[PATCH] Don't try to use agp symbols if we don't have AGP (*cough*sparc*cough*)

[Tom Callaway]

The radeon_buffer.c addition that gets pulled into the DRM driver as
part of Kernel Mode Setting (even if you disable KMS for the Radeon
driver) was assuming that it was safe to use symbols from the AGP code,
but that isn't safe on platforms where __OS_HAS_AGP is 0. Such as
sparc64. :)

This patch gets the code building again. It probably should go into the
actual kms patch, but if you would prefer this go as a separate patch, I
can apply it. With this, we can get rawhide building for sparc64, so I'm
eager to see it applied in one way or another.

Signed-off-by: Tom "spot" Callaway <tcallawa(a)redhat.com&gt;

[Josh Boyer]

On Wed, Mar 11, 2009 at 10:27:51AM -0400, Tom spot Callaway wrote:
> The radeon_buffer.c addition that gets pulled into the DRM driver as
> part of Kernel Mode Setting (even if you disable KMS for the Radeon
> driver) was assuming that it was safe to use symbols from the AGP code,
> but that isn't safe on platforms where __OS_HAS_AGP is 0. Such as
> sparc64. :)
>
> This patch gets the code building again. It probably should go into the
> actual kms patch, but if you would prefer this go as a separate patch, I
> can apply it. With this, we can get rawhide building for sparc64, so I'm
> eager to see it applied in one way or another.
>
> Signed-off-by: Tom "spot" Callaway <tcallawa(a)redhat.com&gt; 
> diff -up
> linux-2.6.28.sparc64/drivers/gpu/drm/radeon/radeon_buffer.c.BAD
> linux-2.6.28.sparc64/drivers/gpu/drm/radeon/radeon_buffer.c
> --- linux-2.6.28.sparc64/drivers/gpu/drm/radeon/radeon_buffer.c.BAD	2009-03-11
> 09:11:01.000000000 -0500
> +++ linux-2.6.28.sparc64/drivers/gpu/drm/radeon/radeon_buffer.c	2009-03-11
> 09:11:27.000000000 -0500
> @@ -37,9 +37,11 @@ struct drm_ttm_backend *radeon_create_tt
>  {
>  	drm_radeon_private_t *dev_priv = dev->dev_private;
>  
> +#if __OS_HAS_AGP 
Not that this is your fault, but shouldn't this be something
like __ARCH_HAS_AGP?

josh

[Kyle McMartin]

On Wed, Mar 11, 2009 at 11:33:25AM -0400, Josh Boyer wrote:
>  >+#if __OS_HAS_AGP
>  
>  Not that this is your fault, but shouldn't this be something
>  like __ARCH_HAS_AGP?
>   
Keep in mind the DRM for better or for worse supposedly supports more
OSes than just Linux.

I suspect the 'ideal' fix for upstream is to stub out drm_agp_init_ttm
on !CONFIG_AGP arches.

regards, Kyle

[Josh Boyer]

On Wed, Mar 11, 2009 at 11:55:53AM -0400, Kyle McMartin wrote:
> On Wed, Mar 11, 2009 at 11:33:25AM -0400, Josh Boyer wrote:
> > >+#if __OS_HAS_AGP
> > 
> > Not that this is your fault, but shouldn't this be something
> > like __ARCH_HAS_AGP?
> > 
>
> Keep in mind the DRM for better or for worse supposedly supports more
> OSes than just Linux. 
I had blissfully forgotten that.

> I suspect the 'ideal' fix for upstream is to stub out
> drm_agp_init_ttm
> on !CONFIG_AGP arches. 
Yeah, something like that.

josh

[David Airlie]

On Wed, 2009-03-11 at 11:55 -0400, Kyle McMartin wrote:
>  On Wed, Mar 11, 2009 at 11:33:25AM -0400, Josh Boyer wrote:
>  > >+#if __OS_HAS_AGP
>  > 
>  > Not that this is your fault, but shouldn't this be something
>  > like __ARCH_HAS_AGP?
>  > 
>  
>  Keep in mind the DRM for better or for worse supposedly supports more
>  OSes than just Linux.
>  
>  I suspect the 'ideal' fix for upstream is to stub out drm_agp_init_ttm
>  on !CONFIG_AGP arches. 
OS_HAS_AGP is just a wrapper around convoluted CONFIG_AGP,
CONFIG_AGP_MODULE.

Its perfectly acceptable to build a kernel with no AGP and DRM enabled
on x86.

Dave.

>  
>  regards, Kyle

[Tom Callaway]

On 03/11/2009 11:33 AM, Josh Boyer wrote:
> > +#if __OS_HAS_AGP
>  
>  Not that this is your fault, but shouldn't this be something
>  like __ARCH_HAS_AGP? 
Maybe, but I'm not about to redefine existing kernel defines for the
purposes of logic. ;)

~spot