Re: [PATCH 04/19] sched: Prepare for Core-wide rq->lock

[Josh Don]

On Thu, Apr 29, 2021 at 2:09 PM Don Hiatt <dhiatt@xxxxxxxxxxxxxxxx> wrote:
>
> On Thu, Apr 29, 2021 at 1:48 PM Josh Don <joshdon@xxxxxxxxxx> wrote:
> >
> > On Wed, Apr 28, 2021 at 9:41 AM Don Hiatt <dhiatt@xxxxxxxxxxxxxxxx> wrote:
> > >
> > > I'm still seeing hard lockups while repeatedly setting cookies on qemu
> > > processes even with
> > > the updated patch. If there is any debug you'd like me to turn on,
> > > just let me know.
> > >
> > > Thanks!
> > >
> > > don
> >
> > Thanks for the added context on your repro configuration. In addition
> > to the updated patch from earlier, could you try the modification to
> > double_rq_lock() from
> > https://lkml.kernel.org/r/CABk29NuS-B3n4sbmavo0NDA1OCCsz6Zf2VDjjFQvAxBMQoJ_Lg@xxxxxxxxxxxxxx
> > ? I have a feeling this is what's causing your lockup.
> >
> > Best,
> > Josh
>
> Hi Josh,
>
> I've been running Aubrey+Peter's patch (attached) for almost 5 hours
> and haven't had a single issue. :)
>
> I'm running a set-cookie script every 5 seconds on the two VMs (each
> vm is running
> 'sysbench --threads=1 --time=0 cpu run' to generate some load in the vm) and
> I'm running two of the same sysbench runs on the HV while setting cookies
> every 5 seconds.
>
> Unless I jinxed us it looks like a great fix. :)
>
> Let me know if there is anything else you'd like me to try. I'm going
> to leave the tests running
> and see what happens. I update with what I find.
>
> Thanks!
>
> don

That's awesome news, thanks for validating. Note that with Aubrey's
patch there is still a race window if sched core is being
enabled/disabled (ie. if you alternate between there being some
cookies in the system, and no cookies). In my reply I posted an
alternative version to avoid that. If your script were to do the
on-off flipping with the old patch, you'd might eventually see another
lockup.