usb: dwc2: regression during boot on Raspberry Pi

[Stefan Wahren]

Am 12.11.2015 um 05:39 schrieb John Youn:
> On 11/11/2015 3:20 PM, Stefan Wahren wrote:
>> Hi John,
>>
>> Am 11.11.2015 um 00:21 schrieb John Youn:
>>> On 11/9/2015 12:43 PM, Stefan Wahren wrote:
>>>> Hi,
>>>
>>> The gadget side required it but it looks like the host side
>>> didn't.
>>>
>>> When Marek did the low-level hw refactor, some of the gadget
>>> stuff was brought up to the common layer, including the returning
>>> of -EPROBE_DEFER when no phys are defined.
>>>
>>> I think this requirement can be removed. At least only enforce it
>>> for the gadget as before.
>>>
>>> Could you see if the following fixes it?
>>
>> the patch looks okay, but it didn't get USB working after applying your
>> patch in combination with my patch.
>>
>> devm_phy_get(hsotg->dev, "usb2-phy") failed, but ret has a value of -38.
>> I don't know where this strange value comes from.
>
> Anyone know where the -38 (-ENOSYS) might be coming from for rpi?
> I couldn't see anything obvious.

I found it. In case IS_ENABLED(CONFIG_GENERIC_PHY) is false
the function devm_phy_get is replace by the following stub:

static inline struct phy *devm_phy_get(struct device *dev, const char 
*string)
{
	return ERR_PTR(-ENOSYS);
}

>
> Do you know if any PHY is defined?

This is one reason for the whole discussion. Currently there is no USB 
PHY in the device tree neither a USB PHY driver for the Raspberry Pi.

>
>>
>> I also experience issues with my rootfs, so i think could be a problem
>> with my build or the linux-next version. Tomorrow i will try the patches
>> on top of Felipe's repo.
>>
>
> Could you also try this patch that I submitted earlier which just
> makes the phy optional and leaves the error handling the same?
>
> http://marc.info/?l=linux-usb&m=144727055711457&w=2

This patch was part of the draft i tested before and has no influence.

>
> Regards,
> John
>
>