Date: Fri, 13 Jul 2007 19:08:29 +0900 From: Pyun YongHyeon <pyunyh@gmail.com> To: "Li-Lun Wang (Leland Wang)" <llwang@infor.org> Cc: freebsd-current@freebsd.org Subject: Re: threadlock and msk watchdog timeout Message-ID: <20070713100829.GC17801@cdnetworks.co.kr> In-Reply-To: <20070713084325.GA47351@Athena.infor.org> References: <20070713084325.GA47351@Athena.infor.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jul 13, 2007 at 04:43:25PM +0800, Li-Lun Wang (Leland Wang) wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Hi, > > After making world a couple of days ago, my msk(4) became very > unstable. Under moderate network load, the interface hung and I > received > > kernel: msk0: watchdog timeout (missed Tx interrupts) -- recovering > > at least once every several minutes and > > kernel: msk0: Rx FIFO overrun! > > occasionally. > > It was so annoying that I took the trouble of binary searching the > kernel version to find the one destabilized my msk(4). > > The outcome of the search turned out te be strange. Instead of > finding a date after which msk(4) became so very unstable, it *seemed* > that the older the kernel version the stabler msk(4) I got, and the > newer the kernel version the easier and more often msk(4) hung. > > I managed to pin down that with the kernel as of 2007.06.04.12.00.00, > it seemed not to give me any msk watchdog timeout at all, and that > with the kernel as of 2007.06.05.12.00.00, msk(4) began to hang and > the watchdog began to timeout once in a while. There may be a latter > commit that made my msk(4) even more unstable, but I am not sure about > this part as it is not easy to measure the level of "unstableness" of > the network. > > It seems that the most significant commit between 2007.06.04.12.00.00 > and 2007.06.05.12.00.00 was threadlock by jeff@. I don't know why or > how it would affect msk(4), though. I was using SCHED_SMP on a C2D, > but switched back to SCHED_ULE when I did the search. > > I discovered a couple other funny phenomena during the search that may > also suggest this be related to threadlock. One is that msk(4) seemed > to hang less frequently when the system was busy building world or > kernel. The other thing is that I seemed to be able to help unhang > the interface by switching the input focus in X Window by move my > mouse cursor to another window. > > My result might not be accurate, though, as I only rebuilt the kernel, > not the whole world, when I did the search. > Does msk(4) use shared interrupt? Show me the output of "vmstat -i". -- Regards, Pyun YongHyeon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070713100829.GC17801>