Re: [PATCH net-next v3 4/4] net: phy: dp83td510: Add support for the DP83TD510 Ethernet PHY

[Dan Murphy <dmurphy@ti.com>]

Andrew

On 11/3/20 11:18 AM, Andrew Lunn wrote:
> On Tue, Nov 03, 2020 at 11:07:00AM -0600, Dan Murphy wrote:
>> Andrew
>>
>> On 10/30/20 3:15 PM, Andrew Lunn wrote:
>>>> +static int dp83td510_config_init(struct phy_device *phydev)
>>>> +{
>>>> +	struct dp83td510_private *dp83td510 = phydev->priv;
>>>> +	int mst_slave_cfg;
>>>> +	int ret = 0;
>>>> +
>>>> +	if (phy_interface_is_rgmii(phydev)) {
>>>> +		if (dp83td510->rgmii_delay) {
>>>> +			ret = phy_set_bits_mmd(phydev, DP83TD510_DEVADDR,
>>>> +					       DP83TD510_MAC_CFG_1, dp83td510->rgmii_delay);
>>>> +			if (ret)
>>>> +				return ret;
>>>> +		}
>>>> +	}
>>> Hi Dan
>>>
>>> I'm getting a bit paranoid about RGMII delays...
>> Not sure what this means.
> See the discussion and breakage around the realtek PHY. It wrongly
> implemented RGMII delays. When it was fixed, lots of board broke
> because the bug in the PHY driver hid bugs in the DT.
>
I will have to go find that thread. Do you have a link?
>>> Please don't use device_property_read_foo API, we don't want to give
>>> the impression it is O.K. to stuff DT properties in ACPI
>>> tables. Please use of_ API calls.
>> Hmm. Is this a new stance in DT handling for the networking tree?
>>
>> If it is should I go back and rework some of my other drivers that use
>> device_property APIs
> There is a slowly growing understanding what ACPI support in this area
> means. It seems to mean that the firmware should actually do all the
> setup, and the kernel should not touch the hardware configuration. But
> some developers are ignoring this, and just stuffing DT properties
> into ACPI tables and letting the kernel configure the hardware, if it
> happens to use the device_property_read API. So i want to make it
> clear that these properties are for device tree, and if you want to
> use ACPI, you should do things the ACPI way.
>
> For new code, i will be pushing for OF only calls. Older code is a bit
> more tricky. There might be boards out there using ACPI, but doing it
> wrongly, and stuffing OF properties into ACPI tables. We should try to
> avoid breaking them.

Got it.  I will move back to of_* calls

Dan


>        Andrew