Date: Mon, 8 Dec 2008 01:20:55 +0100 From: "=?ISO-8859-1?Q?Marius_N=FCnnerich?=" <marius@nuenneri.ch> To: "Poul-Henning Kamp" <phk@phk.freebsd.dk> Cc: freebsd-geom@freebsd.org Subject: Re: Trivial(?) reorganization of topology lock in geom_event Message-ID: <b649e5e0812071620u76306edfv118125457201274@mail.gmail.com> In-Reply-To: <31114.1228597207@critter.freebsd.dk> References: <b649e5e0812041241ja4c3886s526590c98e59f629@mail.gmail.com> <31114.1228597207@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Dec 6, 2008 at 10:00 PM, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote: > In message <b649e5e0812041241ja4c3886s526590c98e59f629@mail.gmail.com>, "=?ISO- > 8859-1?Q?Marius_N=FCnnerich?=" writes: >>Hi, >> >>while working on the DTrace probes for geom I noticed that >>g_topology_lock() is called 20 times per second from the g_event >>thread, even though the thread only runs 10 times per second when >>idle. Maybe it is possible to change the locking like in this patch? I >>also changed the position of one unlocking of g_eventlock. > > In theory the timeout is not necessary, it was added as a stopgap > because there were synchronisation issues long time ago. > > Try dropping the timeout and see if you can provoke problems, > if not, kill it. Did so, please take a look at this patch: http://nuenneri.ch/freebsd/geom_tl2.patch I am running a version of this with the DTrace probes included, I hope the patch is complete. I did a few buildkernels and some of the geom regression tests, so far no problems. I changed the position of the loop to better match how it's like in the up and down threads. What do you think of it? Thanks Marius
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b649e5e0812071620u76306edfv118125457201274>