Nouveau - Aug 2019 - [PATCH 15/15] amdgpu: remove CONFIG_DRM_AMDGPU_USERPTR

On Tue, Aug 6, 2019 at 1:51 PM Kuehling, Felix <Felix.Kuehling at amd.com>
wrote:
>
> On 2019-08-06 13:44, Jason Gunthorpe wrote:
> > On Tue, Aug 06, 2019 at 07:05:53PM +0300, Christoph Hellwig wrote:
> >> The option is just used to select HMM mirror support and has a
very
> >> confusing help text.  Just pull in the HMM mirror code by default
> >> instead.
> >>
> >> Signed-off-by: Christoph Hellwig <hch at lst.de>
> >> ---
> >>   drivers/gpu/drm/Kconfig                 |  2 ++
> >>   drivers/gpu/drm/amd/amdgpu/Kconfig      | 10 ----------
> >>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  6 ------
> >>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 12 ------------
> >>   4 files changed, 2 insertions(+), 28 deletions(-)
> > Felix, was this an effort to avoid the arch restriction on hmm or
> > something? Also can't see why this was like this.
>
> This option predates KFD's support of userptrs, which in turn predates
> HMM. Radeon has the same kind of option, though it doesn't affect HMM
in
> that case.
>
> Alex, Christian, can you think of a good reason to maintain userptr
> support as an option in amdgpu? I suspect it was originally meant as a
> way to allow kernels with amdgpu without MMU notifiers. Now it would
> allow a kernel with amdgpu without HMM or MMU notifiers. I don't know
if
> this is a useful thing to have.
Right.  There were people that didn't have MMU notifiers that wanted
support for the GPU.  For a lot of older APIs, a lack of userptr
support was not a big deal (it just disabled some optimizations and
API extensions), but as it becomes more relevant it may make sense to
just make it a requirement.

Alex
>
> Regards,
>    Felix
>
> >
> > Reviewed-by: Jason Gunthorpe <jgg at mellanox.com>
> >
> > Jason
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

On Tue, Aug 06, 2019 at 02:58:58PM -0400, Alex Deucher
wrote:
> On Tue, Aug 6, 2019 at 1:51 PM Kuehling, Felix <Felix.Kuehling at
amd.com> wrote:
> >
> > On 2019-08-06 13:44, Jason Gunthorpe wrote:
> > > On Tue, Aug 06, 2019 at 07:05:53PM +0300, Christoph Hellwig
wrote:
> > >> The option is just used to select HMM mirror support and has
a very
> > >> confusing help text.  Just pull in the HMM mirror code by
default
> > >> instead.
> > >>
> > >> Signed-off-by: Christoph Hellwig <hch at lst.de>
> > >>   drivers/gpu/drm/Kconfig                 |  2 ++
> > >>   drivers/gpu/drm/amd/amdgpu/Kconfig      | 10 ----------
> > >>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  6 ------
> > >>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 12 ------------
> > >>   4 files changed, 2 insertions(+), 28 deletions(-)
> > > Felix, was this an effort to avoid the arch restriction on hmm or
> > > something? Also can't see why this was like this.
> >
> > This option predates KFD's support of userptrs, which in turn
predates
> > HMM. Radeon has the same kind of option, though it doesn't affect
HMM in
> > that case.
> >
> > Alex, Christian, can you think of a good reason to maintain userptr
> > support as an option in amdgpu? I suspect it was originally meant as a
> > way to allow kernels with amdgpu without MMU notifiers. Now it would
> > allow a kernel with amdgpu without HMM or MMU notifiers. I don't
know if
> > this is a useful thing to have.
> 
> Right.  There were people that didn't have MMU notifiers that wanted
> support for the GPU.
?? Is that even a real thing? mmu_notifier does not have much kconfig
dependency.

Jason

Am 06.08.19 um 22:03 schrieb Jason Gunthorpe:
> On Tue, Aug 06, 2019 at 02:58:58PM -0400, Alex Deucher wrote:
>> On Tue, Aug 6, 2019 at 1:51 PM Kuehling, Felix <Felix.Kuehling at
amd.com> wrote:
>>> On 2019-08-06 13:44, Jason Gunthorpe wrote:
>>>> On Tue, Aug 06, 2019 at 07:05:53PM +0300, Christoph Hellwig
wrote:
>>>>> The option is just used to select HMM mirror support and
has a very
>>>>> confusing help text.  Just pull in the HMM mirror code by
default
>>>>> instead.
>>>>>
>>>>> Signed-off-by: Christoph Hellwig <hch at lst.de>
>>>>>    drivers/gpu/drm/Kconfig                 |  2 ++
>>>>>    drivers/gpu/drm/amd/amdgpu/Kconfig      | 10 ----------
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  6 ------
>>>>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 12
------------
>>>>>    4 files changed, 2 insertions(+), 28 deletions(-)
>>>> Felix, was this an effort to avoid the arch restriction on hmm
or
>>>> something? Also can't see why this was like this.
>>> This option predates KFD's support of userptrs, which in turn
predates
>>> HMM. Radeon has the same kind of option, though it doesn't
affect HMM in
>>> that case.
>>>
>>> Alex, Christian, can you think of a good reason to maintain userptr
>>> support as an option in amdgpu? I suspect it was originally meant
as a
>>> way to allow kernels with amdgpu without MMU notifiers. Now it
would
>>> allow a kernel with amdgpu without HMM or MMU notifiers. I
don't know if
>>> this is a useful thing to have.
>> Right.  There were people that didn't have MMU notifiers that
wanted
>> support for the GPU.
> ?? Is that even a real thing? mmu_notifier does not have much kconfig
> dependency.
Yes, that used to be a very real thing.

Initially a lot of users didn't wanted mmu notifiers to be enabled 
because of the performance overhead they costs.

Then we had the problem that HMM mirror wasn't available on a lot of 
architectures.

Not sure if we still need it separately, but I think that it shouldn't 
be removed without asking around if somebody still needs that.

Christian.
>
> Jason