On 10/03/2021 16:39, Arnd Bergmann wrote:

On Wed, Mar 10, 2021 at 4:06 PM Krzysztof Kozlowski
[off-list ref] wrote:

quoted

On 10/03/2021 15:45, Tom Rix wrote:

quoted

On 3/10/21 1:45 AM, Lee Jones wrote:

Many other architectures do not have vendor prefix (TEGRA, EXYNOS,
ZYNQMP etc). I would call it the same as in ARMv7 - ARCH_SOCFPGA - but
the Altera EDAC driver depends on these symbols to be different.
Anyway, I don't mind using something else for the name.

I agree the name SOCFPGA is confusing, since it is really a class of
device that is made by multiple manufacturers rather than a brand name,
but renaming that now would be equally confusing. If the Intel folks
could suggest a better name that describes all products in the platform
and is less ambiguous, we could rename it to that. I think ARCH_ALTERA
would make sense, but I don't know if that is a name that is getting
phased out. (We once renamed the Marvell Orion platform to ARCH_MVEBU,
but shortly afterwards, Marvell renamed their embedded business unit (EBU)
and the name has no longer made sense since).

I wait then for some ideas from Dinh (or anyone else).

Regardless of what name we end up with, I do think we should have the
same name for 32-bit and 64-bit and instead fix the edac driver to do
runtime detection based on the compatible string.

I can rename ARCH_SOCFPGA on 32-bit ARM as well, however converting edac
driver from #ifdef ARCH_SOCFPGA64 to proper compatible string will be
too much for me: I am not able to test it.

This edac Altera driver is very weird... it uses the same compatible
differently depending whether this is 32-bit or 64-bit (e.g. Stratix
10)! On ARMv7 the compatible means for example one IRQ... On ARMv8, we
have two. It's quite a new code (2019 from Intel), not some ancient
legacy, so it should never have been accepted...

Best regards,
Krzysztof