Re: Re: [RFC PATCH] extcon: Fix sleeping in atomic context after connecting USB cable

Re: Re: [RFC PATCH] extcon: Fix sleeping in atomic context after connecting USB cable

[MyungJoo Ham]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=euc-kr, Size: 4191 bytes --]

> On pon, 2014-09-29 at 04:25 +0000, ÇÔ¸íÁÖ wrote:
> > >   
> > >  Kernel built with extcon and charger-manager.
> > > 
> > > After connecting the USB cable sleeping function was called from atomic
> > > context:
> > > [   63.328648] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:586
> > []
> > > [   63.388743] Workqueue: events max14577_muic_irq_work
> > > [   63.393707] [<c00167a4>] (unwind_backtrace) from [<c0012c08>] (show_stack+0x10/0x14)
> > > [   63.401422] [<c0012c08>] (show_stack) from [<c06232ec>] (dump_stack+0x70/0xbc)
> > > [   63.408625] [<c06232ec>] (dump_stack) from [<c0629d0c>] (mutex_lock_nested+0x24/0x410)
> > > [   63.416525] [<c0629d0c>] (mutex_lock_nested) from [<c0354474>] (regmap_read+0x30/0x60)
> > > [   63.424424] [<c0354474>] (regmap_read) from [<c0406b84>] (max14577_charger_get_property+0x1e4/0x250)
> > > [   63.433535] [<c0406b84>] (max14577_charger_get_property) from [<c0403834>] (is_ext_pwr_online+0x30/0x6c)
> > > [   63.442994] [<c0403834>] (is_ext_pwr_online) from [<c0404a54>] (charger_extcon_notifier+0x40/0x70)
> > > [   63.451934] [<c0404a54>] (charger_extcon_notifier) from [<c044cbe4>] (_call_per_cable+0x40/0x4c)
> > > [   63.460704] [<c044cbe4>] (_call_per_cable) from [<c0051560>] (notifier_call_chain+0x64/0x128)
> > > [   63.469209] [<c0051560>] (notifier_call_chain) from [<c0051640>] (raw_notifier_call_chain+0x18/0x20)
> > > [   63.478321] [<c0051640>] (raw_notifier_call_chain) from [<c044d0fc>] (extcon_update_state+0xa8/0x204)
> > > [   63.487522] [<c044d0fc>] (extcon_update_state) from [<c044e5f8>] (max14577_muic_chg_handler+0xdc/0x180)
> > > [   63.496897] [<c044e5f8>] (max14577_muic_chg_handler) from [<c044e924>] (max14577_muic_irq_work+0x7c/0xd8)
> > > [   63.506445] [<c044e924>] (max14577_muic_irq_work) from [<c004480c>] (process_one_work+0x198/0x66c)
> > > [   63.515385] [<c004480c>] (process_one_work) from [<c0044d4c>] (worker_thread+0x38/0x564)
> > > [   63.523455] [<c0044d4c>] (worker_thread) from [<c004c4ec>] (kthread+0xcc/0xe8)
> > > [   63.530661] [<c004c4ec>] (kthread) from [<c000f1f8>] (ret_from_fork+0x14/0x3c)
> > > [   63.543926] charger_manager: Set current limit of CHARGER : 450000uA ~ 450000uA
> > > [   63.548870] extcon-port jack: jack event CHGDET=usb
> > > [   63.550592] extcon-port jack: jack event CHGDET=charger
> > > [   64.188607] charger-manager charger-manager@0: Failed to get battery temperature
> > > [   64.200684] charger-manager charger-manager@0: CHARGING
> > > 
> > > The first sleeping function is is_ext_pwr_online()
> > > (drivers/power/charger-manager.c). The atomic context initiating the
> > > flow is set up in extcon_update_state() (drivers/extcon/extcon-class.c).
> > > 
> > > The extcon_update_state() uses spin locks which are not necessary
> > > because the function is not called from interrupt service routines.
> > > Instead, the extcon_update_state() is called from:
> > > 1. Threaded interrupt handlers.
> > > 2. Work queues.
> > > 
> > > Replace the spin lock with mutex and update the documentation of this
> > > function.
> > 
> > No. You've done it in the opposite way.
> > 
> > update_state is often called in a interrupt handler that cannot sleep.
> > 
> > For the context you've mentioned, we'd need workqueue invoked by
> > _call_per_cable or notifier callback.
> 
> I'm still not convinced that calling raw_notifier_chain under spin lock
> in extcon_update_state() is safe. I think it is not safe because current
> design implies silently that all notifiers registered with
> extcon_register_notifier() must not sleep.

Even if we remove spin_lock completely, whenever an interrupt handler
notifies state change via extcon, you will get the similar BUG messages.
The mutex of I2C won't work in that context.

> 
> _call_per_cable() is just one of ways to trigger this issue.
> 

Thus, in order to satisfy both interrupt handlers and sleeping notifyees,
we can add an option to let _call_per_cable() use workqueue for those who
want to sleep.


Cheers,
MyungJooÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥