Hi Alan, On Fri, Nov 14, 2014 at 7:17 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > > The reason for the get/put is to force a call to autosuspend_check(). > But in this case, if killing the interrupt URB causes > autosuspend_check() to run then the get/put isn't needed. > > On the other hand, I don't see why killing the interrupt URB would > cause autosuspend_check() to run. Can you explain that? Sorry for the delay in my response. I did some more checking of my particular failure, and my commit message is incorrect. The usb_kill_urb is actually not the cause of this problem. It does not result in autosuspend_check() itself, and is only serving to add some delay. hidraw_release() in hidraw.c calls drop_ref(), which calls the following in sequence upon clearing the last reader : /* close device for last reader */ hid_hw_power(hidraw->hid, PM_HINT_NORMAL); hid_hw_close(hidraw->hid); hid_hw_power results in a usb_autopm_put_interface. In this case, the reference count is decremented to 0, and a delayed autosuspend request is attempted. hid_hw_close leads to usbhid_close, which clears needs_remote_wakeup. However, there's no guarantee that the clear of needs_remote_wakeup will occur before the delayed work ( runtime_idle() -> autosuspend_check() ) runs. Moving usbhid->intf->needs_remote_wakeup = 0 to before the usb_kill_urb(usbhid->urbin) only serves to reduce the amount of time between these events and makes this particular failure less likely. The correct solution is to put get/put around each change of needs_remote_wakeup, as that will correctly trigger another delayed autosuspend_check(), whose result is affected by the state of needs_remote_wakeup. Since autosuspend_check() occurs as delayed work, I think it is appropriate to add get/put around the clear in usbhid_stop as well. -- Benson Leung Software Engineer, Chrom* OS bleung@xxxxxxxxxxxx On Fri, Nov 14, 2014 at 7:17 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Thu, 13 Nov 2014, Benson Leung wrote: > >> Hi Alan, >> >> On Thu, Nov 13, 2014 at 2:11 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: >> > Wait a minute -- in your previous email you said this approach didn't >> > work. So does it work or doesn't it? >> >> Sorry for the confusion. The approach *does* work. >> >> That was actually my original idea to fix the problem, but I saw other >> places in the kernel where it was done with a get/put. > > The reason for the get/put is to force a call to autosuspend_check(). > But in this case, if killing the interrupt URB causes > autosuspend_check() to run then the get/put isn't needed. > > On the other hand, I don't see why killing the interrupt URB would > cause autosuspend_check() to run. Can you explain that? > > Alan Stern > > > -- Benson Leung Software Engineer, Chrom* OS bleung@xxxxxxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html