Kernel-2.6.7-1.494 and Nvidia Drivers

[Brian R Smith]

Hey Stan,

> Then consult nVidia?  What version of the drivers are you using?
> I'm currently using NVIDIA-Linux-x86-1.0-6106-pkg1.run with 
> 2.6.7-1.476 without issue.

NVIDIA-Linux-x86-1.0-6106-pkg1.run with the same kernel as you. Are you
relying heavily on ACPI?  And of course I initially thought it could be
the problem, but:

> Rather than follow the same old mantra of
> adding an acpi=off line to grub (especially since this machine is a
> laptop), I rebuilt the kernel _without_ PREEMPT_VOLUNTARY...

At the end of my post, I said that I rebuilt the kernel _without_
PREEMPT_VOLUNTARY and that afterward, everything worked beautifully, the
NVIDIA Drivers, ACPI and all.  It was strictly a hunch that
PREEMPT_VOLUNTARY was causing problems with ACPI (possibly indirectly?)
and causing that subsystem to enter some race condition, chewing up my
CPU cycles.

My laptop's ACPI implementation shouldn't be broken, especially since
FC2 supported it quite well as does FreeBSD 5.x, in fact, if there are
kernel devs around, do you know if there are any differences between
linux's ACPI system and FreeBSD 5.x's?  It seems the BSD one works
better (of course, my mileage has varied), at least on my Dell Inspiron
2650.

Asside from the newer kernel releases with PREEMPT_VOLUNTARY being
"fixed",  I never noticed any performance gain while using it nor any
benefit so I may not even upgrade later.  But who knows?  I'm always
game for an upgrade.


-brian

On Sun, 2004-07-25 at 12:20, Stan Bubrouski wrote:
> On Sun, 2004-07-25 at 03:45 -0400, Brian R Smith wrote:
> > Hi,
> > 
> > Regarding Kernel-2.6.7-1.494 and Nvidia Drivers,
> > 
> > I don't know if this has been mentioned before, but I've noticed some
> > nasty problems with this kernel release (the rpm) and the Nvidia drivers
> > while running Quake3 based games.  
> > 
> 
> Then consult nVidia?  What version of the drivers are you using?
> I'm currently using NVIDIA-Linux-x86-1.0-6106-pkg1.run with 2.6.7-1.476
> without issue.
> 
> > First of all, performance is terrible.  I could barely get Return to
> > Castle Wolfenstein to run acceptably at 800x600x32 despite the fact that
> > normally, it runs flawlessly on my Geforce2go at 1024x768x32 under
> > normal circumstances.  Secondly, it seems that some of the hardware
> > lighting instructions break under this kernel.
> > 
> 
> This seems like it would be a driver problem...
> 
> > After fooling around with some things to try to fix this problem, I
> > noticed that the kacpid process crashes for some reason, causing it to
> > race and use >50% cpu.
> > 
> 
> Then wouldn't one assume ACPI might be the prob?  There have been
> numerous ACPI in the 2.6.8rcX kernels so this will prolly go away with
> the next kernel RPM release...
> 
> > I did some looking around and noticed that some people have had issues
> > with the PREEMPT_VOLUNTARY patch that was applied to this kernel and had
> > a hunch that perhaps this patch was in some way screwing up the ACPI
> 
> How about ACPI is screwing up because of ACPI bugs?  Preemt Voluntary
> has some issues here, but none are related to video and none are fatal.
> Seems to work GREAT 90% of the time.
> 
> > subsystem in the kernel.  Rather than follow the same old mantra of
> > adding an acpi=off line to grub (especially since this machine is a
> > laptop), I rebuilt the kernel _without_ PREEMPT_VOLUNTARY and with
> 
> Are you sure your laptop's ACPI implementation is sane?  Speaking of
> which you don't mention the laptop brand or hardware at all...
> 
> > PREEMPT and voila, the problems disappeared.  kacpid stopped crashing
> > and 3d under q3 based games worked wonderfully with no apparent
> > performance hit to my setup.
> > 
> 
> There is a newer preempt voluntary in later 2.6.8rcX kernels which fix
> many issues according to Igno, so...whatever just stick to preempt I
> guess the whole point of preempt voluntary is to improve desktop
> performance...
> 
> > I hope that if anyone else runs across this problem that this will help
> > them and that perhaps it could help lead to a bug fix for this.
> > 
> 
> It's probably already been fixed.
> 
> > 
> > -brian
> > 
> > 
> 
> -sb
> -- 
> Stan Bubrouski <stan at ccs.neu.edu>
>