Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
References: <20200429031416.3900-1-bernard@vivo.com>
User-agent: mu4e 1.3.3; emacs 26.3
From:   Jerome Brunet <jbrunet@baylibre.com>
To:     Martin Blumenstingl <martin.blumenstingl@googlemail.com>,
        Bernard Zhao <bernard@vivo.com>,
        Neil Armstrong <narmstrong@baylibre.com>,
        Stephen Boyd <sboyd@kernel.org>,
        Kevin Hilman <khilman@baylibre.com>,
        linux-amlogic@lists.infradead.org, linux-clk@vger.kernel.org,
        linux-kernel@vger.kernel.org
Cc:     opensource.kernel@vivo.com
Subject: Re: [PATCH] clk/meson: fixes memleak issue in init err branch
In-reply-to: <20200429031416.3900-1-bernard@vivo.com>
Date:   Wed, 29 Apr 2020 14:37:09 +0200
Message-ID: <1jpnbqii2y.fsf@starbuckisacylon.baylibre.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On Wed 29 Apr 2020 at 05:14, Bernard Zhao <bernard@vivo.com> wrote:

> In common init function, when run into err branch, we didn`t
> use kfree to release kzmalloc area, this may bring in memleak

Thx for reporting this Bernard.
I'm not a fan of adding kfree everywhere. I'd much prefer a label and
clear error exit path.

That being said, the allocation is probably not the only thing that
needs to be undone in case of error. I guess this is due to conversion
to CLK_OF_DECLARE_DRIVER() which forced to drop all the devm_
This was done because the clock controller was required early in the
boot sequence.

There is 2 paths to properly solve this:
1) Old school: manually undo everything with every error exit condition
   Doable but probably a bit messy
2) Convert back the driver to a real platform driver and use devm_.
   We would still need the controller to register early but I wonder if
   we could use the same method as drivers/clk/mediatek/clk-mt2701.c and
   use arch_initcall() ?

Martin, you did the initial conversion, what do you think of option 2 ?
Would it still answer the problem you were trying to solve back then ?

One added benefit of option 2 is we could drop CLK_OF_DECLARE_DRIVER().
We could even do the same in for the other SoCs, which I suppose would
avoid a fair amount of probe deferral.

>
> Signed-off-by: Bernard Zhao <bernard@vivo.com>
> ---
>  drivers/clk/meson/meson8b.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/clk/meson/meson8b.c b/drivers/clk/meson/meson8b.c
> index 34a70c4b4899..0f07d5a4cd16 100644
> --- a/drivers/clk/meson/meson8b.c
> +++ b/drivers/clk/meson/meson8b.c
> @@ -3687,6 +3687,7 @@ static void __init meson8b_clkc_init_common(struct device_node *np,
>  	if (ret) {
>  		pr_err("%s: Failed to register clkc reset controller: %d\n",
>  		       __func__, ret);
> +		kfree(rstc);
>  		return;
>  	}
>  
> @@ -3710,8 +3711,10 @@ static void __init meson8b_clkc_init_common(struct device_node *np,
>  			continue;
>  
>  		ret = of_clk_hw_register(np, clk_hw_onecell_data->hws[i]);
> -		if (ret)
> +		if (ret) {
> +			kfree(rstc);
>  			return;
> +		}
>  	}
>  
>  	meson8b_cpu_nb_data.cpu_clk = clk_hw_onecell_data->hws[CLKID_CPUCLK];
> @@ -3727,13 +3730,16 @@ static void __init meson8b_clkc_init_common(struct device_node *np,
>  	if (ret) {
>  		pr_err("%s: failed to register the CPU clock notifier\n",
>  		       __func__);
> +		kfree(rstc);
>  		return;
>  	}
>  
>  	ret = of_clk_add_hw_provider(np, of_clk_hw_onecell_get,
>  				     clk_hw_onecell_data);
> -	if (ret)
> +	if (ret) {
>  		pr_err("%s: failed to register clock provider\n", __func__);
> +		kfree(rstc);
> +	}
>  }
>  
>  static void __init meson8_clkc_init(struct device_node *np)