On 3/28/19 11:48 AM, Christoph Hellwig wrote: > On Thu, Mar 28, 2019 at 05:31:29PM +0100, Johannes Thumshirn wrote: >> In all cases I could reproduce it I had nsegs == 2 and >> rq->nr_phys_segments == 1 (rq->bio->bi_phys_segments == 1). > > And all cases are pass through as well? > > Can you check if this still shows up with the series from Ming which > switches the passthrough requests to use multi-page biovecs aswell > and gets rid of the hairy code you quote below. Since I already merged that in my 5.2 but didn't push it out yet, I've made a temporary branch available if you want to test through that: git://git.kernel.dk/linux-block 5.2-tmp -- Jens Axboe
