hv_connection.c revision 299890
1250199Sgrehan/*- 2298446Ssephe * Copyright (c) 2009-2012,2016 Microsoft Corp. 3250199Sgrehan * Copyright (c) 2012 NetApp Inc. 4250199Sgrehan * Copyright (c) 2012 Citrix Inc. 5250199Sgrehan * All rights reserved. 6250199Sgrehan * 7250199Sgrehan * Redistribution and use in source and binary forms, with or without 8250199Sgrehan * modification, are permitted provided that the following conditions 9250199Sgrehan * are met: 10250199Sgrehan * 1. Redistributions of source code must retain the above copyright 11250199Sgrehan * notice unmodified, this list of conditions, and the following 12250199Sgrehan * disclaimer. 13250199Sgrehan * 2. Redistributions in binary form must reproduce the above copyright 14250199Sgrehan * notice, this list of conditions and the following disclaimer in the 15250199Sgrehan * documentation and/or other materials provided with the distribution. 16250199Sgrehan * 17250199Sgrehan * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18250199Sgrehan * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19250199Sgrehan * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20250199Sgrehan * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21250199Sgrehan * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22250199Sgrehan * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23250199Sgrehan * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24250199Sgrehan * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25250199Sgrehan * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26250199Sgrehan * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27250199Sgrehan */ 28250199Sgrehan 29250199Sgrehan#include <sys/param.h> 30296028Ssephe#include <sys/kernel.h> 31250199Sgrehan#include <sys/malloc.h> 32250199Sgrehan#include <sys/systm.h> 33250199Sgrehan#include <sys/lock.h> 34250199Sgrehan#include <sys/mutex.h> 35250199Sgrehan#include <machine/bus.h> 36250199Sgrehan#include <vm/vm.h> 37250199Sgrehan#include <vm/vm_param.h> 38250199Sgrehan#include <vm/pmap.h> 39250199Sgrehan 40250199Sgrehan#include "hv_vmbus_priv.h" 41250199Sgrehan 42250199Sgrehan/* 43250199Sgrehan * Globals 44250199Sgrehan */ 45250199Sgrehanhv_vmbus_connection hv_vmbus_g_connection = 46250199Sgrehan { .connect_state = HV_DISCONNECTED, 47250199Sgrehan .next_gpadl_handle = 0xE1E10, }; 48250199Sgrehan 49282212Swhuuint32_t hv_vmbus_protocal_version = HV_VMBUS_VERSION_WS2008; 50282212Swhu 51282212Swhustatic uint32_t 52282212Swhuhv_vmbus_get_next_version(uint32_t current_ver) 53282212Swhu{ 54282212Swhu switch (current_ver) { 55282212Swhu case (HV_VMBUS_VERSION_WIN7): 56282212Swhu return(HV_VMBUS_VERSION_WS2008); 57282212Swhu 58282212Swhu case (HV_VMBUS_VERSION_WIN8): 59282212Swhu return(HV_VMBUS_VERSION_WIN7); 60282212Swhu 61282212Swhu case (HV_VMBUS_VERSION_WIN8_1): 62282212Swhu return(HV_VMBUS_VERSION_WIN8); 63282212Swhu 64282212Swhu case (HV_VMBUS_VERSION_WS2008): 65282212Swhu default: 66282212Swhu return(HV_VMBUS_VERSION_INVALID); 67282212Swhu } 68282212Swhu} 69282212Swhu 70250199Sgrehan/** 71282212Swhu * Negotiate the highest supported hypervisor version. 72282212Swhu */ 73282212Swhustatic int 74282212Swhuhv_vmbus_negotiate_version(hv_vmbus_channel_msg_info *msg_info, 75282212Swhu uint32_t version) 76282212Swhu{ 77282212Swhu int ret = 0; 78282212Swhu hv_vmbus_channel_initiate_contact *msg; 79282212Swhu 80282212Swhu sema_init(&msg_info->wait_sema, 0, "Msg Info Sema"); 81282212Swhu msg = (hv_vmbus_channel_initiate_contact*) msg_info->msg; 82282212Swhu 83282212Swhu msg->header.message_type = HV_CHANNEL_MESSAGE_INITIATED_CONTACT; 84282212Swhu msg->vmbus_version_requested = version; 85282212Swhu 86282212Swhu msg->interrupt_page = hv_get_phys_addr( 87282212Swhu hv_vmbus_g_connection.interrupt_page); 88282212Swhu 89282212Swhu msg->monitor_page_1 = hv_get_phys_addr( 90295309Ssephe hv_vmbus_g_connection.monitor_page_1); 91282212Swhu 92295308Ssephe msg->monitor_page_2 = hv_get_phys_addr( 93295309Ssephe hv_vmbus_g_connection.monitor_page_2); 94282212Swhu 95282212Swhu /** 96282212Swhu * Add to list before we send the request since we may receive the 97282212Swhu * response before returning from this routine 98282212Swhu */ 99297635Ssephe mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); 100282212Swhu 101282212Swhu TAILQ_INSERT_TAIL( 102282212Swhu &hv_vmbus_g_connection.channel_msg_anchor, 103282212Swhu msg_info, 104282212Swhu msg_list_entry); 105282212Swhu 106297635Ssephe mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); 107282212Swhu 108282212Swhu ret = hv_vmbus_post_message( 109282212Swhu msg, 110282212Swhu sizeof(hv_vmbus_channel_initiate_contact)); 111282212Swhu 112282212Swhu if (ret != 0) { 113297635Ssephe mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); 114282212Swhu TAILQ_REMOVE( 115282212Swhu &hv_vmbus_g_connection.channel_msg_anchor, 116282212Swhu msg_info, 117282212Swhu msg_list_entry); 118297635Ssephe mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); 119282212Swhu return (ret); 120282212Swhu } 121282212Swhu 122282212Swhu /** 123282212Swhu * Wait for the connection response 124282212Swhu */ 125296028Ssephe ret = sema_timedwait(&msg_info->wait_sema, 5 * hz); /* KYS 5 seconds */ 126282212Swhu 127297635Ssephe mtx_lock(&hv_vmbus_g_connection.channel_msg_lock); 128282212Swhu TAILQ_REMOVE( 129282212Swhu &hv_vmbus_g_connection.channel_msg_anchor, 130282212Swhu msg_info, 131282212Swhu msg_list_entry); 132297635Ssephe mtx_unlock(&hv_vmbus_g_connection.channel_msg_lock); 133282212Swhu 134282212Swhu /** 135282212Swhu * Check if successful 136282212Swhu */ 137282212Swhu if (msg_info->response.version_response.version_supported) { 138282212Swhu hv_vmbus_g_connection.connect_state = HV_CONNECTED; 139282212Swhu } else { 140282212Swhu ret = ECONNREFUSED; 141282212Swhu } 142282212Swhu 143282212Swhu return (ret); 144282212Swhu} 145282212Swhu 146282212Swhu/** 147250199Sgrehan * Send a connect request on the partition service connection 148250199Sgrehan */ 149250199Sgrehanint 150250199Sgrehanhv_vmbus_connect(void) { 151250199Sgrehan int ret = 0; 152282212Swhu uint32_t version; 153250199Sgrehan hv_vmbus_channel_msg_info* msg_info = NULL; 154250199Sgrehan 155250199Sgrehan /** 156250199Sgrehan * Make sure we are not connecting or connected 157250199Sgrehan */ 158250199Sgrehan if (hv_vmbus_g_connection.connect_state != HV_DISCONNECTED) { 159250199Sgrehan return (-1); 160250199Sgrehan } 161250199Sgrehan 162250199Sgrehan /** 163250199Sgrehan * Initialize the vmbus connection 164250199Sgrehan */ 165250199Sgrehan hv_vmbus_g_connection.connect_state = HV_CONNECTING; 166250199Sgrehan 167250199Sgrehan TAILQ_INIT(&hv_vmbus_g_connection.channel_msg_anchor); 168250199Sgrehan mtx_init(&hv_vmbus_g_connection.channel_msg_lock, "vmbus channel msg", 169297635Ssephe NULL, MTX_DEF); 170250199Sgrehan 171250199Sgrehan TAILQ_INIT(&hv_vmbus_g_connection.channel_anchor); 172250199Sgrehan mtx_init(&hv_vmbus_g_connection.channel_lock, "vmbus channel", 173282212Swhu NULL, MTX_DEF); 174250199Sgrehan 175250199Sgrehan /** 176250199Sgrehan * Setup the vmbus event connection for channel interrupt abstraction 177250199Sgrehan * stuff 178250199Sgrehan */ 179295309Ssephe hv_vmbus_g_connection.interrupt_page = malloc( 180250199Sgrehan PAGE_SIZE, M_DEVBUF, 181295309Ssephe M_WAITOK | M_ZERO); 182250199Sgrehan 183250199Sgrehan hv_vmbus_g_connection.recv_interrupt_page = 184250199Sgrehan hv_vmbus_g_connection.interrupt_page; 185250199Sgrehan 186250199Sgrehan hv_vmbus_g_connection.send_interrupt_page = 187250199Sgrehan ((uint8_t *) hv_vmbus_g_connection.interrupt_page + 188250199Sgrehan (PAGE_SIZE >> 1)); 189250199Sgrehan 190250199Sgrehan /** 191250199Sgrehan * Set up the monitor notification facility. The 1st page for 192250199Sgrehan * parent->child and the 2nd page for child->parent 193250199Sgrehan */ 194295309Ssephe hv_vmbus_g_connection.monitor_page_1 = malloc( 195295309Ssephe PAGE_SIZE, 196250199Sgrehan M_DEVBUF, 197295309Ssephe M_WAITOK | M_ZERO); 198295309Ssephe hv_vmbus_g_connection.monitor_page_2 = malloc( 199250199Sgrehan PAGE_SIZE, 200295309Ssephe M_DEVBUF, 201295309Ssephe M_WAITOK | M_ZERO); 202250199Sgrehan 203250199Sgrehan msg_info = (hv_vmbus_channel_msg_info*) 204250199Sgrehan malloc(sizeof(hv_vmbus_channel_msg_info) + 205250199Sgrehan sizeof(hv_vmbus_channel_initiate_contact), 206295308Ssephe M_DEVBUF, M_WAITOK | M_ZERO); 207250199Sgrehan 208294553Ssephe hv_vmbus_g_connection.channels = malloc(sizeof(hv_vmbus_channel*) * 209294553Ssephe HV_CHANNEL_MAX_COUNT, 210294553Ssephe M_DEVBUF, M_WAITOK | M_ZERO); 211282212Swhu /* 212282212Swhu * Find the highest vmbus version number we can support. 213250199Sgrehan */ 214282212Swhu version = HV_VMBUS_VERSION_CURRENT; 215250199Sgrehan 216282212Swhu do { 217282212Swhu ret = hv_vmbus_negotiate_version(msg_info, version); 218282212Swhu if (ret == EWOULDBLOCK) { 219282212Swhu /* 220282212Swhu * We timed out. 221282212Swhu */ 222282212Swhu goto cleanup; 223282212Swhu } 224250199Sgrehan 225282212Swhu if (hv_vmbus_g_connection.connect_state == HV_CONNECTED) 226282212Swhu break; 227250199Sgrehan 228282212Swhu version = hv_vmbus_get_next_version(version); 229282212Swhu } while (version != HV_VMBUS_VERSION_INVALID); 230250199Sgrehan 231282212Swhu hv_vmbus_protocal_version = version; 232282212Swhu if (bootverbose) 233293870Ssephe printf("VMBUS: Protocol Version: %d.%d\n", 234282212Swhu version >> 16, version & 0xFFFF); 235250199Sgrehan 236250199Sgrehan sema_destroy(&msg_info->wait_sema); 237250199Sgrehan free(msg_info, M_DEVBUF); 238250199Sgrehan 239250199Sgrehan return (0); 240250199Sgrehan 241250199Sgrehan /* 242250199Sgrehan * Cleanup after failure! 243250199Sgrehan */ 244250199Sgrehan cleanup: 245250199Sgrehan 246250199Sgrehan hv_vmbus_g_connection.connect_state = HV_DISCONNECTED; 247250199Sgrehan 248250199Sgrehan mtx_destroy(&hv_vmbus_g_connection.channel_lock); 249250199Sgrehan mtx_destroy(&hv_vmbus_g_connection.channel_msg_lock); 250250199Sgrehan 251250199Sgrehan if (hv_vmbus_g_connection.interrupt_page != NULL) { 252295964Ssephe free(hv_vmbus_g_connection.interrupt_page, M_DEVBUF); 253250199Sgrehan hv_vmbus_g_connection.interrupt_page = NULL; 254250199Sgrehan } 255250199Sgrehan 256295309Ssephe free(hv_vmbus_g_connection.monitor_page_1, M_DEVBUF); 257295309Ssephe free(hv_vmbus_g_connection.monitor_page_2, M_DEVBUF); 258250199Sgrehan 259250199Sgrehan if (msg_info) { 260250199Sgrehan sema_destroy(&msg_info->wait_sema); 261250199Sgrehan free(msg_info, M_DEVBUF); 262250199Sgrehan } 263250199Sgrehan 264294553Ssephe free(hv_vmbus_g_connection.channels, M_DEVBUF); 265250199Sgrehan return (ret); 266250199Sgrehan} 267250199Sgrehan 268250199Sgrehan/** 269250199Sgrehan * Send a disconnect request on the partition service connection 270250199Sgrehan */ 271250199Sgrehanint 272250199Sgrehanhv_vmbus_disconnect(void) { 273250199Sgrehan int ret = 0; 274295308Ssephe hv_vmbus_channel_unload msg; 275250199Sgrehan 276295308Ssephe msg.message_type = HV_CHANNEL_MESSAGE_UNLOAD; 277250199Sgrehan 278295308Ssephe ret = hv_vmbus_post_message(&msg, sizeof(hv_vmbus_channel_unload)); 279250199Sgrehan 280295964Ssephe free(hv_vmbus_g_connection.interrupt_page, M_DEVBUF); 281250199Sgrehan 282250199Sgrehan mtx_destroy(&hv_vmbus_g_connection.channel_msg_lock); 283250199Sgrehan 284294553Ssephe free(hv_vmbus_g_connection.channels, M_DEVBUF); 285250199Sgrehan hv_vmbus_g_connection.connect_state = HV_DISCONNECTED; 286250199Sgrehan 287250199Sgrehan return (ret); 288250199Sgrehan} 289250199Sgrehan 290250199Sgrehan/** 291250199Sgrehan * Handler for events 292250199Sgrehan */ 293250199Sgrehanvoid 294294886Ssephehv_vmbus_on_events(int cpu) 295250199Sgrehan{ 296282212Swhu int bit; 297250199Sgrehan int dword; 298282212Swhu void *page_addr; 299282212Swhu uint32_t* recv_interrupt_page = NULL; 300250199Sgrehan int rel_id; 301282212Swhu int maxdword; 302282212Swhu hv_vmbus_synic_event_flags *event; 303250199Sgrehan 304282212Swhu KASSERT(cpu <= mp_maxid, ("VMBUS: hv_vmbus_on_events: " 305282212Swhu "cpu out of range!")); 306250199Sgrehan 307297908Ssephe page_addr = hv_vmbus_g_context.syn_ic_event_page[cpu]; 308297908Ssephe event = (hv_vmbus_synic_event_flags *) 309297908Ssephe page_addr + HV_VMBUS_MESSAGE_SINT; 310282212Swhu if ((hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008) || 311282212Swhu (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7)) { 312282212Swhu maxdword = HV_MAX_NUM_CHANNELS_SUPPORTED >> 5; 313282212Swhu /* 314282212Swhu * receive size is 1/2 page and divide that by 4 bytes 315282212Swhu */ 316299889Ssephe if (synch_test_and_clear_bit(0, &event->flags32[0])) { 317297908Ssephe recv_interrupt_page = 318297908Ssephe hv_vmbus_g_connection.recv_interrupt_page; 319299889Ssephe } else { 320299889Ssephe return; 321299889Ssephe } 322282212Swhu } else { 323282212Swhu /* 324282212Swhu * On Host with Win8 or above, the event page can be 325282212Swhu * checked directly to get the id of the channel 326282212Swhu * that has the pending interrupt. 327282212Swhu */ 328282212Swhu maxdword = HV_EVENT_FLAGS_DWORD_COUNT; 329282212Swhu recv_interrupt_page = event->flags32; 330282212Swhu } 331282212Swhu 332250199Sgrehan /* 333250199Sgrehan * Check events 334250199Sgrehan */ 335299889Ssephe for (dword = 0; dword < maxdword; dword++) { 336299889Ssephe if (recv_interrupt_page[dword]) { 337299889Ssephe for (bit = 0; bit < HV_CHANNEL_DWORD_LEN; bit++) { 338299889Ssephe if (synch_test_and_clear_bit(bit, 339299889Ssephe (uint32_t *) &recv_interrupt_page[dword])) { 340299890Ssephe struct hv_vmbus_channel *channel; 341299890Ssephe 342299890Ssephe rel_id = (dword << 5) + bit; 343299890Ssephe channel = hv_vmbus_g_connection.channels[rel_id]; 344299890Ssephe 345299889Ssephe /* if channel is closed or closing */ 346299889Ssephe if (channel == NULL || channel->rxq == NULL) 347299889Ssephe continue; 348250199Sgrehan 349299889Ssephe if (channel->batched_reading) 350299889Ssephe hv_ring_buffer_read_begin(&channel->inbound); 351299889Ssephe taskqueue_enqueue(channel->rxq, &channel->channel_task); 352299889Ssephe } 353299889Ssephe } 354250199Sgrehan } 355250199Sgrehan } 356250199Sgrehan} 357250199Sgrehan 358250199Sgrehan/** 359250199Sgrehan * Send a msg on the vmbus's message connection 360250199Sgrehan */ 361297219Ssepheint hv_vmbus_post_message(void *buffer, size_t bufferLen) 362297219Ssephe{ 363250199Sgrehan hv_vmbus_connection_id connId; 364297219Ssephe sbintime_t time = SBT_1MS; 365297219Ssephe int retries; 366297219Ssephe int ret; 367250199Sgrehan 368297219Ssephe connId.as_uint32_t = 0; 369297219Ssephe connId.u.id = HV_VMBUS_MESSAGE_CONNECTION_ID; 370250199Sgrehan 371297219Ssephe /* 372297219Ssephe * We retry to cope with transient failures caused by host side's 373297219Ssephe * insufficient resources. 20 times should suffice in practice. 374250199Sgrehan */ 375297219Ssephe for (retries = 0; retries < 20; retries++) { 376297219Ssephe ret = hv_vmbus_post_msg_via_msg_ipc(connId, 1, buffer, 377297219Ssephe bufferLen); 378297219Ssephe if (ret == HV_STATUS_SUCCESS) 379297219Ssephe return (0); 380297219Ssephe 381297219Ssephe pause_sbt("pstmsg", time, 0, C_HARDCLOCK); 382297219Ssephe if (time < SBT_1S * 2) 383297219Ssephe time *= 2; 384250199Sgrehan } 385250199Sgrehan 386297219Ssephe KASSERT(ret == HV_STATUS_SUCCESS, 387297219Ssephe ("Error VMBUS: Message Post Failed, ret=%d\n", ret)); 388250199Sgrehan 389297219Ssephe return (EAGAIN); 390250199Sgrehan} 391250199Sgrehan 392250199Sgrehan/** 393250199Sgrehan * Send an event notification to the parent 394250199Sgrehan */ 395250199Sgrehanint 396282212Swhuhv_vmbus_set_event(hv_vmbus_channel *channel) { 397250199Sgrehan int ret = 0; 398282212Swhu uint32_t child_rel_id = channel->offer_msg.child_rel_id; 399250199Sgrehan 400250199Sgrehan /* Each uint32_t represents 32 channels */ 401250199Sgrehan 402250199Sgrehan synch_set_bit(child_rel_id & 31, 403250199Sgrehan (((uint32_t *)hv_vmbus_g_connection.send_interrupt_page 404250199Sgrehan + (child_rel_id >> 5)))); 405282212Swhu ret = hv_vmbus_signal_event(channel->signal_event_param); 406250199Sgrehan 407250199Sgrehan return (ret); 408250199Sgrehan} 409