[PATCH] x86: Fix i386 nmi_watchdog that does not trigger die_nmi
authorGOTO Masanori <gotom@sanori.org>
Wed, 8 Mar 2006 05:55:29 +0000 (21:55 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Wed, 8 Mar 2006 22:14:01 +0000 (14:14 -0800)
Fix i386 nmi_watchdog that does not meet watchdog timeout condition.  It
does not hit die_nmi when it should be triggered, because the current
nmi_watchdog_tick in arch/i386/kernel/nmi.c never count up alert_counter
like this:

void nmi_watchdog_tick (struct pt_regs * regs) {
if (last_irq_sums[cpu] == sum) {
alert_counter[cpu]++; <- count up alert_counter, but
if (alert_counter[cpu] == 5*nmi_hz)
die_nmi(regs, "NMI Watchdog detected LOCKUP");
alert_counter[cpu] = 0; <- reset alert_counter

This patch changes it back to the previous and working version.

This was found and originally written by Kohta NAKASHIMA.

(akpm: also uninline write_watchdog_counter(), saving 184 byets)

Signed-off-by: GOTO Masanori <gotom@sanori.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
arch/i386/kernel/nmi.c

index 63f39a7e2c96b0e3db1ce48ab73696e95db60a0c..be87c5e2ee9584747e05b57f8ce60ee2b29d93ed 100644 (file)
@@ -357,7 +357,7 @@ static void clear_msr_range(unsigned int base, unsigned int n)
                wrmsr(base+i, 0, 0);
 }
 
-static inline void write_watchdog_counter(const char *descr)
+static void write_watchdog_counter(const char *descr)
 {
        u64 count = (u64)cpu_khz * 1000;
 
@@ -544,7 +544,7 @@ void nmi_watchdog_tick (struct pt_regs * regs)
                         * die_nmi will return ONLY if NOTIFY_STOP happens..
                         */
                        die_nmi(regs, "NMI Watchdog detected LOCKUP");
-
+       } else {
                last_irq_sums[cpu] = sum;
                alert_counter[cpu] = 0;
        }