deadlock when CONFIG_KEYBOARD_GPIO is enabled with debug on

[Anders Roxell]

When building kernel: git://git.linaro.org/kernel/linux-linaro-lng.git from
commit 0dc6903b3b36381d95ecdef9a7f1ba2e2146d36d
see .config file: http://people.linaro.org/~anders.roxell/job68289_kernel_config_arndale
I runned into a deadlock see link for more information http://validation.linaro.org/scheduler/job/68289/log_file#L_27_340
here is a snippet from the log:

 [ 3.561043] ======================================================
 [ 3.561046] [ INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected ]
 [ 3.561050] 3.10.6-linaro-arndale #1 Not tainted
 [ 3.561052] ------------------------------------------------------
 [ 3.561056] swapper/0/1 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
 [ 3.561072] (&(&(&bank->slock)->lock)->wait_lock){+.+...}, at: [<804728f9>] rt_spin_lock_slowlock+0x35/0x214
 [ 3.561074]
 [ 3.561074] and this task is already holding:
 [ 3.561089] (&irq_desc_lock_class){-.....}, at: [<80082491>] __setup_irq+0x69/0x354
 [ 3.561090] which would create a new lock dependency:
 [ 3.561101] (&irq_desc_lock_class){-.....} -> (&(&(&bank->slock)->lock)->wait_lock){+.+...}
 [ 3.561103]
 [ 3.561103] but this new dependency connects a HARDIRQ-irq-safe lock:
 [ 3.561110] (&irq_desc_lock_class){-.....}
 [ 3.561110] ... which became HARDIRQ-irq-safe at:
 [ 3.561119] [<80067d89>] __lock_acquire+0x955/0x193c
 [ 3.561125] [<800691db>] lock_acquire+0x7f/0x104
 [ 3.561131] [<80473383>] _raw_spin_lock+0x2b/0x38
 [ 3.561137] [<80083695>] handle_fasteoi_irq+0x15/0xd0
 [ 3.561143] [<80080ed9>] generic_handle_irq+0x21/0x2c
 [ 3.561150] [<80018ea5>] handle_IRQ+0x35/0x70
 [ 3.561155] [<800084a3>] gic_handle_irq+0x33/0x58
 [ 3.561160] [<8001803f>] __irq_svc+0x3f/0x9c
 [ 3.561166] [<8001913a>] arch_cpu_idle+0x26/0x30
 [ 3.561172] [<8005b8bb>] cpu_startup_entry+0x16f/0x200
 [ 3.561178] [<8069c97d>] start_kernel+0x349/0x354
 [ 3.561183] [<4000808d>] 0x4000808d
 [ 3.561185]
 [ 3.561185] to a HARDIRQ-irq-unsafe lock:
 [ 3.561192] (&(&(&bank->slock)->lock)->wait_lock){+.+...}
 [ 3.561192] ... which became HARDIRQ-irq-unsafe at:
 [ 3.561200] ... [<8006790d>] __lock_acquire+0x4d9/0x193c
 [ 3.561206] [<800691db>] lock_acquire+0x7f/0x104
 [ 3.561210] [<80473383>] _raw_spin_lock+0x2b/0x38
 [ 3.561215] [<804728f9>] rt_spin_lock_slowlock+0x35/0x214
 [ 3.561219] [<80472efd>] rt_spin_lock+0x29/0x40
 [ 3.561227] [<8029d259>] samsung_pinconf_rw+0x79/0xd8
 [ 3.561232] [<8029d321>] samsung_pinconf_group_set+0x41/0x54
 [ 3.561237] [<8029c9ab>] pinconf_apply_setting+0x4f/0x118
 [ 3.561245] [<8029a4af>] pinctrl_select_state+0x8b/0x100
 [ 3.561252] [<802dd17f>] pinctrl_bind_pins+0x53/0x8c
 [ 3.561258] [<802cbb2f>] driver_probe_device+0x4b/0x198
 [ 3.561263] [<802cbcf7>] __driver_attach+0x53/0x54
 [ 3.561268] [<802ca837>] bus_for_each_dev+0x33/0x54
 [ 3.561273] [<802cb567>] bus_add_driver+0x127/0x190
 [ 3.561278] [<802cc10b>] driver_register+0x43/0xd0
 [ 3.561283] [<8000862d>] do_one_initcall+0xd1/0x114
 [ 3.561288] [<8069ca93>] kernel_init_freeable+0x10b/0x180
 [ 3.561295] [<804676e5>] kernel_init+0x19/0x110
 [ 3.561299] [<8001866d>] ret_from_fork+0x11/0x20
 [ 3.561301]
 [ 3.561301] other info that might help us debug this:
 [ 3.561301]
 [ 3.561303] Possible interrupt unsafe locking scenario:
 [ 3.561303]
 [ 3.561305] CPU0 CPU1
 [ 3.561306] ---- ----
 [ 3.561310] lock(&(&(&bank->slock)->lock)->wait_lock);
 [ 3.561312] local_irq_disable();
 [ 3.561316] lock(&irq_desc_lock_class);
 [ 3.561320] lock(&(&(&bank->slock)->lock)->wait_lock);
 [ 3.561322] <Interrupt>
 [ 3.561326] lock(&irq_desc_lock_class);
 [ 3.561327]
 [ 3.561327] *** DEADLOCK ***

when I build the same kernel with CONFIG_KEYBOARD_GPIO=n see .config for more information:
http://people.linaro.org/~anders.roxell/job68349_kernel_config_arndale

I do not get any deadlock and the kernel boots, see log:
http://validation.linaro.org/scheduler/job/68349/log_file

[Anders Roxell (aroxell)]

additional information:

kernel version 3.10.6 + rt3

[Mike Holmes (mike-holmes)]

This should be reproducable by adding all the following, or possibly a sub set.

CONFIG_DEBUG_RT_MUTEXES=y
CONFIG_DEBUG_PI_LIST=y
CONFIG_DEBUG_SPINLOCK=y
CONFIG_DEBUG_MUTEXES=y
CONFIG_DEBUG_LOCK_ALLOC=y
CONFIG_PROVE_LOCKING=y
CONFIG_LOCKDEP=y
CONFIG_DEBUG_LOCKDEP=y
CONFIG_TRACE_IRQFLAGS=y
CONFIG_DEBUG_ATOMIC_SLEEP=y
CONFIG_DEBUG_NOTIFIERS=y
CONFIG_DEBUG_PREEMPT=y

[viresh kumar (viresh.kumar)]

I tried fixing this bug and spent quite lot of time onto it.. But couldn't get it fixed :(
I was able to reproduce it with arndale.conf distribution.conf linaro-base.conf preempt-rt.conf + some DEBUG options enabled..
It wasn't reproducible on mainline kernel though..

Still unsure what's causing it to trigger..

[Mike Holmes (mike-holmes)]

Need feedback from Samsung lanfing team, we dont think the driver is RT safe.

[Tushar Behera (tusharbehera)]

Not much update from Samsung LT on this. We spent some time on this, but with little success.