Incorrect ACPI blacklisting of ASUS P4B266 ?

Quoted 5 lines · On Saturday 28 March 2009 16:32:49 you wrote:

Hi.

There are quite a lot of reports of people having problems managing poweroff of their ASUS P4B266 based systems.

Quoted 7 lines

It seems that providing acpi=force as a boot param is quite succesful.

Maybe that should be fixed WRT to the blacklisting... hence reporting as advised in the kernel source (arch/x86/kernel/acpi/boot.c).

I unfortunately couldn't identify the reason why such a blacklisting was setup, as it seems to come from long time ago.

Maybe acpi was broken at that times and the bug was not in the ASUS BIOS, but in the ACPI implementation. Or the BIOS got fixed up.

Why don't you send a patch removing the ASUS from the dmi list and post it on [email protected]
Best also add a pointer to e.g. a discussion where people state that it works better with acpi=force.

Thomas

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi.

Following advice from Thomas Renninger, I hereby propose a patch for the ACPI blacklisting kernel code, that I've successfully applied to kernel 2.6.26-15, that allows ACPI detection on Asus P4B266 mainboards.

Maybe there could be a smarter version that would allow blacklisting for same mainboards with older BIOS versions than the one I'm using, but I don't know if/how that'd be possible. So the patch I propose is pretty obvious.

FYI, here are some reports that mention acpi=force working succesfully for P4B266 mainboards (in english and german) :

http://forums.fedoraforum.org/showpost.php?p=669615&postcount=5 / http://fedoraforum.org/forum/showpost.php?p=669615&postcount=5
https://lists.ubuntu.com/archives/kernel-bugs/2006-November/023486.html / https://bugs.launchpad.net/linux/+bug/43961/comments/145
http://forum.ubuntuusers.de/topic/automatische-abschaltung/#post-249151
http://www.pc-forum24.de/suse-system-installieren/3031-suse-10-2-laesst-sich-nicht-ausschalten.html#post13729

I hope this won't break things for different BIOS versions than mine,

and that this will on the other hand allow lots of users to benefit from working ACPI.

Quoted 6 lines · On Saturday 25 April 2009 21:35:42 Olivier Berger wrote:

Hi.

Following advice from Thomas Renninger, I hereby propose a patch for the ACPI blacklisting kernel code, that I've successfully applied to kernel 2.6.26-15, that allows ACPI detection on Asus P4B266 mainboards.

Hmm, there seem to be different models of this motherboard series:
P4B266
P4B266-C
P4B266-E
P4B266-M
P4B266-SE

P4B266 and P4B266-SE seem to have a similar BIOS history and the latest for both is: Beta Version 1011.003

The rest has a similar version string, but different latest BIOS versions (e.g. 1007)

The dmi blacklisting currently done in the kernel is probably matching all of above and is rather unfortunate.
We could have:
- A BIOS update which makes all of above models work with acpi(=force) well. Then your patch is perfect.

- One or more of above (latest) BIOSes does not work well with acpi=force still.
Then we'd get regressions and some people will moan about that and the patch will get reverted. In this case we should find out which kind of BIOS/mainboard it is and enhance the blacklisting to only match this(these) -> dmidecode is needed to be able to blacklist more fine grained.

Quoted 13 lines

Maybe there could be a smarter version that would allow blacklisting for same mainboards with older BIOS versions than the one I'm using, but I don't know if/how that'd be possible. So the patch I propose is pretty obvious.

FYI, here are some reports that mention acpi=force working succesfully for P4B266 mainboards (in english and german) :

http://forums.fedoraforum.org/showpost.php?p=669615&postcount=5 / http://fedoraforum.org/forum/showpost.php?p=669615&postcount=5
https://lists.ubuntu.com/archives/kernel-bugs/2006-November/023486.html / https://bugs.launchpad.net/linux/+bug/43961/comments/145
http://forum.ubuntuusers.de/topic/automatische-abschaltung/#post-249151
http://www.pc-forum24.de/suse-system-installieren/3031-suse-10-2-laesst-sich-nicht-ausschalten.html#post13729

Quoted 1 line

I hope this won't break things for different BIOS versions than mine,

That's the risk and we should try to find out more first.

I wonder whether it's worth that at all, one of the latest BIOSes is from:
2003/06/10

Quoted 3 lines

and that this will on the other hand allow lots of users to benefit from working ACPI.

It should be best if you state in above references that these guys must use acpi=force.
Others should find it then via google.
That's the easiest and safest way.

I tried to find out who and why this got added, but this exists even before git history...

Quoted 14 lines · Thomas Renninger

Quoted 6 lines · On Saturday 25 April 2009 21:35:42 Olivier Berger wrote:

Hi.

Following advice from Thomas Renninger, I hereby propose a patch for the ACPI blacklisting kernel code, that I've successfully applied to kernel 2.6.26-15, that allows ACPI detection on Asus P4B266 mainboards.

Hmm, there seem to be different models of this motherboard series:
P4B266
P4B266-C
P4B266-E
P4B266-M
P4B266-SE

Quoted 3 lines

P4B266 and P4B266-SE seem to have a similar BIOS history and the latest for both is: Beta Version 1011.003

Where is this information available (other than on Asus support site) , btw ? From the Asus support site.

Quoted 9 lines

The rest has a similar version string, but different latest BIOS versions (e.g. 1007)

The dmi blacklisting currently done in the kernel is probably matching all of above and is rather unfortunate.
We could have:
- A BIOS update which makes all of above models work with acpi(=force) well. Then your patch is perfect.

Quoted 8 lines

- One or more of above (latest) BIOSes does not work well with acpi=force still.
Then we'd get regressions and some people will moan about that and the patch will get reverted. In this case we should find out which kind of BIOS/mainboard it is and enhance the blacklisting to only match this(these) -> dmidecode is needed to be able to blacklist more fine grained.

Isn't there any possibility to compare BIOS versions in the blacklisting code ?

Quoted 13 lines

Quoted 13 lines

Maybe there could be a smarter version that would allow blacklisting for same mainboards with older BIOS versions than the one I'm using, but I don't know if/how that'd be possible. So the patch I propose is pretty obvious.

FYI, here are some reports that mention acpi=force working succesfully for P4B266 mainboards (in english and german) :

http://forums.fedoraforum.org/showpost.php?p=669615&postcount=5 / http://fedoraforum.org/forum/showpost.php?p=669615&postcount=5
https://lists.ubuntu.com/archives/kernel-bugs/2006-November/023486.html / https://bugs.launchpad.net/linux/+bug/43961/comments/145
http://forum.ubuntuusers.de/topic/automatische-abschaltung/#post-249151
http://www.pc-forum24.de/suse-system-installieren/3031-suse-10-2-laesst-sich-nicht-ausschalten.html#post13729

Quoted 5 lines

Quoted 1 line

I hope this won't break things for different BIOS versions than mine,

That's the risk and we should try to find out more first.

I wonder whether it's worth that at all, one of the latest BIOSes is from:
2003/06/10

Of course it's an old mainboard... but I suppose there are still quite a bunch in operation.

Quoted 7 lines

Quoted 3 lines

and that this will on the other hand allow lots of users to benefit from working ACPI.

It should be best if you state in above references that these guys must use acpi=force.
Others should find it then via google.
That's the easiest and safest way.

Well... adding burden on users instead of clean patch ;)

Quoted 3 lines

I tried to find out who and why this got added, but this exists even before git history...

Same for my searches :(

So... there ain't a way to provide an improved patch that wouldn't change anything but for tested BIOS versions (btw, mine is :
BIOS Information
Vendor: Award Software, Inc.
Version: ASUS P4B266 ACPI BIOS Revision 1010 Release Date: 08/06/2002
) ?

That would be sad if the current blacklisting wasn't able to do such checks on versions :(

Anyway, I suppose I will follow your advice and try and report people of the acpi=force as I already started to do in
http://www.olivierberger.com/weblog/index.php?post/2009/03/28/Proper-power-management-on-Asus-P4B266-mainboard

Quoted 3 lines

That would be sad if the current blacklisting wasn't able to do such checks on versions :(

You could either:
1) add a whitelist into the blacklist
2) better limit the blacklist
The first won't break machines, but is ugly.

For the second you must know which BIOS(es) fix the acpi parts to not cause regressions and then list all broken BIOS revisions, e.g.:

	{
	 .callback = force_acpi_ht,
	 .ident = "ASUS P4B266",
	 .matches = {
		     DMI_MATCH(DMI_BOARD_VENDOR, "ASUSTeK Computer INC."),
		     DMI_MATCH(DMI_BOARD_NAME, "P4B266"),
		     DMI_MATCH(DMI_BOARD_VERSION, "ASUS P4B266 ACPI BIOS Revision 1007"),
		     },
	 },
	{
	 .callback = force_acpi_ht,
	 .ident = "ASUS P4B266",
	 .matches = {
		     DMI_MATCH(DMI_BOARD_VENDOR, "ASUSTeK Computer INC."),
		     DMI_MATCH(DMI_BOARD_NAME, "P4B266"),
		     DMI_MATCH(DMI_BOARD_VERSION, "ASUS P4B266 ACPI BIOS Revision 1008"),
		     },
	 },

BTW, the same seem to have happened for the ASUS A7V:

	/*
	 * Boxes that need ACPI PCI IRQ routing disabled
	 */
	{
	 .callback = disable_acpi_irq,
	 .ident = "ASUS A7V",
	 .matches = {
		     DMI_MATCH(DMI_BOARD_VENDOR, "ASUSTeK Computer INC"),
		     DMI_MATCH(DMI_BOARD_NAME, "<A7V>"),
		     /* newer BIOS, Revision 1011, does work */
		     DMI_MATCH(DMI_BIOS_VERSION,
			       "ASUS A7V ACPI BIOS Revision 1007"),
		     },
	 },

One could argue that people have to upgrade to the latest BIOS and your patch is ok. This is IMO a valid argument, but it could be that people with a P4B266-SE board cannot upgrade to version 1010. I'd say, better leave the fingers off...

Quoted 4 lines

Anyway, I suppose I will follow your advice and try and report people of the acpi=force as I already started to do in
http://www.olivierberger.com/weblog/index.php?post/2009/03/28/Proper-power-management-on-Asus-P4B266-mainboard

Everyone with such problems should easily be able to google the acpi=force param for this board then.

Thomas Renninger <[email protected]> writes:

I'm thinking about some kind of warning message that might be provided by the kernel.

Currently, the blacklisting issues something like :
xxx detected : force use of acpi=ht

Maybe in some greylisted cases (like such ASUS P4B266), an additional message may be issued, something like :
xxx detected : may work with acpi=force (test at own risk)
or something like that ?

Quoted 9 lines · On Tuesday 28 April 2009 19:15:47 Olivier Berger wrote:

Thomas Renninger <[email protected]> writes:

I'm thinking about some kind of warning message that might be provided by the kernel.

Currently, the blacklisting issues something like :
xxx detected : force use of acpi=ht

Quoted 4 lines

Maybe in some greylisted cases (like such ASUS P4B266), an additional message may be issued, something like :
xxx detected : may work with acpi=force (test at own risk)
or something like that ?

Sounds like a good idea.
Something like that (not tested at all).
I could imagine Len adds this one if you give it a try and see the message popping up in dmesg.

Quoted 14 lines · Thomas Renninger

Quoted 13 lines · On Tuesday 28 April 2009 19:15:47 Olivier Berger wrote:

Thomas Renninger <[email protected]> writes:

I'm thinking about some kind of warning message that might be provided by the kernel.

Currently, the blacklisting issues something like :
xxx detected : force use of acpi=ht

Maybe in some greylisted cases (like such ASUS P4B266), an additional message may be issued, something like :
xxx detected : may work with acpi=force (test at own risk)
or something like that ?

Quoted 4 lines

Sounds like a good idea.
Something like that (not tested at all).
I could imagine Len adds this one if you give it a try and see the message popping up in dmesg.

I can confirm this works.

The P4B266 DMI entry went into ./arch/i386/kernel/dmi_scan.c back in Aug-2003.

That was when I was just starting to maintain ACPI, and I took a bunch of blacklist entries from SuSE because they were already shipping with ACPI enabled.

In the short term, it may have been the right thing to do, but in the long term it was a mistake and I regret doing it.

Blacklist entries paper-over real bugs, and unless we are lucky enough that somebody like you, Oliver, steps forward, blacklist entries are nearly impossible to ever remove.

So I'm inclined to apply your original patch to delete the blacklist entry entirely. If somebody with one of those boxes has a regression, we'll go fix their box -- which we may have acutally fixed years ago and not known it...

Olivier,
If you can send me a properly formatted patch
to remove the dmi entry,
(Documentation/SubmittingPatches)
including your signed-off, i'll apply it.

thanks,
Len Brown, Intel Open Source Technology Center

From 6f304fa7c0e0c5bac86fb47c53354889017fceb8 Mon Sep 17 00:00:00 2001

Quoted 5 lines · Len Brown

Olivier,
If you can send me a properly formatted patch
to remove the dmi entry,
(Documentation/SubmittingPatches)
including your signed-off, i'll apply it.

Here's it in attachment. I'm not really experienced with git so I hope it suits your needs.

applied for 2.6.31