MUSB dual-role on AM335x behaving weirdly

[Maxime Ripard]

Hi Markus,

On Thu, Jan 22, 2015 at 12:01:13PM +0100, Markus Pargmann wrote:
> Hi,
> 
> On Thu, Jan 22, 2015 at 11:43:30AM +0100, Maxime Ripard wrote:
> > Hi Yegor,
> > 
> > On Thu, Jan 22, 2015 at 08:37:45AM +0100, Yegor Yefremov wrote:
> > > On 21.01.2015 19:53, Bin Liu wrote:
> > > > Hi,
> > > > 
> > > > On Wed, Jan 21, 2015 at 10:06 AM, Maxime Ripard
> > > > <maxime.ripard at free-electrons.com> wrote:
> > > >> Hi Felipe,
> > > >>
> > > >> I'm currently working on a custom AM335x-based board, that has a OTG
> > > >> connector wired to one of the musb controlers, and Linux 3.17
> > > >>
> > > >> This OTG connector seems to behave in a weird way when it comes to
> > > >> switching from one role to another:
> > > >>
> > > >>   - The host mode works with CONFIG_USB_MUSB_DUAL_ROLE set, only if
> > > >>     the dr_mode is set to host in the DT, and only if we plug the USB
> > > >>     device after the kernel has booted. Otherwise, even though the
> > > >>     controller seems to be set up as host in the kernel logs, it
> > > >>     doesn't work. When DR-mode is set to otg, DRVVBUS is not high
> > > >>     which surely doesn't help, but there might be some other issues.
> > > >>
> > > >>   - The gadget mode works with CONFIG_USB_MUSB_DUAL_ROLE, if dr_mode
> > > >>     is set to otg and not host (obviously), and only if we plug the
> > > >>     USB cable after the kernel has booted.
> > > >>
> > > >>   - The gadget mode works with CONFIG_USB_MUSB_GADGET, and with
> > > >>     dr_mode set to otg, even if we cold plug the device.
> > > >>
> > > >>   - As you might expect from the two first items, the runtime
> > > >>     switching from gadget to host when CONFIG_USB_MUSB_DUAL_ROLE is
> > > >>     set and dr_mode is set in otg doesn't work either.
> > > >>
> > > >> It looks like it's only waiting for an interrupt to occur to read the
> > > >> ID pin state, without reading its initial value, which would explain
> > > >> why both the host and gadget enumerate only when a device/cable is
> > > >> hotplugged, and that there's some configuration bits that differ
> > > >> whenever dr_mode changes, but I couldn't really see anything standing
> > > >> out in the driver nor the datasheet.
> > > >>
> > > >> Have you already experienced something alike with that driver?
> > > > 
> > > > I don't use vanilla mainline kernel very often, but I don't have such
> > > > hotplug issues with TI 3.12/14 kernels [1]. I only use
> > > > CONFIG_USB_MUSB_DUAL_ROLE though.
> > > > 
> > > > [1] git://git.ti.com/ti-linux-kernel/ti-linux-kernel.git, branch
> > > > remote/origin/linux-3.14.y
> > > 
> > > I have the same experience with 3.15. The switching is working when
> > > CONFIG_USB_MUSB_DUAL_ROLE is set and dr_mode = "otg". But since 3.16
> > > it seems to be broken. Still had no time to bisect this.
> > 
> > I just gave 3.15 a quick try, and it looks like it can now detect when
> > the device is acting as a gadget, but the host mode is not working at
> > all.
> > 
> > That's a good lead though, I'll dig more into the older versions and
> > see if some are working better than others.
> 
> This issue may be related to the fact that the musb controller does not
> have an ID pin interrupt. Last year I sent a RFC for a soft detection in
> OTG mode. It has a poll loop which toggles the SESSION bit of the
> controller to detect devices as host and drops back to device otherwise.
> This is far away from a good solution.
> 
> http://lists.infradead.org/pipermail/linux-arm-kernel/2013-October/204519.html

I guess it was a combination of two issues: after giving your patch a
try, I found out that (even without your patch), in OTG mode, the
gadget device is properly attached even if plugged before the kernel
boots. I don't really know how I missed that, but anyway...

I also tested to force the ID pin to the ground when a USB device was
plugged in, and it then enumerates properly, both when cold and hot
plugging. It's a bit weird because the adapter I'm using is working
fine on other boards, but maybe the AM335x controller is a bit more
picky with the ID pin, or that the other SoC I was testing it on
(Atmel's sama5d4) is quite relaxed about it.

I'm guessing that when forced in the host mode, the driver ignores
completely the ID pin, hence why it was working properly with that
setup.

Thanks a lot,
Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150122/078ecda9/attachment.sig>