Bug 5636 - Totally bogus BogoMips when calibrate_delay_direct fails; results in sluggish system
Summary: Totally bogus BogoMips when calibrate_delay_direct fails; results in sluggish...
Status: CLOSED PATCH_ALREADY_AVAILABLE
Alias: None
Product: Other
Classification: Unclassified
Component: Other (show other bugs)
Hardware: i386 Linux
: P2 low
Assignee: other_other
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2005-11-23 02:24 UTC by mjacek
Modified: 2006-07-13 09:27 UTC (History)
2 users (show)

See Also:
Kernel Version: 2.6.14.2
Subsystem:
Regression: ---
Bisected commit-id:

Attachments
Add an attachment (proposed patch, testcase, etc.)

Description mjacek 2005-11-23 02:24:10 UTC 
Most recent kernel where this bug did not occur: (last tried: 2.6.12)
Distribution: PLD Linux
Hardware Environment: Asus M2400E laptop
Software Environment: kernel initialization, (kernel/calibrate.c)
Problem Description: calibrate_delay_direct fails ("failed to get a good
estimate for loops_per_jiffy"), then the BogoMIPS number goes ten times larger
than should be. I wish I had a laptop with 27000 BogoMIPS......

The lpj=xxxx kernel parameter setting works OK, but:
1. Doesnt it mean some hidden problem in the new routine?
2. It would be much better to fall to the old routine after the failure.
3. There should be the possibility to switch off the new routine in config.

Steps to reproduce:
Probably by running the new kernel on a platforms with "long platform interrupts"
Comment 1 mjacek 2005-12-01 09:30:08 UTC 
Interesting, but with 2.6.14.3 kernel the calibrate_delay() routine does not fail.
Perhaps the main bug was somewhere else, BEFORE the calibration.....
Comment 2 john stultz 2006-02-03 16:39:41 UTC 
Has this issue been seen w/ 2.6.15 or newer kernels?
Comment 3 mjacek 2006-02-04 14:10:14 UTC 
I've never seen it again after move to newer kernels. (today I use 2.6.14.4)

Haven't tried 2.6.15 yet.
Comment 4 Adrian Bunk 2006-07-13 09:27:47 UTC 
Please reopen this bug if you'll observe it with kernel 2.6.17.
Note You need to log in before you can comment on or make changes to this bug.