Re: [PATCH v6] clk: add si5351 i2c common clock driver

[Sebastian Hesselbarth]

On 04/08/2013 07:46 PM, Guenter Roeck wrote:
> On Mon, Apr 08, 2013 at 06:46:57PM +0200, Sebastian Hesselbarth wrote:
> > This patch adds a common clock driver for Silicon Labs Si5351a/b/c
> > i2c programmable clock generators. Currently, the driver does not
> > support VXCO feature of si5351b. Passing platform_data or DT bindings
> > selectively allow to overwrite stored Si5351 configuration which is
> > very helpful for clock generators with empty eeprom configuration.
> > Corresponding device tree binding documentation is also added.
> >
> > Signed-off-by: Sebastian Hesselbarth<sebastian.hesselbarth@xxxxxxxxx>
>
> [ ... ]
>
> > +
> > +/*
> > + * Si5351 i2c probe and DT
> > + */
> > +#ifdef CONFIG_OF
> > +static const struct of_device_id si5351_dt_ids[] = {
> > +	{ .compatible = "silabs,si5351a", .data = (void *)SI5351_VARIANT_A, },
> > +	{ .compatible = "silabs,si5351a-msop",
> > +					 .data = (void *)SI5351_VARIANT_A3, },
> > +	{ .compatible = "silabs,si5351b", .data = (void *)SI5351_VARIANT_B, },
> > +	{ .compatible = "silabs,si5351c", .data = (void *)SI5351_VARIANT_C, },
> > +	{ }
> > +};
> > +MODULE_DEVICE_TABLE(of, si5351_dt_ids);
> > +
> > +static int si5351_dt_parse(struct i2c_client *client)
> > +{
> > +	struct device_node *child, *np = client->dev.of_node;
> > +	struct si5351_platform_data *pdata;
> > +	const struct of_device_id *match;
> > +	struct property *prop;
> > +	const __be32 *p;
> > +	int num = 0;
> > +	u32 val;
> > +
> > +	if (np == NULL)
> > +		return 0;
> > +
> > +	match = of_match_node(si5351_dt_ids, np);
> > +	if (match == NULL)
> > +		return -EINVAL;
> > +
> > +	pdata = devm_kzalloc(&client->dev, sizeof(*pdata), GFP_KERNEL);
> > +	if (!pdata)
> > +		return -ENOMEM;
> > +
> > +	pdata->variant = (enum si5351_variant)match->data;
> > +	pdata->clk_xtal = of_clk_get(np, 0);
> > +	if (!IS_ERR(pdata->clk_xtal))
> > +		clk_put(pdata->clk_xtal);
> > +	pdata->clk_clkin = of_clk_get(np, 1);
> > +	if (!IS_ERR(pdata->clk_clkin))
> > +		clk_put(pdata->clk_clkin);
> > +
> > +	/*
> > +	 * property silabs,pll-source :<num src>, [<..>]
> > +	 * allow to selectively set pll source
> > +	 */
> > +	of_property_for_each_u32(np, "silabs,pll-source", prop, p, num) {
> > +		if (num>= 2) {
> > +			dev_err(&client->dev,
> > +				"invalid pll %d on pll-source prop\n", num);
> > +			break;
>
> You report dev_err here, but you don't return an error to the caller.
> Is this on purpose ? If it is not a fatal error, maybe it should be dev_info ?

This shouldn't break but continue with one u32 skipped. Actually, it is
a warning because somebody passed an invalid value and should be 
dev_warn(). But see my note below, I will make them all fatal.

> > +		}
> > +
> > +		p = of_prop_next_u32(prop, p,&val);
> > +		if (!p)
> > +			break;
> > +
> > +		switch (val) {
> > +		case 0:
> > +			pdata->pll_src[num] = SI5351_PLL_SRC_XTAL;
> > +			break;
> > +		case 1:
> > +			pdata->pll_src[num] = SI5351_PLL_SRC_CLKIN;
> > +			break;
> > +		default:
> > +			dev_warn(&client->dev,
> > +				 "invalid parent %d for pll %d\n", val, num);
> > +			continue;
>
> Same here, and a couple of times below. Given the context, I think it would
> be better if the error cases would return an error. After all, the condition
> suggests that the device tree data is wrong, meaning one has to assume it
> won't work anyway and should be fixed in the device tree and not be ignored
> by the driver.

I am skipping invalid DT data enties here, but I can make them all
fatal. I will also add more variant checks in the corresponding
_si5351_* functions.

> > +		}
> > +	}
> > +
> > +	/* per clkout properties */
> > +	num = of_get_child_count(np);
> > +
> > +	if (num == 0)
> > +		return 0;
> > +
>
> This doesn't set client->dev.platform_data yet returns no error, meaning the
> calling code will either use provided platform data or fail after all if it is
> NULL. That seems to be inconsistent, given that a dt entry was already detected.
> The user might end up wondering why the provided device tree data is not used,
> not realizing that it is incomplete.
>
> If children are not mandatory, you could just drop the code above and go through
> for_each_child_of_node() anyway; it won't do anything and set
> client->dev.platform_data at the end. If children are mandatory, it might make
> sense to return an eror here (if there is dt information, it should be complete
> and consistent).

Not having children is valid as you only provide them if you want to
overwrite the current config stored in the eeprom. But yes, skipping
for_each_child_of_node below is a left-over from an implementation where
I used dynamically allocated ->pll/->clkout.

> > +	for_each_child_of_node(np, child) {
> > +		if (of_property_read_u32(child, "reg",&num)) {
> > +			dev_err(&client->dev, "missing reg property of %s\n",
> > +				child->name);
> > +			continue;
> > +		}
> > +
> What if "num" returns 100 ? Or 1000 ?

Segmentation fault? ;) I will add a check for 0 <= num < 8.

Thanks for the review,
  Sebastian
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/