AVR32: nmi_enter() without nmi_exit()

AVR32: nmi_enter() without nmi_exit()

[Manfred Spraul]

[-- Attachment #1: Type: text/plain, Size: 384 bytes --]

Hi Haavard,

You are listed as the maintainer for AVR32. If someone else is doing 
that now, could you forward the mail?

While updating the rcu code, I noticed that do_nmi() for AVR32 is odd:
There is an nmi_enter() call without an nmi_exit().
This can't be correct, it breaks rcu (at least the preempt version) and 
lockdep.

Attached is a patch, not tested at all.

--
    Manfred

[-- Attachment #2: patch-avr32-nmi-exit --]
[-- Type: text/plain, Size: 614 bytes --]

diff --git a/arch/avr32/kernel/traps.c b/arch/avr32/kernel/traps.c
index b835c4c..c39cdfc 100644
--- a/arch/avr32/kernel/traps.c
+++ b/arch/avr32/kernel/traps.c
@@ -120,11 +120,11 @@ asmlinkage void do_nmi(unsigned long ecr, struct pt_regs *regs)
 	case NOTIFY_BAD:
 		die("Fatal Non-Maskable Interrupt", regs, SIGINT);
 	default:
+		printk(KERN_ALERT "Got NMI, but nobody cared. Disabling...\n");
+		nmi_disable();
 		break;
 	}
-
-	printk(KERN_ALERT "Got NMI, but nobody cared. Disabling...\n");
-	nmi_disable();
+	nmi_exit();
 }
 
 asmlinkage void do_critical_exception(unsigned long ecr, struct pt_regs *regs)

Re: AVR32: nmi_enter() without nmi_exit()

[Haavard Skinnemoen]

Manfred Spraul <manfred@colorfullife.com> wrote:
> You are listed as the maintainer for AVR32. If someone else is doing 
> that now, could you forward the mail?
> 
> While updating the rcu code, I noticed that do_nmi() for AVR32 is odd:
> There is an nmi_enter() call without an nmi_exit().
> This can't be correct, it breaks rcu (at least the preempt version) and 
> lockdep.
> 
> Attached is a patch, not tested at all.

Totally forgot about this, sorry.

Applied, thanks.

Haavard