Re: [PATCH v3 2/7] clk: rockchip: rk3399: export 480M_SRC clock id for usbphy0/usbphy1

[Frank Wang]

Hi Heiko,

On 2016/8/8 17:55, Frank Wang wrote:
> Hi Heiko,
>
> On 2016/8/6 0:05, Heiko StÃbner wrote:
> > Hi Frank,
> >
> > Am Freitag, 5. August 2016, 16:34:42 schrieb Frank Wang:
> > > On 2016/8/5 3:10, Heiko StÃbner wrote:
> > > > Am Dienstag, 2. August 2016, 15:19:56 schrieb Xing Zheng:
> > > > > Export these source clocks for usbphy.
> > > > >
> > > > > Signed-off-by: Xing Zheng <zhengxing@xxxxxxxxxxxxxx>
> > > > can you please provide a rationale why you need manual control over 
> > > > that
> > > > intermediate clock?
> > > Well, From below graph, you can see that 'clk_usbphyX_480m' is 
> > > generated
> > > from usb2phy, and 'clk_usbphy_480m' which select from
> > > clk_usbphyX_480m_src via a gate (G13[12])  provided 480M clock to other
> > > modules.
> > >
> > >     xin24m
> > >       |__ clk_usb2phy0_ref
> > >       |     |
> > >       |     |__ clk_usbphy0_480m
> > >       |           |
> > >       |           |__clk_usbphy0_480m_src
> > >       |                |
> > >       |                |__clk_usbphy_480m
> > >       |                     |__ ... ...
> > >       |
> > >       |__ clk_usb2phy1_ref
> > >             |
> > >             |__ clk_usbphy1_480m
> > >                   |
> > >                   |__clk_usbphy1_480m_src
> > > > The two usbphys seem to use the clk_usb2phyX_ref clocks, generate the
> > > > 480m
> > > > clocks, but do not seem to need the clk_usbphyX_480m_src gates.
> > > Yeah, they used to be. However, the story went something like this,
> > >
> > > Some PM suspend process related ehci/ohci controller are base on 480m
> > > clocks, unfortunately, usb2-phy suspended earlier than ehci/ohci
> > > (usb2-phy will be auto suspended if no devices plug-in), and the
> > > clk-480m provided by it was disabled if no module used. As a result, 
> > > the
> > > PM suspend process was blocked when it run into ehci/ohci module.
> > ah, so the ehci controller needs that 480m clock as well? Do you 
> > happen to
> > have example patches for the ehci/ohci side already? I'd like to peak 
> > at what
> > you mean with "some PM suspend process related" things.
>
> Actually, no patches for it, I just make below steps manually :-).
> 1. set two usb2-phy into suspend mode.
> 2. disable 480m clock on each usb2-phy (assume only usb2-phy used it).
> 3. press power button let system into PM suspend.
>
> Then, the kernel will be blocked and you can see the following log 
> from console.
> ... ....
> [  123.763848] calling  usb6+ @ 166, parent: xhci-hcd.0.auto
> [  123.764503] call usb6+ returned 0 after 163 usecs
> [  123.765106] calling  usb5+ @ 166, parent: xhci-hcd.0.auto
> [  123.765719] call usb5+ returned 0 after 121 usecs
> [  123.766294] calling  usb4+ @ 166, parent: fe3e0000.usb
> [  123.766917] calling  usb3+ @ 55, parent: fe3a0000.usb

May I know have you tried reproducing on your board or not?

> > Depending on what is actually needed, you could also pull the usbphy 
> > out of
> > autosuspend in a pm-prepare callback of the phy driver itself ... see
> > http://lxr.free-electrons.com/source/include/linux/pm.h#L86
> >
> > Like
> > - in the .prepare callback make sure to unsuspend the phy
> >    and deactivate the autosuspend
> > - ehci/ohci will poweroff the phy in it s suspend callback (already 
> > does that)
>
> Hmm, do you remember that we have previously discussed there are some 
> oddities in ehci/ohci driver? phy_power_on() gets called twice at 
> ehci/ohci driver probe time, one is at pdata->power_on(); another is 
> at usb_add_hcd(), then the power_count of phy increases to 2,  but 
> phy_power_off() is just invoked one time when ehci/ohci goes to PM 
> suspend, so phy->ops->power_off is never be invoked.
>
> In this way,  the usb-phy maybe never go to suspend.
>
> > - suspend -> resume
> > - ehci/ohci will poweron the phy
> > - in the phy's .complete callback you can reactivate the autosuspend 
> > timer
> >
> > Because it looks more like you actually need the phy and not the 
> > clock alone.
> > So it would be nicer to use mechanisms already in place instead of 
> > creating
> > new dependencies.
>
> Theoretically, phy_init() will be invoked when ehci/ohci power on, and 
> the sm_work will be reactivated (have already implemented) in 
> phy->ops->init, but unfortunately, the same issue as phy_power_on() 
> mentioned above, it never run there too .

Would you like to give some comments on above two oddities please?

BR.
Frank

> > > Hence, we are planing to refer clk_usbphyX_480m_src into each ehci/ohci
> > > driver. Maybe you will challenge why not refer clk_usbphy_480m 
> > > directly?
> > > because there are two ehci/ohci connected in the different usb2phy, and
> > > only one clk_usbphy_480m clock was selected in clock tree.
> > Nope, no argument from me as I fully understand that each phy 
> > provides its own
> > 480m clock :-) .
> >
> >
> > Heiko