SIGSEGV on isprime

[Ján Jančár <johny@neuromancer.sk>]

On 10/07/18 23:59, Bill Allombert wrote:
> On Tue, Jul 10, 2018 at 06:35:40PM +0200, Ján Jančár wrote:
>> Hi all,
>>
>> While running pari on some grid computing machines I keep encountering
>> this mysterious error, which only happens when working on certain
>> machines and not on others.
>>
>> To reproduce, I compiled the following:
>>
>> on:
>>     Linux 3.16.0-5-amd64 #1 SMP Debian
>>     3.16.51-3+deb8u1+zs2 (2018-01-12) x86_64 GNU/Linux
>>
>> where it run just fine.
>> However when run(and compiled) on:
>>     Linux 4.9.0-6-amd64 #1 SMP Debian
>>     4.9.82-1+deb9u3 (2018-03-02) x86_64 GNU/Linux
>>
>> It SIGSEGVs in isprime():
>>
>>> Program received signal SIGSEGV, Segmentation fault.
>>> 0x00002aaaab1f771c in red_montgomery (T=0x2aaae879e6a0, N=0x2aaae879efe0, inv=2796584439883844767) at ../src/kernel/gmp/mp.c:1013
>>> 1013	    while (Td < (GEN)av) { t = subllx(*++Td, *++Nd); *Td = t; }
>>> (gdb) bt
>>> #0  0x00002aaaab1f771c in red_montgomery (T=0x2aaae879e6a0, N=0x2aaae879efe0, inv=2796584439883844767) at ../src/kernel/gmp/mp.c:1013
>>> #1  0x00002aaaab33a0ac in _sqr_montred (E=0x2aaae879ef38, x=0x2aaae879e6d0) at ../src/basemath/arith1.c:3311
>>> #2  0x00002aaaab33a138 in _mul2_montred (E=0x2aaae879ef38, x=0x2aaae879e6d0) at ../src/basemath/arith1.c:3326
>>> #3  0x00002aaaab39d641 in gen_pow_fold_i (x=0x2aaae879eef0, N=0x2aaae879efa0, E=0x2aaae879ef38, sqr=0x2aaaab33a068 <_sqr_montred>, msqr=0x2aaaab33a10d <_mul2_montred>)
>>>     at ../src/basemath/bb_group.c:254
>>> #4  0x00002aaaab33abf3 in Fp_pow (A=0x2aaaabaa3998 <readonly_constants+56>, K=0x2aaae879efa0, N=0x2aaae879efe0) at ../src/basemath/arith1.c:3508
>>> #5  0x00002aaaab5ca498 in bad_for_base (S=0x7fffffffe210, a=0x2aaaabaa3998 <readonly_constants+56>) at ../src/basemath/prime.c:95
>>> #6  0x00002aaaab5cbaf2 in BPSW_psp (N=0x2aaae879efe0) at ../src/basemath/prime.c:570
>>> #7  0x00002aaaab5cca58 in isprime (x=0x2aaae879efe0) at ../src/basemath/prime.c:846
>>> #8  0x0000000000400788 in main ()
>>> (gdb) info locals
>>> __value = 8722076062158158581
>>> __arg1 = 144115188075855878
>>> __arg2 = 10180160215563133591
>>> __temp = 18446744073709551615
>>> av = 46913531930272
>>> Te = 0x2aaae867e688
>>> Td = 0x2aaae867e6a0
>>> Ne = 0x2aaae867efe8
>>> Nd = 0x2aaae867f000
>>> scratch = 0x2aaae867e680
>>> i = 2
>>> j = 2
>>> m = 9020403664262637533
>>> t = 8722076062158158581
>>> d = 4
>>> k = 2
>>> carry = 1
>>> hiremainder = 4978068440930021014
>>> overflow = 1
>>> (gdb) info args
>>> T = 0x2aaae867e6a0
>>> N = 0x2aaae867efe0
>>> inv = 2796584439883844767
>>> (gdb) quit
>>
>> I compiled pari 2.9.5 / 2.10.1 / current git master, with
>> ./Configure --enable-tls -g
>> and the error happens in all of the versions.
>>
>> Any ideas on what might be causing this? ldd of the binary on both
>> machines shows the same libraries are used, so it is very mysterious to
>> me that it works on one and not on the other.
> 
> Why are you using --enable-tls ? Does it makes a difference ?
> Are you using the same compiler ? The same processor ?
> This code has not changed between 2.9.5 and 2.10.1, however it is rather
> messy, so maybe it is not compiled correctly.
> You can also try
> ./Configure --kernel=none 
> 

The application I am compiling requires tls, I will test it with the
reproducer without tls but even if it works, it will be unusable for me.
I am indeed using the same compiler, the only obvious difference I can
think of is the processor.

I have now tested:

--enable-tls --kernel=none      --> works
--enable-tls --kernel=none-gmp  --> SEGVs!
--enalbe-tls --kernel=auto-none --> works

Not sure what to make of this, I will try with a newer version of GMP,
6.0.0 is currently installed(on both machines, where it works and
doesn't) but 6.1.2 is available, so I will see if that fixes it.


Cheers,
-- 
Jan
______________________________________________________
   /\  # PGP: 362056ADA8F2F4E421565EF87F4A448FE68F329D
  /__\  # https://neuromancer.sk
 /\  /\  # Eastern Seaboard Phishing Authority
/__\/__\  #

Attachment: signature.asc
Description: OpenPGP digital signature