linux-kernel - Re: NMI watchdog

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <470F5341.6030906@free.fr>
Date:	Fri, 12 Oct 2007 12:58:09 +0200
From:	John Sigler <linux.kernel@...e.fr>
To:	Björn Steinbrink <B.Steinbrink@....de>
CC:	linux-kernel@...r.kernel.org, linux-rt-users@...r.kernel.org
Subject: Re: NMI watchdog

Björn Steinbrink wrote:

> John Sigler wrote:
> 
>> I'm experiencing a full system lockup. I'm using an out-of-tree driver 
>> which I suspect is responsible. I'm trying to enable the NMI watchdog.
>>
>> # cat /proc/version
>> Linux version 2.6.22.1-rt9 (gcc version 3.4.6) #1 PREEMPT RT Tue Oct 9 
>> 12:25:47 CEST 2007
>>
>> # cat /proc/cmdline
>> ro root=/dev/hdc1 console=ttyS0,57600n8 console=tty0 panic=3 apic=debug 
>> nmi_watchdog=2
>>
>> However, after boot, the NMI count does not change.
>>
>> # cat /proc/interrupts ; sleep 10 ; cat /proc/interrupts
> 
> Try running some cpu hog in the background. The performance counters get
> increased only when the CPU is actually doing something. On a mostly
> idle system, it can take quite a while for the next NMI to show up.

You are right. In another shell, I ran while true; do : ; done

# cat /proc/interrupts ; sleep 10 ; cat /proc/interrupts
            CPU0
   0:        100   IO-APIC-edge      timer
   4:         82   IO-APIC-edge      serial
   8:          1   IO-APIC-edge      rtc
   9:          0   IO-APIC-fasteoi   acpi
  15:      13648   IO-APIC-edge      ide1
  16:       1303   IO-APIC-fasteoi   eth0
  17:        575   IO-APIC-fasteoi   eth1
  18:        575   IO-APIC-fasteoi   eth2
  19:        575   IO-APIC-fasteoi   eth3
NMI:       2889
LOC:     115768
ERR:          0
MIS:          0

            CPU0
   0:        100   IO-APIC-edge      timer
   4:         82   IO-APIC-edge      serial
   8:          1   IO-APIC-edge      rtc
   9:          0   IO-APIC-fasteoi   acpi
  15:      13672   IO-APIC-edge      ide1
  16:       1310   IO-APIC-fasteoi   eth0
  17:        580   IO-APIC-fasteoi   eth1
  18:        580   IO-APIC-fasteoi   eth2
  19:        580   IO-APIC-fasteoi   eth3
NMI:       2899
LOC:     116770
ERR:          0
MIS:          0

The performance counter appears to be configured to fire when the event 
count for CPU_CLK_UNHALTED reaches 2,400,000,000 (I have a 2.4 GHz CPU) 
i.e. one NMI per second when the CPU is 100% busy. Is that correct?

On a related note, I have a Pentium 3 which counts CPU_CLK_UNHALTED 
cycles even when the CPU is halted. I was told this is a bug, but it 
actually sounds like a a nice feature!

Is there really no way to have the event counter increment with every 
tick (even when the CPU is halted) on a P4?

Regards.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/