Re: 2.6.30-rc1: invalid opcode with call trace

[Jens Axboe <jens.axboe@oracle.com>]

On Wed, Apr 08 2009, Ingo Molnar wrote:
> 
> * Jens Axboe <jens.axboe@oracle.com> wrote:
> 
> > On Wed, Apr 08 2009, Ingo Molnar wrote:
> > > 
> > > * Ingo Molnar <mingo@elte.hu> wrote:
> > > 
> > > > I too have an async hang/crash, on an old-style SCSI (aic7xxx) box 
> > > > - hang log attached below.
> > > > [...]
> > > >
> > > > ( Full bootlog attached below as well - i'm sending the config as a 
> > > >   reply as this mail is close to lkml size limits already. )
> > > 
> > > Config attached.
> > > 
> > >  known bad  : v2.6.29-9854-gd508afb
> > >  known good : v2.6.29
> > > 
> > > Suspected commit introducing the regression:
> > > 
> > >  9710794: async: remove the temporary (2.6.29) "async is off by default" code
> > > 
> > > (i'll now try a revert of this.)
> > 
> > That's what I figured was the culprit as well, but that does not 
> > really tell us anything about what part of async.c is buggy :-)
> 
> async.c itself is likely not to be buggy - fundamental bugs that 
> deep in the center of the kernel usually cannot hide for long :-)

While it may not be in async.c, the code a) really isn't that old, and
b) hasn't really been used yet. So I'd definitely not rule out a bug in
the async implementation itself.

> What matters more is the _effects_ of having async bootup now, on 
> various subsystems it interacts with. Unexpected parallelism and 
> reordering between init sequences.
> 
> It would have been far better to not have such a 'flip the switch 
> on' moment - but instead a more gradual step by step introduction of 
> async bootup, with accompanied strong testing.

Definitely, switching on single sub systems/drivers one at the time
would be a much saner approach.

-- 
Jens Axboe