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
  31 #include "hyperv_vmbus.h"
  32
  33 void hv_begin_read(struct hv_ring_buffer_info *rbi)
  34 {
  35         rbi->ring_buffer->interrupt_mask = 1;
  36         virt_mb();
  37 }
  38
  39 u32 hv_end_read(struct hv_ring_buffer_info *rbi)
  40 {
  41
  42         rbi->ring_buffer->interrupt_mask = 0;
  43         virt_mb();
  44
  45         /*
  46          * Now check to see if the ring buffer is still empty.
  47          * If it is not, we raced and we need to process new
  48          * incoming messages.
  49          */
  50         return hv_get_bytes_to_read(rbi);
  51 }
  52
  53 /*
  54  * When we write to the ring buffer, check if the host needs to
  55  * be signaled. Here is the details of this protocol:
  56  *
  57  *      1. The host guarantees that while it is draining the
  58  *         ring buffer, it will set the interrupt_mask to
  59  *         indicate it does not need to be interrupted when
  60  *         new data is placed.
  61  *
  62  *      2. The host guarantees that it will completely drain
  63  *         the ring buffer before exiting the read loop. Further,
  64  *         once the ring buffer is empty, it will clear the
  65  *         interrupt_mask and re-check to see if new data has
  66  *         arrived.
  67  */
  68
  69 static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi)
  70 {
  71         virt_mb();
  72         if (READ_ONCE(rbi->ring_buffer->interrupt_mask))
  73                 return false;
  74
  75         /* check interrupt_mask before read_index */
  76         virt_rmb();
  77         /*
  78          * This is the only case we need to signal when the
  79          * ring transitions from being empty to non-empty.
  80          */
  81         if (old_write == READ_ONCE(rbi->ring_buffer->read_index))
  82                 return true;
  83
  84         return false;
  85 }
  86
  87 /* Get the next write location for the specified ring buffer. */
  88 static inline u32
  89 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
  90 {
  91         u32 next = ring_info->ring_buffer->write_index;
  92
  93         return next;
  94 }
  95
  96 /* Set the next write location for the specified ring buffer. */
  97 static inline void
  98 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
  99                      u32 next_write_location)
 100 {
 101         ring_info->ring_buffer->write_index = next_write_location;
 102 }
 103
 104 /* Get the next read location for the specified ring buffer. */
 105 static inline u32
 106 hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
 107 {
 108         u32 next = ring_info->ring_buffer->read_index;
 109
 110         return next;
 111 }
 112
 113 /*
 114  * Get the next read location + offset for the specified ring buffer.
 115  * This allows the caller to skip.
 116  */
 117 static inline u32
 118 hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
 119                                  u32 offset)
 120 {
 121         u32 next = ring_info->ring_buffer->read_index;
 122
 123         next += offset;
 124         next %= ring_info->ring_datasize;
 125
 126         return next;
 127 }
 128
 129 /* Set the next read location for the specified ring buffer. */
 130 static inline void
 131 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
 132                     u32 next_read_location)
 133 {
 134         ring_info->ring_buffer->read_index = next_read_location;
 135 }
 136
 137 /* Get the size of the ring buffer. */
 138 static inline u32
 139 hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
 140 {
 141         return ring_info->ring_datasize;
 142 }
 143
 144 /* Get the read and write indices as u64 of the specified ring buffer. */
 145 static inline u64
 146 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
 147 {
 148         return (u64)ring_info->ring_buffer->write_index << 32;
 149 }
 150
 151 /*
 152  * Helper routine to copy to source from ring buffer.
 153  * Assume there is enough room. Handles wrap-around in src case only!!
 154  */
 155 static u32 hv_copyfrom_ringbuffer(
 156         struct hv_ring_buffer_info      *ring_info,
 157         void                            *dest,
 158         u32                             destlen,
 159         u32                             start_read_offset)
 160 {
 161         void *ring_buffer = hv_get_ring_buffer(ring_info);
 162         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 163
 164         u32 frag_len;
 165
 166         /* wrap-around detected at the src */
 167         if (destlen > ring_buffer_size - start_read_offset) {
 168                 frag_len = ring_buffer_size - start_read_offset;
 169
 170                 memcpy(dest, ring_buffer + start_read_offset, frag_len);
 171                 memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
 172         } else
 173
 174                 memcpy(dest, ring_buffer + start_read_offset, destlen);
 175
 176
 177         start_read_offset += destlen;
 178         start_read_offset %= ring_buffer_size;
 179
 180         return start_read_offset;
 181 }
 182
 183
 184 /*
 185  * Helper routine to copy from source to ring buffer.
 186  * Assume there is enough room. Handles wrap-around in dest case only!!
 187  */
 188 static u32 hv_copyto_ringbuffer(
 189         struct hv_ring_buffer_info      *ring_info,
 190         u32                             start_write_offset,
 191         void                            *src,
 192         u32                             srclen)
 193 {
 194         void *ring_buffer = hv_get_ring_buffer(ring_info);
 195         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 196         u32 frag_len;
 197
 198         /* wrap-around detected! */
 199         if (srclen > ring_buffer_size - start_write_offset) {
 200                 frag_len = ring_buffer_size - start_write_offset;
 201                 memcpy(ring_buffer + start_write_offset, src, frag_len);
 202                 memcpy(ring_buffer, src + frag_len, srclen - frag_len);
 203         } else
 204                 memcpy(ring_buffer + start_write_offset, src, srclen);
 205
 206         start_write_offset += srclen;
 207         start_write_offset %= ring_buffer_size;
 208
 209         return start_write_offset;
 210 }
 211
 212 /* Get various debug metrics for the specified ring buffer. */
 213 void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
 214                             struct hv_ring_buffer_debug_info *debug_info)
 215 {
 216         u32 bytes_avail_towrite;
 217         u32 bytes_avail_toread;
 218
 219         if (ring_info->ring_buffer) {
 220                 hv_get_ringbuffer_availbytes(ring_info,
 221                                         &bytes_avail_toread,
 222                                         &bytes_avail_towrite);
 223
 224                 debug_info->bytes_avail_toread = bytes_avail_toread;
 225                 debug_info->bytes_avail_towrite = bytes_avail_towrite;
 226                 debug_info->current_read_index =
 227                         ring_info->ring_buffer->read_index;
 228                 debug_info->current_write_index =
 229                         ring_info->ring_buffer->write_index;
 230                 debug_info->current_interrupt_mask =
 231                         ring_info->ring_buffer->interrupt_mask;
 232         }
 233 }
 234
 235 /* Initialize the ring buffer. */
 236 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 237                    void *buffer, u32 buflen)
 238 {
 239         if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
 240                 return -EINVAL;
 241
 242         memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 243
 244         ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
 245         ring_info->ring_buffer->read_index =
 246                 ring_info->ring_buffer->write_index = 0;
 247
 248         /* Set the feature bit for enabling flow control. */
 249         ring_info->ring_buffer->feature_bits.value = 1;
 250
 251         ring_info->ring_size = buflen;
 252         ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer);
 253
 254         spin_lock_init(&ring_info->ring_lock);
 255
 256         return 0;
 257 }
 258
 259 /* Cleanup the ring buffer. */
 260 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 261 {
 262 }
 263
 264 /* Write to the ring buffer. */
 265 int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
 266                     struct kvec *kv_list, u32 kv_count, bool *signal, bool lock)
 267 {
 268         int i = 0;
 269         u32 bytes_avail_towrite;
 270         u32 totalbytes_towrite = 0;
 271
 272         u32 next_write_location;
 273         u32 old_write;
 274         u64 prev_indices = 0;
 275         unsigned long flags = 0;
 276
 277         for (i = 0; i < kv_count; i++)
 278                 totalbytes_towrite += kv_list[i].iov_len;
 279
 280         totalbytes_towrite += sizeof(u64);
 281
 282         if (lock)
 283                 spin_lock_irqsave(&outring_info->ring_lock, flags);
 284
 285         bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
 286
 287         /*
 288          * If there is only room for the packet, assume it is full.
 289          * Otherwise, the next time around, we think the ring buffer
 290          * is empty since the read index == write index.
 291          */
 292         if (bytes_avail_towrite <= totalbytes_towrite) {
 293                 if (lock)
 294                         spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 295                 return -EAGAIN;
 296         }
 297
 298         /* Write to the ring buffer */
 299         next_write_location = hv_get_next_write_location(outring_info);
 300
 301         old_write = next_write_location;
 302
 303         for (i = 0; i < kv_count; i++) {
 304                 next_write_location = hv_copyto_ringbuffer(outring_info,
 305                                                      next_write_location,
 306                                                      kv_list[i].iov_base,
 307                                                      kv_list[i].iov_len);
 308         }
 309
 310         /* Set previous packet start */
 311         prev_indices = hv_get_ring_bufferindices(outring_info);
 312
 313         next_write_location = hv_copyto_ringbuffer(outring_info,
 314                                              next_write_location,
 315                                              &prev_indices,
 316                                              sizeof(u64));
 317
 318         /* Issue a full memory barrier before updating the write index */
 319         virt_mb();
 320
 321         /* Now, update the write location */
 322         hv_set_next_write_location(outring_info, next_write_location);
 323
 324
 325         if (lock)
 326                 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 327
 328         *signal = hv_need_to_signal(old_write, outring_info);
 329         return 0;
 330 }
 331
 332 int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info,
 333                        void *buffer, u32 buflen, u32 *buffer_actual_len,
 334                        u64 *requestid, bool *signal, bool raw)
 335 {
 336         u32 bytes_avail_toread;
 337         u32 next_read_location = 0;
 338         u64 prev_indices = 0;
 339         struct vmpacket_descriptor desc;
 340         u32 offset;
 341         u32 packetlen;
 342         int ret = 0;
 343
 344         if (buflen <= 0)
 345                 return -EINVAL;
 346
 347
 348         *buffer_actual_len = 0;
 349         *requestid = 0;
 350
 351         bytes_avail_toread = hv_get_bytes_to_read(inring_info);
 352         /* Make sure there is something to read */
 353         if (bytes_avail_toread < sizeof(desc)) {
 354                 /*
 355                  * No error is set when there is even no header, drivers are
 356                  * supposed to analyze buffer_actual_len.
 357                  */
 358                 return ret;
 359         }
 360
 361         next_read_location = hv_get_next_read_location(inring_info);
 362         next_read_location = hv_copyfrom_ringbuffer(inring_info, &desc,
 363                                                     sizeof(desc),
 364                                                     next_read_location);
 365
 366         offset = raw ? 0 : (desc.offset8 << 3);
 367         packetlen = (desc.len8 << 3) - offset;
 368         *buffer_actual_len = packetlen;
 369         *requestid = desc.trans_id;
 370
 371         if (bytes_avail_toread < packetlen + offset)
 372                 return -EAGAIN;
 373
 374         if (packetlen > buflen)
 375                 return -ENOBUFS;
 376
 377         next_read_location =
 378                 hv_get_next_readlocation_withoffset(inring_info, offset);
 379
 380         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 381                                                 buffer,
 382                                                 packetlen,
 383                                                 next_read_location);
 384
 385         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 386                                                 &prev_indices,
 387                                                 sizeof(u64),
 388                                                 next_read_location);
 389
 390         /*
 391          * Make sure all reads are done before we update the read index since
 392          * the writer may start writing to the read area once the read index
 393          * is updated.
 394          */
 395         virt_mb();
 396
 397         /* Update the read index */
 398         hv_set_next_read_location(inring_info, next_read_location);
 399
 400         *signal = hv_need_to_signal_on_read(inring_info);
 401
 402         return ret;
 403 }