Re: [RFC] mmc: core: transplant ti,wl1251 quirks from to be retired omap_hsmmc

["H. Nikolaus Schaller" <hns@goldelico.com>]

> Am 27.10.2021 um 23:31 schrieb Ulf Hansson <ulf.hansson@linaro.org>:
> 
> On Wed, 27 Oct 2021 at 19:01, H. Nikolaus Schaller <hns@goldelico.com> wrote:
>> 
>> Hi Ulf,
>> 
>>> Am 26.10.2021 um 20:08 schrieb H. Nikolaus Schaller <hns@goldelico.com>:
>>> 
>>> Hi Uf,
>>>> 
>>>> As a matter of fact, the similar problem that you are looking to
>>>> address (applying card quirks based on DT compatibility strings), is
>>>> partly being taken care of in another series [1], being discussed
>>>> right now. I think the solution for the ti,wl1251 should be based upon
>>>> that too. Please have a look and see if you can play with that!?
>>> 
>>> That is interesting.
>>> Yes, maybe it can be the basis. At least for finding the chip and driver.
>> 
>> I have done a first experiment.
>> 
>> It seems as if the series [1] does the opposite of what we need... It just
>> skips entries in struct mmc_fixup if the DT does *not* match.
> 
> Ohh, I didn't look that close. In that case the code isn't doing what
> it *should*. The point is really to match on the compatible string and
> then add quirks if that is found.

That is what I had expected.

> 
> Let me have a closer look - and for sure, I am willing to help if needed.
> 
>> 
>> This new match is not even tried in the wl1251 case since card->cis.vendor
>> and card->cis.device are not properly initialized when mmc_fixup_device() is called.
>> (in the upstream code the init_card function sets these and also sets MMC_QUIRK_NONSTD_SDIO
>> to early abort before sdio_read_cccr, sdio_read_common_cis, and mmc_fixup_device).
> 
> We can call mmc_fixup_device() more than once during initialization
> and provide different struct mmc_fixup* - if that is needed.

Ah, looks good.

> 
>> 
>> What I don't get from the code is how cis.vendor or cis.device can be
>> initialized from device tree for a specific device. As far as I see it can
>> only be checked for and some quirks can be set from a table if vendor and
>> device read from the CIS registers do match.
> 
> Yes. I thought that should be possible, but maybe it is not?

It seems to be a hen or egg issue here. MMC_QUIRK_NONSTD_SDIO should be set
before we can match by vendor and device or compatible. But it can't be set
late.

> 
>> 
>> Instead, we want to match DT and define some values for an otherwise unknown
>> device (i.e. we can't match by vendor or other methods) to help to initialize
>> the interface. So in mmc_fixup_device it is too late and we need something
>> running earlier, based purely on device tree information...
> 
> Okay, I will have a closer look. Maybe we need to extend the card
> quirks interface a bit to make it suitable for this case too.

Combining your suggestions we could do roughly:

in mmc_sdio_init_card():

	if (host->ops->init_card)
		host->ops->init_card(host, card);
	else
		mmc_fixup_device(host, sdio_prepare_fixups_methods);

Next we need a location for the sdio_prepare_fixups_methods table and functions.

For "ti,wl1251" we would then provide the entry in the table and a function doing
the setup. But where should these be defined? Likely not in a header file like
quirks.h? But there is no quirks.c.

Best regards,
Nikolaus