On Tue, 2019-04-02 at 08:53 +-0800, Ming Lei wrote: +AD4 On Mon, Apr 01, 2019 at 02:20:12PM -0700, Bart Van Assche wrote: +AD4 +AD4 diff --git a/block/blk-mq.c b/block/blk-mq.c +AD4 +AD4 index 3ff3d7b49969..652d0c6d5945 100644 +AD4 +AD4 --- a/block/blk-mq.c +AD4 +AD4 +-+-+- b/block/blk-mq.c +AD4 +AD4 +AEAAQA -1499,12 +-1499,20 +AEAAQA void blk+AF8-mq+AF8-run+AF8-hw+AF8-queues(struct request+AF8-queue +ACo-q, bool async) +AD4 +AD4 struct blk+AF8-mq+AF8-hw+AF8-ctx +ACo-hctx+ADs +AD4 +AD4 int i+ADs +AD4 +AD4 +AD4 +AD4 +- /+ACo +AD4 +AD4 +- +ACo Do not run any hardware queues if the queue is frozen or if a +AD4 +AD4 +- +ACo concurrent blk+AF8-cleanup+AF8-queue() call is removing any data +AD4 +AD4 +- +ACo structures used by this function. +AD4 +AD4 +- +ACo-/ +AD4 +AD4 +- if (+ACE-percpu+AF8-ref+AF8-tryget(+ACY-q-+AD4-q+AF8-usage+AF8-counter)) +AD4 +AD4 +- return+ADs +AD4 +AD4 queue+AF8-for+AF8-each+AF8-hw+AF8-ctx(q, hctx, i) +AHs +AD4 +AD4 if (blk+AF8-mq+AF8-hctx+AF8-stopped(hctx)) +AD4 +AD4 continue+ADs +AD4 +AD4 +AD4 +AD4 blk+AF8-mq+AF8-run+AF8-hw+AF8-queue(hctx, async)+ADs +AD4 +AD4 +AH0 +AD4 +AD4 +- percpu+AF8-ref+AF8-put(+ACY-q-+AD4-q+AF8-usage+AF8-counter)+ADs +AD4 +AD4 +AH0 +AD4 +AD4 EXPORT+AF8-SYMBOL(blk+AF8-mq+AF8-run+AF8-hw+AF8-queues)+ADs +AD4 +AD4 I don't see it is necessary to add percpu+AF8-ref+AF8-tryget()/percpu+AF8-ref+AF8-put() +AD4 in the fast path if we simply release all hctx resource in hctx's +AD4 release handler by the following patch: +AD4 +AD4 https://lore.kernel.org/linux-block/20190401044247.29881-2-ming.lei+AEA-redhat.com/T/+ACM-u +AD4 +AD4 Even we can kill the percpu+AF8-ref+AF8-tryget+AF8-live()/percpu+AF8-ref+AF8-put() in +AD4 scsi+AF8-end+AF8-request(). The above approach has the advantages of being easy to review and to maintain. Patch +ACIAWw-PATCH V2 1/3+AF0 blk-mq: free hw queue's resource in hctx's release handler+ACI makes the block layer more complicated because it introduces a new state for hardware queues: block driver cleanup has happened (set-+AD4-ops-+AD4-exit+AF8-hctx(...)) but the hardware queues are still in use by the block layer core. Let's see what other reviewers think. Bart.
