Home » Mailing list archives » linux-kernel » 2007 » October » 30

Re: 32bit builds on x86-64 host.

Previous thread: [PATCH] bttv: uses input functions, should depend on INPUT by Randy Dunlap on Monday, October 29, 2007 - 2:19 pm. (1 message)

Next thread: [Git pull] x86 update by Thomas Gleixner on Monday, October 29, 2007 - 7:48 pm. (1 message)
[view original message]
From: Dave Jones <davej@...>
To: Linux Kernel <linux-kernel@...>
Cc: Sam Ravnborg <sam@...>
Subject: 32bit builds on x86-64 host.
Date: Monday, October 29, 2007 - 7:32 pm

Before the arch merge, I frequently would test 32bit compiles

by doing make ARCH=i386 {bzImage/modules/file.o}


Since commit 47572387d58a9584c60ebbbdee56fc92c627f16f

how does one do this?


	Dave


--

http://www.codemonkey.org.uk

-
[view original message]
From: H. Peter Anvin <hpa@...>
To: Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Monday, October 29, 2007 - 8:07 pm

make ARCH=i386 {bzImage/modules/file.o}


Amazing, isn't it?


	-hpa

-
[view original message]
From: Christoph Hellwig <hch@...>
To: H. Peter Anvin <hpa@...>
Cc: Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Tuesday, October 30, 2007 - 9:10 am

Which seems rather unnatural to me.  All other merge architectures have

ARCH=<archdir> where archdir is the name under arch, e.g. mips, powerpc,

s390.  They then have a CONFIG_FOO_64 and set the utsname and -m32/-m64

respectively.  It would be nice if x86 could behave like all other

architectures in this respect.  Especially as the behaviour of the other

---end quoted text---

-
[view original message]
From: Jeremy Fitzhardinge <jeremy@...>
To: Christoph Hellwig <hch@...>
Cc: H. Peter Anvin <hpa@...>, Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Tuesday, October 30, 2007 - 9:00 pm

Yes, that would be nice.  I keep forgetting the ARCH= in my crossbuild

trees, and end up with strange mismashes.


    J

-
[view original message]
From: Christoph Hellwig <hch@...>
To: Jeremy Fitzhardinge <jeremy@...>
Cc: Christoph Hellwig <hch@...>, H. Peter Anvin <hpa@...>, Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Wednesday, October 31, 2007 - 4:57 am

That's a problem with all architectures currently.  But I think Sam has

a patch that at least stores ARCH when using objdir != srcdir builds.


-
[view original message]
From: Sam Ravnborg <sam@...>
To: Christoph Hellwig <hch@...>
Cc: Jeremy Fitzhardinge <jeremy@...>, H. Peter Anvin <hpa@...>, Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Wednesday, October 31, 2007 - 5:41 am

kbuild check's the asm symlink and error out if it is wrong.

So forgetting ARCH should be captured. But the check does not

store ARCH which is what I hope to have fixed for next merge window.


So for now if you forget ARCH you should be warned.

And in the future kbuild will remember what ARCH (and CROSS_COMPILE)

you used when configuring the kernel.


For x86 the current check does not help use because the asm symlink

is the same for i386 and x86_64 - so for now we are out of luck.


	Sam

-
[view original message]
From: H. Peter Anvin <hpa@...>
To: Jeremy Fitzhardinge <jeremy@...>
Cc: Christoph Hellwig <hch@...>, Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Tuesday, October 30, 2007 - 9:11 pm

That, however, is a bigger problem that also affects, for example,

compiling for MIPS on an x86 host.  I believe Sam is working on a

solution for that.


(That doesn't mean that is not the right thing to do.)


	-hpa

-
[view original message]
From: Dave Jones <davej@...>
To: H. Peter Anvin <hpa@...>
Cc: Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Monday, October 29, 2007 - 8:39 pm

On Mon, Oct 29, 2007 at 05:07:08PM -0700, H. Peter Anvin wrote:

 > Dave Jones wrote:

 > > Before the arch merge, I frequently would test 32bit compiles

 > > by doing make ARCH=i386 {bzImage/modules/file.o}

 > >

 > > Since commit 47572387d58a9584c60ebbbdee56fc92c627f16f

 > > how does one do this?

 > >

 >

 > make ARCH=i386 {bzImage/modules/file.o}

 >

 > Amazing, isn't it?


It's like the future.   I'm not sure why that didn't work for me.

Must be something in my dirty tree.


	Dave

--

http://www.codemonkey.org.uk

-
[view original message]
From: Al Viro <viro@...>
To: Dave Jones <davej@...>, Linux Kernel <linux-kernel@...>, Sam Ravnborg <sam@...>
Subject: Re: 32bit builds on x86-64 host.
Date: Monday, October 29, 2007 - 8:06 pm

Same.  Top-level makefile kludges around that:


SRCARCH         := $(ARCH)


# for i386 and x86_64 we use SRCARCH equal to x86

SRCARCH := $(if $(filter x86_64 i386,$(SRCARCH)),x86,$(SRCARCH))


and then plays with both ARCH and SRCARCH.  Note

# If a arch/$(SRCARCH)/Kconfig.$(ARCH) file exist use it

ifneq ($(wildcard $(srctree)/arch/$(SRCARCH)/Kconfig.$(ARCH)),)

        Kconfig := arch/$(SRCARCH)/Kconfig.$(ARCH)

else

        Kconfig := arch/$(SRCARCH)/Kconfig

endif


in scripts/kconfig/Makefile, BTW - now we use arch/x86/Kconfig.i386 and

arch/x86/Kconfig.x86_64 as starting points for ARCH=i386 and ARCH=x86_64

resp.

-
Previous thread: [PATCH] bttv: uses input functions, should depend on INPUT by Randy Dunlap on Monday, October 29, 2007 - 2:19 pm. (1 message)

Next thread: [Git pull] x86 update by Thomas Gleixner on Monday, October 29, 2007 - 7:48 pm. (1 message)