MIPS failures on 2.14 branch

[Daniel Jacobowitz <drow at mvista dot com>]

Hey MIPS folks ... we've got these, for mipsel-linux:

regexp_diff match failure
regexp "^ 1000 (1010)?0000(1010)? (1210)?0000(1012)? (0c)?000000(0c)? (0e)?000000(0e)? .*$"
line   " 1000 10100000 12100000 0c000000 16000000  ................"
FAIL: ld-elf/merge

  Appears to be an assembler bug.  The .o file looks like:
dump0.o:     file format elf32-tradlittlemips

Contents of section .text:
 0000 00000000 04000000 04000000 10000000  ................

Um... the last two should be four bytes apart... not twelve.  There's also a
valgrind error while running gas on this simple file.  It shows up in
_bfd_elf_write_object_contents, which suggests to me that the in-place
addend was never initialized.

We've also got:
./ld-new  -L/opt/mvista/BUILD/binutils-2.13.90/ld/testsuite/ld-mips-elf -melf32btsmip -shared -o tmpdir/dump tmpdir/dump0.o tmpdir/dump1.o
failed with: <./ld-new: tmpdir/dump0.o: compiled for a little endian system and target is big endian
FAIL: MIPS multi-got-1

And similiarly for rel32/o32, rel32/n32, and rel64/n64.  Those are easy to
fix; add -EB to the as: lines in the relevant .d files.  If you fix
multi-got-1 similarly then you'll get regex failures starting with:

regexp "^000f6710  00201603 R_MIPS_REL32      0009f750   sym_2_3817$"
line   "000f9edc  00201603 R_MIPS_REL32      0009f750   sym_2_3817"

i.e. the addresses have changed.

If you want something _really_ cute, remove the -melf32btsmipn32 from the ld
line in rel32-n32.d and watch ld SIGSEGV.


I'd really appreciate it if someone could look into the two of these which
aren't trivial.  They are also visible on mips-linux (i.e. big endian).  I'd
at least like them casually analyzed before 2.14...

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer