On Sun, May 28, 2000 at 02:05:03PM +0300, you [Ville Herva] claimed:
>
> Basicly I'm having three kind of symptoms:
> 
> o (with "noapic") Tons on "stuck in IPI waits, and the machine locks up
>   for five seconds. These seem to mostly happen when I move mouse, but I
>   can't think of anything else that could be related.

Well updated to 2.2.16pre5 + the patch below and that symptom went away.
Or, at least it hasn't happened for hours whereas it used to happen every
few minutes. Since other stuff changed as well, rebooted to see if it in
fact was that oneliner. Well, even without the oneliner, the ipi waits are
gone. Hmm. I guess the newest reiserfs patch now has work around in it.
Strange that I got those stucks even when not using reiserfs partitions
(module loaded, though).

I wonder if the apic counterpart to the noapic "stuck in ipi wait" could
be the slowdown symptom...?

=================================================================
Date:   Fri, 21 Apr 2000 07:11:46 -0700 (PDT)
From:   Chris Mason <[EMAIL PROTECTED]>
To:     Manfred Spraul <[EMAIL PROTECTED]>
cc:     Hans-Joachim Baader <[EMAIL PROTECTED]>,
        [EMAIL PROTECTED], [EMAIL PROTECTED]
Subject: Re: stuck on TLB IPI wait (CPU#0)

On Fri, 21 Apr 2000, Manfred Spraul wrote:
>
> It's probably a buglet in the main scheduler, and Reiserfs uncovered the
> problem 8-)
> IIRC one of the Reiserfs developers wrote that the newest patch should
> fix that problem.
> 

Exactly.  I work around this in our 2.3 code, but haven't had the chance
to port the fix back yet to the 2.2 code.  Andi Kleen sent along this
patch for sched.c, as far as I know, it has fixed the issue for everyone
that has tried it.

--- linux/kernel/sched.c-o      Wed Feb  9 14:27:20 2000
+++ linux/kernel/sched.c        Wed Mar 29 12:53:41 2000
@@ -803,6 +803,7 @@
        goto handle_bh_back;
 
 handle_tq_scheduler:
+       __sti(); 
        run_task_queue(&tq_scheduler);
        goto tq_scheduler_back;
=================================================================

I've now been running burnBX for a couple of hours, and everything seems
smooth. We'll see...


-- v --

[EMAIL PROTECTED]

/* I only have one bx burning desire; let me stand next to your firewire */
--
=-          To unsubscribe, email [EMAIL PROTECTED] with the       -=
=-                body of "unsubscribe linux-abit".                 -=

Reply via email to