softdog - panic instead of reboot, kernel 2.6.22.9
PATCH 1/1 softdog - panic instead of reboot, kernel 2.6.22.9
Quoted 1 line
From Ken Sugawara <[email protected]>
Description: Add an option to softdog to panic upon timer expiration instead of reboot.
---
This patch is intended to help increase the chance of postmortem analysis in the event of silent system hang where it is impossible to initiate crash dump manually (e.g. no console or network response). We've had occasions where some of our customers had silent system hang problems in which they were unable to obtain vmcore for root cause analysis. Provided that timer interrupts are not blocked, this patch might enable them to get a crash dump in such a situation by activating softdog.
drivers/char/watchdog/softdog.c renamed from drivers/char/watchdog/softdog.c.orig +13 -4
@@ -49,6 +49,7 @@#include <linux/jiffies.h> #include <asm/uaccess.h> +#include <linux/kernel.h> #define PFX "SoftDog: "
@@ -70,6 +71,11 @@ static int soft_noboot = 0;module_param (soft_noboot, int,0 );MODULE_PARM_DESC (soft_noboot,"Softdog action, set to 1 to ignore reboots, 0 to reboot (default depends on ONLY_TESTING)" ); +static int panic_instead =0 ; + +module_param (panic_instead, int,0 ); +MODULE_PARM_DESC (panic_instead,"Softdog action, set to 1 to panic instead of reboot, 0 to reboot (default 0)" ); +/* * Our timer */
@@ -93,8 +99,11 @@ static void watchdog_fire(unsigned longif (soft_noboot) printk(KERN_CRIT PFX "Triggered - Reboot ignored.\n"); -else - { +else if (panic_instead) { + printk(KERN_CRIT PFX "Initiating panic.\n"); +panic ("Watchdog timer expired." ); + printk(KERN_CRIT PFX "Panic didn't ?????\n"); + }else { printk(KERN_CRIT PFX "Initiating system reboot.\n");emergency_restart (); printk(KERN_CRIT PFX "Reboot didn't ?????\n");
@@ -262,7 +271,7 @@ static struct notifier_block softdog_not .notifier_call = softdog_notify_sys, }; -static char banner[]__initdata =KERN_INFO "Software Watchdog Timer: 0.07 initialized. soft_noboot=%d soft_margin=%d sec (nowayout= %d)\n" ; +static char banner[]__initdata =KERN_INFO "Software Watchdog Timer: 0.07 initialized. soft_noboot=%d soft_margin=%d sec panic_instead=%d (nowayout= %d)\n" ;static int __init watchdog_init (void ) {
@@ -290,7 +299,7 @@ static int __init watchdog_init(void)return ret; } -printk (banner, soft_noboot, soft_margin, nowayout); +printk (banner, soft_noboot, soft_margin, panic_instead, nowayout);return 0 ; }
Quoted 3 lines
analysis. Provided that timer interrupts are not blocked, this patch might enable them to get a crash dump in such a situation by activating softdog.
You want the NMI based anti lockup stuff for that. If the box locks up then generally softdog will fail. The NMI based debug (CONFIG_DETECT_SOFTLOCKUP) is the option you want.
Quoted 7 lines · Alan Cox
Quoted 3 lines
analysis. Provided that timer interrupts are not blocked, this patch might enable them to get a crash dump in such a situation by activating softdog.
You want the NMI based anti lockup stuff for that. If the box locks up then generally softdog will fail. The NMI based debug (
CONFIG_DETECT_SOFTLOCKUP) is the option you want.
Thanks for your comment. I'll look into it, but unfortunately, most of our customers are running much older kernels (e.g. 2.6.9-based from RHEL4.x) which don't have the option, and some non-negligible number of them are running on non-IA platforms, whereas right now the option seems to be implemented only in i386 and x86_64 (correct me if I'm wrong).
Anyways, I'll try to look for a way to manually trigger NMI on their non -IA boxes, preferably without any additional hardware.
Regards,
Ken