What are the general causes of frozen system?

What are the general causes of frozen system?

[Larry Finger]

I am trying to help the bcm43xx project develop a driver for the 
Broadcom 43xx wireless chips, using my Linksys WPC54G card. 
Unfortunately, since the group got far enough to turn on RX DMA, my 
system has frozen whenever I load the driver. TX DMA was OK. It seems to 
correlate with the receipt of a beacon from my AP, but that cannot be 
proven. When the freeze happens, I cannot do anything more and have to 
power the system off.

What should I consider as a cause of the freeze? I have reviewed the 
code and do not find any obvious out-of-bounds memory references. I have 
tried various 'printk' statements, but none of them in the bottom-half 
interrupt routine make it to the logs. Are there any tricks that I 
should try?

Thanks,

Larry

Re: What are the general causes of frozen system?

[thockin]

> What should I consider as a cause of the freeze? I have reviewed the 
> code and do not find any obvious out-of-bounds memory references. I have 
> tried various 'printk' statements, but none of them in the bottom-half 
> interrupt routine make it to the logs. Are there any tricks that I 
> should try?

Look for lock-related deadlocks.  Try turning on the nmi watchdog

Re: What are the general causes of frozen system?

[Larry Finger]

thockin@hockin.org wrote:
>>What should I consider as a cause of the freeze? I have reviewed the 
>>code and do not find any obvious out-of-bounds memory references. I have 
>>tried various 'printk' statements, but none of them in the bottom-half 
>>interrupt routine make it to the logs. Are there any tricks that I 
>>should try?
> 
> 
> Look for lock-related deadlocks.  Try turning on the nmi watchdog
> 

Thanks for the quick response. Unfortunately, adding either 
nmi_watchdog=1 or 2 makes my machine lockup in booting - just after ACPI 
is initialized.

Re: What are the general causes of frozen system?

[Larry Finger]

thockin@hockin.org wrote:
> On Sat, Nov 26, 2005 at 09:05:37PM -0600, Larry Finger wrote:
> 
>>>Look for lock-related deadlocks.  Try turning on the nmi watchdog
>>
>>Thanks for the quick response. Unfortunately, adding either 
>>nmi_watchdog=1 or 2 makes my machine lockup in booting - just after ACPI 
>>is initialized.
> 
> 
> Try with acpi=off, too?
> 

It turned out that an 'lacpi' was causing the boot lockup. I tried all 
combinations of acpi on/off and nmi_watchdog 1/2 without any success. 
The NMI count in /proc/interrupts remains stuck at 0.

Re: What are the general causes of frozen system?

[Michael Buesch]

[-- Attachment #1: Type: text/plain, Size: 1216 bytes --]

On Sunday 27 November 2005 03:44, you wrote:
> I am trying to help the bcm43xx project develop a driver for the 
> Broadcom 43xx wireless chips, using my Linksys WPC54G card. 
> Unfortunately, since the group got far enough to turn on RX DMA, my 
> system has frozen whenever I load the driver. TX DMA was OK. It seems to 
> correlate with the receipt of a beacon from my AP, but that cannot be 
> proven. When the freeze happens, I cannot do anything more and have to 
> power the system off.
> 
> What should I consider as a cause of the freeze? I have reviewed the 
> code and do not find any obvious out-of-bounds memory references. I have 
> tried various 'printk' statements, but none of them in the bottom-half 
> interrupt routine make it to the logs. Are there any tricks that I 
> should try?

Enable most of the "Kernel Hacking" Debugging options, like
Memory and spinlock debugging. This will often catch memory corruptions
or deadlocks.

I'm not sure where the freeze happens, from your explanations.
Does it freeze before or after the DMA operations? I am sure we find the
reason, if you talk to us about the problem on IRC or the bcm43xx mailing list.

-- 
Greetings Michael.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: What are the general causes of frozen system?

[Benjamin Herrenschmidt]

On Sat, 2005-11-26 at 20:44 -0600, Larry Finger wrote:
> I am trying to help the bcm43xx project develop a driver for the 
> Broadcom 43xx wireless chips, using my Linksys WPC54G card. 
> Unfortunately, since the group got far enough to turn on RX DMA, my 
> system has frozen whenever I load the driver. TX DMA was OK. It seems to 
> correlate with the receipt of a beacon from my AP, but that cannot be 
> proven. When the freeze happens, I cannot do anything more and have to 
> power the system off.
> 
> What should I consider as a cause of the freeze? I have reviewed the 
> code and do not find any obvious out-of-bounds memory references. I have 
> tried various 'printk' statements, but none of them in the bottom-half 
> interrupt routine make it to the logs. Are there any tricks that I 
> should try?

spinlock debugging is one... if it's a bad DMA to low memory, you can't
do much. Another common cause is a stale interrupt, is your IRQ handler
called over & over again in a loop ?

The problem is that depending on what you are interrupting, printk may
not work ...

Ben.