Re: [v2] clocksrouce/timer-imz-gpt: Prevent resource leaks in error path

From: Thomas Gleixner
Date: Thu Sep 22 2016 - 10:37:19 EST


On Thu, 22 Sep 2016, Arvind Yadav wrote:

Darn. Take your time and think about stuff instead of sending a new patch
out 5 seconds after you got a review comment.

> Subject : [v2] clocksrouce/timer-imz-gpt: Prevent resource leaks in error path

1) This lacks a [PATCH v2] prefix

2) clocksrouce is not a valid subsytem

> Free memory and memory mapping, if mxc_timer_init_dt is not successful.

3) There is no memory mapping to free. Simply because that code maps a
peripheral and not memory.

Stop acting like a shell script or your mail address ends up in one which
sends your mails immediately to /dev/null.

> diff --git a/drivers/clocksource/timer-imx-gpt.c b/drivers/clocksource/timer-imx-gpt.c
> index f595460..19f6860 100644
> --- a/drivers/clocksource/timer-imx-gpt.c
> +++ b/drivers/clocksource/timer-imx-gpt.c
> @@ -489,12 +489,17 @@ static int __init mxc_timer_init_dt(struct device_node *np, enum imx_gpt_type t
> return -ENOMEM;
>
> imxtm->base = of_iomap(np, 0);
> - if (!imxtm->base)
> + if (!imxtm->base) {
> + kfree(imxtm);
> return -ENXIO;
> + }
>
> imxtm->irq = irq_of_parse_and_map(np, 0);
> - if (imxtm->irq <= 0)
> + if (imxtm->irq <= 0) {
> + iounmap(imxtm->base);
> + kfree(imxtm);
> return -EINVAL;
> + }
>
> imxtm->clk_ipg = of_clk_get_by_name(np, "ipg");
>
> @@ -506,8 +511,11 @@ static int __init mxc_timer_init_dt(struct device_node *np, enum imx_gpt_type t
> imxtm->type = type;
>
> ret = _mxc_timer_init(imxtm);
> - if (ret)
> + if (ret) {
> + iounmap(imxtm->base);
> + kfree(imxtm);

Ever heard about the concept of goto? Just adding blindly iounmap and kfree
at every place plugs the leak, but is far away from proper kernel code.

Thanks,

tglx