Ticket #338 (closed defect: fixed)

Opened 3 months ago

Last modified 3 months ago

Crash at start-up on linux-x8632

Reported by: stassats Assigned to: gb
Priority: major Milestone: IA-32 port
Component: Runtime (threads, GC) Version: trunk
Keywords: Cc:

Description

CCL from trunk crashes on linux-x8632 (2.6.24 kernel, glibc-2.7, Pentim 4 prescott CPU). Here is backtrace: 

(#xB7E3CF94) #x1402AD9D : #<Function (:INTERNAL REVIVE-SHARED-LIBRARIES) #x1402AD36> + 103
(#xB7E3CFB0) #x140283FD : #<Function %WALK-SHARED-LIBRARIES #x1402835E> + 159
(#xB7E3CFC4) #x1402A9CD : #<Function REVIVE-SHARED-LIBRARIES #x1402A81E> + 431
(#xB7E3CFE0) #x1402B725 : #<Function REFRESH-EXTERNAL-ENTRYPOINTS #x1402B61E> + 263
(#xB7E3CFE8) #x1480A5ED : #<Function RESTORE-LISP-POINTERS #x1480A5A6> + 71

Registers: 

%eax = 0x00000000
%ecx = 0x00000000
%edx = 0x1490F83E
%ebx = 0xFFFF84B0
%esp = 0xB7DE0F78
%ebp = 0xB7DE0F94
%esi = 0xFFFF84B0
%edi = 0x14026E96
%eip = 0x1402737C

%ebx (arg_z) = -7892
%esi (arg_y) = -7892
------
%edi (fn) = #<Function SONAME-PTR-FROM-LINK-MAP #x14026E96>
------
%ecx (temp0) = 0
%edx (temp1) = #<MACPTR #xB7FC8940>

Change History

09/25/08 20:20:33 changed by gb

  • status changed from new to assigned.

Thanks. I'm fairly sure that this is the same thing as was reported in:

<http://clozure.com/mailman/htdig/openmcl-devel/2007-October/007641.html>

I can reproduce this on a 2.6.24 system here.

If it is the same thing, rebooting with "vdso=0" on the (GRUB or equivalent) command line may work around it; the real problem is in trying to determine if a field in a union that can be used as either way is a relative offset or as an absolute address. Linux generally adheres to the ELF spec, but didn't do so in 2.6.23/24 (at least) with regards to the "vdso" library.

09/26/08 09:45:32 changed by gb

I checked in a change (r10881) that seems to fix this on the 2.6.24 system that I have access to (and also checked in an LX86CL that includes that change.)

Does this fix the problem for you ?

09/26/08 10:07:14 changed by stassats

Yes, it does.

09/26/08 10:09:48 changed by stassats

And I also accidentally noticed, that the same issue was on 64-bit 2.6.26 kernel with 32-bit libraries, and now it's gone too.

09/30/08 19:48:23 changed by stassats

  • status changed from assigned to closed.
  • resolution set to fixed.