rtuDataOperation.c 5.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180
  1. /*
  2. * Functions to sequence FLUSH and FUA writes.
  3. *
  4. * Copyright (C) 2011 Max Planck Institute for Gravitational Physics
  5. * Copyright (C) 2011 Tejun Heo <tj@kernel.org>
  6. *
  7. * This file is released under the GPLv2.
  8. *
  9. * REQ_{FLUSH|FUA} requests are decomposed to sequences consisted of three
  10. * optional steps - PREFLUSH, DATA and POSTFLUSH - according to the request
  11. * properties and hardware capability.
  12. *
  13. * If a request doesn't have data, only REQ_FLUSH makes sense, which
  14. * indicates a simple flush request. If there is data, REQ_FLUSH indicates
  15. * that the device cache should be flushed before the data is executed, and
  16. * REQ_FUA means that the data must be on non-volatile media on request
  17. * completion.
  18. *
  19. * If the device doesn't have writeback cache, FLUSH and FUA don't make any
  20. * difference. The requests are either completed immediately if there's no
  21. * data or executed as normal requests otherwise.
  22. *
  23. * If the device has writeback cache and supports FUA, REQ_FLUSH is
  24. * translated to PREFLUSH but REQ_FUA is passed down directly with DATA.
  25. *
  26. * If the device has writeback cache and doesn't support FUA, REQ_FLUSH is
  27. * translated to PREFLUSH and REQ_FUA to POSTFLUSH.
  28. *
  29. * The actual execution of flush is double buffered. Whenever a request
  30. * needs to execute PRE or POSTFLUSH, it queues at
  31. * q->flush_queue[q->flush_pending_idx]. Once certain criteria are met, a
  32. * flush is issued and the pending_idx is toggled. When the flush
  33. * completes, all the requests which were pending are proceeded to the next
  34. * step. This allows arbitrary merging of different types of FLUSH/FUA
  35. * requests.
  36. *
  37. * Currently, the following conditions are used to determine when to issue
  38. * flush.
  39. *
  40. * C1. At any given time, only one flush shall be in progress. This makes
  41. * double buffering sufficient.
  42. *
  43. * C2. Flush is deferred if any request is executing DATA of its sequence.
  44. * This avoids issuing separate POSTFLUSHes for requests which shared
  45. * PREFLUSH.
  46. *
  47. * C3. The second condition is ignored if there is a request which has
  48. * waited longer than FLUSH_PENDING_TIMEOUT. This is to avoid
  49. * starvation in the unlikely case where there are continuous stream of
  50. * FUA (without FLUSH) requests.
  51. *
  52. * For devices which support FUA, it isn't clear whether C2 (and thus C3)
  53. * is beneficial.
  54. *
  55. * Note that a sequenced FLUSH/FUA request with DATA is completed twice.
  56. * Once while executing DATA and again after the whole sequence is
  57. * complete. The first completion updates the contained bio but doesn't
  58. * finish it so that the bio submitter is notified only after the whole
  59. * sequence is complete. This is implemented by testing REQ_FLUSH_SEQ in
  60. * req_bio_endio().
  61. *
  62. * The above peculiarity requires that each FLUSH/FUA request has only one
  63. * bio attached to it, which is guaranteed as they aren't allowed to be
  64. * merged in the usual way.
  65. */
  66. #include <linux/kernel.h>
  67. #include <linux/module.h>
  68. #include <linux/bio.h>
  69. #include <linux/blkdev.h>
  70. #include <linux/gfp.h>
  71. #include "blk.h"
  72. /* FLUSH/FUA sequences */
  73. enum {
  74. REQ_FSEQ_PREFLUSH = (1 << 0), /* pre-flushing in progress */
  75. REQ_FSEQ_DATA = (1 << 1), /* data write in progress */
  76. REQ_FSEQ_POSTFLUSH = (1 << 2), /* post-flushing in progress */
  77. REQ_FSEQ_DONE = (1 << 3),
  78. REQ_FSEQ_ACTIONS = REQ_FSEQ_PREFLUSH | REQ_FSEQ_DATA |
  79. REQ_FSEQ_POSTFLUSH,
  80. /*
  81. * If flush has been pending longer than the following timeout,
  82. * it's issued even if flush_data requests are still in flight.
  83. */
  84. FLUSH_PENDING_TIMEOUT = 5 * HZ,
  85. };
  86. static bool blk_kick_flush(struct request_queue *q);
  87. static unsigned int blk_flush_policy(unsigned int fflags, struct request *rq)
  88. {
  89. unsigned int policy = 0;
  90. if (blk_rq_sectors(rq))
  91. policy |= REQ_FSEQ_DATA;
  92. if (fflags & REQ_FLUSH) {
  93. if (rq->cmd_flags & REQ_FLUSH)
  94. policy |= REQ_FSEQ_PREFLUSH;
  95. if (!(fflags & REQ_FUA) && (rq->cmd_flags & REQ_FUA))
  96. policy |= REQ_FSEQ_POSTFLUSH;
  97. }
  98. return policy;
  99. }
  100. static unsigned int blk_flush_cur_seq(struct request *rq)
  101. {
  102. return 1 << ffz(rq->flush.seq);
  103. }
  104. static void blk_flush_restore_request(struct request *rq)
  105. {
  106. /*
  107. * After flush data completion, @rq->bio is %NULL but we need to
  108. * complete the bio again. @rq->biotail is guaranteed to equal the
  109. * original @rq->bio. Restore it.
  110. */
  111. rq->bio = rq->biotail;
  112. /* make @rq a normal request */
  113. rq->cmd_flags &= ~REQ_FLUSH_SEQ;
  114. rq->end_io = rq->flush.saved_end_io;
  115. }
  116. /**
  117. * blk_flush_complete_seq - complete flush sequence
  118. * @rq: FLUSH/FUA request being sequenced
  119. * @seq: sequences to complete (mask of %REQ_FSEQ_*, can be zero)
  120. * @error: whether an error occurred
  121. *
  122. * @rq just completed @seq part of its flush sequence, record the
  123. * completion and trigger the next step.
  124. *
  125. * CONTEXT:
  126. * spin_lock_irq(q->queue_lock)
  127. *
  128. * RETURNS:
  129. * %true if requests were added to the dispatch queue, %false otherwise.
  130. */
  131. static bool blk_flush_complete_seq(struct request *rq, unsigned int seq,
  132. int error)
  133. {
  134. struct request_queue *q = rq->q;
  135. struct list_head *pending = &q->flush_queue[q->flush_pending_idx];
  136. bool queued = false;
  137. BUG_ON(rq->flush.seq & seq);
  138. rq->flush.seq |= seq;
  139. if (likely(!error))
  140. seq = blk_flush_cur_seq(rq);
  141. else
  142. seq = REQ_FSEQ_DONE;
  143. switch (seq) {
  144. case REQ_FSEQ_PREFLUSH:
  145. case REQ_FSEQ_POSTFLUSH:
  146. /* queue for flush */
  147. if (list_empty(pending))
  148. q->flush_pending_since = jiffies;
  149. list_move_tail(&rq->flush.list, pending);
  150. break;
  151. case REQ_FSEQ_DATA:
  152. list_move_tail(&rq->flush.list, &q->flush_data_in_flight);
  153. list_add(&rq->queuelist, &q->queue_head);
  154. queued = true;
  155. break;
  156. case REQ_FSEQ_DONE:
  157. /*
  158. * @rq was previously adjusted by blk_flush_issue() for
  159. * flush sequencing and may already have gone through the
  160. * flush data request completion path. Restore @rq for
  161. * normal completion and end it.
  162. */