[PATCH v3 5/7] drm/sun4i: Rely on dma interconnect for our RAM offset

[Robin Murphy]

On 19/02/2019 10:55, Maxime Ripard wrote:
> Hi Robin,
> 
> On Wed, Feb 13, 2019 at 04:40:11PM +0000, Robin Murphy wrote:
>> On 13/02/2019 15:41, Maxime Ripard wrote:
>>> Hi Robin,
>>>
>>> Thanks for your feedback!
>>>
>>> On Tue, Feb 12, 2019 at 06:46:40PM +0000, Robin Murphy wrote:
>>>> On 11/02/2019 15:02, Maxime Ripard wrote:
>>>>> Now that we can express our DMA topology, rely on those property instead of
>>>>> hardcoding an offset from the dma_addr_t which wasn't really great.
>>>>>
>>>>> We still need to add some code to deal with the old DT that would lack that
>>>>> property, but we move the offset to the DRM device dma_pfn_offset to be
>>>>> able to rely on just the dma_addr_t associated to the GEM object.
>>>>>
>>>>> Acked-by: Daniel Vetter <daniel.vetter at ffwll.ch>
>>>>> Signed-off-by: Maxime Ripard <maxime.ripard at bootlin.com>
>>>>> ---
>>>>>     drivers/gpu/drm/sun4i/sun4i_backend.c | 28 +++++++++++++++++++++-------
>>>>>     1 file changed, 21 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/sun4i/sun4i_backend.c b/drivers/gpu/drm/sun4i/sun4i_backend.c
>>>>> index 9e9255ee59cd..1846a1b30fea 100644
>>>>> --- a/drivers/gpu/drm/sun4i/sun4i_backend.c
>>>>> +++ b/drivers/gpu/drm/sun4i/sun4i_backend.c
>>>>> @@ -383,13 +383,6 @@ int sun4i_backend_update_layer_buffer(struct sun4i_backend *backend,
>>>>>     	paddr = drm_fb_cma_get_gem_addr(fb, state, 0);
>>>>>     	DRM_DEBUG_DRIVER("Setting buffer address to %pad\n", &paddr);
>>>>> -	/*
>>>>> -	 * backend DMA accesses DRAM directly, bypassing the system
>>>>> -	 * bus. As such, the address range is different and the buffer
>>>>> -	 * address needs to be corrected.
>>>>> -	 */
>>>>> -	paddr -= PHYS_OFFSET;
>>>>> -
>>>>>     	if (fb->format->is_yuv)
>>>>>     		return sun4i_backend_update_yuv_buffer(backend, fb, paddr);
>>>>> @@ -835,6 +828,27 @@ static int sun4i_backend_bind(struct device *dev, struct device *master,
>>>>>     	dev_set_drvdata(dev, backend);
>>>>>     	spin_lock_init(&backend->frontend_lock);
>>>>> +	if (of_find_property(dev->of_node, "interconnects", NULL)) {
>>>>> +		/*
>>>>> +		 * This assume we have the same DMA constraints for all our the
>>>>> +		 * devices in our pipeline (all the backends, but also the
>>>>> +		 * frontends). This sounds bad, but it has always been the case
>>>>> +		 * for us, and DRM doesn't do per-device allocation either, so
>>>>> +		 * we would need to fix DRM first...
>>>>> +		 */
>>>>> +		ret = of_dma_configure(drm->dev, dev->of_node, true);
>>>>
>>>> It would be even nicer if we could ensure that drm->dev originates from a DT
>>>> node which has the appropriate interconnects property itself, such that we
>>>> can assume it's already configured correctly.
>>>
>>> The thing is drm->dev comes from a node in the DT that is a virtual
>>> node, and therefore doesn't have any resources attached, so I'm not
>>> sure we have any other way, unfortunately.
>>
>> Right, I appreciate that it may not be feasible to swizzle drm->dev for one
>> of the 'real' component devices, but what I was also thinking was that since
>> the virtual device node effectively represents the aggregation of the other
>> component devices, we could just say that it also has to have its own link
>> to the MBUS interconnect (with the ID of the pipeline entrypoint it's
>> associated with, I guess). That ought to be enough to get dma_configure() to
>> do the job, and in fairness is no *less* accurate a description of the
>> hardware, even if might look a little funky to some.
> 
> That virtual device however can work with up to 4 devices that can
> perform DMA operations, all of them going through a different port of
> the MBUS controller that can account for DMA usage on a per-master
> basis.
> 
> Eventually, we should support that feature as well, but the point is
> that from a DT point of view, there's not a single point of DMA access
> for that particular device but more likely 2-4 points, each with a
> different argument to the phandle.
> 
> We could of course have a hack and use only one of them, but that
> would be less accurate than what we currently have. Plus, this is
> really due to a restriction on the DRM side, that could change in the
> future (even though it's unlikely). Fixing that restriction would make
> that property completely useless, confusing and wrong from an hardware
> PoV.

Thanks for the clarification. Much as I'd like to keep 
of_dma_configure() out of drivers as much as possible, I agree that it's 
not sufficient justification for (ab)using DT to sidestep a 
Linux-specific issue which might eventually get properly fixed anyway. 
I'm sure the last thing sunxi needs is any more awkward DT ABI concerns :)

Robin.