Re: [PATCH] mmc: sdhci-pci: Try "cd" for card-detect lookup before using NULL

[Andy Shevchenko]

On Wed, Oct 24, 2018 at 9:03 PM Dmitry Torokhov <dtor@xxxxxxxxxx> wrote:
> On Wed, Oct 24, 2018 at 3:02 AM Andy Shevchenko
> <andy.shevchenko@xxxxxxxxx> wrote:
> > On Mon, Oct 22, 2018 at 04:34:55PM -0700, Rajat Jain wrote:
> > > On Fri, Oct 19, 2018 at 2:13 AM Andy Shevchenko
> > > <andy.shevchenko@xxxxxxxxx> wrote:
> > > > On Fri, Oct 19, 2018 at 12:53 AM Rajat Jain <rajatja@xxxxxxxxxx> wrote:
> >
> > > > > across other users of this API (other MMC host controller drivers).
> > > >
> > > > >         if (slot->cd_idx >= 0) {
> > > > > -               ret = mmc_gpiod_request_cd(host->mmc, NULL, slot->cd_idx,
> > > > > +               ret = mmc_gpiod_request_cd(host->mmc, "cd", slot->cd_idx,
> > > > >                                            slot->cd_override_level, 0, NULL);
> > > >
> > > > Yes.
> > > >
> > > > > +               if (ret && ret != -EPROBE_DEFER)
> > > > > +                       ret = mmc_gpiod_request_cd(host->mmc, NULL,
> > > > > +                                                  slot->cd_idx,
> > > > > +                                                  slot->cd_override_level,
> > > > > +                                                  0, NULL);
> > > >
> > > > And no. Instead of this part you need to provide an ACPI GPIO mapping table.
> > >
> > > Sure, I am willing to do so, and I tried earlier too. However, certain
> > > doubts arose in my mind when I tried that and I posted my questions
> > > earlier (https://lkml.org/lkml/2018/9/28/507) but couldn't elicit any
> > > response. Unfortunately I still do not have answers. My primary
> > > questions are:
> > >
> > > 1) - It seems that 1 SDHCI device may support multiple slots (looking
> > > at the code). It is not clear to me if they could share card detect
> > > interrupts, or should have separate ones?
> >
> > This is more likely question to HW engineers of your platform with a caveat
> > that there should be a way to distinguish exact slot in which card is being
> > inserted.
> >
> > > Also, the driver may not
> > > really know?
> >
> > I think in such case the bug in HW design and / or driver.
>
> Why? You can have a shared or dedicated interrupt and the driver does
> not really need to know if it can poll the status.

Yes, that's my point either we get 1:1 mapping between slot and GPIOs
or have a possibility to read back from some register(s) the actual
status of all of them, otherwise it's a bad design. Sorry if I wasn't
clear about it.

> > > So should I add 1 or two pins using the
> > > devm_acpi_dev_add_driver_gpios().
> >
> > This depends on the above, e.g. HW design, ACPI tables.
>
> Yes, it depends on the HW design and that is exactly why the approach
> with devm_acpi_dev_add_driver_gpios() does not work well here: this is
> a generic driver used on many platforms and you are trying to put the
> platform knowledge into the driver. Here we are lucky I guess as I do
> not believe anyone is using more than one slot, so we can have a tavle
> with a single entry, but actually doing the fallback the way Rajat was
> proposing is more correct. Or you have a table with N entries, where N
> is hopefully sufficiently large.

Yes, unfortunately this is the case. We need to keep somewhere the
list to support old firmwares (see hci_bcm.c as an example how BIOS
can screw things up).
Soonish we start _DSD in BIOSes in a correct way (ha-ha), better for everyone.

> > > Is some one familiar with SDHC
> > > driver can answer these questions, it shall be great.
> >
> > Actually above questions better to ask in linux-mmc mailing list, which by the
> > fact is in Cc list already. So, wait for someone to clarify.
> >
> >
> > > 2) I'm not really sure what should I set "active_low" to? Isn't this
> > > something that should be specified by platform / ACPI too, and driver
> > > should just be able to say say choose whatever the ACPI says?
> > >
> > > struct acpi_gpio_params {
> > >         unsigned int crs_entry_index;
> > >         unsigned int line_index;
> > >         bool active_low;
> > > };
> >
> >
> > ACPI specification misses this property, that's why we have it in the
> > structure. In your case it should be provided by _DSD and thus be consistent
> > with the hardcoded values.
>
> Again, you think as if the driver was platform specific; it is not. I
> have 1000s of systems with different ACPI tables. Let's say half of
> them use one polarity, and half another. Which polarity do you propose
> to use?

Use one table for one half and another for the rest.

-- 
With Best Regards,
Andy Shevchenko