ueagle-atm: fix PHY signal initialization race
A race exists when initializing ueagle-atm devices where the generic atm device may not yet be created before the driver attempts to initialize it's PHY signal state, which checks whether the atm device has been created or not. This often causes the sysfs 'carrier' attribute to be '1' even though no signal has actually been found. uea_probe usbatm_usb_probe driver->bind (uea_bind) uea_boot kthread_run(uea_kthread) uea_kthread usbatm_atm_init uea_start_reset atm_dev_register UPDATE_ATM_SIGNAL UPDATE_ATM_SIGNAL checks whether the ATM device has been created and if not, will not update the PHY signal state. Because of the race that does not always happen in time, and the PHY signal state remains ATM_PHY_SIG_FOUND even though no signal exists. To fix the race, just create the kthread during initialization, and only after initialization is complete, start the thread that reboots the device and initializes PHY state. [ 3030.490931] uea_probe: calling usbatm_usb_probe [ 3030.490946] ueagle-atm 8-2:1.0: usbatm_usb_probe: trying driver ueagle-atm with vendor=1110, product=9031, ifnum 0 [ 3030.493691] uea_bind: setting usbatm [ 3030.496932] usb 8-2: [ueagle-atm] using iso mode [ 3030.497283] ueagle-atm 8-2:1.0: usbatm_usb_probe: using 3021 byte buffer for rx channel 0xffff880125953508 <kthread already started before usbatm_usb_probe() has returned> [ 3030.497292] usb 8-2: [ueagle-atm] (re)booting started <UPDATE_ATM_SIGNAL checks whether ATM device has been created yet before setting PHY state> [ 3030.497298] uea_start_reset: atm dev (null) <and since it hasn't been created yet PHY state is not set> [ 3030.497306] ueagle-atm 8-2:1.0: usbatm_usb_probe: using 3392 byte buffer for tx channel 0xffff8801259535b8 [ 3030.497374] usbatm_usb_probe: about to init [ 3030.497379] usbatm_usb_probe: calling usbatm_atm_init <atm device finally gets created> [ 3030.497384] usbatm_atm_init: creating atm device! Signed-off-by: Dan Williams <dcbw@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Showing
Please register or sign in to comment