Re: [PATCH] drm/sun4i: Add DMA mask and segment size

[Samuel Holland <samuel@sholland.org>]

On 6/16/22 11:13 PM, Jernej Škrabec wrote:
> Dne petek, 17. junij 2022 ob 05:03:11 CEST je Samuel Holland napisal(a):
>> Hi Jernej,
>>
>> On 6/16/22 4:32 PM, Jernej Skrabec wrote:
>>> Kernel occasionally complains that there is mismatch in segment size
>>> when trying to render HW decoded videos and rendering them directly with
>>> sun4i DRM driver.
>>>
>>> Fix that by setting DMA mask and segment size.
>>>
>>> Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
>>> ---
>>>
>>>  drivers/gpu/drm/sun4i/sun4i_drv.c | 4 ++++
>>>  1 file changed, 4 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/sun4i/sun4i_drv.c
>>> b/drivers/gpu/drm/sun4i/sun4i_drv.c index 275f7e4a03ae..83f4e87f77f6
>>> 100644
>>> --- a/drivers/gpu/drm/sun4i/sun4i_drv.c
>>> +++ b/drivers/gpu/drm/sun4i/sun4i_drv.c
>>> @@ -7,6 +7,7 @@
>>>
>>>   */
>>>  
>>>  #include <linux/component.h>
>>>
>>> +#include <linux/dma-mapping.h>
>>>
>>>  #include <linux/kfifo.h>
>>>  #include <linux/module.h>
>>>  #include <linux/of_graph.h>
>>>
>>> @@ -367,6 +368,9 @@ static int sun4i_drv_probe(struct platform_device
>>> *pdev)> 
>>>  	INIT_KFIFO(list.fifo);
>>>
>>> +	dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
>>
>> Isn't this already the default, from of_dma_configure_id or
>> setup_pdev_dma_masks?
> 
> Not sure, I need to check.

Looking at the DE2 and DE3 spec to answer your question, I noticed that all of
the address fields have a corresponding 8-bit "high address" field, so from a
hardware perspective, the DMA mask should be 40 bits.

However, currently we do not use these fields, so we are limited to 32 bits. I
would suggest we comment that, so I am fine with setting the mask even if
redundant, as it provides a good place to add the comment.

>>> +	dma_set_max_seg_size(&pdev->dev, DMA_BIT_MASK(32));
>>
>> This looks like a good change. In fact, I think we need a similar change in
>> some other drivers.
> 
> Should be DMA_BIT_MASK(25) as in your other patch?

No, that limit is specific to the other DMA engine. I do not see any limit for
the display engine. BSP uses sg_tables only for dma-buf, and then it puts the
entire framebuffer in one segment. So I think DMA_BIT_MASK(32) (the maximum
possible value) is correct.

Regards,
Samuel