• Andy Shevchenko's avatar
    mfd: dln2: Run event handler loop under spinlock · 59547851
    Andy Shevchenko authored
    [ Upstream commit 3d858942 ]
    
    The event handler loop must be run with interrupts disabled.
    Otherwise we will have a warning:
    
    [ 1970.785649] irq 31 handler lineevent_irq_handler+0x0/0x20 enabled interrupts
    [ 1970.792739] WARNING: CPU: 0 PID: 0 at kernel/irq/handle.c:159 __handle_irq_event_percpu+0x162/0x170
    [ 1970.860732] RIP: 0010:__handle_irq_event_percpu+0x162/0x170
    ...
    [ 1970.946994] Call Trace:
    [ 1970.949446]  <IRQ>
    [ 1970.951471]  handle_irq_event_percpu+0x2c/0x80
    [ 1970.955921]  handle_irq_event+0x23/0x43
    [ 1970.959766]  handle_simple_irq+0x57/0x70
    [ 1970.963695]  generic_handle_irq+0x42/0x50
    [ 1970.967717]  dln2_rx+0xc1/0x210 [dln2]
    [ 1970.971479]  ? usb_hcd_unmap_urb_for_dma+0xa6/0x1c0
    [ 1970.976362]  __usb_hcd_giveback_urb+0x77/0xe0
    [ 1970.980727]  usb_giveback_urb_bh+0x8e/0xe0
    [ 1970.984837]  tasklet_action_common.isra.0+0x4a/0xe0
    ...
    
    Recently xHCI driver switched to tasklets in the commit 36dc0165
    ("usb: host: xhci: Support running urb giveback in tasklet context").
    
    The handle_irq_event_* functions are expected to be called with interrupts
    disabled and they rightfully complain here because we run in tasklet context
    with interrupts enabled.
    
    Use a event spinlock to protect event handler from being interrupted.
    
    Note, that there are only two users of this GPIO and ADC drivers and both of
    them are using generic_handle_irq() which makes above happen.
    
    Fixes: 338a1281 ("mfd: Add support for Diolan DLN-2 devices")
    Signed-off-by: default avatarAndy Shevchenko <andriy.shevchenko@linux.intel.com>
    Signed-off-by: default avatarLee Jones <lee.jones@linaro.org>
    Signed-off-by: default avatarSasha Levin <sashal@kernel.org>
    59547851
dln2.c 18.4 KB