MUSB dual-role on AM335x behaving weirdly

[Gregory CLEMENT]

Hi Felipe,

On 18/08/2015 16:13, Felipe Balbi wrote:
> Hi,
> 
> On Tue, Aug 18, 2015 at 02:36:13PM +0200, Gregory CLEMENT wrote:
>> Hi again Felipe,
>>
>> I sent this email again without the capture because it prevented to be delivered
>> to the mailing lists.
>>
>> On 04/08/2015 21:32, Felipe Balbi wrote:
>>> On Tue, Aug 04, 2015 at 04:23:02PM +0200, Gregory CLEMENT wrote:
>>>> Hi again,
>>>> On 04/08/2015 15:08, Gregory CLEMENT wrote:
>>>>> Hi Bin,
>>>>>
>>>>> On 02/07/2015 19:05, Bin Liu wrote:
>>>>>> Hi,
>>>>>>
>>>>>> On Thu, Jul 2, 2015 at 2:16 AM, Gregory CLEMENT
>>>>>> <gregory.clement at free-electrons.com> wrote:
>>>>>>> Hi Felipe,
>>>>>>>
>>>>>>> On 27/05/2015 11:42, Alexandre Belloni wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> On 26/05/2015 at 09:51:18 -0500, Felipe Balbi wrote :
>>>>>>>>> On Thu, May 14, 2015 at 04:36:33PM -0500, Bin Liu wrote:
>>>>>>>>>> Alexandre,
>>>>>>>>>>
>>>>>>>>>> On Thu, May 14, 2015 at 4:26 PM, Alexandre Belloni
>>>>>>>>>> <alexandre.belloni at free-electrons.com> wrote:
>>>>>>>>>>> On 14/05/2015 at 16:16:12 -0500, Bin Liu wrote :
>>>>>>>>>>>> I think I found the root cause of the problem: board design issue - I
>>>>>>>>>>>> bet the custom board has too much cap on VBUS line. It should be <
>>>>>>>>>>>> 10uF.
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> We have a custom board that exhibits the issue but it only has a 100nF
>>>>>>>>>>> cap on VBUS.
>>>>>>>>>>
>>>>>>>>>> Have you measured the VBUS discharging? Is there any way to share your
>>>>>>>>>> schematics?
>>>>>>>>>
>>>>>>>>> Alexandre, any further comments ?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Yeah, I have just got more info.
>>>>>>>>
>>>>>>>> This is the relevant part of the schematic:
>>>>>>>> http://free-electrons.com/~alexandre/usb.png
>>>>>>>>
>>>>>>>> The total VBUS capacitance is 200nF and the USB0 pins are connected
>>>>>>>> directly to the AM3358 pins. U1 is actually not fitted.
>>>>>>>>
>>>>>>>> We didn't measure VBUS discharging but we observe the OTG pin sensing
>>>>>>>> stops when plugging an OTG cable without any device.
>>>>>>>
>>>>>>> Do you have any news about this topic?
>>>>>>>
>>>>>>>
>>>>>>> Is there something else that we can do to help solving this issue?
>>>>>>
>>>>>> In the case of CONFIG_USB_MUSB_DUAL_ROLE=y and dr_mode=otg, how is the
>>>>>> gadget driver configured? It has to be a module not built-in.
>>>>>
>>>>> Indeed when I configured CONFIG_USB_MUSB_HDRC=m and CONFIG_USB_MUSB_DSPS=m
>>>>> it worked seamless.
>>>>>
>>>>
>>>> Actually it didn't worked. And now sometimes I even received continuously
>>>> the following message:
>>>>
>>>>  musb_bus_suspend 2484: trying to suspend as a_wait_vfall while active
>>>
>>> this is likely because your VBUS hasn't dropped below 0.8V fast enough.
>>>
>>> I could only trigger this message in that situation. Use a scope to poke
>>> at VBUS and see how long is takes to reach 0.8V, this could all be cause
>>> by too much capacitance on VBUS line.
>>
>> We got some news:
>> "The capacitance on VBUS due to components is 200nF and the additional parasitic
>> capacitance will be much smaller than this"
>>
>> The rail discharge time is ~36ms when an USB drive is removed from the OTG adapter.
>> I attached a capture of this.
>>
>> What do you think about these values?
>>
>>
>> However, "there appears to be a considerable delay between the removal of a usb
>> drive and the initiation of the VBUS discharge (maybe ~1 second, I wasn't able
>> to measure this time)."
> 
> yeah, this is really weird. I can't think of anything that would make
> VBUS discharge slower from a SW point of view. Once you remove the
> cable, VBUS is physically removed and there's nothing else charging it.

I have more feedback about it: "When I look at it on the oscilloscope
this isn't a 'slow discharge' like a slowly draining capacitor, it is
a delay between the removal of a device and the initiation of the
discharge.  The discharge itself is quite fast once it begins, it just
seems as if the SOC/driver is taking a long time to notice the cable
is disconnected. At this stage, this isn't actually a problem, just
odd."

While working on this issue we found that the
tg_state_a_wait_vrise_timeout case seemed not managed by musb_dsps
driver. I've just submitted a patch for it:
https://lkml.org/lkml/2015/8/20/507

I made most of my test on a 3.17 kernel and today by using a 4.1
kernel with the patch I have submitted I didn't manage to reproduce
the issue. I saw that since 3.17, there were some patches related to
the babble interrupts; so maybe it was enough to fix the issue we saw.
It is still weird because one month ago I also tested with a 4.1
kernel and I had issues...

It needs more testing to see if it is really fixed because the issue
comes only time to time. I will keep you inform about our last tests.


Thanks,

Gregory


-- 
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com