We shouldn't call wakeup_source_destroy() from the error path in
wakeup_source_register() because that calls __pm_relax() which takes a
lock that isn't initialized until wakeup_source_add() is called. Add a
new function, wakeup_source_free(), that just does the bare minimum to
free a wakeup source that was created but hasn't been added yet and use
it from the two places it's needed. This fixes the following problem
seen on various x86 server boxes:

 INFO: trying to register non-static key.
 the code is fine but needs lockdep annotation.
 turning off the locking correctness validator.
 CPU: 12 PID: 1 Comm: swapper/0 Not tainted 5.3.0-rc4-
 Hardware name: HP ProLiant XL420 Gen9/ProLiant XL420 Gen9, BIOS U19 12/27/2015
 Call Trace:
  dump_stack+0x62/0x9a
  register_lock_class+0x95a/0x960
  ? __platform_driver_probe+0xcd/0x230
  ? __platform_create_bundle+0xc0/0xe0
  ? i8042_init+0x4ec/0x578
  ? do_one_initcall+0xfe/0x45a
  ? kernel_init_freeable+0x614/0x6a7
  ? kernel_init+0x11/0x138
  ? ret_from_fork+0x35/0x40
  ? is_dynamic_key+0xf0/0xf0
  ? rwlock_bug.part.0+0x60/0x60
  ? __debug_check_no_obj_freed+0x8e/0x250
  __lock_acquire.isra.13+0x5f/0x830
  ? __debug_check_no_obj_freed+0x152/0x250
  lock_acquire+0x107/0x220
  ? __pm_relax.part.2+0x21/0xa0
  _raw_spin_lock_irqsave+0x35/0x50
  ? __pm_relax.part.2+0x21/0xa0
  __pm_relax.part.2+0x21/0xa0
  wakeup_source_destroy.part.3+0x18/0x190
  wakeup_source_register+0x43/0x50

Fixes: c8377adf ("PM / wakeup: Show wakeup sources stats in sysfs")
Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>