On Wed, 09 May 2007 11:12:39 PDT, Jeremy Fitzhardinge said: > It always seems to happen on the last CPU; when I run the kernel as Xen > guest with 4 vcpus, it happens on events/3. Probably something different - at least once, I've caught it on /0 with 2 real cores online (T7200). Looking at kernel/softirq.c, the ksoftirqd() function is basically 2 nested while() loops. Looks like either local_softirq_pending() gets stuck true, but no obvious reason why, or something odd is getting set in the current-> structure so when we call schedule(), we end up returning rather than hitting the idle loop. Off to go see how much mm-bisect I can do before dinnertime - the rest will be later tonight...
Attachment:
pgpT3Y4nm4n3V.pgp
Description: PGP signature
- References:
- 2.6.21-mm2
- From: Andrew Morton <[email protected]>
- Re: 2.6.21-mm2 - 100% CPU on ksoftirqd/1
- From: [email protected]
- Re: 2.6.21-mm2 - 100% CPU on ksoftirqd/1
- From: Jeremy Fitzhardinge <[email protected]>
- 2.6.21-mm2
- Prev by Date: Re: Section mismatch warnings (was Re: [PATCH] early_pfn_to_nid needs to be __meminit)
- Next by Date: Re: [PATCH] doc: volatile considered evil
- Previous by thread: Re: 2.6.21-mm2 - 100% CPU on ksoftirqd/1
- Next by thread: Re: 2.6.21-mm2 - 100% CPU on ksoftirqd/1
- Index(es):
 |