lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Wed, 4 May 2016 15:53:27 +0800
From:	Peter Chen <hzpeterchen@...il.com>
To:	Roger Quadros <rogerq@...com>
Cc:	Jun Li <jun.li@....com>,
	"stern@...land.harvard.edu" <stern@...land.harvard.edu>,
	"balbi@...nel.org" <balbi@...nel.org>,
	"gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
	"peter.chen@...escale.com" <peter.chen@...escale.com>,
	"dan.j.williams@...el.com" <dan.j.williams@...el.com>,
	"jun.li@...escale.com" <jun.li@...escale.com>,
	"mathias.nyman@...ux.intel.com" <mathias.nyman@...ux.intel.com>,
	"tony@...mide.com" <tony@...mide.com>,
	"Joao.Pinto@...opsys.com" <Joao.Pinto@...opsys.com>,
	"abrestic@...omium.org" <abrestic@...omium.org>,
	"r.baldyga@...sung.com" <r.baldyga@...sung.com>,
	"linux-usb@...r.kernel.org" <linux-usb@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linux-omap@...r.kernel.org" <linux-omap@...r.kernel.org>
Subject: Re: [PATCH v6 09/12] usb: gadget: udc: adapt to OTG core

On Wed, May 04, 2016 at 09:37:29AM +0300, Roger Quadros wrote:
> Peter,
> 
> On 04/05/16 06:35, Peter Chen wrote:
> > On Tue, May 03, 2016 at 06:44:46PM +0300, Roger Quadros wrote:
> >> Hi,
> >>
> >> On 03/05/16 10:06, Jun Li wrote:
> >>> Hi
> >>>
> >>>>>>>>>>>  /**
> >>>>>>>>>>> + * usb_gadget_start - start the usb gadget controller and
> >>>>>>>>>>> +connect to bus
> >>>>>>>>>>> + * @gadget: the gadget device to start
> >>>>>>>>>>> + *
> >>>>>>>>>>> + * This is external API for use by OTG core.
> >>>>>>>>>>> + *
> >>>>>>>>>>> + * Start the usb device controller and connect to bus (enable
> >>>> pull).
> >>>>>>>>>>> + */
> >>>>>>>>>>> +static int usb_gadget_start(struct usb_gadget *gadget) {
> >>>>>>>>>>> +	int ret;
> >>>>>>>>>>> +	struct usb_udc *udc = NULL;
> >>>>>>>>>>> +
> >>>>>>>>>>> +	dev_dbg(&gadget->dev, "%s\n", __func__);
> >>>>>>>>>>> +	mutex_lock(&udc_lock);
> >>>>>>>>>>> +	list_for_each_entry(udc, &udc_list, list)
> >>>>>>>>>>> +		if (udc->gadget == gadget)
> >>>>>>>>>>> +			goto found;
> >>>>>>>>>>> +
> >>>>>>>>>>> +	dev_err(gadget->dev.parent, "%s: gadget not registered.\n",
> >>>>>>>>>>> +		__func__);
> >>>>>>>>>>> +	mutex_unlock(&udc_lock);
> >>>>>>>>>>> +	return -EINVAL;
> >>>>>>>>>>> +
> >>>>>>>>>>> +found:
> >>>>>>>>>>> +	ret = usb_gadget_udc_start(udc);
> >>>>>>>>>>> +	if (ret)
> >>>>>>>>>>> +		dev_err(&udc->dev, "USB Device Controller didn't
> >>>>>> start: %d\n",
> >>>>>>>>>>> +			ret);
> >>>>>>>>>>> +	else
> >>>>>>>>>>> +		usb_udc_connect_control(udc);
> >>>>>>>>>>
> >>>>>>>>>> For drd, it's fine, but for real otg, gadget connect should be
> >>>>>>>>>> done by
> >>>>>>>>>> loc_conn() instead of gadget start.
> >>>>>>>>>
> >>>>>>>>> It is upto the OTG state machine to call gadget_start() when it
> >>>>>>>>> needs to connect to the bus (i.e. loc_conn()). I see no point in
> >>>>>>>>> calling gadget start before.
> >>>>>>>>>
> >>>>>>>>> Do you see any issue in doing so?
> >>>>>>>>
> >>>>>>>> This is what OTG state machine does:
> >>>>>>>> case OTG_STATE_B_PERIPHERAL:
> >>>>>>>>      otg_chrg_vbus(otg, 0);
> >>>>>>>>      otg_loc_sof(otg, 0);
> >>>>>>>>      otg_set_protocol(fsm, PROTO_GADGET);
> >>>>>>>>      otg_loc_conn(otg, 1);
> >>>>>>>>      break;
> >>>>>>
> >>>>>> On second thoughts, after seen the OTG state machine.
> >>>>>> otg_set_protocol(fsm, PROTO_GADGET); is always followed by
> >>>>>> otg_loc_conn(otg, 1); And whenever protocol changes to anything other
> >>>>>> the PROTO_GADGET, we use otg_loc_conn(otg, 0);
> >>>>>>
> >>>>>> So otg_loc_conn seems redundant. Can we just get rid of it?
> >>>>>>
> >>>>>> usb_gadget_start() implies that gadget controller starts up and
> >>>>>> enables pull.
> >>>>>> usb_gadget_stop() implies that gadget controller disables pull and
> >>>> stops.
> >>>>>>
> >>>>>>
> >>>>>> Can you please explain why just these 2 APIs are not sufficient for
> >>>>>> full OTG?
> >>>>>>
> >>>>>> Do we want anything to happen between gadget controller start/stop
> >>>>>> and pull on/off?
> >>>>>
> >>>>> "loc_conn" is a standard output parameter in OTG spec, it deserves a
> >>>>> separate api, yes, current implementation of OTG state machine code
> >>>>> seems allow you to combine the 2 things into one, but don't do that,
> >>>>> because they do not always happen together, e.g. for peripheral only B
> >>>>> device (also a part OTG spec: section 7.3), will be fixed in gadget
> >>>>> mode, but it will do gadget connect and disconnect in its diff states,
> >>>>> so, to make the framework common, let's keep them separated.
> >>>>
> >>>> I'm sorry but I didn't understand your comment about "it will do gadget
> >>>> connect and disconnect in its diff states"
> >>>
> >>> Gadget connect means loc_conn(1).
> >>>
> >>>>
> >>>> I am reading the OTG v2.0 specification and loc_conn is always true when
> >>>> b_peripheral or a_peripheral is true and false otherwise.
> >>>
> >>> If you only talk about these 2 states, yes, loc_conn is ture.
> >>>
> >>>>
> >>>> loc_conn is just an internal state variable and it corresponds to our
> >>>> gadget_start/stop() state.
> >>>
> >>> It's not an internal variable, there are OTG state machine
> >>> parameters tables(table 7-x) in OTG spec which have clear lists
> >>> which are "internal variable", which are "input", which are "output"...
> >>>
> >>> Those APIs are driven directly from OTG spec, easily understood so
> >>> code reader can know what's those APIs for. For real OTG, I don't
> >>> see the benefit if get rid of it.
> >>
> >> OK, no issues if we don't get rid of it. But I am still in favor of
> >> doing a connect in usb_gadget_start(), because
> >>
> >> 1) If we split connect/disconnect() and usb_gadget_start/stop() then there is
> >> additional overhead of keeping track whether connect was called or not during
> >> usb_gadget_stop(). Plus we need to take care that users don't call connect/disconnect
> >> outside of start/stop. It is just complicating things.
> >>
> >> 2) for many controllers there is no difference between run/stop and
> >> connect/disconnect. i.e. a single register bit controls both.
> >>
> >> 3) it fits well with the OTG specification. OTG specification says
> >> that loc_conn *variable* must be true *after* the device has signalled a connect.
> >> So OTG state machine can safely set loc_conn variable to true after doing
> >> otg_set_protocol(fsm, PROTO_GADGET); and set it to false otherwise.
> >>
> >> Note, OTG specification does not say to take any action based on loc_conn.
> >> It is just a connect indicator variable. So we might have to fix this in the
> >> OTG state machine.
> >>
> >> My suggestion is to keep it simple for now. Try the OTG implementation,
> >> and later if we find issues then extend it as required.
> >>
> > 
> > Just talked with Jun, he is worried if loc_conn != pullup_dp at some
> > situations. So, how about only calling start gadget at usb_start_gadget,
> 
> Which situations?

Not sure if all SoC works like that.

> 
> > and pullup_dp at drd_set_state (see below).
> > 
> > 
> > static void drd_set_state(struct otg_fsm *fsm, enum usb_otg_state new_state)
> > {
> > 	struct usb_otg *otg = container_of(fsm, struct usb_otg, fsm);
> > 
> > 	if (otg->state == new_state)
> > 		return;
> > 
> > 	fsm->state_changed = 1;
> > 	dev_dbg(otg->dev, "otg: set state: %s\n",
> > 		usb_otg_state_string(new_state));
> > 	switch (new_state) {
> > 	case OTG_STATE_B_IDLE:
> > +		usb_udc_vbus_handler(gadget, false);
> 
> This is redundant, When we switch from PROTO_GADGET to PROTO_UNDEF
> we do a usb_gadget_stop(), and a usb_gadget_disconnect() is done there.

Since fully OTG FSM needs to call usb_gadget_stop too, and it doesn't
want need to call usb_gadget_disconnect at usb_gadget_stop, it calls
usb_gadget_disconnect at otg_fsm->ops->loc_conn in its FSM.

You current code combines otg_fsm->ops->loc_conn into
otg_fsm->ops->start_gadget, but according to spec, ->loc_conn is an action
after ->start_gadget.

Keep two actions alone can be flexible.

> 
> > 		drd_set_protocol(fsm, PROTO_UNDEF);
> > 		otg_drv_vbus(otg, 0);
> > 		break;
> > 	case OTG_STATE_B_PERIPHERAL:
> > 		drd_set_protocol(fsm, PROTO_GADGET);
> > +		usb_udc_vbus_handler(gadget, true);
> 
> This is redundant as well since usb_gadget_start() is doing a
> usb_gadget_connect().
> 

The same for above.

> > 		otg_drv_vbus(otg, 0);
> > 		break;
> > 	......
> > 	};
> > 	
> > }
> > 
> > When the OTG FSM is added to this framework, it can keep usb_fsm->ops->loc_conn,
> > and using the current FSM.
> > 
> 
> I have no strong opinions for or against usb_fsm->ops->loc_conn.
> Although strictly speaking, we shouldn't take any action based on loc_conn.
> It is just a state variable indicator.
> 

No, please check the six outputs at the spec, all of them need to
have an action, but your soc/code can do noop if the action has done at
start/stop_role. If it is only a state variable indicator, and we have no
code to use it, why it needs to be at fsm structure. We need to use
this indicator to know if related action needs to be done again, see below API.

static inline int otg_loc_conn(struct otg_fsm *fsm, int on)
{
	if (!fsm->ops->loc_conn)
		return -EOPNOTSUPP;
	if (fsm->loc_conn != on) {
		fsm->loc_conn = on;
		fsm->ops->loc_conn(fsm, on);
	}
	return 0;
}

Yes, it can work even without loc_conn for fully OTG FSM, but it exists
at OTG spec. We'd better to keep it to reflect the spec.

So, since usb_gadget_stop/usb_gadget_start is common API for both drd
and fully fsm, we keep the same function for both. For drd, you doesn't
need to define otg_fsm->ops->loc_conn at your platform, it calls 
usb_udc_vbus_handler at drd_set_state. 

-- 

Best Regards,
Peter Chen

Powered by blists - more mailing lists