Re: [syzbot] [mm?] INFO: rcu detected stall in kvm_vm_release (4)

[Andrew Morton]

On Fri, 21 Aug 2026 09:06:48 +0800 Ye Liu <ye.liu@xxxxxxxxx> wrote:

> > : The combination of debug tools: KASAN quarantine processing + ORC
> > : unwinder + page_owner stack tracing executed so slowly on the CPU that
> > : it exceeded the RCU stall threshold.
> > 
> > (I think we've recently seen the same thing, different callers)
> > 
> > I can't say I'm terribly motivated.  page_owner is expensive and if you
> > run heavy stress testing with page_owner enabled then such things
> > aren't surprising.
> > 
> > Let me Cc a few page_owner and KASAN people, see if they feel
> > differently.
> 
> Adding cond_resched() to the qlist_free_all() loop would fix this .
> kasan_quarantine_remove_cache() already dose the same in its scanning loop.

I don't think cond_resched() does what we want for RCU stalls.

And cond_resched() is a no-op without CONFIG_PREEMPT_NONE, which wasn't
set here.