[PATCH] i386: don't blindly enable interrupts in die()

[PATCH] i386: don't blindly enable interrupts in die()

[Jan Beulich]

[-- Attachment #1: Type: text/plain, Size: 177 bytes --]

Rather than blindly re-enabling interrupts in die(), save their state
upon entry and then restore that state.

From: Jan Beulich <jbeulich@novell.com>

(actual patch attached)


[-- Attachment #2: linux-2.6.14-i386-die-irq.patch --]
[-- Type: application/octet-stream, Size: 1017 bytes --]

Rather than blindly re-enabling interrupts in die(), save their state
upon entry and then restore that state.

From: Jan Beulich <jbeulich@novell.com>

--- 2.6.14/arch/i386/kernel/traps.c	2005-10-28 02:02:08.000000000 +0200
+++ 2.6.14-i386-die-irq/arch/i386/kernel/traps.c	2005-11-04 17:00:47.000000000 +0100
@@ -306,14 +306,17 @@ void die(const char * str, struct pt_reg
 		.lock_owner_depth =	0
 	};
 	static int die_counter;
+	unsigned long flags;
 
 	if (die.lock_owner != raw_smp_processor_id()) {
 		console_verbose();
-		spin_lock_irq(&die.lock);
+		spin_lock_irqsave(&die.lock, flags);
 		die.lock_owner = smp_processor_id();
 		die.lock_owner_depth = 0;
 		bust_spinlocks(1);
 	}
+	else
+		local_save_flags(flags);
 
 	if (++die.lock_owner_depth < 3) {
 		int nl = 0;
@@ -340,7 +343,7 @@ void die(const char * str, struct pt_reg
 
 	bust_spinlocks(0);
 	die.lock_owner = -1;
-	spin_unlock_irq(&die.lock);
+	spin_unlock_irqrestore(&die.lock, flags);
 
 	if (kexec_should_crash(current))
 		crash_kexec(regs);

Re: [PATCH] i386: don't blindly enable interrupts in die()

[Andrew Morton]

"Jan Beulich" <JBeulich@novell.com> wrote:
>
>  Rather than blindly re-enabling interrupts in die(), save their state
>  upon entry and then restore that state.

What is the reason for this change?

Re: [PATCH] i386: don't blindly enable interrupts in die()

[Jan Beulich]

>>> Andrew Morton <akpm@osdl.org> 08.11.05 23:57:21 >>>
>"Jan Beulich" <JBeulich@novell.com> wrote:
>>
>>  Rather than blindly re-enabling interrupts in die(), save their
state
>>  upon entry and then restore that state.
>
>What is the reason for this change?

If the kernel is in really bad condition and faults with interrupts
disabled, re-enabling them in die() may cause even more trouble,
implying more chances of data corruption.