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 true); 116 if (ret != 0) { 117 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 118 list_del(&msginfo->msglistentry); 119 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, 120 flags); 121 return ret; 122 } 123 124 /* Wait for the connection response */ 125 wait_for_completion(&msginfo->waitevent); 126 127 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 128 list_del(&msginfo->msglistentry); 129 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 130 131 /* Check if successful */ 132 if (msginfo->response.version_response.version_supported) { 133 vmbus_connection.conn_state = CONNECTED; 134 } else { 135 return -ECONNREFUSED; 136 } 137 138 return ret; 139 } 140 141 /* 142 * vmbus_connect - Sends a connect request on the partition service connection 143 */ 144 int vmbus_connect(void) 145 { 146 int ret = 0; 147 struct vmbus_channel_msginfo *msginfo = NULL; 148 __u32 version; 149 150 /* Initialize the vmbus connection */ 151 vmbus_connection.conn_state = CONNECTING; 152 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con"); 153 if (!vmbus_connection.work_queue) { 154 ret = -ENOMEM; 155 goto cleanup; 156 } 157 158 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list); 159 spin_lock_init(&vmbus_connection.channelmsg_lock); 160 161 INIT_LIST_HEAD(&vmbus_connection.chn_list); 162 mutex_init(&vmbus_connection.channel_mutex); 163 164 /* 165 * Setup the vmbus event connection for channel interrupt 166 * abstraction stuff 167 */ 168 vmbus_connection.int_page = 169 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0); 170 if (vmbus_connection.int_page == NULL) { 171 ret = -ENOMEM; 172 goto cleanup; 173 } 174 175 vmbus_connection.recv_int_page = vmbus_connection.int_page; 176 vmbus_connection.send_int_page = 177 (void *)((unsigned long)vmbus_connection.int_page + 178 (PAGE_SIZE >> 1)); 179 180 /* 181 * Setup the monitor notification facility. The 1st page for 182 * parent->child and the 2nd page for child->parent 183 */ 184 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 185 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 186 if ((vmbus_connection.monitor_pages[0] == NULL) || 187 (vmbus_connection.monitor_pages[1] == NULL)) { 188 ret = -ENOMEM; 189 goto cleanup; 190 } 191 192 msginfo = kzalloc(sizeof(*msginfo) + 193 sizeof(struct vmbus_channel_initiate_contact), 194 GFP_KERNEL); 195 if (msginfo == NULL) { 196 ret = -ENOMEM; 197 goto cleanup; 198 } 199 200 /* 201 * Negotiate a compatible VMBUS version number with the 202 * host. We start with the highest number we can support 203 * and work our way down until we negotiate a compatible 204 * version. 205 */ 206 207 version = VERSION_CURRENT; 208 209 do { 210 ret = vmbus_negotiate_version(msginfo, version); 211 if (ret == -ETIMEDOUT) 212 goto cleanup; 213 214 if (vmbus_connection.conn_state == CONNECTED) 215 break; 216 217 version = vmbus_get_next_version(version); 218 } while (version != VERSION_INVAL); 219 220 if (version == VERSION_INVAL) 221 goto cleanup; 222 223 vmbus_proto_version = version; 224 pr_info("Vmbus version:%d.%d\n", 225 version >> 16, version & 0xFFFF); 226 227 kfree(msginfo); 228 return 0; 229 230 cleanup: 231 pr_err("Unable to connect to host\n"); 232 233 vmbus_connection.conn_state = DISCONNECTED; 234 vmbus_disconnect(); 235 236 kfree(msginfo); 237 238 return ret; 239 } 240 241 void vmbus_disconnect(void) 242 { 243 /* 244 * First send the unload request to the host. 245 */ 246 vmbus_initiate_unload(false); 247 248 if (vmbus_connection.work_queue) { 249 drain_workqueue(vmbus_connection.work_queue); 250 destroy_workqueue(vmbus_connection.work_queue); 251 } 252 253 if (vmbus_connection.int_page) { 254 free_pages((unsigned long)vmbus_connection.int_page, 0); 255 vmbus_connection.int_page = NULL; 256 } 257 258 free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0); 259 free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0); 260 vmbus_connection.monitor_pages[0] = NULL; 261 vmbus_connection.monitor_pages[1] = NULL; 262 } 263 264 /* 265 * Map the given relid to the corresponding channel based on the 266 * per-cpu list of channels that have been affinitized to this CPU. 267 * This will be used in the channel callback path as we can do this 268 * mapping in a lock-free fashion. 269 */ 270 static struct vmbus_channel *pcpu_relid2channel(u32 relid) 271 { 272 struct vmbus_channel *channel; 273 struct vmbus_channel *found_channel = NULL; 274 int cpu = smp_processor_id(); 275 struct list_head *pcpu_head = &hv_context.percpu_list[cpu]; 276 277 list_for_each_entry(channel, pcpu_head, percpu_list) { 278 if (channel->offermsg.child_relid == relid) { 279 found_channel = channel; 280 break; 281 } 282 } 283 284 return found_channel; 285 } 286 287 /* 288 * relid2channel - Get the channel object given its 289 * child relative id (ie channel id) 290 */ 291 struct vmbus_channel *relid2channel(u32 relid) 292 { 293 struct vmbus_channel *channel; 294 struct vmbus_channel *found_channel = NULL; 295 struct list_head *cur, *tmp; 296 struct vmbus_channel *cur_sc; 297 298 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex)); 299 300 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 301 if (channel->offermsg.child_relid == relid) { 302 found_channel = channel; 303 break; 304 } else if (!list_empty(&channel->sc_list)) { 305 /* 306 * Deal with sub-channels. 307 */ 308 list_for_each_safe(cur, tmp, &channel->sc_list) { 309 cur_sc = list_entry(cur, struct vmbus_channel, 310 sc_list); 311 if (cur_sc->offermsg.child_relid == relid) { 312 found_channel = cur_sc; 313 break; 314 } 315 } 316 } 317 } 318 319 return found_channel; 320 } 321 322 /* 323 * process_chn_event - Process a channel event notification 324 */ 325 static void process_chn_event(u32 relid) 326 { 327 struct vmbus_channel *channel; 328 void *arg; 329 bool read_state; 330 u32 bytes_to_read; 331 332 /* 333 * Find the channel based on this relid and invokes the 334 * channel callback to process the event 335 */ 336 channel = pcpu_relid2channel(relid); 337 338 if (!channel) 339 return; 340 341 /* 342 * A channel once created is persistent even when there 343 * is no driver handling the device. An unloading driver 344 * sets the onchannel_callback to NULL on the same CPU 345 * as where this interrupt is handled (in an interrupt context). 346 * Thus, checking and invoking the driver specific callback takes 347 * care of orderly unloading of the driver. 348 */ 349 350 if (channel->onchannel_callback != NULL) { 351 arg = channel->channel_callback_context; 352 read_state = channel->batched_reading; 353 /* 354 * This callback reads the messages sent by the host. 355 * We can optimize host to guest signaling by ensuring: 356 * 1. While reading the channel, we disable interrupts from 357 * host. 358 * 2. Ensure that we process all posted messages from the host 359 * before returning from this callback. 360 * 3. Once we return, enable signaling from the host. Once this 361 * state is set we check to see if additional packets are 362 * available to read. In this case we repeat the process. 363 */ 364 365 do { 366 if (read_state) 367 hv_begin_read(&channel->inbound); 368 channel->onchannel_callback(arg); 369 if (read_state) 370 bytes_to_read = hv_end_read(&channel->inbound); 371 else 372 bytes_to_read = 0; 373 } while (read_state && (bytes_to_read != 0)); 374 } 375 } 376 377 /* 378 * vmbus_on_event - Handler for events 379 */ 380 void vmbus_on_event(unsigned long data) 381 { 382 u32 dword; 383 u32 maxdword; 384 int bit; 385 u32 relid; 386 u32 *recv_int_page = NULL; 387 void *page_addr; 388 int cpu = smp_processor_id(); 389 union hv_synic_event_flags *event; 390 391 if (vmbus_proto_version < VERSION_WIN8) { 392 maxdword = MAX_NUM_CHANNELS_SUPPORTED >> 5; 393 recv_int_page = vmbus_connection.recv_int_page; 394 } else { 395 /* 396 * When the host is win8 and beyond, the event page 397 * can be directly checked to get the id of the channel 398 * that has the interrupt pending. 399 */ 400 maxdword = HV_EVENT_FLAGS_DWORD_COUNT; 401 page_addr = hv_context.synic_event_page[cpu]; 402 event = (union hv_synic_event_flags *)page_addr + 403 VMBUS_MESSAGE_SINT; 404 recv_int_page = event->flags32; 405 } 406 407 408 409 /* Check events */ 410 if (!recv_int_page) 411 return; 412 for (dword = 0; dword < maxdword; dword++) { 413 if (!recv_int_page[dword]) 414 continue; 415 for (bit = 0; bit < 32; bit++) { 416 if (sync_test_and_clear_bit(bit, 417 (unsigned long *)&recv_int_page[dword])) { 418 relid = (dword << 5) + bit; 419 420 if (relid == 0) 421 /* 422 * Special case - vmbus 423 * channel protocol msg 424 */ 425 continue; 426 427 process_chn_event(relid); 428 } 429 } 430 } 431 } 432 433 /* 434 * vmbus_post_msg - Send a msg on the vmbus's message connection 435 */ 436 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep) 437 { 438 union hv_connection_id conn_id; 439 int ret = 0; 440 int retries = 0; 441 u32 usec = 1; 442 443 conn_id.asu32 = 0; 444 conn_id.u.id = VMBUS_MESSAGE_CONNECTION_ID; 445 446 /* 447 * hv_post_message() can have transient failures because of 448 * insufficient resources. Retry the operation a couple of 449 * times before giving up. 450 */ 451 while (retries < 100) { 452 ret = hv_post_message(conn_id, 1, buffer, buflen); 453 454 switch (ret) { 455 case HV_STATUS_INVALID_CONNECTION_ID: 456 /* 457 * We could get this if we send messages too 458 * frequently. 459 */ 460 ret = -EAGAIN; 461 break; 462 case HV_STATUS_INSUFFICIENT_MEMORY: 463 case HV_STATUS_INSUFFICIENT_BUFFERS: 464 ret = -ENOMEM; 465 break; 466 case HV_STATUS_SUCCESS: 467 return ret; 468 default: 469 pr_err("hv_post_msg() failed; error code:%d\n", ret); 470 return -EINVAL; 471 } 472 473 retries++; 474 if (can_sleep && usec > 1000) 475 msleep(usec / 1000); 476 else if (usec < MAX_UDELAY_MS * 1000) 477 udelay(usec); 478 else 479 mdelay(usec / 1000); 480 481 if (usec < 256000) 482 usec *= 2; 483 } 484 return ret; 485 } 486 487 /* 488 * vmbus_set_event - Send an event notification to the parent 489 */ 490 void vmbus_set_event(struct vmbus_channel *channel) 491 { 492 u32 child_relid = channel->offermsg.child_relid; 493 494 if (!channel->is_dedicated_interrupt) { 495 /* Each u32 represents 32 channels */ 496 sync_set_bit(child_relid & 31, 497 (unsigned long *)vmbus_connection.send_int_page + 498 (child_relid >> 5)); 499 } 500 501 hv_do_hypercall(HVCALL_SIGNAL_EVENT, channel->sig_event, NULL); 502 } 503 EXPORT_SYMBOL_GPL(vmbus_set_event); 504