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 /* Get various debug metrics for the specified ring buffer. */
 144 int hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
 145                                 struct hv_ring_buffer_debug_info *debug_info)
 146 {
 147         u32 bytes_avail_towrite;
 148         u32 bytes_avail_toread;
 149
 150         if (!ring_info->ring_buffer)
 151                 return -EINVAL;
 152
 153         hv_get_ringbuffer_availbytes(ring_info,
 154                                      &bytes_avail_toread,
 155                                      &bytes_avail_towrite);
 156         debug_info->bytes_avail_toread = bytes_avail_toread;
 157         debug_info->bytes_avail_towrite = bytes_avail_towrite;
 158         debug_info->current_read_index = ring_info->ring_buffer->read_index;
 159         debug_info->current_write_index = ring_info->ring_buffer->write_index;
 160         debug_info->current_interrupt_mask
 161                 = ring_info->ring_buffer->interrupt_mask;
 162         return 0;
 163 }
 164 EXPORT_SYMBOL_GPL(hv_ringbuffer_get_debuginfo);
 165
 166 /* Initialize the ring buffer. */
 167 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 168                        struct page *pages, u32 page_cnt)
 169 {
 170         int i;
 171         struct page **pages_wraparound;
 172
 173         BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE));
 174
 175         memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 176
 177         /*
 178          * First page holds struct hv_ring_buffer, do wraparound mapping for
 179          * the rest.
 180          */
 181         pages_wraparound = kzalloc(sizeof(struct page *) * (page_cnt * 2 - 1),
 182                                    GFP_KERNEL);
 183         if (!pages_wraparound)
 184                 return -ENOMEM;
 185
 186         pages_wraparound[0] = pages;
 187         for (i = 0; i < 2 * (page_cnt - 1); i++)
 188                 pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1];
 189
 190         ring_info->ring_buffer = (struct hv_ring_buffer *)
 191                 vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL);
 192
 193         kfree(pages_wraparound);
 194
 195
 196         if (!ring_info->ring_buffer)
 197                 return -ENOMEM;
 198
 199         ring_info->ring_buffer->read_index =
 200                 ring_info->ring_buffer->write_index = 0;
 201
 202         /* Set the feature bit for enabling flow control. */
 203         ring_info->ring_buffer->feature_bits.value = 1;
 204
 205         ring_info->ring_size = page_cnt << PAGE_SHIFT;
 206         ring_info->ring_datasize = ring_info->ring_size -
 207                 sizeof(struct hv_ring_buffer);
 208
 209         spin_lock_init(&ring_info->ring_lock);
 210
 211         return 0;
 212 }
 213
 214 /* Cleanup the ring buffer. */
 215 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 216 {
 217         vunmap(ring_info->ring_buffer);
 218 }
 219
 220 /* Write to the ring buffer. */
 221 int hv_ringbuffer_write(struct vmbus_channel *channel,
 222                         const struct kvec *kv_list, u32 kv_count)
 223 {
 224         int i;
 225         u32 bytes_avail_towrite;
 226         u32 totalbytes_towrite = sizeof(u64);
 227         u32 next_write_location;
 228         u32 old_write;
 229         u64 prev_indices;
 230         unsigned long flags;
 231         struct hv_ring_buffer_info *outring_info = &channel->outbound;
 232
 233         if (channel->rescind)
 234                 return -ENODEV;
 235
 236         for (i = 0; i < kv_count; i++)
 237                 totalbytes_towrite += kv_list[i].iov_len;
 238
 239         spin_lock_irqsave(&outring_info->ring_lock, flags);
 240
 241         bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
 242
 243         /*
 244          * If there is only room for the packet, assume it is full.
 245          * Otherwise, the next time around, we think the ring buffer
 246          * is empty since the read index == write index.
 247          */
 248         if (bytes_avail_towrite <= totalbytes_towrite) {
 249                 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 250                 return -EAGAIN;
 251         }
 252
 253         /* Write to the ring buffer */
 254         next_write_location = hv_get_next_write_location(outring_info);
 255
 256         old_write = next_write_location;
 257
 258         for (i = 0; i < kv_count; i++) {
 259                 next_write_location = hv_copyto_ringbuffer(outring_info,
 260                                                      next_write_location,
 261                                                      kv_list[i].iov_base,
 262                                                      kv_list[i].iov_len);
 263         }
 264
 265         /* Set previous packet start */
 266         prev_indices = hv_get_ring_bufferindices(outring_info);
 267
 268         next_write_location = hv_copyto_ringbuffer(outring_info,
 269                                              next_write_location,
 270                                              &prev_indices,
 271                                              sizeof(u64));
 272
 273         /* Issue a full memory barrier before updating the write index */
 274         virt_mb();
 275
 276         /* Now, update the write location */
 277         hv_set_next_write_location(outring_info, next_write_location);
 278
 279
 280         spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 281
 282         hv_signal_on_write(old_write, channel);
 283
 284         if (channel->rescind)
 285                 return -ENODEV;
 286
 287         return 0;
 288 }
 289
 290 int hv_ringbuffer_read(struct vmbus_channel *channel,
 291                        void *buffer, u32 buflen, u32 *buffer_actual_len,
 292                        u64 *requestid, bool raw)
 293 {
 294         struct vmpacket_descriptor *desc;
 295         u32 packetlen, offset;
 296
 297         if (unlikely(buflen == 0))
 298                 return -EINVAL;
 299
 300         *buffer_actual_len = 0;
 301         *requestid = 0;
 302
 303         /* Make sure there is something to read */
 304         desc = hv_pkt_iter_first(channel);
 305         if (desc == NULL) {
 306                 /*
 307                  * No error is set when there is even no header, drivers are
 308                  * supposed to analyze buffer_actual_len.
 309                  */
 310                 return 0;
 311         }
 312
 313         offset = raw ? 0 : (desc->offset8 << 3);
 314         packetlen = (desc->len8 << 3) - offset;
 315         *buffer_actual_len = packetlen;
 316         *requestid = desc->trans_id;
 317
 318         if (unlikely(packetlen > buflen))
 319                 return -ENOBUFS;
 320
 321         /* since ring is double mapped, only one copy is necessary */
 322         memcpy(buffer, (const char *)desc + offset, packetlen);
 323
 324         /* Advance ring index to next packet descriptor */
 325         __hv_pkt_iter_next(channel, desc);
 326
 327         /* Notify host of update */
 328         hv_pkt_iter_close(channel);
 329
 330         return 0;
 331 }
 332
 333 /*
 334  * Determine number of bytes available in ring buffer after
 335  * the current iterator (priv_read_index) location.
 336  *
 337  * This is similar to hv_get_bytes_to_read but with private
 338  * read index instead.
 339  */
 340 static u32 hv_pkt_iter_avail(const struct hv_ring_buffer_info *rbi)
 341 {
 342         u32 priv_read_loc = rbi->priv_read_index;
 343         u32 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
 344
 345         if (write_loc >= priv_read_loc)
 346                 return write_loc - priv_read_loc;
 347         else
 348                 return (rbi->ring_datasize - priv_read_loc) + write_loc;
 349 }
 350
 351 /*
 352  * Get first vmbus packet from ring buffer after read_index
 353  *
 354  * If ring buffer is empty, returns NULL and no other action needed.
 355  */
 356 struct vmpacket_descriptor *hv_pkt_iter_first(struct vmbus_channel *channel)
 357 {
 358         struct hv_ring_buffer_info *rbi = &channel->inbound;
 359         struct vmpacket_descriptor *desc;
 360
 361         if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor))
 362                 return NULL;
 363
 364         desc = hv_get_ring_buffer(rbi) + rbi->priv_read_index;
 365         if (desc)
 366                 prefetch((char *)desc + (desc->len8 << 3));
 367
 368         return desc;
 369 }
 370 EXPORT_SYMBOL_GPL(hv_pkt_iter_first);
 371
 372 /*
 373  * Get next vmbus packet from ring buffer.
 374  *
 375  * Advances the current location (priv_read_index) and checks for more
 376  * data. If the end of the ring buffer is reached, then return NULL.
 377  */
 378 struct vmpacket_descriptor *
 379 __hv_pkt_iter_next(struct vmbus_channel *channel,
 380                    const struct vmpacket_descriptor *desc)
 381 {
 382         struct hv_ring_buffer_info *rbi = &channel->inbound;
 383         u32 packetlen = desc->len8 << 3;
 384         u32 dsize = rbi->ring_datasize;
 385
 386         /* bump offset to next potential packet */
 387         rbi->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
 388         if (rbi->priv_read_index >= dsize)
 389                 rbi->priv_read_index -= dsize;
 390
 391         /* more data? */
 392         return hv_pkt_iter_first(channel);
 393 }
 394 EXPORT_SYMBOL_GPL(__hv_pkt_iter_next);
 395
 396 /* How many bytes were read in this iterator cycle */
 397 static u32 hv_pkt_iter_bytes_read(const struct hv_ring_buffer_info *rbi,
 398                                         u32 start_read_index)
 399 {
 400         if (rbi->priv_read_index >= start_read_index)
 401                 return rbi->priv_read_index - start_read_index;
 402         else
 403                 return rbi->ring_datasize - start_read_index +
 404                         rbi->priv_read_index;
 405 }
 406
 407 /*
 408  * Update host ring buffer after iterating over packets.
 409  */
 410 void hv_pkt_iter_close(struct vmbus_channel *channel)
 411 {
 412         struct hv_ring_buffer_info *rbi = &channel->inbound;
 413         u32 curr_write_sz, pending_sz, bytes_read, start_read_index;
 414
 415         /*
 416          * Make sure all reads are done before we update the read index since
 417          * the writer may start writing to the read area once the read index
 418          * is updated.
 419          */
 420         virt_rmb();
 421         start_read_index = rbi->ring_buffer->read_index;
 422         rbi->ring_buffer->read_index = rbi->priv_read_index;
 423
 424         if (!rbi->ring_buffer->feature_bits.feat_pending_send_sz)
 425                 return;
 426
 427         /*
 428          * Issue a full memory barrier before making the signaling decision.
 429          * Here is the reason for having this barrier:
 430          * If the reading of the pend_sz (in this function)
 431          * were to be reordered and read before we commit the new read
 432          * index (in the calling function)  we could
 433          * have a problem. If the host were to set the pending_sz after we
 434          * have sampled pending_sz and go to sleep before we commit the
 435          * read index, we could miss sending the interrupt. Issue a full
 436          * memory barrier to address this.
 437          */
 438         virt_mb();
 439
 440         pending_sz = READ_ONCE(rbi->ring_buffer->pending_send_sz);
 441         if (!pending_sz)
 442                 return;
 443
 444         /*
 445          * Ensure the read of write_index in hv_get_bytes_to_write()
 446          * happens after the read of pending_send_sz.
 447          */
 448         virt_rmb();
 449         curr_write_sz = hv_get_bytes_to_write(rbi);
 450         bytes_read = hv_pkt_iter_bytes_read(rbi, start_read_index);
 451
 452         /*
 453          * If there was space before we began iteration,
 454          * then host was not blocked.
 455          */
 456
 457         if (curr_write_sz - bytes_read > pending_sz)
 458                 return;
 459
 460         /* If pending write will not fit, don't give false hope. */
 461         if (curr_write_sz <= pending_sz)
 462                 return;
 463
 464         vmbus_setevent(channel);
 465 }
 466 EXPORT_SYMBOL_GPL(hv_pkt_iter_close);