ac100 team mailing list archive

[Lee Jones <lee.jones@xxxxxxxxxx>]

On Mon, 23 Feb 2015, Krzysztof Kozlowski wrote:

> Change the ownership of power_supply structure from each driver
> implementing the class to the power supply core.
> 
> The patch changes power_supply_register() function thus all drivers
> implementing power supply class are adjusted.
> 
> Each driver provides the implementation of power supply. However it
> should not be the owner of power supply class instance because it is
> exposed by core to other subsystems with power_supply_get_by_name().
> These other subsystems have no knowledge when the driver will unregister
> the power supply. This leads to several issues when driver is unbound -
> mostly because user of power supply accesses freed memory.
> 
> Instead let the core own the instance of struct 'power_supply'.  Other
> users of this power supply will still access valid memory because it
> will be freed when device reference count reaches 0. Currently this
> means "it will leak" but power_supply_put() call in next patches will
> solve it.
> 
> This solves invalid memory references in following race condition
> scenario:
> 
> Thread 1: charger manager
> Thread 2: power supply driver, used by charger manager
> 
> THREAD 1 (charger manager)         THREAD 2 (power supply driver)
> ==========================         ==============================
> psy = power_supply_get_by_name()
>                                    Driver unbind, .remove
>                                      power_supply_unregister()
>                                      Device fully removed

[...]

>  include/linux/mfd/abx500/ux500_chargalg.h |  11 +-
>  include/linux/mfd/rt5033.h                |   2 +-
>  include/linux/mfd/wm8350/supply.h         |   6 +-

Acked-by: Lee Jones <lee.jones@xxxxxxxxxx>

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog