On Wed, May 29, 2019 at 05:41:07PM +0200, Alexander Graf wrote:
> On 14.05.19 16:06, Leif Lindholm wrote:
> > On Tue, May 14, 2019 at 06:32:06AM -0700, Alexander Graf wrote:
> >>> But, wait...
> >>> arm-linux-gnueabi is the softfloat (v5te) toolchain - if we want to
> >>> test that, fine - but we definitely need to test arm-linux-gnueabihf.
> >>>
> >>> First of all, I would expect that this toolchain will not use T32
> >>> (Thumb-2) instructions by default, so much of the code will end up
> >>> being compiled as A32 anyway.
> >>
> >> Yeah, that's why we don't need to pass in -marm I guess. But I also
> >> don't see an explicit hardfloat cross gcc on kernel.org?
> >>
> >> https://mirrors.kernel.org/pub/tools/crosstool/files/bin/x86_64/8.1.0/
> > Urgh.
> > Yeah, so the default output of that toolchain gives
> > File Attributes
> > Tag_CPU_name: "5T"
> > Tag_CPU_arch: v5T
> > Tag_ARM_ISA_use: Yes
> > Tag_THUMB_ISA_use: Thumb-1
> >
> > In theory, you should be OK just adding
> > "-mfloat-abi=hard -march=armv7-a+vfpv3-d16" to CFLAGS, since we're not
> > using any toolchain-provided libraries:
>
> In configure we're already forcing soft float, so there's little point
> in adding these cflags :). In fact, when I do add them, I only get
> configure errors because setting both -mhard-float and -msoft-float at
> the same time confuses gcc.
OK, that bit is less of an issue then.
We only need -march=armv7-a.
/
Leif
_______________________________________________
Grub-devel mailing list
Grub-devel@gnu.org
https://lists.gnu.org/mailman/listinfo/grub-devel
