Re: [Qemu-devel] [PATCH for-2.0 V3] tests/acpi-test: do not run iasl on big endian machines

[Marcel Apfelbaum]

On Sun, 2014-03-23 at 13:51 +0100, Paolo Bonzini wrote:
> Il 23/03/2014 13:32, Marcel Apfelbaum ha scritto:
> >> > That's the endianness of the machine we're compiling QEMU
> >> > for, not the endianness of the machine we're compiling QEMU
> >> > on. If for instance you're on x86_64 cross-compiling for PPC
> >> > then HOST_WORDS_BIGENDIAN is true, but the iasl you use
> >> > in the build process will be running on a little endian machine.
> > Hi Peter, are you sure about this?
> > I saw the 'target_bigendian' that does what you described above.
> > $bigendian is the result of a little C program that checks *host's* 
> > endian-ness.
> > Of course I might have missed something.
> 
> Peter is right.  There are three systems: build (what you run on), host 
> (what you compile for), target (always x86-64 for now for acpi-test). 
> $bigendian and G_BYTE_ORDER check the host system.  If you wanted to 
> test the build system's endianness, Laszlo's suggestion would be fine.
> 
> However, Michael is also right if we assume that iasl on bigendian can 
> assemble, and that the problems are only on disassembling.  This is 
> because in this particular case you'll be running an executable compiled 
> for the host and you know that host == build.
> 
> I think the assumption is sane.  As far as I remember, upstream iasl 
> doesn't compile at all on big-endian machines.  You need specific 
> patches such as those shared by Fedora and Debian's.  Unfortunately, the 
> patches are incomplete, they only fix assembling because that's what was 
> needed so far to cross-compile SeaBIOS.

Hi Paolo, thanks for the help!
I need a decision here :)

1.
If iasl *does* work for the scenarios used by Qemu until now,
I can conclude that the only part that suffers from it is the
test itself, because it dis-assembles AML code. In this case
I can handle it in the test itself by checking the endian-ness
(V2 already posted), or trying to dis-assemble the expected AML
file and not failing the test if this fails.

2. If iasl causes problems to run a x86_64 guest if one or both build/host
machines are BE, we need to tweak the configuration file.

After this discussion it seems that (2.) is not the case, I think maybe 
re-sending:
 [PATCH for-2.0 V2] tests/acpi-test: do not run iasl on big endian machines

Is everyone OK with this?
Thanks,
Marcel

> Paolo