Hi Alan. On Thu, 2005-12-29 at 12:18, Alan Stern wrote: > On Wed, 28 Dec 2005, Rafael J. Wysocki wrote: > > > I think for a (suspended) device that can be removed, unplugged, undocked, > > etc. (call it "removable") the most natural place in which we can detect > > that the device is no longer accessible is the device driver's .resume() > > routine, at least as far as swsusp is concerned. > > No. The most natural place in which we can detect that a device is no > longer accessible is the place where we already do these detections. Not > in the resume routine. > > > IMO .resume() should check if the device is still there and if not, it should > > put the device on a list of "no-longer-present devices" and just return. > > After the .resume() routines of all devices have completed, the list can > > be processed by something (a kernel thread?) that will do the changes > > on whatever level is necessary. > > Often the device-specific resume routine doesn't have the information > needed for checking whether the device is still there. Such checks are > done by generic bus-oriented routines. > > Yes, the bus's resume handler can do such a check. Why should it bother, > when other parts of the bus code will detect the device removal in the > normal course of events? Or alternatively, why shouldn't the bus's resume > handler simply invoke those other parts of the bus code when it determines > that a device has been removed? I guess because the resume routine is almost certainly the first piece of code that will try to access the hardware (and possibly choke and die it it's not there). Regards, Nigel
