Messages in this thread |  | | Date | Sat, 22 Mar 2008 19:28:02 -0400 (EDT) | From | Alan Stern <> | Subject | Re: [RFC][PATCH] PM: Introduce new top level suspend and hibernation callbacks (rev. 2) |
| |
On Sat, 22 Mar 2008, Rafael J. Wysocki wrote:
> > Yep. The only thing to watch out for is in device_pm_remove(); it > > would be a disaster if somehow a device was removed while it was being > > prepared/suspended/resumed/completed/whatever. I know that's not > > supposed to happen but there's nothing to prevent it, especially if > > the device in question doesn't have a driver. No doubt you can invent > > a way to allow this to happen safely. > > Well, that's a separate issue that IMO should be addressed in a separate patch. > Something like the one below comes to mind. > > The comment removed by the patch is wrong IMO, because it implies that > device_add() may be called with the device semaphore held and that might > deadlock in bus_attach_device().
Are you talking about this comment?
> - * Since device_pm_add() may be called with a device semaphore held, > - * we must never try to acquire a device semaphore while holding > - * dpm_list_mutex.
It isn't wrong. device_add() may indeed be called with a device semaphore held -- just not the semaphore for the device being added. Quite often it is called with device's parent's semaphore held. The implication is not that we may deadlock in bus_attach_device(); rather it is that the order of acquisition must always be device semaphore first, dev_list_mutex second.
> Thus, I think we can acquire dev->sem > in device_pm_add() and in device_pm_remove().
No, you have missed the entire point. The problem doesn't exist in the current code; it exists only if we switch over to using a single list. Routines like dpm_suspend() won't be able to use list_for_each_entry() to traverse the list because entries may be removed by other threads during the traversal. Even list_for_each_entry_safe() won't work correctly without careful attention to details.
Alan Stern
|  |