From: hch@lst.de (Christoph Hellwig) Date: Sat, 27 Apr 2019 07:54:53 +0200 Subject: [PATCH V7 9/9] nvme: hold request queue's refcount in ns's whole lifetime In-Reply-To: <20190426224542.GD31470@ming.t460p> References: <20190424110221.17435-1-ming.lei@redhat.com> <20190424110221.17435-10-ming.lei@redhat.com> <20190424162746.GE23854@lst.de> <20190425010030.GD22636@ming.t460p> <20190426151114.GB20438@lst.de> <20190426224542.GD31470@ming.t460p> Message-ID: <20190427055453.GA21483@lst.de> On Sat, Apr 27, 2019@06:45:43AM +0800, Ming Lei wrote: > Still no difference between your suggestion and the way in this patch, given > driver specific change is a must. Even it is more clean to hold the > queue refocunt by drivers explicitly because we usually do the get/put pair > in one place, instead that getting refcnt in one subsystem, and doing put in > another place. > > I am going to drop this patch in V8 since my original plan is to fix block > layer's race in this patchset. Doing the untegistration and final reference drop is just a bad pattern, leading to possible bugs. I'm fine with you just dropping it for now, I guess I'll just send a scripted conversion moving the put to the drivers to Jens after -rc1 to sort it out. Probably including a rename of blk_cleanup_queue so that everyone notices the difference.