From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; de-DE; rv:1.7.8) Gecko/20050517 Firefox/1.0.4 (Debian package 1.0.4-2)
Description of problem:
Running the MEMORY2 test results in FAILED. Output.log ends like that:
+ echo 'Testing memory read latency (cache-line size detection etc.)'
Testing memory read latency (cache-line size detection etc.)
+ /usr/lib/lmbench/bin/x86_64-linux-gnu/lat_mem_rd 8000 16 32 64 128 256 512 1024
/usr/share/rhr/tests/MEMORY2: line 29: 12246 Segmentation fault $lmbench_dir/lat_mem_rd $MB 16 32 64 128 256 512 1024
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Install RHEL4 Update 2 for x86_64 on machine
2. Install dt-15.14-2.EL4.x86_64.rpm, ltp-20050804-1.EL4.x86_64.rpm,
lmbench-2.0.4-1.EL4.x86_64.rpm and rhr2-2.0-1.EL4.x86_64.rpm
3. Run the MEMORY2 test
Actual Results: MEMORY2 test results with "FAILED".
Expected Results: MEMORY2 test should pass. :-)
I tried this on 2 independent hardware systems:
a) Fujitsu Siemens CELSIUS R630 with 16 GB Memory
b) Fujitsu Siemens CELSIUS M440 with 8 GB Memory
I'll upload attachments with the corresponding output.log and hardware.log files.
Both machines have ECC modules and the BIOS event log shows no errors at all.
On the R630 I was running our own Memory Test utility that we use for RAM module qualification. After 16 hours I got 8 cycles and 0 errors.
From that I have to draw the conclusion that the hardware is ok, but there might be an issue with the new MEMORY2 test.
I tried MEMORY2 as well on a CELSIUS H230 (2 GB RAM, i386 architecture) and there I didn't get any error. So maybe its depending on architecture or size of memory.
Created attachment 123653 [details]
hardware.log for CELSIUS M440
Created attachment 123654 [details]
output.log for CELSIUS M440
Created attachment 123655 [details]
hardware.log for CELSIUS R630
Created attachment 123656 [details]
output.log for CELSIUS R630
I did another test on the CELSIUS M440, this time with only 4 GB of RAM. Then
MEMORY2 works without segfault. Could it be a problem with the 4 GB boundary?
Rainer, with your own Memory Test utility that you ran on the R630, was it run
on the same version of RHEL that the MEMORY2 test failed on?
I'm afraid there is a sort of misunderstanding. Our memory test tool doesn't run
on Linux, its a sort of DOS tool that we boot from CD. Its capable also for
configurations above 4 GB and I guess based on the memtest86 that you find with
some other Linux distributions.
Thanks for the clarification. Our MEMORY2 test checks that Linux can access
every addressable segment of memory. I think we can rule out bad memory given
your report above. Thus, I see two possibilities:
- A bug in the MEMORY2 test
- A bug related to how RHEL and the hardware interact
We will investigate. Any additional insight you may have on the issue would be
Created attachment 123683 [details]
updated lmbench package for RHEL4 - not signed, not for public consumption
This is probably due to a bug in lmbench - on certain arches, attempting to run
lat_mem_rd with >=2048MB RAM causes it to segfault.
I've attached an updated lmbench package above, which should fix this bug. Could
you try upgrading lmbench with this package, and then run MEMORY2 again?
I tried that update of lmbench on my M440 machine with 8 GB. This time no
errors. The R630 is still busy with the old memory test, if its completed I'll
do the test there as well. But so far it looks like the problem is solved.
Thanks for the bug report! Feel free to use the new package for submitting test
results. We will work on issuing an errata for this package.
Hi. I also used the new lmbench RPM and did not see the Seg Fault. It looks like
this is fixed. Thanks.
According to comment comment 7 in Bug 182713, this problem affects at least the
i386 architecture as well.
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.