From: Matt H. <mat...@us...> - 2006-05-20 08:59:27
|
On Fri, 2006-05-19 at 23:00 -0700, Paul Jackson wrote: > Matt wrote: > > cpuset_exit ... It's supposed to be > > asynchronous, but I don't see how just yet (looks like it waits for the > > return code). > > I don't think it waits for the return code; certainly has > no need to. > > Question -- what possible benefit would there be to calling > cpuset_exit via a notifier, rather than simply calling > a plain old fashioned subroutine, as the code does now? Not much by itself -- shortening the length of the copy_process() and do_exit() functions mostly. I'm trying to see how much of the fork/exit code could be pulled out into this notifier chain. Jes Sorensen had some ideas for things that might be able to use per-task notifier chains in these paths. Process events, delay accounting, and resource groups things I know of that could use a global notifier chain in these paths. I've also been looking into audit, profile, cpusets, and wrapping Jes' functionality with it. The goal is to make the interfaces to the chain(s) as useful to existing and future code as possible. Cheers, -Matt Helsley |