check_nmi_watchdog() excessive stack usage

PATCH New Mikael Pettersson 1 participant 1 message Dec 12, 2002

lore

Share & tools

lore

Mikael Pettersson

author

2002-12-12 13:54 UTC #0 lore raw

PATCH 1/1 check_nmi_watchdog() excessive stack usage

Marcelo,

This patch fixes a problem with excessive stack usage in arch/i386/kernel/nmi.c:check_nmi_watchdog(). It's a backport of a fix that has been in the 2.5 kernel for quite a while. There are no behavioural changes.

The problem is that the code copies NR_CPUS irq_cpustat_t values to the stack, even though only NR_CPUS ints actually are needed. irq_cpustat_t values are 24 bytes large, but they are also ____cacheline_aligned, which makes them blow up to 128 bytes on P4s. That's 128*32 == 4 kilobytes of stack on a P4 SMP kernel, when 4*32 == 128 bytes suffices.

/Mikael

arch/i386/kernel/nmi.c +6 -3

diff -ruN linux-2.4.21-pre1/arch/i386/kernel/nmi.c linux-2.4.21-pre1.check_nmi/arch/i386/kernel/nmi.c
@@ -72,18 +72,21 @@
 
 int __init check_nmi_watchdog (void)
 {
-	irq_cpustat_t tmp[NR_CPUS];
+	unsigned int prev_nmi_count[NR_CPUS];
 	int j, cpu;
 
 	printk(KERN_INFO "testing NMI watchdog ... ");
 
-	memcpy(tmp, irq_stat, sizeof(tmp));
+	for (j = 0; j < smp_num_cpus; j++) {
+		cpu = cpu_logical_map(j);
+		prev_nmi_count[cpu] = irq_stat[cpu].__nmi_count;
+	}
 	sti();
 	mdelay((10*1000)/nmi_hz); // wait 10 ticks
 
 	for (j = 0; j < smp_num_cpus; j++) {
 		cpu = cpu_logical_map(j);
-		if (nmi_count(cpu) - tmp[cpu].__nmi_count <= 5) {
+		if (nmi_count(cpu) - prev_nmi_count[cpu] <= 5) {
 			printk("CPU#%d: NMI appears to be stuck!\n", cpu);
 			return -1;
 		}