• Iago Abal's avatar
    usb: gadget: pch_udc: reorder spin_[un]lock to avoid deadlock · 1d23d16a
    Iago Abal authored
    The above commit reordered spin_lock/unlock and now `&dev->lock' is acquired
    (rather than released) before calling `dev->driver->disconnect',
    `dev->driver->setup', `dev->driver->suspend', `usb_gadget_giveback_request', and
    `usb_gadget_udc_reset'.
    
    But this *may* not be the right way to fix the problem pointed by d3cb25a1.
    
    Note that the other usb/gadget/udc drivers do release the lock before calling
    these functions. There are also inconsistencies within pch_udc.c, where
    `dev->driver->disconnect' is called while holding `&dev->lock' in lines 613 and
    1184, but not in line 2739.
    
    Finally, commit d3cb25a1 may have introduced several potential deadlocks.
    
    For instance, EBA (https://github.com/models-team/eba) reports:
    
        Double lock in drivers/usb/gadget/udc/pch_udc.c
        first at 2791: spin_lock(& dev->lock); [pch_udc_isr]
        second at 2694: spin_lock(& dev->lock); [pch_udc_svc_cfg_interrupt]
            after calling from 2793: pch_udc_dev_isr(dev, dev_intr);
            after calling from 2724: pch_udc_svc_cfg_interrupt(dev);
    
    Similarly, other potential deadlocks are 2791 -> 2793 -> 2721 -> 2657; and
    2791 -> 2793 -> 2711 -> 2573 -> 1499 -> 1480.
    
    Fixes: d3cb25a1 ("usb: gadget: udc: fix spin_lock in pch_udc")
    Signed-off-by: default avatarIago Abal <mail@iagoabal.eu>
    Signed-off-by: default avatarFelipe Balbi <felipe.balbi@linux.intel.com>
    1d23d16a
pch_udc.c 86.7 KB