[PATCH 06/10] arm64: entry: Don't preempt with SError or Debug masked

[Jinjie Ruan]

On 2026/4/7 21:16, Mark Rutland wrote:
> On arm64, involuntary kernel preemption has been subtly broken since the
> move to the generic irqentry code. When preemption occurs, the new task
> may run with SError and Debug exceptions masked unexpectedly, leading to
> a loss of RAS events, breakpoints, watchpoints, and single-step
> exceptions.
> 
> Prior to moving to the generic irqentry code, involuntary preemption of
> kernel mode would only occur when returning from regular interrupts, in
> a state where interrupts were masked and all other arm64-specific
> exceptions (SError, Debug, and pseudo-NMI) were unmasked. This is the
> only state in which it is valid to switch tasks.
> 
> As part of moving to the generic irqentry code, the involuntary
> preemption logic was moved such that involuntary preemption could occur
> when returning from any (non-NMI) exception. As most exception handlers
> mask all arm64-specific exceptions before this point, preemption could
> occur in a state where arm64-specific exceptions were masked. This is
> not a valid state to switch tasks, and resulted in the loss of
> exceptions described above.
> 
> As a temporary bodge, avoid the loss of exceptions by avoiding
> involuntary preemption when SError and/or Debug exceptions are masked.
> Practically speaking this means that involuntary preemption will only
> occur when returning from regular interrupts, as was the case before
> moving to the generic irqentry code.
> 
> Fixes: 99eb057ccd67 ("arm64: entry: Move arm64_preempt_schedule_irq() into __exit_to_kernel_mode()")
> Reported-by: Ada Couprie Diaz <ada.coupriediaz at arm.com>
> Reported-by: Vladimir Murzin <vladimir.murzin at arm.com>
> Signed-off-by: Mark Rutland <mark.rutland at arm.com>
> Cc: Andy Lutomirski <luto at kernel.org>
> Cc: Catalin Marinas <catalin.marinas at arm.com>
> Cc: Jinjie Ruan <ruanjinjie at huawei.com>
> Cc: Peter Zijlstra <peterz at infradead.org>
> Cc: Thomas Gleixner <tglx at kernel.org>
> Cc: Will Deacon <will at kernel.org>
> ---
>  arch/arm64/include/asm/entry-common.h | 21 +++++++++++++--------
>  1 file changed, 13 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/entry-common.h b/arch/arm64/include/asm/entry-common.h
> index cab8cd78f6938..20f0a7c7bde15 100644
> --- a/arch/arm64/include/asm/entry-common.h
> +++ b/arch/arm64/include/asm/entry-common.h
> @@ -29,14 +29,19 @@ static __always_inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
>  
>  static inline bool arch_irqentry_exit_need_resched(void)
>  {
> -	/*
> -	 * DAIF.DA are cleared at the start of IRQ/FIQ handling, and when GIC
> -	 * priority masking is used the GIC irqchip driver will clear DAIF.IF
> -	 * using gic_arch_enable_irqs() for normal IRQs. If anything is set in
> -	 * DAIF we must have handled an NMI, so skip preemption.
> -	 */
> -	if (system_uses_irq_prio_masking() && read_sysreg(daif))
> -		return false;
> +	if (system_uses_irq_prio_masking()) {
> +		/*
> +		 * DAIF.DA are cleared at the start of IRQ/FIQ handling, and when GIC
> +		 * priority masking is used the GIC irqchip driver will clear DAIF.IF
> +		 * using gic_arch_enable_irqs() for normal IRQs. If anything is set in
> +		 * DAIF we must have handled an NMI, so skip preemption.
> +		 */
> +		if (read_sysreg(daif))
> +			return false;
> +	} else {
> +		if (read_sysreg(daif) & (PSR_D_BIT | PSR_A_BIT))
> +			return false;

Reviewed-by: Jinjie Ruan <ruanjinjie at huawei.com>

> +	}
>  
>  	/*
>  	 * Preempting a task from an IRQ means we leave copies of PSTATE