On Mon, Sep 21, 2026 at 10:27:29PM +0200, Boqun Feng wrote:
> On Mon, Sep 14, 2026 at 05:34:14PM +0800, Kunwu Chan wrote:
> > The atomic SRCU fastpath enables synchronize_srcu_atomic() to be called
> > from contexts with interrupts disabled. This series extends the
> > rcutorture lockdep tests to cover the atomic SRCU path and fixes an
> > IRQ-state preservation bug discovered along the way.
> >
> > Patch 1-2 add lockdep test coverage for atomic SRCU:
> > - testtype 4: same-type atomic SRCU deadlock
> > - testtype 5: atomic SRCU + raw spinlock dependency cycle
> > - testtype 6: rcu_read_lock() -> synchronize_srcu_atomic()
> > legitimate nesting, must not trigger lockdep warning
> >
> > Patch 3 fixes a bug where the synchronize_srcu_atomic() call chain
> > unconditionally re-enables interrupts on unlock, by switching to the
> > irqsave/irqrestore variants throughout the call chain.
> >
> > Patch 4 fixes a pre-existing double nerrs count in the script.
> >
> > Patch 5 adds testtype 7: cross-CPU IRQ context mismatch detection,
> > where an atomic SRCU reader held with IRQs enabled on one CPU is
> > paired with synchronize_srcu_atomic() via IPI on another CPU.
>
> Acked-by: Boqun Feng <[email protected]>
Queued for testing and further review, thank you both!
Thanx, Paul
> Regards,
> Boqun
>
> > Kunwu Chan (5):
> > rcutorture: Add atomic SRCU lockdep support
> > selftests/rcutorture: Wire atomic SRCU into srcu_lockdep.sh
> > srcutree: Preserve IRQ state in synchronize_srcu_atomic() callchain
> > selftests/rcutorture: Fix double nerrs count in srcu_lockdep.sh
> > rcutorture: Add atomic SRCU cross-CPU IRQ context mismatch test
> >
> > kernel/rcu/rcutorture.c | 109 +++++++++++++++++-
> > kernel/rcu/srcutree.c | 42 ++++---
> > .../selftests/rcutorture/bin/srcu_lockdep.sh | 77 ++++++++++++-
> > 3 files changed, 205 insertions(+), 23 deletions(-)
> >
> > --
> > 2.43.0
> >
> >
