On Tue, Feb 18, 2025 at 11:59 AM Keith Busch <kbusch@xxxxxxxxxx> wrote: > > On Fri, Feb 14, 2025 at 12:38:54PM -0800, Caleb Sander Mateos wrote: > > On Fri, Feb 14, 2025 at 7:45 AM Keith Busch <kbusch@xxxxxxxx> wrote: > > > + > > > + nr_bvecs = blk_rq_nr_phys_segments(rq); > > > > Is this guaranteed to match the number of bvecs in the request? > > Yes. > > > Wouldn't the number of physical segments depend on how the block > > device splits the bvecs? > > Also yes. > > >lo_rw_aio() uses rq_for_each_bvec() to count > > the number of bvecs, for example. > > Hm, that seems unnecessary. The request's nr_phys_segments is > initialized to the number of bvecs rather than page segments, so it can > be used instead of recounting them from a given struct request. > > The initial number of physical segments for a request is set in > bio_split_rw_at(), which uses bio_for_each_bvec(). That's what > rq_for_each_bvec would use, too. The same is used for any bio's that get > merged into the bio. Okay, thanks for verifying! Best, Caleb
