Re: Fix some PR ld/3809 MIPS failures (part 2)

[Hans-Peter Nilsson <hans-peter dot nilsson at axis dot com>]

> From: Richard Sandiford <rdsandiford@googlemail.com>
> Date: Mon, 19 Mar 2012 22:13:21 +0100

> Hans-Peter Nilsson <hans-peter.nilsson@axis.com> writes:
> > Namely, that for 32-bit embedded MIPS targets, there this test failure:
> >
> > Running
> > /mnt/test1/hp/axis/binutils/src/ld/testsuite/ld-mips-elf/mips-elf.exp ...
> > ...
> > FAIL: Global calls from mips16
> >
> > Looking at the dump file and the
> > ld-mips-elf/mips16-call-global.d file, the issue seems to be
> > that sections are 16-byte aligned everywhere else than for
> > mips*-elf and mips*-vxworks; see comment in the patch.  So,
> > don't run the tests on those targets.
> 
> I'd prefer adding ".align 4" to each .s file.  That's preapproved
> if it works.

It doesn't, neither does an explicit --section-start for the
values I've tried (0x10090, 0x40090, 0x400090 (section overlap),
0x4000090).

Different targets have .reginfo and whatnot getting in the way.
The closest I can get still results in:

Executing on host: sh -c {./ld-new   -o tmpdir/mips16-call-global -L/home/hp/binutils/src/ld/testsuite/ld-mips-elf  --section-start=.text=0x4000090 tmpdir/mips16-call-global-1.o tmpdir/mips16-call-global-2.o tmpdir/mips16-call-global-3.o 2>&1}  /dev/null ld.tmp (timeout = 300)
Executing on host: sh -c {/home/hp/binutils/obj/mi/ld/../binutils/objdump -dr tmpdir/mips16-call-global >dump.out 2>ld.stderr}  /dev/null  (timeout = 300)
/home/hp/binutils/obj/mi/ld/../binutils/objdump -dr tmpdir/mips16-call-global
regexp_diff match failure
regexp "^.*0092:	1a00 002e	jal	.*00b8 <x\+0x8>$"
line   " 4000092:	1808 002e	jal	40000b8 <x+0x8>"
regexp_diff match failure
regexp "^.*0098:	1e00 0032	jalx	.*00c8 <z>$"
line   " 4000098:	1c08 0032	jalx	40000c8 <z>"
FAIL: Global calls from mips16

brgds, H-P