Bug 6938 - use of cpufreq (with any cpufreqd-like deamon) has strange influence on the acpi functions of an IBM Thinkpad X21
Summary: use of cpufreq (with any cpufreqd-like deamon) has strange influence on the a...
Status: CLOSED PATCH_ALREADY_AVAILABLE
Alias: None
Product: ACPI
Classification: Unclassified
Component: Other (show other bugs)
Hardware: i386 Linux
: P2 blocking
Assignee: acpi_other
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2006-08-01 07:25 UTC by Tom Herrmann
Modified: 2007-04-20 13:29 UTC (History)
1 user (show)

See Also:
Kernel Version: 2.6.16-rc2 and all newer (2.6.19-rc2 is most recent now)
Subsystem:
Regression: ---
Bisected commit-id:


Attachments
syslog with acpi exeption (and maybe more helpful information) (14.69 KB, text/plain)
2006-10-29 09:47 UTC, Tom Herrmann
Details

Description Tom Herrmann 2006-08-01 07:25:28 UTC
Most recent kernel where this bug did not occur: 2.6.15.X

Distribution: Debian (testing with most recent kernel-package from unstable for
compiling 2.6.18)

Hardware Environment: IBM Thinkpad X21 (Intel Mobile Pentium III [LV], Intel
440ZXM, all but HDD stock components, most recent bios/embeded controler firmware)

Software Environment: normal desktop distribution stuff, cpufreqd (2.1.1) (all
other cpufreq-deamons  dosn't "fix" it so there is no bug)

Problem Description: I normaly use cpufreqd to switch the frequency from this
processor in a "ondemand" way, never had problems with older kernels then 2.6.16.
Because of some strange behavior I had to compile the cpufreq-kernel-option with
debug options, because the relaxed speedstep checks wouldn't work without it,
and are needed for this first generation speedstep.
Then some users reported this "bug" (needed debug option) and someone "fixed"
this in a 2.6.16-rc.
With this and all more recent kernel-versions the error occours:
Booting and the first 2 to 4 minutes of doing something works without a problem,
but after this time (and some freqency switching) I can't use all "special"
functions (as I think) provided by IBM's embeded controler.
switching the volume with the "hardware" buttons gets very slow (normal
switch/respond-time after pressing is <1sek, then it >5sek) and the FN-F*
funktions don't work at all (FN-F7 for Monitor swich, ...)
If I simply don't switch the frequency (disable the deamon) and reboot, the bug
is gone but this is now solution, because the causes mutch more powercounsuming

I have tested any possible cpufreq kernel options combination, but if switching
works, this bug is there...

Steps to reproduce:
compile kernel with any working cpufreq option combination, boot it, switch the
frequency some times, get this bug...
Comment 1 Tom Herrmann 2006-08-20 04:58:06 UTC
switching from
Preemption Model (Preemptible Kernel (Low-Latency Desktop))  --->
      Voluntary Kernel Preemption (Desktop)

to

Preemption Model (Preemptible Kernel (Low-Latency Desktop))  --->
      Preemptible Kernel (Low-Latency Desktop)

causes the same bug with 2.6.15.x kernels...

and the bug bugs the whole acpi, even the power_off (last init.d cmd) needs
mutch more time to to its job (on all kernels)
Comment 2 Tom Herrmann 2006-10-23 05:52:07 UTC
This bug is still present in latest kenel versions.

Because of security and feature lack while using the old 2.6.15.7 kernel I am
going to think about not using linux anymore.

Disable speedstep realy is no option on a (sub-)notebook

It would be greate if someone could even try to fix this bug.
Comment 3 Tom Herrmann 2006-10-29 09:47:14 UTC
Created attachment 9375 [details]
syslog with acpi exeption (and maybe more helpful information)
Comment 4 Tom Herrmann 2006-10-29 09:48:43 UTC
I may have found some helpful ACPI output:
this is from my syslog and the timestamp matches the point where acpi stoped
working correctly. I'll attach the whole file:

Oct 29 18:40:17 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Oct 29 18:40:17 localhost kernel: ibm_acpi: acpi_evalf(TMP0, d, ...) failed: 19
Oct 29 18:40:18 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Oct 29 18:40:18 localhost kernel: ACPI Error (psparse-0537): Method
parse/execution failed [\_SB_.PCI0.ISA_.EC__.AC__._PSR] (Node c136c540), AE_TIME
Oct 29 18:40:18 localhost kernel: ACPI Exception (acpi_ac-0096): AE_TIME, Error
reading AC Adapter state [20060707]
Oct 29 18:40:18 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Oct 29 18:40:18 localhost kernel: ibm_acpi: acpi_evalf(TMP0, d, ...) failed: 19

I hope this helps and someone cares
Comment 5 Venkatesh Pallipadi 2006-10-30 09:29:58 UTC
Moved to ACPI bucket.
Comment 6 Tom Herrmann 2006-11-23 14:46:06 UTC
The bug disappears with the 2.6.18.3 kernel, but is still present in 2.6.19-rc6
(so it is not really fixed now).

With 2.6.18.3 all buttons and acpi functions work, but it still throws a lot of
errors to syslog (even with no debug/verbose option set):

Nov 23 20:45:08 localhost kernel: ACPI: read EC, OB not full
Nov 23 20:45:08 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Nov 23 20:45:08 localhost kernel: ibm_acpi: acpi_evalf(TMP7, d, ...) failed: 19
Nov 23 21:15:48 localhost kernel: ACPI: read EC, OB not full
Nov 23 21:15:48 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Nov 23 21:15:48 localhost kernel: ACPI Error (psparse-0537): Method
parse/execution failed [\_SB_.LID_._LID] (Node c13765e0), AE_TIME
Nov 23 22:53:20 localhost kernel: ACPI: read EC, OB not full
Nov 23 22:53:20 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Nov 23 22:53:20 localhost kernel: ACPI Error (psparse-0537): Method
parse/execution failed [\_SB_.LID_._LID] (Node c13765e0), AE_TIME
Nov 23 23:40:14 localhost kernel: ACPI: read EC, IB not empty
Nov 23 23:40:14 localhost kernel: ACPI: read EC, IB not empty
Nov 23 23:40:14 localhost kernel: ACPI: read EC, OB not full
Nov 23 23:40:14 localhost kernel: ACPI Exception (evregion-0424): AE_TIME,
Returned by Handler for [EmbeddedControl] [20060707]
Nov 23 23:40:14 localhost kernel: ACPI Error (psparse-0537): Method
parse/execution failed [\_SB_.PCI0.ISA_.EC__.AC__._PSR] (Node c136c540), AE_TIME
Nov 23 23:40:14 localhost kernel: ACPI Exception (acpi_ac-0096): AE_TIME, Error
reading AC Adapter state [20060707]
Comment 7 Tom Herrmann 2007-04-20 13:29:22 UTC
With Kernel 2.6.20.X this bug is disappeared, hopefully for ever, as I change my
notebook to a newer one, I cannot check newer kernel releases

Note You need to log in before you can comment on or make changes to this bug.