[PATCH 0/1] usb: dwc3: meson-g12a: fix shared reset control use

[Dan Robertson]

On Tue, Jul 14, 2020 at 08:57:35PM +0530, Anand Moon wrote:
> Hi Dan,
> 
> On Tue, 14 Jul 2020 at 19:00, Dan Robertson <dan at dlrobertson.com> wrote:
> >
> > Hi Anand!
> >
> > On Tue, Jul 14, 2020 at 12:26:57PM +0530, Anand Moon wrote:
> > > hi Dan,
> > >
> > > On Mon, 13 Jul 2020 at 21:37, Dan Robertson <dan at dlrobertson.com> wrote:
> > > >
> > > > When testing suspend for another driver I noticed the following warning:
> > > >
> > > > WARNING: CPU: 1 PID: 5530 at drivers/reset/core.c:355 reset_control_assert+0x184/0x19c
> > > > Hardware name: Hardkernel ODROID-N2 (DT)
> > > > [..]
> > > > pc : reset_control_assert+0x184/0x19c
> > > > lr : dwc3_meson_g12a_suspend+0x68/0x7c
> > > > [..]
> > > > Call trace:
> > > >  reset_control_assert+0x184/0x19c
> > > >  dwc3_meson_g12a_suspend+0x68/0x7c
> > > >  platform_pm_suspend+0x28/0x54
> > > >  __device_suspend+0x590/0xabc
> > > >  dpm_suspend+0x104/0x404
> > > >  dpm_suspend_start+0x84/0x1bc
> > > >  suspend_devices_and_enter+0xc4/0x4fc
> > > >
> > > > In my limited experience and knowlege it appears that we hit this
> > > > because the reset control was switched to shared and the the use
> > > > of the reset control was not changed.
> > > >
> > > > > * Calling reset_control_assert without first calling reset_control_deassert
> > > > > * is not allowed on a shared reset control. Calling reset_control_reset is
> > > > > * also not allowed on a shared reset control.
> > > >
> > > > The above snippet from reset_control_get_shared() seems to indicate that
> > > > this is due to the use of reset_control_reset() in dwc3_meson_g12a_probe()
> > > > and reset_control_deassert is not guaranteed to have been called
> > > > before dwc3_meson_g12a_suspend() and reset_control_assert().
> > > >
> > > > After some basic tests with the following patch I no longer hit the
> > > > warning. Comments and critiques on the patch are welcome. If there is a
> > > > reason for the current use of the reset control, I'd love to learn why!
> > > > Like I said before, I have not really looked at this driver before and
> > > > have verify limited experience with reset controls... Was working on
> > > > another driver, hit the warning, and thought I'd take a shot at the
> > > > fix :-)
> > > >
> > > Thanks, I was also looking into this issue
> >
> > Awesome!
> >
> > > So best Fix to this issue is to drop the call of reset_control_assert
> > > from dwc3_meson_g12a_suspend
> > > and drop the call of reset_control_deassert from dwc3_meson_g12a_resume
> > > With these changes we do not see the warning on suspend and resume
> >
> > We definitely would avoid hitting the warning without the
> > reset_control_(de)assert calls in suspend and resume. That is a valid use of
> > the reset control, but why would that be best?
> >
> > From reset_control_reset():
> 
> Before entering the suspend state the code tries to do following
>      clk_bulk_disable_unprepare
>      regulator_disable
>      phy_power_off
>      phy_exit
> 
> After this operation it's needless to call *reset_control_assert*
> I tried to move this call before all the above operations
> but still no success with this.

How so? Once the reset() is removed prope() and deassert() is guaranteed
to have been called before suspend, like what is in the patch and similar
to other uses of shared reset controllers, this is possible.

> Similarly on resume we should avoid calling resume
> *reset_control_deassert* earlier
> as the core is just reinitializing the devices.
>       clk_bulk_prepare_enable
>       usb_init
>       phy_init
>       phy_power_on
>       regulator_enable
> >
> > > * Consumers must not use reset_control_(de)assert on shared reset lines when
> > > * reset_control_reset has been used.
> >
> > If we use reset_control_reset() then we can not (de)assert the reset line
> > on suspend/resume or any other time. Wouldn't we want to be able to
> > (de)assert the reset line? And why do we no longer want to (de)assert the
> > reset line on suspend/resume?
> >
> > > > Can you try this attached patch?
> > >
> > I think I had already tested something similar. Removing the (de)assert calls
> > but keeping reset will definitely remove the warning, but it means we can not
> > (de)assert the line. My guess is that this is not what we want, but I could be
> > wrong. Thoughts, input, or references to documentation on this would be
> > appreciated.
> >
> 
> As per my knowledge suspend to mem will do complete power down of the
> devices with support suspend and resume feature sequentially and then it tries
> to bring the device up one by one.
> So it should also take care of reset lines as well.

So do we only _actually_ care about the reset line in the probe? Seems like we
should remove the reset controller from the structure if that is the case.

Cheers,

 - Dan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20200715/8d96e30a/attachment.sig>