Index: head/sys/dev/hyperv/vmbus/hv_channel.c =================================================================== --- head/sys/dev/hyperv/vmbus/hv_channel.c (revision 298692) +++ head/sys/dev/hyperv/vmbus/hv_channel.c (revision 298693) @@ -1,1027 +1,1027 @@ /*- * Copyright (c) 2009-2012,2016 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 __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include "hv_vmbus_priv.h" static int vmbus_channel_create_gpadl_header( /* must be phys and virt contiguous*/ void* contig_buffer, /* page-size multiple */ uint32_t size, hv_vmbus_channel_msg_info** msg_info, uint32_t* message_count); static void vmbus_channel_set_event(hv_vmbus_channel* channel); static void VmbusProcessChannelEvent(void* channel, int pending); /** * @brief Trigger an event notification on the specified channel */ static void vmbus_channel_set_event(hv_vmbus_channel *channel) { hv_vmbus_monitor_page *monitor_page; if (channel->offer_msg.monitor_allocated) { /* Each uint32_t represents 32 channels */ synch_set_bit((channel->offer_msg.child_rel_id & 31), ((uint32_t *)hv_vmbus_g_connection.send_interrupt_page + ((channel->offer_msg.child_rel_id >> 5)))); monitor_page = (hv_vmbus_monitor_page *) hv_vmbus_g_connection.monitor_page_2; synch_set_bit(channel->monitor_bit, (uint32_t *)&monitor_page-> trigger_group[channel->monitor_group].u.pending); } else { hv_vmbus_set_event(channel); } } static int vmbus_channel_sysctl_monalloc(SYSCTL_HANDLER_ARGS) { struct hv_vmbus_channel *chan = arg1; int alloc = 0; if (chan->offer_msg.monitor_allocated) alloc = 1; return sysctl_handle_int(oidp, &alloc, 0, req); } static void vmbus_channel_sysctl_create(hv_vmbus_channel* channel) { device_t dev; struct sysctl_oid *devch_sysctl; struct sysctl_oid *devch_id_sysctl, *devch_sub_sysctl; struct sysctl_oid *devch_id_in_sysctl, *devch_id_out_sysctl; struct sysctl_ctx_list *ctx; uint32_t ch_id; uint16_t sub_ch_id; char name[16]; hv_vmbus_channel* primary_ch = channel->primary_channel; if (primary_ch == NULL) { dev = channel->device->device; ch_id = channel->offer_msg.child_rel_id; } else { dev = primary_ch->device->device; ch_id = primary_ch->offer_msg.child_rel_id; sub_ch_id = channel->offer_msg.offer.sub_channel_index; } ctx = device_get_sysctl_ctx(dev); /* This creates dev.DEVNAME.DEVUNIT.channel tree */ devch_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), - OID_AUTO, "channel", CTLFLAG_RD, 0, ""); + OID_AUTO, "channel", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); /* This creates dev.DEVNAME.DEVUNIT.channel.CHANID tree */ snprintf(name, sizeof(name), "%d", ch_id); devch_id_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(devch_sysctl), - OID_AUTO, name, CTLFLAG_RD, 0, ""); + OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); if (primary_ch != NULL) { devch_sub_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(devch_id_sysctl), - OID_AUTO, "sub", CTLFLAG_RD, 0, ""); + OID_AUTO, "sub", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); snprintf(name, sizeof(name), "%d", sub_ch_id); devch_id_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(devch_sub_sysctl), - OID_AUTO, name, CTLFLAG_RD, 0, ""); + OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(devch_id_sysctl), OID_AUTO, "chanid", CTLFLAG_RD, &channel->offer_msg.child_rel_id, 0, "channel id"); } SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(devch_id_sysctl), OID_AUTO, "cpu", CTLFLAG_RD, &channel->target_cpu, 0, "owner CPU id"); SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(devch_id_sysctl), OID_AUTO, - "monitor_allocated", CTLTYPE_INT | CTLFLAG_RD, channel, 0, - vmbus_channel_sysctl_monalloc, "I", + "monitor_allocated", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, + channel, 0, vmbus_channel_sysctl_monalloc, "I", "is monitor allocated to this channel"); devch_id_in_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(devch_id_sysctl), OID_AUTO, "in", - CTLFLAG_RD, 0, ""); + CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); devch_id_out_sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(devch_id_sysctl), OID_AUTO, "out", - CTLFLAG_RD, 0, ""); + CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); hv_ring_buffer_stat(ctx, SYSCTL_CHILDREN(devch_id_in_sysctl), &(channel->inbound), "inbound ring buffer stats"); hv_ring_buffer_stat(ctx, SYSCTL_CHILDREN(devch_id_out_sysctl), &(channel->outbound), "outbound ring buffer stats"); } /** * @brief Open the specified channel */ int hv_vmbus_channel_open( hv_vmbus_channel* new_channel, uint32_t send_ring_buffer_size, uint32_t recv_ring_buffer_size, void* user_data, uint32_t user_data_len, hv_vmbus_pfn_channel_callback pfn_on_channel_callback, void* context) { int ret = 0; void *in, *out; hv_vmbus_channel_open_channel* open_msg; hv_vmbus_channel_msg_info* open_info; mtx_lock(&new_channel->sc_lock); if (new_channel->state == HV_CHANNEL_OPEN_STATE) { new_channel->state = HV_CHANNEL_OPENING_STATE; } else { mtx_unlock(&new_channel->sc_lock); if(bootverbose) printf("VMBUS: Trying to open channel <%p> which in " "%d state.\n", new_channel, new_channel->state); return (EINVAL); } mtx_unlock(&new_channel->sc_lock); new_channel->on_channel_callback = pfn_on_channel_callback; new_channel->channel_callback_context = context; new_channel->rxq = hv_vmbus_g_context.hv_event_queue[new_channel->target_cpu]; TASK_INIT(&new_channel->channel_task, 0, VmbusProcessChannelEvent, new_channel); /* Allocate the ring buffer */ out = contigmalloc((send_ring_buffer_size + recv_ring_buffer_size), M_DEVBUF, M_ZERO, 0UL, BUS_SPACE_MAXADDR, PAGE_SIZE, 0); KASSERT(out != NULL, ("Error VMBUS: contigmalloc failed to allocate Ring Buffer!")); if (out == NULL) return (ENOMEM); in = ((uint8_t *) out + send_ring_buffer_size); new_channel->ring_buffer_pages = out; new_channel->ring_buffer_page_count = (send_ring_buffer_size + recv_ring_buffer_size) >> PAGE_SHIFT; new_channel->ring_buffer_size = send_ring_buffer_size + recv_ring_buffer_size; hv_vmbus_ring_buffer_init( &new_channel->outbound, out, send_ring_buffer_size); hv_vmbus_ring_buffer_init( &new_channel->inbound, in, recv_ring_buffer_size); /* Create sysctl tree for this channel */ vmbus_channel_sysctl_create(new_channel); /** * Establish the gpadl for the ring buffer */ new_channel->ring_buffer_gpadl_handle = 0; ret = hv_vmbus_channel_establish_gpadl(new_channel, new_channel->outbound.ring_buffer, send_ring_buffer_size + recv_ring_buffer_size, &new_channel->ring_buffer_gpadl_handle); /** * Create and init the channel open message */ open_info = (hv_vmbus_channel_msg_info*) malloc( sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_open_channel), M_DEVBUF, M_NOWAIT); KASSERT(open_info != NULL, ("Error VMBUS: malloc failed to allocate Open Channel message!")); if (open_info == NULL) return (ENOMEM); sema_init(&open_info->wait_sema, 0, "Open Info Sema"); open_msg = (hv_vmbus_channel_open_channel*) open_info->msg; open_msg->header.message_type = HV_CHANNEL_MESSAGE_OPEN_CHANNEL; open_msg->open_id = new_channel->offer_msg.child_rel_id; open_msg->child_rel_id = new_channel->offer_msg.child_rel_id; open_msg->ring_buffer_gpadl_handle = new_channel->ring_buffer_gpadl_handle; open_msg->downstream_ring_buffer_page_offset = send_ring_buffer_size >> PAGE_SHIFT; open_msg->target_vcpu = new_channel->target_vcpu; if (user_data_len) memcpy(open_msg->user_data, user_data, user_data_len); mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_INSERT_TAIL( &hv_vmbus_g_connection.channel_msg_anchor, open_info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); ret = hv_vmbus_post_message( open_msg, sizeof(hv_vmbus_channel_open_channel)); if (ret != 0) goto cleanup; ret = sema_timedwait(&open_info->wait_sema, 5 * hz); /* KYS 5 seconds */ if (ret) { if(bootverbose) printf("VMBUS: channel <%p> open timeout.\n", new_channel); goto cleanup; } if (open_info->response.open_result.status == 0) { new_channel->state = HV_CHANNEL_OPENED_STATE; if(bootverbose) printf("VMBUS: channel <%p> open success.\n", new_channel); } else { if(bootverbose) printf("Error VMBUS: channel <%p> open failed - %d!\n", new_channel, open_info->response.open_result.status); } cleanup: mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_REMOVE( &hv_vmbus_g_connection.channel_msg_anchor, open_info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); sema_destroy(&open_info->wait_sema); free(open_info, M_DEVBUF); return (ret); } /** * @brief Create a gpadl for the specified buffer */ static int vmbus_channel_create_gpadl_header( void* contig_buffer, uint32_t size, /* page-size multiple */ hv_vmbus_channel_msg_info** msg_info, uint32_t* message_count) { int i; int page_count; unsigned long long pfn; uint32_t msg_size; hv_vmbus_channel_gpadl_header* gpa_header; hv_vmbus_channel_gpadl_body* gpadl_body; hv_vmbus_channel_msg_info* msg_header; hv_vmbus_channel_msg_info* msg_body; int pfnSum, pfnCount, pfnLeft, pfnCurr, pfnSize; page_count = size >> PAGE_SHIFT; pfn = hv_get_phys_addr(contig_buffer) >> PAGE_SHIFT; /*do we need a gpadl body msg */ pfnSize = HV_MAX_SIZE_CHANNEL_MESSAGE - sizeof(hv_vmbus_channel_gpadl_header) - sizeof(hv_gpa_range); pfnCount = pfnSize / sizeof(uint64_t); if (page_count > pfnCount) { /* if(we need a gpadl body) */ /* fill in the header */ msg_size = sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_gpadl_header) + sizeof(hv_gpa_range) + pfnCount * sizeof(uint64_t); msg_header = malloc(msg_size, M_DEVBUF, M_NOWAIT | M_ZERO); KASSERT( msg_header != NULL, ("Error VMBUS: malloc failed to allocate Gpadl Message!")); if (msg_header == NULL) return (ENOMEM); TAILQ_INIT(&msg_header->sub_msg_list_anchor); msg_header->message_size = msg_size; gpa_header = (hv_vmbus_channel_gpadl_header*) msg_header->msg; gpa_header->range_count = 1; gpa_header->range_buf_len = sizeof(hv_gpa_range) + page_count * sizeof(uint64_t); gpa_header->range[0].byte_offset = 0; gpa_header->range[0].byte_count = size; for (i = 0; i < pfnCount; i++) { gpa_header->range[0].pfn_array[i] = pfn + i; } *msg_info = msg_header; *message_count = 1; pfnSum = pfnCount; pfnLeft = page_count - pfnCount; /* * figure out how many pfns we can fit */ pfnSize = HV_MAX_SIZE_CHANNEL_MESSAGE - sizeof(hv_vmbus_channel_gpadl_body); pfnCount = pfnSize / sizeof(uint64_t); /* * fill in the body */ while (pfnLeft) { if (pfnLeft > pfnCount) { pfnCurr = pfnCount; } else { pfnCurr = pfnLeft; } msg_size = sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_gpadl_body) + pfnCurr * sizeof(uint64_t); msg_body = malloc(msg_size, M_DEVBUF, M_NOWAIT | M_ZERO); KASSERT( msg_body != NULL, ("Error VMBUS: malloc failed to allocate Gpadl msg_body!")); if (msg_body == NULL) return (ENOMEM); msg_body->message_size = msg_size; (*message_count)++; gpadl_body = (hv_vmbus_channel_gpadl_body*) msg_body->msg; /* * gpadl_body->gpadl = kbuffer; */ for (i = 0; i < pfnCurr; i++) { gpadl_body->pfn[i] = pfn + pfnSum + i; } TAILQ_INSERT_TAIL( &msg_header->sub_msg_list_anchor, msg_body, msg_list_entry); pfnSum += pfnCurr; pfnLeft -= pfnCurr; } } else { /* else everything fits in a header */ msg_size = sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_gpadl_header) + sizeof(hv_gpa_range) + page_count * sizeof(uint64_t); msg_header = malloc(msg_size, M_DEVBUF, M_NOWAIT | M_ZERO); KASSERT( msg_header != NULL, ("Error VMBUS: malloc failed to allocate Gpadl Message!")); if (msg_header == NULL) return (ENOMEM); msg_header->message_size = msg_size; gpa_header = (hv_vmbus_channel_gpadl_header*) msg_header->msg; gpa_header->range_count = 1; gpa_header->range_buf_len = sizeof(hv_gpa_range) + page_count * sizeof(uint64_t); gpa_header->range[0].byte_offset = 0; gpa_header->range[0].byte_count = size; for (i = 0; i < page_count; i++) { gpa_header->range[0].pfn_array[i] = pfn + i; } *msg_info = msg_header; *message_count = 1; } return (0); } /** * @brief Establish a GPADL for the specified buffer */ int hv_vmbus_channel_establish_gpadl( hv_vmbus_channel* channel, void* contig_buffer, uint32_t size, /* page-size multiple */ uint32_t* gpadl_handle) { int ret = 0; hv_vmbus_channel_gpadl_header* gpadl_msg; hv_vmbus_channel_gpadl_body* gpadl_body; hv_vmbus_channel_msg_info* msg_info; hv_vmbus_channel_msg_info* sub_msg_info; uint32_t msg_count; hv_vmbus_channel_msg_info* curr; uint32_t next_gpadl_handle; next_gpadl_handle = atomic_fetchadd_int( &hv_vmbus_g_connection.next_gpadl_handle, 1); ret = vmbus_channel_create_gpadl_header( contig_buffer, size, &msg_info, &msg_count); if(ret != 0) { /* * XXX * We can _not_ even revert the above incremental, * if multiple GPADL establishments are running * parallelly, decrement the global next_gpadl_handle * is calling for _big_ trouble. A better solution * is to have a 0-based GPADL id bitmap ... */ return ret; } sema_init(&msg_info->wait_sema, 0, "Open Info Sema"); gpadl_msg = (hv_vmbus_channel_gpadl_header*) msg_info->msg; gpadl_msg->header.message_type = HV_CHANNEL_MESSAGEL_GPADL_HEADER; gpadl_msg->child_rel_id = channel->offer_msg.child_rel_id; gpadl_msg->gpadl = next_gpadl_handle; mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_INSERT_TAIL( &hv_vmbus_g_connection.channel_msg_anchor, msg_info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); ret = hv_vmbus_post_message( gpadl_msg, msg_info->message_size - (uint32_t) sizeof(hv_vmbus_channel_msg_info)); if (ret != 0) goto cleanup; if (msg_count > 1) { TAILQ_FOREACH(curr, &msg_info->sub_msg_list_anchor, msg_list_entry) { sub_msg_info = curr; gpadl_body = (hv_vmbus_channel_gpadl_body*) sub_msg_info->msg; gpadl_body->header.message_type = HV_CHANNEL_MESSAGE_GPADL_BODY; gpadl_body->gpadl = next_gpadl_handle; ret = hv_vmbus_post_message( gpadl_body, sub_msg_info->message_size - (uint32_t) sizeof(hv_vmbus_channel_msg_info)); /* if (the post message failed) give up and clean up */ if(ret != 0) goto cleanup; } } ret = sema_timedwait(&msg_info->wait_sema, 5 * hz); /* KYS 5 seconds*/ if (ret != 0) goto cleanup; *gpadl_handle = gpadl_msg->gpadl; cleanup: mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_REMOVE(&hv_vmbus_g_connection.channel_msg_anchor, msg_info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); sema_destroy(&msg_info->wait_sema); free(msg_info, M_DEVBUF); return (ret); } /** * @brief Teardown the specified GPADL handle */ int hv_vmbus_channel_teardown_gpdal( hv_vmbus_channel* channel, uint32_t gpadl_handle) { int ret = 0; hv_vmbus_channel_gpadl_teardown* msg; hv_vmbus_channel_msg_info* info; info = (hv_vmbus_channel_msg_info *) malloc( sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_gpadl_teardown), M_DEVBUF, M_NOWAIT); KASSERT(info != NULL, ("Error VMBUS: malloc failed to allocate Gpadl Teardown Msg!")); if (info == NULL) { ret = ENOMEM; goto cleanup; } sema_init(&info->wait_sema, 0, "Open Info Sema"); msg = (hv_vmbus_channel_gpadl_teardown*) info->msg; msg->header.message_type = HV_CHANNEL_MESSAGE_GPADL_TEARDOWN; msg->child_rel_id = channel->offer_msg.child_rel_id; msg->gpadl = gpadl_handle; mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_INSERT_TAIL(&hv_vmbus_g_connection.channel_msg_anchor, info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); ret = hv_vmbus_post_message(msg, sizeof(hv_vmbus_channel_gpadl_teardown)); if (ret != 0) goto cleanup; ret = sema_timedwait(&info->wait_sema, 5 * hz); /* KYS 5 seconds */ cleanup: /* * Received a torndown response */ mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); TAILQ_REMOVE(&hv_vmbus_g_connection.channel_msg_anchor, info, msg_list_entry); mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); sema_destroy(&info->wait_sema); free(info, M_DEVBUF); return (ret); } static void hv_vmbus_channel_close_internal(hv_vmbus_channel *channel) { int ret = 0; struct taskqueue *rxq = channel->rxq; hv_vmbus_channel_close_channel* msg; hv_vmbus_channel_msg_info* info; channel->state = HV_CHANNEL_OPEN_STATE; /* * set rxq to NULL to avoid more requests be scheduled */ channel->rxq = NULL; taskqueue_drain(rxq, &channel->channel_task); channel->on_channel_callback = NULL; /** * Send a closing message */ info = (hv_vmbus_channel_msg_info *) malloc( sizeof(hv_vmbus_channel_msg_info) + sizeof(hv_vmbus_channel_close_channel), M_DEVBUF, M_NOWAIT); KASSERT(info != NULL, ("VMBUS: malloc failed hv_vmbus_channel_close!")); if(info == NULL) return; msg = (hv_vmbus_channel_close_channel*) info->msg; msg->header.message_type = HV_CHANNEL_MESSAGE_CLOSE_CHANNEL; msg->child_rel_id = channel->offer_msg.child_rel_id; ret = hv_vmbus_post_message( msg, sizeof(hv_vmbus_channel_close_channel)); /* Tear down the gpadl for the channel's ring buffer */ if (channel->ring_buffer_gpadl_handle) { hv_vmbus_channel_teardown_gpdal(channel, channel->ring_buffer_gpadl_handle); } /* TODO: Send a msg to release the childRelId */ /* cleanup the ring buffers for this channel */ hv_ring_buffer_cleanup(&channel->outbound); hv_ring_buffer_cleanup(&channel->inbound); contigfree(channel->ring_buffer_pages, channel->ring_buffer_size, M_DEVBUF); free(info, M_DEVBUF); } /** * @brief Close the specified channel */ void hv_vmbus_channel_close(hv_vmbus_channel *channel) { hv_vmbus_channel* sub_channel; if (channel->primary_channel != NULL) { /* * We only close multi-channels when the primary is * closed. */ return; } /* * Close all multi-channels first. */ TAILQ_FOREACH(sub_channel, &channel->sc_list_anchor, sc_list_entry) { if (sub_channel->state != HV_CHANNEL_OPENED_STATE) continue; hv_vmbus_channel_close_internal(sub_channel); } /* * Then close the primary channel. */ hv_vmbus_channel_close_internal(channel); } /** * @brief Send the specified buffer on the given channel */ int hv_vmbus_channel_send_packet( hv_vmbus_channel* channel, void* buffer, uint32_t buffer_len, uint64_t request_id, hv_vmbus_packet_type type, uint32_t flags) { int ret = 0; hv_vm_packet_descriptor desc; uint32_t packet_len; uint64_t aligned_data; uint32_t packet_len_aligned; boolean_t need_sig; hv_vmbus_sg_buffer_list buffer_list[3]; packet_len = sizeof(hv_vm_packet_descriptor) + buffer_len; packet_len_aligned = HV_ALIGN_UP(packet_len, sizeof(uint64_t)); aligned_data = 0; /* Setup the descriptor */ desc.type = type; /* HV_VMBUS_PACKET_TYPE_DATA_IN_BAND; */ desc.flags = flags; /* HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED */ /* in 8-bytes granularity */ desc.data_offset8 = sizeof(hv_vm_packet_descriptor) >> 3; desc.length8 = (uint16_t) (packet_len_aligned >> 3); desc.transaction_id = request_id; buffer_list[0].data = &desc; buffer_list[0].length = sizeof(hv_vm_packet_descriptor); buffer_list[1].data = buffer; buffer_list[1].length = buffer_len; buffer_list[2].data = &aligned_data; buffer_list[2].length = packet_len_aligned - packet_len; ret = hv_ring_buffer_write(&channel->outbound, buffer_list, 3, &need_sig); /* TODO: We should determine if this is optional */ if (ret == 0 && need_sig) { vmbus_channel_set_event(channel); } return (ret); } /** * @brief Send a range of single-page buffer packets using * a GPADL Direct packet type */ int hv_vmbus_channel_send_packet_pagebuffer( hv_vmbus_channel* channel, hv_vmbus_page_buffer page_buffers[], uint32_t page_count, void* buffer, uint32_t buffer_len, uint64_t request_id) { int ret = 0; boolean_t need_sig; uint32_t packet_len; uint32_t page_buflen; uint32_t packetLen_aligned; hv_vmbus_sg_buffer_list buffer_list[4]; hv_vmbus_channel_packet_page_buffer desc; uint32_t descSize; uint64_t alignedData = 0; if (page_count > HV_MAX_PAGE_BUFFER_COUNT) return (EINVAL); /* * Adjust the size down since hv_vmbus_channel_packet_page_buffer * is the largest size we support */ descSize = __offsetof(hv_vmbus_channel_packet_page_buffer, range); page_buflen = sizeof(hv_vmbus_page_buffer) * page_count; packet_len = descSize + page_buflen + buffer_len; packetLen_aligned = HV_ALIGN_UP(packet_len, sizeof(uint64_t)); /* Setup the descriptor */ desc.type = HV_VMBUS_PACKET_TYPE_DATA_USING_GPA_DIRECT; desc.flags = HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; /* in 8-bytes granularity */ desc.data_offset8 = (descSize + page_buflen) >> 3; desc.length8 = (uint16_t) (packetLen_aligned >> 3); desc.transaction_id = request_id; desc.range_count = page_count; buffer_list[0].data = &desc; buffer_list[0].length = descSize; buffer_list[1].data = page_buffers; buffer_list[1].length = page_buflen; buffer_list[2].data = buffer; buffer_list[2].length = buffer_len; buffer_list[3].data = &alignedData; buffer_list[3].length = packetLen_aligned - packet_len; ret = hv_ring_buffer_write(&channel->outbound, buffer_list, 4, &need_sig); /* TODO: We should determine if this is optional */ if (ret == 0 && need_sig) { vmbus_channel_set_event(channel); } return (ret); } /** * @brief Send a multi-page buffer packet using a GPADL Direct packet type */ int hv_vmbus_channel_send_packet_multipagebuffer( hv_vmbus_channel* channel, hv_vmbus_multipage_buffer* multi_page_buffer, void* buffer, uint32_t buffer_len, uint64_t request_id) { int ret = 0; uint32_t desc_size; boolean_t need_sig; uint32_t packet_len; uint32_t packet_len_aligned; uint32_t pfn_count; uint64_t aligned_data = 0; hv_vmbus_sg_buffer_list buffer_list[3]; hv_vmbus_channel_packet_multipage_buffer desc; pfn_count = HV_NUM_PAGES_SPANNED( multi_page_buffer->offset, multi_page_buffer->length); if ((pfn_count == 0) || (pfn_count > HV_MAX_MULTIPAGE_BUFFER_COUNT)) return (EINVAL); /* * Adjust the size down since hv_vmbus_channel_packet_multipage_buffer * is the largest size we support */ desc_size = sizeof(hv_vmbus_channel_packet_multipage_buffer) - ((HV_MAX_MULTIPAGE_BUFFER_COUNT - pfn_count) * sizeof(uint64_t)); packet_len = desc_size + buffer_len; packet_len_aligned = HV_ALIGN_UP(packet_len, sizeof(uint64_t)); /* * Setup the descriptor */ desc.type = HV_VMBUS_PACKET_TYPE_DATA_USING_GPA_DIRECT; desc.flags = HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; desc.data_offset8 = desc_size >> 3; /* in 8-bytes granularity */ desc.length8 = (uint16_t) (packet_len_aligned >> 3); desc.transaction_id = request_id; desc.range_count = 1; desc.range.length = multi_page_buffer->length; desc.range.offset = multi_page_buffer->offset; memcpy(desc.range.pfn_array, multi_page_buffer->pfn_array, pfn_count * sizeof(uint64_t)); buffer_list[0].data = &desc; buffer_list[0].length = desc_size; buffer_list[1].data = buffer; buffer_list[1].length = buffer_len; buffer_list[2].data = &aligned_data; buffer_list[2].length = packet_len_aligned - packet_len; ret = hv_ring_buffer_write(&channel->outbound, buffer_list, 3, &need_sig); /* TODO: We should determine if this is optional */ if (ret == 0 && need_sig) { vmbus_channel_set_event(channel); } return (ret); } /** * @brief Retrieve the user packet on the specified channel */ int hv_vmbus_channel_recv_packet( hv_vmbus_channel* channel, void* Buffer, uint32_t buffer_len, uint32_t* buffer_actual_len, uint64_t* request_id) { int ret; uint32_t user_len; uint32_t packet_len; hv_vm_packet_descriptor desc; *buffer_actual_len = 0; *request_id = 0; ret = hv_ring_buffer_peek(&channel->inbound, &desc, sizeof(hv_vm_packet_descriptor)); if (ret != 0) return (0); packet_len = desc.length8 << 3; user_len = packet_len - (desc.data_offset8 << 3); *buffer_actual_len = user_len; if (user_len > buffer_len) return (EINVAL); *request_id = desc.transaction_id; /* Copy over the packet to the user buffer */ ret = hv_ring_buffer_read(&channel->inbound, Buffer, user_len, (desc.data_offset8 << 3)); return (0); } /** * @brief Retrieve the raw packet on the specified channel */ int hv_vmbus_channel_recv_packet_raw( hv_vmbus_channel* channel, void* buffer, uint32_t buffer_len, uint32_t* buffer_actual_len, uint64_t* request_id) { int ret; uint32_t packetLen; hv_vm_packet_descriptor desc; *buffer_actual_len = 0; *request_id = 0; ret = hv_ring_buffer_peek( &channel->inbound, &desc, sizeof(hv_vm_packet_descriptor)); if (ret != 0) return (0); packetLen = desc.length8 << 3; *buffer_actual_len = packetLen; if (packetLen > buffer_len) return (ENOBUFS); *request_id = desc.transaction_id; /* Copy over the entire packet to the user buffer */ ret = hv_ring_buffer_read(&channel->inbound, buffer, packetLen, 0); return (0); } /** * Process a channel event notification */ static void VmbusProcessChannelEvent(void* context, int pending) { void* arg; uint32_t bytes_to_read; hv_vmbus_channel* channel = (hv_vmbus_channel*)context; boolean_t is_batched_reading; /** * Find the channel based on this relid and invokes * the channel callback to process the event */ if (channel == NULL) { return; } /** * To deal with the race condition where we might * receive a packet while the relevant driver is * being unloaded, dispatch the callback while * holding the channel lock. The unloading driver * will acquire the same channel lock to set the * callback to NULL. This closes the window. */ if (channel->on_channel_callback != NULL) { arg = channel->channel_callback_context; is_batched_reading = channel->batched_reading; /* * Optimize host to guest signaling by ensuring: * 1. While reading the channel, we disable interrupts from * host. * 2. Ensure that we process all posted messages from the host * before returning from this callback. * 3. Once we return, enable signaling from the host. Once this * state is set we check to see if additional packets are * available to read. In this case we repeat the process. */ do { if (is_batched_reading) hv_ring_buffer_read_begin(&channel->inbound); channel->on_channel_callback(arg); if (is_batched_reading) bytes_to_read = hv_ring_buffer_read_end(&channel->inbound); else bytes_to_read = 0; } while (is_batched_reading && (bytes_to_read != 0)); } } Index: head/sys/dev/hyperv/vmbus/hv_ring_buffer.c =================================================================== --- head/sys/dev/hyperv/vmbus/hv_ring_buffer.c (revision 298692) +++ head/sys/dev/hyperv/vmbus/hv_ring_buffer.c (revision 298693) @@ -1,552 +1,552 @@ /*- * Copyright (c) 2009-2012,2016 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 "hv_vmbus_priv.h" /* Amount of space to write to */ #define HV_BYTES_AVAIL_TO_WRITE(r, w, z) ((w) >= (r))? \ ((z) - ((w) - (r))):((r) - (w)) static int hv_rbi_sysctl_stats(SYSCTL_HANDLER_ARGS) { hv_vmbus_ring_buffer_info* rbi; uint32_t read_index, write_index, interrupt_mask, sz; uint32_t read_avail, write_avail; char rbi_stats[256]; rbi = (hv_vmbus_ring_buffer_info*)arg1; read_index = rbi->ring_buffer->read_index; write_index = rbi->ring_buffer->write_index; interrupt_mask = rbi->ring_buffer->interrupt_mask; sz = rbi->ring_data_size; write_avail = HV_BYTES_AVAIL_TO_WRITE(read_index, write_index, sz); read_avail = sz - write_avail; snprintf(rbi_stats, sizeof(rbi_stats), "r_idx:%d " "w_idx:%d " "int_mask:%d " "r_avail:%d " "w_avail:%d", read_index, write_index, interrupt_mask, read_avail, write_avail); return (sysctl_handle_string(oidp, rbi_stats, sizeof(rbi_stats), req)); } void hv_ring_buffer_stat( struct sysctl_ctx_list *ctx, struct sysctl_oid_list *tree_node, hv_vmbus_ring_buffer_info *rbi, const char *desc) { SYSCTL_ADD_PROC(ctx, tree_node, OID_AUTO, "ring_buffer_stats", - CTLTYPE_STRING|CTLFLAG_RD, rbi, 0, + CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE, rbi, 0, hv_rbi_sysctl_stats, "A", desc); } /** * @brief Get number of bytes available to read and to write to * for the specified ring buffer */ static inline void get_ring_buffer_avail_bytes( hv_vmbus_ring_buffer_info* rbi, uint32_t* read, uint32_t* write) { uint32_t read_loc, write_loc; /* * Capture the read/write indices before they changed */ read_loc = rbi->ring_buffer->read_index; write_loc = rbi->ring_buffer->write_index; *write = HV_BYTES_AVAIL_TO_WRITE( read_loc, write_loc, rbi->ring_data_size); *read = rbi->ring_data_size - *write; } /** * @brief Get the next write location for the specified ring buffer */ static inline uint32_t get_next_write_location(hv_vmbus_ring_buffer_info* ring_info) { uint32_t next = ring_info->ring_buffer->write_index; return (next); } /** * @brief Set the next write location for the specified ring buffer */ static inline void set_next_write_location( hv_vmbus_ring_buffer_info* ring_info, uint32_t next_write_location) { ring_info->ring_buffer->write_index = next_write_location; } /** * @brief Get the next read location for the specified ring buffer */ static inline uint32_t get_next_read_location(hv_vmbus_ring_buffer_info* ring_info) { uint32_t next = ring_info->ring_buffer->read_index; return (next); } /** * @brief Get the next read location + offset for the specified ring buffer. * This allows the caller to skip. */ static inline uint32_t get_next_read_location_with_offset( hv_vmbus_ring_buffer_info* ring_info, uint32_t offset) { uint32_t next = ring_info->ring_buffer->read_index; next += offset; next %= ring_info->ring_data_size; return (next); } /** * @brief Set the next read location for the specified ring buffer */ static inline void set_next_read_location( hv_vmbus_ring_buffer_info* ring_info, uint32_t next_read_location) { ring_info->ring_buffer->read_index = next_read_location; } /** * @brief Get the start of the ring buffer */ static inline void * get_ring_buffer(hv_vmbus_ring_buffer_info* ring_info) { return (void *) ring_info->ring_buffer->buffer; } /** * @brief Get the size of the ring buffer. */ static inline uint32_t get_ring_buffer_size(hv_vmbus_ring_buffer_info* ring_info) { return ring_info->ring_data_size; } /** * Get the read and write indices as uint64_t of the specified ring buffer. */ static inline uint64_t get_ring_buffer_indices(hv_vmbus_ring_buffer_info* ring_info) { return (uint64_t) ring_info->ring_buffer->write_index << 32; } void hv_ring_buffer_read_begin( hv_vmbus_ring_buffer_info* ring_info) { ring_info->ring_buffer->interrupt_mask = 1; mb(); } uint32_t hv_ring_buffer_read_end( hv_vmbus_ring_buffer_info* ring_info) { uint32_t read, write; ring_info->ring_buffer->interrupt_mask = 0; mb(); /* * Now check to see if the ring buffer is still empty. * If it is not, we raced and we need to process new * incoming messages. */ get_ring_buffer_avail_bytes(ring_info, &read, &write); return (read); } /* * When we write to the ring buffer, check if the host needs to * be signaled. Here is the details of this protocol: * * 1. The host guarantees that while it is draining the * ring buffer, it will set the interrupt_mask to * indicate it does not need to be interrupted when * new data is placed. * * 2. The host guarantees that it will completely drain * the ring buffer before exiting the read loop. Further, * once the ring buffer is empty, it will clear the * interrupt_mask and re-check to see if new data has * arrived. */ static boolean_t hv_ring_buffer_needsig_on_write( uint32_t old_write_location, hv_vmbus_ring_buffer_info* rbi) { mb(); if (rbi->ring_buffer->interrupt_mask) return (FALSE); /* Read memory barrier */ rmb(); /* * This is the only case we need to signal when the * ring transitions from being empty to non-empty. */ if (old_write_location == rbi->ring_buffer->read_index) return (TRUE); return (FALSE); } static uint32_t copy_to_ring_buffer( hv_vmbus_ring_buffer_info* ring_info, uint32_t start_write_offset, char* src, uint32_t src_len); static uint32_t copy_from_ring_buffer( hv_vmbus_ring_buffer_info* ring_info, char* dest, uint32_t dest_len, uint32_t start_read_offset); /** * @brief Get the interrupt mask for the specified ring buffer. */ uint32_t hv_vmbus_get_ring_buffer_interrupt_mask(hv_vmbus_ring_buffer_info *rbi) { return rbi->ring_buffer->interrupt_mask; } /** * @brief Initialize the ring buffer. */ int hv_vmbus_ring_buffer_init( hv_vmbus_ring_buffer_info* ring_info, void* buffer, uint32_t buffer_len) { memset(ring_info, 0, sizeof(hv_vmbus_ring_buffer_info)); ring_info->ring_buffer = (hv_vmbus_ring_buffer*) buffer; ring_info->ring_buffer->read_index = ring_info->ring_buffer->write_index = 0; ring_info->ring_size = buffer_len; ring_info->ring_data_size = buffer_len - sizeof(hv_vmbus_ring_buffer); mtx_init(&ring_info->ring_lock, "vmbus ring buffer", NULL, MTX_SPIN); return (0); } /** * @brief Cleanup the ring buffer. */ void hv_ring_buffer_cleanup(hv_vmbus_ring_buffer_info* ring_info) { mtx_destroy(&ring_info->ring_lock); } /** * @brief Write to the ring buffer. */ int hv_ring_buffer_write( hv_vmbus_ring_buffer_info* out_ring_info, hv_vmbus_sg_buffer_list sg_buffers[], uint32_t sg_buffer_count, boolean_t *need_sig) { int i = 0; uint32_t byte_avail_to_write; uint32_t byte_avail_to_read; uint32_t old_write_location; uint32_t total_bytes_to_write = 0; volatile uint32_t next_write_location; uint64_t prev_indices = 0; for (i = 0; i < sg_buffer_count; i++) { total_bytes_to_write += sg_buffers[i].length; } total_bytes_to_write += sizeof(uint64_t); mtx_lock_spin(&out_ring_info->ring_lock); get_ring_buffer_avail_bytes(out_ring_info, &byte_avail_to_read, &byte_avail_to_write); /* * If there is only room for the packet, assume it is full. * Otherwise, the next time around, we think the ring buffer * is empty since the read index == write index */ if (byte_avail_to_write <= total_bytes_to_write) { mtx_unlock_spin(&out_ring_info->ring_lock); return (EAGAIN); } /* * Write to the ring buffer */ next_write_location = get_next_write_location(out_ring_info); old_write_location = next_write_location; for (i = 0; i < sg_buffer_count; i++) { next_write_location = copy_to_ring_buffer(out_ring_info, next_write_location, (char *) sg_buffers[i].data, sg_buffers[i].length); } /* * Set previous packet start */ prev_indices = get_ring_buffer_indices(out_ring_info); next_write_location = copy_to_ring_buffer( out_ring_info, next_write_location, (char *) &prev_indices, sizeof(uint64_t)); /* * Full memory barrier before upding the write index. */ mb(); /* * Now, update the write location */ set_next_write_location(out_ring_info, next_write_location); mtx_unlock_spin(&out_ring_info->ring_lock); *need_sig = hv_ring_buffer_needsig_on_write(old_write_location, out_ring_info); return (0); } /** * @brief Read without advancing the read index. */ int hv_ring_buffer_peek( hv_vmbus_ring_buffer_info* in_ring_info, void* buffer, uint32_t buffer_len) { uint32_t bytesAvailToWrite; uint32_t bytesAvailToRead; uint32_t nextReadLocation = 0; mtx_lock_spin(&in_ring_info->ring_lock); get_ring_buffer_avail_bytes(in_ring_info, &bytesAvailToRead, &bytesAvailToWrite); /* * Make sure there is something to read */ if (bytesAvailToRead < buffer_len) { mtx_unlock_spin(&in_ring_info->ring_lock); return (EAGAIN); } /* * Convert to byte offset */ nextReadLocation = get_next_read_location(in_ring_info); nextReadLocation = copy_from_ring_buffer( in_ring_info, (char *)buffer, buffer_len, nextReadLocation); mtx_unlock_spin(&in_ring_info->ring_lock); return (0); } /** * @brief Read and advance the read index. */ int hv_ring_buffer_read( hv_vmbus_ring_buffer_info* in_ring_info, void* buffer, uint32_t buffer_len, uint32_t offset) { uint32_t bytes_avail_to_write; uint32_t bytes_avail_to_read; uint32_t next_read_location = 0; uint64_t prev_indices = 0; if (buffer_len <= 0) return (EINVAL); mtx_lock_spin(&in_ring_info->ring_lock); get_ring_buffer_avail_bytes( in_ring_info, &bytes_avail_to_read, &bytes_avail_to_write); /* * Make sure there is something to read */ if (bytes_avail_to_read < buffer_len) { mtx_unlock_spin(&in_ring_info->ring_lock); return (EAGAIN); } next_read_location = get_next_read_location_with_offset( in_ring_info, offset); next_read_location = copy_from_ring_buffer( in_ring_info, (char *) buffer, buffer_len, next_read_location); next_read_location = copy_from_ring_buffer( in_ring_info, (char *) &prev_indices, sizeof(uint64_t), next_read_location); /* * Make sure all reads are done before we update the read index since * the writer may start writing to the read area once the read index * is updated. */ wmb(); /* * Update the read index */ set_next_read_location(in_ring_info, next_read_location); mtx_unlock_spin(&in_ring_info->ring_lock); return (0); } /** * @brief Helper routine to copy from source to ring buffer. * * Assume there is enough room. Handles wrap-around in dest case only! */ uint32_t copy_to_ring_buffer( hv_vmbus_ring_buffer_info* ring_info, uint32_t start_write_offset, char* src, uint32_t src_len) { char *ring_buffer = get_ring_buffer(ring_info); uint32_t ring_buffer_size = get_ring_buffer_size(ring_info); uint32_t fragLen; if (src_len > ring_buffer_size - start_write_offset) { /* wrap-around detected! */ fragLen = ring_buffer_size - start_write_offset; memcpy(ring_buffer + start_write_offset, src, fragLen); memcpy(ring_buffer, src + fragLen, src_len - fragLen); } else { memcpy(ring_buffer + start_write_offset, src, src_len); } start_write_offset += src_len; start_write_offset %= ring_buffer_size; return (start_write_offset); } /** * @brief Helper routine to copy to source from ring buffer. * * Assume there is enough room. Handles wrap-around in src case only! */ uint32_t copy_from_ring_buffer( hv_vmbus_ring_buffer_info* ring_info, char* dest, uint32_t dest_len, uint32_t start_read_offset) { uint32_t fragLen; char *ring_buffer = get_ring_buffer(ring_info); uint32_t ring_buffer_size = get_ring_buffer_size(ring_info); if (dest_len > ring_buffer_size - start_read_offset) { /* wrap-around detected at the src */ fragLen = ring_buffer_size - start_read_offset; memcpy(dest, ring_buffer + start_read_offset, fragLen); memcpy(dest + fragLen, ring_buffer, dest_len - fragLen); } else { memcpy(dest, ring_buffer + start_read_offset, dest_len); } start_read_offset += dest_len; start_read_offset %= ring_buffer_size; return (start_read_offset); }