Re: PATCH - acpica-20031203

On Tuesday 09 December 2003 7:44 pm, Nate Lawson wrote:
> On Tue, 9 Dec 2003, Andy Fawcett wrote:
> > On Monday 08 December 2003 19:47, Nate Lawson wrote:
> > > I've built and run the latest ACPI-CA dist and it looks pretty
> > > good. It has a lot of fixes from FreeBSD as well as Linux. 
> > > Please give it a try and let me know if it gives you any new
> > > problems.
> > >
> > >     http://www.root.org/~nate/freebsd/acpica-20031203.diff.gz
>
> The about patch has been committed, along with some other related
> changes. Please cvsup to today and test again.

I will do that later, please see below first.

> > I've just installed 5.2-CURRENT (post -BETA branching) to my Asus
> > P2B-DS based system, and I am seeing slow performance. Looking at
> > the output of "ps -axu" I can see massive amounts of Interrupt
> > activity (removed VSZ and RSS fields to save space):
> >
> > USER    PID %CPU %MEM   TT  STAT STARTED      TIME COMMAND
> > root     30 95.3  0.0   ??  RL    1:08PM   4:11.77  (irq20: acpi0)
> > root     12 53.9  0.0   ??  RL    1:08PM   5:08.86  (idle: cpu0)
> > root     11 50.0  0.0   ??  RL    1:08PM   4:28.89  (idle: cpu1)
> >
> > # uname -a
> > FreeBSD nobby.int.athame.co.uk 5.2-CURRENT FreeBSD 5.2-CURRENT #2:
> > Mon Dec  8 21:32:07 EET 2003    
> > root_at_nobby.int.athame.co.uk:/usr/obj/usr/ src/sys/NOBBY  i386
> >
> > This happens both before and after installing the 20031203
> > patchset.
>
> Since you have this problem after applying the patch, cvsuping
> probably won't help.

Indeed.

> > The box was previously running 5.1R with no apparent problems.
> >
> > From dmesg:
> > acpi0: Overriding SCI Interrupt from IRQ 9 to IRQ 20
> >
> > I've read through as much of the -current archives as possible, but
> > I can't see anything that helps me diagnose the problem.
>
> The issue is likely that you have a broken BIOS.

It's quite possible, the BIOS on this system (while being the latest 
available from Asus) is not very recent.

> Before 5.2, we
> didn't use the APIC on non-SMP boxes.  Now we use it if it's
> available.

I didn't quite understand this, since my box _is_ SMP...


> Try this in loader.conf to disable APIC irq routing:
>
>   hint.apic.0.disabled=1

That has removed the spurious interrupt problem from my system. I'll do 
some more experimenting later, and provide some feedback if I find 
anything out.

Thanks,


Andy


-- 
Andy Fawcett                                     | andy_at_athame.co.uk
                                                 | tap_at_kde.org
"In an open world without walls and fences,      | tap_at_lspace.org
  we wouldn't need Windows and Gates."  -- anon  | tap_at_fruitsalad.org