Re: [PATCH v3 2/4] net: stmmac: dwc-qos: Add FSD EQoS support

From: Krzysztof Kozlowski
Date: Fri Aug 18 2023 - 05:28:23 EST


On 16/08/2023 08:38, Sriranjani P wrote:
>
>
>> -----Original Message-----
>> From: Krzysztof Kozlowski [mailto:krzysztof.kozlowski@xxxxxxxxxx]
>> Sent: 15 August 2023 01:21
>> To: Sriranjani P <sriranjani.p@xxxxxxxxxxx>; davem@xxxxxxxxxxxxx;
>> edumazet@xxxxxxxxxx; kuba@xxxxxxxxxx; pabeni@xxxxxxxxxx;
>> robh+dt@xxxxxxxxxx; krzysztof.kozlowski+dt@xxxxxxxxxx;
>> conor+dt@xxxxxxxxxx; richardcochran@xxxxxxxxx;
>> alexandre.torgue@xxxxxxxxxxx; joabreu@xxxxxxxxxxxx;
>> mcoquelin.stm32@xxxxxxxxx; alim.akhtar@xxxxxxxxxxx; linux-
>> fsd@xxxxxxxxx; pankaj.dubey@xxxxxxxxxxx; swathi.ks@xxxxxxxxxxx;
>> ravi.patel@xxxxxxxxxxx
>> Cc: netdev@xxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-
>> kernel@xxxxxxxxxxxxxxx; linux-samsung-soc@xxxxxxxxxxxxxxx; linux-arm-
>> kernel@xxxxxxxxxxxxxxxxxxx; Chandrasekar R <rcsekar@xxxxxxxxxxx>;
>> Suresh Siddha <ssiddha@xxxxxxxxx>
>> Subject: Re: [PATCH v3 2/4] net: stmmac: dwc-qos: Add FSD EQoS support
>>
>> On 14/08/2023 13:25, Sriranjani P wrote:
>>> The FSD SoC contains two instance of the Synopsys DWC ethernet QOS IP
>> core.
>>> The binding that it uses is slightly different from existing ones
>>> because of the integration (clocks, resets).
>>>
>>> For FSD SoC, a mux switch is needed between internal and external clocks.
>>> By default after reset internal clock is used but for receiving
>>> packets properly, external clock is needed. Mux switch to external
>>> clock happens only when the external clock is present.
>>>
>>> Signed-off-by: Chandrasekar R <rcsekar@xxxxxxxxxxx>
>>> Signed-off-by: Suresh Siddha <ssiddha@xxxxxxxxx>
>>> Signed-off-by: Swathi K S <swathi.ks@xxxxxxxxxxx>
>>> Signed-off-by: Sriranjani P <sriranjani.p@xxxxxxxxxxx>
>>> ---
>>
>>
>>> +static int dwc_eqos_setup_rxclock(struct platform_device *pdev, int
>>> +ins_num) {
>>> + struct device_node *np = pdev->dev.of_node;
>>> + struct regmap *syscon;
>>> + unsigned int reg;
>>> +
>>> + if (np && of_property_read_bool(np, "fsd-rx-clock-skew")) {
>>> + syscon = syscon_regmap_lookup_by_phandle_args(np,
>>> + "fsd-rx-clock-
>> skew",
>>> + 1, &reg);
>>> + if (IS_ERR(syscon)) {
>>> + dev_err(&pdev->dev,
>>> + "couldn't get the rx-clock-skew syscon!\n");
>>> + return PTR_ERR(syscon);
>>> + }
>>> +
>>> + regmap_write(syscon, reg, rx_clock_skew_val[ins_num]);
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int fsd_eqos_clk_init(struct fsd_eqos_plat_data *plat,
>>> + struct plat_stmmacenet_data *data) {
>>> + int ret = 0, i;
>>> +
>>> + const struct fsd_eqos_variant *fsd_eqos_v_data =
>>> + plat->fsd_eqos_inst_var;
>>> +
>>> + plat->clks = devm_kcalloc(plat->dev, fsd_eqos_v_data->num_clks,
>>> + sizeof(*plat->clks), GFP_KERNEL);
>>> + if (!plat->clks)
>>> + return -ENOMEM;
>>> +
>>> + for (i = 0; i < fsd_eqos_v_data->num_clks; i++)
>>> + plat->clks[i].id = fsd_eqos_v_data->clk_list[i];
>>> +
>>> + ret = devm_clk_bulk_get(plat->dev, fsd_eqos_v_data->num_clks,
>>> + plat->clks);
>>
>> Instead of duplicating entire clock management with existing code, you
>> should extend/rework existing one.
>>
>> This code is unfortunately great example how not to stuff vendor code into
>> upstream project. :(
>
> I will check again if I can extend existing one to support FSD platform specific requirement.
>
>>
>>> +
>>> + return ret;
>>> +}
>>> +
>>> +static int fsd_clks_endisable(void *priv, bool enabled) {
>>> + int ret, num_clks;
>>> + struct fsd_eqos_plat_data *plat = priv;
>>> +
>>> + num_clks = plat->fsd_eqos_inst_var->num_clks;
>>> +
>>> + if (enabled) {
>>> + ret = clk_bulk_prepare_enable(num_clks, plat->clks);
>>> + if (ret) {
>>> + dev_err(plat->dev, "Clock enable failed, err = %d\n",
>> ret);
>>> + return ret;
>>> + }
>>> + } else {
>>> + clk_bulk_disable_unprepare(num_clks, plat->clks);
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int fsd_eqos_probe(struct platform_device *pdev,
>>> + struct plat_stmmacenet_data *data,
>>> + struct stmmac_resources *res)
>>> +{
>>> + struct fsd_eqos_plat_data *priv_plat;
>>> + struct device_node *np = pdev->dev.of_node;
>>> + int ret = 0;
>>> +
>>> + priv_plat = devm_kzalloc(&pdev->dev, sizeof(*priv_plat),
>> GFP_KERNEL);
>>> + if (!priv_plat) {
>>> + ret = -ENOMEM;
>>
>> return -ENOMEM
>
> Will fix this in v4.
>
>>
>>> + goto error;
>>> + }
>>> +
>>> + priv_plat->dev = &pdev->dev;
>>> + data->bus_id = of_alias_get_id(np, "eth");
>>
>> No, you cannot do like this. Aliases are board specific and are based on
>> labeling on the board.
>
> So if I understood this correctly, I need to move alias in the board specific DTS file

This part: yes

> and I can use this, because we have to handle rx-clock-skew differently for the two instances in the FSD platform.

Not really. Do you expect it to work correctly if given EQoS instance
receives different alias, e.g. 5?

> Another approach we took in v1, by specifying the value to be programmed in rx-clock-skew property itself, but it seems it is not a preferred approach.
> I can see that in drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c +436 common code is already using this API and getting alias id, so I can probably use the same rather getting same again here, but I have to specify alias in DTS file.

Getting alias ID is okay in general. It is used to provide user-visible
ID of the devices (see mmc). Using such alias to configure hardware is
abuse of the alias, because of the reasons I said - how is it supposed
to work if alias is 5 (this is perfectly valid alias)?

I suspect that all this is to substitute missing abstractions, like
clocks, phys or resets...

Best regards,
Krzysztof