backporting timer fix to dapper

Brian Eaton eaton.lists at gmail.com
Wed Jul 19 03:55:57 BST 2006


Hi folks -

Is there any chance that some fixes from the 2.6.16 kernel tree could
be backported to Dapper?  There is a bug in the 2.6.15 kernel that
causes timer interrupts to go missing on some laptops, leading to
serious performance problems.  Symptoms include horrible bad graphics,
audio and keyboard performance, and high power consumption.  The only
consistent workaround seems to be to set max_cstate to either 1 or 2,
though some people report that switching from the 686 to the 386
kernel helps.

An explanation of the problem may be found here:

http://lkml.org/lkml/2005/12/8/304

Venkatesh's initial series of three patches was later supplemented
with a couple of others.  I've done a little bit of digging, and I
think a total of six patches are actually needed.  These apply cleanly
to the current dapper kernel, and together they fix the problem:

5a07a30c3cc4dc438494d6416ffa74008a2194b3
6eb0a0fd059598ee0d49c6283ce25cccd743e9fc
d25bf7e5fe73b5b6d2246ab0be08ae35d718456b
76b461c21468f41837283b7888d55f1c0671f719
bd6633476922b7b51227f7f704c2546e763ae5ed
0b5c59a1e41636afa77b90d34e8c394d8d929733

There are several open dapper bugs that I think all have the same root
cause, the missing timer interrupts.  I didn't do anything like an
exhaustive search through the bug database, but I suspect all of these
are related:

30557 - definitely this problem
34641 - maybe
44391 - maybe
44912 - probably
50153 - probably
50341 - definitely
53404 - definitely

This seems like a fairly large change to make to the dapper kernel,
but the main fix has been in the kernel.org trees for about six
months, with the last tweak (making the fix apply to AMD CPUs as well
as Intel) in March.  The fix ought to be solid at this point.

Thanks for the work you've put in to Ubuntu.  It's a pleasure to use.

Regards,
Brian



More information about the kernel-team mailing list