Re: Frustrated with Linux, Asus, and nVidia, and AMD

Re: Frustrated with Linux, Asus, and nVidia, and AMD

[Robert Hancock]

Bill Davidsen wrote:
> 2.6.18 is the latest released kernel, I don't think calling one release 
> back "ancient" is really advancing the solution. The problem seems to 
> have been reported in August, and is still not fixed, I do understand 
> that he would feel there is no progress.
> 
> How long will you wait before putting in the fix Marc Perkel suggested, 
> perhaps with a warning logged that it's a band-aid? Many users will not 
> be astute enough to find this discussion, the bug report, the fix, 
> configure and build a kernel, etc. And not all distributions will 
> address it either.

As far as the "fix" of disabling the skip-ACPI-timer-override, that is 
not something that can be put in the kernel as it will break other 
boards that require the ACPI timer override not to be used (like many 
nForce2 boards for example). Breaking working setups in order to fix 
others isn't acceptable.

There are clearly some NVIDIA chipsets which require the override be 
skipped, and some which require it not be. I think the ball is currently 
in NVIDIA's court to provide a way of figuring out which chipsets 
require the quirk and which don't..

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: Frustrated with Linux, Asus, and nVidia, and AMD

[Bill Davidsen]

Robert Hancock wrote:
> Bill Davidsen wrote:
>> 2.6.18 is the latest released kernel, I don't think calling one 
>> release back "ancient" is really advancing the solution. The problem 
>> seems to have been reported in August, and is still not fixed, I do 
>> understand that he would feel there is no progress.
>>
>> How long will you wait before putting in the fix Marc Perkel 
>> suggested, perhaps with a warning logged that it's a band-aid? Many 
>> users will not be astute enough to find this discussion, the bug 
>> report, the fix, configure and build a kernel, etc. And not all 
>> distributions will address it either.
> 
> As far as the "fix" of disabling the skip-ACPI-timer-override, that is 
> not something that can be put in the kernel as it will break other 
> boards that require the ACPI timer override not to be used (like many 
> nForce2 boards for example). Breaking working setups in order to fix 
> others isn't acceptable.
> 
> There are clearly some NVIDIA chipsets which require the override be 
> skipped, and some which require it not be. I think the ball is currently 
> in NVIDIA's court to provide a way of figuring out which chipsets 
> require the quirk and which don't..
> 
The kernel seems to handle lots of other quirks, I agree that this 
should not be the default behavior, but it certainly seems possible to 
catch this on the more common cases for which it is needed.

The boards which need it probably should be the exception, 2.6 kernels 
did without it for several years, and breaking hardware which worked up 
through 2.6.15 is probably not the best way to cope with new boards. I 
don't think ignoring the problem while waiting for nVidia is good 
policy, since it's a regression.

-- 
Bill Davidsen <davidsen@tmr.com>
   Obscure bug of 2004: BASH BUFFER OVERFLOW - if bash is being run by a
normal user and is setuid root, with the "vi" line edit mode selected,
and the character set is "big5," an off-by-one errors occurs during
wildcard (glob) expansion.

lspci output needed was Re: Frustrated with Linux, Asus, and nVidia, and AMD

[Andi Kleen]

> There are clearly some NVIDIA chipsets which require the override be 
> skipped, and some which require it not be. I think the ball is currently 
> in NVIDIA's court to provide a way of figuring out which chipsets 
> require the quirk and which don't..

My current plan is to switch in 2.6.20 to automatic probing of more pins
for the timer routing (suggested by Tim Hockin, I've got a test patch). 

But that's too risky for .19.

For 2.6.19 we'll likely add some more PCI-IDs disabling the quirk
and a command line option to disable the skip timer override quirk. 

Doing this per PCI ID isn't that bad because afaik Vista certification
requires enabling the HPET table and I assume most boards will get
Vista certification soon. This will force Asus to fix their BIOS.

Can people who use a Nvidia based AM2/SocketF board (especially when they have timer 
troubles but otherwise would be useful too) please report their lspcis in private 
mail to me?

-Andi

Re: lspci output needed was Re: Frustrated with Linux, Asus, and nVidia, and AMD

[Bill Davidsen]

Andi Kleen wrote:

>>There are clearly some NVIDIA chipsets which require the override be 
>>skipped, and some which require it not be. I think the ball is currently 
>>in NVIDIA's court to provide a way of figuring out which chipsets 
>>require the quirk and which don't..
>>    
>>
>
>My current plan is to switch in 2.6.20 to automatic probing of more pins
>for the timer routing (suggested by Tim Hockin, I've got a test patch). 
>  
>
Thank you, Tim!

>But that's too risky for .19.
>
>For 2.6.19 we'll likely add some more PCI-IDs disabling the quirk
>and a command line option to disable the skip timer override quirk. 
>  
>
That should be safe, and timer override as an option should give 
everyone a way to get what they need on any given system.

>Doing this per PCI ID isn't that bad because afaik Vista certification
>requires enabling the HPET table and I assume most boards will get
>Vista certification soon. This will force Asus to fix their BIOS.
>  
>
And nVidia to release more information? Hopefully.

>Can people who use a Nvidia based AM2/SocketF board (especially when they have timer 
>troubles but otherwise would be useful too) please report their lspcis in private 
>mail to me?
>
>-Andi
>
>
>  
>


-- 
bill davidsen <davidsen@tmr.com>
  CTO TMR Associates, Inc
  Doing interesting things with small computers since 1979

Re: lspci output needed was Re: Frustrated with Linux, Asus, and nVidia, and AMD

[Gerd v. Egidy]

Hi,

> Can people who use a Nvidia based AM2/SocketF board (especially when they
> have timer troubles but otherwise would be useful too) please report their
> lspcis in private mail to me?

I'll send the lspci in private as requested in a minute.

I have a ASUS M2N-SLI Deluxe (nForce 570) with a Athlon64 X2 4600+ (AM2) and 
Bios version 0307 or 0402 beta. The problem with this board seems to be a 
little different than the problem the others posting here are experiencing. 
Maybe it's a different bug but it seems to be timer-related too.

Booting 2.6.18.1 (or some other 2.6er I tried) stops with

[...]
ENABLING IO-APIC IRQs
..TIMER: vector=0x31 apic1=0 pin1=0 apci2=-1 pin2=-1
..MP-BIOS bug: 8254 timer not connected to IO-APIC
....trying to set up timer (IRQ0) through the 8259A ...  failed.
....trying to set up timer as Virtual Wire IRQ... failed.
....trying to set up timer as ExtINT IRQ... failed :(.
Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug
and send a report.  Then try booting with the 'noapic' option

All kernels were SMP and i386 (not x86_64).

I tried to use the acpi_skip_timer_override=0 patch that helped the other 
posters but it did not help - same error. Of course I patched the 
corresponding code for i386 in arch/i386/kernel/acpi/earlyquirk.c. Booting 
with the no_timer_check boot parameter as some people suggested does not help 
too.

Booting with apic=debug does not reveal any additional info. Only booting with 
noapic makes the machine boot.

Kind regards,

Gerd