On Wed, Sep 29, 2021 at 09:07:33AM +0200, Christoph Hellwig wrote: > On Tue, Sep 28, 2021 at 10:32:51PM +0800, Ming Lei wrote: > > Hello Christoph, > > > > On Tue, Sep 28, 2021 at 07:22:06AM +0200, Christoph Hellwig wrote: > > > Hi Jens, > > > > > > this series removes the rq_disk field in struct request, which isn't > > > needed now that we can get the disk from the request_queue. > > > > Can we hold on this series until q->disk becomes really reliable[1][2]? > > It's not like q->disk isn't reliable. It is that we don't kill all bios q->disk will be cleaned in gendisk's release handler, which may happen before or after blk_cleanup_queue(). > when tearing down the gendisk, which is an old problem that got worse. It is fine to not kill all bios when tearing down gendisk if BDI isn't retrieved from q->disk->bdi since gendisk has same lifetime with bdev. So what is the old problem? > That being said I'm resending this again. No, either your patch or mine can't fix the issue, and I will comment on your resending. Thanks, Ming
