Re: [PATCH 2/3] gpio: Fix gpiochip_add_data_with_key() error path

[Linus Walleij]

On Thu, Mar 28, 2019 at 2:13 PM Geert Uytterhoeven
<geert+renesas@xxxxxxxxx> wrote:

> The err_remove_chip block is too coarse, and may perform cleanup that
> must not be done.  E.g. if of_gpiochip_add() fails, of_gpiochip_remove()
> is still called, causing:
>
>     OF: ERROR: Bad of_node_put() on /soc/gpio@e6050000
>     CPU: 1 PID: 20 Comm: kworker/1:1 Not tainted 5.1.0-rc2-koelsch+ #407
>     Hardware name: Generic R-Car Gen2 (Flattened Device Tree)
>     Workqueue: events deferred_probe_work_func
>     [<c020ec74>] (unwind_backtrace) from [<c020ae58>] (show_stack+0x10/0x14)
>     [<c020ae58>] (show_stack) from [<c07c1224>] (dump_stack+0x7c/0x9c)
>     [<c07c1224>] (dump_stack) from [<c07c5a80>] (kobject_put+0x94/0xbc)
>     [<c07c5a80>] (kobject_put) from [<c0470420>] (gpiochip_add_data_with_key+0x8d8/0xa3c)
>     [<c0470420>] (gpiochip_add_data_with_key) from [<c0473738>] (gpio_rcar_probe+0x1d4/0x314)
>     [<c0473738>] (gpio_rcar_probe) from [<c052fca8>] (platform_drv_probe+0x48/0x94)
>
> and later, if a GPIO consumer tries to use a GPIO from a failed
> controller:
>
>     WARNING: CPU: 0 PID: 1 at lib/refcount.c:156 kobject_get+0x38/0x4c
>     refcount_t: increment on 0; use-after-free.
>     Modules linked in:
>     CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.1.0-rc2-koelsch+ #407
>     Hardware name: Generic R-Car Gen2 (Flattened Device Tree)
>     [<c020ec74>] (unwind_backtrace) from [<c020ae58>] (show_stack+0x10/0x14)
>     [<c020ae58>] (show_stack) from [<c07c1224>] (dump_stack+0x7c/0x9c)
>     [<c07c1224>] (dump_stack) from [<c0221580>] (__warn+0xd0/0xec)
>     [<c0221580>] (__warn) from [<c02215e0>] (warn_slowpath_fmt+0x44/0x6c)
>     [<c02215e0>] (warn_slowpath_fmt) from [<c07c58fc>] (kobject_get+0x38/0x4c)
>     [<c07c58fc>] (kobject_get) from [<c068b3ec>] (of_node_get+0x14/0x1c)
>     [<c068b3ec>] (of_node_get) from [<c0686f24>] (of_find_node_by_phandle+0xc0/0xf0)
>     [<c0686f24>] (of_find_node_by_phandle) from [<c0686fbc>] (of_phandle_iterator_next+0x68/0x154)
>     [<c0686fbc>] (of_phandle_iterator_next) from [<c0687fe4>] (__of_parse_phandle_with_args+0x40/0xd0)
>     [<c0687fe4>] (__of_parse_phandle_with_args) from [<c0688204>] (of_parse_phandle_with_args_map+0x100/0x3ac)
>     [<c0688204>] (of_parse_phandle_with_args_map) from [<c0471240>] (of_get_named_gpiod_flags+0x38/0x380)
>     [<c0471240>] (of_get_named_gpiod_flags) from [<c046f864>] (gpiod_get_from_of_node+0x24/0xd8)
>     [<c046f864>] (gpiod_get_from_of_node) from [<c0470aa4>] (devm_fwnode_get_index_gpiod_from_child+0xa0/0x144)
>     [<c0470aa4>] (devm_fwnode_get_index_gpiod_from_child) from [<c05f425c>] (gpio_keys_probe+0x418/0x7bc)
>     [<c05f425c>] (gpio_keys_probe) from [<c052fca8>] (platform_drv_probe+0x48/0x94)
>
> Fix this by splitting the cleanup block, and adding a missing call to
> gpiochip_irqchip_remove().
>
> Fixes: 28355f81969962cf ("gpio: defer probe if pinctrl cannot be found")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> ---
> I'm not so sure about the need for the call to
> gpiochip_irqchip_remove(), as add/remove are not really symmetrical.
> Any comments?

I need to think about this.

Also Mukesh had some comments on naming.

I will try to read through the code closer for v2 and
figure out if it is what we want to do. But I think we need it.

Yours,
Linus Walleij