Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

[Hans-Peter Nilsson]

My autotester noticed:
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/elf.exp ...
FAIL: ld-elf/commonpage1
FAIL: ld-elf/maxpage1

With the message in ld.log:
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/elf.exp ...
/n/asic_slask/hp/autobinutest/ld/../gas/as-new    -o tmpdir/dump0.o /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/maxpage1.s
./ld-new  -L/h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf  -z max-page-size=0x200000 -z common-page-size=0x100000 -o tmpdir/dump tmpdir/dump0.o
failed with: <./ld-new: unrecognized option '-z'
./ld-new: use the --help option for usage information>, expected: <>
./ld-new: unrecognized option '-z'
./ld-new: use the --help option for usage information
/n/asic_slask/hp/autobinutest/ld/../binutils/readelf  -l --wide tmpdir/dump > tmpdir/dump.out
readelf: Error: 'tmpdir/dump': No such file
FAIL: ld-elf/commonpage1

and similar for maxpage1.

It looked like a port buglet, but it also fails for
--target=arm-elf so apparently not the case.

Is the -z option supposed to be Linux-specific and so a
"target: *-*-linux*" added to the top of the .d files the
obvious solution?  Looking at elf32.em
gld${EMULATION_NAME}_add_options it seems -z should be a general
ELF-option, but de facto not.

FWIW, the
tests pass for native i686-pc-linux-gnu and cross to
cris-axis-linux-gnu.

brgds, H-P

Re: Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

[Thiemo Seufer]

Hans-Peter Nilsson wrote:
> My autotester noticed:
> Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/elf.exp ...
> FAIL: ld-elf/commonpage1
> FAIL: ld-elf/maxpage1

For mips-linux:

FAIL: ld-elf/commonpage1
FAIL: ld-elf/maxpage1
FAIL: MIPS multi-got-no-shared


/home/ths/gcc/binutils-upstream/head-build/ld/../gas/as-new    -o tmpdir/dump0.o /home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-elf/maxpage1.s
./ld-new  -L/home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-elf  -z max-page-size=0x200000 -z common-page-size=0x100000 -o tmpdir/dump tmpdir/dump0.o
succeeded with: <./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0>, expected: <>
./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0
succeeded with: <./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0>, expected: <>
FAIL: ld-elf/commonpage1
[snip]
/home/ths/gcc/binutils-upstream/head-build/ld/../gas/as-new    -o tmpdir/dump0.o /home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-elf/maxpage1.s
./ld-new  -L/home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-elf  -z max-page-size=0x200000 -o tmpdir/dump tmpdir/dump0.o
succeeded with: <./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0>, expected: <>
./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0
succeeded with: <./ld-new: warning: cannot find entry symbol __start; defaulting to 00000000004000b0>, expected: <>
FAIL: ld-elf/maxpage1
[snip]
/home/ths/gcc/binutils-upstream/head-build/ld/../gas/as-new  -EB -32 -KPIC -mno-shared  -o tmpdir/dump0.o /home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-mips-elf/multi-got-no-shared-1.s
/home/ths/gcc/binutils-upstream/head-build/ld/../gas/as-new  -EB -32 -KPIC -mno-shared  -o tmpdir/dump1.o /home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-mips-elf/multi-got-no-shared-2.s
./ld-new  -L/home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-mips-elf  -melf32btsmip --entry func1 -o tmpdir/dump tmpdir/dump0.o tmpdir/dump1.o
/home/ths/gcc/binutils-upstream/head-build/ld/../binutils/objdump  -D -j .text --prefix-addresses --show-raw-insn tmpdir/dump > tmpdir/dump.out
regexp_diff match failure
regexp "^004000b0 <[^>]*> 3c1c0046 	lui	gp,0x46$"
line   "004000b0 <func1> 3c1c0043 	lui	gp,0x43"
extra regexps in /home/ths/gcc/binutils-upstream/head/ld/testsuite/ld-mips-elf/multi-got-no-shared.d starting with "^00408d60 <[^>]*> 3c1c0047 	lui	gp,0x47$"
EOF from tmpdir/dump.out
FAIL: MIPS multi-got-no-shared


The ld-elf failures are just the usual missing __start for mips, I
haven't looked into multi-got-no-shared yet.


Thiemo

Re: Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

[Alan Modra]

On Wed, May 31, 2006 at 12:05:45PM +0200, Hans-Peter Nilsson wrote:
> ./ld-new: unrecognized option '-z'
> ./ld-new: use the --help option for usage information

The failure is because cris-elf doesn't support shared libs.
GENERATE_SHLIB_SCRIPT isn't set.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

[Hans-Peter Nilsson]

> Date: Wed, 31 May 2006 23:07:41 +0930
> From: Alan Modra <amodra@bigpond.net.au>

> On Wed, May 31, 2006 at 12:05:45PM +0200, Hans-Peter Nilsson wrote:
> > ./ld-new: unrecognized option '-z'
> > ./ld-new: use the --help option for usage information
> 
> The failure is because cris-elf

And arm-elf, and...

> doesn't support shared libs.
> GENERATE_SHLIB_SCRIPT isn't set.

Inferring that all run_dump-tests using -z options should use
"#target:" or "#notarget:".

Tested cris-elf, arm-elf native i686-pc-linux-gnu and cross to
cris-axis-linux-gnu.  Ok to commit?

ld/testsuite:
	* ld-elf/commonpage1.d, ld-elf/maxpage1.d: Restrict to target
	*-*-linux*.


Index: ld-elf/commonpage1.d
===================================================================
RCS file: /cvs/src/src/ld/testsuite/ld-elf/commonpage1.d,v
retrieving revision 1.1
diff -u -p -r1.1 commonpage1.d
--- ld-elf/commonpage1.d	30 May 2006 16:45:32 -0000	1.1
+++ ld-elf/commonpage1.d	31 May 2006 14:28:40 -0000
@@ -1,5 +1,6 @@
 #source: maxpage1.s
 #ld: -z max-page-size=0x200000 -z common-page-size=0x100000
+#target: *-*-linux*
 #readelf: -l --wide
 
 #...
Index: ld-elf/maxpage1.d
===================================================================
RCS file: /cvs/src/src/ld/testsuite/ld-elf/maxpage1.d,v
retrieving revision 1.1
diff -u -p -r1.1 maxpage1.d
--- ld-elf/maxpage1.d	30 May 2006 16:45:32 -0000	1.1
+++ ld-elf/maxpage1.d	31 May 2006 14:28:40 -0000
@@ -1,5 +1,6 @@
 #source: maxpage1.s
 #ld: -z max-page-size=0x200000
+#target: *-*-linux*
 #readelf: -l --wide
 
 #...

brgds, H-P

[RFA:] Re: Ld option -z mistreat (new tests ld-elf/commonpage1 ld-elf/maxpage1 fail for cris-elf arm-elf)

[Hans-Peter Nilsson]

> Date: Wed, 31 May 2006 16:30:40 +0200
> From: Hans-Peter Nilsson <hp@axis.com>

> Inferring that all run_dump-tests using -z options should use
> "#target:" or "#notarget:".

Similarly other tests needing to use the ld -z option, of
course.  Like these one, which get UNRESOLVED otherwise, thus a
nonzero "make check" exit.  These are the only tests in
binutils.exp and there's a is_elf_format test at the top, but
future tests could presumably be added that don't use -z, so I
put the condition close to the test.  Checked cross to
cris-axis-elf.

Ok to commit?

	* ld-elf/binutils.exp: Restrict tests using ld option -z to target
	*-*-linux*.

Index: binutils.exp
===================================================================
RCS file: /cvs/src/src/ld/testsuite/ld-elf/binutils.exp,v
retrieving revision 1.1
diff -u -p -r1.1 binutils.exp
--- binutils.exp	30 May 2006 16:45:32 -0000	1.1
+++ binutils.exp	31 May 2006 16:22:42 -0000
@@ -76,5 +76,10 @@ proc strip_test { ld_options test } {
     pass "$ld_options"
 }
 
-strip_test "-z max-page-size=0x200000" maxpage1
-strip_test "-z max-page-size=0x200000 -z common-page-size=0x100000" maxpage1
+# Use ld -z options only for GNU/Linux targets; they're unrecognized for
+# targets that don't support dynamic linking.
+
+if [istarget *-*-linux*] {
+    strip_test "-z max-page-size=0x200000" maxpage1
+    strip_test "-z max-page-size=0x200000 -z common-page-size=0x100000" maxpage1
+}

brgds, H-P