[Buildroot] [PATCH 0/5] Add ILP32 support in aarch64

[Bamvor Zhang]

2015-3-17 21:30  "Gustavo Zacarias" <gustavo at zacarias.com.ar> wrote:
>
> On 03/16/2015 07:00 AM, Zhang Jian(Bamvor) wrote:
> > LP64 is the default ABI on arm64(Implies -mabi=lp64 while compiling),
and the
> > kernel remains in LP64 no matter ILP32 enables or not.
> >
> > These series patches introduce the big endian support in aarch64 and
then add
> > ILP32 special filename and menuconfig step by step.
> >
> > The test of build is based on linaro kernel [1] and toolchain built by
linaro
> > ABE environment[2].
> >
> > This is my first time I try to contribute the buildroot, feedback is
very
> > appreciated. Sorry if there are some coding style issues.
>
> Hi.
> A couple of issues i've discovered with this patchset are:
>
Hi, Gustavo

> 1) You're using -mabi unconditionally.
> Current default gcc is 4.8.x which doesn't understand -mabi for aarch64.
Yes.
>
> 2) You're not taking care of the internal toolchain backend for ILP32.
> Currently we default to gcc 4.8.x which doesn't handle ilp32 at all
> hence errors out for aarch64 in that configuration (doesn't understand
> -mabi, point 1).
> The proper action would be to lock down available versions in
> package/gcc/Config.in.host via "depends on !BR2_AARCH64_ILP32"
> (basically just for 4.8.x since previous versions are already locked out
> for aarch64 in general).
I will do it in next version.
> This will only fix the ILP32 problem, doesn't help for -mabi=lp64 though
> (point 1).
Well, do you mean I should handle -mabi=lp64 as well? It is the default api
in aarch64 gcc 4.9.x. It may be useful when build the kernel.
> 3) ILP32 little endian doesn't build with gcc 4.9.x/binutils 2.25/glibc
> 2.20: it results in an internal compiler error while building glibc, so
> very likely gcc's fault. This should be handled as well, by fixing it or
> disabling the internal toolchain backend for this combination.
I guess linaro has the patch in it while I hope I could deal with it after
the external toolchain patches is accept. Do you comfortable with only
external toolchain support at this time?
>
> 3) Big endian support is broken for the internal toolchain.
> The glibc package isn't enabled for aarch64 BE, so basically it builds a
> toolchain without a libc, not nice.
> Fix in toolchain/toolchain-buildroot/Config.in - add appropiate
> BR2_aarch64_be depends.
Ok,  i will do it.
>
> 4) We don't know/have any external toolchains to test this.
> Suggestions? :)
We could test it with linaro gcc built from linaro abe system.
>
> 5) You're not setting KERNEL_ARCH appropiately for big endian, right now
> the regex is: -e s/aarch64/arm64/
> It should be: -e s/aarch64.*/arm64/
> Otherwise we end up with KERNEL_ARCH being arm64_be which obviously
> fails badly, not only for kernel builds, but also for internal
> toolchains builds (headers), possibly some packages and linux extensions.
> Fix in top Makefile.
Ok.
>
> There are probably other details missing, these are the quick/big ones i
> could see.
maybe I should add RFC for my next verion of patches.

regards

bamvor
>
> Regards.
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20150317/7cfdc158/attachment-0002.html>