Re: [PATCH] serial: vt8500: Use devm_clk_get_enabled()

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Low] serial: vt8500: malformed commit message and missing
  Signed-off-by
It also noticed 1 related pre-existing issue this patch does not
introduce:
- [Medium] serial: vt8500: port index leak in vt8500_serial_probe() [1]
--

commit 8b3881e4e3251512b27bf8982412d7a23856190f
Author: Wentao Liang <vulab@iscas.ac.cn>

> serial: vt8500: Use devm_clk_get_enabled()
>
> Hi Alexey,
>
> Sorry for the very late reply. Working through the review comments I got
> in September has taken me far longer than it should have.
>
> Your suggestion was right, but the leak fix went in before I could act on
> it: it is commit 9d8955281d92 ("serial: vt8500: Fix clock reference leak
> in vt8500_serial_probe()") in mainline, and it kept the clk_put() you
> asked me to replace. It has been backported to the stable trees by now as
> well. So the patch below is not a v2 of that one, which would only resend
> something that is already applied, but a follow-up that does the
> conversion on top of it.
>
> devm_clk_get_enabled() gets the clock and enables it in one call and
> registers the disable and put with devm, so both the manual error handling
> and the clk_put() go away. Probe now returns the error code from the clock
> lookup instead of a hard-coded -EINVAL, and the clock is released
> automatically through devres. The success path is unchanged and nothing
> else in the driver is touched.
>
> For the record on where this came from: the candidate came from our static
> analysis tooling, and DeepSeek and GPT-6.1 Sol were used to audit it and to
> help write the change. I reviewed the result and I am responsible for it.
> The patch is build-tested only: it compiles cleanly with W=1 for arm
> (vt8500_v6_v7_defconfig, CONFIG_SERIAL_VT8500=y) and passes
> checkpatch.pl --strict; I have no VT8500 hardware to test it on.
>
> Thanks for the review,
> Wentao
>
> From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
> From: Wentao Liang <vulab@iscas.ac.cn>
> Date: Sun, 8 Nov 2026 11:41:30 +0000
> Subject: [PATCH] serial: vt8500: Use devm_clk_get_enabled()
>
> vt8500_serial_probe() gets the clock with of_clk_get() and enables it
> with clk_prepare_enable(), so the reference has to be dropped by hand on
> the error path, and the clock is never disabled or released afterwards,
> even though the rest of the probe is devm-managed.
>
> Use devm_clk_get_enabled() instead, which gets and enables the clock in
> one call and registers the teardown with devm. The manual error handling
> and the clk_put() can then go, and probe returns the actual error code
> from the clock lookup rather than a hard-coded -EINVAL.
>
> The clock still stays enabled as long as the port is in use, so the
> success path is unchanged.

[Severity: Low]
The commit message contains an unstripped email reply and embedded git
format-patch headers, and is missing a Signed-off-by: trailer. Personal
commentary and build-testing notes belong below the --- separator (or
above a scissors line) so they do not end up in the permanent git log.

> diff --git a/drivers/tty/serial/vt8500_serial.c b/drivers/tty/serial/vt8500_serial.c
> index fc6fdc751b167..73922523dff64 100644
> --- a/drivers/tty/serial/vt8500_serial.c
> +++ b/drivers/tty/serial/vt8500_serial.c
[ ... ]
> @@ -647,17 +646,10 @@ static int vt8500_serial_probe(struct platform_device *pdev)
>  	if (IS_ERR(vt8500_port->uart.membase))
>  		return PTR_ERR(vt8500_port->uart.membase);
>  
> -	vt8500_port->clk = of_clk_get(pdev->dev.of_node, 0);
> +	vt8500_port->clk = devm_clk_get_enabled(&pdev->dev, NULL);
>  	if (IS_ERR(vt8500_port->clk)) {
>  		dev_err(&pdev->dev, "failed to get clock\n");
> -		return  -EINVAL;
> -	}
> -
> -	ret = clk_prepare_enable(vt8500_port->clk);
> -	if (ret) {
> -		dev_err(&pdev->dev, "failed to enable clock\n");
> -		clk_put(vt8500_port->clk);
> -		return ret;
> +		return PTR_ERR(vt8500_port->clk);
>  	}

[Severity: Medium]
This problem wasn't introduced by this patch, but does returning on error
here (or from the earlier devm_kzalloc() and
devm_platform_get_and_ioremap_resource() checks) leak the port index in
vt8500_ports_in_use [1]?

drivers/tty/serial/vt8500_serial.c:vt8500_serial_probe() {
	...
	/* reserve the port id */
	if (test_and_set_bit(port, vt8500_ports_in_use)) {
		/* port already in use - shouldn't really happen */
		return -EBUSY;
	}
	...
}

If devm_clk_get_enabled() fails (for example with -EPROBE_DEFER), the bit
in vt8500_ports_in_use is never cleared, so a later retry or re-bind will
find the port still marked in use and fail with -EBUSY (or exhaust
VT8500_MAX_PORTS for dynamically assigned ports).

[1] https://sashiko.dev/#/bug/linux-ca000721-4a5b-4d88-bdb8-7f1e438dc81e

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/1794138380.2883200.vulab@iscas.ac.cn?part=1