Re: [RFC PATCH 2/9] arm64/entry: Make debug_exception_enter/exit() noinstr

[Mark Rutland <mark.rutland@arm.com>]

On Mon, Jul 27, 2026 at 12:25:36PM +0000, Hongyan Xia wrote:
> From: Hongyan Xia <hongyan.xia@transsion.com>
> 
> Commit 879a6754d3d11e30af24b7dc486f561510d62641 ran into a crash because
> debug_exception_enter/exit() triggered page faults caused by perf dwarf
> call graph tracing. That patch was a band-aid on top. Instead of trying
> to band-aid all possible paths that can happen during instrumentation or
> perf tracing, simply make both functions noinstr.
> 
> Drop the RCU_LOCKDEP_WARN(): arm64_enter_el1_dbg() runs first and
> enters NMI context via ct_nmi_enter(), so RCU is always watching by the
> time debug_exception_enter() runs.
> 
> Signed-off-by: Hongyan Xia <hongyan.xia@transsion.com>
> ---
>  arch/arm64/kernel/entry-common.c | 24 +++++++++++++++---------
>  1 file changed, 15 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c
> index 466529cce1c3..f06b5e2437cd 100644
> --- a/arch/arm64/kernel/entry-common.c
> +++ b/arch/arm64/kernel/entry-common.c
> @@ -293,20 +293,26 @@ static __always_inline void fpsimd_syscall_exit(void)
>   * accidentally schedule in exception context and it will force a warning
>   * if we somehow manage to schedule by accident.
>   */
> -static void debug_exception_enter(struct pt_regs *regs)
> +static void noinstr debug_exception_enter(struct pt_regs *regs)
>  {
> -	preempt_disable();
> -
> -	/* This code is a bit fragile.  Test it. */
> -	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
> +	/*
> +	 * debug_exception_enter/exit() can be quite delicate. The normal
> +	 * preempt_disable/enable() can be instrumented or traced by perf,
> +	 * which leads to a can of worms including triggering page faults.
> +	 *
> +	 * Instead of trying to make all of them work properly here, just
> +	 * open-code the simpler version of preempt_disable/enable() and make
> +	 * enter/exit() both noinstr to avoid the entire complexity.
> +	 */
> +	__preempt_count_inc();
> +	barrier();
>  }
> -NOKPROBE_SYMBOL(debug_exception_enter);

This is an open-coded version of preempt_disable_notrace(), and there's
no reason the existing code needs to be out-of-line in the first place.
Please make this:

static __always_inline void debug_exception_enter(struct pt_regs *regs)
{
	preempt_disable_notrace();
}

> -static void debug_exception_exit(struct pt_regs *regs)
> +static void noinstr debug_exception_exit(struct pt_regs *regs)
>  {
> -	preempt_enable_no_resched();
> +	barrier();
> +	__preempt_count_dec();
>  }
> -NOKPROBE_SYMBOL(debug_exception_exit);

Similarly, this is an open-coded copy of
preempt_enable_no_resched_notrace(). Please make this:

static __always_inline void debug_exception_exit(struct pt_regs *regs)
{
	preempt_enable_no_resched_notrace();
}

In both cases I'm happy with keeping the regs argument for now.

Mark.