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 /*
 223  * create_gpadl_header - Creates a gpadl for the specified buffer
 224  */
 225 static int create_gpadl_header(void *kbuffer, u32 size,
 226                                          struct vmbus_channel_msginfo **msginfo,
 227                                          u32 *messagecount)
 228 {
 229         int i;
 230         int pagecount;
 231         struct vmbus_channel_gpadl_header *gpadl_header;
 232         struct vmbus_channel_gpadl_body *gpadl_body;
 233         struct vmbus_channel_msginfo *msgheader;
 234         struct vmbus_channel_msginfo *msgbody = NULL;
 235         u32 msgsize;
 236
 237         int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
 238
 239         pagecount = size >> PAGE_SHIFT;
 240
 241         /* do we need a gpadl body msg */
 242         pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 243                   sizeof(struct vmbus_channel_gpadl_header) -
 244                   sizeof(struct gpa_range);
 245         pfncount = pfnsize / sizeof(u64);
 246
 247         if (pagecount > pfncount) {
 248                 /* we need a gpadl body */
 249                 /* fill in the header */
 250                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 251                           sizeof(struct vmbus_channel_gpadl_header) +
 252                           sizeof(struct gpa_range) + pfncount * sizeof(u64);
 253                 msgheader =  kzalloc(msgsize, GFP_KERNEL);
 254                 if (!msgheader)
 255                         goto nomem;
 256
 257                 INIT_LIST_HEAD(&msgheader->submsglist);
 258                 msgheader->msgsize = msgsize;
 259
 260                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 261                         msgheader->msg;
 262                 gpadl_header->rangecount = 1;
 263                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 264                                          pagecount * sizeof(u64);
 265                 gpadl_header->range[0].byte_offset = 0;
 266                 gpadl_header->range[0].byte_count = size;
 267                 for (i = 0; i < pfncount; i++)
 268                         gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
 269                                 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
 270                 *msginfo = msgheader;
 271                 *messagecount = 1;
 272
 273                 pfnsum = pfncount;
 274                 pfnleft = pagecount - pfncount;
 275
 276                 /* how many pfns can we fit */
 277                 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
 278                           sizeof(struct vmbus_channel_gpadl_body);
 279                 pfncount = pfnsize / sizeof(u64);
 280
 281                 /* fill in the body */
 282                 while (pfnleft) {
 283                         if (pfnleft > pfncount)
 284                                 pfncurr = pfncount;
 285                         else
 286                                 pfncurr = pfnleft;
 287
 288                         msgsize = sizeof(struct vmbus_channel_msginfo) +
 289                                   sizeof(struct vmbus_channel_gpadl_body) +
 290                                   pfncurr * sizeof(u64);
 291                         msgbody = kzalloc(msgsize, GFP_KERNEL);
 292
 293                         if (!msgbody) {
 294                                 struct vmbus_channel_msginfo *pos = NULL;
 295                                 struct vmbus_channel_msginfo *tmp = NULL;
 296                                 /*
 297                                  * Free up all the allocated messages.
 298                                  */
 299                                 list_for_each_entry_safe(pos, tmp,
 300                                         &msgheader->submsglist,
 301                                         msglistentry) {
 302
 303                                         list_del(&pos->msglistentry);
 304                                         kfree(pos);
 305                                 }
 306
 307                                 goto nomem;
 308                         }
 309
 310                         msgbody->msgsize = msgsize;
 311                         (*messagecount)++;
 312                         gpadl_body =
 313                                 (struct vmbus_channel_gpadl_body *)msgbody->msg;
 314
 315                         /*
 316                          * Gpadl is u32 and we are using a pointer which could
 317                          * be 64-bit
 318                          * This is governed by the guest/host protocol and
 319                          * so the hypervisor gurantees that this is ok.
 320                          */
 321                         for (i = 0; i < pfncurr; i++)
 322                                 gpadl_body->pfn[i] = slow_virt_to_phys(
 323                                         kbuffer + PAGE_SIZE * (pfnsum + i)) >>
 324                                         PAGE_SHIFT;
 325
 326                         /* add to msg header */
 327                         list_add_tail(&msgbody->msglistentry,
 328                                       &msgheader->submsglist);
 329                         pfnsum += pfncurr;
 330                         pfnleft -= pfncurr;
 331                 }
 332         } else {
 333                 /* everything fits in a header */
 334                 msgsize = sizeof(struct vmbus_channel_msginfo) +
 335                           sizeof(struct vmbus_channel_gpadl_header) +
 336                           sizeof(struct gpa_range) + pagecount * sizeof(u64);
 337                 msgheader = kzalloc(msgsize, GFP_KERNEL);
 338                 if (msgheader == NULL)
 339                         goto nomem;
 340                 msgheader->msgsize = msgsize;
 341
 342                 gpadl_header = (struct vmbus_channel_gpadl_header *)
 343                         msgheader->msg;
 344                 gpadl_header->rangecount = 1;
 345                 gpadl_header->range_buflen = sizeof(struct gpa_range) +
 346                                          pagecount * sizeof(u64);
 347                 gpadl_header->range[0].byte_offset = 0;
 348                 gpadl_header->range[0].byte_count = size;
 349                 for (i = 0; i < pagecount; i++)
 350                         gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
 351                                 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
 352
 353                 *msginfo = msgheader;
 354                 *messagecount = 1;
 355         }
 356
 357         return 0;
 358 nomem:
 359         kfree(msgheader);
 360         kfree(msgbody);
 361         return -ENOMEM;
 362 }
 363
 364 /*
 365  * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
 366  *
 367  * @channel: a channel
 368  * @kbuffer: from kmalloc or vmalloc
 369  * @size: page-size multiple
 370  * @gpadl_handle: some funky thing
 371  */
 372 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
 373                                u32 size, u32 *gpadl_handle)
 374 {
 375         struct vmbus_channel_gpadl_header *gpadlmsg;
 376         struct vmbus_channel_gpadl_body *gpadl_body;
 377         struct vmbus_channel_msginfo *msginfo = NULL;
 378         struct vmbus_channel_msginfo *submsginfo;
 379         u32 msgcount;
 380         struct list_head *curr;
 381         u32 next_gpadl_handle;
 382         unsigned long flags;
 383         int ret = 0;
 384
 385         next_gpadl_handle =
 386                 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
 387
 388         ret = create_gpadl_header(kbuffer, size, &msginfo, &msgcount);
 389         if (ret)
 390                 return ret;
 391
 392         init_completion(&msginfo->waitevent);
 393
 394         gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
 395         gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
 396         gpadlmsg->child_relid = channel->offermsg.child_relid;
 397         gpadlmsg->gpadl = next_gpadl_handle;
 398
 399
 400         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 401         list_add_tail(&msginfo->msglistentry,
 402                       &vmbus_connection.chn_msg_list);
 403
 404         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 405
 406         ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
 407                                sizeof(*msginfo));
 408         if (ret != 0)
 409                 goto cleanup;
 410
 411         if (msgcount > 1) {
 412                 list_for_each(curr, &msginfo->submsglist) {
 413
 414                         submsginfo = (struct vmbus_channel_msginfo *)curr;
 415                         gpadl_body =
 416                              (struct vmbus_channel_gpadl_body *)submsginfo->msg;
 417
 418                         gpadl_body->header.msgtype =
 419                                 CHANNELMSG_GPADL_BODY;
 420                         gpadl_body->gpadl = next_gpadl_handle;
 421
 422                         ret = vmbus_post_msg(gpadl_body,
 423                                                submsginfo->msgsize -
 424                                                sizeof(*submsginfo));
 425                         if (ret != 0)
 426                                 goto cleanup;
 427
 428                 }
 429         }
 430         wait_for_completion(&msginfo->waitevent);
 431
 432         /* At this point, we received the gpadl created msg */
 433         *gpadl_handle = gpadlmsg->gpadl;
 434
 435 cleanup:
 436         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 437         list_del(&msginfo->msglistentry);
 438         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 439
 440         kfree(msginfo);
 441         return ret;
 442 }
 443 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
 444
 445 /*
 446  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
 447  */
 448 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
 449 {
 450         struct vmbus_channel_gpadl_teardown *msg;
 451         struct vmbus_channel_msginfo *info;
 452         unsigned long flags;
 453         int ret;
 454
 455         info = kmalloc(sizeof(*info) +
 456                        sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
 457         if (!info)
 458                 return -ENOMEM;
 459
 460         init_completion(&info->waitevent);
 461
 462         msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
 463
 464         msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
 465         msg->child_relid = channel->offermsg.child_relid;
 466         msg->gpadl = gpadl_handle;
 467
 468         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 469         list_add_tail(&info->msglistentry,
 470                       &vmbus_connection.chn_msg_list);
 471         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 472         ret = vmbus_post_msg(msg,
 473                                sizeof(struct vmbus_channel_gpadl_teardown));
 474
 475         if (ret)
 476                 goto post_msg_err;
 477
 478         wait_for_completion(&info->waitevent);
 479
 480 post_msg_err:
 481         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 482         list_del(&info->msglistentry);
 483         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 484
 485         kfree(info);
 486         return ret;
 487 }
 488 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
 489
 490 static void reset_channel_cb(void *arg)
 491 {
 492         struct vmbus_channel *channel = arg;
 493
 494         channel->onchannel_callback = NULL;
 495 }
 496
 497 static int vmbus_close_internal(struct vmbus_channel *channel)
 498 {
 499         struct vmbus_channel_close_channel *msg;
 500         struct tasklet_struct *tasklet;
 501         int ret;
 502
 503         /*
 504          * process_chn_event(), running in the tasklet, can race
 505          * with vmbus_close_internal() in the case of SMP guest, e.g., when
 506          * the former is accessing channel->inbound.ring_buffer, the latter
 507          * could be freeing the ring_buffer pages.
 508          *
 509          * To resolve the race, we can serialize them by disabling the
 510          * tasklet when the latter is running here.
 511          */
 512         tasklet = hv_context.event_dpc[channel->target_cpu];
 513         tasklet_disable(tasklet);
 514
 515         channel->state = CHANNEL_OPEN_STATE;
 516         channel->sc_creation_callback = NULL;
 517         /* Stop callback and cancel the timer asap */
 518         if (channel->target_cpu != get_cpu()) {
 519                 put_cpu();
 520                 smp_call_function_single(channel->target_cpu, reset_channel_cb,
 521                                          channel, true);
 522         } else {
 523                 reset_channel_cb(channel);
 524                 put_cpu();
 525         }
 526
 527         /* Send a closing message */
 528
 529         msg = &channel->close_msg.msg;
 530
 531         msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
 532         msg->child_relid = channel->offermsg.child_relid;
 533
 534         ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
 535
 536         if (ret) {
 537                 pr_err("Close failed: close post msg return is %d\n", ret);
 538                 /*
 539                  * If we failed to post the close msg,
 540                  * it is perhaps better to leak memory.
 541                  */
 542                 goto out;
 543         }
 544
 545         /* Tear down the gpadl for the channel's ring buffer */
 546         if (channel->ringbuffer_gpadlhandle) {
 547                 ret = vmbus_teardown_gpadl(channel,
 548                                            channel->ringbuffer_gpadlhandle);
 549                 if (ret) {
 550                         pr_err("Close failed: teardown gpadl return %d\n", ret);
 551                         /*
 552                          * If we failed to teardown gpadl,
 553                          * it is perhaps better to leak memory.
 554                          */
 555                         goto out;
 556                 }
 557         }
 558
 559         /* Cleanup the ring buffers for this channel */
 560         hv_ringbuffer_cleanup(&channel->outbound);
 561         hv_ringbuffer_cleanup(&channel->inbound);
 562
 563         free_pages((unsigned long)channel->ringbuffer_pages,
 564                 get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
 565
 566 out:
 567         tasklet_enable(tasklet);
 568
 569         return ret;
 570 }
 571
 572 /*
 573  * vmbus_close - Close the specified channel
 574  */
 575 void vmbus_close(struct vmbus_channel *channel)
 576 {
 577         struct list_head *cur, *tmp;
 578         struct vmbus_channel *cur_channel;
 579
 580         if (channel->primary_channel != NULL) {
 581                 /*
 582                  * We will only close sub-channels when
 583                  * the primary is closed.
 584                  */
 585                 return;
 586         }
 587         /*
 588          * Close all the sub-channels first and then close the
 589          * primary channel.
 590          */
 591         list_for_each_safe(cur, tmp, &channel->sc_list) {
 592                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 593                 if (cur_channel->state != CHANNEL_OPENED_STATE)
 594                         continue;
 595                 vmbus_close_internal(cur_channel);
 596         }
 597         /*
 598          * Now close the primary.
 599          */
 600         vmbus_close_internal(channel);
 601 }
 602 EXPORT_SYMBOL_GPL(vmbus_close);
 603
 604 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
 605                            u32 bufferlen, u64 requestid,
 606                            enum vmbus_packet_type type, u32 flags, bool kick_q)
 607 {
 608         struct vmpacket_descriptor desc;
 609         u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
 610         u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 611         struct kvec bufferlist[3];
 612         u64 aligned_data = 0;
 613         int ret;
 614         bool signal = false;
 615         int num_vecs = ((bufferlen != 0) ? 3 : 1);
 616
 617
 618         /* Setup the descriptor */
 619         desc.type = type; /* VmbusPacketTypeDataInBand; */
 620         desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
 621         /* in 8-bytes granularity */
 622         desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
 623         desc.len8 = (u16)(packetlen_aligned >> 3);
 624         desc.trans_id = requestid;
 625
 626         bufferlist[0].iov_base = &desc;
 627         bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
 628         bufferlist[1].iov_base = buffer;
 629         bufferlist[1].iov_len = bufferlen;
 630         bufferlist[2].iov_base = &aligned_data;
 631         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 632
 633         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
 634                                   &signal);
 635
 636         /*
 637          * Signalling the host is conditional on many factors:
 638          * 1. The ring state changed from being empty to non-empty.
 639          *    This is tracked by the variable "signal".
 640          * 2. The variable kick_q tracks if more data will be placed
 641          *    on the ring. We will not signal if more data is
 642          *    to be placed.
 643          *
 644          * Based on the channel signal state, we will decide
 645          * which signaling policy will be applied.
 646          *
 647          * If we cannot write to the ring-buffer; signal the host
 648          * even if we may not have written anything. This is a rare
 649          * enough condition that it should not matter.
 650          */
 651
 652         if (channel->signal_policy)
 653                 signal = true;
 654         else
 655                 kick_q = true;
 656
 657         if (((ret == 0) && kick_q && signal) || (ret))
 658                 vmbus_setevent(channel);
 659
 660         return ret;
 661 }
 662 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
 663
 664 /**
 665  * vmbus_sendpacket() - Send the specified buffer on the given channel
 666  * @channel: Pointer to vmbus_channel structure.
 667  * @buffer: Pointer to the buffer you want to receive the data into.
 668  * @bufferlen: Maximum size of what the the buffer will hold
 669  * @requestid: Identifier of the request
 670  * @type: Type of packet that is being send e.g. negotiate, time
 671  * packet etc.
 672  *
 673  * Sends data in @buffer directly to hyper-v via the vmbus
 674  * This will send the data unparsed to hyper-v.
 675  *
 676  * Mainly used by Hyper-V drivers.
 677  */
 678 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
 679                            u32 bufferlen, u64 requestid,
 680                            enum vmbus_packet_type type, u32 flags)
 681 {
 682         return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
 683                                     type, flags, true);
 684 }
 685 EXPORT_SYMBOL(vmbus_sendpacket);
 686
 687 /*
 688  * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
 689  * packets using a GPADL Direct packet type. This interface allows you
 690  * to control notifying the host. This will be useful for sending
 691  * batched data. Also the sender can control the send flags
 692  * explicitly.
 693  */
 694 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
 695                                      struct hv_page_buffer pagebuffers[],
 696                                      u32 pagecount, void *buffer, u32 bufferlen,
 697                                      u64 requestid,
 698                                      u32 flags,
 699                                      bool kick_q)
 700 {
 701         int ret;
 702         int i;
 703         struct vmbus_channel_packet_page_buffer desc;
 704         u32 descsize;
 705         u32 packetlen;
 706         u32 packetlen_aligned;
 707         struct kvec bufferlist[3];
 708         u64 aligned_data = 0;
 709         bool signal = false;
 710
 711         if (pagecount > MAX_PAGE_BUFFER_COUNT)
 712                 return -EINVAL;
 713
 714
 715         /*
 716          * Adjust the size down since vmbus_channel_packet_page_buffer is the
 717          * largest size we support
 718          */
 719         descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
 720                           ((MAX_PAGE_BUFFER_COUNT - pagecount) *
 721                           sizeof(struct hv_page_buffer));
 722         packetlen = descsize + bufferlen;
 723         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 724
 725         /* Setup the descriptor */
 726         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
 727         desc.flags = flags;
 728         desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
 729         desc.length8 = (u16)(packetlen_aligned >> 3);
 730         desc.transactionid = requestid;
 731         desc.rangecount = pagecount;
 732
 733         for (i = 0; i < pagecount; i++) {
 734                 desc.range[i].len = pagebuffers[i].len;
 735                 desc.range[i].offset = pagebuffers[i].offset;
 736                 desc.range[i].pfn        = pagebuffers[i].pfn;
 737         }
 738
 739         bufferlist[0].iov_base = &desc;
 740         bufferlist[0].iov_len = descsize;
 741         bufferlist[1].iov_base = buffer;
 742         bufferlist[1].iov_len = bufferlen;
 743         bufferlist[2].iov_base = &aligned_data;
 744         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 745
 746         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
 747
 748         /*
 749          * Signalling the host is conditional on many factors:
 750          * 1. The ring state changed from being empty to non-empty.
 751          *    This is tracked by the variable "signal".
 752          * 2. The variable kick_q tracks if more data will be placed
 753          *    on the ring. We will not signal if more data is
 754          *    to be placed.
 755          *
 756          * Based on the channel signal state, we will decide
 757          * which signaling policy will be applied.
 758          *
 759          * If we cannot write to the ring-buffer; signal the host
 760          * even if we may not have written anything. This is a rare
 761          * enough condition that it should not matter.
 762          */
 763
 764         if (channel->signal_policy)
 765                 signal = true;
 766         else
 767                 kick_q = true;
 768
 769         if (((ret == 0) && kick_q && signal) || (ret))
 770                 vmbus_setevent(channel);
 771
 772         return ret;
 773 }
 774 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
 775
 776 /*
 777  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
 778  * packets using a GPADL Direct packet type.
 779  */
 780 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
 781                                      struct hv_page_buffer pagebuffers[],
 782                                      u32 pagecount, void *buffer, u32 bufferlen,
 783                                      u64 requestid)
 784 {
 785         u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 786         return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
 787                                                buffer, bufferlen, requestid,
 788                                                flags, true);
 789
 790 }
 791 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
 792
 793 /*
 794  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
 795  * using a GPADL Direct packet type.
 796  * The buffer includes the vmbus descriptor.
 797  */
 798 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
 799                               struct vmbus_packet_mpb_array *desc,
 800                               u32 desc_size,
 801                               void *buffer, u32 bufferlen, u64 requestid)
 802 {
 803         int ret;
 804         u32 packetlen;
 805         u32 packetlen_aligned;
 806         struct kvec bufferlist[3];
 807         u64 aligned_data = 0;
 808         bool signal = false;
 809
 810         packetlen = desc_size + bufferlen;
 811         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 812
 813         /* Setup the descriptor */
 814         desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
 815         desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 816         desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
 817         desc->length8 = (u16)(packetlen_aligned >> 3);
 818         desc->transactionid = requestid;
 819         desc->rangecount = 1;
 820
 821         bufferlist[0].iov_base = desc;
 822         bufferlist[0].iov_len = desc_size;
 823         bufferlist[1].iov_base = buffer;
 824         bufferlist[1].iov_len = bufferlen;
 825         bufferlist[2].iov_base = &aligned_data;
 826         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 827
 828         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
 829
 830         if (ret == 0 && signal)
 831                 vmbus_setevent(channel);
 832
 833         return ret;
 834 }
 835 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
 836
 837 /*
 838  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
 839  * using a GPADL Direct packet type.
 840  */
 841 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
 842                                 struct hv_multipage_buffer *multi_pagebuffer,
 843                                 void *buffer, u32 bufferlen, u64 requestid)
 844 {
 845         int ret;
 846         struct vmbus_channel_packet_multipage_buffer desc;
 847         u32 descsize;
 848         u32 packetlen;
 849         u32 packetlen_aligned;
 850         struct kvec bufferlist[3];
 851         u64 aligned_data = 0;
 852         bool signal = false;
 853         u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
 854                                          multi_pagebuffer->len);
 855
 856         if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
 857                 return -EINVAL;
 858
 859         /*
 860          * Adjust the size down since vmbus_channel_packet_multipage_buffer is
 861          * the largest size we support
 862          */
 863         descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
 864                           ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
 865                           sizeof(u64));
 866         packetlen = descsize + bufferlen;
 867         packetlen_aligned = ALIGN(packetlen, sizeof(u64));
 868
 869
 870         /* Setup the descriptor */
 871         desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
 872         desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 873         desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
 874         desc.length8 = (u16)(packetlen_aligned >> 3);
 875         desc.transactionid = requestid;
 876         desc.rangecount = 1;
 877
 878         desc.range.len = multi_pagebuffer->len;
 879         desc.range.offset = multi_pagebuffer->offset;
 880
 881         memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
 882                pfncount * sizeof(u64));
 883
 884         bufferlist[0].iov_base = &desc;
 885         bufferlist[0].iov_len = descsize;
 886         bufferlist[1].iov_base = buffer;
 887         bufferlist[1].iov_len = bufferlen;
 888         bufferlist[2].iov_base = &aligned_data;
 889         bufferlist[2].iov_len = (packetlen_aligned - packetlen);
 890
 891         ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
 892
 893         if (ret == 0 && signal)
 894                 vmbus_setevent(channel);
 895
 896         return ret;
 897 }
 898 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
 899
 900 /**
 901  * vmbus_recvpacket() - Retrieve the user packet on the specified channel
 902  * @channel: Pointer to vmbus_channel structure.
 903  * @buffer: Pointer to the buffer you want to receive the data into.
 904  * @bufferlen: Maximum size of what the the buffer will hold
 905  * @buffer_actual_len: The actual size of the data after it was received
 906  * @requestid: Identifier of the request
 907  *
 908  * Receives directly from the hyper-v vmbus and puts the data it received
 909  * into Buffer. This will receive the data unparsed from hyper-v.
 910  *
 911  * Mainly used by Hyper-V drivers.
 912  */
 913 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
 914                         u32 bufferlen, u32 *buffer_actual_len, u64 *requestid)
 915 {
 916         struct vmpacket_descriptor desc;
 917         u32 packetlen;
 918         u32 userlen;
 919         int ret;
 920         bool signal = false;
 921
 922         *buffer_actual_len = 0;
 923         *requestid = 0;
 924
 925
 926         ret = hv_ringbuffer_peek(&channel->inbound, &desc,
 927                              sizeof(struct vmpacket_descriptor));
 928         if (ret != 0)
 929                 return 0;
 930
 931         packetlen = desc.len8 << 3;
 932         userlen = packetlen - (desc.offset8 << 3);
 933
 934         *buffer_actual_len = userlen;
 935
 936         if (userlen > bufferlen) {
 937
 938                 pr_err("Buffer too small - got %d needs %d\n",
 939                            bufferlen, userlen);
 940                 return -ETOOSMALL;
 941         }
 942
 943         *requestid = desc.trans_id;
 944
 945         /* Copy over the packet to the user buffer */
 946         ret = hv_ringbuffer_read(&channel->inbound, buffer, userlen,
 947                              (desc.offset8 << 3), &signal);
 948
 949         if (signal)
 950                 vmbus_setevent(channel);
 951
 952         return 0;
 953 }
 954 EXPORT_SYMBOL(vmbus_recvpacket);
 955
 956 /*
 957  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
 958  */
 959 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
 960                               u32 bufferlen, u32 *buffer_actual_len,
 961                               u64 *requestid)
 962 {
 963         struct vmpacket_descriptor desc;
 964         u32 packetlen;
 965         int ret;
 966         bool signal = false;
 967
 968         *buffer_actual_len = 0;
 969         *requestid = 0;
 970
 971
 972         ret = hv_ringbuffer_peek(&channel->inbound, &desc,
 973                              sizeof(struct vmpacket_descriptor));
 974         if (ret != 0)
 975                 return 0;
 976
 977
 978         packetlen = desc.len8 << 3;
 979
 980         *buffer_actual_len = packetlen;
 981
 982         if (packetlen > bufferlen)
 983                 return -ENOBUFS;
 984
 985         *requestid = desc.trans_id;
 986
 987         /* Copy over the entire packet to the user buffer */
 988         ret = hv_ringbuffer_read(&channel->inbound, buffer, packetlen, 0,
 989                                  &signal);
 990
 991         if (signal)
 992                 vmbus_setevent(channel);
 993
 994         return ret;
 995 }
 996 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);