Message-ID: <54EC4ECA.2090300@ti.com>
Date:	Tue, 24 Feb 2015 12:13:30 +0200
From:	Roger Quadros <rogerq@...com>
To:	zhangfei <zhangfei.gao@...aro.org>, <balbi@...com>
CC:	Kishon Vijay Abraham I <kishon@...com>, <mark.rutland@....com>,
	Peter Chen <peter.chen@...escale.com>,
	Sergei Shtylyov <sergei.shtylyov@...entembedded.com>,
	"dan . zhao" <dan.zhao@...ilicon.com>,
	Wangbinghui <wangbinghui@...ilicon.com>,
	<linux-kernel@...r.kernel.org>,
	<linux-arm-kernel@...ts.infradead.org>,
	<devicetree@...r.kernel.org>, <linux-usb@...r.kernel.org>
Subject: Re: [PATCH v4 4/4] phy: add phy-hi6220-usb

Hi Zhangfei,

On 22/02/15 05:10, zhangfei wrote:
> Hi, Balbi
> 
> On 02/22/2015 12:21 AM, Felipe Balbi wrote:
>> Hi,
>>
>> On Sat, Feb 21, 2015 at 11:03:05PM +0800, zhangfei wrote:
>>>>>>> +static void hi6220_start_peripheral(struct hi6220_priv *priv, bool on)
>>>>>>> +{
>>>>>>> +    struct usb_otg *otg = priv->phy.otg;
>>>>>>> +
>>>>>>> +    if (!otg->gadget)
>>>>>>> +        return;
>>>>>>> +
>>>>>>> +    if (on)
>>>>>>> +        usb_gadget_connect(otg->gadget);
>>>>>>> +    else
>>>>>>> +        usb_gadget_disconnect(otg->gadget);
>>>>>>
>>>>>> why is the PHY fiddling with pullups ?
>>>>>
>>>>> We use this to enable/disable otg gadget mode.
>>>>
>>>> I got that, but the pullups don't belong to the PHY, they belong to the
>>>> gadget.
>>>>
>>>>> The gpio_id & gpio_vbus are used to distinguish otg gadget mode or
>>>>> host mode.
>>>>> When micro usb or otg device attached to otg, gpio_vbus falling down.
>>>>> And gpio_id = 1 is micro usb, gpio_id = 0 is otg device.
>>>>
>>>> all of that I understood clearly :-)
>>>>
>>>>> So when micro usb attached, we enable gadget mode; while micro usb
>>>>> detached, we disable gadget mode, and dwc2 will automatically set to
>>>>> host mode.
>>>>
>>>> that's all fine, I'm concerned about letting the PHY fiddle with
>>>> something it doesn't own. If I am to change pullups rules in udc-core,
>>>> this is likely to break down miserably and I don't want to have to go
>>>> through that.
>>>
>>> Thanks for the clarifying.
>>
>> no problem.
>>
>>> How about using usb_gadget_vbus_connect/disconnect, which are used in many
>>> files under drivers/usb/phy.
>>> There is no vbus_session in dwc2/gadget.c, I thought it would be same as
>>> pullup.
>>>
>>> However, usb_gadget_vbus_connect still need para gadget, where should we put
>>> this file, drivers/usb/phy or drivers/phy
>>
>> drivers/phy, if the framework misses anything you need, it's a great
>> opportunity to give back to the community by extending the framework.
> 
> Sorry, I am a little confused.
> I need some concrete suggestion for the next step of this patch, which is required for the community board, hikey board.
> 
> Do you mean in the future we need use hsotg->phy instead of hsotg->uphy.
>         struct phy *phy;
>         struct usb_phy *uphy;
> usb_phy has many members that struct phy does not have, including otg.
> struct usb_otg          *otg;
> Is that mean we need port such member from usb_phy to phy.

In my opinion otg structure should belong to the USB core part that takes care
of the OTG/DRD state machine. We still don't have a clear solution here and
I'm currently investigating this.
My current work is to get Dual role functionality working with DWC3 controller and TI
platforms.

Currently phy drivers take care of OTG operation themselves but there is an opportunity
to share code and centralize USB role switching.
The USB core should be the owner of the Host controller, Gadget controller and the OTG phy
and should take care of the that.

> 
> Besides, are you ok with using usb_gadget_vbus_connect/disconnect.

I don't think PHY is the right place for this even though older drivers seem to be doing so.
But at the same time there is nowhere else to add this at the moment.
The right place should be the USB core that is aware of host/gadget, phy and the state of the bus.

> 
>>
>> Scratching one's own itch kinda thing...
>>
>>>>>>> +static void hi6220_detect_work(struct work_struct *work)
>>>>>>> +{
>>>>>>> +    struct hi6220_priv *priv =
>>>>>>> +        container_of(work, struct hi6220_priv, work.work);
>>>>>>> +    int gpio_id, gpio_vbus;
>>>>>>> +    enum usb_otg_state state;
>>>>>>> +
>>>>>>> +    if (!gpio_is_valid(priv->gpio_id) || !gpio_is_valid(priv->gpio_vbus))
>>>>>>> +        return;
>>>>>>> +
>>>>>>> +    gpio_id = gpio_get_value_cansleep(priv->gpio_id);
>>>>>>> +    gpio_vbus = gpio_get_value_cansleep(priv->gpio_vbus);
>>>>>>
>>>>>> looks like this should be using extcon
>>>>> Not used extcon before.
>>>>> However, we need gpio_vbus interrupt.
>>>>> Checked phy-tahvo.c and phy-omap-otg.c, not find extcon related with
>>>>> interrupt.
>>>>> Will investigate tomorrow.
>>>>
>>>> drivers/extcon/extcon-gpio.c
>>> I think there is no need to use extcon, gpio is clear enough.
>>> extcon-gpio.c even do not support dt.
>>
>> well, add DT. The whole idea of free software is that we improve on
>> things we already have. EXTCON is *the* API to handle such things.
> 

I wrote the extcon-gpio-usb.c driver for exactly your use case. It is
queued for v4.1
https://lkml.org/lkml/2015/2/2/187

It takes care of debouncing for you. Although currently it supports only ID gpio,
it should be very easy to extend to VBUS sense GPIO.

> I think I am still not understanding extcon-gpio, not sure why need use this API here.

several reasons. Let me list a few.
1) Code reuse. Every PHY driver doesn't need to implement GPIO/interrupt handling and debouncing.
It just registers what cable events it wants to hear and gets a notification.
2) The events (ID/VBUS) are not only interesting for the PHY driver but also the controller
driver and the OTG state machine (whenever it exists at a common place) ;).
3) standardization because of common API.

> 
> Here two gpio requires, one gpio as interrupt, in the interrupt handler, we detect the gpio status judging the otg status.
> extcon-gpio.c use the interrupt, then can we also use the gpio interrupt.
> Using extcon-gpio is used for saving gpio_request?
> 
>>
>> Quite frankly, though, Roger Quadros (now in Cc) has been working to get
>> DT support on extcon-gpio, so perhaps wait for that and base your
>> patches on top of his.
>>
>> Now your statement that GPIO is clear enough is completely bogus to me.
>>
>> Why do we have fixed regulators with GPIO enable signals, right ? Might
>> as well just fiddle with the GPIO directly, right ? Wrong, the idea of
>> using these frameworks is to encapsulate implementation details and make
>> sure that if things change from one board to another, we can still use
>> our SW without major modifications.
>>
>>>>>>> +    if (gpio_vbus == 0) {
>>>>>>> +        if (gpio_id == 1)
>>>>>>> +            state = OTG_STATE_B_PERIPHERAL;
>>>>>>> +        else
>>>>>>> +            state = OTG_STATE_A_HOST;
>>>>>>> +    } else {
>>>>>>> +        state = OTG_STATE_A_HOST;
>>>>>>> +    }
>>>>>>> +
>>>>>>> +    if (priv->state != state) {
>>>>>>> +        hi6220_start_peripheral(priv, state == OTG_STATE_B_PERIPHERAL);
>>>>>>> +        priv->state = state;
>>>>>>> +    }
>>>>>>> +}
>>>>>>> +
>>>>>>> +static irqreturn_t hiusb_gpio_intr(int irq, void *data)
>>>>>>> +{
>>>>>>> +    struct hi6220_priv *priv = (struct hi6220_priv *)data;
>>>>>>> +
>>>>>>> +    /* add debounce time */
>>>>>>> +    schedule_delayed_work(&priv->work, msecs_to_jiffies(100));
>>>>>>
>>>>>> this is really bad. We have threaded interrupt support, right ?
>>>>>
>>>>> Since we use two gpio to distinguish gadget mode or host mode.
>>>>> Debounce time can introduce more accuracy.
>>>>
>>>> gpio_set_debounce() ?
>>> Not all gpio.c support set_debounce, including gpio-pl061.c.
>>
>> then the framework should implement it in SW. That was the whole idea of
>> adding set_debounce() to gpiolib. If the controller can't handle it,
>> gpiolib handles it in SW. Again, encapsulating details.
>>
>>>>> I think threaded interrupt can not be used for adding debounce time.
>>>>> Here add debounce is just for safety.
>>>>
>>>> add the debounce to the gpio itself.
>>>
>>> Here the debounce added only for safety.
>>> gpio_id may mis-report when unplug usb, but it is correct for plug usb & otg
>>> device.
>>> So debounce can be omitted.
>>> If you think using delayed work for debounce is ugly, it is fine switch to
>>> threaded_irq.
>>
>> debounce might be very well needed. We *do* want to filter out the
>> transient period. I'm just telling you there are better ways of doing
>> so; and your response to that is "let's just remove it" and I'm not
>> really comfortable with that attitude.
>>
>> That's the attitude of a lazy person which, I hope, you are not ;)
> 
> Understand.
> What I mean here is gpio_id is not used when unplug, it is only used after plug.
> 
> Thanks

cheers,
-roger
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists