Re: A quick architecture status report

[Joseph Myers <joseph at codesourcery dot com>]

On Sun, 9 Jul 2017, Zack Weinberg wrote:

> Since I was tinkering with build-many-glibcs anyway, here is a report
> on all the failures I currently see.
> 
> Overall things seem to be in good shape, but several architectures
> have ELF-lint test failures:
> 
> FAIL: glibcs-ia64-linux-gnu check
>     FAIL: elf/check-execstack
> 
> FAIL: glibcs-microblaze-linux-gnu check
>     FAIL: elf/check-execstack
> 
> FAIL: glibcs-microblazeel-linux-gnu check
>     FAIL: elf/check-execstack
> 
> FAIL: glibcs-powerpc64le-linux-gnu check
>     FAIL: elf/check-localplt

The powerpc64le-linux-gnu failure is the issue under discussion with a 
libgcc IFUNC doing things that aren't safe in IFUNCs.  (And build for 
powerpc64le-linux-gnu fails since --enable-profile was added to 
build-many-glibcs.py.)

> FAIL: glibcs-hppa-linux-gnu check
>     (with JDA's patches for HPPA pthreads applied)
>     FAIL: elf/check-execstack
>     FAIL: elf/check-localplt
>     FAIL: elf/check-textrel
> 
> These shouldn't be hard to fix, for someone who knows the
> architecture.  I don't, but I might try to fix them anyway.

There are at least three possible explanations for check-execstack 
failures, each of which should be addressed in a different way.

* Architecture does not support the page permissions required for non-exec 
stacks: XFAIL with a comment to that effect.

* Architecture has the required support, all that's needed is a 
straightforward GCC change to include the required section in 
compiler-generated objects and libgcc objects from assembly sources: make 
that GCC change.

* Support is possible, but more complicated toolchain / kernel changes are 
needed for some architecture-specific reason: XFAIL with a comment 
explaining the issues, as on MIPS.

-- 
Joseph S. Myers
joseph@codesourcery.com