[PATCH v1 2/3] riscv: cache: Add CBO instructions

[Conor Dooley]

On Wed, Aug 21, 2024 at 09:27:03PM +0530, Mayuresh Chitale wrote:
> On Wed, Aug 21, 2024 at 3:28 PM Conor Dooley <conor.dooley at microchip.com> wrote:
> >
> > On Tue, Aug 20, 2024 at 02:14:01PM +0200, Heinrich Schuchardt wrote:
> > > On 20.08.24 11:37, Mayuresh Chitale wrote:
> > > > +void riscv_zicbom_init(void)
> > > > +{
> > > > +   struct udevice *dev;
> > > > +
> > > > +   if (!CONFIG_IS_ENABLED(RISCV_ISA_ZICBOM))
> > > > +           return;
> > > > +
> > > > +   uclass_first_device(UCLASS_CPU, &dev);
> > > > +   if (!dev) {
> > > > +           log_err("Failed to get cpu device!\n");
> > > > +           return;
> > >
> > > Please, return an error code.
> > >
> > > > +   }
> > > > +
> > > > +   (void)dev_read_u32(dev, "riscv,cbom-block-size", &zicbom_block_size);
> > >
> > > Please, do not ignore errors.
> >
> > I'm curious what the policy of U-Boot is w.r.t. extension probing. Is it
> > okay to enable the option for Zicbom even if your hardware does not
> > support it, in which case riscv_zicbom_init() would be expected to fail
> > gracefully and no CMOs done?
> > Say, for example, you had two very similar chips, one with DMA non-coherent
> > peripherals and one that only differed by having DMA coherent ones, and you
> > wanted to run the same U-Boot binary on both devices using a devicetree
> > passed from firmware.
> I think in such a case the config option can always be enabled with
> the proper dt being passed by the firmware.

I don't understand what you mean, sorry. Both cases I mention would have
a "proper" devicetree that matches the hardware. I'm interested in
whether or not U-Boot's maintainers think that enabling the config option
should warn on, or preclude use of, systems that do not not support
Zicbom.

Cheers,
Conor.

Also, technically there's no guarantee that "riscv,cbom-block-size" being
present means that zicbom also is. Realistically, it should never be the
case that it is present without zicbom in the isa string representation,
but correctness would require checking that it is. supports_extension()
only supports single-letter extensions at present, making it support
multi-letter extensions should not be too difficult.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20240821/77043564/attachment.sig>