Re: [coreboot] Any plans to fix payload builds on 64-bit machines?

[Sean McNeil]

I've tried your suggestion. Building binutils on my 64-bit Fedora 20 system:

../../../binutils-2.23.2/bfd/doc/bfd.texinfo:325: unknown command `colophon'
../../../binutils-2.23.2/bfd/doc/bfd.texinfo:336: unknown command `cygnus'
make[3]: *** [bfd.info] Error 1

I have texinfo-5.1-4.fc20.x86_64 installed.

On 05/22/2014 02:42 AM, David Hendricks wrote:
> On Tue, May 20, 2014 at 8:51 PM, Sean McNeil <[e-mail] 
> > wrote:
> Are you bullding on 64-bit machine and have:
>
>
> CONFIG_COMPILER_GCC=y
> CONFIG_ANY_TOOLCHAIN=y
>
> ?
>
>
> CONFIG_ANY_TOOLCHAIN should not be used in most cases. What does your 
> .xcompile file look like?
>
> I'm building on a 64-bit host using the crossgcc toolchain. Try 
> running the buildgcc script in util/crossgcc, then rm -f .xcompile and 
> run make and see if that helps.
>
> For reference, here are my .xcompile and .config files as well as the 
> output from make.
> ​
> xcompile.txt 
> https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNLVhwSHh4V0VWY28/edit?usp=drive_web
> ​​
> config.txt 
> https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNZ2hEcENPQ2kwSHM/edit?usp=drive_web
> ​​
> make.txt 
> https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNY1phV0c4c0MtM0U/edit?usp=drive_web
> ​
> arch/x86/exec.S:1:0: note: this is the location of the previous
> definition
>  /*
>  ^
> arch/x86/exec.S: Assembler messages:
> arch/x86/exec.S:43: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:45: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:52: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:53: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:54: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:61: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:62: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:70: Error: invalid instruction suffix for `push'
> arch/x86/exec.S:73: Error: operand type mismatch for `call'
> arch/x86/exec.S:81: Error: invalid instruction suffix for `pop'
> arch/x86/exec.S:91: Error: invalid instruction suffix for `pop'
> arch/x86/exec.S:92: Error: invalid instruction suffix for `pop'
> arch/x86/exec.S:93: Error: invalid instruction suffix for `pop'
> arch/x86/exec.S:97: Error: invalid instruction suffix for `pop'
> make[2]: *** [build/arch/x86/exec.libc.o] Error 1
> make[1]: *** [libpayload] Error 2
> make: *** [filo] Error 2
>
>
>
> On 05/21/2014 10:36 AM, David Hendricks wrote:
> > On Tue, May 20, 2014 at 6:31 PM, Sean McNeil
> > <[e-mail] <mailto:[e-mail]>> wrote:
> >
> >     All the recent changes to how the compiler is defined have
> >     now caused payloads like FILO and SeaBIOS to not build on
> >     X86_64 machines. Is anyone working to fix this?
> >
> >
> > I just tried building images with each of those as payloads and
> > they seemed to compile fine for me. Try "rm -f .xcompile" before
> > compiling and see if that helps.
> >
> > If that doesn't work, can you post more details such as the git
> > version you're on and output of "make" ?
> >
> > -- 
> > David Hendricks (dhendrix)
> > Systems Software Engineer, Google Inc.

Any plans to fix payload builds on 64-bit machines?

[Sean McNeil]

The below problem is known and fixed in the newer binutils. I changed 
buildgcc to:

-BINUTILS_VERSION=2.23.2
+BINUTILS_VERSION=2.24

and everything builds fine. So please confirm that CONFIG_ANY_TOOLCHAIN 
is deprecated and to be removed. If there are no plans to support it, in 
my opinion it should not be provided as a choice.

Cheers,
Sean

On 05/22/2014 09:59 AM, Sean McNeil wrote:
> I've tried your suggestion. Building binutils on my 64-bit Fedora 20 
> system:
>
> ../../../binutils-2.23.2/bfd/doc/bfd.texinfo:325: unknown command 
> `colophon'
> ../../../binutils-2.23.2/bfd/doc/bfd.texinfo:336: unknown command `cygnus'
> make[3]: *** [bfd.info] Error 1
>
> I have texinfo-5.1-4.fc20.x86_64 installed.
>
> On 05/22/2014 02:42 AM, David Hendricks wrote:
> > On Tue, May 20, 2014 at 8:51 PM, Sean McNeil <[e-mail] 
> > > wrote:
> > Are you bullding on 64-bit machine and have:
> >
> >
> > CONFIG_COMPILER_GCC=y
> > CONFIG_ANY_TOOLCHAIN=y
> >
> > ?
> >
> >
> > CONFIG_ANY_TOOLCHAIN should not be used in most cases. What does your 
> > .xcompile file look like?
> >
> > I'm building on a 64-bit host using the crossgcc toolchain. Try 
> > running the buildgcc script in util/crossgcc, then rm -f .xcompile 
> > and run make and see if that helps.
> >
> > For reference, here are my .xcompile and .config files as well as the 
> > output from make.
> > ​
> > xcompile.txt 
> > https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNLVhwSHh4V0VWY28/edit?usp=drive_web
> > ​​
> > config.txt 
> > https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNZ2hEcENPQ2kwSHM/edit?usp=drive_web
> > ​​
> > make.txt 
> > https://docs.google.com/a/google.com/file/d/0BwyScGuaZ2uNY1phV0c4c0MtM0U/edit?usp=drive_web
> > ​
> > arch/x86/exec.S:1:0: note: this is the location of the previous
> > definition
> >  /*
> >  ^
> > arch/x86/exec.S: Assembler messages:
> > arch/x86/exec.S:43: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:45: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:52: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:53: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:54: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:61: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:62: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:70: Error: invalid instruction suffix for `push'
> > arch/x86/exec.S:73: Error: operand type mismatch for `call'
> > arch/x86/exec.S:81: Error: invalid instruction suffix for `pop'
> > arch/x86/exec.S:91: Error: invalid instruction suffix for `pop'
> > arch/x86/exec.S:92: Error: invalid instruction suffix for `pop'
> > arch/x86/exec.S:93: Error: invalid instruction suffix for `pop'
> > arch/x86/exec.S:97: Error: invalid instruction suffix for `pop'
> > make[2]: *** [build/arch/x86/exec.libc.o] Error 1
> > make[1]: *** [libpayload] Error 2
> > make: *** [filo] Error 2
> >
> >
> >
> > On 05/21/2014 10:36 AM, David Hendricks wrote:
> > > On Tue, May 20, 2014 at 6:31 PM, Sean McNeil
> > > <[e-mail] <mailto:[e-mail]>> wrote:
> > >
> > >     All the recent changes to how the compiler is defined have
> > >     now caused payloads like FILO and SeaBIOS to not build on
> > >     X86_64 machines. Is anyone working to fix this?
> > >
> > >
> > > I just tried building images with each of those as payloads and
> > > they seemed to compile fine for me. Try "rm -f .xcompile" before
> > > compiling and see if that helps.
> > >
> > > If that doesn't work, can you post more details such as the git
> > > version you're on and output of "make" ?
> > >
> > > -- 
> > > David Hendricks (dhendrix)
> > > Systems Software Engineer, Google Inc.

Any plans to fix payload builds on 64-bit machines?

[Patrick Georgi]

Am 22.05.2014 05:36, schrieb Sean McNeil:
> and everything builds fine. So please confirm that CONFIG_ANY_TOOLCHAIN
> is deprecated and to be removed. If there are no plans to support it, in
> my opinion it should not be provided as a choice.

It's not that there aren't plans to support it (I think Google needs
it), but it's certainly of secondary concern to us.
We certainly prefer people to use "our" compiler since that allows us to
reproduce issues more easily.

That said I think the (ad-hoc) patch I sent a couple of hours ago should
fix things for seabios with your host compiler. Other payloads can be
adapted similarily.

As a result, ANY_TOOLCHAIN will stay and you're free to use it if you
prefer, but expect the occasional issue arising that you'll have to work
on yourself. Patches accepted :-)

Patrick

Any plans to fix payload builds on 64-bit machines?

[Peter Stuge]

Patrick Georgi wrote:
> As a result, ANY_TOOLCHAIN will stay and you're free to use it if you
> prefer,

Let's put it behind CONFIG_EXPERT.

//Peter