Re: [PATCH v10 06/11] usb: dwc3: qcom: Refactor IRQ handling in QCOM Glue driver

[Krishna Kurapati PSSNV <quic_kriskura@quicinc.com>]

On 8/11/2023 10:35 PM, Konrad Dybcio wrote:
>>>>>> +static int dwc3_get_port_irq(struct platform_device *pdev, u8 port_index)
>>>>>> +{
>>>>>> +    struct dwc3_qcom *qcom = platform_get_drvdata(pdev);
>>>>>> +    bool is_mp_supported = (qcom->data->num_ports > 1) ? true : false;
>>>>>> +    const struct dwc3_acpi_pdata *pdata = qcom->acpi_pdata;
>>>>>> +    char *disp_name;
>>>>>> +    int acpi_index;
>>>>>> +    char *dt_name;
>>>>>> +    int ret;
>>>>>> +    int irq;
>>>>>> +    int i;
>>>>>> +
>>>>>> +    /*
>>>>>> +     * We need to read only DP/DM/SS IRQ's here.
>>>>>> +     * So loop over from 1->3 and accordingly modify respective phy_irq[].
>>>>>> +     */
>>>>>> +    for (i = 1; i < MAX_PHY_IRQ; i++) {
>>>>>> +
>>>>>> +        if (!is_mp_supported && (port_index == 0)) {
>>>>>> +            if (i == DP_HS_PHY_IRQ) {
>>>>>> +                dt_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "dp_hs_phy_irq");
>>>>>> +                disp_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "qcom_dwc3 DP_HS");
>>>>>> +            } else if (i == DM_HS_PHY_IRQ) {
>>>>>> +                dt_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "dm_hs_phy_irq");
>>>>>> +                disp_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "qcom_dwc3 DM_HS");
>>>>>> +            } else if (i == SS_PHY_IRQ) {
>>>>>> +                dt_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "ss_phy_irq");
>>>>>> +                disp_name = devm_kasprintf(&pdev->dev, GFP_KERNEL,
>>>>>> +                    "qcom_dwc3 SS");
>>>> Bjorn, Konrad,
>>>>
>>>> If we are to remove this repetitive loops, we might need to make a 2D array for all of Dp/Dm/Ss interrutps and make a global array of names to be used for irq lookup and use them to reduce the if-else-if stuff here. If that is fine, I can make those changes, else I would like to stick to this approach for now because if we don't add the global array of names, prepping them seperately for dp/dm/ss would again lead us to making if-else loops like above.
>>>>
>>>> Please let me know your thoughts on this.
>>> Can we not just reuse the associated interrupt-names from the devicetree
>>> if present?
>>>
>> Hi Konrad,
>>
>>   Thanks for the comments but one more confirmation.
>> We can read the interrupts from DT but I believe the compatible would still need to stay. We need the num_ports information not just for registering interrupts but for modifying the pwr_event_irq registers during suspend/resume. If we rely on the interrupts to find the number of ports, the user is free to remove any IRQ and we might end up in a situation where glue and core are not having same view of how many number of ports present. So I believe its best to keep the compatible and get num_ports info from there and rely on reading interrupt-names to get interrupts cleanly. Can you let me know your view on the same.
> So is "is it okay to add SoC-specific compatibles and add the port number in
> match data" what you're asking?
> 
> If so, that doesn't seem right.
> 
> The user should not "feel free to remove any IRQ", modifying the devicetree to
> depict a subset of the hardware is not something we want to support. The driver
> has to work with the "full" description in accordance with the bindings.
> 

Hi Konrad,

Thanks for the review.

While I agree with you that we must not skip any hw specifics in DT, 
there is nothing stopping the user from doing so right ?

And whatever be the case, we must be a fool-proof and fail safe system 
able to handle all such situations. While we can read interrupt IRQ 
prefixes to get port count in one way or other, adding a compatible 
would be the least ambiguous path. Is there any other concern you see in 
adding a compatible ? I might be missing something because even Bjorn's 
suggestion too was to try and avoid a new compatible addition and to add 
it only if we have no other way of reliably reading the port count 
(which I believe would be an issue if we need to rely on interrupt name 
reading).

Regards,
Krishna,