Time warp on 2.6.19-rc6-rt4

From: Clark Williams
Date: Mon Nov 20 2006 - 17:19:07 EST


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Ingo/Thomas/John/et al,

I've seen sporadic time warp messages while running my pi_stress program
(http://people.redhat.com/williams/tests/pi_tests-1.7.tar.gz). The
frequency of time warp occurrences has been gong down over time, but I
just got one on a dual Opteron running 2.6.19-rc6-rt4:

- --------------
$ sudo ./pi_stress --signal
Starting PI Stress test
Test threads using scheduler policy: SCHED_FIFO
Admin priority: 99
Reporter priority: 98
10 groups of 3 threads will be created
High priority: 97
Med priority: 96
Low priority: 95
Admin threads running on processor: 0
Test threads running on processor: 1
Press Ctrl-C to stop test
Current Inversions: 1032940
Stoppping test
Total inversion performed: 6836686
- --------------

The Current Inversions line is dynamically updated, but after I noticed
it had stopped updating, I looked a the console and found:

- --------------
BUG: time warp detected!
prev > now, 10277e5c84132156 > 10277e5b6c9e0baf:
= 4688516519 delta, on CPU#0

Call Trace:
[<ffffffff8026e647>] dump_trace+0xaa/0x3f2
[<ffffffff8026e9c9>] show_trace+0x3a/0x58
[<ffffffff8026ec03>] dump_stack+0x15/0x17
[<ffffffff8025cd91>] getnstimeofday+0x142/0x150
[<ffffffff802a2aff>] sys_clock_gettime+0x4d/0x98
[<ffffffff802601cf>] system_call+0xef/0x15c
DWARF2 unwinder stuck at system_call+0xef/0x15c

Leftover inexact backtrace:


skipping trace printing on CPU#0 != -1
hdb: lost interrupt
hdb: lost interrupt
- --------------

The system was still running, but in my test the timer wasn't updating.
This was actually one of Ingo's prepackaged kernel-rt rpms, so I don't
have much debug on.

Anything you want me to try?

Clark
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iD8DBQFFYijXHyuj/+TTEp0RAslPAKCdeI1cbm8hINu0b2aR38LCeEaYZgCdHPAG
Heb3RlEy2UnmZX1rUpBvmN4=
=zDdd
-----END PGP SIGNATURE-----
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/