śr., 22 cze 2022 o 16:20 Andy Shevchenko
[off-list ref] napisał(a):

On Wed, Jun 22, 2022 at 01:22:15PM +0200, Andrew Lunn wrote:

quoted

quoted

quoted

It's not device on MDIO bus, but the MDIO controller's register itself
(this _CSR belongs to the parent, subnodes do not refer to it in any
way). The child device requires only _ADR (or whatever else is needed
for the case the DSA device is attached to SPI/I2C controllers).

More and more the idea of standardizing the MDIOSerialBus() resource looks
plausible. The _ADR() usage is a bit grey area in ACPI specification. Maybe
someone can also make it descriptive, so Microsoft and others won't utilize
_ADR() in any level of weirdness.

I don't know if it makes any difference, but there are two protocols
spoken over MDIO, c22 and c45, specified in clause 22 and clause 45 of
the 802.3 specification. In some conditions, you need to specify which
protocol to speak to a device at a particular address. In DT we
indicate this with the compatible string, when maybe it should really
be considered as an extension of the address.

If somebody does produce a draft for MDIOSerialBus() i'm happy to
review it.

I also can review it. Marcin, would it be hard for you to prepare a formal
proposal for ACPI specification?

I've just consulted this to get an understanding of the process.
* I will initiate it with the code-first ECR using the linux-acpi
mailing list, where all the technical review will take place.
* At the same time a ticket and the formal process for this will be
triggered within UEFI Forum.
* Once everything gets approved, an official confirmation will be
provided and from that moment, it would allow us to proceed with
implementation without need of waiting months for another ACPI
Specification release.

Unless anyone objects, I will include this thread recipients to take
part in review of the proposed MDIOSerialBus _CRS resource macro
contents, so it contains all relevant information.

Note: Once this hopefully gets accepted one day and allow us proceed
with Linux handling, it should be easy to satisfy backward
compatibility with current users of MDIO+PHY in ACPI.

Best regards,
Marcin