Re: [PATCH v6 00/13] GLIBC LoongArch PATCHES

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 08/07/22 04:13, Xi Ruoyao via Libc-alpha wrote:
> On Fri, 2022-07-08 at 14:52 +0800, caiyinyu wrote:
> 
>> binutils:       https://github.com/loongson/binutils-gdb/tree/dev/new_reloc
> 
> There are too much changes in this branch.  Could you isolate a minimal
> binutils patch enough for eliminating R_LARCH_NONE and R_LARCH_IRELATIVE
> from .rela.plt, get it reviewed, and push it into GNU binutils repo?  I
> don't think it's a good idea to make *GNU* libc depend on some external
> linker code which is not reviewed by GNU binutils developers.
> 
> And we need to do this ASAP: we already missed binutils 2.39 release
> branch cut date [1] and if we don't move fast we'll miss glibc-2.36
> release date.
> 
> [1]: https://sourceware.org/pipermail/binutils/2022-July/121567.html
> 
> 

I finished my review for the port and it looks ok in general, however there
some pieces that would require a v7:

  1. Add R_LARCH_NONE handling on bootstrap, binutils 2.38 does generated it 
     and from previous discussion it should be considered a missed 
     optimizations instead of a linker error.

  2. Remove the __loongarch_soft_float parts, since there is no support for
     soft floating-point.

  3. Remove sysdeps/unix/sysv/linux/loongarch/ldconfig.h, this file is not
     required.

  4. Some minor style issues.

  5. Either remove HAVE_GETTIMEOFDAY_VSYSCALL or add a gettimeofday ifunc
     optimization.

The only part really prevent port inclusion is 1.  I am also assuming ifunc 
is support (at least you have added support on the Linux ABI part) with
binutils 2.38.  If not, you will need to remove support until you fix it on
binutils.

And also, you need to check and report the test results using the expected
defined release branches, using out-of-tree branches are not acceptable for
inclusion.