Re: [PATCH] usb: musb: dsps: handle the otg_state_a_wait_vrise_timeout case

From: Felipe Balbi
Date: Mon Dec 07 2015 - 14:27:33 EST



Hi,

Gregory CLEMENT <gregory.clement@xxxxxxxxxxxxxxxxxx> writes:
> Hi Felipe,
>
> I am going back on this subject (again :) )
>
> On mar., oct. 20 2015, Gregory CLEMENT <gregory.clement@xxxxxxxxxxxxxxxxxx> wrote:
>
>> Hi Felipe,
>>
>> On lun., oct. 05 2015, Felipe Balbi <balbi@xxxxxx> wrote:
>>
>>
>>>> So after many tests on different devices, 200ms is enough for most of
>>>> them, but for one, 2000ms (2s) was needed!
>>>>
>>>> I see several option:
>>>> - adding a sysfs entry to setup the time
>>>> - adding a debugs entry entry
>>>> - adding configuration option in menuconfig
>>>> - using 2000ms and hopping it was enough for everyone
>>>>
>>>> My preference would go to the first option, what is your opinion?
>>>
>>> I'm not sure if either of them is good. man 2s is just too large. If the
>
> Well 2s is lon I agree, but currently instead of 2s we have an infinite
> wait, which is worse!
>
>>> device isn't following the specification, I'm afraid that device needs
>>> to be fixed.
>
> I agree but these devices are for most of them USB stick that we find in
> retail. We have no influence on them, so we have to do with them, even
> if they do not follow the sepc.
>
> So what about using configfs or sysfs to let the user confgiure this
> value if needed?

iff we have to; I'm still not 100% convinced.

> I go back on this because, your suggestion didn't work. At least, I
> didn't manage to make it improve the situation.
>
> Thanks,
>
>>>
>>> I think the real issue here is the lack of a disconnect IRQ from AM335x
>>> devices. But here's something I've been meaning to test but never had
>>> time. If you look into AM335x address space, there's a bit in the
>>> wrapper which tells it to use the standard MUSB registers for IRQ,
>>> instead of the TI-specific thing. Can you see if we get a disconnect IRQ
>>> with that ?
>>>
>>> TRM is at [1], see page 2566. Basically, if you set bit 3 in register
>>> offset 0x1014, then it should use Mentor IRQ registers. If that works,
>>> quite a few problems will actually vanish :-p
>>
>> So I tried it with the following patch:
>>
>> -------------------------------------
>> diff --git a/drivers/usb/musb/musb_dsps.c b/drivers/usb/musb/musb_dsps.c
>> index c791ba5..c714500 100644
>> --- a/drivers/usb/musb/musb_dsps.c
>> +++ b/drivers/usb/musb/musb_dsps.c
>> @@ -470,6 +470,7 @@ static int dsps_musb_init(struct musb *musb)
>>
>> /* Reset the musb */
>> dsps_writel(reg_base, wrp->control, (1 << wrp->reset));
>> + dsps_writel(musb->ctrl_base, wrp->control, BIT(3));

overwritting reset.

>> musb->isr = dsps_interrupt;
>>
>> @@ -625,6 +626,7 @@ static int dsps_musb_reset(struct musb *musb)
>> if (session_restart || !glue->sw_babble_enabled) {
>> dev_info(musb->controller, "Restarting MUSB to recover from Babble\n");
>> dsps_writel(musb->ctrl_base, wrp->control, (1 << wrp->reset));
>> + dsps_writel(musb->ctrl_base, wrp->control, BIT(3));

here too. No wonder it won't work, right :-)

>> I am not very familiar with that driver but my understanding was that
>> the Mentor IRQ registeres are managed by the musb_interrupt() function
>> which is called from the dsps_interrupt() interrupt handler.
>>
>> Am I right?

yeah, however the way IRQs are reported is a different thing. AM335x
introduced its own IRQ reporting scheme which, basically, reads MUSB
generic IRQ status registers and reports on an AM335x specific
register. No idea why TI did that for AM335x devices.

>> if it is the case then it didn't fix the issue I had.
>>
>> I activated the following debug line:
>>
>> [musb_hdrc]musb_interrupt =_ "** IRQ %s usb%04x tx%04x rx%04x\012"
>> [musb_dsps]dsps_interrupt =p "usbintr (%x) epintr(%x)\012"
>>
>> But I didn't get any interrupt while disconnecting the cable without any
>> device connected on it (whereas I got an interrupt when I connected it).
>>
>> Note that I applied this patch instead of the "usb: musb: dsps: handle
>> the otg_state_a_wait_vrise_timeout case", is what you had in mind ?

yeah, that's what I had in mind. But your patch seems wrong :-)

I tried writing a more correct version here and found 2 issues:

a) bit 3 doesn't do anything :-p I cannot read IRQs from mentor's
registers

b) when setting RESET_ISOLATION bit, reads of CTRL register hang. Note
that according to TRM, RESET_ISOLATION _must_ be set prior to a soft
reset and cleared afterwards. But right after setting RESET_ISOLATION,
if I try a read of CTRL, it'll hang forever.

Bin, do you know about these problems ? They both seem rather alarming
to me.

--
balbi

Attachment: signature.asc
Description: PGP signature