Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Peter Zijlstra <peterz@infradead.org>]

On Sat, May 30, 2026 at 12:26:20PM +0200, Paolo Bonzini wrote:

> From f2c604a94cdea9bb96bb21c8a29288eae6cbac67 Mon Sep 17 00:00:00 2001
> From: Paolo Bonzini <pbonzini@redhat.com>
> Date: Sat, 30 May 2026 10:54:36 +0200
> Subject: [PATCH] seqcount: allow raw_seqcount_try_begin in atomic context
> 
> When raw_seqcount_try_begin() is used, the typical action when
> the seqcount is busy is to take the same lock that the writer uses.
> Another possibility, found in kernel/events/uprobes.c, is to delay
> the work to a safe point using RCU.  Either way there is no need
> to guarantee progress of the writer under CONFIG_PREEMPT_RT,
> because the caller is not going to call raw_seqcount_try_begin() in a loop.
> 
> However, raw_seqcount_try_begin() currently uses seqprop_sequence()
> via raw_read_seqcount(), and therefore does a lock/unlock of the
> write-side lock on PREEMPT_RT kernel.  This prevents it from being
> used in atomic context.  Use __seqprop_sequence() instead of
> raw_read_seqcount() to allow it.
> 
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Sean Christopherson <seanjc@google.com>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  include/linux/seqlock.h | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
> index 5a40252b8334..9adbf5d9667c 100644
> --- a/include/linux/seqlock.h
> +++ b/include/linux/seqlock.h
> @@ -338,7 +338,10 @@ SEQCOUNT_LOCKNAME(mutex,        struct mutex,    true,     mutex)
>   */
>  #define raw_seqcount_try_begin(s, start)				\
>  ({									\
> -	start = raw_read_seqcount(s);					\
> +	start = __seqprop_sequence(s);					\
> +									\
> +	if (!(start & 1))						\
> +		kcsan_atomic_next(KCSAN_SEQLOCK_REGION_MAX);		\
>  	!(start & 1);							\
>  })

Yeah, I think this makes sense. Thanks!