Date: Sat, 3 Aug 2013 21:51:22 -0400 From: Mark Johnston <markj@freebsd.org> To: Andriy Gapon <avg@FreeBSD.org> Cc: freebsd-dtrace@FreeBSD.org Subject: Re: [RFC] reworking FreeBSD's SDT implementation Message-ID: <20130804015122.GA3259@raichu> In-Reply-To: <51F372C8.1000107@FreeBSD.org> References: <20130703041023.GA82673@raichu> <20130711024500.GA67976@raichu> <20130711210215.GB7506@gmail.com> <20130713234200.GA40803@raichu> <20130714075634.GC2832@gmail.com> <20130722022811.GA14288@raichu> <51F14150.7000509@FreeBSD.org> <20130727022656.GB67227@raichu> <51F372C8.1000107@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Jul 27, 2013 at 10:12:08AM +0300, Andriy Gapon wrote: > on 27/07/2013 05:26 Mark Johnston said the following: > > On Thu, Jul 25, 2013 at 06:16:32PM +0300, Andriy Gapon wrote: > >> on 22/07/2013 05:28 markj@freebsd.org said the following: > >>> http://people.freebsd.org/~markj/patches/sdt-module-info/20130721-sdt-module-info.diff > >> > >> Mark, > >> > >> this is a minor suggestion only partially related to your patch. > >> I think that it would be nice if module loading and unloading events were posted > >> via EVENTHANDLER(9) mechanism. Then instead of introducing yet more DTrace > >> related hooks in the kernel code, DTrace modules could just subscribe to those > >> events. Also, those events could be potentially useful to other consumers > >> beyond DTrace. > >> What do you think? > > > > Hm, now that I look at this, I'm not sure if it can work. The unload > > hooks need to be able to veto a module unload in the case that one of > > its probes is enabled. This is done by checking whether lf->nenabled > 0, > > and it needs to be done with the dtrace lock held to prevent races. > > > > I've done this by having the unload hooks return a non-zero value if > > there are probes enabled, but EVENTHANDLER(9) doesn't give me a way to > > look at a handler's return value. Do you see a way to get around this? > > Hmm, I didn't think about this problem in advance... > Having looked around I think that it should be possible to handle this situation > in a way similar to watchdog_list. watchdog(9) documents how that works. Of > course, all handlers will have to be careful to not override error if it's > already set. I implemented the module load/unload events a couple of days ago using EVENTHANDLER(9), and it looks like everything works properly. I copied the trick used for watchdog_list and just added an error pointer argument to the unload handlers. The revised patch is here: http://people.freebsd.org/~markj/patches/sdt-module-info/20130803-sdt-module-info.diff Thanks! -Mark
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130804015122.GA3259>