[RFC PATCH 2/3] mmc: host: omap_hsmmc: Enable ADMA2

Felipe Balbi balbi at kernel.org
Mon May 23 00:18:54 PDT 2016


Hi Kishon,

Kishon Vijay Abraham I <kishon at ti.com> writes:
> Hi Felipe,
>
> On Friday 20 May 2016 12:06 AM, Felipe Balbi wrote:
>> 
>> Hi,
>> 
>> Tony Lindgren <tony at atomide.com> writes:
>>> * Peter Ujfalusi <peter.ujfalusi at ti.com> [160519 01:10]:
>>>> On 05/18/2016 10:30 PM, Tony Lindgren wrote:
>>>>> Ideally the adma support would be a separate loadable module,
>>>>> similar how the cppi41dma is a child of the OTG controller.
>>>>
>>>> The Master DMA is part of the hsmmc IP block. If the same ADMA module is
>>>> present on other IPs it might be beneficial to have a helper library to handle
>>>> it (allocating the descriptor pool, wrinting, updating descriptors, etc).
>>>
>>> OK. Yeah if it's part of the MMC controller it makes no sense to
>>> separate it. So then the conecrns are using alternate DMA
>>> implementations and keeping PM runtime working :)
>>>
>>> BTW, Felipe mentioned that the best thing to do in the long run would
>>> be to set up sdhci-omap.c operating in ADMA mode.
>>>
>>> Felipe, care to summarize what you had in mind?
>> 
>> yeah, just write a new sdhci-omap.c to start moving away from
>> omap-hsmmc.c, just like it was done for 8250-omap.
>> 
>> At the beginning, it could be just the bare minimum to get it working
>> and slowly move over stuff like pm runtime, dmaengine, PIO. Move more
>> platforms over to that driver and, eventually, get rid of omap-hsmmc.c
>> altogether.
>> 
>> That way, development can be focussed on generic layers (SDHCI) to which
>> OMAP MMC controller is compliant (apart from the VERSION register
>> quirk).
>
> About an year back, when I tried using SDHCI for OMAP I ran into
> issues and was not able to get it working. IIRC SDHCI_PRESENT_STATE
> (or OMAP_HSMMC_PSTATE) was not showing the correct state for card
> present and is unable to raise an interrupt when a card is inserted. I
> didn't debug this further.

I'd say this is a bug in hsmmc. I remember seeing some bits in some
TI-specific register (before SDHCI address space starts) which can be
used to keep parts of SDHCI powered on exactly so normal WP and CD pins
work as expected.

In any case, adding support for GPIO-based card detect to generic SDHCI
shouldn't be too difficult :-)

> It also kept me wondering why gpio interrupt was always used for card
> detect instead of using mmci_sdcd line of the controller.

Probably a really, really old bug which nobody ever debugged properly ;-)

ps: you don't need that ADMA2 DT property, btw. There's a bit in another
register which you can check if $this controller was configured with
ADMA2 support or not. IIRC, OMAP5's TRM describes them.

-- 
balbi



More information about the linux-arm-kernel mailing list