Re: [patch 0/4] gpu/drm, x86, PAT: Add PAT support for io_mapping_* interface

Re: [patch 0/4] gpu/drm, x86, PAT: Add PAT support for io_mapping_* interface

[Ingo Molnar]

* Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> wrote:

> io_mapping_create_wc(), io_mapping_free() and associated map/unmap routines
> are relatively new set of APIs for drivers, that change memory type.
> 
> Currently, these interfaces hardcode WC type on x86 and does not go through
> "PAT enforcement". This patchset adds PAT support for this interface.
> 
> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>

hm, the patches look sane and clean and mostly factor out 
existing code and checks and apply them to the io_mapping_*() 
interfaces - so it's something worth having for .29.

But it's all rather awfully late in the cycle and ioremap type 
of bugs typically have 1-2 rc's of latency to be fully 
discovered.

They will probably fix a few regressions and are easy to revert, 
plus they affect the one driver whose authors wrote these 
facilities:

  drivers/gpu/drm/i915/i915_dma.c
  drivers/gpu/drm/i915/i915_gem.c

so maybe we should try it as it's all a clear step forward. I'll 
give them a spin in tip:x86/urgent. Dunno.

Linus, do you have any preference?

	Ingo

RE: [patch 0/4] gpu/drm, x86, PAT: Add PAT support for io_mapping_* interface

[Pallipadi, Venkatesh]

 

>-----Original Message-----
>From: Ingo Molnar [mailto:mingo@elte.hu] 
>Sent: Wednesday, February 25, 2009 4:09 AM
>To: Pallipadi, Venkatesh; Linus Torvalds
>Cc: Dave Airlie; Jesse Barnes; Eric Anholt; Thomas Gleixner; H 
>Peter Anvin; Keith Packard; linux-kernel@vger.kernel.org
>Subject: Re: [patch 0/4] gpu/drm, x86, PAT: Add PAT support 
>for io_mapping_* interface
>
>
>* Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> wrote:
>
>> io_mapping_create_wc(), io_mapping_free() and associated 
>map/unmap routines
>> are relatively new set of APIs for drivers, that change memory type.
>> 
>> Currently, these interfaces hardcode WC type on x86 and does 
>not go through
>> "PAT enforcement". This patchset adds PAT support for this interface.
>> 
>> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
>> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
>
>hm, the patches look sane and clean and mostly factor out 
>existing code and checks and apply them to the io_mapping_*() 
>interfaces - so it's something worth having for .29.
>
>But it's all rather awfully late in the cycle and ioremap type 
>of bugs typically have 1-2 rc's of latency to be fully 
>discovered.
>
>They will probably fix a few regressions and are easy to revert, 
>plus they affect the one driver whose authors wrote these 
>facilities:
>
>  drivers/gpu/drm/i915/i915_dma.c
>  drivers/gpu/drm/i915/i915_gem.c
>
>so maybe we should try it as it's all a clear step forward. I'll 
>give them a spin in tip:x86/urgent. Dunno.
>

Actual PAT support (patch 3 and 4) can probably wait until .30. As
there is only one user of this interface. Patch 1 and 2 are more in
bugfix category and probably considered for .29.

Thanks,
Venki

Re: [patch 0/4] gpu/drm, x86, PAT: Add PAT support for io_mapping_* interface

[Ingo Molnar]

* Pallipadi, Venkatesh <venkatesh.pallipadi@intel.com> wrote:

> >They will probably fix a few regressions and are easy to 
> >revert, plus they affect the one driver whose authors wrote 
> >these facilities:
> >
> >  drivers/gpu/drm/i915/i915_dma.c
> >  drivers/gpu/drm/i915/i915_gem.c
> >
> >so maybe we should try it as it's all a clear step forward. 
> >I'll give them a spin in tip:x86/urgent. Dunno.
> 
> Actual PAT support (patch 3 and 4) can probably wait until 
> .30. As there is only one user of this interface. Patch 1 and 
> 2 are more in bugfix category and probably considered for .29.

ok, i've delayed #3/#4 to .30 and kept #1/#2 in tip:x86/urgent. 
No complications so far with any of them.

	Ingo