Re: [PATCH v1 5/6] arm64: cpufeature: Track BBML3 exclusions with erratum options

From: Ryan Roberts

Date: Sun Nov 22 2026 - 04:08:12 EST


On 18/11/2026 17:59, Catalin Marinas wrote:
> On Fri, Nov 13, 2026 at 02:50:19PM +0000, Ryan Roberts wrote:
>> Several CPUs implementing the BBML3 no-abort semantics advertise only
>> BBML2, but cannot safely use those semantics because of hardware errata.
>> Omitting their MIDRs from the allow list hides the reason for the
>> exclusion and makes the relevant workarounds difficult to track.
>>
>> Add a default-enabled Kconfig option for each of the following errata,
>> and include the affected CPUs in the BBML3 allow list only when the
>> corresponding workaround is disabled:
>>
>> 3043240 Cortex-X925
>> 3043263 Cortex-X4
>> 3053180 Neoverse-V3 and Neoverse-V3AE
>> 3683289 C1-Ultra and C1-Premium
>> 3696239 Cortex-X3
>> 3696242 Neoverse-V2
>> 3696244 Cortex-A710 and Cortex-X2
>> 3696250 Neoverse-N2
>> 3696285 Neoverse-V1
>> 4196850 C2-Ultra
>
> I'm not sure what the benefit of this is. These CPUs are not listed in
> the BBML3 capable list anyway, so the patch adds them but gated by a
> Kconfig option, default y. Is it only an elaborate way to tell people
> why we couldn't enable BBML3 support for them? Maybe the documentation
> part is sufficient.
>

I'm just trying to make it very clear why certain CPUs are not in the list. With
the exception of A510, in general, Arm-designed CPUs that advertise BBML2 are
also BBML3 compliant and should therefore be in this list. But some of the CPUs
have BBML2 errata which prevent them from being in the list.

The information is all very scattered internally, and I've personally had to do
the exercise 3 times now, of figuring out which CPUs are excluded from the list
because they haven't been added yet vs which are excluded because of an erratum.

Perhaps it would be simpler just to create a file in Documentation/arch/arm64 to
track this and add a comment next to the midr list referring to it? I don't mind
how we do it, but I do think it's valuable to capture the info somehow. I did it
the way I've done it because it looked like the standard way of documenting
workarounds.

Thanks,
Ryan