Date: Wed, 11 Jan 2006 09:06:09 -0500 From: John Baldwin <jhb@freebsd.org> To: Scott Long <scottl@samsco.org> Cc: cvs-src@freebsd.org, src-committers@freebsd.org, Scott Long <scottl@freebsd.org>, cvs-all@freebsd.org Subject: Re: cvs commit: src/sys/kern subr_taskqueue.c Message-ID: <200601110906.12293.jhb@freebsd.org> In-Reply-To: <43C50E9B.5050508@samsco.org> References: <200601110037.k0B0bDv4009424@repoman.freebsd.org> <200601110847.08614.jhb@freebsd.org> <43C50E9B.5050508@samsco.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wednesday 11 January 2006 08:56 am, Scott Long wrote: > John Baldwin wrote: > > On Tuesday 10 January 2006 07:37 pm, Scott Long wrote: > >>scottl 2006-01-11 00:37:13 UTC > >> > >> FreeBSD src repository > >> > >> Modified files: > >> sys/kern subr_taskqueue.c > >> Log: > >> The interlock in taskqueue_terminate() is completely wrong for > >> taskqueues that use spinlocks. Remove it for now. > > > > Eh? It's waiting for the wakeup that comes from kthread_exit() after t= he > > thread has exited which is locked via the proc lock. Sleeping on the > > taskqueue itself doesn't buy you anything. (In fact, it might sleep > > forever.) The simplest solution might be to acquire the proc lock a l= ot > > earlier before the taskqueue lock in this function so that you don't ha= ve > > to acquire it while holding the taskqueue lock since that is what gives > > you problems. > > With the code the way it was, kthread_exit() in taskqueue_thread_loop > can wind up blocking on the proc lock while the lock is still held in > taskqueue_terminate. I don't know why this is actually a problem, but > turning on WITNESS to investigate revealed the immediate problem of > trying to grab the proc lock with a spinlock already held. The > interlock is really just a protection against drivers that don't > adequately quiesce themselves, so I removed it for now until we can > figure out something better. The interlock to make sure the thread has terminated before the function=20 returns. This would be important if the taskqueue routine for this thread= =20 was in a kernel module that was being unloaded to avoid having a kernel pag= e=20 fault. The solution for the LOR you saw is probably to just lock the proc= =20 lock earlier. taskqueue_terminate() is not really a critical path operatio= n=20 (if it is that's a bug) so holding the proc lock a bit longer there won't=20 kill. As it is, there isn't a matching wakeup to ever resume the sleeper i= n=20 taskqueue_terminate(), so the calling thread will block indefinitely (unles= s=20 it is using a timeout). =2D-=20 John Baldwin <jhb@FreeBSD.org> =A0<>< =A0http://www.FreeBSD.org/~jhb/ "Power Users Use the Power to Serve" =A0=3D =A0http://www.FreeBSD.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200601110906.12293.jhb>