Re: [PATCH/RFC] oops and panic message logging to MTD

[Richard Purdie <rpurdie@openedhand.com>]

On Tue, 2007-07-03 at 12:47 +0300, Jarkko Lavinen wrote:
> On Tue, Jun 19, 2007 at 11:00:54AM +0100, Richard Purdie wrote:
> > Its not a case of formatting the whole partition. The whole point of
> > this code is the following use case:
> > 
> > 1. Device crashes
> > 2. Device reboots
> > 3. mtdoops partition has a log of why it crashed
> 
> The oops logger uses oops_in_progress variable to detect the begin and the 
> end of an oops.  The end is detected when the first non-oops line comes and
> oops_in_progress is false.
> 
> This works if the kernel is still running after the oops and gemerates some 
> non-oops messages.  But if there is no non-oops line following an oops, no
> flushing will occur and there won't be a log on flash.

There was a printk within bust_spinlocks which flushed the klogd queues
and hence flushed the mtd_oops queue too.

I've noticed this has recently been removed and replaced with a wait
queue [1] and hence the problem mentioned above now exists (but didn't
when the driver was developed).

[1] http://git.o-hand.com/?p=linux-rpurdie;a=commitdiff;h=e3e8a75d2acfc61ebf25524666a0a2c6abb0620c

This raises the question of how to known when the oops has completed.
The neatest solution I can see would be to add some kind of optional
sync function pointer to struct console. Would that be acceptable?

Richard