lib/ringbuffer/ring_buffer_iterator.c

   1 /*
   2  * ring_buffer_iterator.c
   3  *
   4  * (C) Copyright 2010 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   5  *
   6  * Ring buffer and channel iterators. Get each event of a channel in order. Uses
   7  * a prio heap for per-cpu buffers, giving a O(log(NR_CPUS)) algorithmic
   8  * complexity for the "get next event" operation.
   9  *
  10  * Author:
  11  *      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  12  *
  13  * Dual LGPL v2.1/GPL v2 license.
  14  */
  15
  16 #include "../../wrapper/ringbuffer/iterator.h"
  17 #include <linux/jiffies.h>
  18 #include <linux/delay.h>
  19 #include <linux/module.h>
  20
  21 /*
  22  * Safety factor taking into account internal kernel interrupt latency.
  23  * Assuming 250ms worse-case latency.
  24  */
  25 #define MAX_SYSTEM_LATENCY      250
  26
  27 /*
  28  * Maximum delta expected between trace clocks. At most 1 jiffy delta.
  29  */
  30 #define MAX_CLOCK_DELTA         (jiffies_to_usecs(1) * 1000)
  31
  32 /**
  33  * lib_ring_buffer_get_next_record - Get the next record in a buffer.
  34  * @chan: channel
  35  * @buf: buffer
  36  *
  37  * Returns the size of the event read, -EAGAIN if buffer is empty, -ENODATA if
  38  * buffer is empty and finalized. The buffer must already be opened for reading.
  39  */
  40 ssize_t lib_ring_buffer_get_next_record(struct channel *chan,
  41                                         struct lib_ring_buffer *buf)
  42 {
  43         const struct lib_ring_buffer_config *config = chan->backend.config;
  44         struct lib_ring_buffer_iter *iter = &buf->iter;
  45         int ret;
  46
  47 restart:
  48         switch (iter->state) {
  49         case ITER_GET_SUBBUF:
  50                 ret = lib_ring_buffer_get_next_subbuf(buf);
  51                 if (ret && !ACCESS_ONCE(buf->finalized)
  52                     && config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
  53                         /*
  54                          * Use "pull" scheme for global buffers. The reader
  55                          * itself flushes the buffer to "pull" data not visible
  56                          * to readers yet. Flush current subbuffer and re-try.
  57                          *
  58                          * Per-CPU buffers rather use a "push" scheme because
  59                          * the IPI needed to flush all CPU's buffers is too
  60                          * costly. In the "push" scheme, the reader waits for
  61                          * the writer periodic deferrable timer to flush the
  62                          * buffers (keeping track of a quiescent state
  63                          * timestamp). Therefore, the writer "pushes" data out
  64                          * of the buffers rather than letting the reader "pull"
  65                          * data from the buffer.
  66                          */
  67                         lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE);
  68                         ret = lib_ring_buffer_get_next_subbuf(buf);
  69                 }
  70                 if (ret)
  71                         return ret;
  72                 iter->consumed = buf->cons_snapshot;
  73                 iter->data_size = lib_ring_buffer_get_read_data_size(config, buf);
  74                 iter->read_offset = iter->consumed;
  75                 /* skip header */
  76                 iter->read_offset += config->cb.subbuffer_header_size();
  77                 iter->state = ITER_TEST_RECORD;
  78                 goto restart;
  79         case ITER_TEST_RECORD:
  80                 if (iter->read_offset - iter->consumed >= iter->data_size) {
  81                         iter->state = ITER_PUT_SUBBUF;
  82                 } else {
  83                         CHAN_WARN_ON(chan, !config->cb.record_get);
  84                         config->cb.record_get(config, chan, buf,
  85                                               iter->read_offset,
  86                                               &iter->header_len,
  87                                               &iter->payload_len,
  88                                               &iter->timestamp);
  89                         iter->read_offset += iter->header_len;
  90                         subbuffer_consume_record(config, &buf->backend);
  91                         iter->state = ITER_NEXT_RECORD;
  92                         return iter->payload_len;
  93                 }
  94                 goto restart;
  95         case ITER_NEXT_RECORD:
  96                 iter->read_offset += iter->payload_len;
  97                 iter->state = ITER_TEST_RECORD;
  98                 goto restart;
  99         case ITER_PUT_SUBBUF:
 100                 lib_ring_buffer_put_next_subbuf(buf);
 101                 iter->state = ITER_GET_SUBBUF;
 102                 goto restart;
 103         default:
 104                 CHAN_WARN_ON(chan, 1);  /* Should not happen */
 105                 return -EPERM;
 106         }
 107 }
 108 EXPORT_SYMBOL_GPL(lib_ring_buffer_get_next_record);
 109
 110 static int buf_is_higher(void *a, void *b)
 111 {
 112         struct lib_ring_buffer *bufa = a;
 113         struct lib_ring_buffer *bufb = b;
 114
 115         /* Consider lowest timestamps to be at the top of the heap */
 116         return (bufa->iter.timestamp < bufb->iter.timestamp);
 117 }
 118
 119 static
 120 void lib_ring_buffer_get_empty_buf_records(const struct lib_ring_buffer_config *config,
 121                                            struct channel *chan)
 122 {
 123         struct ptr_heap *heap = &chan->iter.heap;
 124         struct lib_ring_buffer *buf, *tmp;
 125         ssize_t len;
 126
 127         list_for_each_entry_safe(buf, tmp, &chan->iter.empty_head,
 128                                  iter.empty_node) {
 129                 len = lib_ring_buffer_get_next_record(chan, buf);
 130
 131                 /*
 132                  * Deal with -EAGAIN and -ENODATA.
 133                  * len >= 0 means record contains data.
 134                  * -EBUSY should never happen, because we support only one
 135                  * reader.
 136                  */
 137                 switch (len) {
 138                 case -EAGAIN:
 139                         /* Keep node in empty list */
 140                         break;
 141                 case -ENODATA:
 142                         /*
 143                          * Buffer is finalized. Don't add to list of empty
 144                          * buffer, because it has no more data to provide, ever.
 145                          */
 146                         list_del(&buf->iter.empty_node);
 147                         break;
 148                 case -EBUSY:
 149                         CHAN_WARN_ON(chan, 1);
 150                         break;
 151                 default:
 152                         /*
 153                          * Insert buffer into the heap, remove from empty buffer
 154                          * list. The heap should never overflow.
 155                          */
 156                         CHAN_WARN_ON(chan, len < 0);
 157                         list_del(&buf->iter.empty_node);
 158                         CHAN_WARN_ON(chan, heap_insert(heap, buf) != NULL);
 159                 }
 160         }
 161 }
 162
 163 static
 164 void lib_ring_buffer_wait_for_qs(const struct lib_ring_buffer_config *config,
 165                                  struct channel *chan)
 166 {
 167         u64 timestamp_qs;
 168         unsigned long wait_msecs;
 169
 170         /*
 171          * No need to wait if no empty buffers are present.
 172          */
 173         if (list_empty(&chan->iter.empty_head))
 174                 return;
 175
 176         timestamp_qs = config->cb.ring_buffer_clock_read(chan);
 177         /*
 178          * We need to consider previously empty buffers.
 179          * Do a get next buf record on each of them. Add them to
 180          * the heap if they have data. If at least one of them
 181          * don't have data, we need to wait for
 182          * switch_timer_interval + MAX_SYSTEM_LATENCY (so we are sure the
 183          * buffers have been switched either by the timer or idle entry) and
 184          * check them again, adding them if they have data.
 185          */
 186         lib_ring_buffer_get_empty_buf_records(config, chan);
 187
 188         /*
 189          * No need to wait if no empty buffers are present.
 190          */
 191         if (list_empty(&chan->iter.empty_head))
 192                 return;
 193
 194         /*
 195          * We need to wait for the buffer switch timer to run. If the
 196          * CPU is idle, idle entry performed the switch.
 197          * TODO: we could optimize further by skipping the sleep if all
 198          * empty buffers belong to idle or offline cpus.
 199          */
 200         wait_msecs = jiffies_to_msecs(chan->switch_timer_interval);
 201         wait_msecs += MAX_SYSTEM_LATENCY;
 202         msleep(wait_msecs);
 203         lib_ring_buffer_get_empty_buf_records(config, chan);
 204         /*
 205          * Any buffer still in the empty list here cannot possibly
 206          * contain an event with a timestamp prior to "timestamp_qs".
 207          * The new quiescent state timestamp is the one we grabbed
 208          * before waiting for buffer data.  It is therefore safe to
 209          * ignore empty buffers up to last_qs timestamp for fusion
 210          * merge.
 211          */
 212         chan->iter.last_qs = timestamp_qs;
 213 }
 214
 215 /**
 216  * channel_get_next_record - Get the next record in a channel.
 217  * @chan: channel
 218  * @ret_buf: the buffer in which the event is located (output)
 219  *
 220  * Returns the size of new current event, -EAGAIN if all buffers are empty,
 221  * -ENODATA if all buffers are empty and finalized. The channel must already be
 222  * opened for reading.
 223  */
 224
 225 ssize_t channel_get_next_record(struct channel *chan,
 226                                 struct lib_ring_buffer **ret_buf)
 227 {
 228         const struct lib_ring_buffer_config *config = chan->backend.config;
 229         struct lib_ring_buffer *buf;
 230         struct ptr_heap *heap;
 231         ssize_t len;
 232
 233         if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
 234                 *ret_buf = channel_get_ring_buffer(config, chan, 0);
 235                 return lib_ring_buffer_get_next_record(chan, *ret_buf);
 236         }
 237
 238         heap = &chan->iter.heap;
 239
 240         /*
 241          * get next record for topmost buffer.
 242          */
 243         buf = heap_maximum(heap);
 244         if (buf) {
 245                 len = lib_ring_buffer_get_next_record(chan, buf);
 246                 /*
 247                  * Deal with -EAGAIN and -ENODATA.
 248                  * len >= 0 means record contains data.
 249                  */
 250                 switch (len) {
 251                 case -EAGAIN:
 252                         buf->iter.timestamp = 0;
 253                         list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 254                         /* Remove topmost buffer from the heap */
 255                         CHAN_WARN_ON(chan, heap_remove(heap) != buf);
 256                         break;
 257                 case -ENODATA:
 258                         /*
 259                          * Buffer is finalized. Remove buffer from heap and
 260                          * don't add to list of empty buffer, because it has no
 261                          * more data to provide, ever.
 262                          */
 263                         CHAN_WARN_ON(chan, heap_remove(heap) != buf);
 264                         break;
 265                 case -EBUSY:
 266                         CHAN_WARN_ON(chan, 1);
 267                         break;
 268                 default:
 269                         /*
 270                          * Reinsert buffer into the heap. Note that heap can be
 271                          * partially empty, so we need to use
 272                          * heap_replace_max().
 273                          */
 274                         CHAN_WARN_ON(chan, len < 0);
 275                         CHAN_WARN_ON(chan, heap_replace_max(heap, buf) != buf);
 276                         break;
 277                 }
 278         }
 279
 280         buf = heap_maximum(heap);
 281         if (!buf || buf->iter.timestamp > chan->iter.last_qs) {
 282                 /*
 283                  * Deal with buffers previously showing no data.
 284                  * Add buffers containing data to the heap, update
 285                  * last_qs.
 286                  */
 287                 lib_ring_buffer_wait_for_qs(config, chan);
 288         }
 289
 290         *ret_buf = buf = heap_maximum(heap);
 291         if (buf) {
 292                 /*
 293                  * If this warning triggers, you probably need to check your
 294                  * system interrupt latency. Typical causes: too many printk()
 295                  * output going to a serial console with interrupts off.
 296                  * Allow for MAX_CLOCK_DELTA ns timestamp delta going backward.
 297                  * Observed on SMP KVM setups with trace_clock().
 298                  */
 299                 if (chan->iter.last_timestamp
 300                     > (buf->iter.timestamp + MAX_CLOCK_DELTA)) {
 301                         printk(KERN_WARNING "ring_buffer: timestamps going "
 302                                "backward. Last time %llu ns, cpu %d, "
 303                                "current time %llu ns, cpu %d, "
 304                                "delta %llu ns.\n",
 305                                chan->iter.last_timestamp, chan->iter.last_cpu,
 306                                buf->iter.timestamp, buf->backend.cpu,
 307                                chan->iter.last_timestamp - buf->iter.timestamp);
 308                         CHAN_WARN_ON(chan, 1);
 309                 }
 310                 chan->iter.last_timestamp = buf->iter.timestamp;
 311                 chan->iter.last_cpu = buf->backend.cpu;
 312                 return buf->iter.payload_len;
 313         } else {
 314                 /* Heap is empty */
 315                 if (list_empty(&chan->iter.empty_head))
 316                         return -ENODATA;        /* All buffers finalized */
 317                 else
 318                         return -EAGAIN;         /* Temporarily empty */
 319         }
 320 }
 321 EXPORT_SYMBOL_GPL(channel_get_next_record);
 322
 323 static
 324 void lib_ring_buffer_iterator_init(struct channel *chan, struct lib_ring_buffer *buf)
 325 {
 326         if (buf->iter.allocated)
 327                 return;
 328
 329         buf->iter.allocated = 1;
 330         if (chan->iter.read_open && !buf->iter.read_open) {
 331                 CHAN_WARN_ON(chan, lib_ring_buffer_open_read(buf) != 0);
 332                 buf->iter.read_open = 1;
 333         }
 334
 335         /* Add to list of buffers without any current record */
 336         if (chan->backend.config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 337                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 338 }
 339
 340 #ifdef CONFIG_HOTPLUG_CPU
 341 static
 342 int __cpuinit channel_iterator_cpu_hotplug(struct notifier_block *nb,
 343                                            unsigned long action,
 344                                            void *hcpu)
 345 {
 346         unsigned int cpu = (unsigned long)hcpu;
 347         struct channel *chan = container_of(nb, struct channel,
 348                                             hp_iter_notifier);
 349         struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu);
 350         const struct lib_ring_buffer_config *config = chan->backend.config;
 351
 352         if (!chan->hp_iter_enable)
 353                 return NOTIFY_DONE;
 354
 355         CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL);
 356
 357         switch (action) {
 358         case CPU_DOWN_FAILED:
 359         case CPU_DOWN_FAILED_FROZEN:
 360         case CPU_ONLINE:
 361         case CPU_ONLINE_FROZEN:
 362                 lib_ring_buffer_iterator_init(chan, buf);
 363                 return NOTIFY_OK;
 364         default:
 365                 return NOTIFY_DONE;
 366         }
 367 }
 368 #endif
 369
 370 int channel_iterator_init(struct channel *chan)
 371 {
 372         const struct lib_ring_buffer_config *config = chan->backend.config;
 373         struct lib_ring_buffer *buf;
 374
 375         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 376                 int cpu, ret;
 377
 378                 INIT_LIST_HEAD(&chan->iter.empty_head);
 379                 ret = heap_init(&chan->iter.heap,
 380                                 num_possible_cpus()
 381                                 * sizeof(struct lib_ring_buffer *),
 382                                 GFP_KERNEL, buf_is_higher);
 383                 if (ret)
 384                         return ret;
 385                 /*
 386                  * In case of non-hotplug cpu, if the ring-buffer is allocated
 387                  * in early initcall, it will not be notified of secondary cpus.
 388                  * In that off case, we need to allocate for all possible cpus.
 389                  */
 390 #ifdef CONFIG_HOTPLUG_CPU
 391                 chan->hp_iter_notifier.notifier_call =
 392                         channel_iterator_cpu_hotplug;
 393                 chan->hp_iter_notifier.priority = 10;
 394                 register_cpu_notifier(&chan->hp_iter_notifier);
 395                 get_online_cpus();
 396                 for_each_online_cpu(cpu) {
 397                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 398                         lib_ring_buffer_iterator_init(chan, buf);
 399                 }
 400                 chan->hp_iter_enable = 1;
 401                 put_online_cpus();
 402 #else
 403                 for_each_possible_cpu(cpu) {
 404                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 405                         lib_ring_buffer_iterator_init(chan, buf);
 406                 }
 407 #endif
 408         } else {
 409                 buf = channel_get_ring_buffer(config, chan, 0);
 410                 lib_ring_buffer_iterator_init(chan, buf);
 411         }
 412         return 0;
 413 }
 414
 415 void channel_iterator_unregister_notifiers(struct channel *chan)
 416 {
 417         const struct lib_ring_buffer_config *config = chan->backend.config;
 418
 419         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 420                 chan->hp_iter_enable = 0;
 421                 unregister_cpu_notifier(&chan->hp_iter_notifier);
 422         }
 423 }
 424
 425 void channel_iterator_free(struct channel *chan)
 426 {
 427         const struct lib_ring_buffer_config *config = chan->backend.config;
 428
 429         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 430                 heap_free(&chan->iter.heap);
 431 }
 432
 433 int lib_ring_buffer_iterator_open(struct lib_ring_buffer *buf)
 434 {
 435         struct channel *chan = buf->backend.chan;
 436         const struct lib_ring_buffer_config *config = chan->backend.config;
 437         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 438         return lib_ring_buffer_open_read(buf);
 439 }
 440 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_open);
 441
 442 /*
 443  * Note: Iterators must not be mixed with other types of outputs, because an
 444  * iterator can leave the buffer in "GET" state, which is not consistent with
 445  * other types of output (mmap, splice, raw data read).
 446  */
 447 void lib_ring_buffer_iterator_release(struct lib_ring_buffer *buf)
 448 {
 449         lib_ring_buffer_release_read(buf);
 450 }
 451 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_release);
 452
 453 int channel_iterator_open(struct channel *chan)
 454 {
 455         const struct lib_ring_buffer_config *config = chan->backend.config;
 456         struct lib_ring_buffer *buf;
 457         int ret = 0, cpu;
 458
 459         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 460
 461         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 462                 get_online_cpus();
 463                 /* Allow CPU hotplug to keep track of opened reader */
 464                 chan->iter.read_open = 1;
 465                 for_each_channel_cpu(cpu, chan) {
 466                         buf = channel_get_ring_buffer(config, chan, cpu);
 467                         ret = lib_ring_buffer_iterator_open(buf);
 468                         if (ret)
 469                                 goto error;
 470                         buf->iter.read_open = 1;
 471                 }
 472                 put_online_cpus();
 473         } else {
 474                 buf = channel_get_ring_buffer(config, chan, 0);
 475                 ret = lib_ring_buffer_iterator_open(buf);
 476         }
 477         return ret;
 478 error:
 479         /* Error should always happen on CPU 0, hence no close is required. */
 480         CHAN_WARN_ON(chan, cpu != 0);
 481         put_online_cpus();
 482         return ret;
 483 }
 484 EXPORT_SYMBOL_GPL(channel_iterator_open);
 485
 486 void channel_iterator_release(struct channel *chan)
 487 {
 488         const struct lib_ring_buffer_config *config = chan->backend.config;
 489         struct lib_ring_buffer *buf;
 490         int cpu;
 491
 492         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 493                 get_online_cpus();
 494                 for_each_channel_cpu(cpu, chan) {
 495                         buf = channel_get_ring_buffer(config, chan, cpu);
 496                         if (buf->iter.read_open) {
 497                                 lib_ring_buffer_iterator_release(buf);
 498                                 buf->iter.read_open = 0;
 499                         }
 500                 }
 501                 chan->iter.read_open = 0;
 502                 put_online_cpus();
 503         } else {
 504                 buf = channel_get_ring_buffer(config, chan, 0);
 505                 lib_ring_buffer_iterator_release(buf);
 506         }
 507 }
 508 EXPORT_SYMBOL_GPL(channel_iterator_release);
 509
 510 void lib_ring_buffer_iterator_reset(struct lib_ring_buffer *buf)
 511 {
 512         struct channel *chan = buf->backend.chan;
 513
 514         if (buf->iter.state != ITER_GET_SUBBUF)
 515                 lib_ring_buffer_put_next_subbuf(buf);
 516         buf->iter.state = ITER_GET_SUBBUF;
 517         /* Remove from heap (if present). */
 518         if (heap_cherrypick(&chan->iter.heap, buf))
 519                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 520         buf->iter.timestamp = 0;
 521         buf->iter.header_len = 0;
 522         buf->iter.payload_len = 0;
 523         buf->iter.consumed = 0;
 524         buf->iter.read_offset = 0;
 525         buf->iter.data_size = 0;
 526         /* Don't reset allocated and read_open */
 527 }
 528
 529 void channel_iterator_reset(struct channel *chan)
 530 {
 531         const struct lib_ring_buffer_config *config = chan->backend.config;
 532         struct lib_ring_buffer *buf;
 533         int cpu;
 534
 535         /* Empty heap, put into empty_head */
 536         while ((buf = heap_remove(&chan->iter.heap)) != NULL)
 537                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 538
 539         for_each_channel_cpu(cpu, chan) {
 540                 buf = channel_get_ring_buffer(config, chan, cpu);
 541                 lib_ring_buffer_iterator_reset(buf);
 542         }
 543         /* Don't reset read_open */
 544         chan->iter.last_qs = 0;
 545         chan->iter.last_timestamp = 0;
 546         chan->iter.last_cpu = 0;
 547         chan->iter.len_left = 0;
 548 }
 549
 550 /*
 551  * Ring buffer payload extraction read() implementation.
 552  */
 553 static
 554 ssize_t channel_ring_buffer_file_read(struct file *filp,
 555                                       char __user *user_buf,
 556                                       size_t count,
 557                                       loff_t *ppos,
 558                                       struct channel *chan,
 559                                       struct lib_ring_buffer *buf,
 560                                       int fusionmerge)
 561 {
 562         const struct lib_ring_buffer_config *config = chan->backend.config;
 563         size_t read_count = 0, read_offset;
 564         ssize_t len;
 565
 566         might_sleep();
 567         if (!access_ok(VERIFY_WRITE, user_buf, count))
 568                 return -EFAULT;
 569
 570         /* Finish copy of previous record */
 571         if (*ppos != 0) {
 572                 if (read_count < count) {
 573                         len = chan->iter.len_left;
 574                         read_offset = *ppos;
 575                         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU
 576                             && fusionmerge)
 577                                 buf = heap_maximum(&chan->iter.heap);
 578                         CHAN_WARN_ON(chan, !buf);
 579                         goto skip_get_next;
 580                 }
 581         }
 582
 583         while (read_count < count) {
 584                 size_t copy_len, space_left;
 585
 586                 if (fusionmerge)
 587                         len = channel_get_next_record(chan, &buf);
 588                 else
 589                         len = lib_ring_buffer_get_next_record(chan, buf);
 590 len_test:
 591                 if (len < 0) {
 592                         /*
 593                          * Check if buffer is finalized (end of file).
 594                          */
 595                         if (len == -ENODATA) {
 596                                 /* A 0 read_count will tell about end of file */
 597                                 goto nodata;
 598                         }
 599                         if (filp->f_flags & O_NONBLOCK) {
 600                                 if (!read_count)
 601                                         read_count = -EAGAIN;
 602                                 goto nodata;
 603                         } else {
 604                                 int error;
 605
 606                                 /*
 607                                  * No data available at the moment, return what
 608                                  * we got.
 609                                  */
 610                                 if (read_count)
 611                                         goto nodata;
 612
 613                                 /*
 614                                  * Wait for returned len to be >= 0 or -ENODATA.
 615                                  */
 616                                 if (fusionmerge)
 617                                         error = wait_event_interruptible(
 618                                           chan->read_wait,
 619                                           ((len = channel_get_next_record(chan,
 620                                                 &buf)), len != -EAGAIN));
 621                                 else
 622                                         error = wait_event_interruptible(
 623                                           buf->read_wait,
 624                                           ((len = lib_ring_buffer_get_next_record(
 625                                                   chan, buf)), len != -EAGAIN));
 626                                 CHAN_WARN_ON(chan, len == -EBUSY);
 627                                 if (error) {
 628                                         read_count = error;
 629                                         goto nodata;
 630                                 }
 631                                 CHAN_WARN_ON(chan, len < 0 && len != -ENODATA);
 632                                 goto len_test;
 633                         }
 634                 }
 635                 read_offset = buf->iter.read_offset;
 636 skip_get_next:
 637                 space_left = count - read_count;
 638                 if (len <= space_left) {
 639                         copy_len = len;
 640                         chan->iter.len_left = 0;
 641                         *ppos = 0;
 642                 } else {
 643                         copy_len = space_left;
 644                         chan->iter.len_left = len - copy_len;
 645                         *ppos = read_offset + copy_len;
 646                 }
 647                 if (__lib_ring_buffer_copy_to_user(&buf->backend, read_offset,
 648                                                &user_buf[read_count],
 649                                                copy_len)) {
 650                         /*
 651                          * Leave the len_left and ppos values at their current
 652                          * state, as we currently have a valid event to read.
 653                          */
 654                         return -EFAULT;
 655                 }
 656                 read_count += copy_len;
 657         };
 658         return read_count;
 659
 660 nodata:
 661         *ppos = 0;
 662         chan->iter.len_left = 0;
 663         return read_count;
 664 }
 665
 666 /**
 667  * lib_ring_buffer_file_read - Read buffer record payload.
 668  * @filp: file structure pointer.
 669  * @buffer: user buffer to read data into.
 670  * @count: number of bytes to read.
 671  * @ppos: file read position.
 672  *
 673  * Returns a negative value on error, or the number of bytes read on success.
 674  * ppos is used to save the position _within the current record_ between calls
 675  * to read().
 676  */
 677 static
 678 ssize_t lib_ring_buffer_file_read(struct file *filp,
 679                                   char __user *user_buf,
 680                                   size_t count,
 681                                   loff_t *ppos)
 682 {
 683         struct inode *inode = filp->f_dentry->d_inode;
 684         struct lib_ring_buffer *buf = inode->i_private;
 685         struct channel *chan = buf->backend.chan;
 686
 687         return channel_ring_buffer_file_read(filp, user_buf, count, ppos,
 688                                              chan, buf, 0);
 689 }
 690
 691 /**
 692  * channel_file_read - Read channel record payload.
 693  * @filp: file structure pointer.
 694  * @buffer: user buffer to read data into.
 695  * @count: number of bytes to read.
 696  * @ppos: file read position.
 697  *
 698  * Returns a negative value on error, or the number of bytes read on success.
 699  * ppos is used to save the position _within the current record_ between calls
 700  * to read().
 701  */
 702 static
 703 ssize_t channel_file_read(struct file *filp,
 704                           char __user *user_buf,
 705                           size_t count,
 706                           loff_t *ppos)
 707 {
 708         struct inode *inode = filp->f_dentry->d_inode;
 709         struct channel *chan = inode->i_private;
 710         const struct lib_ring_buffer_config *config = chan->backend.config;
 711
 712         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 713                 return channel_ring_buffer_file_read(filp, user_buf, count,
 714                                                      ppos, chan, NULL, 1);
 715         else {
 716                 struct lib_ring_buffer *buf =
 717                         channel_get_ring_buffer(config, chan, 0);
 718                 return channel_ring_buffer_file_read(filp, user_buf, count,
 719                                                      ppos, chan, buf, 0);
 720         }
 721 }
 722
 723 static
 724 int lib_ring_buffer_file_open(struct inode *inode, struct file *file)
 725 {
 726         struct lib_ring_buffer *buf = inode->i_private;
 727         int ret;
 728
 729         ret = lib_ring_buffer_iterator_open(buf);
 730         if (ret)
 731                 return ret;
 732
 733         file->private_data = buf;
 734         ret = nonseekable_open(inode, file);
 735         if (ret)
 736                 goto release_iter;
 737         return 0;
 738
 739 release_iter:
 740         lib_ring_buffer_iterator_release(buf);
 741         return ret;
 742 }
 743
 744 static
 745 int lib_ring_buffer_file_release(struct inode *inode, struct file *file)
 746 {
 747         struct lib_ring_buffer *buf = inode->i_private;
 748
 749         lib_ring_buffer_iterator_release(buf);
 750         return 0;
 751 }
 752
 753 static
 754 int channel_file_open(struct inode *inode, struct file *file)
 755 {
 756         struct channel *chan = inode->i_private;
 757         int ret;
 758
 759         ret = channel_iterator_open(chan);
 760         if (ret)
 761                 return ret;
 762
 763         file->private_data = chan;
 764         ret = nonseekable_open(inode, file);
 765         if (ret)
 766                 goto release_iter;
 767         return 0;
 768
 769 release_iter:
 770         channel_iterator_release(chan);
 771         return ret;
 772 }
 773
 774 static
 775 int channel_file_release(struct inode *inode, struct file *file)
 776 {
 777         struct channel *chan = inode->i_private;
 778
 779         channel_iterator_release(chan);
 780         return 0;
 781 }
 782
 783 const struct file_operations channel_payload_file_operations = {
 784         .open = channel_file_open,
 785         .release = channel_file_release,
 786         .read = channel_file_read,
 787         .llseek = lib_ring_buffer_no_llseek,
 788 };
 789 EXPORT_SYMBOL_GPL(channel_payload_file_operations);
 790
 791 const struct file_operations lib_ring_buffer_payload_file_operations = {
 792         .open = lib_ring_buffer_file_open,
 793         .release = lib_ring_buffer_file_release,
 794         .read = lib_ring_buffer_file_read,
 795         .llseek = lib_ring_buffer_no_llseek,
 796 };
 797 EXPORT_SYMBOL_GPL(lib_ring_buffer_payload_file_operations);