Re: [EXT] Re: [PATCH v2 2/3] usb: chipidea: imx: support disabling runtime-pm

[Luca Ceresoli]

Hi Xu,

On Thu, 20 Jul 2023 10:13:57 +0000
Xu Yang <xu.yang_2@xxxxxxx> wrote:

> Hi Luca,
> 
> > > > -----Original Message-----
> > > >
> > > > Hello Xu,
> > > >
> > > > On Tue, 18 Jul 2023 08:31:48 +0000
> > > > Xu Yang <xu.yang_2@xxxxxxx> wrote:
> > > >  
> > > > > > -----Original Message-----
> > > > > >
> > > > > > Ciao Francesco,
> > > > > >
> > > > > > On Thu, 6 Jul 2023 12:23:43 +0200
> > > > > > Francesco Dolcini <francesco@xxxxxxxxxx> wrote:
> > > > > >  
> > > > > > > Hello Luca,
> > > > > > >
> > > > > > > On Tue, May 30, 2023 at 11:22:51AM +0000, Jun Li wrote:  
> > > > > > > > Yes, your understanding is correct, talked with Xu(in CC), he will take this
> > > > > > > > soon.  
> > > > > > >
> > > > > > > A series was posted
> > > > > > >
> > > > > > > I had no time to try or look at it yet.  
> > > > > >
> > > > > > Thanks for keeping me up to date on this topic, which is still totally
> > > > > > relevant to me.
> > > > > >
> > > > > > I looked at the series, but it does not seem to be addressing the
> > > > > > problem with USB host not detecting new devices when VBUS is not
> > > > > > directly connected, e.g. in the Colibri imx6ull SoM.
> > > > > >
> > > > > > Xu, do you confirm the series at the link is _not_ solving the problem
> > > > > > being discussed here?  
> > > > >
> > > > > Have you tried this patchset? The upstream driver couldn't get correct
> > > > > USB role from HW_USBPHY_CTRL register when the ID pin is float. This is
> > > > > what this patchset is trying to fix. With this patch, condition
> > > > > "(!vbus_is_on && !mxs_phy_is_otg_host(mxs_phy)" will always be false when
> > > > > controller acts as host role, then __mxs_phy_disconnect_line(phy, true)
> > > > > will never be called. So I think it doesn't matter whether VBUS is connected
> > > > > or not when act as host mode. If you still have issue after apply this patchset,
> > > > > please let me know.  
> > > >
> > > > I tested this patchset on top of v6.5-rc2 and I confirm USB detection
> > > > is still broken on the Colibri iMX6ULL. With or without the patches
> > > > the behavior is the same: USB devices are detected only during boot,
> > > > and anything connected after boot are never detected.  
> > >
> > > Thanks for your feedback. As you said this issue will disappear with below change, right?
> > >
> > >       diff --git a/drivers/usb/phy/phy-mxs-usb.c b/drivers/usb/phy/phy-mxs-usb.c
> > >       index e1a2b2ea098b..ec5ee790455e 100644
> > >       --- a/drivers/usb/phy/phy-mxs-usb.c
> > >       +++ b/drivers/usb/phy/phy-mxs-usb.c
> > >       @@ -178,7 +178,6 @@ static const struct mxs_phy_data imx6sx_phy_data = {
> > >        };
> > >
> > >        static const struct mxs_phy_data imx6ul_phy_data = {
> > >       -       .flags = MXS_PHY_DISCONNECT_LINE_WITHOUT_VBUS,
> > >        };
> > >
> > >        static const struct mxs_phy_data imx7ulp_phy_data = {  
> > 
> > Exactly.
> >   
> > > So I guess something in __mxs_phy_disconnect_line(mxs_phy, true) is causing this behavior.
> > > Could you please help to find which line to comment to make this issue disappear?  
> 
> To correct what I said:  __mxs_phy_disconnect_line(mxs_phy, false) should
> be called.
> 
> I think the enable wakeup sequence should be follow for host-only port:
> mxs_phy_set_wakeup(mxs_phy, true)
>     mxs_phy_disconnect_line(mxs_phy, true);
>         __mxs_phy_disconnect_line(mxs_phy, false);
> 
> And disable wakeup sequence:
> mxs_phy_set_wakeup(mxs_phy, false)
>     mxs_phy_disconnect_line(mxs_phy, false);
>         __mxs_phy_disconnect_line(mxs_phy, false);
> 
> So "bool variable disconnect is false" all the time.
> 
> > 
> > I did some tests and detection works by doing _any_ of the following
> > two changes (or both of them).
> > 
> > Change 1:
> > 
> > --- a/drivers/usb/phy/phy-mxs-usb.c
> > +++ b/drivers/usb/phy/phy-mxs-usb.c
> > @@ -359,10 +359,6 @@ static void __mxs_phy_disconnect_line(struct mxs_phy *mxs_phy, bool disconnect)
> >         void __iomem *base = mxs_phy->phy.io_priv;
> >         u32 reg;
> > 
> > -       if (disconnect)
> > -               writel_relaxed(BM_USBPHY_DEBUG_CLKGATE,
> > -                       base + HW_USBPHY_DEBUG_CLR);
> > -  
> 
> Since disconnect = false, this code didn't get executed all the time.
> Remove this will have no impact. But your test results didn't align
> to this. Could you please help check the sequence? Is disconnect
> true or false when __mxs_phy_disconnect_line is called?

What I observe is that __mxs_phy_disconnect_line(..., true) is called.
This happens during boot and after unplugging a device in case one was
detected during boot.

This is because in mxs_phy_disconnect_line() [0]:
 - on = 1
 - !vbus_is_on = 1
 - !mxs_phy_is_otg_host(mxs_phy) = 1

Which one(s) of those three would you expect to be 0?

Some additional info about the !mxs_phy_is_otg_host(mxs_phy) value:
with or without CONFIG_USB_OTG, it always has the same value because
phyctrl always has the BM_USBPHY_CTRL_OTG_ID_VALUE bit set.

[0]
https://elixir.bootlin.com/linux/v6.5-rc2/source/drivers/usb/phy/phy-mxs-usb.c#L415

Luca

-- 
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com