2.4.19 on dual 1gig G4 funny slow down to kernel death experience

2.4.19 on dual 1gig G4 funny slow down to kernel death experience

[Kevin B. Hendricks]

Hi,

I just experienced the funny slow down until kernel death experience again
that I reported many months ago on my dual 1gig G4 machine (smp with 768
meg of memory)

The dmesg log is full of a long stream of the following

wait_on_irq, CPU 0:
irq:  -1 [0 0]
bh:   0 [0 0]

Everything got slower and slower (I could actually watch the sequence of
events that happen when an icon is clicked!)

And processing would only continue if I hit <enter> for some reason.

I was able to restart the machine after syncing the drive but my restart
ended up with lost interrupt messages and literally hung on the hde
partition check.

I literally had to unplug the machine for awhile and then restart it to
continue.

Once rebooted everything is lightning fast again.


Can anyone tell me what this warning/error message really means and how I
might go about tracking this down myself.

Thanks,

Kevin


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: 2.4.19 on dual 1gig G4 funny slow down to kernel death experience

[Benjamin Herrenschmidt]

On Fri, 2003-02-07 at 22:29, Kevin B. Hendricks wrote:

> I just experienced the funny slow down until kernel death experience again
> that I reported many months ago on my dual 1gig G4 machine (smp with 768
> meg of memory)
>
> The dmesg log is full of a long stream of the following
>
> wait_on_irq, CPU 0:
> irq:  -1 [0 0]
> bh:   0 [0 0]

Argh !

The global irq count should never _ever_ get to -1, that's very bad...

Either some weird codepath I haven't manage to figure out, or,
and that's what I'm afraid of, for some reasons, atomic operations
between CPUs failed, which should obviously never happen !

Remind me which rev of the CPU it is ?

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: 2.4.19 on dual 1gig G4 funny slow down to kernel death experience

[Kevin B. Hendricks]

Hi Ben,

I don't even remember anymore.  :-)

But the of device tree under the cpu has this cpu-version info:

8001 0201

Which from cputables.c looks like a 7455

Also, this is a very rare event (it has only happned twice in 6 months or
so).  The only thing in common with both events is that I have several
very large compiles going on in multiple processes at the same time (I
have two different xterms running independent compiles of OpenOffice.org),
while also surfing and things (killing time while the compiles continue)

I know this is not a lot to go on but it is now repeatable.

Thanks,

Kevin

On February 8, 2003 08:27, Benjamin Herrenschmidt wrote:
> On Fri, 2003-02-07 at 22:29, Kevin B. Hendricks wrote:
> > I just experienced the funny slow down until kernel death experience
> > again that I reported many months ago on my dual 1gig G4 machine (smp
> > with 768 meg of memory)
> >
> > The dmesg log is full of a long stream of the following
> >
> > wait_on_irq, CPU 0:
> > irq:  -1 [0 0]
> > bh:   0 [0 0]
>
> Argh !
>
> The global irq count should never _ever_ get to -1, that's very bad...
>
> Either some weird codepath I haven't manage to figure out, or,
> and that's what I'm afraid of, for some reasons, atomic operations
> between CPUs failed, which should obviously never happen !
>
> Remind me which rev of the CPU it is ?
>
> Ben.
>
>


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/