On Wed, Dec 18, 2024 at 05:50:05PM +0100, Frederic Weisbecker wrote:
> 5c0930ccaad5 ("hrtimers: Push pending hrtimers away from outgoing CPU
> earlier")
> was introduced to fix stalls with scheduler bandwidth timers getting
> migrated while some kthreads handling CPU hotplug rely on bandwidth.
>
> However this has introduced several other issues which used to be
> confined to RCU. But not anymore as it is spreading to hotplug code
> itself
> (https://lore.kernel.org/all/20241213203739.1519801-1-usamaarif...@gmail.com/)
>
> Instead of introducing yet another new hackery, fix the problem in
> hrtimers for everyone.
The good news is that this passes 12 hours of 400*TREE03. (Yay!!!)
The so-so news is that this gives only about 70% confidence that these
patches help, but on the other hand, it also gives much higher confidence
that these patches are not hurting anything.
At least for TREE03.
The not-so-good news is that this series causes build failures for
rcutorture scenarios (such as SRCU-T) that build with CONFIG_SMP=n:
------------------------------------------------------------------------
kernel/time/hrtimer.c: In function ‘enqueue_hrtimer_offline’:
kernel/time/hrtimer.c:1229:42: error: ‘migration_base’ undeclared (first use in
this function); did you mean ‘is_migration_base’?
------------------------------------------------------------------------
When built with KCSAN enabled (--kcsan to kvm.sh), there is this
additional build failure on that same line of code:
------------------------------------------------------------------------
kernel/time/hrtimer.c:1229:3: error: incompatible pointer types assigning to
'volatile typeof (timer->base)' (aka 'struct hrtimer_clock_base *volatile')
from 'bool (*)(struct hrtimer_clock_base *)' (aka '_Bool (*)(struct
hrtimer_clock_base *)') [-Werror,-Wincompatible-pointer-types]
1229 | WRITE_ONCE(timer->base, &migration_base);
------------------------------------------------------------------------
Me, I am a bit surprised that enqueue_hrtimer_offline() is even built
in a CONFIG_SMP=n kernel.
But there might be some reason why #ifdef-ing out that function's body
would be a bad idea, so over to you! ;-)
Thanx, Paul
> Frederic Weisbecker (3):
> hrtimers: Force migrate away hrtimers queued after
> CPUHP_AP_HRTIMERS_DYING
> rcu: Remove swake_up_one_online() bandaid
> Revert "rcu/nocb: Fix rcuog wake-up from offline softirq"
>
> include/linux/hrtimer_defs.h | 1 +
> kernel/rcu/tree.c | 34 +-------------------
> kernel/rcu/tree_exp.h | 2 +-
> kernel/rcu/tree_nocb.h | 10 ++----
> kernel/time/hrtimer.c | 60 +++++++++++++++++++++++++++++++-----
> 5 files changed, 58 insertions(+), 49 deletions(-)
>
> --
> 2.46.0
>
