Nouveau - Apr 2014 - [PATCH v2 04/10] drm/nouveau/fb: add GK20A support

On Wed, Apr 23, 2014 at 11:07 AM, Alexandre Courbot <acourbot at
nvidia.com> wrote:
> On 04/22/2014 07:40 PM, Thierry Reding wrote:
>>
>> * PGP Signed by an unknown key
>>
>>
>> On Mon, Apr 21, 2014 at 03:02:16PM +0900, Alexandre Courbot wrote:
>> [...]
>>>
>>> diff --git a/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
>>> b/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
>>
>> [...]
>>>
>>> +               pages = dma_alloc_from_contiguous(dev, ncmin,
order);
>>> +               if (!pages) {
>>> +                       gk20a_ram_put(pfb, &mem);
>>> +                       return -ENOMEM;
>>> +               }
>>> +
>>> +               dma_addr =
pfn_to_dma(nv_device_base(nv_device(pfb)),
>>> +                                     page_to_pfn(pages));
>>
>>
>> This breaks compilation on x86 because neither pfn_to_dma() nor
>> dma_to_pfn() are available. Is there some other way this can be
>> allocated so that these functions don't need to be called?
>
>
> Mmm, this is bad. There is probably another more portable way to do this.
> Let me look for it.
page_to_phys()/phys_to_page() can be used by drivers and will work
just fine here since the CPU and GPU use the same physical addresses
to access memory.

Thanks,
Alex.

On Wed, Apr 23, 2014 at 03:11:01PM +0900, Alexandre Courbot
wrote:
> On Wed, Apr 23, 2014 at 11:07 AM, Alexandre Courbot <acourbot at
nvidia.com> wrote:
> > On 04/22/2014 07:40 PM, Thierry Reding wrote:
> >>
> >> * PGP Signed by an unknown key
> >>
> >>
> >> On Mon, Apr 21, 2014 at 03:02:16PM +0900, Alexandre Courbot wrote:
> >> [...]
> >>>
> >>> diff --git a/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
> >>> b/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
> >>
> >> [...]
> >>>
> >>> +               pages = dma_alloc_from_contiguous(dev, ncmin,
order);
> >>> +               if (!pages) {
> >>> +                       gk20a_ram_put(pfb, &mem);
> >>> +                       return -ENOMEM;
> >>> +               }
> >>> +
> >>> +               dma_addr =
pfn_to_dma(nv_device_base(nv_device(pfb)),
> >>> +                                     page_to_pfn(pages));
> >>
> >>
> >> This breaks compilation on x86 because neither pfn_to_dma() nor
> >> dma_to_pfn() are available. Is there some other way this can be
> >> allocated so that these functions don't need to be called?
> >
> >
> > Mmm, this is bad. There is probably another more portable way to do
this.
> > Let me look for it.
> 
> page_to_phys()/phys_to_page() can be used by drivers and will work
> just fine here since the CPU and GPU use the same physical addresses
> to access memory.
I'm wondering how this is going to pan out when we try adding IOMMU
support. But I guess we can cross that bridge when we come to it.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: not available
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20140428/e952897e/attachment-0001.sig>

On Mon, Apr 28, 2014 at 8:44 PM, Thierry Reding
<thierry.reding at gmail.com> wrote:
> On Wed, Apr 23, 2014 at 03:11:01PM +0900, Alexandre Courbot wrote:
>> On Wed, Apr 23, 2014 at 11:07 AM, Alexandre Courbot <acourbot at
nvidia.com> wrote:
>> > On 04/22/2014 07:40 PM, Thierry Reding wrote:
>> >>
>> >> * PGP Signed by an unknown key
>> >>
>> >>
>> >> On Mon, Apr 21, 2014 at 03:02:16PM +0900, Alexandre Courbot
wrote:
>> >> [...]
>> >>>
>> >>> diff --git
a/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
>> >>> b/drivers/gpu/drm/nouveau/core/subdev/fb/ramgk20a.c
>> >>
>> >> [...]
>> >>>
>> >>> +               pages = dma_alloc_from_contiguous(dev,
ncmin, order);
>> >>> +               if (!pages) {
>> >>> +                       gk20a_ram_put(pfb, &mem);
>> >>> +                       return -ENOMEM;
>> >>> +               }
>> >>> +
>> >>> +               dma_addr =
pfn_to_dma(nv_device_base(nv_device(pfb)),
>> >>> +                                     page_to_pfn(pages));
>> >>
>> >>
>> >> This breaks compilation on x86 because neither pfn_to_dma()
nor
>> >> dma_to_pfn() are available. Is there some other way this can
be
>> >> allocated so that these functions don't need to be called?
>> >
>> >
>> > Mmm, this is bad. There is probably another more portable way to
do this.
>> > Let me look for it.
>>
>> page_to_phys()/phys_to_page() can be used by drivers and will work
>> just fine here since the CPU and GPU use the same physical addresses
>> to access memory.
>
> I'm wondering how this is going to pan out when we try adding IOMMU
> support. But I guess we can cross that bridge when we come to it.
Agreed. Besides I hope the day won't come where we have to go through
2 layers of memory translation for the GPU...