[PATCH v5 3/4] USB: ehci-s5p: Add phy driver support

[Doug Anderson]

On Wed, Dec 19, 2012 at 10:37 PM, Vivek Gautam
<gautamvivek1987 at gmail.com> wrote:
>
> On Thu, Dec 20, 2012 at 5:00 AM, Doug Anderson <dianders at chromium.org> wrote:
>>
>> On Tue, Dec 18, 2012 at 6:43 AM, Vivek Gautam <gautam.vivek at samsung.com> wrote:
>>> +static void s5p_ehci_phy_enable(struct s5p_ehci_hcd *s5p_ehci)
>>> +{
>>> +       struct platform_device *pdev = to_platform_device(s5p_ehci->dev);
>>> +
>>> +       if (s5p_ehci->phy) {
>>> +               samsung_usbphy_set_type(s5p_ehci->phy, USB_PHY_TYPE_HOST);
>>
>> This confuses me.  Why are you setting the type to host here?
>>
> Being in host controller, before calling usb_phy_init() we set type to
> Host since, with certain SOCs
> like 4210, same register has different bit settings for HOST type and
> device type. So setting this
> to Host type here make the flow of usb_phy_init to go in the direction of Host.

OK.  I think I need to study the code more...

>>>
>>> +       phy = devm_usb_get_phy(&pdev->dev, USB_PHY_TYPE_USB2);
>>> +       if (IS_ERR_OR_NULL(phy)) {
>>> +               /* Fallback to pdata */
>>> +               if (!pdata) {
>>> +                       dev_err(&pdev->dev, "no platform data or transceiver defined\n");
>>> +                       return -EPROBE_DEFER;
>>
>> Shouldn't you return -EINVAL like the old code did?
>
> We are deferring the probe since the usb-phy transceiver may get
> probed after ehci/ohci controllers.
> And if we return -EINVAL like the previous code, we would end up not
> setting the phy.

OK.  Something is wrong here then, since this really isn't an error:
* It should be commented about why you're deferring.
* You shouldn't have a dev_err for something that's not fatal.

Ideally we'd want something that would force probing to happen in the
right order.  I spent a little time looking and didn't see anything,
but maybe I'm missing something obvious.  If nothing pops out, the
defer seems OK as long as it is commented well.