On Mon, Aug 17, 2015 at 3:13 PM, Mika Westerberg
[off-list ref] wrote:

When running -rt kernel and an interrupt happens on a GPIO line controlled by
Intel Cherryview/Braswell pinctrl driver we get:

 BUG: sleeping function called from invalid context at kernel/locking/rtmutex.c:917
 in_atomic(): 1, irqs_disabled(): 1, pid: 0, name: swapper/0
 Preemption disabled at:[<ffffffff81092e9f>] cpu_startup_entry+0x17f/0x480

 CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.1.5-rt5 #16
  ...
 Call Trace:
  <IRQ>  [<ffffffff816283c6>] dump_stack+0x4a/0x61
  [<ffffffff81077e17>] ___might_sleep+0xe7/0x170
  [<ffffffff8162d6cf>] rt_spin_lock+0x1f/0x50
  [<ffffffff812e52ed>] chv_gpio_irq_ack+0x3d/0xa0
  [<ffffffff810a72f5>] handle_edge_irq+0x75/0x180
  [<ffffffff810a3457>] generic_handle_irq+0x27/0x40
  [<ffffffff812e57de>] chv_gpio_irq_handler+0x7e/0x110
  [<ffffffff810050aa>] handle_irq+0xaa/0x190
  ...

This is because desc->lock is raw_spinlock and is held when chv_gpio_irq_ack()
is called by the genirq core. chv_gpio_irq_ack() in turn takes pctrl->lock
which in -rt is an rt-mutex causing might_sleep() rightfully to complain about
sleeping function called from invalid context.

In order to keep -rt happy but at the same time make sure that register
accesses get serialized, convert the driver to use raw_spinlock instead.

Suggested-by: Linus Walleij <redacted>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>

Patch applied. Again splendid job here.

Yours,
Linus Walleij