Date: Wed, 21 Oct 2009 10:36:04 +1300 From: Andrew Thompson <thompsa@FreeBSD.org> To: Alexander Motin <mav@FreeBSD.org> Cc: FreeBSD-Current <freebsd-current@freebsd.org>, Scott Long <scottl@FreeBSD.org> Subject: Re: CAM problem Message-ID: <20091020213604.GA63951@citylink.fud.org.nz> In-Reply-To: <4ADD7683.7040907@FreeBSD.org> References: <mailpost.1255999338.6409497.5480.mailing.freebsd.current@FreeBSD.cs.nctu.edu.tw> <4ADD7683.7040907@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 20, 2009 at 11:36:19AM +0300, Alexander Motin wrote: > Andrew Thompson wrote: > > I have a cam problem that is noticeable with usb devices. It relates to > > the ordering of xpt_release_device() and the CAM_DEV_UNCONFIGURED flag > > when yanking a device that has stalled. This then causes a problem with > > the usb explore thread which will end up waiting on simfree forever, > > blocking any further usb attach/detach on the controller. > > > > As I can see, you are returning CAM_TID_INVALID error here. There is no > special error handling for this error, comparing to CAM_SEL_TIMEOUT. If > you return CAM_SEL_TIMEOUT there, device will be killed immediately and > probably workaround this specific problem. Setting the error to CAM_SEL_TIMEOUT does indeed workaround the problem. I have committed this so at least it can be merged to 8.0. > > scsi_dev_async: set dev dev3 unconfigured > > > > ^^^ dev3 gets the CAM_DEV_UNCONFIGURED flag cleared here > > ... but removing configured status does not call deallocation, as > unreferencing does. > > > xpt_bus_deregister: xpt_release_bus > > xpt_release_bus: ref=4 -> 3 > > xpt_release_device dev4 OK > > xpt_release_target: xpt_release_bus > > xpt_release_bus: ref=3 -> 2 > > xpt_release_path: xpt_release_bus > > xpt_release_bus: ref=2 -> 1 > > umass_cam_detach_sim: > > umass-sim0: waiting... ref = 1 > > > > ^^^ wait on "simfree" forever. > > I think correct solution will be to additionally increment reference > counter before clearing CAM_DEV_UNCONFIGURED and decrement it back after > setting CAM_DEV_UNCONFIGURED back. Check for CAM_DEV_UNCONFIGURED inside > xpt_release_device() then could be removed or turned into assertion. I agree, this looks like the best solution. Andrew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20091020213604.GA63951>