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 <asm/mshyperv.h> 36 37 #include "hyperv_vmbus.h" 38 39 40 struct vmbus_connection vmbus_connection = { 41 .conn_state = DISCONNECTED, 42 .next_gpadl_handle = ATOMIC_INIT(0xE1E10), 43 }; 44 EXPORT_SYMBOL_GPL(vmbus_connection); 45 46 /* 47 * Negotiated protocol version with the host. 48 */ 49 __u32 vmbus_proto_version; 50 EXPORT_SYMBOL_GPL(vmbus_proto_version); 51 52 static __u32 vmbus_get_next_version(__u32 current_version) 53 { 54 switch (current_version) { 55 case (VERSION_WIN7): 56 return VERSION_WS2008; 57 58 case (VERSION_WIN8): 59 return VERSION_WIN7; 60 61 case (VERSION_WIN8_1): 62 return VERSION_WIN8; 63 64 case (VERSION_WIN10): 65 return VERSION_WIN8_1; 66 67 case (VERSION_WS2008): 68 default: 69 return VERSION_INVAL; 70 } 71 } 72 73 static int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo, 74 __u32 version) 75 { 76 int ret = 0; 77 struct vmbus_channel_initiate_contact *msg; 78 unsigned long flags; 79 80 init_completion(&msginfo->waitevent); 81 82 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg; 83 84 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT; 85 msg->vmbus_version_requested = version; 86 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page); 87 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]); 88 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]); 89 /* 90 * We want all channel messages to be delivered on CPU 0. 91 * This has been the behavior pre-win8. This is not 92 * perf issue and having all channel messages delivered on CPU 0 93 * would be ok. 94 * For post win8 hosts, we support receiving channel messagges on 95 * all the CPUs. This is needed for kexec to work correctly where 96 * the CPU attempting to connect may not be CPU 0. 97 */ 98 if (version >= VERSION_WIN8_1) { 99 msg->target_vcpu = 100 hv_cpu_number_to_vp_number(smp_processor_id()); 101 vmbus_connection.connect_cpu = smp_processor_id(); 102 } else { 103 msg->target_vcpu = 0; 104 vmbus_connection.connect_cpu = 0; 105 } 106 107 /* 108 * Add to list before we send the request since we may 109 * receive the response before returning from this routine 110 */ 111 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 112 list_add_tail(&msginfo->msglistentry, 113 &vmbus_connection.chn_msg_list); 114 115 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 116 117 ret = vmbus_post_msg(msg, 118 sizeof(struct vmbus_channel_initiate_contact), 119 true); 120 121 trace_vmbus_negotiate_version(msg, ret); 122 123 if (ret != 0) { 124 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 125 list_del(&msginfo->msglistentry); 126 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, 127 flags); 128 return ret; 129 } 130 131 /* Wait for the connection response */ 132 wait_for_completion(&msginfo->waitevent); 133 134 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 135 list_del(&msginfo->msglistentry); 136 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 137 138 /* Check if successful */ 139 if (msginfo->response.version_response.version_supported) { 140 vmbus_connection.conn_state = CONNECTED; 141 } else { 142 return -ECONNREFUSED; 143 } 144 145 return ret; 146 } 147 148 /* 149 * vmbus_connect - Sends a connect request on the partition service connection 150 */ 151 int vmbus_connect(void) 152 { 153 int ret = 0; 154 struct vmbus_channel_msginfo *msginfo = NULL; 155 __u32 version; 156 157 /* Initialize the vmbus connection */ 158 vmbus_connection.conn_state = CONNECTING; 159 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con"); 160 if (!vmbus_connection.work_queue) { 161 ret = -ENOMEM; 162 goto cleanup; 163 } 164 165 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list); 166 spin_lock_init(&vmbus_connection.channelmsg_lock); 167 168 INIT_LIST_HEAD(&vmbus_connection.chn_list); 169 mutex_init(&vmbus_connection.channel_mutex); 170 171 /* 172 * Setup the vmbus event connection for channel interrupt 173 * abstraction stuff 174 */ 175 vmbus_connection.int_page = 176 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0); 177 if (vmbus_connection.int_page == NULL) { 178 ret = -ENOMEM; 179 goto cleanup; 180 } 181 182 vmbus_connection.recv_int_page = vmbus_connection.int_page; 183 vmbus_connection.send_int_page = 184 (void *)((unsigned long)vmbus_connection.int_page + 185 (PAGE_SIZE >> 1)); 186 187 /* 188 * Setup the monitor notification facility. The 1st page for 189 * parent->child and the 2nd page for child->parent 190 */ 191 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 192 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0); 193 if ((vmbus_connection.monitor_pages[0] == NULL) || 194 (vmbus_connection.monitor_pages[1] == NULL)) { 195 ret = -ENOMEM; 196 goto cleanup; 197 } 198 199 msginfo = kzalloc(sizeof(*msginfo) + 200 sizeof(struct vmbus_channel_initiate_contact), 201 GFP_KERNEL); 202 if (msginfo == NULL) { 203 ret = -ENOMEM; 204 goto cleanup; 205 } 206 207 /* 208 * Negotiate a compatible VMBUS version number with the 209 * host. We start with the highest number we can support 210 * and work our way down until we negotiate a compatible 211 * version. 212 */ 213 214 version = VERSION_CURRENT; 215 216 do { 217 ret = vmbus_negotiate_version(msginfo, version); 218 if (ret == -ETIMEDOUT) 219 goto cleanup; 220 221 if (vmbus_connection.conn_state == CONNECTED) 222 break; 223 224 version = vmbus_get_next_version(version); 225 } while (version != VERSION_INVAL); 226 227 if (version == VERSION_INVAL) 228 goto cleanup; 229 230 vmbus_proto_version = version; 231 pr_info("Vmbus version:%d.%d\n", 232 version >> 16, version & 0xFFFF); 233 234 kfree(msginfo); 235 return 0; 236 237 cleanup: 238 pr_err("Unable to connect to host\n"); 239 240 vmbus_connection.conn_state = DISCONNECTED; 241 vmbus_disconnect(); 242 243 kfree(msginfo); 244 245 return ret; 246 } 247 248 void vmbus_disconnect(void) 249 { 250 /* 251 * First send the unload request to the host. 252 */ 253 vmbus_initiate_unload(false); 254 255 if (vmbus_connection.work_queue) { 256 drain_workqueue(vmbus_connection.work_queue); 257 destroy_workqueue(vmbus_connection.work_queue); 258 } 259 260 if (vmbus_connection.int_page) { 261 free_pages((unsigned long)vmbus_connection.int_page, 0); 262 vmbus_connection.int_page = NULL; 263 } 264 265 free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0); 266 free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0); 267 vmbus_connection.monitor_pages[0] = NULL; 268 vmbus_connection.monitor_pages[1] = NULL; 269 } 270 271 /* 272 * relid2channel - Get the channel object given its 273 * child relative id (ie channel id) 274 */ 275 struct vmbus_channel *relid2channel(u32 relid) 276 { 277 struct vmbus_channel *channel; 278 struct vmbus_channel *found_channel = NULL; 279 struct list_head *cur, *tmp; 280 struct vmbus_channel *cur_sc; 281 282 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex)); 283 284 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 285 if (channel->offermsg.child_relid == relid) { 286 found_channel = channel; 287 break; 288 } else if (!list_empty(&channel->sc_list)) { 289 /* 290 * Deal with sub-channels. 291 */ 292 list_for_each_safe(cur, tmp, &channel->sc_list) { 293 cur_sc = list_entry(cur, struct vmbus_channel, 294 sc_list); 295 if (cur_sc->offermsg.child_relid == relid) { 296 found_channel = cur_sc; 297 break; 298 } 299 } 300 } 301 } 302 303 return found_channel; 304 } 305 306 /* 307 * vmbus_on_event - Process a channel event notification 308 * 309 * For batched channels (default) optimize host to guest signaling 310 * by ensuring: 311 * 1. While reading the channel, we disable interrupts from host. 312 * 2. Ensure that we process all posted messages from the host 313 * before returning from this callback. 314 * 3. Once we return, enable signaling from the host. Once this 315 * state is set we check to see if additional packets are 316 * available to read. In this case we repeat the process. 317 * If this tasklet has been running for a long time 318 * then reschedule ourselves. 319 */ 320 void vmbus_on_event(unsigned long data) 321 { 322 struct vmbus_channel *channel = (void *) data; 323 unsigned long time_limit = jiffies + 2; 324 325 trace_vmbus_on_event(channel); 326 327 do { 328 void (*callback_fn)(void *); 329 330 /* A channel once created is persistent even when 331 * there is no driver handling the device. An 332 * unloading driver sets the onchannel_callback to NULL. 333 */ 334 callback_fn = READ_ONCE(channel->onchannel_callback); 335 if (unlikely(callback_fn == NULL)) 336 return; 337 338 (*callback_fn)(channel->channel_callback_context); 339 340 if (channel->callback_mode != HV_CALL_BATCHED) 341 return; 342 343 if (likely(hv_end_read(&channel->inbound) == 0)) 344 return; 345 346 hv_begin_read(&channel->inbound); 347 } while (likely(time_before(jiffies, time_limit))); 348 349 /* The time limit (2 jiffies) has been reached */ 350 tasklet_schedule(&channel->callback_event); 351 } 352 353 /* 354 * vmbus_post_msg - Send a msg on the vmbus's message connection 355 */ 356 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep) 357 { 358 union hv_connection_id conn_id; 359 int ret = 0; 360 int retries = 0; 361 u32 usec = 1; 362 363 conn_id.asu32 = 0; 364 conn_id.u.id = VMBUS_MESSAGE_CONNECTION_ID; 365 366 /* 367 * hv_post_message() can have transient failures because of 368 * insufficient resources. Retry the operation a couple of 369 * times before giving up. 370 */ 371 while (retries < 100) { 372 ret = hv_post_message(conn_id, 1, buffer, buflen); 373 374 switch (ret) { 375 case HV_STATUS_INVALID_CONNECTION_ID: 376 /* 377 * We could get this if we send messages too 378 * frequently. 379 */ 380 ret = -EAGAIN; 381 break; 382 case HV_STATUS_INSUFFICIENT_MEMORY: 383 case HV_STATUS_INSUFFICIENT_BUFFERS: 384 ret = -ENOBUFS; 385 break; 386 case HV_STATUS_SUCCESS: 387 return ret; 388 default: 389 pr_err("hv_post_msg() failed; error code:%d\n", ret); 390 return -EINVAL; 391 } 392 393 retries++; 394 if (can_sleep && usec > 1000) 395 msleep(usec / 1000); 396 else if (usec < MAX_UDELAY_MS * 1000) 397 udelay(usec); 398 else 399 mdelay(usec / 1000); 400 401 if (retries < 22) 402 usec *= 2; 403 } 404 return ret; 405 } 406 407 /* 408 * vmbus_set_event - Send an event notification to the parent 409 */ 410 void vmbus_set_event(struct vmbus_channel *channel) 411 { 412 u32 child_relid = channel->offermsg.child_relid; 413 414 if (!channel->is_dedicated_interrupt) 415 vmbus_send_interrupt(child_relid); 416 417 ++channel->sig_events; 418 419 hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, channel->sig_event); 420 } 421 EXPORT_SYMBOL_GPL(vmbus_set_event); 422