2.6.18 Hard lockup advice needed: do_IRQ: stack overflow 508

2.6.18 Hard lockup advice needed: do_IRQ: stack overflow 508

[Steve Rottinger]

Hi,

I have a small Pentium 4 file server set up using version 2.6.18 of the Kernel.  The system contains a 
SiI 3124 based SATA controller, and I have a software RAID set up, with LVM enabled.   During normal daily operation, the system is stable.    However, when I perform nightly backups, and there is a lot of disk and network activity, as the data is transferred to a backup server, the system eventually locks up.   Unfortunately, there is no indication of a failure in the system logs.   I also enabled the NMI_Watchdog without any additional visibility.    By disabling the screen blanking on the console windows,  I do get a single message reported before the lockup:

do_IRQ: stack overflow 508
do_IRQ+0x69/0xbc 0xc0104ce4

If I downgrade to 2.6.17 of the Kernel, I see the same problem, but it takes longer for it to occur.

-Steve

Re: 2.6.18 Hard lockup advice needed: do_IRQ: stack overflow 508

[Jarek Poplawski]

On 18-10-2006 17:50, Steve Rottinger wrote:
> Hi,
> 
> I have a small Pentium 4 file server set up using version 2.6.18 of the 
> Kernel.  The system contains a SiI 3124 based SATA controller, and I 
> have a software RAID set up, with LVM enabled.   During normal daily 
> operation, the system is stable.    However, when I perform nightly 
> backups, and there is a lot of disk and network activity, as the data is 
> transferred to a backup server, the system eventually locks up.   
> Unfortunately, there is no indication of a failure in the system logs.   
> I also enabled the NMI_Watchdog without any additional visibility.    By 
> disabling the screen blanking on the console windows,  I do get a single 
> message reported before the lockup:
> 
> do_IRQ: stack overflow 508
> do_IRQ+0x69/0xbc 0xc0104ce4
> 
> If I downgrade to 2.6.17 of the Kernel, I see the same problem, but it 
> takes longer for it to occur.

You can try to change "Use 4Kb for kernel stacks instead of 8Kb" 
setting under "Kernel hacking" and if no result enable more
debugging options.

Jarek P.