[Nouveau] [PATCH 4/6] drm: enable big page mapping for small pages when IOMMU is available

[Alexandre Courbot]

On Fri, Apr 17, 2015 at 4:28 PM, Vince Hsu <vinceh at nvidia.com> wrote:
>
>
> On 04/17/2015 02:26 PM, Alexandre Courbot wrote:
>>
>> On Thu, Apr 16, 2015 at 8:06 PM, Vince Hsu <vinceh at nvidia.com> wrote:
>>>
>>> Some platforms have IOMMU to map non-contiguous physical memory into
>>> contiguous GPU virtual address. We can use this feature to enable big
>>> pages
>>> mapping on scattered small pages. To achieve that, we also need changes
>>> in
>>> subdev/mmu as well.
>>>
>>> Signed-off-by: Vince Hsu <vinceh at nvidia.com>
>>> ---
>>>   drm/nouveau/nouveau_bo.c | 9 +++++++++
>>>   1 file changed, 9 insertions(+)
>>>
>>> diff --git a/drm/nouveau/nouveau_bo.c b/drm/nouveau/nouveau_bo.c
>>> index 77326e344dad..da76ee1121e4 100644
>>> --- a/drm/nouveau/nouveau_bo.c
>>> +++ b/drm/nouveau/nouveau_bo.c
>>> @@ -221,6 +221,11 @@ nouveau_bo_new(struct drm_device *dev, int size, int
>>> align,
>>>          if (drm->client.vm) {
>>>                  if (!(flags & TTM_PL_FLAG_TT) && size > 256 * 1024)
>>>                          nvbo->page_shift =
>>> drm->client.vm->mmu->lpg_shift;
>>> +
>>> +               if ((flags & TTM_PL_FLAG_TT) &&
>>> +                               drm->client.vm->mmu->iommu_capable &&
>>> +                               (size % (1 <<
>>> drm->client.vm->mmu->lpg_shift)) == 0)
>>> +                       nvbo->page_shift =
>>> drm->client.vm->mmu->lpg_shift;
>>
>> I wonder if we should not just use the same size heuristics as for VRAM
>> above?
>>
>> Here, unless your buffer size is an exact multiple of the big page
>> size (128K for GK20A), you will not use big pages at all. In effect,
>> many buffers will be rejected for this reason. A behavior like "if the
>> buffer size of more than 256KB, increase the size of the buffer to the
>> next multiple of 128K and use big pages" would probably yield better
>> results.
>
> I'm told that the user space would align the buffer to the big page size. So
> I made this patch based on that impression. I'm not pretty familiar with the
> user space behavior, so sorry if I say something wrong here.

We would have to add this logic into Mesa, and only for GPUs that
feature no VRAM. Not impossible, but let's gather more informed
opinions about this.