_NDIS_SET_CMPLT: case REMOTE_NDIS_RESET_CMPLT: case REMOTE_NDIS_KEEPALIVE_CMPLT: hv_rf_receive_response(rndis_dev, rndis_hdr); break; /* notification message */ case REMOTE_NDIS_INDICATE_STATUS_MSG: hv_rf_receive_indicate_status(rndis_dev, rndis_hdr); break; default: printf("hv_rf_on_receive(): Unknown msg_type 0x%x\n", rndis_hdr->ndis_msg_type); break; } return (0); } /* * RNDIS filter query device */ static int hv_rf_query_device(rndis_device *device, uint32_t oid, void *result, uint32_t *result_size) { rndis_request *request; uint32_t in_result_size = *result_size; rndis_query_request *query; rndis_query_complete *query_complete; int ret = 0; *result_size = 0; request = hv_rndis_request(device, REMOTE_NDIS_QUERY_MSG, RNDIS_MESSAGE_SIZE(rndis_query_request)); if (request == NULL) { ret = -1; goto cleanup; } /* Set up the rndis query */ query = &request->request_msg.msg.query_request; query->oid = oid; query->info_buffer_offset = sizeof(rndis_query_request); query->info_buffer_length = 0; query->device_vc_handle = 0; ret = hv_rf_send_request(device, request, REMOTE_NDIS_QUERY_MSG); if (ret != 0) { /* Fixme: printf added */ printf("RNDISFILTER request failed to Send!\n"); goto cleanup; } sema_wait(&request->wait_sema); /* Copy the response back */ query_complete = &request->response_msg.msg.query_complete; if (query_complete->info_buffer_length > in_result_size) { ret = EINVAL; goto cleanup; } memcpy(result, (void *)((unsigned long)query_complete + query_complete->info_buffer_offset), query_complete->info_buffer_length); *result_size = query_complete->info_buffer_length; cleanup: if (request != NULL) hv_put_rndis_request(device, request); return (ret); } /* * RNDIS filter query device MAC address */ static inline int hv_rf_query_device_mac(rndis_device *device) { uint32_t size = HW_MACADDR_LEN; return (hv_rf_query_device(device, RNDIS_OID_802_3_PERMANENT_ADDRESS, device->hw_mac_addr, &size)); } /* * RNDIS filter query device link status */ static inline int hv_rf_query_device_link_status(rndis_device *device) { uint32_t size = sizeof(uint32_t); return (hv_rf_query_device(device, RNDIS_OID_GEN_MEDIA_CONNECT_STATUS, &device->link_status, &size)); } /* * RNDIS filter set packet filter * Sends an rndis request with the new filter, then waits for a response * from the host. * Returns zero on success, non-zero on failure. */ static int hv_rf_set_packet_filter(rndis_device *device, uint32_t new_filter) { rndis_request *request; rndis_set_request *set; rndis_set_complete *set_complete; uint32_t status; int ret; request = hv_rndis_request(device, REMOTE_NDIS_SET_MSG, RNDIS_MESSAGE_SIZE(rndis_set_request) + sizeof(uint32_t)); if (request == NULL) { ret = -1; goto cleanup; } /* Set up the rndis set */ set = &request->request_msg.msg.set_request; set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER; set->info_buffer_length = sizeof(uint32_t); set->info_buffer_offset = sizeof(rndis_set_request); memcpy((void *)((unsigned long)set + sizeof(rndis_set_request)), &new_filter, sizeof(uint32_t)); ret = hv_rf_send_request(device, request, REMOTE_NDIS_SET_MSG); if (ret != 0) { goto cleanup; } /* * Wait for the response from the host. Another thread will signal * us when the response has arrived. In the failure case, * sema_timedwait() returns a non-zero status after waiting 5 seconds. */ ret = sema_timedwait(&request->wait_sema, 500); if (ret == 0) { /* Response received, check status */ set_complete = &request->response_msg.msg.set_complete; status = set_complete->status; if (status != RNDIS_STATUS_SUCCESS) { /* Bad response status, return error */ ret = -2; } } else { /* * We cannot deallocate the request since we may still * receive a send completion for it. */ goto exit; } cleanup: if (request != NULL) { hv_put_rndis_request(device, request); } exit: return (ret); } /* * RNDIS filter init device */ static int hv_rf_init_device(rndis_device *device) { rndis_request *request; rndis_initialize_request *init; rndis_initialize_complete *init_complete; uint32_t status; int ret; request = hv_rndis_request(device, REMOTE_NDIS_INITIALIZE_MSG, RNDIS_MESSAGE_SIZE(rndis_initialize_request)); if (!request) { ret = -1; goto cleanup; } /* Set up the rndis set */ init = &request->request_msg.msg.init_request; init->major_version = RNDIS_MAJOR_VERSION; init->minor_version = RNDIS_MINOR_VERSION; /* * Per the RNDIS document, this should be set to the max MTU * plus the header size. However, 2048 works fine, so leaving * it as is. */ init->max_xfer_size = 2048; device->state = RNDIS_DEV_INITIALIZING; ret = hv_rf_send_request(device, request, REMOTE_NDIS_INITIALIZE_MSG); if (ret != 0) { device->state = RNDIS_DEV_UNINITIALIZED; goto cleanup; } sema_wait(&request->wait_sema); init_complete = &request->response_msg.msg.init_complete; status = init_complete->status; if (status == RNDIS_STATUS_SUCCESS) { device->state = RNDIS_DEV_INITIALIZED; ret = 0; } else { device->state = RNDIS_DEV_UNINITIALIZED; ret = -1; } cleanup: if (request) { hv_put_rndis_request(device, request); } return (ret); } #define HALT_COMPLETION_WAIT_COUNT 25 /* * RNDIS filter halt device */ static int hv_rf_halt_device(rndis_device *device) { rndis_request *request; rndis_halt_request *halt; int i, ret; /* Attempt to do a rndis device halt */ request = hv_rndis_request(device, REMOTE_NDIS_HALT_MSG, RNDIS_MESSAGE_SIZE(rndis_halt_request)); if (request == NULL) { return (-1); } /* initialize "poor man's semaphore" */ request->halt_complete_flag = 0; /* Set up the rndis set */ halt = &request->request_msg.msg.halt_request; halt->request_id = atomic_fetchadd_int(&device->new_request_id, 1); /* Increment to get the new value (call above returns old value) */ halt->request_id += 1; ret = hv_rf_send_request(device, request, REMOTE_NDIS_HALT_MSG); if (ret != 0) { return (-1); } /* * Wait for halt response from halt callback. We must wait for * the transaction response before freeing the request and other * resources. */ for (i=HALT_COMPLETION_WAIT_COUNT; i > 0; i--) { if (request->halt_complete_flag != 0) { break; } DELAY(400); } if (i == 0) { return (-1); } device->state = RNDIS_DEV_UNINITIALIZED; if (request != NULL) { hv_put_rndis_request(device, request); } return (0); } /* * RNDIS filter open device */ static int hv_rf_open_device(rndis_device *device) { int ret; if (device->state != RNDIS_DEV_INITIALIZED) { return (0); } if (hv_promisc_mode != 1) { ret = hv_rf_set_packet_filter(device, NDIS_PACKET_TYPE_BROADCAST | NDIS_PACKET_TYPE_ALL_MULTICAST | NDIS_PACKET_TYPE_DIRECTED); } else { ret = hv_rf_set_packet_filter(device, NDIS_PACKET_TYPE_PROMISCUOUS); } if (ret == 0) { device->state = RNDIS_DEV_DATAINITIALIZED; } return (ret); } /* * RNDIS filter close device */ static int hv_rf_close_device(rndis_device *device) { int ret; if (device->state != RNDIS_DEV_DATAINITIALIZED) { return (0); } ret = hv_rf_set_packet_filter(device, 0); if (ret == 0) { device->state = RNDIS_DEV_INITIALIZED; } return (ret); } /* * RNDIS filter on device add */ int hv_rf_on_device_add(struct hv_device *device, void *additl_info) { int ret; netvsc_dev *net_dev; rndis_device *rndis_dev; rndis_offload_params offloads; netvsc_device_info *dev_info = (netvsc_device_info *)additl_info; device_t dev = device->device; rndis_dev = hv_get_rndis_device(); if (rndis_dev == NULL) { return (ENOMEM); } /* * Let the inner driver handle this first to create the netvsc channel * NOTE! Once the channel is created, we may get a receive callback * (hv_rf_on_receive()) before this call is completed. * Note: Earlier code used a function pointer here. */ net_dev = hv_nv_on_device_add(device, additl_info); if (!net_dev) { hv_put_rndis_device(rndis_dev); return (ENOMEM); } /* * Initialize the rndis device */ net_dev->extension = rndis_dev; rndis_dev->net_dev = net_dev; /* Send the rndis initialization message */ ret = hv_rf_init_device(rndis_dev); if (ret != 0) { /* * TODO: If rndis init failed, we will need to shut down * the channel */ } /* Get the mac address */ ret = hv_rf_query_device_mac(rndis_dev); if (ret != 0) { /* TODO: shut down rndis device and the channel */ } /* config csum offload and send request to host */ memset(&offloads, 0, sizeof(offloads)); offloads.ipv4_csum = RNDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED; offloads.tcp_ipv4_csum = RNDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED; offloads.udp_ipv4_csum = RNDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED; offloads.tcp_ipv6_csum = RNDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED; offloads.udp_ipv6_csum = RNDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED; offloads.lso_v2_ipv4 = RNDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED; ret = hv_rf_send_offload_request(device, &offloads); if (ret != 0) { /* TODO: shut down rndis device and the channel */ device_printf(dev, "hv_rf_send_offload_request failed, ret=%d\n", ret); } memcpy(dev_info->mac_addr, rndis_dev->hw_mac_addr, HW_MACADDR_LEN); hv_rf_query_device_link_status(rndis_dev); dev_info->link_state = rndis_dev->link_status; return (ret); } /* * RNDIS filter on device remove */ int hv_rf_on_device_remove(struct hv_device *device, boolean_t destroy_channel) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev; rndis_device *rndis_dev = (rndis_device *)net_dev->extension; int ret; /* Halt and release the rndis device */ ret = hv_rf_halt_device(rndis_dev); hv_put_rndis_device(rndis_dev); net_dev->extension = NULL; /* Pass control to inner driver to remove the device */ ret |= hv_nv_on_device_remove(device, destroy_channel); return (ret); } /* * RNDIS filter on open */ int hv_rf_on_open(struct hv_device *device) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev; return (hv_rf_open_device((rndis_device *)net_dev->extension)); } /* * RNDIS filter on close */ int hv_rf_on_close(struct hv_device *device) { hn_softc_t *sc = device_get_softc(device->device); netvsc_dev *net_dev = sc->net_dev; return (hv_rf_close_device((rndis_device *)net_dev->extension)); } /* * RNDIS filter on send request completion callback */ static void hv_rf_on_send_request_completion(void *context) { } /* * RNDIS filter on send request (halt only) completion callback */ static void hv_rf_on_send_request_halt_completion(void *context) { rndis_request *request = context; /* * Notify hv_rf_halt_device() about halt completion. * The halt code must wait for completion before freeing * the transaction resources. */ request->halt_complete_flag = 1; } /* * RNDIS filter when "all" reception is done */ void hv_rf_receive_rollup(netvsc_dev *net_dev) { rndis_device *rndis_dev; rndis_dev = (rndis_device *)net_dev->extension; netvsc_recv_rollup(rndis_dev->net_dev->dev); } void hv_rf_channel_rollup(netvsc_dev *net_dev) { rndis_device *rndis_dev; rndis_dev = (rndis_device *)net_dev->extension; /* * This could be called pretty early, so we need * to make sure everything has been setup. */ if (rndis_dev == NULL || rndis_dev->net_dev == NULL || rndis_dev->net_dev->dev == NULL) return; netvsc_channel_rollup(rndis_dev->net_dev->dev); } Index: head/sys/dev/hyperv/vmbus/hv_connection.c =================================================================== --- head/sys/dev/hyperv/vmbus/hv_connection.c (revision 295307) +++ head/sys/dev/hyperv/vmbus/hv_connection.c (revision 295308) @@ -1,450 +1,420 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2012 NetApp Inc. * Copyright (c) 2012 Citrix Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, 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 ``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 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. */ #include #include #include #include #include #include #include #include #include #include "hv_vmbus_priv.h" /* * Globals */ hv_vmbus_connection hv_vmbus_g_connection = { .connect_state = HV_DISCONNECTED, .next_gpadl_handle = 0xE1E10, }; uint32_t hv_vmbus_protocal_version = HV_VMBUS_VERSION_WS2008; static uint32_t hv_vmbus_get_next_version(uint32_t current_ver) { switch (current_ver) { case (HV_VMBUS_VERSION_WIN7): return(HV_VMBUS_VERSION_WS2008); case (HV_VMBUS_VERSION_WIN8): return(HV_VMBUS_VERSION_WIN7); case (HV_VMBUS_VERSION_WIN8_1): return(HV_VMBUS_VERSION_WIN8); case (HV_VMBUS_VERSION_WS2008): default: return(HV_VMBUS_VERSION_INVALID); } } /** * Negotiate the highest supported hypervisor version. */ static int hv_vmbus_negotiate_version(hv_vmbus_channel_msg_info *msg_info, uint32_t version) { int ret = 0; hv_vmbus_channel_initiate_contact *msg; sema_init(&msg_info->wait_sema, 0, "Msg Info Sema"); msg = (hv_vmbus_channel_initiate_contact*) msg_info->msg; msg->header.message_type = HV_CHANNEL_MESSAGE_INITIATED_CONTACT; msg->vmbus_version_requested = version; msg->interrupt_page = hv_get_phys_addr( hv_vmbus_g_connection.interrupt_page); msg->monitor_page_1 = hv_get_phys_addr( hv_vmbus_g_connection.monitor_pages); - msg->monitor_page_2 = - hv_get_phys_addr( + msg->monitor_page_2 = hv_get_phys_addr( ((uint8_t *) hv_vmbus_g_connection.monitor_pages + PAGE_SIZE)); /** * Add to list before we send the request since we may receive the * response before returning from this routine */ mtx_lock_spin(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_INSERT_TAIL( &hv_vmbus_g_connection.channel_msg_anchor, msg_info, msg_list_entry); mtx_unlock_spin(&hv_vmbus_g_connection.channel_msg_lock); ret = hv_vmbus_post_message( msg, sizeof(hv_vmbus_channel_initiate_contact)); if (ret != 0) { mtx_lock_spin(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_REMOVE( &hv_vmbus_g_connection.channel_msg_anchor, msg_info, msg_list_entry); mtx_unlock_spin(&hv_vmbus_g_connection.channel_msg_lock); return (ret); } /** * Wait for the connection response */ ret = sema_timedwait(&msg_info->wait_sema, 500); /* KYS 5 seconds */ mtx_lock_spin(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_REMOVE( &hv_vmbus_g_connection.channel_msg_anchor, msg_info, msg_list_entry); mtx_unlock_spin(&hv_vmbus_g_connection.channel_msg_lock); /** * Check if successful */ if (msg_info->response.version_response.version_supported) { hv_vmbus_g_connection.connect_state = HV_CONNECTED; } else { ret = ECONNREFUSED; } return (ret); } /** * Send a connect request on the partition service connection */ int hv_vmbus_connect(void) { int ret = 0; uint32_t version; hv_vmbus_channel_msg_info* msg_info = NULL; /** * Make sure we are not connecting or connected */ if (hv_vmbus_g_connection.connect_state != HV_DISCONNECTED) { return (-1); } /** * Initialize the vmbus connection */ hv_vmbus_g_connection.connect_state = HV_CONNECTING; TAILQ_INIT(&hv_vmbus_g_connection.channel_msg_anchor); mtx_init(&hv_vmbus_g_connection.channel_msg_lock, "vmbus channel msg", NULL, MTX_SPIN); TAILQ_INIT(&hv_vmbus_g_connection.channel_anchor); mtx_init(&hv_vmbus_g_connection.channel_lock, "vmbus channel", NULL, MTX_DEF); /** * Setup the vmbus event connection for channel interrupt abstraction * stuff */ hv_vmbus_g_connection.interrupt_page = contigmalloc( PAGE_SIZE, M_DEVBUF, - M_NOWAIT | M_ZERO, 0UL, + M_WAITOK | M_ZERO, 0UL, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); - KASSERT(hv_vmbus_g_connection.interrupt_page != NULL, - ("Error VMBUS: malloc failed to allocate Channel" - " Request Event message!")); - if (hv_vmbus_g_connection.interrupt_page == NULL) { - ret = ENOMEM; - goto cleanup; - } hv_vmbus_g_connection.recv_interrupt_page = hv_vmbus_g_connection.interrupt_page; hv_vmbus_g_connection.send_interrupt_page = ((uint8_t *) hv_vmbus_g_connection.interrupt_page + (PAGE_SIZE >> 1)); /** * Set up the monitor notification facility. The 1st page for * parent->child and the 2nd page for child->parent */ hv_vmbus_g_connection.monitor_pages = contigmalloc( 2 * PAGE_SIZE, M_DEVBUF, - M_NOWAIT | M_ZERO, + M_WAITOK | M_ZERO, 0UL, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); - KASSERT(hv_vmbus_g_connection.monitor_pages != NULL, - ("Error VMBUS: malloc failed to allocate Monitor Pages!")); - if (hv_vmbus_g_connection.monitor_pages == NULL) { - ret = ENOMEM; - goto cleanup; - } msg_info = (hv_vmbus_channel_msg_info*) malloc(sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_initiate_contact), - M_DEVBUF, M_NOWAIT | M_ZERO); - KASSERT(msg_info != NULL, - ("Error VMBUS: malloc failed for Initiate Contact message!")); - if (msg_info == NULL) { - ret = ENOMEM; - goto cleanup; - } + M_DEVBUF, M_WAITOK | M_ZERO); hv_vmbus_g_connection.channels = malloc(sizeof(hv_vmbus_channel*) * HV_CHANNEL_MAX_COUNT, M_DEVBUF, M_WAITOK | M_ZERO); /* * Find the highest vmbus version number we can support. */ version = HV_VMBUS_VERSION_CURRENT; do { ret = hv_vmbus_negotiate_version(msg_info, version); if (ret == EWOULDBLOCK) { /* * We timed out. */ goto cleanup; } if (hv_vmbus_g_connection.connect_state == HV_CONNECTED) break; version = hv_vmbus_get_next_version(version); } while (version != HV_VMBUS_VERSION_INVALID); hv_vmbus_protocal_version = version; if (bootverbose) printf("VMBUS: Protocol Version: %d.%d\n", version >> 16, version & 0xFFFF); sema_destroy(&msg_info->wait_sema); free(msg_info, M_DEVBUF); return (0); /* * Cleanup after failure! */ cleanup: hv_vmbus_g_connection.connect_state = HV_DISCONNECTED; mtx_destroy(&hv_vmbus_g_connection.channel_lock); mtx_destroy(&hv_vmbus_g_connection.channel_msg_lock); if (hv_vmbus_g_connection.interrupt_page != NULL) { contigfree( hv_vmbus_g_connection.interrupt_page, PAGE_SIZE, M_DEVBUF); hv_vmbus_g_connection.interrupt_page = NULL; } if (hv_vmbus_g_connection.monitor_pages != NULL) { contigfree( hv_vmbus_g_connection.monitor_pages, 2 * PAGE_SIZE, M_DEVBUF); hv_vmbus_g_connection.monitor_pages = NULL; } if (msg_info) { sema_destroy(&msg_info->wait_sema); free(msg_info, M_DEVBUF); } free(hv_vmbus_g_connection.channels, M_DEVBUF); return (ret); } /** * Send a disconnect request on the partition service connection */ int hv_vmbus_disconnect(void) { int ret = 0; - hv_vmbus_channel_unload* msg; + hv_vmbus_channel_unload msg; - msg = malloc(sizeof(hv_vmbus_channel_unload), - M_DEVBUF, M_NOWAIT | M_ZERO); - KASSERT(msg != NULL, - ("Error VMBUS: malloc failed to allocate Channel Unload Msg!")); - if (msg == NULL) - return (ENOMEM); + msg.message_type = HV_CHANNEL_MESSAGE_UNLOAD; - msg->message_type = HV_CHANNEL_MESSAGE_UNLOAD; + ret = hv_vmbus_post_message(&msg, sizeof(hv_vmbus_channel_unload)); - ret = hv_vmbus_post_message(msg, sizeof(hv_vmbus_channel_unload)); - - contigfree(hv_vmbus_g_connection.interrupt_page, PAGE_SIZE, M_DEVBUF); mtx_destroy(&hv_vmbus_g_connection.channel_msg_lock); free(hv_vmbus_g_connection.channels, M_DEVBUF); hv_vmbus_g_connection.connect_state = HV_DISCONNECTED; - - free(msg, M_DEVBUF); return (ret); } /** * Handler for events */ void hv_vmbus_on_events(int cpu) { int bit; int dword; void *page_addr; uint32_t* recv_interrupt_page = NULL; int rel_id; int maxdword; hv_vmbus_synic_event_flags *event; /* int maxdword = PAGE_SIZE >> 3; */ KASSERT(cpu <= mp_maxid, ("VMBUS: hv_vmbus_on_events: " "cpu out of range!")); if ((hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008) || (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7)) { maxdword = HV_MAX_NUM_CHANNELS_SUPPORTED >> 5; /* * receive size is 1/2 page and divide that by 4 bytes */ recv_interrupt_page = hv_vmbus_g_connection.recv_interrupt_page; } else { /* * On Host with Win8 or above, the event page can be * checked directly to get the id of the channel * that has the pending interrupt. */ maxdword = HV_EVENT_FLAGS_DWORD_COUNT; page_addr = hv_vmbus_g_context.syn_ic_event_page[cpu]; event = (hv_vmbus_synic_event_flags *) page_addr + HV_VMBUS_MESSAGE_SINT; recv_interrupt_page = event->flags32; } /* * Check events */ if (recv_interrupt_page != NULL) { for (dword = 0; dword < maxdword; dword++) { if (recv_interrupt_page[dword]) { for (bit = 0; bit < HV_CHANNEL_DWORD_LEN; bit++) { if (synch_test_and_clear_bit(bit, (uint32_t *) &recv_interrupt_page[dword])) { rel_id = (dword << 5) + bit; if (rel_id == 0) { /* * Special case - * vmbus channel protocol msg. */ continue; } else { hv_vmbus_channel * channel = hv_vmbus_g_connection.channels[rel_id]; /* if channel is closed or closing */ if (channel == NULL || channel->rxq == NULL) continue; if (channel->batched_reading) hv_ring_buffer_read_begin(&channel->inbound); taskqueue_enqueue_fast(channel->rxq, &channel->channel_task); } } } } } } return; } /** * Send a msg on the vmbus's message connection */ int hv_vmbus_post_message(void *buffer, size_t bufferLen) { int ret = 0; hv_vmbus_connection_id connId; unsigned retries = 0; /* NetScaler delays from previous code were consolidated here */ static int delayAmount[] = {100, 100, 100, 500, 500, 5000, 5000, 5000}; /* for(each entry in delayAmount) try to post message, * delay a little bit before retrying */ for (retries = 0; retries < sizeof(delayAmount)/sizeof(delayAmount[0]); retries++) { connId.as_uint32_t = 0; connId.u.id = HV_VMBUS_MESSAGE_CONNECTION_ID; ret = hv_vmbus_post_msg_via_msg_ipc(connId, 1, buffer, bufferLen); if (ret != HV_STATUS_INSUFFICIENT_BUFFERS) break; /* TODO: KYS We should use a blocking wait call */ DELAY(delayAmount[retries]); } KASSERT(ret == 0, ("Error VMBUS: Message Post Failed\n")); return (ret); } /** * Send an event notification to the parent */ int hv_vmbus_set_event(hv_vmbus_channel *channel) { int ret = 0; uint32_t child_rel_id = channel->offer_msg.child_rel_id; /* Each uint32_t represents 32 channels */ synch_set_bit(child_rel_id & 31, (((uint32_t *)hv_vmbus_g_connection.send_interrupt_page + (child_rel_id >> 5)))); ret = hv_vmbus_signal_event(channel->signal_event_param); return (ret); } Index: head/sys/dev/hyperv/vmbus/hv_hv.c =================================================================== --- head/sys/dev/hyperv/vmbus/hv_hv.c (revision 295307) +++ head/sys/dev/hyperv/vmbus/hv_hv.c (revision 295308) @@ -1,429 +1,425 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2012 NetApp Inc. * Copyright (c) 2012 Citrix Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, 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 ``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 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. */ /** * Implements low-level interactions with Hypver-V/Azure */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include "hv_vmbus_priv.h" #define HV_NANOSECONDS_PER_SEC 1000000000L static u_int hv_get_timecount(struct timecounter *tc); /** * Globals */ hv_vmbus_context hv_vmbus_g_context = { .syn_ic_initialized = FALSE, .hypercall_page = NULL, }; static struct timecounter hv_timecounter = { hv_get_timecount, 0, ~0u, HV_NANOSECONDS_PER_SEC/100, "Hyper-V", HV_NANOSECONDS_PER_SEC/100 }; static u_int hv_get_timecount(struct timecounter *tc) { u_int now = rdmsr(HV_X64_MSR_TIME_REF_COUNT); return (now); } /** * @brief Query the cpuid for presence of windows hypervisor */ int hv_vmbus_query_hypervisor_presence(void) { if (vm_guest != VM_GUEST_HV) return (0); return (hv_high >= HV_X64_CPUID_MIN && hv_high <= HV_X64_CPUID_MAX); } /** * @brief Get version of the windows hypervisor */ static int hv_vmbus_get_hypervisor_version(void) { u_int regs[4]; unsigned int maxLeaf; unsigned int op; /* * Its assumed that this is called after confirming that * Viridian is present * Query id and revision. */ op = HV_CPU_ID_FUNCTION_HV_VENDOR_AND_MAX_FUNCTION; do_cpuid(op, regs); maxLeaf = regs[0]; op = HV_CPU_ID_FUNCTION_HV_INTERFACE; do_cpuid(op, regs); if (maxLeaf >= HV_CPU_ID_FUNCTION_MS_HV_VERSION) { op = HV_CPU_ID_FUNCTION_MS_HV_VERSION; do_cpuid(op, regs); } return (maxLeaf); } /** * @brief Invoke the specified hypercall */ static uint64_t hv_vmbus_do_hypercall(uint64_t control, void* input, void* output) { #ifdef __x86_64__ uint64_t hv_status = 0; uint64_t input_address = (input) ? hv_get_phys_addr(input) : 0; uint64_t output_address = (output) ? hv_get_phys_addr(output) : 0; volatile void* hypercall_page = hv_vmbus_g_context.hypercall_page; __asm__ __volatile__ ("mov %0, %%r8" : : "r" (output_address): "r8"); __asm__ __volatile__ ("call *%3" : "=a"(hv_status): "c" (control), "d" (input_address), "m" (hypercall_page)); return (hv_status); #else uint32_t control_high = control >> 32; uint32_t control_low = control & 0xFFFFFFFF; uint32_t hv_status_high = 1; uint32_t hv_status_low = 1; uint64_t input_address = (input) ? hv_get_phys_addr(input) : 0; uint32_t input_address_high = input_address >> 32; uint32_t input_address_low = input_address & 0xFFFFFFFF; uint64_t output_address = (output) ? hv_get_phys_addr(output) : 0; uint32_t output_address_high = output_address >> 32; uint32_t output_address_low = output_address & 0xFFFFFFFF; volatile void* hypercall_page = hv_vmbus_g_context.hypercall_page; __asm__ __volatile__ ("call *%8" : "=d"(hv_status_high), "=a"(hv_status_low) : "d" (control_high), "a" (control_low), "b" (input_address_high), "c" (input_address_low), "D"(output_address_high), "S"(output_address_low), "m" (hypercall_page)); return (hv_status_low | ((uint64_t)hv_status_high << 32)); #endif /* __x86_64__ */ } /** * @brief Main initialization routine. * * This routine must be called * before any other routines in here are called */ int hv_vmbus_init(void) { int max_leaf; hv_vmbus_x64_msr_hypercall_contents hypercall_msr; void* virt_addr = 0; memset( hv_vmbus_g_context.syn_ic_event_page, 0, sizeof(hv_vmbus_handle) * MAXCPU); memset( hv_vmbus_g_context.syn_ic_msg_page, 0, sizeof(hv_vmbus_handle) * MAXCPU); if (vm_guest != VM_GUEST_HV) goto cleanup; max_leaf = hv_vmbus_get_hypervisor_version(); /* * Write our OS info */ uint64_t os_guest_info = HV_FREEBSD_GUEST_ID; wrmsr(HV_X64_MSR_GUEST_OS_ID, os_guest_info); hv_vmbus_g_context.guest_id = os_guest_info; /* * See if the hypercall page is already set */ hypercall_msr.as_uint64_t = rdmsr(HV_X64_MSR_HYPERCALL); - virt_addr = malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT | M_ZERO); - KASSERT(virt_addr != NULL, - ("Error VMBUS: malloc failed to allocate page during init!")); - if (virt_addr == NULL) - goto cleanup; + virt_addr = malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK | M_ZERO); hypercall_msr.u.enable = 1; hypercall_msr.u.guest_physical_address = (hv_get_phys_addr(virt_addr) >> PAGE_SHIFT); wrmsr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64_t); /* * Confirm that hypercall page did get set up */ hypercall_msr.as_uint64_t = 0; hypercall_msr.as_uint64_t = rdmsr(HV_X64_MSR_HYPERCALL); if (!hypercall_msr.u.enable) goto cleanup; hv_vmbus_g_context.hypercall_page = virt_addr; tc_init(&hv_timecounter); /* register virtual timecount */ hv_et_init(); return (0); cleanup: if (virt_addr != NULL) { if (hypercall_msr.u.enable) { hypercall_msr.as_uint64_t = 0; wrmsr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64_t); } free(virt_addr, M_DEVBUF); } return (ENOTSUP); } /** * @brief Cleanup routine, called normally during driver unloading or exiting */ void hv_vmbus_cleanup(void) { hv_vmbus_x64_msr_hypercall_contents hypercall_msr; if (hv_vmbus_g_context.guest_id == HV_FREEBSD_GUEST_ID) { if (hv_vmbus_g_context.hypercall_page != NULL) { hypercall_msr.as_uint64_t = 0; wrmsr(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64_t); free(hv_vmbus_g_context.hypercall_page, M_DEVBUF); hv_vmbus_g_context.hypercall_page = NULL; } } } /** * @brief Post a message using the hypervisor message IPC. * (This involves a hypercall.) */ hv_vmbus_status hv_vmbus_post_msg_via_msg_ipc( hv_vmbus_connection_id connection_id, hv_vmbus_msg_type message_type, void* payload, size_t payload_size) { struct alignedinput { uint64_t alignment8; hv_vmbus_input_post_message msg; }; hv_vmbus_input_post_message* aligned_msg; hv_vmbus_status status; size_t addr; if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) return (EMSGSIZE); addr = (size_t) malloc(sizeof(struct alignedinput), M_DEVBUF, M_ZERO | M_NOWAIT); KASSERT(addr != 0, ("Error VMBUS: malloc failed to allocate message buffer!")); if (addr == 0) return (ENOMEM); aligned_msg = (hv_vmbus_input_post_message*) (HV_ALIGN_UP(addr, HV_HYPERCALL_PARAM_ALIGN)); aligned_msg->connection_id = connection_id; aligned_msg->message_type = message_type; aligned_msg->payload_size = payload_size; memcpy((void*) aligned_msg->payload, payload, payload_size); status = hv_vmbus_do_hypercall( HV_CALL_POST_MESSAGE, aligned_msg, 0) & 0xFFFF; free((void *) addr, M_DEVBUF); return (status); } /** * @brief Signal an event on the specified connection using the hypervisor * event IPC. (This involves a hypercall.) */ hv_vmbus_status hv_vmbus_signal_event(void *con_id) { hv_vmbus_status status; status = hv_vmbus_do_hypercall( HV_CALL_SIGNAL_EVENT, con_id, 0) & 0xFFFF; return (status); } /** * @brief hv_vmbus_synic_init */ void hv_vmbus_synic_init(void *arg) { int cpu; uint64_t hv_vcpu_index; hv_vmbus_synic_simp simp; hv_vmbus_synic_siefp siefp; hv_vmbus_synic_scontrol sctrl; hv_vmbus_synic_sint shared_sint; uint64_t version; hv_setup_args* setup_args = (hv_setup_args *)arg; cpu = PCPU_GET(cpuid); if (hv_vmbus_g_context.hypercall_page == NULL) return; /* * TODO: Check the version */ version = rdmsr(HV_X64_MSR_SVERSION); hv_vmbus_g_context.syn_ic_msg_page[cpu] = setup_args->page_buffers[2 * cpu]; hv_vmbus_g_context.syn_ic_event_page[cpu] = setup_args->page_buffers[2 * cpu + 1]; /* * Setup the Synic's message page */ simp.as_uint64_t = rdmsr(HV_X64_MSR_SIMP); simp.u.simp_enabled = 1; simp.u.base_simp_gpa = ((hv_get_phys_addr( hv_vmbus_g_context.syn_ic_msg_page[cpu])) >> PAGE_SHIFT); wrmsr(HV_X64_MSR_SIMP, simp.as_uint64_t); /* * Setup the Synic's event page */ siefp.as_uint64_t = rdmsr(HV_X64_MSR_SIEFP); siefp.u.siefp_enabled = 1; siefp.u.base_siefp_gpa = ((hv_get_phys_addr( hv_vmbus_g_context.syn_ic_event_page[cpu])) >> PAGE_SHIFT); wrmsr(HV_X64_MSR_SIEFP, siefp.as_uint64_t); /*HV_SHARED_SINT_IDT_VECTOR + 0x20; */ shared_sint.as_uint64_t = 0; shared_sint.u.vector = setup_args->vector; shared_sint.u.masked = FALSE; shared_sint.u.auto_eoi = TRUE; wrmsr(HV_X64_MSR_SINT0 + HV_VMBUS_MESSAGE_SINT, shared_sint.as_uint64_t); /* Enable the global synic bit */ sctrl.as_uint64_t = rdmsr(HV_X64_MSR_SCONTROL); sctrl.u.enable = 1; wrmsr(HV_X64_MSR_SCONTROL, sctrl.as_uint64_t); hv_vmbus_g_context.syn_ic_initialized = TRUE; /* * Set up the cpuid mapping from Hyper-V to FreeBSD. * The array is indexed using FreeBSD cpuid. */ hv_vcpu_index = rdmsr(HV_X64_MSR_VP_INDEX); hv_vmbus_g_context.hv_vcpu_index[cpu] = (uint32_t)hv_vcpu_index; return; } /** * @brief Cleanup routine for hv_vmbus_synic_init() */ void hv_vmbus_synic_cleanup(void *arg) { hv_vmbus_synic_sint shared_sint; hv_vmbus_synic_simp simp; hv_vmbus_synic_siefp siefp; if (!hv_vmbus_g_context.syn_ic_initialized) return; shared_sint.as_uint64_t = rdmsr( HV_X64_MSR_SINT0 + HV_VMBUS_MESSAGE_SINT); shared_sint.u.masked = 1; /* * Disable the interrupt */ wrmsr( HV_X64_MSR_SINT0 + HV_VMBUS_MESSAGE_SINT, shared_sint.as_uint64_t); simp.as_uint64_t = rdmsr(HV_X64_MSR_SIMP); simp.u.simp_enabled = 0; simp.u.base_simp_gpa = 0; wrmsr(HV_X64_MSR_SIMP, simp.as_uint64_t); siefp.as_uint64_t = rdmsr(HV_X64_MSR_SIEFP); siefp.u.siefp_enabled = 0; siefp.u.base_siefp_gpa = 0; wrmsr(HV_X64_MSR_SIEFP, siefp.as_uint64_t); } Index: head/sys/dev/hyperv/vmbus/hv_vmbus_drv_freebsd.c =================================================================== --- head/sys/dev/hyperv/vmbus/hv_vmbus_drv_freebsd.c (revision 295307) +++ head/sys/dev/hyperv/vmbus/hv_vmbus_drv_freebsd.c (revision 295308) @@ -1,742 +1,732 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2012 NetApp Inc. * Copyright (c) 2012 Citrix Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, 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 ``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 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. */ /* * VM Bus Driver Implementation */ #include __FBSDID("$FreeBSD$"); #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 "hv_vmbus_priv.h" #include #include "acpi_if.h" static device_t vmbus_devp; static int vmbus_inited; static hv_setup_args setup_args; /* only CPU 0 supported at this time */ static char *vmbus_ids[] = { "VMBUS", NULL }; /** * @brief Software interrupt thread routine to handle channel messages from * the hypervisor. */ static void vmbus_msg_swintr(void *arg) { int cpu; void* page_addr; hv_vmbus_channel_msg_header *hdr; hv_vmbus_channel_msg_table_entry *entry; hv_vmbus_channel_msg_type msg_type; hv_vmbus_message* msg; cpu = (int)(long)arg; KASSERT(cpu <= mp_maxid, ("VMBUS: vmbus_msg_swintr: " "cpu out of range!")); page_addr = hv_vmbus_g_context.syn_ic_msg_page[cpu]; msg = (hv_vmbus_message*) page_addr + HV_VMBUS_MESSAGE_SINT; for (;;) { if (msg->header.message_type == HV_MESSAGE_TYPE_NONE) break; /* no message */ hdr = (hv_vmbus_channel_msg_header *)msg->u.payload; msg_type = hdr->message_type; if (msg_type >= HV_CHANNEL_MESSAGE_COUNT) { printf("VMBUS: unknown message type = %d\n", msg_type); goto handled; } entry = &g_channel_message_table[msg_type]; if (entry->messageHandler) entry->messageHandler(hdr); handled: msg->header.message_type = HV_MESSAGE_TYPE_NONE; /* * Make sure the write to message_type (ie set to * HV_MESSAGE_TYPE_NONE) happens before we read the * message_pending and EOMing. Otherwise, the EOMing will * not deliver any more messages * since there is no empty slot */ wmb(); if (msg->header.message_flags.u.message_pending) { /* * This will cause message queue rescan to possibly * deliver another msg from the hypervisor */ wrmsr(HV_X64_MSR_EOM, 0); } } } /** * @brief Interrupt filter routine for VMBUS. * * The purpose of this routine is to determine the type of VMBUS protocol * message to process - an event or a channel message. */ static inline int hv_vmbus_isr(struct trapframe *frame) { int cpu; hv_vmbus_message* msg; hv_vmbus_synic_event_flags* event; void* page_addr; cpu = PCPU_GET(cpuid); /* * The Windows team has advised that we check for events * before checking for messages. This is the way they do it * in Windows when running as a guest in Hyper-V */ page_addr = hv_vmbus_g_context.syn_ic_event_page[cpu]; event = (hv_vmbus_synic_event_flags*) page_addr + HV_VMBUS_MESSAGE_SINT; if ((hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008) || (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7)) { /* Since we are a child, we only need to check bit 0 */ if (synch_test_and_clear_bit(0, &event->flags32[0])) { hv_vmbus_on_events(cpu); } } else { /* * On host with Win8 or above, we can directly look at * the event page. If bit n is set, we have an interrupt * on the channel with id n. * Directly schedule the event software interrupt on * current cpu. */ hv_vmbus_on_events(cpu); } /* Check if there are actual msgs to be process */ page_addr = hv_vmbus_g_context.syn_ic_msg_page[cpu]; msg = (hv_vmbus_message*) page_addr + HV_VMBUS_MESSAGE_SINT; /* we call eventtimer process the message */ if (msg->header.message_type == HV_MESSAGE_TIMER_EXPIRED) { msg->header.message_type = HV_MESSAGE_TYPE_NONE; /* * Make sure the write to message_type (ie set to * HV_MESSAGE_TYPE_NONE) happens before we read the * message_pending and EOMing. Otherwise, the EOMing will * not deliver any more messages * since there is no empty slot */ wmb(); if (msg->header.message_flags.u.message_pending) { /* * This will cause message queue rescan to possibly * deliver another msg from the hypervisor */ wrmsr(HV_X64_MSR_EOM, 0); } hv_et_intr(frame); return (FILTER_HANDLED); } if (msg->header.message_type != HV_MESSAGE_TYPE_NONE) { swi_sched(hv_vmbus_g_context.msg_swintr[cpu], 0); } return (FILTER_HANDLED); } u_long *hv_vmbus_intr_cpu[MAXCPU]; void hv_vector_handler(struct trapframe *trap_frame) { int cpu; /* * Disable preemption. */ critical_enter(); /* * Do a little interrupt counting. */ cpu = PCPU_GET(cpuid); (*hv_vmbus_intr_cpu[cpu])++; hv_vmbus_isr(trap_frame); /* * Enable preemption. */ critical_exit(); } static int vmbus_read_ivar( device_t dev, device_t child, int index, uintptr_t* result) { struct hv_device *child_dev_ctx = device_get_ivars(child); switch (index) { case HV_VMBUS_IVAR_TYPE: *result = (uintptr_t) &child_dev_ctx->class_id; return (0); case HV_VMBUS_IVAR_INSTANCE: *result = (uintptr_t) &child_dev_ctx->device_id; return (0); case HV_VMBUS_IVAR_DEVCTX: *result = (uintptr_t) child_dev_ctx; return (0); case HV_VMBUS_IVAR_NODE: *result = (uintptr_t) child_dev_ctx->device; return (0); } return (ENOENT); } static int vmbus_write_ivar( device_t dev, device_t child, int index, uintptr_t value) { switch (index) { case HV_VMBUS_IVAR_TYPE: case HV_VMBUS_IVAR_INSTANCE: case HV_VMBUS_IVAR_DEVCTX: case HV_VMBUS_IVAR_NODE: /* read-only */ return (EINVAL); } return (ENOENT); } struct hv_device* hv_vmbus_child_device_create( hv_guid type, hv_guid instance, hv_vmbus_channel* channel) { hv_device* child_dev; /* * Allocate the new child device */ child_dev = malloc(sizeof(hv_device), M_DEVBUF, - M_NOWAIT | M_ZERO); - KASSERT(child_dev != NULL, - ("Error VMBUS: malloc failed to allocate hv_device!")); + M_WAITOK | M_ZERO); - if (child_dev == NULL) - return (NULL); - child_dev->channel = channel; memcpy(&child_dev->class_id, &type, sizeof(hv_guid)); memcpy(&child_dev->device_id, &instance, sizeof(hv_guid)); return (child_dev); } static void print_dev_guid(struct hv_device *dev) { int i; unsigned char guid_name[100]; for (i = 0; i < 32; i += 2) sprintf(&guid_name[i], "%02x", dev->class_id.data[i / 2]); if(bootverbose) printf("VMBUS: Class ID: %s\n", guid_name); } int hv_vmbus_child_device_register(struct hv_device *child_dev) { device_t child; int ret = 0; print_dev_guid(child_dev); child = device_add_child(vmbus_devp, NULL, -1); child_dev->device = child; device_set_ivars(child, child_dev); mtx_lock(&Giant); ret = device_probe_and_attach(child); mtx_unlock(&Giant); return (0); } int hv_vmbus_child_device_unregister(struct hv_device *child_dev) { int ret = 0; /* * XXXKYS: Ensure that this is the opposite of * device_add_child() */ mtx_lock(&Giant); ret = device_delete_child(vmbus_devp, child_dev->device); mtx_unlock(&Giant); return(ret); } static int vmbus_probe(device_t dev) { if (ACPI_ID_PROBE(device_get_parent(dev), dev, vmbus_ids) == NULL || device_get_unit(dev) != 0) return (ENXIO); device_set_desc(dev, "Vmbus Devices"); return (BUS_PROBE_DEFAULT); } #ifdef HYPERV extern inthand_t IDTVEC(rsvd), IDTVEC(hv_vmbus_callback); /** * @brief Find a free IDT slot and setup the interrupt handler. */ static int vmbus_vector_alloc(void) { int vector; uintptr_t func; struct gate_descriptor *ip; /* * Search backwards form the highest IDT vector available for use * as vmbus channel callback vector. We install 'hv_vmbus_callback' * handler at that vector and use it to interrupt vcpus. */ vector = APIC_SPURIOUS_INT; while (--vector >= APIC_IPI_INTS) { ip = &idt[vector]; func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset); if (func == (uintptr_t)&IDTVEC(rsvd)) { #ifdef __i386__ setidt(vector , IDTVEC(hv_vmbus_callback), SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); #else setidt(vector , IDTVEC(hv_vmbus_callback), SDT_SYSIGT, SEL_KPL, 0); #endif return (vector); } } return (0); } /** * @brief Restore the IDT slot to rsvd. */ static void vmbus_vector_free(int vector) { uintptr_t func; struct gate_descriptor *ip; if (vector == 0) return; KASSERT(vector >= APIC_IPI_INTS && vector < APIC_SPURIOUS_INT, ("invalid vector %d", vector)); ip = &idt[vector]; func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset); KASSERT(func == (uintptr_t)&IDTVEC(hv_vmbus_callback), ("invalid vector %d", vector)); setidt(vector, IDTVEC(rsvd), SDT_SYSIGT, SEL_KPL, 0); } #else /* HYPERV */ static int vmbus_vector_alloc(void) { return(0); } static void vmbus_vector_free(int vector) { } #endif /* HYPERV */ /** * @brief Main vmbus driver initialization routine. * * Here, we * - initialize the vmbus driver context * - setup various driver entry points * - invoke the vmbus hv main init routine * - get the irq resource * - invoke the vmbus to add the vmbus root device * - setup the vmbus root device * - retrieve the channel offers */ static int vmbus_bus_init(void) { int i, j, n, ret; char buf[MAXCOMLEN + 1]; cpuset_t cpu_mask; if (vmbus_inited) return (0); vmbus_inited = 1; ret = hv_vmbus_init(); if (ret) { if(bootverbose) printf("Error VMBUS: Hypervisor Initialization Failed!\n"); return (ret); } /* * Find a free IDT slot for vmbus callback. */ hv_vmbus_g_context.hv_cb_vector = vmbus_vector_alloc(); if (hv_vmbus_g_context.hv_cb_vector == 0) { if(bootverbose) printf("Error VMBUS: Cannot find free IDT slot for " "vmbus callback!\n"); goto cleanup; } if(bootverbose) printf("VMBUS: vmbus callback vector %d\n", hv_vmbus_g_context.hv_cb_vector); /* * Notify the hypervisor of our vector. */ setup_args.vector = hv_vmbus_g_context.hv_cb_vector; CPU_FOREACH(j) { hv_vmbus_g_context.hv_msg_intr_event[j] = NULL; hv_vmbus_g_context.msg_swintr[j] = NULL; snprintf(buf, sizeof(buf), "cpu%d:hyperv", j); intrcnt_add(buf, &hv_vmbus_intr_cpu[j]); for (i = 0; i < 2; i++) setup_args.page_buffers[2 * j + i] = NULL; } /* * Per cpu setup. */ CPU_FOREACH(j) { /* * Setup taskqueue to handle events */ hv_vmbus_g_context.hv_event_queue[j] = taskqueue_create_fast("hyperv event", M_WAITOK, taskqueue_thread_enqueue, &hv_vmbus_g_context.hv_event_queue[j]); if (hv_vmbus_g_context.hv_event_queue[j] == NULL) { if (bootverbose) printf("VMBUS: failed to setup taskqueue\n"); goto cleanup1; } CPU_SETOF(j, &cpu_mask); taskqueue_start_threads_cpuset(&hv_vmbus_g_context.hv_event_queue[j], 1, PI_NET, &cpu_mask, "hvevent%d", j); /* * Setup software interrupt thread and handler for msg handling. */ ret = swi_add(&hv_vmbus_g_context.hv_msg_intr_event[j], "hv_msg", vmbus_msg_swintr, (void *)(long)j, SWI_CLOCK, 0, &hv_vmbus_g_context.msg_swintr[j]); if (ret) { if(bootverbose) printf("VMBUS: failed to setup msg swi for " "cpu %d\n", j); goto cleanup1; } /* * Bind the swi thread to the cpu. */ ret = intr_event_bind(hv_vmbus_g_context.hv_msg_intr_event[j], j); if (ret) { if(bootverbose) printf("VMBUS: failed to bind msg swi thread " "to cpu %d\n", j); goto cleanup1; } /* * Prepare the per cpu msg and event pages to be called on each cpu. */ for(i = 0; i < 2; i++) { setup_args.page_buffers[2 * j + i] = - malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT | M_ZERO); - if (setup_args.page_buffers[2 * j + i] == NULL) { - KASSERT(setup_args.page_buffers[2 * j + i] != NULL, - ("Error VMBUS: malloc failed!")); - goto cleanup1; - } + malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK | M_ZERO); } } if (bootverbose) printf("VMBUS: Calling smp_rendezvous, smp_started = %d\n", smp_started); smp_rendezvous(NULL, hv_vmbus_synic_init, NULL, &setup_args); /* * Connect to VMBus in the root partition */ ret = hv_vmbus_connect(); if (ret != 0) goto cleanup1; hv_vmbus_request_channel_offers(); return (ret); cleanup1: /* * Free pages alloc'ed */ for (n = 0; n < 2 * MAXCPU; n++) if (setup_args.page_buffers[n] != NULL) free(setup_args.page_buffers[n], M_DEVBUF); /* * remove swi and vmbus callback vector; */ CPU_FOREACH(j) { if (hv_vmbus_g_context.hv_event_queue[j] != NULL) taskqueue_free(hv_vmbus_g_context.hv_event_queue[j]); if (hv_vmbus_g_context.msg_swintr[j] != NULL) swi_remove(hv_vmbus_g_context.msg_swintr[j]); hv_vmbus_g_context.hv_msg_intr_event[j] = NULL; } vmbus_vector_free(hv_vmbus_g_context.hv_cb_vector); cleanup: hv_vmbus_cleanup(); return (ret); } static int vmbus_attach(device_t dev) { if(bootverbose) device_printf(dev, "VMBUS: attach dev: %p\n", dev); vmbus_devp = dev; /* * If the system has already booted and thread * scheduling is possible indicated by the global * cold set to zero, we just call the driver * initialization directly. */ if (!cold) vmbus_bus_init(); return (0); } static void vmbus_init(void) { if (vm_guest != VM_GUEST_HV) return; /* * If the system has already booted and thread * scheduling is possible, as indicated by the * global cold set to zero, we just call the driver * initialization directly. */ if (!cold) vmbus_bus_init(); } static void vmbus_bus_exit(void) { int i; hv_vmbus_release_unattached_channels(); hv_vmbus_disconnect(); smp_rendezvous(NULL, hv_vmbus_synic_cleanup, NULL, NULL); for(i = 0; i < 2 * MAXCPU; i++) { if (setup_args.page_buffers[i] != 0) free(setup_args.page_buffers[i], M_DEVBUF); } hv_vmbus_cleanup(); /* remove swi */ CPU_FOREACH(i) { if (hv_vmbus_g_context.hv_event_queue[i] != NULL) taskqueue_free(hv_vmbus_g_context.hv_event_queue[i]); if (hv_vmbus_g_context.msg_swintr[i] != NULL) swi_remove(hv_vmbus_g_context.msg_swintr[i]); hv_vmbus_g_context.hv_msg_intr_event[i] = NULL; } vmbus_vector_free(hv_vmbus_g_context.hv_cb_vector); return; } static void vmbus_exit(void) { vmbus_bus_exit(); } static int vmbus_detach(device_t dev) { vmbus_exit(); return (0); } static void vmbus_mod_load(void) { if(bootverbose) printf("VMBUS: load\n"); } static void vmbus_mod_unload(void) { if(bootverbose) printf("VMBUS: unload\n"); } static int vmbus_modevent(module_t mod, int what, void *arg) { switch (what) { case MOD_LOAD: vmbus_mod_load(); break; case MOD_UNLOAD: vmbus_mod_unload(); break; } return (0); } static device_method_t vmbus_methods[] = { /** Device interface */ DEVMETHOD(device_probe, vmbus_probe), DEVMETHOD(device_attach, vmbus_attach), DEVMETHOD(device_detach, vmbus_detach), DEVMETHOD(device_shutdown, bus_generic_shutdown), DEVMETHOD(device_suspend, bus_generic_suspend), DEVMETHOD(device_resume, bus_generic_resume), /** Bus interface */ DEVMETHOD(bus_add_child, bus_generic_add_child), DEVMETHOD(bus_print_child, bus_generic_print_child), DEVMETHOD(bus_read_ivar, vmbus_read_ivar), DEVMETHOD(bus_write_ivar, vmbus_write_ivar), { 0, 0 } }; static char driver_name[] = "vmbus"; static driver_t vmbus_driver = { driver_name, vmbus_methods,0, }; devclass_t vmbus_devclass; DRIVER_MODULE(vmbus, acpi, vmbus_driver, vmbus_devclass, vmbus_modevent, 0); MODULE_DEPEND(vmbus, acpi, 1, 1, 1); MODULE_VERSION(vmbus, 1); /* We want to be started after SMP is initialized */ SYSINIT(vmb_init, SI_SUB_SMP + 1, SI_ORDER_FIRST, vmbus_init, NULL);