Messages in this thread | | | Date | Thu, 20 Apr 2006 16:24:56 -0700 | From | Anil S Keshavamurthy <> | Subject | [(take 2)patch 0/7] Notify page fault call chain |
| |
Currently in the do_page_fault() code path, we call notify_die(DIE_PAGE_FAULT, ...) to notify the page fault. Since notify_die() is highly overloaded, this page fault notification is currently being sent to all the components registered with register_die_notification() which uses the same die_chain to loop for all the registered components which is unnecessary.
In order to optimize the do_page_fault() code path, this critical page fault notification is now moved to different call chain and the test results showed great improvements.
And the kprobes which is interested in this notifications, now registers onto this new call chain only when it need to, i.e Kprobes now registers for page fault notification only when their are an active probes and unregisters from this page fault notification when no probes are active.
I have incorporated all the feedback given by Ananth and Keith and everyone, and thanks for all the review feedback.
Andrew, please apply this patch to your mm tree and I would gladly fix if their are any other issues.
thanks, Anil Keshavamurthy
-- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |