[4.12 REGRESSION] pinctrl: rockchip: sleeping function called from atomic context

[Brian Norris]

Hi,

Looks like we've added a mutex in the ->bus_lock() callback for
Rockchip's pinctrl irqchip, which triggers a CONFIG_DEBUG_ATOMIC_SLEEP
warning when entering system suspend:

[  151.406483] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:238
[  151.420321] in_atomic(): 0, irqs_disabled(): 0, pid: 2447, name: cat
[  151.427457] CPU: 2 PID: 2447 Comm: cat Tainted: G        W     4.12.0-rc1+
[  151.435922] Hardware name: Google Kevin (DT)
[  151.440687] Call trace:
[  151.443440] [<ffffff900808b2fc>] dump_backtrace+0x0/0x2b8
[  151.449485] [<ffffff900808b5d4>] show_stack+0x20/0x28
[  151.455137] [<ffffff900848f738>] dump_stack+0xa4/0xcc
[  151.460797] [<ffffff90080e3118>] ___might_sleep+0x16c/0x188
[  151.467036] [<ffffff90080e31f4>] __might_sleep+0xc0/0xd4
[  151.472979] [<ffffff9008a70ca0>] mutex_lock+0x2c/0x68
[  151.478637] [<ffffff90084e806c>] rockchip_irq_bus_lock+0x48/0x54
[  151.485364] [<ffffff900811e3ac>] __irq_get_desc_lock+0x9c/0xc8
[  151.491895] [<ffffff900811fa3c>] irq_set_irq_wake+0x44/0x178
[  151.498235] [<ffffff900860d190>] dev_pm_arm_wake_irq+0x78/0x84
[  151.504769] [<ffffff90086137bc>] device_wakeup_arm_wake_irqs+0x48/0x70
[  151.512077] [<ffffff90086109c8>] dpm_suspend_noirq+0x180/0x51c
[  151.518608] [<ffffff90081171a4>] suspend_devices_and_enter+0x1dc/0xde4
[  151.525914] [<ffffff90081187a4>] pm_suspend+0x9f8/0xa38
[...]

The warning goes away if I revert commit 88bb94216f59 ("pinctrl:
rockchip: avoid hardirq-unsafe functions in irq_chip").

The thing is, the documentation (and apparent design) suggest that
calling sleeping functions from ->irq_bus_lock() is perfectly valid. I'm
not 100% following the ___might_sleep() logic, but is this complaining
because of the RCU read locking in device_wakeup_arm_wake_irqs()? I have
CONFIG_PREEMPT_RCU and CONFIG_PREEMPT enabled, FWIW.

Brian