Re: [arm-gnu] urgent toolchain

[Gilles Chanteperdrix <gilles.chanteperdrix@xxxxxxxxxxx>]

Gilles Chanteperdrix wrote:
> Carlos O'Donell wrote:
>> Gilles Chanteperdrix wrote:
>>>> Using "-mfpu=vfp -mfloat-abi=softfp" instead of "-msoft-float" will 
>>>> enable VFP hardware floating-point code-generation for your program. The 
>>>> run-time libraries will still use software floating-point. Routines 
>>>> compiled for VFP are compatible with software floating-point.
>>> It looks like run-time libraries do contain some FP code. Here is an
>>> excerpt of disassembling
>>> arm-2007q3/arm-none-linux-gnueabi/libc/lib/libc-2.5.so:
>> The run-time may contain VFP instructions as long as they are never 
>> executed on hardware that does not have a VFP.
>>
>>>    29b10:	0a000002 	beq	29b20 <Lno_vfp>
>>>    29b14:	ecac8b11 	fstmiax	ip!, {d8-d15}
>>>    29b18:	eef12a10 	fmrx	r2, fpscr
>>>    29b1c:	e48c2004 	str	r2, [ip], #4
>> The GNU C Library contains run-time CPU feature detection. The 'beq' 
>> instruction will check the variable "no_vfp" (No VFP available) and skip 
>> the VFP instructions on non-VFP hardware.
>>
>> Are you having a problem with this?
> 
> I do not have checked any further, but I have a VFP platform where the
> Linux kernel fails executing init without CONFIG_VFP enabled.

It turns out that I had made modifications to the Linux kernel which
were bugous and were causing the killing of init with CONFIG_VFP disabled.
Fixing this let init start with CONFIG_VFP disabled. So, sorry for the
noise.

However, I still have the issue that some busybox applets (ps, top)
segfault when busybox is compiled with -march=armv6.

-- 
					    Gilles.