Re: 1352 NUL bytes at the end of a page? (was Re: Assertion `s && s->tree' failed: The saga continues.)

[Andrea Arcangeli <andrea@suse.de>]

On Sun, May 16, 2004 at 04:11:16PM -0600, Steven Cole wrote:
> On Sunday 16 May 2004 03:29 pm, Andrew Morton wrote:
> > Steven Cole <elenstev@mesatop.com> wrote:
> > >
> > > Anyway, although the regression for my particular machine for this
> > >  particular load may be interesting, the good news is that I've seen
> > >  none of the failures which started this whole thread, which are relatively
> > >  easily reproduceable with PREEMPT set.  
> > 
> > So...  would it be correct to say that with CONFIG_PREEMPT, ppp or its
> > underlying driver stack
> > 
> > a) screws up the connection and hangs and
> > 
> > b) scribbles on pagecache?
> > 
> > Because if so, the same will probably happen on SMP.
> > 
> Perhaps someone has the hardware to test this.
> 
> To summarize my experience with the past 24 hours of testing:
> Without PREEMPT , everything is rock solid. 

so we've two separate problems: the first is the ppp instability with
preempt, the second is a regresion in the vm heuristics between 2.6.3
and 2.6.5.

> and I (cringes at the thought) may repeat some bk pulls with
> PREEMPT set.

I've heard other reports of preempt being unstable with some sound
stuff, just in case are you using sound drivers at all during that
workload?