On Wed 06-04-22 18:03:59, Jan Kara wrote: > Hello! > > On Wed 06-04-22 09:44:55, yukuai (C) wrote: > > 在 2022/04/04 22:35, Paolo Valente 写道: > > > This seems to have to do with Jan's patches on tag allocation. I'm CCing him too. Jan, I'm willing to provide my usual dev version for testing, if useful. > > > > I had observed this warning for a long time, and it's very to reporduce > > by our reporducer for the problem that Jan fixed recently. > > > > I thougt this warning is due to bfqq is associated with wrong cgroup, > > which should be fixed by Jan's patch. > > > > What do you think, Jan ? > > Hum, the warning suggests that the blkcg's idea of cgroup hierarchy is not in > sync with BFQ's idea of cgroup hierarchy. One possibility how this could > happen is if bfqq is moved from one bfqg to another between the depth check > and the for_each_entity() loop. This should actually get fixed as a > side-effect of my patches because they change the logic so that alive bfqq > can change parent only if they contain a single process. But nevertheless > we should make the logic inside bfqq_request_over_limit() more robust in > place of process moving between cgroups. So I'll send a patch for that on > top of my patches. Posted: https://lore.kernel.org/all/20220407140738.9723-1-jack@xxxxxxx Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
