Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

[David Howells]

Has anyone had any success building a gcc-4.7.0 that can build SH64 binaries
without dying?  I am seeing ICEs and SEGVs:

include/linux/genhd.h:214:3: internal compiler error: Segmentation fault
mm/madvise.c:285:2: internal compiler error: Segmentation fault
mm/percpu.c: In function 'pcpu_free_area':
mm/percpu.c:602:1: error: unrecognizable insn:
(insn 74 73 75 8 (set (reg:SI 279)
        (abs:SI (reg:SI 280))) mm/percpu.c:575 -1
     (nil))
mm/percpu.c:602:1: internal compiler error: in extract_insn, at recog.c:2123

I'm not entirely certain what target triplet I should be using.  I'm currently
using sh64-linux-gnu.  It does build a compiler (and it can find its
binutils).  I've also tried sh64-elf, but that doesn't work either.

David

Re: Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

[Oleg Endo]

On Tue, 2012-05-29 at 18:23 +0100, David Howells wrote:
> Has anyone had any success building a gcc-4.7.0 that can build SH64 binaries
> without dying?  I am seeing ICEs and SEGVs:
> 
> include/linux/genhd.h:214:3: internal compiler error: Segmentation fault
> mm/madvise.c:285:2: internal compiler error: Segmentation fault
> mm/percpu.c: In function 'pcpu_free_area':
> mm/percpu.c:602:1: error: unrecognizable insn:
> (insn 74 73 75 8 (set (reg:SI 279)
>         (abs:SI (reg:SI 280))) mm/percpu.c:575 -1
>      (nil))
> mm/percpu.c:602:1: internal compiler error: in extract_insn, at recog.c:2123
> 

Hmm .. abs:SI ... I'm afraid that was me.  Sorry about that!
Somebody noticed the problem with abs:SI on SH64 a while ago, too.
However, I don't remember whether the fix made it into 4.7.0 or into the
4.7.1 branch.  Could you try with the 4.7.1 branch?

Cheers,
Oleg

Re: Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

[Kaz Kojima]

Oleg Endo <oleg.endo@t-online.de> wrote:
> Hmm .. abs:SI ... I'm afraid that was me.  Sorry about that!
> Somebody noticed the problem with abs:SI on SH64 a while ago, too.
> However, I don't remember whether the fix made it into 4.7.0 or into the
> 4.7.1 branch.  Could you try with the 4.7.1 branch?

There were 3 patches for sh64-elf gcc build failure on trunk:

http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00208.html
http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00209.html
http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00210.html

which are generic for sh64 targets.  The 3rd one is for the abs:SI
issue.  I thought that they are included in 4.7 but they aren't
because they were applied just after branching 4.8 done unfortunately.
I guess that these patches should be backported to 4.7.1 branch.

Regards,
	kaz

Re: Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

[David Howells]

Kaz Kojima <kkojima@rr.iij4u.or.jp> wrote:

> There were 3 patches for sh64-elf gcc build failure on trunk:
> 
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00208.html
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00209.html
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00210.html

They seem to fix the problem, thanks!

Oh, and I think these also fix this:

	http://gcc.gnu.org/bugzilla/show_bug.cgi?idR971

Am I right in that?

David

Re: Has anyone got gcc-4.7.0 to build an SH64 kernel without dying?

[Oleg Endo]

On Wed, 2012-05-30 at 13:53 +0900, Kaz Kojima wrote:
> Oleg Endo <oleg.endo@t-online.de> wrote:
> > Hmm .. abs:SI ... I'm afraid that was me.  Sorry about that!
> > Somebody noticed the problem with abs:SI on SH64 a while ago, too.
> > However, I don't remember whether the fix made it into 4.7.0 or into the
> > 4.7.1 branch.  Could you try with the 4.7.1 branch?
> 
> There were 3 patches for sh64-elf gcc build failure on trunk:
> 
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00208.html
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00209.html
> http://gcc.gnu.org/ml/gcc-patches/2012-03/msg00210.html
> 
> which are generic for sh64 targets.  The 3rd one is for the abs:SI
> issue.  I thought that they are included in 4.7 but they aren't
> because they were applied just after branching 4.8 done unfortunately.
> I guess that these patches should be backported to 4.7.1 branch.

Done :)

Cheers,
Oleg