libc and ARM_INSTRUCTION_SET

libc and ARM_INSTRUCTION_SET

[Andrea Adami]

Hi,
I'm testing uclibcgnueabi on c7x0.

Two doubts:

1) arm / thumb

Now armv5te machine defaults to "thumb" and uclibc fails do_compile:
http://tinderbox.openembedded.net/public/logs/2601805.txt

My solution offhand is to set ARM_INSTRUCTION_SET = "arm" in
uclibc.inc but perhaps someone can sort out a proper patch.

I checked how the settings are for glibc and well, the same string is in:
glibc.inc and (why?) in glibc-2.5.bb and glibc-2.7.bb which are
including glibc.inc...

2) is it actual and correct that uclibc and klibc are machine-specific?

#recent versions uclibc require real kernel headers
PACKAGE_ARCH = "${MACHINE_ARCH}"

Well, for klibc I'd say yes, it's bound to the kernel but still...I ask.

Thanks in advance

Andrea

Re: libc and ARM_INSTRUCTION_SET

[Koen Kooi]

On 22-12-08 15:56, Andrea Adami wrote:
> Hi,
> I'm testing uclibcgnueabi on c7x0.
>
> Two doubts:
>
> 1) arm / thumb
>
> Now armv5te machine defaults to "thumb" and uclibc fails do_compile:
> http://tinderbox.openembedded.net/public/logs/2601805.txt
>
> My solution offhand is to set ARM_INSTRUCTION_SET = "arm" in
> uclibc.inc but perhaps someone can sort out a proper patch.

Having uclibc (and hence libm) in ARM mode would probably be a good thing.


> 2) is it actual and correct that uclibc and klibc are machine-specific?

uclibc is machine specific since the defconfigs are machine specific, 
because at the time OE lacked fine grained architecture overrides.

regards,

Koen

Re: libc and ARM_INSTRUCTION_SET

[Henning Heinold]

On Mon, Dec 22, 2008 at 03:56:20PM +0100, Andrea Adami wrote:
> Hi,
> I'm testing uclibcgnueabi on c7x0.
> 
> Two doubts:
> 
> 1) arm / thumb
> 
> Now armv5te machine defaults to "thumb" and uclibc fails do_compile:
> http://tinderbox.openembedded.net/public/logs/2601805.txt
> 
> My solution offhand is to set ARM_INSTRUCTION_SET = "arm" in
> uclibc.inc but perhaps someone can sort out a proper patch.
> 
> I checked how the settings are for glibc and well, the same string is in:
> glibc.inc and (why?) in glibc-2.5.bb and glibc-2.7.bb which are
> including glibc.inc...
> 
> 2) is it actual and correct that uclibc and klibc are machine-specific?
> 
> #recent versions uclibc require real kernel headers
> PACKAGE_ARCH = "${MACHINE_ARCH}"
> 
> Well, for klibc I'd say yes, it's bound to the kernel but still...I ask.
> 
> Thanks in advance
> 
> Andrea

Philipp Blundel told me that this is probably fixed in the gcc-4.4 branch,
 unfornatly I didnt test it yet my self, for now it would be okay to turn
off thumb.

Bye Henning