Re: [PATCH v8 6/8] cgroup/cpuset: Change cpuset_rwsem and hotplug lock order

[Peter Zijlstra]

On Mon, Jul 01, 2019 at 08:52:33AM +0200, Juri Lelli wrote:
> Hi,
> 
> On 28/06/19 15:03, Peter Zijlstra wrote:
> > On Fri, Jun 28, 2019 at 10:06:16AM +0200, Juri Lelli wrote:
> > > cpuset_rwsem is going to be acquired from sched_setscheduler() with a
> > > following patch. There are however paths (e.g., spawn_ksoftirqd) in
> > > which sched_scheduler() is eventually called while holding hotplug lock;
> > > this creates a dependecy between hotplug lock (to be always acquired
> > > first) and cpuset_rwsem (to be always acquired after hotplug lock).
> > > 
> > > Fix paths which currently take the two locks in the wrong order (after
> > > a following patch is applied).
> > > Signed-off-by: Juri Lelli <juri.lelli@xxxxxxxxxx>
> > 
> > This all reminds me of this:
> > 
> >   https://lkml.kernel.org/r/1510755615-25906-1-git-send-email-prsood@xxxxxxxxxxxxxx
> > 
> > Which sadly got reverted again. If we do this now (I've always been a
> > proponent), then we can make that rebuild synchronous again, which
> > should also help here IIRC.
> 
> Why was that reverted? Perf regression of some type?

IIRC TJ figured it wasn't strictly required to fix the lock invertion at
that time and they sorted it differently. If I (re)read the thread
correctly the other day, he didn't have fundamental objections against
it, but wanted the simpler fix.