drivers/staging/lttng/lib/ringbuffer/ring_buffer_iterator.c

   1 /*
   2  * ring_buffer_iterator.c
   3  *
   4  * Ring buffer and channel iterators. Get each event of a channel in order. Uses
   5  * a prio heap for per-cpu buffers, giving a O(log(NR_CPUS)) algorithmic
   6  * complexity for the "get next event" operation.
   7  *
   8  * Copyright (C) 2010-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   9  *
  10  * This library is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; only
  13  * version 2.1 of the License.
  14  *
  15  * This library is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with this library; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  *
  24  * Author:
  25  *      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  26  */
  27
  28 #include "../../wrapper/ringbuffer/iterator.h"
  29 #include <linux/jiffies.h>
  30 #include <linux/delay.h>
  31 #include <linux/module.h>
  32
  33 /*
  34  * Safety factor taking into account internal kernel interrupt latency.
  35  * Assuming 250ms worse-case latency.
  36  */
  37 #define MAX_SYSTEM_LATENCY      250
  38
  39 /*
  40  * Maximum delta expected between trace clocks. At most 1 jiffy delta.
  41  */
  42 #define MAX_CLOCK_DELTA         (jiffies_to_usecs(1) * 1000)
  43
  44 /**
  45  * lib_ring_buffer_get_next_record - Get the next record in a buffer.
  46  * @chan: channel
  47  * @buf: buffer
  48  *
  49  * Returns the size of the event read, -EAGAIN if buffer is empty, -ENODATA if
  50  * buffer is empty and finalized. The buffer must already be opened for reading.
  51  */
  52 ssize_t lib_ring_buffer_get_next_record(struct channel *chan,
  53                                         struct lib_ring_buffer *buf)
  54 {
  55         const struct lib_ring_buffer_config *config = &chan->backend.config;
  56         struct lib_ring_buffer_iter *iter = &buf->iter;
  57         int ret;
  58
  59 restart:
  60         switch (iter->state) {
  61         case ITER_GET_SUBBUF:
  62                 ret = lib_ring_buffer_get_next_subbuf(buf);
  63                 if (ret && !ACCESS_ONCE(buf->finalized)
  64                     && config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
  65                         /*
  66                          * Use "pull" scheme for global buffers. The reader
  67                          * itself flushes the buffer to "pull" data not visible
  68                          * to readers yet. Flush current subbuffer and re-try.
  69                          *
  70                          * Per-CPU buffers rather use a "push" scheme because
  71                          * the IPI needed to flush all CPU's buffers is too
  72                          * costly. In the "push" scheme, the reader waits for
  73                          * the writer periodic deferrable timer to flush the
  74                          * buffers (keeping track of a quiescent state
  75                          * timestamp). Therefore, the writer "pushes" data out
  76                          * of the buffers rather than letting the reader "pull"
  77                          * data from the buffer.
  78                          */
  79                         lib_ring_buffer_switch_slow(buf, SWITCH_ACTIVE);
  80                         ret = lib_ring_buffer_get_next_subbuf(buf);
  81                 }
  82                 if (ret)
  83                         return ret;
  84                 iter->consumed = buf->cons_snapshot;
  85                 iter->data_size = lib_ring_buffer_get_read_data_size(config, buf);
  86                 iter->read_offset = iter->consumed;
  87                 /* skip header */
  88                 iter->read_offset += config->cb.subbuffer_header_size();
  89                 iter->state = ITER_TEST_RECORD;
  90                 goto restart;
  91         case ITER_TEST_RECORD:
  92                 if (iter->read_offset - iter->consumed >= iter->data_size) {
  93                         iter->state = ITER_PUT_SUBBUF;
  94                 } else {
  95                         CHAN_WARN_ON(chan, !config->cb.record_get);
  96                         config->cb.record_get(config, chan, buf,
  97                                               iter->read_offset,
  98                                               &iter->header_len,
  99                                               &iter->payload_len,
 100                                               &iter->timestamp);
 101                         iter->read_offset += iter->header_len;
 102                         subbuffer_consume_record(config, &buf->backend);
 103                         iter->state = ITER_NEXT_RECORD;
 104                         return iter->payload_len;
 105                 }
 106                 goto restart;
 107         case ITER_NEXT_RECORD:
 108                 iter->read_offset += iter->payload_len;
 109                 iter->state = ITER_TEST_RECORD;
 110                 goto restart;
 111         case ITER_PUT_SUBBUF:
 112                 lib_ring_buffer_put_next_subbuf(buf);
 113                 iter->state = ITER_GET_SUBBUF;
 114                 goto restart;
 115         default:
 116                 CHAN_WARN_ON(chan, 1);  /* Should not happen */
 117                 return -EPERM;
 118         }
 119 }
 120 EXPORT_SYMBOL_GPL(lib_ring_buffer_get_next_record);
 121
 122 static int buf_is_higher(void *a, void *b)
 123 {
 124         struct lib_ring_buffer *bufa = a;
 125         struct lib_ring_buffer *bufb = b;
 126
 127         /* Consider lowest timestamps to be at the top of the heap */
 128         return (bufa->iter.timestamp < bufb->iter.timestamp);
 129 }
 130
 131 static
 132 void lib_ring_buffer_get_empty_buf_records(const struct lib_ring_buffer_config *config,
 133                                            struct channel *chan)
 134 {
 135         struct lttng_ptr_heap *heap = &chan->iter.heap;
 136         struct lib_ring_buffer *buf, *tmp;
 137         ssize_t len;
 138
 139         list_for_each_entry_safe(buf, tmp, &chan->iter.empty_head,
 140                                  iter.empty_node) {
 141                 len = lib_ring_buffer_get_next_record(chan, buf);
 142
 143                 /*
 144                  * Deal with -EAGAIN and -ENODATA.
 145                  * len >= 0 means record contains data.
 146                  * -EBUSY should never happen, because we support only one
 147                  * reader.
 148                  */
 149                 switch (len) {
 150                 case -EAGAIN:
 151                         /* Keep node in empty list */
 152                         break;
 153                 case -ENODATA:
 154                         /*
 155                          * Buffer is finalized. Don't add to list of empty
 156                          * buffer, because it has no more data to provide, ever.
 157                          */
 158                         list_del(&buf->iter.empty_node);
 159                         break;
 160                 case -EBUSY:
 161                         CHAN_WARN_ON(chan, 1);
 162                         break;
 163                 default:
 164                         /*
 165                          * Insert buffer into the heap, remove from empty buffer
 166                          * list.
 167                          */
 168                         CHAN_WARN_ON(chan, len < 0);
 169                         list_del(&buf->iter.empty_node);
 170                         CHAN_WARN_ON(chan, lttng_heap_insert(heap, buf));
 171                 }
 172         }
 173 }
 174
 175 static
 176 void lib_ring_buffer_wait_for_qs(const struct lib_ring_buffer_config *config,
 177                                  struct channel *chan)
 178 {
 179         u64 timestamp_qs;
 180         unsigned long wait_msecs;
 181
 182         /*
 183          * No need to wait if no empty buffers are present.
 184          */
 185         if (list_empty(&chan->iter.empty_head))
 186                 return;
 187
 188         timestamp_qs = config->cb.ring_buffer_clock_read(chan);
 189         /*
 190          * We need to consider previously empty buffers.
 191          * Do a get next buf record on each of them. Add them to
 192          * the heap if they have data. If at least one of them
 193          * don't have data, we need to wait for
 194          * switch_timer_interval + MAX_SYSTEM_LATENCY (so we are sure the
 195          * buffers have been switched either by the timer or idle entry) and
 196          * check them again, adding them if they have data.
 197          */
 198         lib_ring_buffer_get_empty_buf_records(config, chan);
 199
 200         /*
 201          * No need to wait if no empty buffers are present.
 202          */
 203         if (list_empty(&chan->iter.empty_head))
 204                 return;
 205
 206         /*
 207          * We need to wait for the buffer switch timer to run. If the
 208          * CPU is idle, idle entry performed the switch.
 209          * TODO: we could optimize further by skipping the sleep if all
 210          * empty buffers belong to idle or offline cpus.
 211          */
 212         wait_msecs = jiffies_to_msecs(chan->switch_timer_interval);
 213         wait_msecs += MAX_SYSTEM_LATENCY;
 214         msleep(wait_msecs);
 215         lib_ring_buffer_get_empty_buf_records(config, chan);
 216         /*
 217          * Any buffer still in the empty list here cannot possibly
 218          * contain an event with a timestamp prior to "timestamp_qs".
 219          * The new quiescent state timestamp is the one we grabbed
 220          * before waiting for buffer data.  It is therefore safe to
 221          * ignore empty buffers up to last_qs timestamp for fusion
 222          * merge.
 223          */
 224         chan->iter.last_qs = timestamp_qs;
 225 }
 226
 227 /**
 228  * channel_get_next_record - Get the next record in a channel.
 229  * @chan: channel
 230  * @ret_buf: the buffer in which the event is located (output)
 231  *
 232  * Returns the size of new current event, -EAGAIN if all buffers are empty,
 233  * -ENODATA if all buffers are empty and finalized. The channel must already be
 234  * opened for reading.
 235  */
 236
 237 ssize_t channel_get_next_record(struct channel *chan,
 238                                 struct lib_ring_buffer **ret_buf)
 239 {
 240         const struct lib_ring_buffer_config *config = &chan->backend.config;
 241         struct lib_ring_buffer *buf;
 242         struct lttng_ptr_heap *heap;
 243         ssize_t len;
 244
 245         if (config->alloc == RING_BUFFER_ALLOC_GLOBAL) {
 246                 *ret_buf = channel_get_ring_buffer(config, chan, 0);
 247                 return lib_ring_buffer_get_next_record(chan, *ret_buf);
 248         }
 249
 250         heap = &chan->iter.heap;
 251
 252         /*
 253          * get next record for topmost buffer.
 254          */
 255         buf = lttng_heap_maximum(heap);
 256         if (buf) {
 257                 len = lib_ring_buffer_get_next_record(chan, buf);
 258                 /*
 259                  * Deal with -EAGAIN and -ENODATA.
 260                  * len >= 0 means record contains data.
 261                  */
 262                 switch (len) {
 263                 case -EAGAIN:
 264                         buf->iter.timestamp = 0;
 265                         list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 266                         /* Remove topmost buffer from the heap */
 267                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 268                         break;
 269                 case -ENODATA:
 270                         /*
 271                          * Buffer is finalized. Remove buffer from heap and
 272                          * don't add to list of empty buffer, because it has no
 273                          * more data to provide, ever.
 274                          */
 275                         CHAN_WARN_ON(chan, lttng_heap_remove(heap) != buf);
 276                         break;
 277                 case -EBUSY:
 278                         CHAN_WARN_ON(chan, 1);
 279                         break;
 280                 default:
 281                         /*
 282                          * Reinsert buffer into the heap. Note that heap can be
 283                          * partially empty, so we need to use
 284                          * lttng_heap_replace_max().
 285                          */
 286                         CHAN_WARN_ON(chan, len < 0);
 287                         CHAN_WARN_ON(chan, lttng_heap_replace_max(heap, buf) != buf);
 288                         break;
 289                 }
 290         }
 291
 292         buf = lttng_heap_maximum(heap);
 293         if (!buf || buf->iter.timestamp > chan->iter.last_qs) {
 294                 /*
 295                  * Deal with buffers previously showing no data.
 296                  * Add buffers containing data to the heap, update
 297                  * last_qs.
 298                  */
 299                 lib_ring_buffer_wait_for_qs(config, chan);
 300         }
 301
 302         *ret_buf = buf = lttng_heap_maximum(heap);
 303         if (buf) {
 304                 /*
 305                  * If this warning triggers, you probably need to check your
 306                  * system interrupt latency. Typical causes: too many printk()
 307                  * output going to a serial console with interrupts off.
 308                  * Allow for MAX_CLOCK_DELTA ns timestamp delta going backward.
 309                  * Observed on SMP KVM setups with trace_clock().
 310                  */
 311                 if (chan->iter.last_timestamp
 312                     > (buf->iter.timestamp + MAX_CLOCK_DELTA)) {
 313                         printk(KERN_WARNING "ring_buffer: timestamps going "
 314                                "backward. Last time %llu ns, cpu %d, "
 315                                "current time %llu ns, cpu %d, "
 316                                "delta %llu ns.\n",
 317                                chan->iter.last_timestamp, chan->iter.last_cpu,
 318                                buf->iter.timestamp, buf->backend.cpu,
 319                                chan->iter.last_timestamp - buf->iter.timestamp);
 320                         CHAN_WARN_ON(chan, 1);
 321                 }
 322                 chan->iter.last_timestamp = buf->iter.timestamp;
 323                 chan->iter.last_cpu = buf->backend.cpu;
 324                 return buf->iter.payload_len;
 325         } else {
 326                 /* Heap is empty */
 327                 if (list_empty(&chan->iter.empty_head))
 328                         return -ENODATA;        /* All buffers finalized */
 329                 else
 330                         return -EAGAIN;         /* Temporarily empty */
 331         }
 332 }
 333 EXPORT_SYMBOL_GPL(channel_get_next_record);
 334
 335 static
 336 void lib_ring_buffer_iterator_init(struct channel *chan, struct lib_ring_buffer *buf)
 337 {
 338         if (buf->iter.allocated)
 339                 return;
 340
 341         buf->iter.allocated = 1;
 342         if (chan->iter.read_open && !buf->iter.read_open) {
 343                 CHAN_WARN_ON(chan, lib_ring_buffer_open_read(buf) != 0);
 344                 buf->iter.read_open = 1;
 345         }
 346
 347         /* Add to list of buffers without any current record */
 348         if (chan->backend.config.alloc == RING_BUFFER_ALLOC_PER_CPU)
 349                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 350 }
 351
 352 #ifdef CONFIG_HOTPLUG_CPU
 353 static
 354 int channel_iterator_cpu_hotplug(struct notifier_block *nb,
 355                                            unsigned long action,
 356                                            void *hcpu)
 357 {
 358         unsigned int cpu = (unsigned long)hcpu;
 359         struct channel *chan = container_of(nb, struct channel,
 360                                             hp_iter_notifier);
 361         struct lib_ring_buffer *buf = per_cpu_ptr(chan->backend.buf, cpu);
 362         const struct lib_ring_buffer_config *config = &chan->backend.config;
 363
 364         if (!chan->hp_iter_enable)
 365                 return NOTIFY_DONE;
 366
 367         CHAN_WARN_ON(chan, config->alloc == RING_BUFFER_ALLOC_GLOBAL);
 368
 369         switch (action) {
 370         case CPU_DOWN_FAILED:
 371         case CPU_DOWN_FAILED_FROZEN:
 372         case CPU_ONLINE:
 373         case CPU_ONLINE_FROZEN:
 374                 lib_ring_buffer_iterator_init(chan, buf);
 375                 return NOTIFY_OK;
 376         default:
 377                 return NOTIFY_DONE;
 378         }
 379 }
 380 #endif
 381
 382 int channel_iterator_init(struct channel *chan)
 383 {
 384         const struct lib_ring_buffer_config *config = &chan->backend.config;
 385         struct lib_ring_buffer *buf;
 386
 387         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 388                 int cpu, ret;
 389
 390                 INIT_LIST_HEAD(&chan->iter.empty_head);
 391                 ret = lttng_heap_init(&chan->iter.heap,
 392                                 num_possible_cpus(),
 393                                 GFP_KERNEL, buf_is_higher);
 394                 if (ret)
 395                         return ret;
 396                 /*
 397                  * In case of non-hotplug cpu, if the ring-buffer is allocated
 398                  * in early initcall, it will not be notified of secondary cpus.
 399                  * In that off case, we need to allocate for all possible cpus.
 400                  */
 401 #ifdef CONFIG_HOTPLUG_CPU
 402                 chan->hp_iter_notifier.notifier_call =
 403                         channel_iterator_cpu_hotplug;
 404                 chan->hp_iter_notifier.priority = 10;
 405                 register_cpu_notifier(&chan->hp_iter_notifier);
 406                 get_online_cpus();
 407                 for_each_online_cpu(cpu) {
 408                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 409                         lib_ring_buffer_iterator_init(chan, buf);
 410                 }
 411                 chan->hp_iter_enable = 1;
 412                 put_online_cpus();
 413 #else
 414                 for_each_possible_cpu(cpu) {
 415                         buf = per_cpu_ptr(chan->backend.buf, cpu);
 416                         lib_ring_buffer_iterator_init(chan, buf);
 417                 }
 418 #endif
 419         } else {
 420                 buf = channel_get_ring_buffer(config, chan, 0);
 421                 lib_ring_buffer_iterator_init(chan, buf);
 422         }
 423         return 0;
 424 }
 425
 426 void channel_iterator_unregister_notifiers(struct channel *chan)
 427 {
 428         const struct lib_ring_buffer_config *config = &chan->backend.config;
 429
 430         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 431                 chan->hp_iter_enable = 0;
 432                 unregister_cpu_notifier(&chan->hp_iter_notifier);
 433         }
 434 }
 435
 436 void channel_iterator_free(struct channel *chan)
 437 {
 438         const struct lib_ring_buffer_config *config = &chan->backend.config;
 439
 440         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 441                 lttng_heap_free(&chan->iter.heap);
 442 }
 443
 444 int lib_ring_buffer_iterator_open(struct lib_ring_buffer *buf)
 445 {
 446         struct channel *chan = buf->backend.chan;
 447         const struct lib_ring_buffer_config *config = &chan->backend.config;
 448         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 449         return lib_ring_buffer_open_read(buf);
 450 }
 451 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_open);
 452
 453 /*
 454  * Note: Iterators must not be mixed with other types of outputs, because an
 455  * iterator can leave the buffer in "GET" state, which is not consistent with
 456  * other types of output (mmap, splice, raw data read).
 457  */
 458 void lib_ring_buffer_iterator_release(struct lib_ring_buffer *buf)
 459 {
 460         lib_ring_buffer_release_read(buf);
 461 }
 462 EXPORT_SYMBOL_GPL(lib_ring_buffer_iterator_release);
 463
 464 int channel_iterator_open(struct channel *chan)
 465 {
 466         const struct lib_ring_buffer_config *config = &chan->backend.config;
 467         struct lib_ring_buffer *buf;
 468         int ret = 0, cpu;
 469
 470         CHAN_WARN_ON(chan, config->output != RING_BUFFER_ITERATOR);
 471
 472         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 473                 get_online_cpus();
 474                 /* Allow CPU hotplug to keep track of opened reader */
 475                 chan->iter.read_open = 1;
 476                 for_each_channel_cpu(cpu, chan) {
 477                         buf = channel_get_ring_buffer(config, chan, cpu);
 478                         ret = lib_ring_buffer_iterator_open(buf);
 479                         if (ret)
 480                                 goto error;
 481                         buf->iter.read_open = 1;
 482                 }
 483                 put_online_cpus();
 484         } else {
 485                 buf = channel_get_ring_buffer(config, chan, 0);
 486                 ret = lib_ring_buffer_iterator_open(buf);
 487         }
 488         return ret;
 489 error:
 490         /* Error should always happen on CPU 0, hence no close is required. */
 491         CHAN_WARN_ON(chan, cpu != 0);
 492         put_online_cpus();
 493         return ret;
 494 }
 495 EXPORT_SYMBOL_GPL(channel_iterator_open);
 496
 497 void channel_iterator_release(struct channel *chan)
 498 {
 499         const struct lib_ring_buffer_config *config = &chan->backend.config;
 500         struct lib_ring_buffer *buf;
 501         int cpu;
 502
 503         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU) {
 504                 get_online_cpus();
 505                 for_each_channel_cpu(cpu, chan) {
 506                         buf = channel_get_ring_buffer(config, chan, cpu);
 507                         if (buf->iter.read_open) {
 508                                 lib_ring_buffer_iterator_release(buf);
 509                                 buf->iter.read_open = 0;
 510                         }
 511                 }
 512                 chan->iter.read_open = 0;
 513                 put_online_cpus();
 514         } else {
 515                 buf = channel_get_ring_buffer(config, chan, 0);
 516                 lib_ring_buffer_iterator_release(buf);
 517         }
 518 }
 519 EXPORT_SYMBOL_GPL(channel_iterator_release);
 520
 521 void lib_ring_buffer_iterator_reset(struct lib_ring_buffer *buf)
 522 {
 523         struct channel *chan = buf->backend.chan;
 524
 525         if (buf->iter.state != ITER_GET_SUBBUF)
 526                 lib_ring_buffer_put_next_subbuf(buf);
 527         buf->iter.state = ITER_GET_SUBBUF;
 528         /* Remove from heap (if present). */
 529         if (lttng_heap_cherrypick(&chan->iter.heap, buf))
 530                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 531         buf->iter.timestamp = 0;
 532         buf->iter.header_len = 0;
 533         buf->iter.payload_len = 0;
 534         buf->iter.consumed = 0;
 535         buf->iter.read_offset = 0;
 536         buf->iter.data_size = 0;
 537         /* Don't reset allocated and read_open */
 538 }
 539
 540 void channel_iterator_reset(struct channel *chan)
 541 {
 542         const struct lib_ring_buffer_config *config = &chan->backend.config;
 543         struct lib_ring_buffer *buf;
 544         int cpu;
 545
 546         /* Empty heap, put into empty_head */
 547         while ((buf = lttng_heap_remove(&chan->iter.heap)) != NULL)
 548                 list_add(&buf->iter.empty_node, &chan->iter.empty_head);
 549
 550         for_each_channel_cpu(cpu, chan) {
 551                 buf = channel_get_ring_buffer(config, chan, cpu);
 552                 lib_ring_buffer_iterator_reset(buf);
 553         }
 554         /* Don't reset read_open */
 555         chan->iter.last_qs = 0;
 556         chan->iter.last_timestamp = 0;
 557         chan->iter.last_cpu = 0;
 558         chan->iter.len_left = 0;
 559 }
 560
 561 /*
 562  * Ring buffer payload extraction read() implementation.
 563  */
 564 static
 565 ssize_t channel_ring_buffer_file_read(struct file *filp,
 566                                       char __user *user_buf,
 567                                       size_t count,
 568                                       loff_t *ppos,
 569                                       struct channel *chan,
 570                                       struct lib_ring_buffer *buf,
 571                                       int fusionmerge)
 572 {
 573         const struct lib_ring_buffer_config *config = &chan->backend.config;
 574         size_t read_count = 0, read_offset;
 575         ssize_t len;
 576
 577         might_sleep();
 578         if (!access_ok(VERIFY_WRITE, user_buf, count))
 579                 return -EFAULT;
 580
 581         /* Finish copy of previous record */
 582         if (*ppos != 0) {
 583                 if (read_count < count) {
 584                         len = chan->iter.len_left;
 585                         read_offset = *ppos;
 586                         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU
 587                             && fusionmerge)
 588                                 buf = lttng_heap_maximum(&chan->iter.heap);
 589                         CHAN_WARN_ON(chan, !buf);
 590                         goto skip_get_next;
 591                 }
 592         }
 593
 594         while (read_count < count) {
 595                 size_t copy_len, space_left;
 596
 597                 if (fusionmerge)
 598                         len = channel_get_next_record(chan, &buf);
 599                 else
 600                         len = lib_ring_buffer_get_next_record(chan, buf);
 601 len_test:
 602                 if (len < 0) {
 603                         /*
 604                          * Check if buffer is finalized (end of file).
 605                          */
 606                         if (len == -ENODATA) {
 607                                 /* A 0 read_count will tell about end of file */
 608                                 goto nodata;
 609                         }
 610                         if (filp->f_flags & O_NONBLOCK) {
 611                                 if (!read_count)
 612                                         read_count = -EAGAIN;
 613                                 goto nodata;
 614                         } else {
 615                                 int error;
 616
 617                                 /*
 618                                  * No data available at the moment, return what
 619                                  * we got.
 620                                  */
 621                                 if (read_count)
 622                                         goto nodata;
 623
 624                                 /*
 625                                  * Wait for returned len to be >= 0 or -ENODATA.
 626                                  */
 627                                 if (fusionmerge)
 628                                         error = wait_event_interruptible(
 629                                           chan->read_wait,
 630                                           ((len = channel_get_next_record(chan,
 631                                                 &buf)), len != -EAGAIN));
 632                                 else
 633                                         error = wait_event_interruptible(
 634                                           buf->read_wait,
 635                                           ((len = lib_ring_buffer_get_next_record(
 636                                                   chan, buf)), len != -EAGAIN));
 637                                 CHAN_WARN_ON(chan, len == -EBUSY);
 638                                 if (error) {
 639                                         read_count = error;
 640                                         goto nodata;
 641                                 }
 642                                 CHAN_WARN_ON(chan, len < 0 && len != -ENODATA);
 643                                 goto len_test;
 644                         }
 645                 }
 646                 read_offset = buf->iter.read_offset;
 647 skip_get_next:
 648                 space_left = count - read_count;
 649                 if (len <= space_left) {
 650                         copy_len = len;
 651                         chan->iter.len_left = 0;
 652                         *ppos = 0;
 653                 } else {
 654                         copy_len = space_left;
 655                         chan->iter.len_left = len - copy_len;
 656                         *ppos = read_offset + copy_len;
 657                 }
 658                 if (__lib_ring_buffer_copy_to_user(&buf->backend, read_offset,
 659                                                &user_buf[read_count],
 660                                                copy_len)) {
 661                         /*
 662                          * Leave the len_left and ppos values at their current
 663                          * state, as we currently have a valid event to read.
 664                          */
 665                         return -EFAULT;
 666                 }
 667                 read_count += copy_len;
 668         };
 669         return read_count;
 670
 671 nodata:
 672         *ppos = 0;
 673         chan->iter.len_left = 0;
 674         return read_count;
 675 }
 676
 677 /**
 678  * lib_ring_buffer_file_read - Read buffer record payload.
 679  * @filp: file structure pointer.
 680  * @buffer: user buffer to read data into.
 681  * @count: number of bytes to read.
 682  * @ppos: file read position.
 683  *
 684  * Returns a negative value on error, or the number of bytes read on success.
 685  * ppos is used to save the position _within the current record_ between calls
 686  * to read().
 687  */
 688 static
 689 ssize_t lib_ring_buffer_file_read(struct file *filp,
 690                                   char __user *user_buf,
 691                                   size_t count,
 692                                   loff_t *ppos)
 693 {
 694         struct inode *inode = filp->f_dentry->d_inode;
 695         struct lib_ring_buffer *buf = inode->i_private;
 696         struct channel *chan = buf->backend.chan;
 697
 698         return channel_ring_buffer_file_read(filp, user_buf, count, ppos,
 699                                              chan, buf, 0);
 700 }
 701
 702 /**
 703  * channel_file_read - Read channel record payload.
 704  * @filp: file structure pointer.
 705  * @buffer: user buffer to read data into.
 706  * @count: number of bytes to read.
 707  * @ppos: file read position.
 708  *
 709  * Returns a negative value on error, or the number of bytes read on success.
 710  * ppos is used to save the position _within the current record_ between calls
 711  * to read().
 712  */
 713 static
 714 ssize_t channel_file_read(struct file *filp,
 715                           char __user *user_buf,
 716                           size_t count,
 717                           loff_t *ppos)
 718 {
 719         struct inode *inode = filp->f_dentry->d_inode;
 720         struct channel *chan = inode->i_private;
 721         const struct lib_ring_buffer_config *config = &chan->backend.config;
 722
 723         if (config->alloc == RING_BUFFER_ALLOC_PER_CPU)
 724                 return channel_ring_buffer_file_read(filp, user_buf, count,
 725                                                      ppos, chan, NULL, 1);
 726         else {
 727                 struct lib_ring_buffer *buf =
 728                         channel_get_ring_buffer(config, chan, 0);
 729                 return channel_ring_buffer_file_read(filp, user_buf, count,
 730                                                      ppos, chan, buf, 0);
 731         }
 732 }
 733
 734 static
 735 int lib_ring_buffer_file_open(struct inode *inode, struct file *file)
 736 {
 737         struct lib_ring_buffer *buf = inode->i_private;
 738         int ret;
 739
 740         ret = lib_ring_buffer_iterator_open(buf);
 741         if (ret)
 742                 return ret;
 743
 744         file->private_data = buf;
 745         ret = nonseekable_open(inode, file);
 746         if (ret)
 747                 goto release_iter;
 748         return 0;
 749
 750 release_iter:
 751         lib_ring_buffer_iterator_release(buf);
 752         return ret;
 753 }
 754
 755 static
 756 int lib_ring_buffer_file_release(struct inode *inode, struct file *file)
 757 {
 758         struct lib_ring_buffer *buf = inode->i_private;
 759
 760         lib_ring_buffer_iterator_release(buf);
 761         return 0;
 762 }
 763
 764 static
 765 int channel_file_open(struct inode *inode, struct file *file)
 766 {
 767         struct channel *chan = inode->i_private;
 768         int ret;
 769
 770         ret = channel_iterator_open(chan);
 771         if (ret)
 772                 return ret;
 773
 774         file->private_data = chan;
 775         ret = nonseekable_open(inode, file);
 776         if (ret)
 777                 goto release_iter;
 778         return 0;
 779
 780 release_iter:
 781         channel_iterator_release(chan);
 782         return ret;
 783 }
 784
 785 static
 786 int channel_file_release(struct inode *inode, struct file *file)
 787 {
 788         struct channel *chan = inode->i_private;
 789
 790         channel_iterator_release(chan);
 791         return 0;
 792 }
 793
 794 const struct file_operations channel_payload_file_operations = {
 795         .owner = THIS_MODULE,
 796         .open = channel_file_open,
 797         .release = channel_file_release,
 798         .read = channel_file_read,
 799         .llseek = vfs_lib_ring_buffer_no_llseek,
 800 };
 801 EXPORT_SYMBOL_GPL(channel_payload_file_operations);
 802
 803 const struct file_operations lib_ring_buffer_payload_file_operations = {
 804         .owner = THIS_MODULE,
 805         .open = lib_ring_buffer_file_open,
 806         .release = lib_ring_buffer_file_release,
 807         .read = lib_ring_buffer_file_read,
 808         .llseek = vfs_lib_ring_buffer_no_llseek,
 809 };
 810 EXPORT_SYMBOL_GPL(lib_ring_buffer_payload_file_operations);