Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

we know the problem is related with allocation of large data structures
on the stack, the trap (SIGBUS) is inside of a kind of the "alloca_probe"
routine. However, we do consider this as the kernel bug, not the
Firefly's internal problem, because the proper signal in the case
of any problems (if any) would be SIGSEGV, not SIGBUS. Actually,
this memory is mapped into Firefly's address space, so the problem
is most likely with OS/hardware internal data structures like page
tables. Another possibility could be that the bad kernel allows
mprotect to be called on unmapped pages and returns no error in this case.

If those of you who like Ubuntu would also like to further investigate
this bug and then file bug report to OS developers, we can provide
a set of small test programs, which should help to elucidate the exact problem.

Regards,
Alex Granovsky



On Thu Sep 2 '10 3:41pm, Reinaldo Pis Diez wrote
------------------------------------------------
>Thanks Davide! Your post confirm that it is in fact a problem related to one of the last updates in Ubuntu, both in 9.10 and 10.04.
>
>
>On Thu Sep 2 '10 3:11pm, Davide Vanossi wrote
>---------------------------------------------
>>I have the same kind of problem after a kernel update of ubuntu 10.04 64 bit. Under the last kernel (2.6.32-24 #42) installed from the official repository FireFly does not work. The error is of the same kind of the one reported by Reinaldo and it affects both serial and parallel runs (under different version of MPI; I currently use the FireFly version 7.1.G with MPICH, MPICH2 and openMPI).
>>Using the kernel 2.6.31-21 (#31) FF works without any problem.
>>Regards,

>>     Davide Vanossi
>>
>>
>>On Thu Sep 2 '10 0:26am, Reinaldo Pis Diez wrote
>>------------------------------------------------
>>>Dear Jim,

>>>On Wed Sep 1 '10 11:52pm, Jim Kress wrote
>>>-----------------------------------------
>>>>Try MPICH2 or OpenMPI version.  MPICH1 is obsolete, not supported and buggy.

>>>I didn't mention but I had the same results using MPICH2. As I didn't install OpenMPI libraries FF complains about the lack of them.
>>>As I have the same behavior with both MPICH1 (static and dynamic) and MPICH2, I believe the problem is not related to FF.

>>>Reinaldo  
>>>
>>>
>>>>Jim

>>>>On Wed Sep 1 '10 10:16pm, Reinaldo Pis Diez wrote
>>>>-------------------------------------------------
>>>>>Dear folks,

>>>>>I use FF, version 7.1.G, build 5618 in my laptop under Ubuntu 9.10. I've installed the MPICH1 version with libraries statically linked. Until last week or so I was running FF without any problem.
>>>>>Then, the automatic package manager facility in Ubuntu upgraded the kernel to 2.6.31-22 and after that I was unable to run FF again.
>>>>>The program starts and stops at different points, depending on the job, and immediately prints on the screen the following:

>>>>>TID 5898 caught signal 7, exiting.                      

>>>>>Dump of registers follows

>>>>>eax :: 0x00048914, edx :: 0xffffa868
>>>>>ecx :: 0xfffe8634, ebx :: 0xffffa648
>>>>>esi :: 0xf5ea2498, edi :: 0xffffa6d4
>>>>>ebp :: 0xffffa690, esp :: 0xffffa62c
>>>>>eip :: 0xf5e84177, eflags :: 0x00210206

>>>>>cs  :: 0x0023
>>>>>ds  :: 0x002b
>>>>>es  :: 0x002b
>>>>>ss  :: 0x002b
>>>>>fs  :: 0x00d7
>>>>>gs  :: 0x0000

>>>>>Stack backtrace

>>>>>esp :: 0xffffa690, ebp :: 0x00000002, eip :: 0xf5e8d5fc

>>>>>Any ideas/hints about what's going on?
>>>>>Regards,

>>>>>Reinaldo
>>>>>

[ This message was edited on Fri Sep 3 '10 at 6:31pm by the author ]