Re: [Qemu-devel] [PATCH] linux-user: Support explicit targets for PowerPC

[Laurent Vivier <laurent@vivier.eu>]

Le 07/11/2017 à 10:29, Peter Maydell a écrit :
> On 7 November 2017 at 06:57, Laurent Vivier <laurent@vivier.eu> wrote:
>> Le 07/11/2017 à 03:50, Aaron Sierra a écrit :
>>> Enable building PowerPC targets supporting a specific CPU, without
>>> having to set QEMU_CPU via the environment. For example these build
>>> targets (and many more) become available:
>>>
>>>     qemu-ppc.e500mc
>>>     qemu-ppc.e500v2
>>>     qemu-ppc.e5500
>>>     qemu-ppc.e600
>>>     qemu-ppc.e6500
>>>
>>> These (statically compiled) binaries have proven useful for emulating
>>> PowerPC CPUs within Docker containers, where it's hard to reliably
>>> define environment variables that are available for every process.
>>
>> I agree with you.
>>
>> I have patches to do that for years, perhaps we can merge our both
>> solutions to have something working:
>>
>> linux-user: define default cpu model in configure instead of
>> linux-user/main.c
>> https://github.com/vivier/qemu-m68k/commit/1488df8c9fa3a5b099bbc4f41c077a6c69305cc2
>>
>> linux-user: specify the cpu model during configure
>> https://github.com/vivier/qemu-m68k/commit/496f804fd97d5f0ef26da1090e1d95ad15bcc50e
>>
>> The part missing in my series is to rename the qemu binaries with the
>> suffix of the default CPU.
> 
> I don't really like this -- it means we'll end up with a huge
> extra range of executables, and every time a new CPU type is
> added a new binary has to be plumbed in to our own 'make
> install' machinery and into every distro's packaging.

I understand, so I don't push the idea more than that...

> The 'look at argv[0]' approach is perhaps a bit more
> palatable. Or you could just write and use wrapper
> scripts.

It's a personal opinion: I don't like the wrapper idea.
If we could go with the 'look at argv[0]' approach, I'd be happy.

> I take it there's no good "-cpu any" option for PPC that
> would run all these binaries?

Yes, there are the embedded version and the server version, that are
described in different books (books and booke). I didn't check, but I
think some instructions differ.

But I have the same problem with m68k between coldfire and 680x0, and I
know we can't use the "-cpu any" in this case.

Thanks,
Laurent