Re: [RFC PATCH 00/40] Toward class init of cpu features

[Alex Bennée]

Peter Maydell <peter.maydell@linaro.org> writes:

> On Fri, 6 Jan 2023 at 19:29, Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> On 1/6/23 11:12, Peter Maydell wrote:
>> > The trouble with this idea is that not all instances of the same
>> > class are actually necessarily the same. For instance, if you
>> > have a system with both (a) a Cortex-A53 with a PMU, and
>> > (b) a Cortex-A53 without a PMU, then they're both instances of
>> > the same class, but they shouldn't be sharing the coprocessor
>> > register hashtable because they don't have an identical set of
>> > system registers.
>> >
>> > This kind of same-CPU-type-heterogenous-configuration system is
>> > not something we're currently using on A-profile, but we do have
>> > it for M-profile (the sse200 has a dual-core setup where only one
>> > of the CPUs has an FPU), so it's not totally outlandish.
>>
>> Yes, I know.  See patch 29 where I moved the vfp and dsp properties off of 
>> the m-profile
>> cpus and created new cpu classes instead, specifically for the sse220.
>>
>> It's not scalable, I'll grant you, but it's hard to design for something 
>> we're not using.
>> What we use now, apart from the sse200, are common properties set on the 
>> command-line.
>
> We also set some properties in code -- eg aspeed_ast2600.c clears
> the 'neon' property on its CPUs, lots of the boards clear
> has_el3 and has_el2, etc. I hadn't got as far as patch 29, but
> looking at it now that looks like a pretty strong indication
> that this is the wrong way to go. It creates 3 extra
> cortex-m33 CPU classes, and if we find another thing that
> ought to be a CPU property then we'll be up to 8; and the
> mess propagates up into the SSE-200 class, which is also
> no longer able to be configured in the normal way by setting
> properties on it, and it becomes visible in user-facing
> command line stuff.
>
> I think our object model pretty strongly wants "create object;
> set properties on it that only affect this object you created;
> realize it", and having one particular subset of objects that
> doesn't work the same way is going to be very confusing.

What about cloning objects after they are realised? After all that is
what we do for the core CPUClass in user-mode.

>
> thanks
> -- PMM


-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro