Re: [PATCH] rseq: Protect event mask against membarrier IPI

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

On 2025-08-13 11:02, Thomas Gleixner wrote:
> From: Thomas Gleixner <tglx@linutronix.de>
> Date: Tue, 12 Aug 2025 16:34:43 +0200
> 
> rseq_need_restart() reads and clears task::rseq_event_mask with preemption
> disabled to guard against the scheduler.
> 
> But membarrier() uses an IPI and sets the PREEMPT bit in the event mask
> from the IPI, which leaves that RMW operation unprotected.
> 
> Use guard(irq) if CONFIG_MEMBARRIER is enabled to fix that.
> 
> Fixes: 2a36ab717e8f ("rseq/membarrier: Add MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ")
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: "Paul E. McKenney" <paulmck@kernel.org>
> Cc: Boqun Feng <boqun.feng@gmail.com>
> Cc: stable@vger.kernel.org
> ---
>   include/linux/rseq.h |   11 ++++++++---
>   kernel/rseq.c        |   10 +++++-----
>   2 files changed, 13 insertions(+), 8 deletions(-)
> 
> --- a/include/linux/rseq.h
> +++ b/include/linux/rseq.h
> @@ -7,6 +7,12 @@
>   #include <linux/preempt.h>
>   #include <linux/sched.h>
>   
> +#ifdef CONFIG_MEMBARRIER
> +# define RSEQ_EVENT_GUARD	irq
> +#else
> +# define RSEQ_EVENT_GUARD	preempt
> +#endif

We should also update this comment in include/linux/sched.h:

         /*
          * RmW on rseq_event_mask must be performed atomically
          * with respect to preemption.
          */
         unsigned long rseq_event_mask;

to e.g.:

/*
  * RmW on rseq_event_mask must be performed atomically
  * with respect to preemption and membarrier IPIs.
  */

> +
>   /*
>    * Map the event mask on the user-space ABI enum rseq_cs_flags
>    * for direct mask checks.
> @@ -41,9 +47,8 @@ static inline void rseq_handle_notify_re
>   static inline void rseq_signal_deliver(struct ksignal *ksig,
>   				       struct pt_regs *regs)
>   {
> -	preempt_disable();
> -	__set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
> -	preempt_enable();
> +	scoped_guard(RSEQ_EVENT_GUARD)
> +		__set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);

Then we have more to worry about interaction of the following
rseq events with membarrier IPI:

- rseq_preempt, rseq_migrate, rseq_signal_deliver.

Both rseq_preempt and rseq_migrate are documented as only being required
to be called with preempt off, not irq off.

I don't see the point in sharing the same rseq_event_mask across all of
those rseq event sources.

Can we just move the event sources requiring preempt-off to their own
word, and use a separate word for membarrier IPI instead ? This would
allow us to partition the problem into two distinct states each
protected by their respective mechanism.

>   	rseq_handle_notify_resume(ksig, regs);
>   }
>   
> --- a/kernel/rseq.c
> +++ b/kernel/rseq.c
> @@ -342,12 +342,12 @@ static int rseq_need_restart(struct task
>   
>   	/*
>   	 * Load and clear event mask atomically with respect to
> -	 * scheduler preemption.
> +	 * scheduler preemption and membarrier IPIs.
>   	 */
> -	preempt_disable();
> -	event_mask = t->rseq_event_mask;
> -	t->rseq_event_mask = 0;
> -	preempt_enable();
> +	scoped_guard(RSEQ_EVENT_GUARD) {
> +		event_mask = t->rseq_event_mask;
> +		t->rseq_event_mask = 0;
> +	}

Instead we could sample both the preempt-off and the irq-off
words here.

Thanks,

Mathieu

>   
>   	return !!event_mask;
>   }


-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com