linux/drivers/hv/ring_buffer.c
<<
>>
Prefs
   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
  30#include "hyperv_vmbus.h"
  31
  32void hv_begin_read(struct hv_ring_buffer_info *rbi)
  33{
  34        rbi->ring_buffer->interrupt_mask = 1;
  35        mb();
  36}
  37
  38u32 hv_end_read(struct hv_ring_buffer_info *rbi)
  39{
  40        u32 read;
  41        u32 write;
  42
  43        rbi->ring_buffer->interrupt_mask = 0;
  44        mb();
  45
  46        /*
  47         * Now check to see if the ring buffer is still empty.
  48         * If it is not, we raced and we need to process new
  49         * incoming messages.
  50         */
  51        hv_get_ringbuffer_availbytes(rbi, &read, &write);
  52
  53        return read;
  54}
  55
  56/*
  57 * When we write to the ring buffer, check if the host needs to
  58 * be signaled. Here is the details of this protocol:
  59 *
  60 *      1. The host guarantees that while it is draining the
  61 *         ring buffer, it will set the interrupt_mask to
  62 *         indicate it does not need to be interrupted when
  63 *         new data is placed.
  64 *
  65 *      2. The host guarantees that it will completely drain
  66 *         the ring buffer before exiting the read loop. Further,
  67 *         once the ring buffer is empty, it will clear the
  68 *         interrupt_mask and re-check to see if new data has
  69 *         arrived.
  70 */
  71
  72static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi)
  73{
  74        mb();
  75        if (rbi->ring_buffer->interrupt_mask)
  76                return false;
  77
  78        /* check interrupt_mask before read_index */
  79        rmb();
  80        /*
  81         * This is the only case we need to signal when the
  82         * ring transitions from being empty to non-empty.
  83         */
  84        if (old_write == rbi->ring_buffer->read_index)
  85                return true;
  86
  87        return false;
  88}
  89
  90/*
  91 * To optimize the flow management on the send-side,
  92 * when the sender is blocked because of lack of
  93 * sufficient space in the ring buffer, potential the
  94 * consumer of the ring buffer can signal the producer.
  95 * This is controlled by the following parameters:
  96 *
  97 * 1. pending_send_sz: This is the size in bytes that the
  98 *    producer is trying to send.
  99 * 2. The feature bit feat_pending_send_sz set to indicate if
 100 *    the consumer of the ring will signal when the ring
 101 *    state transitions from being full to a state where
 102 *    there is room for the producer to send the pending packet.
 103 */
 104
 105static bool hv_need_to_signal_on_read(u32 old_rd,
 106                                         struct hv_ring_buffer_info *rbi)
 107{
 108        u32 prev_write_sz;
 109        u32 cur_write_sz;
 110        u32 r_size;
 111        u32 write_loc = rbi->ring_buffer->write_index;
 112        u32 read_loc = rbi->ring_buffer->read_index;
 113        u32 pending_sz = rbi->ring_buffer->pending_send_sz;
 114
 115        /*
 116         * If the other end is not blocked on write don't bother.
 117         */
 118        if (pending_sz == 0)
 119                return false;
 120
 121        r_size = rbi->ring_datasize;
 122        cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) :
 123                        read_loc - write_loc;
 124
 125        prev_write_sz = write_loc >= old_rd ? r_size - (write_loc - old_rd) :
 126                        old_rd - write_loc;
 127
 128
 129        if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz))
 130                return true;
 131
 132        return false;
 133}
 134
 135/*
 136 * hv_get_next_write_location()
 137 *
 138 * Get the next write location for the specified ring buffer
 139 *
 140 */
 141static inline u32
 142hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
 143{
 144        u32 next = ring_info->ring_buffer->write_index;
 145
 146        return next;
 147}
 148
 149/*
 150 * hv_set_next_write_location()
 151 *
 152 * Set the next write location for the specified ring buffer
 153 *
 154 */
 155static inline void
 156hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
 157                     u32 next_write_location)
 158{
 159        ring_info->ring_buffer->write_index = next_write_location;
 160}
 161
 162/*
 163 * hv_get_next_read_location()
 164 *
 165 * Get the next read location for the specified ring buffer
 166 */
 167static inline u32
 168hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
 169{
 170        u32 next = ring_info->ring_buffer->read_index;
 171
 172        return next;
 173}
 174
 175/*
 176 * hv_get_next_readlocation_withoffset()
 177 *
 178 * Get the next read location + offset for the specified ring buffer.
 179 * This allows the caller to skip
 180 */
 181static inline u32
 182hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
 183                                 u32 offset)
 184{
 185        u32 next = ring_info->ring_buffer->read_index;
 186
 187        next += offset;
 188        next %= ring_info->ring_datasize;
 189
 190        return next;
 191}
 192
 193/*
 194 *
 195 * hv_set_next_read_location()
 196 *
 197 * Set the next read location for the specified ring buffer
 198 *
 199 */
 200static inline void
 201hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
 202                    u32 next_read_location)
 203{
 204        ring_info->ring_buffer->read_index = next_read_location;
 205}
 206
 207
 208/*
 209 *
 210 * hv_get_ring_buffer()
 211 *
 212 * Get the start of the ring buffer
 213 */
 214static inline void *
 215hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info)
 216{
 217        return (void *)ring_info->ring_buffer->buffer;
 218}
 219
 220
 221/*
 222 *
 223 * hv_get_ring_buffersize()
 224 *
 225 * Get the size of the ring buffer
 226 */
 227static inline u32
 228hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
 229{
 230        return ring_info->ring_datasize;
 231}
 232
 233/*
 234 *
 235 * hv_get_ring_bufferindices()
 236 *
 237 * Get the read and write indices as u64 of the specified ring buffer
 238 *
 239 */
 240static inline u64
 241hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
 242{
 243        return (u64)ring_info->ring_buffer->write_index << 32;
 244}
 245
 246/*
 247 *
 248 * hv_copyfrom_ringbuffer()
 249 *
 250 * Helper routine to copy to source from ring buffer.
 251 * Assume there is enough room. Handles wrap-around in src case only!!
 252 *
 253 */
 254static u32 hv_copyfrom_ringbuffer(
 255        struct hv_ring_buffer_info      *ring_info,
 256        void                            *dest,
 257        u32                             destlen,
 258        u32                             start_read_offset)
 259{
 260        void *ring_buffer = hv_get_ring_buffer(ring_info);
 261        u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 262
 263        u32 frag_len;
 264
 265        /* wrap-around detected at the src */
 266        if (destlen > ring_buffer_size - start_read_offset) {
 267                frag_len = ring_buffer_size - start_read_offset;
 268
 269                memcpy(dest, ring_buffer + start_read_offset, frag_len);
 270                memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
 271        } else
 272
 273                memcpy(dest, ring_buffer + start_read_offset, destlen);
 274
 275
 276        start_read_offset += destlen;
 277        start_read_offset %= ring_buffer_size;
 278
 279        return start_read_offset;
 280}
 281
 282
 283/*
 284 *
 285 * hv_copyto_ringbuffer()
 286 *
 287 * Helper routine to copy from source to ring buffer.
 288 * Assume there is enough room. Handles wrap-around in dest case only!!
 289 *
 290 */
 291static u32 hv_copyto_ringbuffer(
 292        struct hv_ring_buffer_info      *ring_info,
 293        u32                             start_write_offset,
 294        void                            *src,
 295        u32                             srclen)
 296{
 297        void *ring_buffer = hv_get_ring_buffer(ring_info);
 298        u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 299        u32 frag_len;
 300
 301        /* wrap-around detected! */
 302        if (srclen > ring_buffer_size - start_write_offset) {
 303                frag_len = ring_buffer_size - start_write_offset;
 304                memcpy(ring_buffer + start_write_offset, src, frag_len);
 305                memcpy(ring_buffer, src + frag_len, srclen - frag_len);
 306        } else
 307                memcpy(ring_buffer + start_write_offset, src, srclen);
 308
 309        start_write_offset += srclen;
 310        start_write_offset %= ring_buffer_size;
 311
 312        return start_write_offset;
 313}
 314
 315/*
 316 *
 317 * hv_ringbuffer_get_debuginfo()
 318 *
 319 * Get various debug metrics for the specified ring buffer
 320 *
 321 */
 322void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
 323                            struct hv_ring_buffer_debug_info *debug_info)
 324{
 325        u32 bytes_avail_towrite;
 326        u32 bytes_avail_toread;
 327
 328        if (ring_info->ring_buffer) {
 329                hv_get_ringbuffer_availbytes(ring_info,
 330                                        &bytes_avail_toread,
 331                                        &bytes_avail_towrite);
 332
 333                debug_info->bytes_avail_toread = bytes_avail_toread;
 334                debug_info->bytes_avail_towrite = bytes_avail_towrite;
 335                debug_info->current_read_index =
 336                        ring_info->ring_buffer->read_index;
 337                debug_info->current_write_index =
 338                        ring_info->ring_buffer->write_index;
 339                debug_info->current_interrupt_mask =
 340                        ring_info->ring_buffer->interrupt_mask;
 341        }
 342}
 343
 344/*
 345 *
 346 * hv_ringbuffer_init()
 347 *
 348 *Initialize the ring buffer
 349 *
 350 */
 351int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 352                   void *buffer, u32 buflen)
 353{
 354        if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
 355                return -EINVAL;
 356
 357        memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 358
 359        ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
 360        ring_info->ring_buffer->read_index =
 361                ring_info->ring_buffer->write_index = 0;
 362
 363        ring_info->ring_size = buflen;
 364        ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer);
 365
 366        spin_lock_init(&ring_info->ring_lock);
 367
 368        return 0;
 369}
 370
 371/*
 372 *
 373 * hv_ringbuffer_cleanup()
 374 *
 375 * Cleanup the ring buffer
 376 *
 377 */
 378void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 379{
 380}
 381
 382/*
 383 *
 384 * hv_ringbuffer_write()
 385 *
 386 * Write to the ring buffer
 387 *
 388 */
 389int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
 390                    struct scatterlist *sglist, u32 sgcount, bool *signal)
 391{
 392        int i = 0;
 393        u32 bytes_avail_towrite;
 394        u32 bytes_avail_toread;
 395        u32 totalbytes_towrite = 0;
 396
 397        struct scatterlist *sg;
 398        u32 next_write_location;
 399        u32 old_write;
 400        u64 prev_indices = 0;
 401        unsigned long flags;
 402
 403        for_each_sg(sglist, sg, sgcount, i)
 404        {
 405                totalbytes_towrite += sg->length;
 406        }
 407
 408        totalbytes_towrite += sizeof(u64);
 409
 410        spin_lock_irqsave(&outring_info->ring_lock, flags);
 411
 412        hv_get_ringbuffer_availbytes(outring_info,
 413                                &bytes_avail_toread,
 414                                &bytes_avail_towrite);
 415
 416
 417        /* If there is only room for the packet, assume it is full. */
 418        /* Otherwise, the next time around, we think the ring buffer */
 419        /* is empty since the read index == write index */
 420        if (bytes_avail_towrite <= totalbytes_towrite) {
 421                spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 422                return -EAGAIN;
 423        }
 424
 425        /* Write to the ring buffer */
 426        next_write_location = hv_get_next_write_location(outring_info);
 427
 428        old_write = next_write_location;
 429
 430        for_each_sg(sglist, sg, sgcount, i)
 431        {
 432                next_write_location = hv_copyto_ringbuffer(outring_info,
 433                                                     next_write_location,
 434                                                     sg_virt(sg),
 435                                                     sg->length);
 436        }
 437
 438        /* Set previous packet start */
 439        prev_indices = hv_get_ring_bufferindices(outring_info);
 440
 441        next_write_location = hv_copyto_ringbuffer(outring_info,
 442                                             next_write_location,
 443                                             &prev_indices,
 444                                             sizeof(u64));
 445
 446        /* Issue a full memory barrier before updating the write index */
 447        mb();
 448
 449        /* Now, update the write location */
 450        hv_set_next_write_location(outring_info, next_write_location);
 451
 452
 453        spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 454
 455        *signal = hv_need_to_signal(old_write, outring_info);
 456        return 0;
 457}
 458
 459
 460/*
 461 *
 462 * hv_ringbuffer_peek()
 463 *
 464 * Read without advancing the read index
 465 *
 466 */
 467int hv_ringbuffer_peek(struct hv_ring_buffer_info *Inring_info,
 468                   void *Buffer, u32 buflen)
 469{
 470        u32 bytes_avail_towrite;
 471        u32 bytes_avail_toread;
 472        u32 next_read_location = 0;
 473        unsigned long flags;
 474
 475        spin_lock_irqsave(&Inring_info->ring_lock, flags);
 476
 477        hv_get_ringbuffer_availbytes(Inring_info,
 478                                &bytes_avail_toread,
 479                                &bytes_avail_towrite);
 480
 481        /* Make sure there is something to read */
 482        if (bytes_avail_toread < buflen) {
 483
 484                spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
 485
 486                return -EAGAIN;
 487        }
 488
 489        /* Convert to byte offset */
 490        next_read_location = hv_get_next_read_location(Inring_info);
 491
 492        next_read_location = hv_copyfrom_ringbuffer(Inring_info,
 493                                                Buffer,
 494                                                buflen,
 495                                                next_read_location);
 496
 497        spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
 498
 499        return 0;
 500}
 501
 502
 503/*
 504 *
 505 * hv_ringbuffer_read()
 506 *
 507 * Read and advance the read index
 508 *
 509 */
 510int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, void *buffer,
 511                   u32 buflen, u32 offset, bool *signal)
 512{
 513        u32 bytes_avail_towrite;
 514        u32 bytes_avail_toread;
 515        u32 next_read_location = 0;
 516        u64 prev_indices = 0;
 517        unsigned long flags;
 518        u32 old_read;
 519
 520        if (buflen <= 0)
 521                return -EINVAL;
 522
 523        spin_lock_irqsave(&inring_info->ring_lock, flags);
 524
 525        hv_get_ringbuffer_availbytes(inring_info,
 526                                &bytes_avail_toread,
 527                                &bytes_avail_towrite);
 528
 529        old_read = bytes_avail_toread;
 530
 531        /* Make sure there is something to read */
 532        if (bytes_avail_toread < buflen) {
 533                spin_unlock_irqrestore(&inring_info->ring_lock, flags);
 534
 535                return -EAGAIN;
 536        }
 537
 538        next_read_location =
 539                hv_get_next_readlocation_withoffset(inring_info, offset);
 540
 541        next_read_location = hv_copyfrom_ringbuffer(inring_info,
 542                                                buffer,
 543                                                buflen,
 544                                                next_read_location);
 545
 546        next_read_location = hv_copyfrom_ringbuffer(inring_info,
 547                                                &prev_indices,
 548                                                sizeof(u64),
 549                                                next_read_location);
 550
 551        /* Make sure all reads are done before we update the read index since */
 552        /* the writer may start writing to the read area once the read index */
 553        /*is updated */
 554        mb();
 555
 556        /* Update the read index */
 557        hv_set_next_read_location(inring_info, next_read_location);
 558
 559        spin_unlock_irqrestore(&inring_info->ring_lock, flags);
 560
 561        *signal = hv_need_to_signal_on_read(old_read, inring_info);
 562
 563        return 0;
 564}
 565
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.