[0/3] sched: migrate_disable() vs per-CPU access safety checks

Message

Valentin Schneider July 21, 2021, 11:51 a.m. UTC

Hi folks,

I've hit a few warnings when taking v5.13-rt1 out for a spin on my arm64
Juno. Those are due to regions that become preemptible under PREEMPT_RT, but
remain safe wrt per-CPU accesses due to migrate_disable() + a sleepable lock.

This adds a helper that looks at not just preemptability but also affinity and
migrate disable, and plasters the warning sites.

Cheers,
Valentin

Valentin Schneider (3):
  sched: Introduce is_pcpu_safe()
  rcu/nocb: Check for migratability rather than pure preemptability
  arm64: mm: Make arch_faults_on_old_pte() check for migratability

 arch/arm64/include/asm/pgtable.h |  2 +-
 include/linux/sched.h            | 10 ++++++++++
 kernel/rcu/tree_plugin.h         |  3 +--
 3 files changed, 12 insertions(+), 3 deletions(-)

--
2.25.1

Comments

Thomas Gleixner July 27, 2021, 7:45 p.m. UTC | #1

On Wed, Jul 21 2021 at 12:51, Valentin Schneider wrote:
> Hi folks,
>
> I've hit a few warnings when taking v5.13-rt1 out for a spin on my arm64
> Juno. Those are due to regions that become preemptible under PREEMPT_RT, but
> remain safe wrt per-CPU accesses due to migrate_disable() + a sleepable lock.
>
> This adds a helper that looks at not just preemptability but also affinity and
> migrate disable, and plasters the warning sites.

Nice!

I just pulled that into the RT queue and it will show up with the next
release.

Thanks,

        tglx