Re: [RFC][PATCH] sched,livepatch: Untangle cond_resched() and live-patching
From: Peter Zijlstra
Date: Wed Mar 26 2025 - 10:49:28 EST
On Wed, Mar 26, 2025 at 03:37:50PM +0100, Miroslav Benes wrote:
> If I remember correctly, we had something like this in the old kGraft
> implementation of the live patching (SUSE way). We exactly had a hook
> somewhere in the kthread freezing code. This looks much cleaner and as far
> as I know the fridge went through improvements recently.
Yeah, I rewrote it a while ago :-)
> Peter, so that I understand it correctly... we would rely on all kthreads
> becoming freezable eventually so that both suspend and livepatch benefit.
> Is that what you meant by the above?
Well, IIRC (its been a while already) all kthreads should have a
FREEZABLE already. Things like suspend-to-idle don't hit the hotplug
path at all anymore and everything must freeze, otherwise they fail.
I was more meaning the time-to-freeze; if some kthreads take a long time
to freeze/patch then this would want improving on both ends.