2020-11-20 11:52:38

by Peter Zijlstra

[permalink] [raw]
Subject: [PATCH 0/2] More RCU vs idle fixes


Both arm64 and s390 are tripping over arch_cpu_idle() RCU,tracing,lockdep
interaction. While looking at that I also found fail in inte_idle.

Please consider for this cycle.


2020-11-24 14:23:48

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH 0/2] More RCU vs idle fixes

On Tue, Nov 24, 2020 at 02:47:27PM +0100, Sven Schnelle wrote:
> Peter Zijlstra <[email protected]> writes:
>
> > Both arm64 and s390 are tripping over arch_cpu_idle() RCU,tracing,lockdep
> > interaction. While looking at that I also found fail in inte_idle.
> >
> > Please consider for this cycle.
>
> Is anyone taking this patchset?

I think I'll stuff it in x86/urgent for lack of a better place.

> For s390, we also need to change the
> local_irq_safe/restore to the raw variants in enabled_wait() in
> arch/s390/kernel/idle.c.

Duh, I'll add that.



2020-11-24 22:18:20

by Sven Schnelle

[permalink] [raw]
Subject: Re: [PATCH 0/2] More RCU vs idle fixes

Peter Zijlstra <[email protected]> writes:

> Both arm64 and s390 are tripping over arch_cpu_idle() RCU,tracing,lockdep
> interaction. While looking at that I also found fail in inte_idle.
>
> Please consider for this cycle.

Is anyone taking this patchset? For s390, we also need to change the
local_irq_safe/restore to the raw variants in enabled_wait() in
arch/s390/kernel/idle.c. I can make a patch and carry that via the
s390 tree, but i want to make sure the s390 change in this patchset
also reaches linus' tree.

Thanks
Sven

2020-11-25 02:08:48

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH 0/2] More RCU vs idle fixes

On Tue, Nov 24, 2020 at 03:18:53PM +0100, Peter Zijlstra wrote:
> On Tue, Nov 24, 2020 at 02:47:27PM +0100, Sven Schnelle wrote:
> > Peter Zijlstra <[email protected]> writes:
> >
> > > Both arm64 and s390 are tripping over arch_cpu_idle() RCU,tracing,lockdep
> > > interaction. While looking at that I also found fail in inte_idle.
> > >
> > > Please consider for this cycle.
> >
> > Is anyone taking this patchset?
>
> I think I'll stuff it in x86/urgent for lack of a better place.

Ah, locking/urgent might be a better place I suppose.