Sergey Bugaev writes: Hello Sergey,
> On Fri, Nov 22, 2024 at 1:52 PM <jann...@gnu.org> wrote:
>> Hi!
>
> Hi Janneke,
>
>> I have bisected the problem to be in sigaction.o: when linking with
>> sigaction.o from Debian's libcrt.a it passes, when using Guix's
>> sigaction.o it fails. Problem solved, you would say?
>
> I assume you mean sigreturn.o, not sigaction.o
Sure, sigreturn.o, what a silly typo to make!
>> Anyway, attached are the disassembled versions of Debian's and Guix's
>> sigaction.o.d. Maybe some of you can spot the significant difference
>> and where it might come from?
>
> From diff'ing the two files, one of them has
>
> 46: 64 44 89 24 25 80 00 mov %r12d,%fs:0x80
>
> after the call to __mach_port_mod_refs (side note: it would be easier
> to analyze disassembly if you passed --reloc to objdump), where %r12d
> was previously
>
> 6: 41 89 d4 mov %edx,%r12d
>
> this corresponds to
>
> THREAD_SETMEM (THREAD_SELF, reply_port, sc_reply_port);
>
> in the source code. The other one doesn't do this, and seems to
> clobber %edx without saving it.
Oh wow, I would never have found that! We're cross-compiling this
bootstrap-glibc for Guix, so we would have to check if everything
was configured OK. I restarted the build and killed it after
sigreturn.o was compiled, and verified that the assembly is the
same. So far so good.
In sysdeps/mach/hurd/x86_64/sigreturn.c, I added a redefine of
THREAD_SETMEM
--8<---------------cut here---------------start------------->8---
#include <hurd/msg.h>
#include <stdlib.h>
+#define THREAD_SETMEM foobar
+
/* This is run on the thread stack after restoring it, to be able to
--8<---------------cut here---------------end--------------->8---
to find out which one is being used and then compiled sigreturn.c again
using the (amazing) command-line that glibc creates
--8<---------------cut here---------------start------------->8---
$ x86_64-pc-gnu-gcc ../sysdeps/mach/hurd/x86_64/sigreturn.c -c -std=gnu11
-fgnu89-inline -g -O2 -Wall -Wwrite-strings -Wundef -fmerge-all-constants
-frounding-math -fno-stack-protector -fno-common -Wp,-U_FORTIFY_SOURCE
-Wno-parentheses -Wstrict-prototypes -Wold-style-definition -fmath-errno
-fPIE -fno-stack-protector -DSTACK_PROTECTOR_LEVEL=0
-ftls-model=initial-exec -I../include
-I/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build/signal
-I/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build
-I../sysdeps/mach/hurd/x86_64 -I../sysdeps/mach/hurd/x86
-I../sysdeps/mach/hurd/x86_64/htl -I../sysdeps/mach/hurd/htl
-I../sysdeps/hurd/htl -I../sysdeps/mach/htl -I../sysdeps/htl/include
-I../sysdeps/htl -I../sysdeps/pthread -I../sysdeps/x86_64/htl
-I../sysdeps/x86/htl -I../sysdeps/mach/hurd -I../sysdeps/gnu
-I../sysdeps/unix/bsd -I../sysdeps/unix/inet -I../sysdeps/mach/x86_64
-I../sysdeps/mach/x86 -I../sysdeps/mach/include -I../sysdeps/mach
-I../sysdeps/x86_64/64 -I../sysdeps/x86_64/fpu/multiarch
-I../sysdeps/x86_64/fpu -I../sysdeps/x86/fpu -I../sysdeps/x86_64/multiarch
-I../sysdeps/x86_64 -I../sysdeps/x86/include -I../sysdeps/x86
-I../sysdeps/ieee754/float128 -I../sysdeps/ieee754/ldbl-96/include
-I../sysdeps/ieee754/ldbl-96 -I../sysdeps/ieee754/dbl-64
-I../sysdeps/ieee754/flt-32 -I../sysdeps/hurd/include -I../sysdeps/hurd
-I../sysdeps/unix -I../sysdeps/posix -I../sysdeps/wordsize-64
-I../sysdeps/ieee754 -I../sysdeps/generic -I../hurd
-I/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build/hurd/ -I../mach
-I/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build/mach/ -I..
-I../libio -I. -nostdinc -isystem
/gnu/store/6nfg25y6hfp8mx7j1cwq995mcm0kf6hw-gcc-cross-sans-libc-x86_64-pc-gnu-14.2.0-lib/lib/gcc/x86_64-pc-gnu/14.2.0/include
-isystem
/gnu/store/6nfg25y6hfp8mx7j1cwq995mcm0kf6hw-gcc-cross-sans-libc-x86_64-pc-gnu-14.2.0-lib/lib/gcc/x86_64-pc-gnu/14.2.0/include-fixed
-isystem
/gnu/store/axql682fcn94kkdyqgy1vv9pcx7n24zv-hurd-core-headers-cross-x86_64-pc-gnu-0.9.git20240714/include
-D_LIBC_REENTRANT -include
/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build/libc-modules.h
-DMODULE_NAME=libc -include ../include/libc-symbols.h -DPIC
-DTOP_NAMESPACE=glibc -o
/tmp/guix-build-glibc-cross-x86_64-pc-gnu-2.39.drv-0/build/signal/sigreturn.o
../sysdeps/mach/hurd/x86_64/sigreturn.c:23:9: warning: "THREAD_SETMEM" redefined
23 | #define THREAD_SETMEM foobar
| ^~~~~~~~~~~~~
In file included from ../sysdeps/x86/atomic-machine.h:23,
from ../include/atomic.h:49,
from ../mach/lowlevellock.h:24,
from ../mach/lock-intern.h:23,
from ../sysdeps/mach/include/lock-intern.h:2,
from ../mach/spin-lock.h:22,
from ../hurd/hurd/signal.h:41,
from ../sysdeps/hurd/include/hurd/signal.h:9,
from ../hurd/hurd/userlink.h:29,
from ../hurd/hurd/port.h:25,
from ../sysdeps/hurd/include/hurd/port.h:2,
from ../hurd/hurd.h:41,
from ../sysdeps/hurd/include/hurd.h:2,
from ../sysdeps/mach/hurd/x86_64/sigreturn.c:18:
../sysdeps/mach/hurd/x86_64/tls.h:92:10: note: this is the location of the
previous definition
92 | # define THREAD_SETMEM(descr, member, value)
\
| ^~~~~~~~~~~~~
--8<---------------cut here---------------end--------------->8---
Hmm, that doesn't look so bad? That file is unmodified
<https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/mach/hurd/x86_64/tls.h;h=3d3253963b5de04d10adf93fe52c5956108e2fd4;hb=ef321e23c20eebc6d6fb4044425c00e6df27b05f#l92>
86 # define THREAD_SELF
\
87 (*(tcbhead_t * __seg_fs *) offsetof (tcbhead_t, tcb))
88 /* Read member of the thread descriptor directly. */
89 # define THREAD_GETMEM(descr, member)
\
90 (*(__typeof (descr->member) __seg_fs *) offsetof (tcbhead_t, member))
91 /* Write member of the thread descriptor directly. */
92 # define THREAD_SETMEM(descr, member, value)
\
93 (*(__typeof (descr->member) __seg_fs *) offsetof (tcbhead_t, member)
= value)
...hmm, might offsetof be the problem here? Same trick...
../sysdeps/mach/hurd/x86_64/sigreturn.c:24:9: warning: "offsetof" redefined
24 | #define offsetof barbaz
| ^~~~~~~~
In file included from ../posix/sched.h:29,
from ../include/sched.h:2,
from ../sysdeps/htl/pthread.h:49,
from ../sysdeps/htl/include/pthread.h:2,
from ../include/pthread.h:1,
from ../sysdeps/htl/libc-lockP.h:22,
from ../sysdeps/mach/libc-lock.h:223,
from ../sysdeps/htl/libc-lock.h:22,
from ../sysdeps/hurd/include/hurd/port.h:5:
/gnu/store/6nfg25y6hfp8mx7j1cwq995mcm0kf6hw-gcc-cross-sans-libc-x86_64-pc-gnu-14.2.0-lib/lib/gcc/x86_64-pc-gnu/14.2.0/include/stddef.h:416:9:
note: this is the location of the previous definition
416 | #define offsetof(TYPE, MEMBER) __builtin_offsetof (TYPE, MEMBER)
| ^~~~~~~~
eh...hmm? Last thing I tried was marking the THREAD_SETMEM calls with
NOPS
reply_port = THREAD_GETMEM (THREAD_SELF, reply_port);
asm ("nop");
THREAD_SETMEM (THREAD_SELF, reply_port, MACH_PORT_DEAD);
asm ("nop"); asm ("nop");
if (__glibc_likely (MACH_PORT_VALID (reply_port)))
(void) __mach_port_mod_refs (__mach_task_self (), reply_port,
MACH_PORT_RIGHT_RECEIVE, -1);
asm ("nop"); asm ("nop"); asm ("nop");
THREAD_SETMEM (THREAD_SELF, reply_port, sc_reply_port);
asm ("nop"); asm ("nop"); asm ("nop"); asm ("nop");
and I have attached the the objdump --reloc of this. Can you make
anything of this?
>> Note: on our 32bit Hurd which uses the same patch set for glibc, the
>> test runs fine. For now I have disabled running check on
>> findutils...but I'm afraid this test will keep coming back and fail for
>> every package that uses gnulib :-(
>
> Well, sigreturn is arch-specific, i386-gnu and x86_64-gnu (and
> aarch64-gnu) all have different implementations.
Right.
Greetings,
Janneke
sigreturn+nop.o.d
Description: Binary data
-- Janneke Nieuwenhuizen <jann...@gnu.org> | GNU LilyPond https://LilyPond.org Freelance IT https://www.JoyOfSource.com | Avatar® https://AvatarAcademy.com
