On Thu, 2018-12-20 at 15:50 -0700, Jens Axboe wrote: +AD4 +-static void blk+AF8-fq+AF8-rcu+AF8-free(struct work+AF8-struct +ACo-work) +AD4 +-+AHs +AD4 +- struct blk+AF8-flush+AF8-queue +ACo-fq +AD0 container+AF8-of(to+AF8-rcu+AF8-work(work), +AD4 +- struct blk+AF8-flush+AF8-queue, +AD4 +- rcu+AF8-work)+ADs +AD4 +- +AD4 +- kfree(fq-+AD4-flush+AF8-rq)+ADs +AD4 +- kfree(fq)+ADs +AD4 +-+AH0 +AD4 +- +AD4 void blk+AF8-free+AF8-flush+AF8-queue(struct blk+AF8-flush+AF8-queue +ACo-fq) +AD4 +AHs +AD4 /+ACo bio based request queue hasn't flush queue +ACo-/ +AD4 if (+ACE-fq) +AD4 return+ADs +AD4 +AD4 - kfree(fq-+AD4-flush+AF8-rq)+ADs +AD4 - kfree(fq)+ADs +AD4 +- INIT+AF8-RCU+AF8-WORK(+ACY-fq-+AD4-rcu+AF8-work, blk+AF8-fq+AF8-rcu+AF8-free)+ADs +AD4 +- queue+AF8-rcu+AF8-work(system+AF8-wq, +ACY-fq-+AD4-rcu+AF8-work)+ADs +AD4 +AH0 Can INIT+AF8-RCU+AF8-WORK() +- queue+AF8-rcu+AF8-work() be changed into call+AF8-rcu()? The latter namely uses a smaller data structure. +AD4 diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c +AD4 index 2089c6c62f44..c39b58391ae8 100644 +AD4 --- a/block/blk-mq-tag.c +AD4 +-+-+- b/block/blk-mq-tag.c +AD4 +AEAAQA -228,13 +-228,15 +AEAAQA static bool bt+AF8-iter(struct sbitmap +ACo-bitmap, unsigned int bitnr, void +ACo-data) +AD4 +AD4 if (+ACE-reserved) +AD4 bitnr +-+AD0 tags-+AD4-nr+AF8-reserved+AF8-tags+ADs +AD4 - rq +AD0 tags-+AD4-rqs+AFs-bitnr+AF0AOw +AD4 +- if (tags-+AD4-rqs+AFs-bitnr+AF0.queue +ACEAPQ hctx-+AD4-queue) +AD4 +- return true+ADs Since blk+AF8-mq+AF8-tag+AF8-set+AF8-rq() is not serialized against this function I doubt that the tags-+AD4-rqs+AFs-bitnr+AF0.queue +ACEAPQ hctx-+AD4-queue check helps. Can it be left out? +AD4 +-struct rq+AF8-tag+AF8-entry +AHs +AD4 +- struct request+AF8-queue +ACo-queue+ADs +AD4 +- struct request +ACo-rq+ADs If the new test can be left out from bt+AF8-iter(), can this new data structure be left out too? In other words, keep the existing approach of only storing the request pointer and not the queue pointer. Thanks, Bart.
