bcm63xx_udc_stop() clears udc->driver and then calls msleep(100) plus the PHY/clock teardown while still holding udc->lock, which is taken with spin_lock_irqsave(). Sleeping in that atomic context triggers "BUG: scheduling while atomic" (and a CONFIG_DEBUG_ATOMIC_SLEEP splat) when the gadget driver is unbound. The lock is only needed to clear udc->driver, which serializes against the interrupt handler. The settle delay and the bcm63xx_select_phy_mode() / set_clocks() teardown only touch hardware and are already called without the lock during init, so drop the lock right after clearing udc->driver and run the sleep and teardown unlocked. Fixes: 613065e53cb1 ("usb: gadget: bcm63xx UDC driver") Cc: stable@vger.kernel.org Signed-off-by: Hyunjung Ko --- drivers/usb/gadget/udc/bcm63xx_udc.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/drivers/usb/gadget/udc/bcm63xx_udc.c b/drivers/usb/gadget/udc/bcm63xx_udc.c index ba91b1ebbf57..269063d19f10 100644 --- a/drivers/usb/gadget/udc/bcm63xx_udc.c +++ b/drivers/usb/gadget/udc/bcm63xx_udc.c @@ -1847,8 +1847,8 @@ static int bcm63xx_udc_stop(struct usb_gadget *gadget) unsigned long flags; spin_lock_irqsave(&udc->lock, flags); - udc->driver = NULL; + spin_unlock_irqrestore(&udc->lock, flags); /* * If we switch the PHY too abruptly after dropping D+, the host @@ -1861,8 +1861,6 @@ static int bcm63xx_udc_stop(struct usb_gadget *gadget) bcm63xx_select_phy_mode(udc, false); set_clocks(udc, false); - spin_unlock_irqrestore(&udc->lock, flags); - return 0; } -- 2.43.0