Re: [PATCH V3] net: phy: tja11xx: Add TJA11xx PHY driver

[Heiner Kallweit <hkallweit1@gmail.com>]

On 23.12.2018 10:16, Marek Vasut wrote:
> On 12/22/18 6:39 PM, Heiner Kallweit wrote:
>> On 22.12.2018 00:35, Marek Vasut wrote:
>>> Add driver for the NXP TJA1100 and TJA1101 PHYs. These PHYs are special
>>> BroadRReach 100BaseT1 PHYs used in automotive.
>>>
>>> Signed-off-by: Marek Vasut <marex@denx.de>
>>> Cc: Andrew Lunn <andrew@lunn.ch>
>>> Cc: Florian Fainelli <f.fainelli@gmail.com>
>>> Cc: Heiner Kallweit <hkallweit1@gmail.com>
>>> ---
>>> V2: - Use phy_modify(), phy_{set,clear}_bits()
>>>     - Drop enable argument of tja11xx_enable_link_control()
>>>     - Use PHY_BASIC_T1_FEATURES and dont modify supported/advertised
>>>       features in config_init callback
>>>     - Use genphy_soft_reset() instead of opencoding the reset sequence.
>>>     - Drop the aneg parts, since the PHY datasheet claims it does not
>>>       support aneg
>>> V3: - Replace clr with mask
>>>     - Add hwmon support
>>>     - Check commstat in tja11xx_read_status() only if link is up
>>>     - Use PHY_ID_MATCH_MODEL()
>>> ---
>>>  drivers/net/phy/Kconfig       |   6 +
>>>  drivers/net/phy/Makefile      |   1 +
>>>  drivers/net/phy/nxp-tja11xx.c | 424 ++++++++++++++++++++++++++++++++++
>>>  3 files changed, 431 insertions(+)
>>>  create mode 100644 drivers/net/phy/nxp-tja11xx.c
>>>
>> [...]
>>> +
>>> +struct tja11xx_phy_stats {
>>> +	const char	*string;
>>> +	u8		reg;
>>> +	u8		off;
>>> +	u16		mask;
>>> +};
>>> +
>> As written in my other mail, you could think of using
>> FIELD_GET() again. Things like
>> ... n, BIT(n), 
>> ... m, BIT(m), 
>> are simply redundant.
> 
> Done
> 
>>> +static struct tja11xx_phy_stats tja11xx_hw_stats[] = {
>>> +	{ "phy_symbol_error_count", 20, 0, 0xffff },
>>> +	{ "phy_polarity_detect", 25, 6, BIT(6) },
>>> +	{ "phy_open_detect", 25, 7, BIT(7) },
>>> +	{ "phy_short_detect", 25, 8, BIT(8) },
>>> +	{ "phy_rem_rcvr_count", 26, 0, 0xff },
>>> +	{ "phy_loc_rcvr_count", 26, 8, 0xff },
>>
>> Shouldn't mask in the last line be 0xff00 ?
>> In the relevant code you do: val = (reg & mask) >> off
> 
> Yes, fixed, thanks
> 
>>> +static int tja11xx_probe(struct phy_device *phydev)
>>> +{
>>> +	struct device *dev = &phydev->mdio.dev;
>>> +	struct tja11xx_priv *priv;
>>> +	int i;
>>> +
>>> +	priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL);
>>> +	if (!priv)
>>> +		return -ENOMEM;
>>> +
>>> +	priv->hwmon_name = devm_kstrdup(dev, dev_name(dev), GFP_KERNEL);
>>> +	if (!priv->hwmon_name)
>>> +		return -ENODEV;
>>
>> Do you really need to make a copy of the device name?
>> Why not simply priv->hwmon_name = dev_name(dev) ?
> 
> Fine by me, but then maybe I don't quite understand why the other
> drivers duplicate the name, eg. the sfp.c one.
> 
It's a question of object lifetime. If the original object can go away
before your object, then you need to make a copy of the name.
However in our case I don't think priv can live longer than dev.

>> And if devm_kstrdup fails, then most likely you have an out-of-memory
>> error, so why not return -ENOMEM as usual?
> 
> Fixed
> 
>>> +
>>> +	for (i = 0; priv->hwmon_name[i]; i++)
>>> +		if (hwmon_is_bad_char(priv->hwmon_name[i]))
>>> +			priv->hwmon_name[i] = '_';
>>> +
>>> +	priv->hwmon_dev =
>>> +		devm_hwmon_device_register_with_info(dev, priv->hwmon_name,
>>> +						     phydev,
>>> +						     &tja11xx_hwmon_chip_info,
>>> +						     NULL);
>>> +
>> Prerequisite for this call is that HWMON is configured in the kernel and
>> it's reachable. Something like "IS_REACHABLE(CONFIG_HWMON)" would be
>> needed. You can see driver rtc-ds1307 for an example.
> 
> The driver depends on HWMON, so that should be sufficient ?
> 
Missed that, that's sufficient. Just something to think about:
Often HWMON is seen as an optional add-on feature. The driver itself would
work perfectly fine also w/o HWMON. In this case you don't want the hard
dependency. So it's up to you whether you want to allow that the driver is
used on systems w/o HWMON support.