Messages in this thread | | | Date | Fri, 30 Jan 2009 12:23:11 +0100 | From | Jens Axboe <> | Subject | Re: [PATCH -v2] use per cpu data for single cpu ipi calls |
| |
Hi,
A little late into this discussion, was nasty ill the past days.
Anyway, v2 with the suggested patches looks fine to me. That is, use local cpu storage (and thus lose the lock) and retain the the kmalloc() for the !wait case. That should still scale fine, as far as I can tell. The block layer usage, for which this was originally concocted, doesn't really care that much, since it always provides its own csd storage and thus use the __ variant of the helper.
Peter, can you post a final complete patch for review and acks?
-- Jens Axboe
| |