You are not logged in.

#1 2017-10-27 15:05:09

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

valgrind

Running on an older machine:

shell> valgrind ls -al
vex x86->IR: unhandled instruction bytes: 0x67 0xE8 0xD3 0x8B

Offline

#2 2017-10-27 15:24:37

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

layout asm in gdb shows me something like:

>│0x40021c7       addr16 call 0x401ada0                                                               │

So it's not SSE3 or AVX2 code creeping into the builds.

I'm trying to rebuild valgrind, maybe it's a glibc interface change issue..

Offline

#3 2017-10-27 15:38:17

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

Doesn't look like it..

Offline

#4 2017-10-27 18:18:06

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

So, a small test program:

#include <stdio.h>

int main
( void ) {
        puts( "hello" );
        return 0;
}

Using gdbserver I see:

Program received signal SIGILL, Illegal instruction.
0x040021c7 in dl_main () from /lib/ld-linux.so.2
(gdb) x $pc
0x40021c7 <dl_main+1047>:       0x8bd3e867

the sequence fits to the unknown opcodes reported by valgrind, and

>│0x40021c7 <dl_main+1047>        addr16 call 0x401ada0 <memcmp>                                      │

So it fails calling memcmp?

Offline

#5 2017-10-27 18:36:00

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

Actually, I start to think valgrind 3.13.0 has issues with glibc 2.26 on 32-bit. SIGILL is not signalling an illegal opcode,
it's just a signal which may easier pass to the users attention and actually kill the instrumented process.
I'll test with the vanilla and the git version of valgrind..

Offline

#6 2017-10-27 18:50:04

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

Ok, same happens with 3.14.0.GIT c1eace647ca4f670ef9bec0d0fe72cdd25a96394, I think, I have to take this upstream..

Offline

#7 2017-10-27 19:06:19

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

Ah, there it is https://bugs.kde.org/show_bug.cgi?id=386115.
https://bugs.kde.org/show_bug.cgi?id=384230
addr16 prefix for call address. :-)

I'll try to backport the patch and test again.

Offline

#8 2017-10-27 19:34:30

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Offline

#9 2017-10-27 19:39:07

andreas_baumann
Administrator
From: Zurich, Switzerland
Registered: 2017-08-10
Posts: 833
Website

Re: valgrind

Reported: https://bugs.archlinux.org/task/56141, maybe a backport still makes sense?

Offline

Board footer

Powered by FluxBB