On 5/16/20 12:27 PM, Christoph Hellwig wrote: > Hi Jens, > > the way we track reference on q_usage_counter is a little weird at the > moment, in that we often have to grab another reference in addition to > the current one. This small series reshuffles that to avoid the extra > references in the normal I/O path. > > Changes since v2: > - increase the q_usage_counter critical section a bit in > blk_mq_alloc_request_hctx > > Changes since v1: > - rebased to the lastest for-5.8/block tree with the blk-crypt addition Applied, thanks. -- Jens Axboe