Re: [PATCH 2/4] target/arm: Move cortex-m related functions to new file v7m.c

[Thomas Huth <thuth@redhat.com>]

On 23/09/2019 20.45, Peter Maydell wrote:
> On Mon, 23 Sep 2019 at 19:09, Thomas Huth <thuth@redhat.com> wrote:
>> However, there's one thing I currently don't quite understand in this
>> code (since I'm not an ARM guy, sorry) : There are references to "v8" in
>> m_helper.c, too. Is that related to a separate CPU type, ie. should the
>> v8 code also be available when CONFIG_ARM_V7M is disabled? Or can the
>> code in m_helper.c be disabled completely if CONFIG_ARM_V7M is not set?
> 
> QEMU's naming conventions here is a bit confusing, for
> historical reasons.
> 
> Architecturally what we have is:
>  * "M-profile" -- this is the flavour of Arm architecture for
> microcontrollers; it has some big differences from A and R
> profile (eg the exception mechanism is different and it has
> a built-in NVIC interrupt controller). All the Cortex-M<anything>
> CPUs are M-profile
>  * "Arm-v7M" -- this is the v7 flavour of the M-profile
> architecture, eg Cortex-M3.
>  * "Arm-v6M" -- this looks like it ought to mean "v6 flavour
> of M-profile", but if you look at what features it has it's
> more like "cut down version of v7M" (fewer instructions, cut
> down exception model, etc, but some things which on A-profile
> don't appear until v7A are present in v6M). Cortex-M0 and -M1.
>  * "Arm-v8M" -- v8 flavour of M-profile. The big change here
> is support for TrustZone. Cortex-M33. v8M comes in two
> sub-profiles: "mainline", which has all the features like v7M,
> and "baseline", which is cut-down in the same way v6M is a
> cut-down v7M.
> 
> In QEMU, we implemented Cortex-M3 first, and then added -M0 and
> -M33 later. So mostly our function naming convention uses
> "v7m" when it means "any M-profile"; a few v8M-specific
> functions use a "v8m" prefix. Everything in m_helper.c is
> M-profile specific; most of it is used by all M-profile cores,
> and a few bits are v8M-only or v7M-and-v8M-mainline only.
> 
> From a CONFIG switch point of view I don't think it's
> worth being able to #ifdef out the various flavours of
> M-profile individually.

Ok, thanks a lot for the explanation! It's much clearer to me now.
So I think it likely would be best to disable the whole m_helper.c code
instead of introducing a new file like v7m.c (but that likely requires a
stub file instead or some #ifdeffing in translate.c).

It also sounds like CONFIG_ARM_V7M should rather be renamed to
CONFIG_ARM_MPROFILE or something similar?

 Thomas