Skip to contentSkip to conversation

powerpc: oops fixes

PATCH Applied Nicholas Piggin 3 participants 8 messages Nov 8, 2016 – Nov 22, 2016 PowerPC

lore
Share & tools
lore

Nicholas Piggin

author reviewer
#0 lore raw

cover powerpc: oops fixes

Hi,

Couple of small things I've come across. After this, injecting nmi to enter xmon (via qemu) and then recovering works nicely.

Thanks,
Nick

Nicholas Piggin (2):
powerpc: fix graceful debugger recovery
powerpc: fix second nested oops hang

 arch/powerpc/kernel/traps.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
Signature · 1 line
2.10.2
author reviewer
#1 lore raw

PATCH 1/2 powerpc: fix graceful debugger recovery

When exiting xmon with 'x' (exit and recover), oops_begin bails out immediately, but die then calls __die() and oops_end(), which cause a lot of bad things to happen.

If the debugger was attached then went to graceful recovery, exit from die() immediately.

Signed-off-by Nicholas Piggin <[email protected]>
---
 arch/powerpc/kernel/traps.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
arch/powerpc/kernel/traps.c +5 -4
index 7480902..26c3ba4 100644
@@ -122,9 +122,6 @@ static unsigned long oops_begin(struct pt_regs *regs)
 	int cpu;
 	unsigned long flags;
 
-	if (debugger(regs))
-		return 1;
-
 	oops_enter();
 
 	/* racy, but better than risking deadlock. */

@@ -227,8 +224,12 @@ NOKPROBE_SYMBOL(__die);
 
 void die(const char *str, struct pt_regs *regs, long err)
 {
-	unsigned long flags = oops_begin(regs);
+	unsigned long flags;
+
+	if (debugger(regs))
+		return;
 
+	flags = oops_begin(regs);
 	if (__die(str, regs, err))
 		err = 0;
 	oops_end(flags, regs, err);
Signature · 1 line
2.10.2
reviewer
#2 lore raw

Re: [PATCH 1/2] powerpc: fix graceful debugger recovery

Quoted 3 lines · Nicholas Piggin

When exiting xmon with 'x' (exit and recover), oops_begin bails out immediately, but die then calls __die() and oops_end(), which cause a lot of bad things to happen.

In fact oops_begin() returns 1, which oops_end() then passes directly to raw_local_irq_restore() as flags. On 64-bit that actually works because arch_local_irq_restore() takes just "en" (enable), not real flags. But on 32-bit it's supposed to be the MSR value. So that's impressively broken.

Quoted 2 lines

If the debugger was attached then went to graceful recovery, exit from die() immediately.

Right. Crucially it doesn't change anything in terms of the actual logic of oops_begin(), ie. previously oops_begin() did nothing prior to calling debugger(), and after this patch that remains the same (which you did mention above but just spelling it out for myself).

cheers

author reviewer
#3 lore raw

Re: [PATCH 1/2] powerpc: fix graceful debugger recovery

Quoted 11 lines · On Thu, 10 Nov 2016 12:35:59 +1100 Michael Ellerman <[email protected]> wrote:
Quoted 3 lines · Nicholas Piggin

When exiting xmon with 'x' (exit and recover), oops_begin bails out immediately, but die then calls __die() and oops_end(), which cause a lot of bad things to happen.

In fact oops_begin() returns 1, which oops_end() then passes directly to raw_local_irq_restore() as flags. On 64-bit that actually works because arch_local_irq_restore() takes just "en" (enable), not real flags. But on 32-bit it's supposed to be the MSR value. So that's impressively broken.

Yeah, I guess most of the time you either go to debugger with sysrq, or in case of a crash don't try to graceful recover.

When sending debug NMIs down via system reset it becomes a problem!

Quoted 8 lines
Quoted 3 lines

If the debugger was attached then went to graceful recovery, exit from die() immediately.

Right. Crucially it doesn't change anything in terms of the actual logic of oops_begin(), ie. previously oops_begin() did nothing prior to calling debugger(), and after this patch that remains the same (which you did mention above but just spelling it out for myself).

Right.

Thanks,
Nick

reviewer
#4 lore raw

Re: [PATCH 1/2] powerpc: fix graceful debugger recovery

Quoted 9 lines · Nicholas Piggin
Quoted 6 lines · On Thu, 10 Nov 2016 12:35:59 +1100 Michael Ellerman <[email protected]> wrote:
Quoted 3 lines · Nicholas Piggin

When exiting xmon with 'x' (exit and recover), oops_begin bails out immediately, but die then calls __die() and oops_end(), which cause a lot of bad things to happen.

Quoted 8 lines
Quoted 5 lines

In fact oops_begin() returns 1, which oops_end() then passes directly to raw_local_irq_restore() as flags. On 64-bit that actually works because arch_local_irq_restore() takes just "en" (enable), not real flags. But on 32-bit it's supposed to be the MSR value. So that's impressively broken.

Yeah, I guess most of the time you either go to debugger with sysrq, or in case of a crash don't try to graceful recover.

Yeah. It's debatable whether we should even allow graceful recovery, but it's useful sometimes and regular users probably shouldn't have a debugger enabled anyway.

cheers

Applied
#5 lore raw

Re: [1/2] powerpc: fix graceful debugger recovery

In reply to Nicholas Piggin
Quoted 8 lines · Nicholas Piggin

When exiting xmon with 'x' (exit and recover), oops_begin bails out immediately, but die then calls __die() and oops_end(), which cause a lot of bad things to happen.

If the debugger was attached then went to graceful recovery, exit from die() immediately.

Signed-off-by Nicholas Piggin <[email protected]>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/6f44b20ee9b4130345c189c0c90ef6

cheers

author reviewer
#6 lore raw

PATCH 2/2 powerpc: fix second nested oops hang

In reply to Nicholas Piggin

When ending an oops, don't clear die_owner unless the nest count went to zero. This prevents a second nested oops from hanging forever on the die_lock.

Signed-off-by Nicholas Piggin <[email protected]>
---
 arch/powerpc/kernel/traps.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
arch/powerpc/kernel/traps.c +3 -2
index 26c3ba4..ba78e38 100644
@@ -147,14 +147,15 @@ static void oops_end(unsigned long flags, struct pt_regs *regs,
 			       int signr)
 {
 	bust_spinlocks(0);
-	die_owner = -1;
 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 	die_nest_count--;
 	oops_exit();
 	printk("\n");
-	if (!die_nest_count)
+	if (!die_nest_count) {
 		/* Nest count reaches zero, release the lock. */
+		die_owner = -1;
 		arch_spin_unlock(&die_lock);
+	}
 	raw_local_irq_restore(flags);
 
 	crash_fadump(regs, "die oops");
Signature · 1 line
2.10.2