On 4/10/25 7:53 PM, Ming Lei wrote: > > I just sent out the whole patchset, which did one thing basically: move kobject > & debugfs & cpuhp out of queue freezing, then no any lockdep is observed in my > test VM after running blktests of './check block/'. Sure, I think our emails crossed over. I will review your changes. > > So the point is _not_ related with elevator lock or its order with > freeze lock. What matters is actually "do not connect freeze lock with > other subsystem(debugfs, sysfs, cpuhp, ...)", because freeze_lock relies > on fs_reclaim directly with commit ffa1e7ada456, but other subsystem easily > depends on fs_reclaim again. > > I did have patches to follow your idea to reorder elevator lock vs. freeze > lock, but it didn't make difference, even though after I killed most of > elevator_lock. > > Finally I realized the above point. Ok great:) Thanks, --Nilay
