On Sat, Mar 18, 2017 at 1:39 AM, Bart Van Assche
[off-list ref] wrote:

On Fri, 2017-03-17 at 17:57 +0800, Ming Lei wrote:

quoted

+/*
+ * When we reach here because queue is busy, REQ_ATOM_COMPLETE
+ * flag isn't set yet, so there may be race with timeout hanlder,
+ * but given rq->deadline is just set in .queue_rq() under
+ * this sitation, the race won't be possible in reality because
+ * rq->timeout should be set as big enough to cover the window
+ * between blk_mq_start_request() called from .queue_rq() and
+ * clearing REQ_ATOM_STARTED here.
+ */
 static void __blk_mq_requeue_request(struct request *rq)
 {
      struct request_queue *q = rq->q;

@@ -700,6 +709,19 @@ static void blk_mq_check_expired(struct blk_mq_hw_ctx *hctx,
      if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
              return;

+     /*
+      * The rq being checked may have been freed and reallocated
+      * out already here, we avoid this race by checking rq->deadline
+      * and REQ_ATOM_COMPLETE flag together:
+      *
+      * - if rq->deadline is observed as new value because of
+      *   reusing, the rq won't be timed out because of timing.
+      * - if rq->deadline is observed as previous value,
+      *   REQ_ATOM_COMPLETE flag won't be cleared in reuse path
+      *   because we put a barrier between setting rq->deadline
+      *   and clearing the flag in blk_mq_start_request(), so
+      *   this rq won't be timed out too.
+      */
      if (time_after_eq(jiffies, rq->deadline)) {
              if (!blk_mark_rq_complete(rq))
                      blk_mq_rq_timed_out(rq, reserved);

Since this explanation applies to the same race addressed by patch 1/3,

First, this explains how we deal with the race of reuse vs. timeout, and 1/3
fixes another race or rq corruption. Did you see anywhere I mentioned STARTED
flag in above comment?

In case of 1/3, the rq to be dispatched can be destroyed simply by the
blk_mq_end_request() from timeout. Or even it can survive, the same rq
can be allocated into another I/O path, and this situation is different with
reuse vs. timeout. And I can't see any help from the comment for explaining
1/3's issue, can you? Maybe I need to mention rq corruption in 1/3 explicitly.

Secondly introducing this comment to 1/3 just causes unnecessary
backporting burden, as we have to make it into -stable.

please consider squashing this patch into patch 1/3.

So please do not consider that.

Thanks,
Ming Lei