On 19/03/2024 17:32, Naushir Patuck wrote:
On Tue, 19 Mar 2024 at 14:03, Tomi Valkeinen
<tomi.valkeinen@xxxxxxxxxxxxxxxx> wrote:
On 19/03/2024 15:05, Naushir Patuck wrote:
On Tue, 19 Mar 2024 at 13:02, Krzysztof Kozlowski
<krzysztof.kozlowski@xxxxxxxxxx> wrote:
On 19/03/2024 13:57, Naushir Patuck wrote:
See writing bindings. Compatibles should be SoC specific. In some cases
generic fallbacks make sense, in some note. But don't just choose
"generic fallback" because you want. Provide rationale.
If the compatible is SoC specific, I suppose "raspberrypi,rp1-cfe"
would be the correct string.
Sure, but then please think what if rp1 is on Rpi6, called exactly the
same (rp1), with some minor differences? Could it be?
Yes, this is definitely possible. In such cases, I would expect the
hardware to have a version register that would be queried by the
driver to adjust for minor differences, and the compatible string
remains the same. Does that seem reasonable?
The "would expect" is concerning. The register(s) must be there already,
with proper value.
A version register already exists in the current hardware, so we will
update it to identify future hardware revisions.
But that's a version register for the FE block, not for the whole
module, right? Are you suggesting that you'll make sure the FE version
will be changed every time anything in the bigger CFE block is changed,
and thus the FE version would also reflect the whole CFE version?
Yes, we will update the FE versioning when either CSI2 / FE blocks are updated.
Can there be versions without the FE block, with just the CSI-2 parts?
There is no version register just in the CSI2 block in isolation, so
this is not possible.
I meant could there be a future RPx with only the CSI-2 parts on it, no
FE? In which case we would not have any register for the version. But
then, that would be a rather big change, so a different compatible
string would be in order.
So, while it's not exactly a perfect version register, I think it will
work fine, assuming the HW people will actually increase the version
also for changes outside FE.
Also, I'm still wondering about the RP1 part there in the compatible
string. Is it necessary? The CFE is located in the RP1 co-processor, but
is that relevant?
Is there a versioning for the whole RP1 chip? Maybe it's going to the
wrong direction if we use the board/SoC for this compatible name, as
it's actually the RP1 where the CFE is located in, not the SoC.
I don't really know the conversion required to answer this one.
Logically CFE is on RP1, so it makes sense to me to have "rp1" in the
string, but I will follow the judgment of the maintainers.
Well, my thinking here was that if we have a register from which to read
the version, and Raspberry Pi would create a new co-processor, RP2, with
the same CFE. Would we then have "raspberrypi,rp1-cfe" and
"raspberrypi,rp2-cfe", even if there are no changes? Or would a plain
"raspberrypi,cfe" do for both?
In other words, if we don't need the "rp1" for versioning purposes,
should it then be dropped?
On the other hand, maybe it is safer to just keep the "rp1" there anyway...
Tomi
