Re: Arm drivers: are device tree descriptions required for every device?

[Brook Milligan <brook%nmsu.edu@localhost>]

> On Apr 10, 2023, at 7:01 PM, Mouse <mouse%Rodents-Montreal.ORG@localhost> wrote:
> 
> There are at least three things here that can reasonably be called the
> device tree.
> 
> (1) The tree passed from the booter to the kernel.
> 
> (2) The tree (or more properly directed graph) of attachments described
> by the kernel config.
> 
> (3) The tree actually found by autoconf.

Thank you for being more explicit than my question.  That context probably helps frame my question better.

The goal is to understand what has to be specified in (1) and (2) in order to achieve a desired (3) at runtime.

My question could have been better phased in these terms by asking if (1) has to include everytihing intended to be in (3).

It seems that the answer is mixed:

- For busses and devices that cannot (by design) enumerate themselves, the answer is yes, (1) has to be all inclusive.

- For busses and devices that can enumerate themselves, e.g., USB, the answer is no, (1) need not include them.

I’m assuming that in all cases (2) must include all the drivers, though.

In addition, there seem to be corner cases, where it might be desirable to pass specific information from (1) for a device that otherwise might not need it.

There may be additional corner cases.  How does one discover whether a particular device is a corner case or not?

I hope that is a reasonably concise summary.

Thanks to everyone who responded so far.

Cheers,
Brook