drivers/hv/channel.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/sched.h>
  25 #include <linux/wait.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/hyperv.h>
  30 #include <linux/uio.h>
  31 #include <linux/interrupt.h>
  32
  33 #include "hyperv_vmbus.h"
  34
  35 #define NUM_PAGES_SPANNED(addr, len) \
  36 ((PAGE_ALIGN(addr + len) >> PAGE_SHIFT) - (addr >> PAGE_SHIFT))
  37
  38 /*
  39  * vmbus_setevent- Trigger an event notification on the specified
  40  * channel.
  41  */
  42 static void vmbus_setevent(struct vmbus_channel *channel)
  43 {
  44         struct hv_monitor_page *monitorpage;
  45
  46         if (channel->offermsg.monitor_allocated) {
  47                 /* Each u32 represents 32 channels */
  48                 sync_set_bit(channel->offermsg.child_relid & 31,
  49                         (unsigned long *) vmbus_connection.send_int_page +
  50                         (channel->offermsg.child_relid >> 5));
  51
  52                 /* Get the child to parent monitor page */
  53                 monitorpage = vmbus_connection.monitor_pages[1];
  54
  55                 sync_set_bit(channel->monitor_bit,
  56                         (unsigned long *)&monitorpage->trigger_group
  57                                         [channel->monitor_grp].pending);
  58
  59         } else {
  60                 vmbus_set_event(channel);
  61         }
  62 }
  63
  64 /*
  65  * vmbus_open - Open the specified channel.
  66  */
  67 int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
  68                      u32 recv_ringbuffer_size, void *userdata, u32 userdatalen,
  69                      void (*onchannelcallback)(void *context), void *context)
  70 {
  71         struct vmbus_channel_open_channel *open_msg;
  72         struct vmbus_channel_msginfo *open_info = NULL;
  73         void *in, *out;
  74         unsigned long flags;
  75         int ret, err = 0;
  76         unsigned long t;
  77         struct page *page;
  78
  79         spin_lock_irqsave(&newchannel->lock, flags);
  80         if (newchannel->state == CHANNEL_OPEN_STATE) {
  81                 newchannel->state = CHANNEL_OPENING_STATE;
  82         } else {
  83                 spin_unlock_irqrestore(&newchannel->lock, flags);
  84                 return -EINVAL;
  85         }
  86         spin_unlock_irqrestore(&newchannel->lock, flags);
  87
  88         newchannel->onchannel_callback = onchannelcallback;
  89         newchannel->channel_callback_context = context;
  90
  91         /* Allocate the ring buffer */
  92         page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
  93                                 GFP_KERNEL|__GFP_ZERO,
  94                                 get_order(send_ringbuffer_size +
  95                                 recv_ringbuffer_size));
  96
  97         if (!page)
  98                 out = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
  99                                                get_order(send_ringbuffer_size +
 100                                                recv_ringbuffer_size));
 101         else
 102                 out = (void *)page_address(page);
 103
 104         if (!out) {
 105                 err = -ENOMEM;
 106                 goto error0;
 107         }
 108
 109         in = (void *)((unsigned long)out + send_ringbuffer_size);
 110
 111         newchannel->ringbuffer_pages = out;
 112         newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
 113                                            recv_ringbuffer_size) >> PAGE_SHIFT;
 114
 115         ret = hv_ringbuffer_init(
 116                 &newchannel->outbound, out, send_ringbuffer_size);
 117
 118         if (ret != 0) {
 119                 err = ret;
 120                 goto error0;
 121         }
 122
 123         ret = hv_ringbuffer_init(
 124                 &newchannel->inbound, in, recv_ringbuffer_size);
 125         if (ret != 0) {
 126                 err = ret;
 127                 goto error0;
 128         }
 129
 130
 131         /* Establish the gpadl for the ring buffer */
 132         newchannel->ringbuffer_gpadlhandle = 0;
 133
 134         ret = vmbus_establish_gpadl(newchannel,
 135                                          newchannel->outbound.ring_buffer,
 136                                          send_ringbuffer_size +
 137                                          recv_ringbuffer_size,
 138                                          &newchannel->ringbuffer_gpadlhandle);
 139
 140         if (ret != 0) {
 141                 err = ret;
 142                 goto error0;
 143         }
 144
 145         /* Create and init the channel open message */
 146         open_info = kmalloc(sizeof(*open_info) +
 147                            sizeof(struct vmbus_channel_open_channel),
 148                            GFP_KERNEL);
 149         if (!open_info) {
 150                 err = -ENOMEM;
 151                 goto error_gpadl;
 152         }
 153
 154         init_completion(&open_info->waitevent);
 155
 156         open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
 157         open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
 158         open_msg->openid = newchannel->offermsg.child_relid;
 159         open_msg->child_relid = newchannel->offermsg.child_relid;
 160         open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
 161         open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
 162                                                   PAGE_SHIFT;
 163         open_msg->target_vp = newchannel->target_vp;
 164
 165         if (userdatalen > MAX_USER_DEFINED_BYTES) {
 166                 err = -EINVAL;
 167                 goto error_gpadl;
 168         }
 169
 170         if (userdatalen)
 171                 memcpy(open_msg->userdata, userdata, userdatalen);
 172
 173         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 174         list_add_tail(&open_info->msglistentry,
 175                       &vmbus_connection.chn_msg_list);
 176         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 177
 178         ret = vmbus_post_msg(open_msg,
 179                                sizeof(struct vmbus_channel_open_channel));
 180
 181         if (ret != 0) {
 182                 err = ret;
 183                 goto error1;
 184         }
 185
 186         t = wait_for_completion_timeout(&open_info->waitevent, 5*HZ);
 187         if (t == 0) {
 188                 err = -ETIMEDOUT;
 189                 goto error1;
 190         }
 191
 192         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 193         list_del(&open_info->msglistentry);
 194         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 195
 196         if (open_info->response.open_result.status) {
 197                 err = -EAGAIN;
 198                 goto error_gpadl;
 199         }
 200
 201         newchannel->state = CHANNEL_OPENED_STATE;
 202         kfree(open_info);
 203         return 0;
 204
 205 error1:
 206         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 207         list_del(&open_info->msglistentry);
 208         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 209
 210 error_gpadl:
 211         vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
 212
 213 error0:
 214         free_pages((unsigned long)out,
 215                 get_order(send_ringbuffer_size + recv_ringbuffer_size));
 216         kfree(open_info);
 217         newchannel->state = CHANNEL_OPEN_STATE;
 218         return err;
 219 }
 220 EXPORT_SYMBOL_GPL(vmbus_open);
 221
 222 /* Used for Hyper-V Socket: a guest client's connect() to the host */
 223 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
 224                                   const uuid_le *shv_host_servie_id)
 225 {
 226         struct vmbus_channel_tl_connect_request conn_msg;
 227
 228         memset(&conn_msg, 0, sizeof(conn_msg));
 229         conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
 230         conn_msg.guest_endpoint_id = *shv_guest_servie_id;
 231         conn_msg.host_service_id = *shv_host_servie_id;
 232
 233         return vmbus_post_msg(&conn_msg, sizeof(conn_msg));
 234 }
 235 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
 236
 237 /*
 238  * create_gpadl_header - Creates a gpadl for the specified buffer
 239  */
 240 static int create_gpadl_header(void *kbuffer, u32 size,
 241                                struct vmbus_channel_msginfo **msginfo)
 242 {
 243         int i;
 244         int pagecount;
 245         struct vmbus_channel_gpadl_header *gpadl_header;
 246         struct vmbus_channel_gpadl_body *gpadl_body;
 247         struct vmbus_channel_msginfo *msgheader;
 248         struct vmbus_channel_msginfo *msgbody = NULL;
 249         u32 msgsize;
 250
 251         int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
 252
 253         pagecount = size >> PAGE_SHIFT;
 254
 255         /* do we need a gpadl body msg */
 256         pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 257                   sizeof(struct vmbus_channel_gpadl_header) -
 258                   sizeof(struct gpa_range);
 259         pfncount = pfnsize / sizeof(u64);
 260
 261         if (pagecount > pfncount) {
 262                 /* we need a gpadl body */
 263                 /* fill in the header */
 264                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 265                           sizeof(struct vmbus_channel_gpadl_header) +
 266                           sizeof(struct gpa_range) + pfncount * sizeof(u64);
 267                 msgheader =  kzalloc(msgsize, GFP_KERNEL);
 268                 if (!msgheader)
 269                         goto nomem;
 270
 271                 INIT_LIST_HEAD(&msgheader->submsglist);
 272                 msgheader->msgsize = msgsize;
 273
 274                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 275                         msgheader->msg;
 276                 gpadl_header->rangecount = 1;
 277                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 278                                          pagecount * sizeof(u64);
 279                 gpadl_header->range[0].byte_offset = 0;
 280                 gpadl_header->range[0].byte_count = size;
 281                 for (i = 0; i < pfncount; i++)
 282                         gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
 283                                 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
 284                 *msginfo = msgheader;
 285
 286                 pfnsum = pfncount;
 287                 pfnleft = pagecount - pfncount;
 288
 289                 /* how many pfns can we fit */
 290                 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 291                           sizeof(struct vmbus_channel_gpadl_body);
 292                 pfncount = pfnsize / sizeof(u64);
 293
 294                 /* fill in the body */
 295                 while (pfnleft) {
 296                         if (pfnleft > pfncount)
 297                                 pfncurr = pfncount;
 298                         else
 299                                 pfncurr = pfnleft;
 300
 301                         msgsize = sizeof(struct vmbus_channel_msginfo) +
 302                                   sizeof(struct vmbus_channel_gpadl_body) +
 303                                   pfncurr * sizeof(u64);
 304                         msgbody = kzalloc(msgsize, GFP_KERNEL);
 305
 306                         if (!msgbody) {
 307                                 struct vmbus_channel_msginfo *pos = NULL;
 308                                 struct vmbus_channel_msginfo *tmp = NULL;
 309                                 /*
 310                                  * Free up all the allocated messages.
 311                                  */
 312                                 list_for_each_entry_safe(pos, tmp,
 313                                         &msgheader->submsglist,
 314                                         msglistentry) {
 315
 316                                         list_del(&pos->msglistentry);
 317                                         kfree(pos);
 318                                 }
 319
 320                                 goto nomem;
 321                         }
 322
 323                         msgbody->msgsize = msgsize;
 324                         gpadl_body =
 325                                 (struct vmbus_channel_gpadl_body *)msgbody->msg;
 326
 327                         /*
 328                          * Gpadl is u32 and we are using a pointer which could
 329                          * be 64-bit
 330                          * This is governed by the guest/host protocol and
 331                          * so the hypervisor gurantees that this is ok.
 332                          */
 333                         for (i = 0; i < pfncurr; i++)
 334                                 gpadl_body->pfn[i] = slow_virt_to_phys(
 335                                         kbuffer + PAGE_SIZE * (pfnsum + i)) >>
 336                                         PAGE_SHIFT;
 337
 338                         /* add to msg header */
 339                         list_add_tail(&msgbody->msglistentry,
 340                                       &msgheader->submsglist);
 341                         pfnsum += pfncurr;
 342                         pfnleft -= pfncurr;
 343                 }
 344         } else {
 345                 /* everything fits in a header */
 346                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 347                           sizeof(struct vmbus_channel_gpadl_header) +
 348                           sizeof(struct gpa_range) + pagecount * sizeof(u64);
 349                 msgheader = kzalloc(msgsize, GFP_KERNEL);
 350                 if (msgheader == NULL)
 351                         goto nomem;
 352
 353                 INIT_LIST_HEAD(&msgheader->submsglist);
 354                 msgheader->msgsize = msgsize;
 355
 356                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 357                         msgheader->msg;
 358                 gpadl_header->rangecount = 1;
 359                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 360                                          pagecount * sizeof(u64);
 361                 gpadl_header->range[0].byte_offset = 0;
 362                 gpadl_header->range[0].byte_count = size;
 363                 for (i = 0; i < pagecount; i++)
 364                         gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
 365                                 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
 366
 367                 *msginfo = msgheader;
 368         }
 369
 370         return 0;
 371 nomem:
 372         kfree(msgheader);
 373         kfree(msgbody);
 374         return -ENOMEM;
 375 }
 376
 377 /*
 378  * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
 379  *
 380  * @channel: a channel
 381  * @kbuffer: from kmalloc or vmalloc
 382  * @size: page-size multiple
 383  * @gpadl_handle: some funky thing
 384  */
 385 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 386                                u32 size, u32 *gpadl_handle)
 387 {
 388         struct vmbus_channel_gpadl_header *gpadlmsg;
 389         struct vmbus_channel_gpadl_body *gpadl_body;
 390         struct vmbus_channel_msginfo *msginfo = NULL;
 391         struct vmbus_channel_msginfo *submsginfo, *tmp;
 392         struct list_head *curr;
 393         u32 next_gpadl_handle;
 394         unsigned long flags;
 395         int ret = 0;
 396
 397         next_gpadl_handle =
 398                 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
 399
 400         ret = create_gpadl_header(kbuffer, size, &msginfo);
 401         if (ret)
 402                 return ret;
 403
 404         init_completion(&msginfo->waitevent);
 405
 406         gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
 407         gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
 408         gpadlmsg->child_relid = channel->offermsg.child_relid;
 409         gpadlmsg->gpadl = next_gpadl_handle;
 410
 411
 412         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 413         list_add_tail(&msginfo->msglistentry,
 414                       &vmbus_connection.chn_msg_list);
 415
 416         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 417
 418         ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
 419                                sizeof(*msginfo));
 420         if (ret != 0)
 421                 goto cleanup;
 422
 423         list_for_each(curr, &msginfo->submsglist) {
 424                 submsginfo = (struct vmbus_channel_msginfo *)curr;
 425                 gpadl_body =
 426                         (struct vmbus_channel_gpadl_body *)submsginfo->msg;
 427
 428                 gpadl_body->header.msgtype =
 429                         CHANNELMSG_GPADL_BODY;
 430                 gpadl_body->gpadl = next_gpadl_handle;
 431
 432                 ret = vmbus_post_msg(gpadl_body,
 433                                      submsginfo->msgsize -
 434                                      sizeof(*submsginfo));
 435                 if (ret != 0)
 436                         goto cleanup;
 437
 438         }
 439         wait_for_completion(&msginfo->waitevent);
 440
 441         /* At this point, we received the gpadl created msg */
 442         *gpadl_handle = gpadlmsg->gpadl;
 443
 444 cleanup:
 445         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 446         list_del(&msginfo->msglistentry);
 447         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 448         list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
 449                                  msglistentry) {
 450                 kfree(submsginfo);
 451         }
 452
 453         kfree(msginfo);
 454         return ret;
 455 }
 456 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
 457
 458 /*
 459  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
 460  */
 461 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
 462 {
 463         struct vmbus_channel_gpadl_teardown *msg;
 464         struct vmbus_channel_msginfo *info;
 465         unsigned long flags;
 466         int ret;
 467
 468         info = kmalloc(sizeof(*info) +
 469                        sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
 470         if (!info)
 471                 return -ENOMEM;
 472
 473         init_completion(&info->waitevent);
 474
 475         msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
 476
 477         msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
 478         msg->child_relid = channel->offermsg.child_relid;
 479         msg->gpadl = gpadl_handle;
 480
 481         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 482         list_add_tail(&info->msglistentry,
 483                       &vmbus_connection.chn_msg_list);
 484         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 485         ret = vmbus_post_msg(msg,
 486                                sizeof(struct vmbus_channel_gpadl_teardown));
 487
 488         if (ret)
 489                 goto post_msg_err;
 490
 491         wait_for_completion(&info->waitevent);
 492
 493 post_msg_err:
 494         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 495         list_del(&info->msglistentry);
 496         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 497
 498         kfree(info);
 499         return ret;
 500 }
 501 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
 502
 503 static void reset_channel_cb(void *arg)
 504 {
 505         struct vmbus_channel *channel = arg;
 506
 507         channel->onchannel_callback = NULL;
 508 }
 509
 510 static int vmbus_close_internal(struct vmbus_channel *channel)
 511 {
 512         struct vmbus_channel_close_channel *msg;
 513         struct tasklet_struct *tasklet;
 514         int ret;
 515
 516         /*
 517          * process_chn_event(), running in the tasklet, can race
 518          * with vmbus_close_internal() in the case of SMP guest, e.g., when
 519          * the former is accessing channel->inbound.ring_buffer, the latter
 520          * could be freeing the ring_buffer pages.
 521          *
 522          * To resolve the race, we can serialize them by disabling the
 523          * tasklet when the latter is running here.
 524          */
 525         tasklet = hv_context.event_dpc[channel->target_cpu];
 526         tasklet_disable(tasklet);
 527
 528         /*
 529          * In case a device driver's probe() fails (e.g.,
 530          * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
 531          * rescinded later (e.g., we dynamically disble an Integrated Service
 532          * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
 533          * here we should skip most of the below cleanup work.
 534          */
 535         if (channel->state != CHANNEL_OPENED_STATE) {
 536                 ret = -EINVAL;
 537                 goto out;
 538         }
 539
 540         channel->state = CHANNEL_OPEN_STATE;
 541         channel->sc_creation_callback = NULL;
 542         /* Stop callback and cancel the timer asap */
 543         if (channel->target_cpu != get_cpu()) {
 544                 put_cpu();
 545                 smp_call_function_single(channel->target_cpu, reset_channel_cb,
 546                                          channel, true);
 547         } else {
 548                 reset_channel_cb(channel);
 549                 put_cpu();
 550         }
 551
 552         /* Send a closing message */
 553
 554         msg = &channel->close_msg.msg;
 555
 556         msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
 557         msg->child_relid = channel->offermsg.child_relid;
 558
 559         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
 560
 561         if (ret) {
 562                 pr_err("Close failed: close post msg return is %d\n", ret);
 563                 /*
 564                  * If we failed to post the close msg,
 565                  * it is perhaps better to leak memory.
 566                  */
 567                 goto out;
 568         }
 569
 570         /* Tear down the gpadl for the channel's ring buffer */
 571         if (channel->ringbuffer_gpadlhandle) {
 572                 ret = vmbus_teardown_gpadl(channel,
 573                                            channel->ringbuffer_gpadlhandle);
 574                 if (ret) {
 575                         pr_err("Close failed: teardown gpadl return %d\n", ret);
 576                         /*
 577                          * If we failed to teardown gpadl,
 578                          * it is perhaps better to leak memory.
 579                          */
 580                         goto out;
 581                 }
 582         }
 583
 584         /* Cleanup the ring buffers for this channel */
 585         hv_ringbuffer_cleanup(&channel->outbound);
 586         hv_ringbuffer_cleanup(&channel->inbound);
 587
 588         free_pages((unsigned long)channel->ringbuffer_pages,
 589                 get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
 590
 591 out:
 592         tasklet_enable(tasklet);
 593
 594         return ret;
 595 }
 596
 597 /*
 598  * vmbus_close - Close the specified channel
 599  */
 600 void vmbus_close(struct vmbus_channel *channel)
 601 {
 602         struct list_head *cur, *tmp;
 603         struct vmbus_channel *cur_channel;
 604
 605         if (channel->primary_channel != NULL) {
 606                 /*
 607                  * We will only close sub-channels when
 608                  * the primary is closed.
 609                  */
 610                 return;
 611         }
 612         /*
 613          * Close all the sub-channels first and then close the
 614          * primary channel.
 615          */
 616         list_for_each_safe(cur, tmp, &channel->sc_list) {
 617                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 618                 if (cur_channel->state != CHANNEL_OPENED_STATE)
 619                         continue;
 620                 vmbus_close_internal(cur_channel);
 621         }
 622         /*
 623          * Now close the primary.
 624          */
 625         vmbus_close_internal(channel);
 626 }
 627 EXPORT_SYMBOL_GPL(vmbus_close);
 628
 629 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
 630                            u32 bufferlen, u64 requestid,
 631                            enum vmbus_packet_type type, u32 flags, bool kick_q)
 632 {
 633         struct vmpacket_descriptor desc;
 634         u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
 635         u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 636         struct kvec bufferlist[3];
 637         u64 aligned_data = 0;
 638         int ret;
 639         bool signal = false;
 640         bool lock = channel->acquire_ring_lock;
 641         int num_vecs = ((bufferlen != 0) ? 3 : 1);
 642
 643
 644         /* Setup the descriptor */
 645         desc.type = type; /* VmbusPacketTypeDataInBand; */
 646         desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
 647         /* in 8-bytes granularity */
 648         desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
 649         desc.len8 = (u16)(packetlen_aligned >> 3);
 650         desc.trans_id = requestid;
 651
 652         bufferlist[0].iov_base = &desc;
 653         bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
 654         bufferlist[1].iov_base = buffer;
 655         bufferlist[1].iov_len = bufferlen;
 656         bufferlist[2].iov_base = &aligned_data;
 657         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 658
 659         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
 660                                   &signal, lock);
 661
 662         /*
 663          * Signalling the host is conditional on many factors:
 664          * 1. The ring state changed from being empty to non-empty.
 665          *    This is tracked by the variable "signal".
 666          * 2. The variable kick_q tracks if more data will be placed
 667          *    on the ring. We will not signal if more data is
 668          *    to be placed.
 669          *
 670          * Based on the channel signal state, we will decide
 671          * which signaling policy will be applied.
 672          *
 673          * If we cannot write to the ring-buffer; signal the host
 674          * even if we may not have written anything. This is a rare
 675          * enough condition that it should not matter.
 676          * NOTE: in this case, the hvsock channel is an exception, because
 677          * it looks the host side's hvsock implementation has a throttling
 678          * mechanism which can hurt the performance otherwise.
 679          */
 680
 681         if (channel->signal_policy)
 682                 signal = true;
 683         else
 684                 kick_q = true;
 685
 686         if (((ret == 0) && kick_q && signal) ||
 687             (ret && !is_hvsock_channel(channel)))
 688                 vmbus_setevent(channel);
 689
 690         return ret;
 691 }
 692 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
 693
 694 /**
 695  * vmbus_sendpacket() - Send the specified buffer on the given channel
 696  * @channel: Pointer to vmbus_channel structure.
 697  * @buffer: Pointer to the buffer you want to receive the data into.
 698  * @bufferlen: Maximum size of what the the buffer will hold
 699  * @requestid: Identifier of the request
 700  * @type: Type of packet that is being send e.g. negotiate, time
 701  * packet etc.
 702  *
 703  * Sends data in @buffer directly to hyper-v via the vmbus
 704  * This will send the data unparsed to hyper-v.
 705  *
 706  * Mainly used by Hyper-V drivers.
 707  */
 708 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
 709                            u32 bufferlen, u64 requestid,
 710                            enum vmbus_packet_type type, u32 flags)
 711 {
 712         return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
 713                                     type, flags, true);
 714 }
 715 EXPORT_SYMBOL(vmbus_sendpacket);
 716
 717 /*
 718  * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
 719  * packets using a GPADL Direct packet type. This interface allows you
 720  * to control notifying the host. This will be useful for sending
 721  * batched data. Also the sender can control the send flags
 722  * explicitly.
 723  */
 724 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
 725                                      struct hv_page_buffer pagebuffers[],
 726                                      u32 pagecount, void *buffer, u32 bufferlen,
 727                                      u64 requestid,
 728                                      u32 flags,
 729                                      bool kick_q)
 730 {
 731         int ret;
 732         int i;
 733         struct vmbus_channel_packet_page_buffer desc;
 734         u32 descsize;
 735         u32 packetlen;
 736         u32 packetlen_aligned;
 737         struct kvec bufferlist[3];
 738         u64 aligned_data = 0;
 739         bool signal = false;
 740         bool lock = channel->acquire_ring_lock;
 741
 742         if (pagecount > MAX_PAGE_BUFFER_COUNT)
 743                 return -EINVAL;
 744
 745
 746         /*
 747          * Adjust the size down since vmbus_channel_packet_page_buffer is the
 748          * largest size we support
 749          */
 750         descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
 751                           ((MAX_PAGE_BUFFER_COUNT - pagecount) *
 752                           sizeof(struct hv_page_buffer));
 753         packetlen = descsize + bufferlen;
 754         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 755
 756         /* Setup the descriptor */
 757         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
 758         desc.flags = flags;
 759         desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
 760         desc.length8 = (u16)(packetlen_aligned >> 3);
 761         desc.transactionid = requestid;
 762         desc.rangecount = pagecount;
 763
 764         for (i = 0; i < pagecount; i++) {
 765                 desc.range[i].len = pagebuffers[i].len;
 766                 desc.range[i].offset = pagebuffers[i].offset;
 767                 desc.range[i].pfn        = pagebuffers[i].pfn;
 768         }
 769
 770         bufferlist[0].iov_base = &desc;
 771         bufferlist[0].iov_len = descsize;
 772         bufferlist[1].iov_base = buffer;
 773         bufferlist[1].iov_len = bufferlen;
 774         bufferlist[2].iov_base = &aligned_data;
 775         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 776
 777         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 778                                   &signal, lock);
 779
 780         /*
 781          * Signalling the host is conditional on many factors:
 782          * 1. The ring state changed from being empty to non-empty.
 783          *    This is tracked by the variable "signal".
 784          * 2. The variable kick_q tracks if more data will be placed
 785          *    on the ring. We will not signal if more data is
 786          *    to be placed.
 787          *
 788          * Based on the channel signal state, we will decide
 789          * which signaling policy will be applied.
 790          *
 791          * If we cannot write to the ring-buffer; signal the host
 792          * even if we may not have written anything. This is a rare
 793          * enough condition that it should not matter.
 794          */
 795
 796         if (channel->signal_policy)
 797                 signal = true;
 798         else
 799                 kick_q = true;
 800
 801         if (((ret == 0) && kick_q && signal) || (ret))
 802                 vmbus_setevent(channel);
 803
 804         return ret;
 805 }
 806 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
 807
 808 /*
 809  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
 810  * packets using a GPADL Direct packet type.
 811  */
 812 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
 813                                      struct hv_page_buffer pagebuffers[],
 814                                      u32 pagecount, void *buffer, u32 bufferlen,
 815                                      u64 requestid)
 816 {
 817         u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 818         return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
 819                                                buffer, bufferlen, requestid,
 820                                                flags, true);
 821
 822 }
 823 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
 824
 825 /*
 826  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
 827  * using a GPADL Direct packet type.
 828  * The buffer includes the vmbus descriptor.
 829  */
 830 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
 831                               struct vmbus_packet_mpb_array *desc,
 832                               u32 desc_size,
 833                               void *buffer, u32 bufferlen, u64 requestid)
 834 {
 835         int ret;
 836         u32 packetlen;
 837         u32 packetlen_aligned;
 838         struct kvec bufferlist[3];
 839         u64 aligned_data = 0;
 840         bool signal = false;
 841         bool lock = channel->acquire_ring_lock;
 842
 843         packetlen = desc_size + bufferlen;
 844         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 845
 846         /* Setup the descriptor */
 847         desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
 848         desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 849         desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
 850         desc->length8 = (u16)(packetlen_aligned >> 3);
 851         desc->transactionid = requestid;
 852         desc->rangecount = 1;
 853
 854         bufferlist[0].iov_base = desc;
 855         bufferlist[0].iov_len = desc_size;
 856         bufferlist[1].iov_base = buffer;
 857         bufferlist[1].iov_len = bufferlen;
 858         bufferlist[2].iov_base = &aligned_data;
 859         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 860
 861         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 862                                   &signal, lock);
 863
 864         if (ret == 0 && signal)
 865                 vmbus_setevent(channel);
 866
 867         return ret;
 868 }
 869 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
 870
 871 /*
 872  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
 873  * using a GPADL Direct packet type.
 874  */
 875 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
 876                                 struct hv_multipage_buffer *multi_pagebuffer,
 877                                 void *buffer, u32 bufferlen, u64 requestid)
 878 {
 879         int ret;
 880         struct vmbus_channel_packet_multipage_buffer desc;
 881         u32 descsize;
 882         u32 packetlen;
 883         u32 packetlen_aligned;
 884         struct kvec bufferlist[3];
 885         u64 aligned_data = 0;
 886         bool signal = false;
 887         bool lock = channel->acquire_ring_lock;
 888         u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
 889                                          multi_pagebuffer->len);
 890
 891         if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
 892                 return -EINVAL;
 893
 894         /*
 895          * Adjust the size down since vmbus_channel_packet_multipage_buffer is
 896          * the largest size we support
 897          */
 898         descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
 899                           ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
 900                           sizeof(u64));
 901         packetlen = descsize + bufferlen;
 902         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 903
 904
 905         /* Setup the descriptor */
 906         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
 907         desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 908         desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
 909         desc.length8 = (u16)(packetlen_aligned >> 3);
 910         desc.transactionid = requestid;
 911         desc.rangecount = 1;
 912
 913         desc.range.len = multi_pagebuffer->len;
 914         desc.range.offset = multi_pagebuffer->offset;
 915
 916         memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
 917                pfncount * sizeof(u64));
 918
 919         bufferlist[0].iov_base = &desc;
 920         bufferlist[0].iov_len = descsize;
 921         bufferlist[1].iov_base = buffer;
 922         bufferlist[1].iov_len = bufferlen;
 923         bufferlist[2].iov_base = &aligned_data;
 924         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 925
 926         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 927                                   &signal, lock);
 928
 929         if (ret == 0 && signal)
 930                 vmbus_setevent(channel);
 931
 932         return ret;
 933 }
 934 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
 935
 936 /**
 937  * vmbus_recvpacket() - Retrieve the user packet on the specified channel
 938  * @channel: Pointer to vmbus_channel structure.
 939  * @buffer: Pointer to the buffer you want to receive the data into.
 940  * @bufferlen: Maximum size of what the the buffer will hold
 941  * @buffer_actual_len: The actual size of the data after it was received
 942  * @requestid: Identifier of the request
 943  *
 944  * Receives directly from the hyper-v vmbus and puts the data it received
 945  * into Buffer. This will receive the data unparsed from hyper-v.
 946  *
 947  * Mainly used by Hyper-V drivers.
 948  */
 949 static inline int
 950 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
 951                    u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
 952                    bool raw)
 953 {
 954         int ret;
 955         bool signal = false;
 956
 957         ret = hv_ringbuffer_read(&channel->inbound, buffer, bufferlen,
 958                                  buffer_actual_len, requestid, &signal, raw);
 959
 960         if (signal)
 961                 vmbus_setevent(channel);
 962
 963         return ret;
 964 }
 965
 966 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
 967                      u32 bufferlen, u32 *buffer_actual_len,
 968                      u64 *requestid)
 969 {
 970         return __vmbus_recvpacket(channel, buffer, bufferlen,
 971                                   buffer_actual_len, requestid, false);
 972 }
 973 EXPORT_SYMBOL(vmbus_recvpacket);
 974
 975 /*
 976  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
 977  */
 978 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
 979                               u32 bufferlen, u32 *buffer_actual_len,
 980                               u64 *requestid)
 981 {
 982         return __vmbus_recvpacket(channel, buffer, bufferlen,
 983                                   buffer_actual_len, requestid, true);
 984 }
 985 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);