Re: 2.5.67-mm3: Bad: scheduling while atomic with IEEE1394 then hard freeze ( lockup on CPU0)

["Philippe Gramoullé" <philippe.gramoulle@mmania.com>]

Hello,

On Tue, 15 Apr 2003 16:05:30 -0700
Andrew Morton <akpm@digeo.com> wrote:

  | Philippe Gramoullé  <philippe.gramoulle@mmania.com> wrote:
  | >
  | > 
  | > http://www.philou.org/2.5.67-mm3/2.5.67-mm3.log
  | 
  | This is a great bug report.  Thanks.

Well, i finally managed to get some output when i learned about the
nmi_watchdog boot option ( reading another thread about debugging hard hangs)
so i'm pleased if this report helped :)
  | 
  | The 1394 warnings are known about and I think Ben is working on it.

Ok, great. I think this is one of the latest thing that prevents me to
use 2.5.x almost full time.

  | 
  | The NMI watchdog hit is nasty:
  | 
[snip]
  | 
  | What has happened here is that you were in the middle of a kobject_get(),
  | holding spin_lock(&kobj_lock) when an interrupt came in.  The USB interrupt
  | handler comes in and ends up calling kobject_get() again.  This CPU already
  | holds the lock and blamyouredead.
  | 
  | Turning kobj_lock into an IRQ-safe lock would appear to be a sufficient fix.

I'll wait for the fix and will happily try it once it's available.

Thanks,

Philippe