On Thu, 4 Mar 2010, Rafael J. Wysocki wrote: > On Thursday 04 March 2010, Alan Stern wrote: > > On Thu, 4 Mar 2010, Rafael J. Wysocki wrote: > > > > > > Very well. Then we still need a solution to the original problem: > > > > Devices sometimes need to be unregistered during resume, but > > > > del_gendisk() blocks on the writeback thread, which is frozen until > > > > after the resume finishes. How do you suggest this be fixed? > > > > > > I thought about thawing the writeback thread earlier in such cases. > > > > > > Would that makes sense / is it doable at all? > > > > My thought exactly. This is the only approach that also solves the > > following race: > > > > A driver is unloaded at the same time as a suspend starts. > > > > The writeback thread gets frozen. > > > > Then before the rmmod thread is frozen, it calls del_gendisk. > > > > Delaying things by means of a workqueue (or the equivalent) might also > > work, but it doesn't seem as safe. For example, some important > > writebacks might end up getting delayed until too late. > > OK, so what exactly should trigger the thawing of the writeback thread? > > Should we do that unconditionally at one point or wait for a specific situation > to happen, and how to recognize that situation in the latter case? My thought was that del_gendisk would do this unconditionally. It would make the task unfreezable and then thaw it, to prevent a race. This assumes that del_gendisk never gets called in the middle of a sleep transition unless the device is really gone. If that's not the case, Pavel's suggestion (make the resume routine do the device removal asynchronously) might be better. Alan Stern _______________________________________________ linux-pm mailing list linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/linux-pm
