[PATCH] powerpc/64s/interrupt: Fix lost interrupts when returning to soft-masked context

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 <Jason@zx2c4.com>
Reported-by: Sachin Sant <redacted>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Fixes: e485f6c751e0a ("powerpc/64/interrupt: Fix return to masked context after hard-mask irq becomes pending")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
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

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 <Jason@zx2c4.com>
Reported-by: Sachin Sant <redacted>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Fixes: e485f6c751e0a ("powerpc/64/interrupt: Fix return to masked context after hard-mask irq becomes pending")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>

I thought I sent this before, but I am not sure if it got lost
since I don't see it in the powerpc patchwork.

Tested-by: Guenter Roeck <linux@roeck-us.net>

Guenter

On Thu, 13 Oct 2022 16:44:18 +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).

[...]

Applied to powerpc/fixes.

[1/1] powerpc/64s/interrupt: Fix lost interrupts when returning to soft-masked context
      https://git.kernel.org/powerpc/c/a4cb3651a174366cc85a677da9e3681fbe97fdae

cheers