Re: [Gems-users] SIMICS SEG FAULT problem..

[BYONG WU CHONG <bernard.chong@xxxxxxxx>]

 

Hi, Kisup

 

Running STAMP benchmark on GEMS is quite hard.

 

You should check following.

1.       Did you change all the malloc system calls to user-land malloc calls?

2.       STAMP sometime uses P_MALLOC inside a transaction and it’s better to use all user-land malloc calls for P_MALLOC and TM_MALLOC.

3.       Did your simulation warmed up the system? TLB and Solaris software cache should be warmed up.

4.       Did your benchmark used Solaris memory alignment on malloc calls?

5.       Did you make right processor cabinets and bind them correctly?

6.       Did the benchmark make transactions only after set_transaction_regiter() call?

7.       Are you sure there are no system calls within a transaction? pthread_*(), puts(), printf(), memcpy(), memset() are to watch for.

8.       Did you set cpu_switch_time to 1?

 

If your answer to these questions are all yes, then I you might be dealing with much harder problem.

 

BTW, “SIMICS SEG FAULT” means there is a seg fault reported from the benchmark.

You will probably want to test run Lock version of the binary in 1~16 thread mode and test run TM version in 1 thread mode in the real Solaris-SPARC machine.

If you don’t get error from above test run. Then check again for memory alignment and possibility of two threads accessing a same line without volatile access.

I see that you are running in EETM mode. Try LLTM and see if you still get error.

 

Good luck,

 

- Byong-Wu

 

 

From: gems-users-bounces@xxxxxxxxxxx [mailto:gems-users-bounces@xxxxxxxxxxx] On Behalf Of Kisup, Hong (홍기섭)
Sent: Monday, August 31, 2009 4:20 AM
To: gems-users@xxxxxxxxxxx
Subject: [Gems-users] SIMICS SEG FAULT problem..

 

Hi, all.

 

I'm simulating the STAMP application(the bayes) using Log-TM.

And, I got the following error.

I'm using Simics-3.0.29 and GEMS-2.1. (Protocol is the MESI_CMP_filter_directory)

The number of processors are 4. And g_MEMORY_SIZE_BYTES is 4294967296 (4GB)

And the MAX_LOG_SIZE is 700000.

If anyone could help, I would greatly appreciate it.

 

 

142225122 0 [0,0] Begin ESCAPE ACTION - ESCAPE DEPTH: 1 PC [0x18ac8, line 0x18ac0]
 142225157 0 [0,0] END ESCAPE ACTION - ESCAPE DEPTH: 0 PC [0x18c64, line 0x18c40]
142225671 0 [0,0] ISOLATE XACT STORE [0x4a089a00, line 0x4a089a00] XACT LEVEL: 1 PC = [0x1acb4, line 0x1ac80]
 142225671 0 [0,0] ADD UNDO LOG ENTRY: [0x6dda00, line 0x6dda00] [0x4a089a00, line 0x4a089a00] LogAddress: [0x1bc640, line 0x1bc640] 1
142225698 0 [0,0] ISOLATE XACT LOAD VA: [0x6dd9c0, line 0x6dd9c0] PA: [0x4a0899c0, line 0x4a0899c0] XACT LEVEL: 1 PC = [0x1acfc, line 0x1acc0]
142225703 0 [0,0] ISOLATE XACT LOAD VA: [0x6dd9c0, line 0x6dd9c0] PA: [0x4a0899c0, line 0x4a0899c0] XACT LEVEL: 1 PC = [0x1ad0c, line 0x1ad00]
…

142226415 0 [0,0] ISOLATE XACT STORE [0x4a089a00, line 0x4a089a00] XACT LEVEL: 1 PC = [0x1aca4, line 0x1ac80]
142226418 0 [0,0] ISOLATE XACT LOAD VA: [0x2e480, line 0x2e480] PA: [0x6e1da480, line 0x6e1da480] XACT LEVEL: 1 PC = [0x1ab70, line 0x1ab40]
142226433 0 [0,0] ISOLATE XACT LOAD VA: [0xff1e0080, line 0xff1e0080] PA: [0x4b384080, line 0x4b384080] XACT LEVEL: 1 PC = [0xff0e5168, line 0xff0e5140]
 142226438 0 [0,0] Begin ESCAPE ACTION - ESCAPE DEPTH: 1 PC [0x18ac8, line 0x18ac0]
SIMICS SEG FAULT  for thread 0 of proc 0
SIMICS SEG FAULT
ASSERT FAILURE at PC 18b00
None

 

 

Kisup, Hong.