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
  33 #include "hyperv_vmbus.h"
  34
  35 void hv_begin_read(struct hv_ring_buffer_info *rbi)
  36 {
  37         rbi->ring_buffer->interrupt_mask = 1;
  38         virt_mb();
  39 }
  40
  41 u32 hv_end_read(struct hv_ring_buffer_info *rbi)
  42 {
  43
  44         rbi->ring_buffer->interrupt_mask = 0;
  45         virt_mb();
  46
  47         /*
  48          * Now check to see if the ring buffer is still empty.
  49          * If it is not, we raced and we need to process new
  50          * incoming messages.
  51          */
  52         return hv_get_bytes_to_read(rbi);
  53 }
  54
  55 /*
  56  * When we write to the ring buffer, check if the host needs to
  57  * be signaled. Here is the details of this protocol:
  58  *
  59  *      1. The host guarantees that while it is draining the
  60  *         ring buffer, it will set the interrupt_mask to
  61  *         indicate it does not need to be interrupted when
  62  *         new data is placed.
  63  *
  64  *      2. The host guarantees that it will completely drain
  65  *         the ring buffer before exiting the read loop. Further,
  66  *         once the ring buffer is empty, it will clear the
  67  *         interrupt_mask and re-check to see if new data has
  68  *         arrived.
  69  */
  70
  71 static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi,
  72                               enum hv_signal_policy policy)
  73 {
  74         virt_mb();
  75         if (READ_ONCE(rbi->ring_buffer->interrupt_mask))
  76                 return false;
  77
  78         /*
  79          * When the client wants to control signaling,
  80          * we only honour the host interrupt mask.
  81          */
  82         if (policy == HV_SIGNAL_POLICY_EXPLICIT)
  83                 return true;
  84
  85         /* check interrupt_mask before read_index */
  86         virt_rmb();
  87         /*
  88          * This is the only case we need to signal when the
  89          * ring transitions from being empty to non-empty.
  90          */
  91         if (old_write == READ_ONCE(rbi->ring_buffer->read_index))
  92                 return true;
  93
  94         return false;
  95 }
  96
  97 /* Get the next write location for the specified ring buffer. */
  98 static inline u32
  99 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
 100 {
 101         u32 next = ring_info->ring_buffer->write_index;
 102
 103         return next;
 104 }
 105
 106 /* Set the next write location for the specified ring buffer. */
 107 static inline void
 108 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
 109                      u32 next_write_location)
 110 {
 111         ring_info->ring_buffer->write_index = next_write_location;
 112 }
 113
 114 /* Get the next read location for the specified ring buffer. */
 115 static inline u32
 116 hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
 117 {
 118         u32 next = ring_info->ring_buffer->read_index;
 119
 120         return next;
 121 }
 122
 123 /*
 124  * Get the next read location + offset for the specified ring buffer.
 125  * This allows the caller to skip.
 126  */
 127 static inline u32
 128 hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
 129                                  u32 offset)
 130 {
 131         u32 next = ring_info->ring_buffer->read_index;
 132
 133         next += offset;
 134         next %= ring_info->ring_datasize;
 135
 136         return next;
 137 }
 138
 139 /* Set the next read location for the specified ring buffer. */
 140 static inline void
 141 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
 142                     u32 next_read_location)
 143 {
 144         ring_info->ring_buffer->read_index = next_read_location;
 145         ring_info->priv_read_index = next_read_location;
 146 }
 147
 148 /* Get the size of the ring buffer. */
 149 static inline u32
 150 hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
 151 {
 152         return ring_info->ring_datasize;
 153 }
 154
 155 /* Get the read and write indices as u64 of the specified ring buffer. */
 156 static inline u64
 157 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
 158 {
 159         return (u64)ring_info->ring_buffer->write_index << 32;
 160 }
 161
 162 /*
 163  * Helper routine to copy to source from ring buffer.
 164  * Assume there is enough room. Handles wrap-around in src case only!!
 165  */
 166 static u32 hv_copyfrom_ringbuffer(
 167         struct hv_ring_buffer_info      *ring_info,
 168         void                            *dest,
 169         u32                             destlen,
 170         u32                             start_read_offset)
 171 {
 172         void *ring_buffer = hv_get_ring_buffer(ring_info);
 173         u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
 174
 175         memcpy(dest, ring_buffer + start_read_offset, destlen);
 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
 197         memcpy(ring_buffer + start_write_offset, src, srclen);
 198
 199         start_write_offset += srclen;
 200         start_write_offset %= ring_buffer_size;
 201
 202         return start_write_offset;
 203 }
 204
 205 /* Get various debug metrics for the specified ring buffer. */
 206 void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
 207                             struct hv_ring_buffer_debug_info *debug_info)
 208 {
 209         u32 bytes_avail_towrite;
 210         u32 bytes_avail_toread;
 211
 212         if (ring_info->ring_buffer) {
 213                 hv_get_ringbuffer_availbytes(ring_info,
 214                                         &bytes_avail_toread,
 215                                         &bytes_avail_towrite);
 216
 217                 debug_info->bytes_avail_toread = bytes_avail_toread;
 218                 debug_info->bytes_avail_towrite = bytes_avail_towrite;
 219                 debug_info->current_read_index =
 220                         ring_info->ring_buffer->read_index;
 221                 debug_info->current_write_index =
 222                         ring_info->ring_buffer->write_index;
 223                 debug_info->current_interrupt_mask =
 224                         ring_info->ring_buffer->interrupt_mask;
 225         }
 226 }
 227
 228 /* Initialize the ring buffer. */
 229 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
 230                        struct page *pages, u32 page_cnt)
 231 {
 232         int i;
 233         struct page **pages_wraparound;
 234
 235         BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE));
 236
 237         memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
 238
 239         /*
 240          * First page holds struct hv_ring_buffer, do wraparound mapping for
 241          * the rest.
 242          */
 243         pages_wraparound = kzalloc(sizeof(struct page *) * (page_cnt * 2 - 1),
 244                                    GFP_KERNEL);
 245         if (!pages_wraparound)
 246                 return -ENOMEM;
 247
 248         pages_wraparound[0] = pages;
 249         for (i = 0; i < 2 * (page_cnt - 1); i++)
 250                 pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1];
 251
 252         ring_info->ring_buffer = (struct hv_ring_buffer *)
 253                 vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL);
 254
 255         kfree(pages_wraparound);
 256
 257
 258         if (!ring_info->ring_buffer)
 259                 return -ENOMEM;
 260
 261         ring_info->ring_buffer->read_index =
 262                 ring_info->ring_buffer->write_index = 0;
 263
 264         /* Set the feature bit for enabling flow control. */
 265         ring_info->ring_buffer->feature_bits.value = 1;
 266
 267         ring_info->ring_size = page_cnt << PAGE_SHIFT;
 268         ring_info->ring_datasize = ring_info->ring_size -
 269                 sizeof(struct hv_ring_buffer);
 270
 271         spin_lock_init(&ring_info->ring_lock);
 272
 273         return 0;
 274 }
 275
 276 /* Cleanup the ring buffer. */
 277 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
 278 {
 279         vunmap(ring_info->ring_buffer);
 280 }
 281
 282 /* Write to the ring buffer. */
 283 int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
 284                     struct kvec *kv_list, u32 kv_count, bool *signal, bool lock,
 285                     enum hv_signal_policy policy)
 286 {
 287         int i = 0;
 288         u32 bytes_avail_towrite;
 289         u32 totalbytes_towrite = 0;
 290
 291         u32 next_write_location;
 292         u32 old_write;
 293         u64 prev_indices = 0;
 294         unsigned long flags = 0;
 295
 296         for (i = 0; i < kv_count; i++)
 297                 totalbytes_towrite += kv_list[i].iov_len;
 298
 299         totalbytes_towrite += sizeof(u64);
 300
 301         if (lock)
 302                 spin_lock_irqsave(&outring_info->ring_lock, flags);
 303
 304         bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
 305
 306         /*
 307          * If there is only room for the packet, assume it is full.
 308          * Otherwise, the next time around, we think the ring buffer
 309          * is empty since the read index == write index.
 310          */
 311         if (bytes_avail_towrite <= totalbytes_towrite) {
 312                 if (lock)
 313                         spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 314                 return -EAGAIN;
 315         }
 316
 317         /* Write to the ring buffer */
 318         next_write_location = hv_get_next_write_location(outring_info);
 319
 320         old_write = next_write_location;
 321
 322         for (i = 0; i < kv_count; i++) {
 323                 next_write_location = hv_copyto_ringbuffer(outring_info,
 324                                                      next_write_location,
 325                                                      kv_list[i].iov_base,
 326                                                      kv_list[i].iov_len);
 327         }
 328
 329         /* Set previous packet start */
 330         prev_indices = hv_get_ring_bufferindices(outring_info);
 331
 332         next_write_location = hv_copyto_ringbuffer(outring_info,
 333                                              next_write_location,
 334                                              &prev_indices,
 335                                              sizeof(u64));
 336
 337         /* Issue a full memory barrier before updating the write index */
 338         virt_mb();
 339
 340         /* Now, update the write location */
 341         hv_set_next_write_location(outring_info, next_write_location);
 342
 343
 344         if (lock)
 345                 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
 346
 347         *signal = hv_need_to_signal(old_write, outring_info, policy);
 348         return 0;
 349 }
 350
 351 int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info,
 352                        void *buffer, u32 buflen, u32 *buffer_actual_len,
 353                        u64 *requestid, bool *signal, bool raw)
 354 {
 355         u32 bytes_avail_toread;
 356         u32 next_read_location = 0;
 357         u64 prev_indices = 0;
 358         struct vmpacket_descriptor desc;
 359         u32 offset;
 360         u32 packetlen;
 361         int ret = 0;
 362
 363         if (buflen <= 0)
 364                 return -EINVAL;
 365
 366
 367         *buffer_actual_len = 0;
 368         *requestid = 0;
 369
 370         bytes_avail_toread = hv_get_bytes_to_read(inring_info);
 371         /* Make sure there is something to read */
 372         if (bytes_avail_toread < sizeof(desc)) {
 373                 /*
 374                  * No error is set when there is even no header, drivers are
 375                  * supposed to analyze buffer_actual_len.
 376                  */
 377                 return ret;
 378         }
 379
 380         next_read_location = hv_get_next_read_location(inring_info);
 381         next_read_location = hv_copyfrom_ringbuffer(inring_info, &desc,
 382                                                     sizeof(desc),
 383                                                     next_read_location);
 384
 385         offset = raw ? 0 : (desc.offset8 << 3);
 386         packetlen = (desc.len8 << 3) - offset;
 387         *buffer_actual_len = packetlen;
 388         *requestid = desc.trans_id;
 389
 390         if (bytes_avail_toread < packetlen + offset)
 391                 return -EAGAIN;
 392
 393         if (packetlen > buflen)
 394                 return -ENOBUFS;
 395
 396         next_read_location =
 397                 hv_get_next_readlocation_withoffset(inring_info, offset);
 398
 399         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 400                                                 buffer,
 401                                                 packetlen,
 402                                                 next_read_location);
 403
 404         next_read_location = hv_copyfrom_ringbuffer(inring_info,
 405                                                 &prev_indices,
 406                                                 sizeof(u64),
 407                                                 next_read_location);
 408
 409         /*
 410          * Make sure all reads are done before we update the read index since
 411          * the writer may start writing to the read area once the read index
 412          * is updated.
 413          */
 414         virt_mb();
 415
 416         /* Update the read index */
 417         hv_set_next_read_location(inring_info, next_read_location);
 418
 419         *signal = hv_need_to_signal_on_read(inring_info);
 420
 421         return ret;
 422 }