Re: [PATCH -next v1 05/12] srcu: Don't queue workqueue handlers to never-online CPUs

[Boqun Feng]

On Sun, May 17, 2026 at 09:11:10PM +0200, Thorsten Leemhuis wrote:
> On 5/17/26 20:03, Paul E. McKenney wrote:
> > On Sun, May 17, 2026 at 02:47:55PM +0200, Uladzislau Rezki wrote:
> >> On Wed, May 13, 2026 at 12:38:01PM +0200, Thorsten Leemhuis wrote:
> >>> On 5/11/26 19:54, Uladzislau Rezki (Sony) wrote:
> >>>> From: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> >>>>
> >>>> While an srcu_struct structure is in the midst of switching from CPU-0
> >>>> to all-CPUs state, it can attempt to invoke callbacks for CPUs that
> >>>> have never been online.  Worse yet, it can attempt in invoke callbacks
> >>>> for CPUs that never will be online, even including imaginary CPUs not in
> >>>> cpu_possible_mask.  This can cause hangs on s390,
> >>>
> >>> Uladzislau, Paul, according to the fixes tag below this change fixes a
> >>> change that went into 7.0-rc6 -- and apparently causes a "hang" on some
> >>> architectures. So shouldn't this be heading to mainline instead of
> >>> -next? Ideally with a stable tag to ensure backporting to 7.0.y, but
> >>> that is a separate decision?
> >>>

FWIW, I'm happy to create an urgent PR to Linus with this fix. Since it
fixes a commit in the previous urgent fix PR from me. Ulad, let me know
if you want me to do it.

Regards,
Boqun

> >> I assume it will be picked to stable
> 
> Thx for the reply. This answered the second aspect I raised. What about
> the first: shouldn't this be heading to mainline this cycle, ideally
> quite soon? Or is this the plan and I misinterpreted the "-next" in the
> subject?
> 
> Sadly Jiri (now CCed) afaics didn't yet report if this fix helped to fix
> the issues they were seeing, but it sounded a lot like it was likely:
> https://lore.kernel.org/all/f0351414-41af-4f74-899c-87fb5bb45621@xxxxxxxxxx/
> 
> >> since there is a "Fixes:" tag. From
> >> the other hand i can explicitly update commit message with Cc: stable@xxxxxxxxxxxxxxx
> >>
> >> Paul, any thoughts?
> > 
> > The "Fixes" tag should cover it.
> 
> Often it does, yes, but it might be silently dropped for one reason or
> another. So in the end only a stable tag will guarantee backporting
> attempt by the stable team.
> 
> Site note: I recently also tried to persuade Greg to also take those
> with a stable tag first in case there is a huge backlog.
> 
> Ciao, Thorsten