Re: [PATCH] hw/riscv/virt.c: re-insert and deprecate 'riscv, delegate'

[Daniel Henrique Barboza]

On 7/13/24 3:57 PM, Conor Dooley wrote:
> On Sat, Jul 13, 2024 at 02:43:25PM -0300, Daniel Henrique Barboza wrote:
> > Commit b1f1e9dcfa renamed 'riscv,delegate' to 'riscv,delegation' since
> > it is the correct name as per dt-bindings, and the absence of the
> > correct name will result in validation fails when dumping the dtb and
> > using dt-validate.
> >
> > But this change has a side-effect: every other firmware available that
> > is AIA capable is using 'riscv,delegate', and it will fault/misbehave if
> > this property isn't present. The property was added back in QEMU 7.0,
> > meaning we have 2 years of firmware development using the wrong
> > property.
> >
> > Re-introducing 'riscv,delegate' while keeping 'riscv,delegation' will
> > make current firmwares to keep booting with the 'virt' machine and
> > dt-validate won't complain about it since we're still using the expected
> > property 'riscv,delegation'. 'riscv,delegate' is then marked for future
> > deprecation and its use is being discouraged.
>
> dt-validate /should/ complain about it - does yours not? It's probably a
> bug if it doesn't. Whether dt-validate complains or not here I don't
> think is relevant though, only the impact on firmware of removing it.

We'll keep the property as a temporary documented band-aid that we'll remove
a few releases from now. OpenSBI was also updated to use the right property
recently so we shouldn't have any problems.

Oh, and dt-validate will complain about it if we use the right command line that
creates the controller ... something that I failed to do in v1 :(  I'll re-send
the patch with an updated commit msg.


Thanks,

Daniel

>