Re: >100K interrupts/s on IOAPIC 0 Pin 9

On Wed, Sep 12, 2007 at 06:18:33PM +0200, Edgar Fu? wrote:
> > There have been several efforts to add code to NetBSD to fix this
> > but I don't believe any have been committed as each turned out to
> > be not quite right in some way.
> Any way I can help to improve that? Anything I can test?

Please try -current from today and see if that helps. There was a pretty
nasty bug where the SCI overwrite was not used correctly if the BIOS
requested anything but an identity mapping. If that still, contact me
directly, I don't follow this lists regulary.

Joerg

ACPI test kernels for *all* users

Hi all,
it seems that the fix for ACPI SCI handling broke some machines that
were working before. The jmcneill-pm branch has a change that it
should fix all cases, but it is aggressive and might break older machines.

Please test http://www.netbsd.org/~joerg/netbsd-i386 or 
http://www.netbsd.org/~joerg/netbsd-amd64 (depending on your
architecture). The kernel should boot without issues. Power down should
still work with it. "vmstat -i" on the command line (or "show event" in
ddb) should not show any interrupt storms.

I am especially interested in reports from systems without IOAPIC.

Joerg

Re: Adding a aprint_debug to the errata patch path

Hi Bernd,

On Sun, Sep 30, 2007 at 10:54:07AM +0200, Bernd Ernesti wrote:

> I would like to commit the following patch, so it possible to see which
> errata was patched, if you boot the kernel with the debug option.

Looks fine to me - go for it.

Thanks,
Andrew

LKMs support for compat_linux(8)

Hi,

I just finished to add support for compat_linux(8) using the lkm(4)
framework.

njoly <at> lanfeust [~]> uname -a
NetBSD lanfeust.sis.pasteur.fr 4.99.31 NetBSD 4.99.31 (GENERIC) #10: Thu Oct  4 13:45:21 CEST 2007 
njoly <at> lanfeust.sis.pasteur.fr:/local/src/NetBSD/obj/amd64/sys/arch/amd64/compile/GENERIC amd64

njoly <at> lanfeust [~]> /emul/linux/bin/uname -a
zsh: exec format error: /emul/linux/bin/uname
njoly <at> lanfeust [~]> sudo modload /usr/lkm/compat_linux.o
Module loaded as ID 0
njoly <at> lanfeust [~]> sudo modload /usr/lkm/exec_linux_elf.o
Module loaded as ID 1
njoly <at> lanfeust [~]> /emul/linux/bin/uname -a
Linux lanfeust.sis.pasteur.fr 2.4.18 #0 Wed Feb 20 20:00:02 CET 2002 x86_64 x86_64 x86_64 GNU/Linux
njoly <at> lanfeust [~]> modstat
Type    Id   Offset Loadaddr         Size Info             Rev Module Name
COMPAT    0       - ffffffff80ba2000 0050 ffffffff80bb3260   2 compat_linux
EXEC      1       - ffffffff80bb6000 0004 ffffffff80bb6b40   2 exec_linux_elf

The attached patch does mostly the following :

- Add needed COMPAT_OSSAUDIO to GENERIC kernel config.

- Add missing includes to mostly all compat linux files, needed by
  `linux_syscallargs.h'.

Re: LKMs support for compat_linux(8)

Excellent, LKM support for this would be convenient wrt: diagnosing
problems.  BTW (and I'm sure others have thought about this), do you
know if there's a make variable that would allow me to pass
-DMULTIPROCESSOR to the LKM build?  I almost always use an MP kernel,
and the source tree builds LKMs for a UP kernel, which is annoying. ;)

--Blair

Re: LKMs support for compat_linux(8)

On Thu, Oct 04, 2007 at 09:50:34AM -0400, Blair Sadewitz wrote:
> Excellent, LKM support for this would be convenient wrt: diagnosing
> problems.  BTW (and I'm sure others have thought about this), do you
> know if there's a make variable that would allow me to pass
> -DMULTIPROCESSOR to the LKM build?  I almost always use an MP kernel,
> and the source tree builds LKMs for a UP kernel, which is annoying. ;)

Unfortunately no ...

--

-- 
Nicolas Joly

Biological Software and Databanks.
Institut Pasteur, Paris.

Re: ACPI test kernels for *all* users

On Wed, Oct 03, 2007 at 01:35:24AM +0200, Joerg Sonnenberger wrote:
> it seems that the fix for ACPI SCI handling broke some machines that
> were working before. The jmcneill-pm branch has a change that it
> should fix all cases, but it is aggressive and might break older machines.

Attached is the backport of that specific change. It can be applied with
some fuzz to current as well.

Please test this. GENERIC kernels for netbsd4 can be found at
http://www.netbsd.org/~joerg/netbsd4-i386.bz2 and
http://www.netbsd.org/~joerg/netbsd4-amd64.bz2

I will ask for a pullup of this change only, if noone steps up with a
regression relative to RC1 for this patch. In that case the original
pullup will be backed out as well.

Joerg

Index: include/mpacpi.h
===================================================================
RCS file: /cvsroot/src/sys/arch/x86/include/mpacpi.h,v
retrieving revision 1.4
diff -u -r1.4 mpacpi.h
--- include/mpacpi.h	4 Jul 2006 00:30:22 -0000	1.4
+++ include/mpacpi.h	5 Oct 2007 15:01:56 -0000
 <at>  <at>  -14,4 +14,6  <at>  <at> 
 struct mp_intr_map;
 int mpacpi_findintr_linkdev(struct mp_intr_map *);

Re: ACPI test kernels for *all* users

On Fri, 5 Oct 2007, Joerg Sonnenberger wrote:

> On Wed, Oct 03, 2007 at 01:35:24AM +0200, Joerg Sonnenberger wrote:
>> it seems that the fix for ACPI SCI handling broke some machines that
>> were working before. The jmcneill-pm branch has a change that it
>> should fix all cases, but it is aggressive and might break older machines.
>
> Attached is the backport of that specific change. It can be applied with
> some fuzz to current as well.
>
> Please test this. GENERIC kernels for netbsd4 can be found at
> http://www.netbsd.org/~joerg/netbsd4-i386.bz2 and
> http://www.netbsd.org/~joerg/netbsd4-amd64.bz2
>
> I will ask for a pullup of this change only, if noone steps up with a
> regression relative to RC1 for this patch. In that case the original
> pullup will be backed out as well.

         Just to confirm this patch fixed my RC2 on thinkpad T41p issue.

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

Re: ACPI test kernels for *all* users

On Fri, Oct 05, 2007 at 05:23:25PM +0200, Joerg Sonnenberger wrote:
> Please test this. GENERIC kernels for netbsd4 can be found at
> http://www.netbsd.org/~joerg/netbsd4-i386.bz2 and
> http://www.netbsd.org/~joerg/netbsd4-amd64.bz2

http://www.netbsd.org/~joerg/netbsd4-mp-i386.bz2
http://www.netbsd.org/~joerg/netbsd4-mp-amd64.bz2

are the corresponding MP kernels as requested.

Joerg

Re: ACPI test kernels for *all* users

On Fri, Oct 05, 2007 at 09:01:04PM +0200, Joerg Sonnenberger wrote:
> On Fri, Oct 05, 2007 at 05:23:25PM +0200, Joerg Sonnenberger wrote:
> > Please test this. GENERIC kernels for netbsd4 can be found at
> > http://www.netbsd.org/~joerg/netbsd4-i386.bz2 and
> > http://www.netbsd.org/~joerg/netbsd4-amd64.bz2
> 
> http://www.netbsd.org/~joerg/netbsd4-mp-i386.bz2
> http://www.netbsd.org/~joerg/netbsd4-mp-amd64.bz2
> 
> are the corresponding MP kernels as requested.

Thank you.

Lets see.

- netbsd4-mp-i386 boots, but I couldn't break into ddb from the serial
console:

sdubp{e0r}v> isor trap page fault, code=0                                                           
Faulted in DDB; continuing...
db{1}kernel: supervisor trap page fault, code=0
Faulted in DDB; continuing...

- netbsd4-i386 seems to work and I can break into ddb, but a reboot
from ddb just hangs:

db> reboot
syncing disks... done
unmounting file systems...