linux-wandboard

MLK-13457 ARM: imx: busfreq: fix deadlock detected by lockdep

The deadlock scenario is the following:

1. We schedule low_bus_freq_handle() but it does not run yet.

2. We run set_high_bus_freq() or some other function, that does the
following two things: (a) takes the busfreq mutex and (b)
synchronously cancel the low_bus_freq_handle work

If between (a) and (b) the low_bus_freq_handle work starts running, it
will take the bus freq mutex and block which will cause (b) to
deadlock since the work will never finish now.

To fix this issue avoid synchronously canceling the work and instead
use a new global variable (protected by the busfreq mutex) to mark the
cancellation and abort the work when it is scheduled. In order to
avoid unnecessary schedules we also try to cancel the work with
cancel_delayed_work().

 ======================================================
 [ INFO: possible circular locking dependency detected ]
 4.9.0-rc4-00776-gd4f2779 #348 Tainted: G        W
 -------------------------------------------------------
 kworker/3:1/68 is trying to acquire lock:
  (
 bus_freq_mutex
 ){+.+...}
 , at:
 [<c0128a20>] reduce_bus_freq_handler+0x1c/0x30

 but task is already holding lock:
  (
 (&(&low_bus_freq_handler)->work)
 ){+.+...}
 , at:
 [<c014f4ec>] process_one_work+0x128/0x418

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #1
  (
 (&(&low_bus_freq_handler)->work)
 ){+.+...}
 :

 [<c014dafc>] flush_work+0x44/0x234
 [<c0150348>] __cancel_work_timer+0x98/0x1c8
 [<c01504a4>] cancel_delayed_work_sync+0x14/0x18
 [<c0129d9c>] request_bus_freq+0x9c/0x150
 [<c06b2b28>] imx6q_cpufreq_init+0x8c/0xb8
 [<c06afc9c>] cpufreq_online+0xc0/0x67c
 [<c06b0308>] cpufreq_add_dev+0xb0/0xd4
 [<c05251b0>] subsys_interface_register+0x9c/0xd8
 [<c06af124>] cpufreq_register_driver+0x130/0x1dc
 [<c06b3224>] imx6q_cpufreq_probe+0x5c8/0x8a0
 [<c0528768>] platform_drv_probe+0x54/0xb8
 [<c0526bf8>] driver_probe_device+0x20c/0x2c4
 [<c0526e4c>] __device_attach_driver+0x9c/0xb4
 [<c0524e3c>] bus_for_each_drv+0x6c/0xa0
 [<c05268c8>] __device_attach+0xb8/0x11c
 [<c0526fc4>] device_initial_probe+0x14/0x18
 [<c0525ee8>] bus_probe_device+0x90/0x98
 [<c052401c>] device_add+0x3c8/0x578
 [<c052846c>] platform_device_add+0xa8/0x208
 [<c0529030>] platform_device_register+0x28/0x2c
 [<c0d0f63c>] imx6q_init_late+0x180/0x1c8
 [<c0d03880>] init_machine_late+0x24/0x98
 [<c01019ec>] do_one_initcall+0x44/0x180
 [<c0d00e28>] kernel_init_freeable+0x12c/0x1f4
 [<c0978ba8>] kernel_init+0x10/0x120
 [<c0107ff0>] ret_from_fork+0x14/0x24

 -> #0
  (
 bus_freq_mutex
 ){+.+...}
 :

 [<c01811e4>] lock_acquire+0x78/0x98
 [<c097cedc>] mutex_lock_nested+0x54/0x3e4
 [<c0128a20>] reduce_bus_freq_handler+0x1c/0x30
 [<c014f558>] process_one_work+0x194/0x418
 [<c014f810>] worker_thread+0x34/0x4fc
 [<c0155e44>] kthread+0xdc/0xf8
 [<c0107ff0>] ret_from_fork+0x14/0x24

 other info that might help us debug this:

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock( (&(&low_bus_freq_handler)->work) );
                                lock( bus_freq_mutex );
                                lock( (&(&low_bus_freq_handler)->work) );
   lock( bus_freq_mutex );

  *** DEADLOCK ***

 2 locks held by kworker/3:1/68:
  #0:
  (
 "events"
 ){.+.+.+}
 , at:
 [<c014f4ec>] process_one_work+0x128/0x418
  #1:
  (
 (&(&low_bus_freq_handler)->work)
 ){+.+...}
 , at:
 [<c014f4ec>] process_one_work+0x128/0x418

 stack backtrace:
 CPU: 3 PID: 68 Comm: kworker/3:1 Tainted: G        W       4.9.0-rc4-00776-gd4f2779 #348
 Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
 Workqueue: events reduce_bus_freq_handler
 Backtrace:
 [<c010c538>] (dump_backtrace) from [<c010c730>] (show_stack+0x18/0x1c)
 [<c010c718>] (show_stack) from [<c0403a58>] (dump_stack+0xb4/0xe8)
 [<c04039a4>] (dump_stack) from [<c017d4f0>] (print_circular_bug+0x1d4/0x318)
 [<c017d31c>] (print_circular_bug) from [<c0180bb4>] (__lock_acquire+0x1864/0x1ad4)
 [<c017f350>] (__lock_acquire) from [<c01811e4>] (lock_acquire+0x78/0x98)
 [<c018116c>] (lock_acquire) from [<c097cedc>] (mutex_lock_nested+0x54/0x3e4)
 [<c097ce88>] (mutex_lock_nested) from [<c0128a20>] (reduce_bus_freq_handler+0x1c/0x30)
 [<c0128a04>] (reduce_bus_freq_handler) from [<c014f558>] (process_one_work+0x194/0x418)
 [<c014f3c4>] (process_one_work) from [<c014f810>] (worker_thread+0x34/0x4fc)
 [<c014f7dc>] (worker_thread) from [<c0155e44>] (kthread+0xdc/0xf8)
 [<c0155d68>] (kthread) from [<c0107ff0>] (ret_from_fork+0x14/0x24)

Signed-off-by: Octavian Purdila <octavian.purdila@nxp.com>
Reviewed-by: Ranjani Vaidyanathan <ranjani.vaidyanathan@nxp.com>