[PATCH] mci: dw_mmc: set dma mask to work correctly on 64bit platforms

[Ahmad Fatoum]

Hi,

On 11.06.21 12:02, Sascha Hauer wrote:
> On Fri, Jun 11, 2021 at 10:23:25AM +0200, Ahmad Fatoum wrote:
>> Hello Sascha,
>>
>> On 10.06.21 15:10, Sascha Hauer wrote:
>>> The variant we support currently can only do 32bit DMA. Adjust dma mask
>>> accordingly. Also use dma_map_single() rather than dma_sync_single() to
>>> actually get errors when the mapping fails.
>>>
>>> Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
>>
>> Tested-by: Ahmad Fatoum <a.fatoum at pengutronix.de>
>>
>> On a StarFive JH7100 (64-bit CPU, 32-bit MMC controller, non-1:1 mapping for
>> coherent DMA).
>>
>>>  	host->idmac = dma_alloc_coherent(sizeof(*host->idmac) * DW_MMC_NUM_IDMACS,
>>> -					 DMA_ADDRESS_BROKEN);
>>> +					 &host->idmac_dma);
>>> +	if (!host->idmac)
>>
>> That works for Linux, but not for barebox: barebox dma_alloc_coherent doesn't have
>>
>> a dev parameter, so it can't check for dma_mapping_error() internally.
> 
> Yeah, I know. I have that on my mental todo list and hoped nobody would
> realize.

Tss. tss. ;)

> Adding a dev parameter to dma_alloc_coherent() is one thing. With that
> we can check for errors. The next step of course would be to allocate
> memory in the allowed area, not only to complain.

What you can do for now is checking for dma_mapping_error here and abort
the probe if you exceed the mask.

> 
>>
>> dma_alloc_coherent also never returns NULL in barebox, all implementations, except
>>
>> for kvx, abort if no memory could be allocated.
>>
>>> +		return PTR_ERR(-ENOMEM);
>>
>> -ENOMEM is no pointer.
> 
> Yes, fixed.
> 
> Sascha
> 

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |