Re: [Qemu-devel] [RFC PATCH 06/11] target/arm: use the common interface for WRITE0/WRITEC in arm-semi

[Miroslav Rezanina <mrezanin@redhat.com>]

----- Original Message -----
> From: "Peter Maydell" <peter.maydell@linaro.org>
> To: "Alex Bennée" <alex.bennee@linaro.org>
> Cc: "Miroslav Rezanina" <mrezanin@redhat.com>, "QEMU Developers" <qemu-devel@nongnu.org>, "Riku Voipio"
> <riku.voipio@iki.fi>, "qemu-arm" <qemu-arm@nongnu.org>, "Laurent Vivier" <laurent@vivier.eu>
> Sent: Friday, May 31, 2019 4:38:04 PM
> Subject: Re: [Qemu-devel] [RFC PATCH 06/11] target/arm: use the common interface for WRITE0/WRITEC in arm-semi
> 
> On Fri, 31 May 2019 at 15:28, Alex Bennée <alex.bennee@linaro.org> wrote:
> > Miroslav Rezanina <mrezanin@redhat.com> writes:
> > >From: "Alex Bennée" <alex.bennee@linaro.org>
> > >> OK - so from the upstream source tree CONFIG_SEMIHOSTING is still =y
> > >> (but I can see most of them are now =n). Isn't the simplest solution to
> > >> fix-up your version of the default_config file to include SEMIHOSTING?
> > >>
> > >
> > > It's fix but it goes against our policy of handling CONFIG options so we
> > > would prefer to have this fixed - otherwise there's no meaning in having
> > > config option if you can't disable it.
> >
> > Is that what it means? For my part it means we don't build in
> > CONFIG_SEMIHOSTING for the arches that don't need it (which we were
> > before). Granted it only really simplified the vl.c parsing and dropped
> > support for semihosting for the linux-user targets (except ARM).
> 
> Yes, that would be my interpretation of it. If we had
> a 'config FOO' stanza for CPUs, then Arm CPUs would
> "select SEMIHOSTING". If RedHat would like it to be possible
> to build Arm CPUs without CONFIG_SEMIHOSTING then they're
> free to submit patches for that, but that's a new feature
> upstream doesn't currently support, not a bug in upstream.
> (Also I'd be a bit dubious because it means that previously working
> guest setups that use semihosting will break.)

I partially agree here - I see difference between disabling
config and omitting it. We are not not disabling CONFIG_SEMIHOSTING,
we just ignore it. So we got error because it is not properly handled.
Proper handling should be either auto-include it as dependency or
successful build with option disabled.

As there's currently no way to auto-include it through dependency,
it would be good to have comment in default_config file next to it stating
that it's required option. This will allow us to see it and
add to our default_config we used instead upstream one.

Mirek
> 
> PS: if we had a 'config FOO' stanza for CPUs that would also
> allow us to say "building Arm CPUs requires the NVIC" and
> similarly for things which in QEMU are devices but which are
> architecturally tightly-coupled non-optional parts of the CPU.
> 
> thanks
> -- PMM
> 

-- 
Miroslav Rezanina
Software Engineer - Virtualization Team Maintainer