Re: [Qemu-devel] [PATCH] configure: adding ppc64le to supported host CPUs

[Daniel Henrique Barboza <danielhb@linux.vnet.ibm.com>]

On 11/01/2017 04:23 PM, Peter Maydell wrote:
> On 1 November 2017 at 17:53, Daniel Henrique Barboza
> <danielhb@linux.vnet.ibm.com> wrote:
>> When executing 'configure' in a fresh QEMU clone, in a fresh
>> OS install running in a ppc64le host, this is the error
>> shown:
>>
>> -----
>>
>> ../configure --enable-trace-backend=simple --enable-debug
>>      --target-list=ppc64-softmmu
>>
>> ERROR: Unsupported CPU = ppc64le, try --enable-tcg-interpreter
>>
>> -----
>>
>> This isn't true, ppc64le host CPU is supported.
>>
>> This patch adds 'ppc64le' to the list of supported host CPUs. After
>> this patch, the next error when executing configure in the
>> scenario mentioned is:
>>
>> -----
>>
>> ../configure --enable-trace-backend=simple --enable-debug
>>      --target-list=ppc64-softmmu
>>
>> ERROR: "cc" either does not exist or does not work
>>
>> -----
>>
>> Indicating that now we need a valid compiler to proceed.
>>
>> Signed-off-by: Daniel Henrique Barboza <danielhb@linux.vnet.ibm.com>
>> ---
> I think this is either unnecessary or it should be a lot more
> changes than this. The problem here is that how we determine
> the host CPU type is by using the C compiler. So if you don't
> have a C compiler then we're going to fail to autodetect,
> and the (kind of bogus) fallback is via "uname -a". What
> we ought to do is (a) detect "no C compiler" before we
> try to use it and (b) not have that uname -a fallback.

Now that you've said, if I install GCC before running configure for the
first time I don't see this error happening. Perhaps alternative
(a) is a good approach and avoid a lot of changes just for the
sake of getting rid of that error message.


Daniel

> I suggest you install a C compiler and then try again with
> an unmodified configure to see how that does.
>
> (I think that the correct value of the "cpu" variable for
> ppc64le is "ppc64", same as for bigendian, which we'll
> autodetect if we have a compiler to do it with. If that's
> right, then you could argue for having our "normalise
> CPU host name" code map ppc64le to ppc64. Certainly a
> lot of places in configure assume "ppc64" is the only
> kind...)
>
> thanks
> -- PMM
>