Re: [Qemu-devel] Can't compile QEMU 2.5.0 on Arch Linux ARM

[Peter Maydell]

On 21 February 2016 at 10:30, Xavier de Rauville
<address@hidden> wrote:
> Greetings
>
> I am attempting to compile QEMU for Arch Liux ARM using the build files from
> here:
> https://projects.archlinux.org/svntogit/packages.git/tree/trunk?h=packages/qemu
>
> I am targeting the ARMv7hf architecture. Unfortunately, I am getting an
> error during the build process. I created a log file from makepkg's stdout
> and stderr streams.

I do build QEMU for 32-bit ARM (on Ubuntu), so this is likely one of:
 * your compile environment is broken somehow
 * something particular to the QXL/spice code that's being built
   (which my build config may well not be exercising)

> -c -o hw/display/qxl.o hw/display/qxl.c
> {standard input}: Assembler messages:
> {standard input}:2410: Error: bad instruction `lock'
> {standard input}:2410: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:2518: Error: bad instruction `lock'
> {standard input}:2518: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:3045: Error: bad instruction `lock'
> {standard input}:3045: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:3370: Error: bad instruction `lock'
> {standard input}:3370: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:7470: Error: bad instruction `lock'
> {standard input}:7470: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:9097: Error: bad instruction `lock'
> {standard input}:9097: Error: bad instruction `addl $0,0(%rsp)'
> {standard input}:9412: Error: bad instruction `lock'
> {standard input}:9412: Error: bad instruction `addl $0,0(%rsp)'
> /home/share/qemu-devel/QEMU/src/qemu-2.5.0/rules.mak:57: recipe for target
> 'hw/display/qxl.o' failed
> make: *** [hw/display/qxl.o] Error 1
> ==> ERROR: A failure occurred in build().
>     Aborting...

This should really not happen because it implies that we
somehow ended up with x86 inline assembly.

Some things to test:
 (1) if you configure QEMU with --disable-spice, does it build
OK, or does it just fail later on on some other source file?
 (2) does this fail with the current git master?
 (3) can you get the compiler to produce the preprocessed
source for us and put it somewhere? (you can rerun the specific
failing compiler command and add -save-temps, then it will
output a .i file which is the preprocessed output.) That
should help to track down exactly what is resulting in these
x86 insns getting into the output.

thanks
-- PMM