Re: agp/intel: Fix the pre-9xx chipset flush

Re: agp/intel: Fix the pre-9xx chipset flush

[Jan Beulich]

Eric,

in this patch you removed from intel_i830_setup_flush() a call to
map_page_into_agp(), but you left in place the counterpart call from
intel_i830_fini_flush() to unmap_page_from_agp()? Isn't it mere luck
that on x86 this is only a call to set_pages_wb()?

Further, is there really a point in doing the respective allocation
with GFP_DMA32? Nor does __GFP_ZERO seem really necessary...

Thanks, Jan

Re: agp/intel: Fix the pre-9xx chipset flush

[Eric Anholt]

[-- Attachment #1: Type: text/plain, Size: 882 bytes --]

On Thu, 16 Sep 2010 13:40:48 +0100, "Jan Beulich" <JBeulich@novell.com> wrote:
> Eric,
> 
> in this patch you removed from intel_i830_setup_flush() a call to
> map_page_into_agp(), but you left in place the counterpart call from
> intel_i830_fini_flush() to unmap_page_from_agp()? Isn't it mere luck
> that on x86 this is only a call to set_pages_wb()?

Sounds like just an oversight.

> Further, is there really a point in doing the respective allocation
> with GFP_DMA32? Nor does __GFP_ZERO seem really necessary...

The hardware in question doesn't have memory above 32 bits, so either
way doesn't matter much.  Yeah, GFP_ZERO doesn't seem needed.

If you've got patches for agp, Chris Wilson's the kernel module
maintainer now, and Daniel Vetter's got a huge changeset queued to -next
to clean up the disaster that is the "agp" code for integrated graphics.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]