[Freedreno] [PATCH 2/3] drm/msm: add DRM_MSM_GEM_SYNC_CACHE for non-coherent cache maintenance

[Jonathan Marek]

On 10/5/20 4:29 AM, Christoph Hellwig wrote:
> On Fri, Oct 02, 2020 at 08:46:35AM -0400, Jonathan Marek wrote:
>>>> +void msm_gem_sync_cache(struct drm_gem_object *obj, uint32_t flags,
>>>> +		size_t range_start, size_t range_end)
>>>> +{
>>>> +	struct msm_gem_object *msm_obj = to_msm_bo(obj);
>>>> +
>>>> +	/* TODO: sync only the required range, and don't invalidate on clean */
>>>> +
>>>> +	if (flags & MSM_GEM_SYNC_CACHE_CLEAN)
>>>> +		sync_for_device(msm_obj);
>>>> +
>>>> +	if (flags & MSM_GEM_SYNC_CACHE_INVALIDATE)
>>>> +		sync_for_cpu(msm_obj);
>>>
>>> And make to these ones as well.  They are complete abuses of the DMA
>>> API, and while we had to live with that for now to not cause regressions
>>> they absoutely must not be exposed in a userspace ABI like this.
>>>
>>
>> How do you propose that cached non-coherent memory be implemented? It is a
>> useful feature for userspace.
> 
> If the driver is using the DMA API you need to use dma_alloc_noncoherent
> and friends as of 5.10 (see the iommu list for the discussion).
> 
> If you use the raw IOMMU API (which I think the msm drm driver does) you
> need to work with the maintainers to implement a cache synchronization
> API that is not tied to the DMA API.
> 

The cache synchronization doesn't have anything to do with IOMMU (for 
example: cache synchronization would be useful in cases where drm/msm 
doesn't use IOMMU).

What is needed is to call arch_sync_dma_for_{cpu,device} (which is what 
I went with initially, but then decided to re-use drm/msm's 
sync_for_{cpu,device}). But you are also saying those functions aren't 
for driver use, and I doubt IOMMU maintainers will want to add wrappers 
for these functions just to satisfy this "not for driver use" requirement.