On Thu, Oct 13, 2022 at 04:44:18PM +1000, Nicholas Piggin wrote:
> It's possible for an interrupt returning to an irqs-disabled context to
> lose a pending soft-masked irq because it branches to part of the exit
> code for irqs-enabled contexts, which is meant to clear only the
> PACA_IRQS_HARD_DIS flag from PACAIRQHAPPENED by zeroing the byte. This
> just looks like a simple thinko from a recent commit (if there was no
> hard mask pending, there would be no reason to clear it anyway).
> 
> This also adds comment to the code that actually does need to clear the
> flag.
> 
> Cc: Jason A. Donenfeld <[email protected]>
> Reported-by: Sachin Sant <[email protected]>
> Reported-by: Guenter Roeck <[email protected]>
> Fixes: e485f6c751e0a ("powerpc/64/interrupt: Fix return to masked context 
> after hard-mask irq becomes pending")
> Signed-off-by: Nicholas Piggin <[email protected]>

Tested-by: Guenter Roeck <[email protected]>

> ---
> I credit Sachin as well because he likely ran into it here,
> 
> https://lists.ozlabs.org/pipermail/linuxppc-dev/2022-March/240971.html
> 
> It was much more difficult to hit on his setup so when I asked him to
> re-test with a later kernel and it didn't reproduce, I thought it
> could have been unrelated. I stared at the patch for ages back then and
> didn't see the bug. I should have known better :(
> 
> I'm pretty confident this is the hang problem. Reproducer is intermittent
> for me, but I did catch it losing pending irqs here using debug code, so
> this certainly is _a_ bug that can explain the symptoms.
> 
> Thanks,
> Nick
> 
>  arch/powerpc/kernel/interrupt_64.S | 15 +++++++++++++--
>  1 file changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/powerpc/kernel/interrupt_64.S 
> b/arch/powerpc/kernel/interrupt_64.S
> index 904a5608cbe3..978a173eb339 100644
> --- a/arch/powerpc/kernel/interrupt_64.S
> +++ b/arch/powerpc/kernel/interrupt_64.S
> @@ -538,7 +538,7 @@ _ASM_NOKPROBE_SYMBOL(interrupt_return_\srr\()_kernel)
>       beq     .Lfast_kernel_interrupt_return_\srr\() // EE already disabled
>       lbz     r11,PACAIRQHAPPENED(r13)
>       andi.   r10,r11,PACA_IRQ_MUST_HARD_MASK
> -     beq     1f // No HARD_MASK pending
> +     beq     .Lfast_kernel_interrupt_return_\srr\() // No HARD_MASK pending
>  
>       /* Must clear MSR_EE from _MSR */
>  #ifdef CONFIG_PPC_BOOK3S
> @@ -555,12 +555,23 @@ _ASM_NOKPROBE_SYMBOL(interrupt_return_\srr\()_kernel)
>       b       .Lfast_kernel_interrupt_return_\srr\()
>  
>  .Linterrupt_return_\srr\()_soft_enabled:
> +     /*
> +      * In the soft-enabled case, need to double-check that we have no
> +      * pending interrupts that might have come in before we reached the
> +      * restart section of code, and restart the exit so those can be
> +      * handled.
> +      *
> +      * If there are none, it is be possible that the interrupt still
> +      * has PACA_IRQ_HARD_DIS set, which needs to be cleared for the
> +      * interrupted context. This clear will not clobber a new pending
> +      * interrupt coming in, because we're in the restart section, so
> +      * such would return to the restart location.
> +      */
>  #ifdef CONFIG_PPC_BOOK3S
>       lbz     r11,PACAIRQHAPPENED(r13)
>       andi.   r11,r11,(~PACA_IRQ_HARD_DIS)@l
>       bne-    interrupt_return_\srr\()_kernel_restart
>  #endif
> -1:
>       li      r11,0
>       stb     r11,PACAIRQHAPPENED(r13) // clear the possible HARD_DIS
>  
> -- 
> 2.37.2
> 

Reply via email to