[xfs-masters] Re: 2.6.19-rc6 : Spontaneous reboots, stack overflows - seems to implicate xfs, scsi, networking, SMP

["Jesper Juhl" <jesper.juhl@xxxxxxxxx>]

On 24/11/06, David Chinner <dgc@xxxxxxx> wrote:
> On Thu, Nov 23, 2006 at 07:37:00PM +0100, Arjan van de Ven wrote:
> > On Thu, 2006-11-23 at 14:16 +0100, Ingo Oeser wrote:
> > > Hi there,
> > >
> > > David Chinner schrieb:
> > > > If the softirqs were run on a different stack, then a lot of these
> >
> > softirqs DO run on their own stack!
>
> So they run on a separate stack for 4k stacks on x86?
>

Yes, with 4K stacks there's sepperate IRQ stack.

From the help text for CONFIG_4KSTACKS :

"If you say Y here the kernel will use a 4Kb stacksize for the
 kernel stack attached to each process/thread. This facilitates
 running more threads on a system and also reduces the pressure
 on the VM subsystem for higher order allocations. This option
 will also use IRQ stacks to compensate for the reduced stackspace."


> They don't run on a separate stack for 8k stacks on x86 -
> Jesper's traces show that - so this may indicate an issue
> with the methodology used to generate the stack overflow
> traces inteh first place. i.e. if 4k stacks use a separate
> stack, then most of the reported overflows are spurious
> and would not normally occur on 4k stack systems..
>

Well, some of the traces show that we were down to ~3K stack free with
8K stacks, so ~5K used. Even with 4K stacks and sepperate stack for
IRQs we will still be uncomfortably close to the edge in those cases.
Also, I did manage to capture a single line via netconsole while
running with 4K stacks :
    do_IRQ: stack overflow: 492
Unfortunately that was the only line that made it to the remote log
server, so I don't have the actual trace for that one. But it does
show that there really is an issue when running with 4K stacks, IRQ
stacks or no.


-- 
Jesper Juhl <jesper.juhl@xxxxxxxxx>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html