Re: [PATCH v6 2/2] phy: qualcomm: Add Synopsys High-Speed USB PHY driver

From: Manu Gautam
Date: Fri Dec 21 2018 - 03:13:23 EST


Hi,

On 12/20/2018 4:39 PM, Shawn Guo wrote:
> Hi Manu,
>
> On Thu, Dec 20, 2018 at 09:33:43AM +0530, mgautam@xxxxxxxxxxxxxx wrote:
>> Hi Shawn,
>>
>> On 2018-12-20 06:31, Shawn Guo wrote:
>>> It adds Synopsys 28nm Femto High-Speed USB PHY driver support, which
>>> is usually paired with Synopsys DWC3 USB controllers on Qualcomm SoCs.
>>>
>>> Signed-off-by: Shawn Guo <shawn.guo@xxxxxxxxxx>
>>> ---
>>
...
>>> +struct hsphy_priv {
>> nit-pick - indentation for following structure members?
> Hmm, my personal taste says no, because I found that it's hard to keep
> the indentation when adding new members later.
ok
>
>>> + void __iomem *base;
>>> + struct clk_bulk_data *clks;
>>> + int num_clks;
>>> + struct reset_control *phy_reset;
>>> + struct reset_control *por_reset;
>>> + struct regulator_bulk_data vregs[VREG_NUM];
>>> + unsigned int voltages[VREG_NUM][VOL_NUM];
>>> + const struct hsphy_data *data;
>>> + bool cable_connected;
>> You can get cable-connected state from "enum phy_mode mode" which
>> is present in this driver.
>> E.g. cable_connected is false if mode is neither HOST nor DEVICE.
>>
>>
>>> + struct extcon_dev *vbus_edev;
>>> + struct notifier_block vbus_notify;
>> extcons not needed if you use "mode" for the same purpose.
> The extcon is there for indicating cable connection status. I'm not
> sure that phy_mode can be used for that purpose. For example, what
> value would phy core set phy_mode to, if we disconnect the cable from
> phy_mode being HOST or DEVICE?

it depends how it is used. Looks like it is used to decide whether to turn-off
regulators or not. Unless you plan to add low power support as part of
runtime-suspend of PHY during host mode, there is no reason to not turn-off
regulators on pm_suspend(). Please refer to my comments below.

>
>>
>>> + enum phy_mode mode;
>>> +};
>>> +
>>
>>> +
>>> +static int qcom_snps_hsphy_vbus_notifier(struct notifier_block *nb,
>>> + unsigned long event, void *ptr)
>>> +{
>>> + struct hsphy_priv *priv = container_of(nb, struct hsphy_priv,
>>> + vbus_notify);
>>> + priv->cable_connected = !!event;
>>> + return 0;
>>> +}
>>> +
>>> +static int qcom_snps_hsphy_power_on(struct phy *phy)
>> Can you instead merge this power_on function with phy_init?
> I can do that, but what's the gain/advantage from doing that?

dwc3 core calls phy_init() before power_on(). AFAIK, PHY regulators
need to be ON before initializing it.

>
>>> +{
>>> + struct hsphy_priv *priv = phy_get_drvdata(phy);
>>> + int ret;
>>> +
>>> + if (priv->cable_connected) {
>> Why distinguish between cable connected vs not-connected?
> This is based on the vendor driver implementation. It does a more
> aggressive low power management in case that cable is not connected,
> i.e. turning off regulator and entering retention mode.

I believe 'aggressive low power management' will be triggered on
pm_suspend?
And dwc3 core will in any case perform phy_exit()->phy_init() across
pm_suspend/resume which will reset PHYs. Hence, there is no need to check
for cable connected state here.


>
>>> + ret = clk_bulk_prepare_enable(priv->num_clks, priv->clks);
>>> + if (ret)
>>> + return ret;
>>> + qcom_snps_hsphy_disable_hv_interrupts(priv);
>>> + } else {
>>> + ret = qcom_snps_hsphy_enable_regulators(priv);
>>> + if (ret)
>>> + return ret;
>>> + ret = clk_bulk_prepare_enable(priv->num_clks, priv->clks);
>>> + if (ret)
>>> + return ret;
>>> + qcom_snps_hsphy_exit_retention(priv);
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> +static int qcom_snps_hsphy_power_off(struct phy *phy)
>>> +{
>>> + struct hsphy_priv *priv = phy_get_drvdata(phy);
>>> +
>>> + if (priv->cable_connected) {
>>> + qcom_snps_hsphy_enable_hv_interrupts(priv);
>>> + clk_bulk_disable_unprepare(priv->num_clks, priv->clks);
>>> + } else {
>>> + qcom_snps_hsphy_enter_retention(priv);
>>> + clk_bulk_disable_unprepare(priv->num_clks, priv->clks);
>>> + qcom_snps_hsphy_disable_regulators(priv);
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>
>>
>> ..
>>> +static const struct phy_ops qcom_snps_hsphy_ops = {
>>> + .init = qcom_snps_hsphy_init,
>>> + .power_on = qcom_snps_hsphy_power_on,
>>> + .power_off = qcom_snps_hsphy_power_off,
>>> + .set_mode = qcom_snps_hsphy_set_mode,
>> .phy_exit()?
>> I believe that is needed as dwc3 core driver performs
>> phy_exit/phy_init across pm_suspend/resume.
> I just do not see anything that we should be doing in .exit hook right
> now.

After you merge power_on() with phy_init() as per my previous comment,
you can rely on phy_exit() to take care of putting PHY in low power state
and turn off regulators as well.

>
> Shawn
>
>>
>>> + .owner = THIS_MODULE,
>>> +};
>>> +

--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project