Date: Wed, 21 Jul 2004 12:20:54 +0100 From: Doug Rabson <dfr@nlsystems.com> To: Poul-Henning Kamp <phk@phk.freebsd.dk> Cc: freebsd-arch@freebsd.org Subject: Re: kldunload DIAGNOSTIC idea... Message-ID: <1090408854.7114.8.camel@builder02.qubesoft.com> In-Reply-To: <81662.1090407761@critter.freebsd.dk> References: <81662.1090407761@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 2004-07-21 at 12:02, Poul-Henning Kamp wrote: > In message <1090406270.7114.2.camel@builder02.qubesoft.com>, Doug Rabson writes > : > >On Wed, 2004-07-21 at 10:21, Poul-Henning Kamp wrote: > >> In message <200407211010.08159.dfr@nlsystems.com>, Doug Rabson writes: > >> > >> >The original intention was that drivers use the > >> >device_busy()/device_unbusy() counter to handle these things. In some > >> >cases, just calling device_busy() from fooopen() and device_unbusy() > >> >from fooclose() is sufficient. > >> > >> That is not enough. All methods in cdevsw, and things not in cdevsw > >> (clone handlers, call backs, etc etc) needs to refcount. > >> > >> I have a lot of this working in a tree here, and will commit it once > >> I have gone over it a few more times. > > > >Methods in cdevsw which can't be called unless the device is opened can > >rely on a single counter managed by open/close in most cases. Other > >callbacks may or may not need extra handling depending on whether or not > >the callback can persist past close. > > The problem is that if you are sleeping in for instance a read, you > need to wake up the thread, and you can still only hope that it > will at some point in the future do a close. > > That is why the devsw/cdev code is designed so that you can forego > your close (and do it yourself) by destroying the cdev. You still > need to evict all threads of course, but you don't need to wait > (potentially for ever) for them to come back and call close. > > >Will you use the existing device_busy() counter or will each driver use > >its own counter? > > device_busy doesn't work well because it cannot happen until I am > already inside the driver and because there is no known relationship > between cdevs and device_t's. > > There are three parts to it, a refcount on cdevsw which tells us if > any thread is inside the driver using that route, a refcount on the > individual cdev and a linkage between the two. The device_busy() counter is still simplest (as long as there is a device_t at all). The implementation of devclass_delete_driver() will automatically veto the unload (when its called from driver_module_handler()).
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1090408854.7114.8.camel>