Re: [PATCH 5/6] platform/chrome: cros_ec_lpc: Correct ACPI name for Framework Laptop

[Tzung-Bi Shih]

On Mon, May 27, 2024 at 07:06:40PM +0100, Ben Walsh wrote:
> Tzung-Bi Shih <tzungbi@xxxxxxxxxx> writes:
> 
> > On Fri, May 24, 2024 at 07:35:22PM +0100, Ben Walsh wrote:
> 
> >> I could add a new quirk which provides an alternative ACPI match table
> >> to be used instead of the default. In the default case the match_table
> >> will contain only "GOOG0004" as before. But in the Framework EC case the
> >> match table will be "PNP0C09".
> >
> > I think it doesn't work as the current quirk is handling in
> > cros_ec_lpc_probe() which is after matching.
> 
> I was thinking of a new quirk called CROS_EC_LPC_QUIRK_ACPI_MATCH, and
> putting it in cros_ec_lpc_init(), not cros_ec_lpc_probe(). Do we have to
> do all quirk handling in cros_ec_lpc_probe()?

No, but there is already code in cros_ec_lpc_probe() for handling quirks and
we would like to reuse them if we could.

Also a possible issue: MODULE_DEVICE_TABLE() wouldn't work if the table changes
during runtime.

> 
> > My original idea: would it be possible to get the `adev` in cros_ec_lpc_probe()
> > via any lookup API?  If yes, it could still use DMI match and get `adev` if
> > required.
> 
> That works; I've tested it.
> 
> In this scenario we're not using the existing PNP0C09 platform device,
> which means I can't look at
> /sys/bus/acpi/devices/PNP0C09\:00/physical_node/driver and see the
> driver. Is this OK?
> 
> (Note that ACPI_COMPANION_SET() doesn't fix this. You can use
> acpi_bind_one() but that seems more like internal plumbing).

As long as ACPI_COMPANION() in the driver can get the correct `adev`, I guess
it's fine.  Otherwise, put the `adev` to somewhere device specific (e.g.
struct cros_ec_lpc) should also work.