Messages in this thread | | | Date | Fri, 16 Jan 2009 10:03:03 -0800 | From | Mike Travis <> | Subject | Re: [PULL}: latest tip/cpus4096 changes |
| |
Ingo Molnar wrote: > * Mike Travis <travis@sgi.com> wrote: > >> commit b758cdbee5da0b8fb7e34a68651e6ccc5310b48a >> Author: Rusty Russell <rusty@rustcorp.com.au> >> Date: Thu Jan 15 16:29:16 2009 -0800 >> >> work_on_cpu: Use our own workqueue. >> >> Impact: remove potential circular lock dependency with generic kevent workqueue >> >> Annoyingly, some places we want to use work_on_cpu are already in >> workqueues. As per Ingo's suggestion, we create a different workqueue >> for work_on_cpu. > > btw., that's a nice fix - were you able to reproduce any of the lockdep > asserts that i got in testing, and did those go away with this patch? > > If yes then that's nice and makes work_on_cpu() a lot more usable IMO. > > If not then that should generally be declared in the pull request: > "beware, different approach than before but might still trigger lockdep > warnings" > > Ingo
Hi,
I've been trying all sorts of overlapping testing and I've not gotten the lockdep warnings any more. (I do occasionally get that debug object warning that I mentioned to Thomas, though that comes and goes irregardless of any patches.) The two fixes that Rusty did (lose the get_online_cpus() and use a separate work queue) seems to have relieved work_on_cpu of it's primary problems, and I've not found any new ones to replace them yet. ;-)
So I'll "un-push" the entire patchset, and then re-push the x86 only ones, and send the others to their respective maintainers after fixing the problems you mentioned. (Posting patches to the list was way more fun, and less "committable". ;-)
Thanks, Mike
| |