A 4 core Opteron needs longer than 10 ticks for this.
Signed-off-by: Andi Kleen <ak@suse.de>
        for_each_possible_cpu(cpu)
                prev_nmi_count[cpu] = per_cpu(irq_stat, cpu).__nmi_count;
        local_irq_enable();
-       mdelay((10*1000)/nmi_hz); // wait 10 ticks
+       mdelay((20*1000)/nmi_hz); // wait 20 ticks
 
        for_each_possible_cpu(cpu) {
 #ifdef CONFIG_SMP
 
        for (cpu = 0; cpu < NR_CPUS; cpu++)
                counts[cpu] = cpu_pda(cpu)->__nmi_count;
        local_irq_enable();
-       mdelay((10*1000)/nmi_hz); // wait 10 ticks
+       mdelay((20*1000)/nmi_hz); // wait 20 ticks
 
        for_each_online_cpu(cpu) {
                if (!per_cpu(nmi_watchdog_ctlblk, cpu).enabled)