Date:	Fri, 01 Jul 2016 12:41:43 +0300
From:	Felipe Balbi <balbi@...nel.org>
To:	Roger Quadros <rogerq@...com>, tony@...mide.com
Cc:	stern@...land.harvard.edu, sre@...nel.org, peter.chen@....com,
	linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org,
	linux-omap@...r.kernel.org
Subject: Re: [PATCH] usb: host: Allow EHCI_OMAP to be built-in when USB_GADGET is 'm'


Hi,

Roger Quadros <rogerq@...com> writes:
>>>>>>> NOP_USB_XCEIV is used not only by gadget drivers but by
>>>>>>> host drivers as well e.g. EHCI_OMAP.
>>>>>>>
>>>>>>> commit 5a8d651a2bde ("usb: gadget: move gadget API functions to udc-core")
>>>>>>> made it so that NOP_USB_XCEIV can't be built-in if USB_GADGET is 'm'.
>>>>>>> But this prevents EHCI_OMAP to be built-in if USB_GADGET is 'm'.
>>>>>>>
>>>>>>> Fix this undesired behaviour by moving usb_gadget_vbus_connect/disconnect()
>>>>>>> to usb/gadget.h so that NOP_USB_XCEIV has no build dependency
>>>>>>> on USB_GADGET.
>>>>>>>
>>>>>>> Retain the original Kconfig behaviour i.e. NOP_USB_XCEIV is selected
>>>>>>> by drivers that need it.
>>>>>>
>>>>>> no, this is the wrong way to fix this. NOP _has_ a dependency on the
>>>>>> Gadget API if it calls Gadget API functions. Dependencies are proper.
>>>>>>
>>>>>> Maybe the reason for the problem is that we ended up adding far too much
>>>>>> code to phy-generic.c itself. Maybe it shouldn't know about clks and
>>>>>> interrupts. The original idea of that driver was to simply satisfy a
>>>>>> requirement to have a valid transceiver by some platforms. Maybe we
>>>>>> should fix that instead. Moving functions around to workaround a problem
>>>>>> is not the way to go, sorry.
>>>>>>
>>>>> OK but something that was working all these years is broken by your
>>>>> patch.
>>>>> Do you mind fixing it please? Or at least let me know how you want to
>>>>> get it fixed.
>>>>
>>>> diff --git a/drivers/usb/host/Kconfig b/drivers/usb/host/Kconfig
>>>> index 2e710a4cca52..89fd095ca33d 100644
>>>> --- a/drivers/usb/host/Kconfig
>>>> +++ b/drivers/usb/host/Kconfig
>>>> @@ -180,7 +180,7 @@ config USB_EHCI_MXC
>>>>  config USB_EHCI_HCD_OMAP
>>>>         tristate "EHCI support for OMAP3 and later chips"
>>>>         depends on ARCH_OMAP
>>>> -       depends on NOP_USB_XCEIV
>>>> +       depends on USB_PHY
>>>>         default y
>>>>         ---help---
>>>>           Enables support for the on-chip EHCI controller on
>>>>
>>>>
>>> This doesn't fix the real problem. i.e. we can't have NFS root on pandaboard
>>> and omap5_uevm because network device is on USB EHCI, unless we have
>>> both USB_GADGET and NOP_USB_XCEVI built-in, which is not really that
>>> desirable.
>> 
>> Well, NOP calls a symbol that belongs to the Gadget API. So NOP cannot
>> be built-in if gadget API is a module. Satisfy the dependencies and
>> everything will work for you.
>> 
>> Don't get me wrong, I see what you mean; but you _are_ relying in a
>> driver that calls into the Gadget API. Consider if it was the other way
>> around: say NOP was calling something from usbcore for whatever reason
>> and MUSB depended on NOP. You wouldn't have MUSB built-in unless NOP and
>> usbcore were also built-in, right?
>> 
> I agree with you. As far as EHCI_OMAP is concerned we're only interested
> in configuring the PHY clocks and issuing a GPIO reset.
> As of now because of dependency issues with GADGET, NOP_USB_XCEIV doesn't
> seem to be the right place. I'm OK with using some other driver
> that doesn't have such dependency, even if it breaks DT compatibility with
> some OMAP boards. Or I'll just let EHCI_OMAP do it all by itself.
>
> Unless Tony has any objections.

I don't have memory of why EHCI-omap is using PHY, actually. Is it
really resetting a PHY?

-- 
balbi

Download attachment "signature.asc" of type "application/pgp-signature" (819 bytes)

Powered by blists - more mailing lists