1 /* 2 * 3 * Copyright (c) 2009, Microsoft Corporation. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 * 14 * You should have received a copy of the GNU General Public License along with 15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple 16 * Place - Suite 330, Boston, MA 02111-1307 USA. 17 * 18 * Authors: 19 * Haiyang Zhang <haiyangz@microsoft.com> 20 * Hank Janssen <hjanssen@microsoft.com> 21 * 22 */ 23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 24 25 #include <linux/kernel.h> 26 #include <linux/sched.h> 27 #include <linux/wait.h> 28 #include <linux/delay.h> 29 #include <linux/mm.h> 30 #include <linux/slab.h> 31 #include <linux/vmalloc.h> 32 #include <linux/hyperv.h> 33 #include <linux/export.h> 34 #include <asm/hyperv.h> 35 #include "hyperv_vmbus.h" 36 37 38 struct vmbus_connection vmbus_connection = { 39 .conn_state = DISCONNECTED, 40 .next_gpadl_handle = ATOMIC_INIT(0xE1E10), 41 }; 42 EXPORT_SYMBOL_GPL(vmbus_connection); 43 44 /* 45 * Negotiated protocol version with the host. 46 */ 47 __u32 vmbus_proto_version; 48 EXPORT_SYMBOL_GPL(vmbus_proto_version); 49 50 static __u32 vmbus_get_next_version(__u32 current_version) 51 { 52 switch (current_version) { 53 case (VERSION_WIN7): 54 return VERSION_WS2008; 55 56 case (VERSION_WIN8): 57 return VERSION_WIN7; 58 59 case (VERSION_WIN8_1): 60 return VERSION_WIN8; 61 62 case (VERSION_WIN10): 63 return VERSION_WIN8_1; 64 65 case (VERSION_WS2008): 66 default: 67 return VERSION_INVAL; 68 } 69 } 70 71 static int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, 72 __u32 version) 73 { 74 int ret = 0; 75 struct vmbus_channel_initiate_contact *msg; 76 unsigned long flags; 77 78 init_completion(&msginfo->waitevent); 79 80 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg; 81 82 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT; 83 msg->vmbus_version_requested = version; 84 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page); 85 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]); 86 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]); 87 /* 88 * We want all channel messages to be delivered on CPU 0. 89 * This has been the behavior pre-win8. This is not 90 * perf issue and having all channel messages delivered on CPU 0 91 * would be ok. 92 * For post win8 hosts, we support receiving channel messagges on 93 * all the CPUs. This is needed for kexec to work correctly where 94 * the CPU attempting to connect may not be CPU 0. 95 */ 96 if (version >= VERSION_WIN8_1) { 97 msg->target_vcpu = hv_context.vp_index[get_cpu()]; 98 put_cpu(); 99 } else { 100 msg->target_vcpu = 0; 101 } 102 103 /* 104 * Add to list before we send the request since we may 105 * receive the response before returning from this routine 106 */ 107 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 108 list_add_tail(&msginfo->msglistentry, 109 &vmbus_connection.chn_msg_list); 110 111 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 112 113 ret = vmbus_post_msg(msg, 114 sizeof(struct vmbus_channel_initiate_contact)); 115 if (ret != 0) { 116 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 117 list_del(&msginfo->msglistentry); 118 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, 119 flags); 120 return ret; 121 } 122 123 /* Wait for the connection response */ 124 wait_for_completion(&msginfo->waitevent); 125 126 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 127 list_del(&msginfo->msglistentry); 128 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 129 130 /* Check if successful */ 131 if (msginfo->response.version_response.version_supported) { 132 vmbus_connection.conn_state = CONNECTED; 133 } else { 134 return -ECONNREFUSED; 135 } 136 137 return ret; 138 } 139 140 /* 141 * vmbus_connect - Sends a connect request on the partition service connection 142 */ 143 int vmbus_connect(void) 144 { 145 int ret = 0; 146 struct vmbus_channel_msginfo *msginfo = NULL; 147 __u32 version; 148 149 /* Initialize the vmbus connection */ 150 vmbus_connection.conn_state = CONNECTING; 151 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con"); 152 if (!vmbus_connection.work_queue) { 153 ret = -ENOMEM; 154 goto cleanup; 155 } 156 157 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list); 158 spin_lock_init(&vmbus_connection.channelmsg_lock); 159 160 INIT_LIST_HEAD(&vmbus_connection.chn_list); 161 mutex_init(&vmbus_connection.channel_mutex); 162 163 /* 164 * Setup the vmbus event connection for channel interrupt 165 * abstraction stuff 166 */ 167 vmbus_connection.int_page = 168 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0); 169 if (vmbus_connection.int_page == NULL) { 170 ret = -ENOMEM; 171 goto cleanup; 172 } 173 174 vmbus_connection.recv_int_page = vmbus_connection.int_page; 175 vmbus_connection.send_int_page = 176 (void *)((unsigned long)vmbus_connection.int_page + 177 (PAGE_SIZE >> 1)); 178 179 /* 180 * Setup the monitor notification facility. The 1st page for 181 * parent->child and the 2nd page for child->parent 182 */ 183 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 184 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 185 if ((vmbus_connection.monitor_pages[0] == NULL) || 186 (vmbus_connection.monitor_pages[1] == NULL)) { 187 ret = -ENOMEM; 188 goto cleanup; 189 } 190 191 msginfo = kzalloc(sizeof(*msginfo) + 192 sizeof(struct vmbus_channel_initiate_contact), 193 GFP_KERNEL); 194 if (msginfo == NULL) { 195 ret = -ENOMEM; 196 goto cleanup; 197 } 198 199 /* 200 * Negotiate a compatible VMBUS version number with the 201 * host. We start with the highest number we can support 202 * and work our way down until we negotiate a compatible 203 * version. 204 */ 205 206 version = VERSION_CURRENT; 207 208 do { 209 ret = vmbus_negotiate_version(msginfo, version); 210 if (ret == -ETIMEDOUT) 211 goto cleanup; 212 213 if (vmbus_connection.conn_state == CONNECTED) 214 break; 215 216 version = vmbus_get_next_version(version); 217 } while (version != VERSION_INVAL); 218 219 if (version == VERSION_INVAL) 220 goto cleanup; 221 222 vmbus_proto_version = version; 223 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d; Vmbus version:%d.%d\n", 224 host_info_eax, host_info_ebx >> 16, 225 host_info_ebx & 0xFFFF, host_info_ecx, 226 host_info_edx >> 24, host_info_edx & 0xFFFFFF, 227 version >> 16, version & 0xFFFF); 228 229 kfree(msginfo); 230 return 0; 231 232 cleanup: 233 pr_err("Unable to connect to host\n"); 234 235 vmbus_connection.conn_state = DISCONNECTED; 236 vmbus_disconnect(); 237 238 kfree(msginfo); 239 240 return ret; 241 } 242 243 void vmbus_disconnect(void) 244 { 245 /* 246 * First send the unload request to the host. 247 */ 248 vmbus_initiate_unload(false); 249 250 if (vmbus_connection.work_queue) { 251 drain_workqueue(vmbus_connection.work_queue); 252 destroy_workqueue(vmbus_connection.work_queue); 253 } 254 255 if (vmbus_connection.int_page) { 256 free_pages((unsigned long)vmbus_connection.int_page, 0); 257 vmbus_connection.int_page = NULL; 258 } 259 260 free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0); 261 free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0); 262 vmbus_connection.monitor_pages[0] = NULL; 263 vmbus_connection.monitor_pages[1] = NULL; 264 } 265 266 /* 267 * Map the given relid to the corresponding channel based on the 268 * per-cpu list of channels that have been affinitized to this CPU. 269 * This will be used in the channel callback path as we can do this 270 * mapping in a lock-free fashion. 271 */ 272 static struct vmbus_channel *pcpu_relid2channel(u32 relid) 273 { 274 struct vmbus_channel *channel; 275 struct vmbus_channel *found_channel = NULL; 276 int cpu = smp_processor_id(); 277 struct list_head *pcpu_head = &hv_context.percpu_list[cpu]; 278 279 list_for_each_entry(channel, pcpu_head, percpu_list) { 280 if (channel->offermsg.child_relid == relid) { 281 found_channel = channel; 282 break; 283 } 284 } 285 286 return found_channel; 287 } 288 289 /* 290 * relid2channel - Get the channel object given its 291 * child relative id (ie channel id) 292 */ 293 struct vmbus_channel *relid2channel(u32 relid) 294 { 295 struct vmbus_channel *channel; 296 struct vmbus_channel *found_channel = NULL; 297 struct list_head *cur, *tmp; 298 struct vmbus_channel *cur_sc; 299 300 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex)); 301 302 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 303 if (channel->offermsg.child_relid == relid) { 304 found_channel = channel; 305 break; 306 } else if (!list_empty(&channel->sc_list)) { 307 /* 308 * Deal with sub-channels. 309 */ 310 list_for_each_safe(cur, tmp, &channel->sc_list) { 311 cur_sc = list_entry(cur, struct vmbus_channel, 312 sc_list); 313 if (cur_sc->offermsg.child_relid == relid) { 314 found_channel = cur_sc; 315 break; 316 } 317 } 318 } 319 } 320 321 return found_channel; 322 } 323 324 /* 325 * process_chn_event - Process a channel event notification 326 */ 327 static void process_chn_event(u32 relid) 328 { 329 struct vmbus_channel *channel; 330 void *arg; 331 bool read_state; 332 u32 bytes_to_read; 333 334 /* 335 * Find the channel based on this relid and invokes the 336 * channel callback to process the event 337 */ 338 channel = pcpu_relid2channel(relid); 339 340 if (!channel) 341 return; 342 343 /* 344 * A channel once created is persistent even when there 345 * is no driver handling the device. An unloading driver 346 * sets the onchannel_callback to NULL on the same CPU 347 * as where this interrupt is handled (in an interrupt context). 348 * Thus, checking and invoking the driver specific callback takes 349 * care of orderly unloading of the driver. 350 */ 351 352 if (channel->onchannel_callback != NULL) { 353 arg = channel->channel_callback_context; 354 read_state = channel->batched_reading; 355 /* 356 * This callback reads the messages sent by the host. 357 * We can optimize host to guest signaling by ensuring: 358 * 1. While reading the channel, we disable interrupts from 359 * host. 360 * 2. Ensure that we process all posted messages from the host 361 * before returning from this callback. 362 * 3. Once we return, enable signaling from the host. Once this 363 * state is set we check to see if additional packets are 364 * available to read. In this case we repeat the process. 365 */ 366 367 do { 368 if (read_state) 369 hv_begin_read(&channel->inbound); 370 channel->onchannel_callback(arg); 371 if (read_state) 372 bytes_to_read = hv_end_read(&channel->inbound); 373 else 374 bytes_to_read = 0; 375 } while (read_state && (bytes_to_read != 0)); 376 } 377 } 378 379 /* 380 * vmbus_on_event - Handler for events 381 */ 382 void vmbus_on_event(unsigned long data) 383 { 384 u32 dword; 385 u32 maxdword; 386 int bit; 387 u32 relid; 388 u32 *recv_int_page = NULL; 389 void *page_addr; 390 int cpu = smp_processor_id(); 391 union hv_synic_event_flags *event; 392 393 if (vmbus_proto_version < VERSION_WIN8) { 394 maxdword = MAX_NUM_CHANNELS_SUPPORTED >> 5; 395 recv_int_page = vmbus_connection.recv_int_page; 396 } else { 397 /* 398 * When the host is win8 and beyond, the event page 399 * can be directly checked to get the id of the channel 400 * that has the interrupt pending. 401 */ 402 maxdword = HV_EVENT_FLAGS_DWORD_COUNT; 403 page_addr = hv_context.synic_event_page[cpu]; 404 event = (union hv_synic_event_flags *)page_addr + 405 VMBUS_MESSAGE_SINT; 406 recv_int_page = event->flags32; 407 } 408 409 410 411 /* Check events */ 412 if (!recv_int_page) 413 return; 414 for (dword = 0; dword < maxdword; dword++) { 415 if (!recv_int_page[dword]) 416 continue; 417 for (bit = 0; bit < 32; bit++) { 418 if (sync_test_and_clear_bit(bit, 419 (unsigned long *)&recv_int_page[dword])) { 420 relid = (dword << 5) + bit; 421 422 if (relid == 0) 423 /* 424 * Special case - vmbus 425 * channel protocol msg 426 */ 427 continue; 428 429 process_chn_event(relid); 430 } 431 } 432 } 433 } 434 435 /* 436 * vmbus_post_msg - Send a msg on the vmbus's message connection 437 */ 438 int vmbus_post_msg(void *buffer, size_t buflen) 439 { 440 union hv_connection_id conn_id; 441 int ret = 0; 442 int retries = 0; 443 u32 usec = 1; 444 445 conn_id.asu32 = 0; 446 conn_id.u.id = VMBUS_MESSAGE_CONNECTION_ID; 447 448 /* 449 * hv_post_message() can have transient failures because of 450 * insufficient resources. Retry the operation a couple of 451 * times before giving up. 452 */ 453 while (retries < 20) { 454 ret = hv_post_message(conn_id, 1, buffer, buflen); 455 456 switch (ret) { 457 case HV_STATUS_INVALID_CONNECTION_ID: 458 /* 459 * We could get this if we send messages too 460 * frequently. 461 */ 462 ret = -EAGAIN; 463 break; 464 case HV_STATUS_INSUFFICIENT_MEMORY: 465 case HV_STATUS_INSUFFICIENT_BUFFERS: 466 ret = -ENOMEM; 467 break; 468 case HV_STATUS_SUCCESS: 469 return ret; 470 default: 471 pr_err("hv_post_msg() failed; error code:%d\n", ret); 472 return -EINVAL; 473 } 474 475 retries++; 476 udelay(usec); 477 if (usec < 2048) 478 usec *= 2; 479 } 480 return ret; 481 } 482 483 /* 484 * vmbus_set_event - Send an event notification to the parent 485 */ 486 void vmbus_set_event(struct vmbus_channel *channel) 487 { 488 u32 child_relid = channel->offermsg.child_relid; 489 490 if (!channel->is_dedicated_interrupt) { 491 /* Each u32 represents 32 channels */ 492 sync_set_bit(child_relid & 31, 493 (unsigned long *)vmbus_connection.send_int_page + 494 (child_relid >> 5)); 495 } 496 497 hv_do_hypercall(HVCALL_SIGNAL_EVENT, channel->sig_event, NULL); 498 } 499 EXPORT_SYMBOL_GPL(vmbus_set_event); 500