Comment 97 for bug 183461

Created an attachment (id=415179)
tasklet debug patch

(In reply to comment #54)
> ** novelty ** The first time for the ksoftirqd 100% cpu usage problem several
> stack dumps were taken (by Alt-PrnScr-L) to let you see in which execution
> state the CPU was. So just have a look at this.

The 2.6.37 traces are useless. The 2.6.34.7 ones are helpful though. Also /proc/softirq clearly shows that some kind of shit schedules a tasklet way too often.

I'm attaching a patch to track that down. Also I'm building a kernel to test and it will appear at:
http://labs.suse.cz/jslaby/bug-465039

Watch for tasklet_action in the logs when this happens. Maybe there will be false positives. Then I'll increase the limit. Let's see.