Bug#278837: ia64 problems
Package: libc6.1
Version: 2.3.2.ds1-18
Tags: patch
This mail was sent to the debian-ia64 list; I'm filing the appropriate
bugs. Please consider including this patch.
----- Forwarded message from David Mosberger <davidm@napali.hpl.hp.com> -----
I have grown increasingly concerned that even with the current
Debian/unstable, there are several failures reported by "make check"
in libunwind. This is disappointing because most of the necessary
toolchain/libc fixes have been out there for a long time but things
just haven't moved far along enough yet for Debian to pick them up in
the normal course of syncing with upstream. To make matters worse, I
recently found on a SuSE 9 system that even with the existing fixes in
place, there were two other unwind-related bugs in GCC. Those have
been fixed in upstream now and as of now, there are no known
unwind-related bugs left. Thus, it would be really good to get the
Debian toolchain in sync. Since Debian won't upgrade to gcc-3.4 and
the latest libc over night, I thought I'd spend the effort to backport
the minimal fixes to get things working right. The result is the 3
patches below, which fix, respectively, GCC v3.3, libc6.1, and
binutils.
Now the problem is that I'd really need help to get these patches
integrated into Debian. Any help in that direction would be greatly
appreciated (especially since I'll be distracted with my move back to
California over the next few days).
In terms of the safety of these patches: they all have been checked
into the mainline CVS trees and most have been there for months. I'm
convinced they're all safe for ia64 and even more so for all other
arches (since they're for the most part 100% unaffected). So really
hope that these could be merged quickly. Of course, I'd be happy to
answer any questions in case something comes up.
Also, it would be good if GCC on ia64 could be built against
libunwind. To do so, we'd probably want to make sure that Debian has
libunwind v0.98.2 (which I'll release next week and contains just two
minor bug fixes vs. v0.98.1) installed when building GCC and glibc.
The rest should be automatic.
With the patches below installed, "make check" in libunwind completes
without failures. Also, I single-stepped a test-program
(test-ptrace-misc) from beginning to end, unwinding after each
instruction and no failures where detected! Also, the GCC patches
fixes the 2 remaining unexpected libjava in GCC 3.3.5 (with no other
changes), so I really do think this gets Debian/ia64 into much better
shape.
Anybody willing to help getting these into Debian?
--david
Content-Description: glibc fixes
Makeconfig | 18 ++++++++-
Makerules | 4 +-
config.h.in | 3 +
config.make.in | 1
configure.in | 17 +++++++++
nptl/sysdeps/unix/sysv/linux/ia64/pt-initfini.c | 45 +++++++++++++++++++++---
6 files changed, 78 insertions(+), 10 deletions(-)
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makeconfig glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makeconfig
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makeconfig 2004-10-29 09:19:49.973173566 -0700
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makeconfig 2004-10-28 10:47:59.000000000 -0700
@@ -498,12 +498,12 @@
# The static libraries.
ifeq (yes,$(build-static))
-link-libc-static = $(common-objpfx)libc.a $(gnulib) $(common-objpfx)libc.a
+link-libc-static = $(common-objpfx)libc.a $(static-gnulib) $(common-objpfx)libc.a
link-extra-libs-static = $(foreach lib,$(LDLIBS-$(@F)),$(common-objpfx)$(lib).a)
else
ifeq (yes,$(build-shared))
# We can try to link the programs with lib*_pic.a...
-link-libc-static = $(gnulib) $(common-objpfx)libc_pic.a
+link-libc-static = $(static-gnulib) $(common-objpfx)libc_pic.a
link-extra-libs-static = $(link-extra-libs)
endif
endif
@@ -511,7 +511,19 @@
link-extra-libs-bounded = $(foreach lib,$(LDLIBS-$(@F:%-bp=%)),$(common-objpfx)$(lib)_b.a)
ifndef gnulib
-gnulib := -lgcc -lgcc_eh
+ifneq ($(have-cc-with-libunwind),yes)
+ libunwind =
+else
+ libunwind = -lunwind
+endif
+ifneq ($(have-as-needed),yes)
+ libgcc_eh := -lgcc_eh $(libunwind)
+else
+ libgcc_eh := -Wl,--as-needed -lgcc_s$(libgcc_s_suffix) $(libunwind) -Wl,--no-as-needed
+endif
+gnulib := -lgcc $(libgcc_eh)
+static-gnulib := -lgcc -lgcc_eh $(libunwind)
+libc.so-gnulib := -lgcc
endif
ifeq ($(elf),yes)
+preinit = $(addprefix $(csu-objpfx),crti.o)
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makerules glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makerules
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makerules 2004-10-29 09:19:50.451682658 -0700
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/Makerules 2004-10-28 07:58:29.000000000 -0700
@@ -569,8 +569,8 @@
# Also omits crti.o and crtn.o, which we do not want
# since we define our own `.init' section specially.
LDFLAGS-c.so = -nostdlib -nostartfiles
-# But we still want to link libc.so against $(gnulib).
-LDLIBS-c.so += $(gnulib)
+# But we still want to link libc.so against $(libc.so-gnulib).
+LDLIBS-c.so += $(libc.so-gnulib)
# Give libc.so an entry point and make it directly runnable itself.
LDFLAGS-c.so += -e __libc_main
# Pre-link the objects of libc_pic.a so that we can locally resolve
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.h.in glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.h.in
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.h.in 2004-10-29 09:19:46.026938378 -0700
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.h.in 2004-10-28 08:02:30.000000000 -0700
@@ -153,6 +153,9 @@
sections. */
#undef HAVE_INITFINI_ARRAY
+/* Define if the compiler's exception support is based on libunwind. */
+#undef HAVE_CC_WITH_LIBUNWIND
+
/* Define if the access to static and hidden variables is position independent
and does not need relocations. */
#undef PI_STATIC_AND_HIDDEN
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.make.in glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.make.in
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.make.in 2004-10-29 09:19:50.853044366 -0700
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/config.make.in 2004-10-28 09:11:43.000000000 -0700
@@ -53,6 +53,7 @@
have-cpp-asm-debuginfo = @libc_cv_cpp_asm_debuginfo@
enable-check-abi = @enable_check_abi@
have-forced-unwind = @libc_cv_forced_unwind@
+have-cc-with-libunwind = @libc_cv_cc_with_libunwind@
fno-unit-at-a-time = @fno_unit_at_a_time@
static-libgcc = @libc_cv_gcc_static_libgcc@
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/configure.in glibc-2.3.2.ds1/build-tree/glibc-2.3.2/configure.in
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/configure.in 2004-10-29 09:19:50.849138169 -0700
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/configure.in 2004-10-28 07:55:27.000000000 -0700
@@ -1251,6 +1251,23 @@
AC_DEFINE(HAVE_INITFINI_ARRAY)
fi
+ AC_CACHE_CHECK(for libunwind-support in compiler,
+ libc_cv_cc_with_libunwind, [
+ cat > conftest.c <<EOF
+int main (void) { return 0; }
+EOF
+ if ${CC-cc} $CFLAGS $CPPFLAGS $LDFLAGS -static -o conftest \
+ conftest.c -v 2>&1 >/dev/null | grep -q " -lunwind "; then
+ libc_cv_cc_with_libunwind=yes
+ else
+ libc_cv_cc_with_libunwind=no
+ fi
+ rm -f conftest*])
+ AC_SUBST(libc_cv_cc_with_libunwind)
+ if test $libc_cv_cc_with_libunwind = yes; then
+ AC_DEFINE(HAVE_CC_WITH_LIBUNWIND)
+ fi
+
AC_CACHE_CHECK(for -z nodelete option,
libc_cv_z_nodelete, [dnl
cat > conftest.c <<EOF
diff -urN orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/nptl/sysdeps/unix/sysv/linux/ia64/pt-initfini.c glibc-2.3.2.ds1/build-tree/glibc-2.3.2/nptl/sysdeps/unix/sysv/linux/ia64/pt-initfini.c
--- orig/glibc-2.3.2.ds1/build-tree/glibc-2.3.2/nptl/sysdeps/unix/sysv/linux/ia64/pt-initfini.c 2003-03-11 01:20:41.000000000 -0800
+++ glibc-2.3.2.ds1/build-tree/glibc-2.3.2/nptl/sysdeps/unix/sysv/linux/ia64/pt-initfini.c 2004-10-29 08:42:00.835176367 -0700
@@ -36,6 +36,22 @@
* crtn.s puts the corresponding function epilogues
in the .init and .fini sections. */
+#include <stddef.h>
+
+#ifdef HAVE_INITFINI_ARRAY
+
+__asm__ ("\n\
+#include \"defs.h\"\n\
+\n\
+/*@HEADER_ENDS*/\n\
+\n\
+/*@_init_PROLOG_BEGINS*/\n\
+ .xdata8 \".init_array\",@fptr(__pthread_initialize_minimal_internal)\n\
+/*@_init_PROLOG_ENDS*/\n\
+");
+
+#else
+
__asm__ ("\n\
\n\
#include \"defs.h\"\n\
@@ -48,13 +64,16 @@
.global _init#\n\
.proc _init#\n\
_init:\n\
+ .prologue\n\
+ .save ar.pfs, r34\n\
alloc r34 = ar.pfs, 0, 3, 0, 0\n\
+ .vframe r32\n\
mov r32 = r12\n\
+ .save rp, r33\n\
mov r33 = b0\n\
+ .body\n\
adds r12 = -16, r12\n\
;;\n\
-/* we could use r35 to save gp, but we use the stack since that's what\n\
- * all the other init routines will do --davidm 00/04/05 */\n\
st8 [r12] = gp, -16\n\
br.call.sptk.many b0 = __pthread_initialize_minimal_internal# ;;\n\
;;\n\
@@ -62,14 +81,18 @@
;;\n\
ld8 gp = [r12]\n\
;;\n\
- .align 16\n\
.endp _init#\n\
\n\
/*@_init_PROLOG_ENDS*/\n\
\n\
/*@_init_EPILOG_BEGINS*/\n\
.section .init\n\
- .regstk 0,2,0,0\n\
+ .proc _init#\n\
+ .prologue\n\
+ .save ar.pfs, r34\n\
+ .vframe r32\n\
+ .save rp, r33\n\
+ .body\n\
mov r12 = r32\n\
mov ar.pfs = r34\n\
mov b0 = r33\n\
@@ -83,18 +106,28 @@
.global _fini#\n\
.proc _fini#\n\
_fini:\n\
+ .prologue\n\
+ .save ar.pfs, r34\n\
alloc r34 = ar.pfs, 0, 3, 0, 0\n\
+ .vframe r32\n\
mov r32 = r12\n\
+ .save rp, r33\n\
mov r33 = b0\n\
+ .body\n\
adds r12 = -16, r12\n\
;;\n\
- .align 16\n\
.endp _fini#\n\
\n\
/*@_fini_PROLOG_ENDS*/\n\
\n\
/*@_fini_EPILOG_BEGINS*/\n\
.section .fini\n\
+ .proc _fini#\n\
+ .prologue\n\
+ .save ar.pfs, r34\n\
+ .vframe r32\n\
+ .save rp, r33\n\
+ .body\n\
mov r12 = r32\n\
mov ar.pfs = r34\n\
mov b0 = r33\n\
@@ -106,3 +139,5 @@
/*@TRAILER_BEGINS*/\n\
.weak __gmon_start__#\n\
");
+
+#endif
----- End forwarded message -----
--
"Next the statesmen will invent cheap lies, putting the blame upon
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince
himself that the war is just, and will thank God for the better sleep
he enjoys after this process of grotesque self-deception." -- Mark Twain
Reply to: