drivers/hv/ring_buffer.c

   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  *   K. Y. Srinivasan <kys@microsoft.com>
  22  *
  23  */
  24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/kernel.h>
  27 #include <linux/mm.h>
  28 #include <linux/hyperv.h>
  29 #include <linux/uio.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/slab.h>
  32 #include <linux/prefetch.h>
  33
  34 #include "hyperv_vmbus.h"
  35
  36 #define VMBUS_PKT_TRAILER       8
  37
  38 /*
  39  * When we write to the ring buffer, check if the host needs to
  40  * be signaled. Here is the details of this protocol:
  41  *
  42  *      1. The host guarantees that while it is draining the
  43  *         ring buffer, it will set the interrupt_mask to
  44  *         indicate it does not need to be interrupted when
  45  *         new data is placed.
  46  *
  47  *      2. The host guarantees that it will completely drain
  48  *         the ring buffer before exiting the read loop. Further,
  49  *         once the ring buffer is empty, it will clear the
  50  *         interrupt_mask and re-check to see if new data has
  51  *         arrived.
  52  *
  53  * KYS: Oct. 30, 2016:
  54  * It looks like Windows hosts have logic to deal with DOS attacks that
  55  * can be triggered if it receives interrupts when it is not expecting
  56  * the interrupt. The host expects interrupts only when the ring
  57  * transitions from empty to non-empty (or full to non full on the guest
  58  * to host ring).
  59  * So, base the signaling decision solely on the ring state until the
  60  * host logic is fixed.
  61  */
  62
  63 static void hv_signal_on_write(u32 old_write, struct vmbus_channel *channel)
  64 {
  65         struct hv_ring_buffer_info *rbi = &channel->outbound;
  66
  67         virt_mb();
  68         if (READ_ONCE(rbi->ring_buffer->interrupt_mask))
  69                 return;
  70
  71         /* check interrupt_mask before read_index */
  72         virt_rmb();
  73         /*
  74          * This is the only case we need to signal when the
  75          * ring transitions from being empty to non-empty.
  76          */
  77         if (old_write == READ_ONCE(rbi->ring_buffer->read_index))
  78                 vmbus_setevent(channel);
  79 }
  80
  81 /* Get the next write location for the specified ring buffer. */
  82 static inline u32
  83 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
  84 {
  85         u32 next = ring_info->ring_buffer->write_index;
  86
  87         return next;
  88 }
  89
  90 /* Set the next write location for the specified ring buffer. */
  91 static inline void
  92 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
  93                      u32 next_write_location)
  94 {
  95         ring_info->ring_buffer->write_index = next_write_location;
  96 }
  97
  98 /* Set the next read location for the specified ring buffer. */
  99 static inline void
 100 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
 101                     u32 next_read_location)
 102 {
 103         ring_info->ring_buffer->read_index = next_read_location;
 104         ring_info->priv_read_index = next_read_location;
 105 }
 106
 107 /* Get the size of the ring buffer. */
 108 static inline u32
 109 hv_get_ring_buffersize(const struct hv_ring_buffer_info *ring_info)
 110 {
 111         return ring_info->ring_datasize;
 112 }
 113
 114 /* Get the read and write indices as u64 of the specified ring buffer. */
 115 static inline u64
 116 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
 117 {
 118         return (u64)ring_info->ring_buffer->write_index << 32;
 119 }
 120
 121 /*
 122  * Helper routine to copy from source to ring buffer.
 123  * Assume there is enough room. Handles wrap-around in dest case only!!
 124  */
 125 static u32 hv_copyto_ringbuffer(
 126         struct hv_ring_buffer_info      *ring_info,
 127         u32                             start_write_offset,
 128         const void                      *src,
 129         u32                             srclen)
 130 {
 131         void *ring_buffer = hv_get_ring_buffer(ring_info);
 132         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 133
 134         memcpy(ring_buffer + start_write_offset, src, srclen);
 135
 136         start_write_offset += srclen;
 137         if (start_write_offset >= ring_buffer_size)
 138                 start_write_offset -= ring_buffer_size;
 139
 140         return start_write_offset;
 141 }
 142
 143 /*
 144  *
 145  * hv_get_ringbuffer_availbytes()
 146  *
 147  * Get number of bytes available to read and to write to
 148  * for the specified ring buffer
 149  */
 150 static void
 151 hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info *rbi,
 152                              u32 *read, u32 *write)
 153 {
 154         u32 read_loc, write_loc, dsize;
 155
 156         /* Capture the read/write indices before they changed */
 157         read_loc = READ_ONCE(rbi->ring_buffer->read_index);
 158         write_loc = READ_ONCE(rbi->ring_buffer->write_index);
 159         dsize = rbi->ring_datasize;
 160
 161         *write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
 162                 read_loc - write_loc;
 163         *read = dsize - *write;
 164 }
 165
 166 /* Get various debug metrics for the specified ring buffer. */
 167 void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
 168                                  struct hv_ring_buffer_debug_info *debug_info)
 169 {
 170         u32 bytes_avail_towrite;
 171         u32 bytes_avail_toread;
 172
 173         if (ring_info->ring_buffer) {
 174                 hv_get_ringbuffer_availbytes(ring_info,
 175                                         &bytes_avail_toread,
 176                                         &bytes_avail_towrite);
 177
 178                 debug_info->bytes_avail_toread = bytes_avail_toread;
 179                 debug_info->bytes_avail_towrite = bytes_avail_towrite;
 180                 debug_info->current_read_index =
 181                         ring_info->ring_buffer->read_index;
 182                 debug_info->current_write_index =
 183                         ring_info->ring_buffer->write_index;
 184                 debug_info->current_interrupt_mask =
 185                         ring_info->ring_buffer->interrupt_mask;
 186         }
 187 }
 188 EXPORT_SYMBOL_GPL(hv_ringbuffer_get_debuginfo);
 189
 190 /* Initialize the ring buffer. */
 191 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 192                        struct page *pages, u32 page_cnt)
 193 {
 194         int i;
 195         struct page **pages_wraparound;
 196
 197         BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE));
 198
 199         memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 200
 201         /*
 202          * First page holds struct hv_ring_buffer, do wraparound mapping for
 203          * the rest.
 204          */
 205         pages_wraparound = kcalloc(page_cnt * 2 - 1, sizeof(struct page *),
 206                                    GFP_KERNEL);
 207         if (!pages_wraparound)
 208                 return -ENOMEM;
 209
 210         pages_wraparound[0] = pages;
 211         for (i = 0; i < 2 * (page_cnt - 1); i++)
 212                 pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1];
 213
 214         ring_info->ring_buffer = (struct hv_ring_buffer *)
 215                 vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL);
 216
 217         kfree(pages_wraparound);
 218
 219
 220         if (!ring_info->ring_buffer)
 221                 return -ENOMEM;
 222
 223         ring_info->ring_buffer->read_index =
 224                 ring_info->ring_buffer->write_index = 0;
 225
 226         /* Set the feature bit for enabling flow control. */
 227         ring_info->ring_buffer->feature_bits.value = 1;
 228
 229         ring_info->ring_size = page_cnt << PAGE_SHIFT;
 230         ring_info->ring_size_div10_reciprocal =
 231                 reciprocal_value(ring_info->ring_size / 10);
 232         ring_info->ring_datasize = ring_info->ring_size -
 233                 sizeof(struct hv_ring_buffer);
 234
 235         spin_lock_init(&ring_info->ring_lock);
 236
 237         return 0;
 238 }
 239
 240 /* Cleanup the ring buffer. */
 241 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 242 {
 243         vunmap(ring_info->ring_buffer);
 244         ring_info->ring_buffer = NULL;
 245 }
 246
 247 /* Write to the ring buffer. */
 248 int hv_ringbuffer_write(struct vmbus_channel *channel,
 249                         const struct kvec *kv_list, u32 kv_count)
 250 {
 251         int i;
 252         u32 bytes_avail_towrite;
 253         u32 totalbytes_towrite = sizeof(u64);
 254         u32 next_write_location;
 255         u32 old_write;
 256         u64 prev_indices;
 257         unsigned long flags;
 258         struct hv_ring_buffer_info *outring_info = &channel->outbound;
 259
 260         if (channel->rescind)
 261                 return -ENODEV;
 262
 263         for (i = 0; i < kv_count; i++)
 264                 totalbytes_towrite += kv_list[i].iov_len;
 265
 266         spin_lock_irqsave(&outring_info->ring_lock, flags);
 267
 268         bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
 269
 270         /*
 271          * If there is only room for the packet, assume it is full.
 272          * Otherwise, the next time around, we think the ring buffer
 273          * is empty since the read index == write index.
 274          */
 275         if (bytes_avail_towrite <= totalbytes_towrite) {
 276                 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 277                 return -EAGAIN;
 278         }
 279
 280         /* Write to the ring buffer */
 281         next_write_location = hv_get_next_write_location(outring_info);
 282
 283         old_write = next_write_location;
 284
 285         for (i = 0; i < kv_count; i++) {
 286                 next_write_location = hv_copyto_ringbuffer(outring_info,
 287                                                      next_write_location,
 288                                                      kv_list[i].iov_base,
 289                                                      kv_list[i].iov_len);
 290         }
 291
 292         /* Set previous packet start */
 293         prev_indices = hv_get_ring_bufferindices(outring_info);
 294
 295         next_write_location = hv_copyto_ringbuffer(outring_info,
 296                                              next_write_location,
 297                                              &prev_indices,
 298                                              sizeof(u64));
 299
 300         /* Issue a full memory barrier before updating the write index */
 301         virt_mb();
 302
 303         /* Now, update the write location */
 304         hv_set_next_write_location(outring_info, next_write_location);
 305
 306
 307         spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 308
 309         hv_signal_on_write(old_write, channel);
 310
 311         if (channel->rescind)
 312                 return -ENODEV;
 313
 314         return 0;
 315 }
 316
 317 int hv_ringbuffer_read(struct vmbus_channel *channel,
 318                        void *buffer, u32 buflen, u32 *buffer_actual_len,
 319                        u64 *requestid, bool raw)
 320 {
 321         struct vmpacket_descriptor *desc;
 322         u32 packetlen, offset;
 323
 324         if (unlikely(buflen == 0))
 325                 return -EINVAL;
 326
 327         *buffer_actual_len = 0;
 328         *requestid = 0;
 329
 330         /* Make sure there is something to read */
 331         desc = hv_pkt_iter_first(channel);
 332         if (desc == NULL) {
 333                 /*
 334                  * No error is set when there is even no header, drivers are
 335                  * supposed to analyze buffer_actual_len.
 336                  */
 337                 return 0;
 338         }
 339
 340         offset = raw ? 0 : (desc->offset8 << 3);
 341         packetlen = (desc->len8 << 3) - offset;
 342         *buffer_actual_len = packetlen;
 343         *requestid = desc->trans_id;
 344
 345         if (unlikely(packetlen > buflen))
 346                 return -ENOBUFS;
 347
 348         /* since ring is double mapped, only one copy is necessary */
 349         memcpy(buffer, (const char *)desc + offset, packetlen);
 350
 351         /* Advance ring index to next packet descriptor */
 352         __hv_pkt_iter_next(channel, desc);
 353
 354         /* Notify host of update */
 355         hv_pkt_iter_close(channel);
 356
 357         return 0;
 358 }
 359
 360 /*
 361  * Determine number of bytes available in ring buffer after
 362  * the current iterator (priv_read_index) location.
 363  *
 364  * This is similar to hv_get_bytes_to_read but with private
 365  * read index instead.
 366  */
 367 static u32 hv_pkt_iter_avail(const struct hv_ring_buffer_info *rbi)
 368 {
 369         u32 priv_read_loc = rbi->priv_read_index;
 370         u32 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
 371
 372         if (write_loc >= priv_read_loc)
 373                 return write_loc - priv_read_loc;
 374         else
 375                 return (rbi->ring_datasize - priv_read_loc) + write_loc;
 376 }
 377
 378 /*
 379  * Get first vmbus packet from ring buffer after read_index
 380  *
 381  * If ring buffer is empty, returns NULL and no other action needed.
 382  */
 383 struct vmpacket_descriptor *hv_pkt_iter_first(struct vmbus_channel *channel)
 384 {
 385         struct hv_ring_buffer_info *rbi = &channel->inbound;
 386         struct vmpacket_descriptor *desc;
 387
 388         if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor))
 389                 return NULL;
 390
 391         desc = hv_get_ring_buffer(rbi) + rbi->priv_read_index;
 392         if (desc)
 393                 prefetch((char *)desc + (desc->len8 << 3));
 394
 395         return desc;
 396 }
 397 EXPORT_SYMBOL_GPL(hv_pkt_iter_first);
 398
 399 /*
 400  * Get next vmbus packet from ring buffer.
 401  *
 402  * Advances the current location (priv_read_index) and checks for more
 403  * data. If the end of the ring buffer is reached, then return NULL.
 404  */
 405 struct vmpacket_descriptor *
 406 __hv_pkt_iter_next(struct vmbus_channel *channel,
 407                    const struct vmpacket_descriptor *desc)
 408 {
 409         struct hv_ring_buffer_info *rbi = &channel->inbound;
 410         u32 packetlen = desc->len8 << 3;
 411         u32 dsize = rbi->ring_datasize;
 412
 413         /* bump offset to next potential packet */
 414         rbi->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
 415         if (rbi->priv_read_index >= dsize)
 416                 rbi->priv_read_index -= dsize;
 417
 418         /* more data? */
 419         return hv_pkt_iter_first(channel);
 420 }
 421 EXPORT_SYMBOL_GPL(__hv_pkt_iter_next);
 422
 423 /* How many bytes were read in this iterator cycle */
 424 static u32 hv_pkt_iter_bytes_read(const struct hv_ring_buffer_info *rbi,
 425                                         u32 start_read_index)
 426 {
 427         if (rbi->priv_read_index >= start_read_index)
 428                 return rbi->priv_read_index - start_read_index;
 429         else
 430                 return rbi->ring_datasize - start_read_index +
 431                         rbi->priv_read_index;
 432 }
 433
 434 /*
 435  * Update host ring buffer after iterating over packets. If the host has
 436  * stopped queuing new entries because it found the ring buffer full, and
 437  * sufficient space is being freed up, signal the host. But be careful to
 438  * only signal the host when necessary, both for performance reasons and
 439  * because Hyper-V protects itself by throttling guests that signal
 440  * inappropriately.
 441  *
 442  * Determining when to signal is tricky. There are three key data inputs
 443  * that must be handled in this order to avoid race conditions:
 444  *
 445  * 1. Update the read_index
 446  * 2. Read the pending_send_sz
 447  * 3. Read the current write_index
 448  *
 449  * The interrupt_mask is not used to determine when to signal. The
 450  * interrupt_mask is used only on the guest->host ring buffer when
 451  * sending requests to the host. The host does not use it on the host->
 452  * guest ring buffer to indicate whether it should be signaled.
 453  */
 454 void hv_pkt_iter_close(struct vmbus_channel *channel)
 455 {
 456         struct hv_ring_buffer_info *rbi = &channel->inbound;
 457         u32 curr_write_sz, pending_sz, bytes_read, start_read_index;
 458
 459         /*
 460          * Make sure all reads are done before we update the read index since
 461          * the writer may start writing to the read area once the read index
 462          * is updated.
 463          */
 464         virt_rmb();
 465         start_read_index = rbi->ring_buffer->read_index;
 466         rbi->ring_buffer->read_index = rbi->priv_read_index;
 467
 468         /*
 469          * Older versions of Hyper-V (before WS2102 and Win8) do not
 470          * implement pending_send_sz and simply poll if the host->guest
 471          * ring buffer is full.  No signaling is needed or expected.
 472          */
 473         if (!rbi->ring_buffer->feature_bits.feat_pending_send_sz)
 474                 return;
 475
 476         /*
 477          * Issue a full memory barrier before making the signaling decision.
 478          * If reading pending_send_sz were to be reordered and happen
 479          * before we commit the new read_index, a race could occur.  If the
 480          * host were to set the pending_send_sz after we have sampled
 481          * pending_send_sz, and the ring buffer blocks before we commit the
 482          * read index, we could miss sending the interrupt. Issue a full
 483          * memory barrier to address this.
 484          */
 485         virt_mb();
 486
 487         /*
 488          * If the pending_send_sz is zero, then the ring buffer is not
 489          * blocked and there is no need to signal.  This is far by the
 490          * most common case, so exit quickly for best performance.
 491          */
 492         pending_sz = READ_ONCE(rbi->ring_buffer->pending_send_sz);
 493         if (!pending_sz)
 494                 return;
 495
 496         /*
 497          * Ensure the read of write_index in hv_get_bytes_to_write()
 498          * happens after the read of pending_send_sz.
 499          */
 500         virt_rmb();
 501         curr_write_sz = hv_get_bytes_to_write(rbi);
 502         bytes_read = hv_pkt_iter_bytes_read(rbi, start_read_index);
 503
 504         /*
 505          * We want to signal the host only if we're transitioning
 506          * from a "not enough free space" state to a "enough free
 507          * space" state.  For example, it's possible that this function
 508          * could run and free up enough space to signal the host, and then
 509          * run again and free up additional space before the host has a
 510          * chance to clear the pending_send_sz.  The 2nd invocation would
 511          * be a null transition from "enough free space" to "enough free
 512          * space", which doesn't warrant a signal.
 513          *
 514          * Exactly filling the ring buffer is treated as "not enough
 515          * space". The ring buffer always must have at least one byte
 516          * empty so the empty and full conditions are distinguishable.
 517          * hv_get_bytes_to_write() doesn't fully tell the truth in
 518          * this regard.
 519          *
 520          * So first check if we were in the "enough free space" state
 521          * before we began the iteration. If so, the host was not
 522          * blocked, and there's no need to signal.
 523          */
 524         if (curr_write_sz - bytes_read > pending_sz)
 525                 return;
 526
 527         /*
 528          * Similarly, if the new state is "not enough space", then
 529          * there's no need to signal.
 530          */
 531         if (curr_write_sz <= pending_sz)
 532                 return;
 533
 534         vmbus_setevent(channel);
 535 }
 536 EXPORT_SYMBOL_GPL(hv_pkt_iter_close);