On 2/14/25 6:18 PM, Linus Torvalds wrote:
Adding more people: Peter / Phil / Waiman. Juri was already on the list earlier.
On Fri, 14 Feb 2025 at 02:12, Holger Hoffstätte
<holger@xxxxxxxxxxxxxxxxxxxxxx> wrote:
Whoop! Whoop! The sound of da police!
2ce2a62881abcd379b714bf41aa671ad7657bdd2 is the first bad commit
commit 2ce2a62881abcd379b714bf41aa671ad7657bdd2 (HEAD)
Author: Juri Lelli <juri.lelli@xxxxxxxxxx>
Date: Fri Nov 15 11:48:29 2024 +0000
sched/deadline: Check bandwidth overflow earlier for hotplug
[ Upstream commit 53916d5fd3c0b658de3463439dd2b7ce765072cb ]
With this reverted it reliably suspends again.
Can you check that it works (or - more likely - doesn't work) in upstream?
That commit 53916d5fd3c0 ("sched/deadline: Check bandwidth overflow
earlier for hotplug") got merged during the current merge window, so
it would be lovely if you can check whether current -git (or just the
latest 6.14-rc) works for you, or has the same breakage.
Background for new people on the participants list: original report at
https://lore.kernel.org/all/e7096ec2-68db-fc3e-9c48-f20d3e80df72@xxxxxxxxxxxxxxxxxxxxxx/
which says
Common symptom on all machines seems to be
[ +0.000134] Disabling non-boot CPUs ...
[ +0.000072] Error taking CPU15 down: -16
[ +0.000002] Non-boot CPUs are not disabled
and this bisection result is from
https://lore.kernel.org/all/9a44f314-c101-4ed1-98ad-547c84df7cdd@xxxxxxxxxxxxxxxxxxxxxx/
and if it breaks in 6.13 -stable, I would expect the same in the
current tree. Unless there's some non-obvious interaction with
something else ?
Commit 53916d5fd3c0 ("sched/deadline: Check bandwidth overflow earlier
for hotplug") is the last patch of the 3 patch series.
1) commit 41d4200b7103 ("sched/deadline: Restore dl_server bandwidth
on non-destructive root domain changes")
2) commit d4742f6ed7ea ("sched/deadline: Correctly account for
allocated bandwidth during hotplug")
3) commit 53916d5fd3c0 ("sched/deadline: Check bandwidth overflow
earlier for hotplug")
It looks like 6.13.3-rc1 has patches 2 and 3, but not patch 1. It is
possible that patch 3 has a dependency on patch 1. My suggestion is to
either take patch 1 as well or none of them.
Cheers,
Longman
