Re: [RFC 5/5] USB: Eliminate delays involved in root hub initialization during HNP

[Pavan Kondeti <pkondeti@codeaurora.org>]

On 12/16/2010 5:45 PM, Felipe Balbi wrote:
> On Thu, Dec 16, 2010 at 04:56:15PM +0530, Pavan Kondeti wrote:
>> On 12/15/2010 5:51 PM, Felipe Balbi wrote:
>>> On Wed, Dec 15, 2010 at 04:44:14PM +0530, Pavankumar Kondeti wrote:
>>>> Some USB controllers have common resources (IRQ, register address
>>>> space) for Host, Peripheral and OTG.  So HCD is added only before
>>>> entering into Host mode.  Root hub initialization is done in
>>>> different steps to decrease boot up time.  But this makes B-device
>>>> difficult to meet HNP timings.  Hence eliminate delays involved in
>>>> root hub initialization for B-host.
>>>
>>> I wonder if this is the best approach. Would it be easier to not touch
>>> usbcore, probe the entire stack during boot but have a "core" layer
>>> handling synchronization to shared resources ?
>>>
>> The implementation is like this:
>>
>> Actually OTG synchronizes UDC and HCD. OTG driver probe will be called
>> first and it takes care of turning on clocks, resetting controller and
>> PHY and enabling interrupts (VBUS and Id). UDC and HCD will not modify
>> hardware registers in their probe functions. So HCD will not call
>> usb_add_hcd() function in it's probe(). After HCD and UDC registers with
>> OTG via otg_set_xxx(), OTG activates HCD (by calling usb_add_hcd()) or
>> UDC (usb_gadget_connect()) based on Id/VBUS status. When gadget is
>> active, HCD is detached from USB core (usb_remove_hcd()). So user space
>> can not poke into sysfs/debugfs nodes provided in the usb core.
>>
>>> Maybe you could make your device an MFD device that allocates
>>> platform_devices for its children (HCI, UDC, OTG, etc) and pass in a
>>> bunch of read/write functions as platform_data for them to use as
>>> accessors to shared register space ?!? Would that work ?
>>>
>> The easy way to deal this problem is removing HCD when Mini/Micro -A
>> cable is detached.
> 
> But why do you need to do that isn't clear. What's the problem in
> keeping HCD always available ? If it's a mini/micro-B cable you will
> enter peripheral role anyway.
> 
If you keep the HCD, root hub auto/system suspend can happen, user may
un-configure the root hub or remove the HCD etc. There is not standard
way of guarding against PM & user for this kind of HCD.

We had a discussion about this some time back.

Please see http://www.spinics.net/lists/linux-usb/msg35847.html

Thanks,
Pavan

-- 
Sent by a consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.