[PATCH 0/1] iwlwifi: trace debug messages

Herton Ronaldo Krzesinski herton.krzesinski at canonical.com
Tue Sep 4 13:37:07 UTC 2012

On Tue, Sep 04, 2012 at 12:35:09PM +0100, Luis Henriques wrote:
> On Tue, Sep 04, 2012 at 12:04:59PM +0100, Andy Whitcroft wrote:
> > On Mon, Sep 03, 2012 at 01:52:44PM +0100, Luis Henriques wrote:
> > 
> > > It looks like the issue is due to an overflow of debug messages to the
> > > kernel logs.  After bisecting the kernel, commit:
> > > 
> > > 2655e314c4b204966008689eaf3e87ba1f38d55c iwlwifi: trace debug messages
> > > 
> > > does fix the problem.  This commit had to be backported to the Precise
> > > kernel as it didn't applied cleanly.  The original bug reporter claims
> > > the issue is fixed with this backport.
> > 
> > This is rather a large patch which must fix the issue as a side effect
> > rather than really fixing the original issue.  I am not against fixing
> > this this way I guess, but it does feel a little heavy handed.  Did we
> > have any idea which debug message was triggering the crash?
> I understand your concerns and its ok for me to look for a different
> solution.
> I was able to reproduce the issue with a similar HW, and the last
> message I'm able to get is:
> iwlwifi 0000:0c:00.0: Invalid tbl->lq_type 0
> Note that I get this message *before* the system actually freezes.
> After a few seconds, I'm not able to get anything out of the system, and
> a reboot will not show anything else on the logs.
> I guess that I could spend some more time debugging it, but the solution
> would probably end up being a SAUCE patch instead of a
> cherry-pick/backport.  I simply bisected the kernel and this was the
> commit that fixed it.
> Another issue with this approach is that there is code brought in by this
> commit that is not used on the driver (e.g., the IWL_DEBUG_QUIET_RFKILL
> is introduced by this patch, but the 3.2 driver is not using it).
> Having dead code (even if it isn't compiled in this example) always
> brings maintenance issues...
> Anyway, I'm ok dropping this patch and spend some more time debugging
> it.

Did you try disabling CONFIG_IWLWIFI_DEVICE_TRACING? It seems the limits
brought by commit 2655e31 on tracing is what would make the difference, so
perhaps disabling all the tracing would be better to understand and
confirm this. If it works, I'm not suggesting disabling the config
option as the ideal solution, but at least would isolate and clear any
concerns about commit 2655e31 I think.

> Cheers,
> --
> Luis


More information about the kernel-team mailing list