Re: [PATCH v1] arm64: dts: imx8mm: fix missing pgc_vpu_* power domain parent

From: Vitor Soares
Date: Wed Apr 10 2024 - 07:01:54 EST


Hi Lucas,

On Tue, 2024-04-09 at 17:44 +0100, Vitor Soares wrote:
> On Tue, 2024-04-09 at 16:36 +0200, Lucas Stach wrote:
> > Am Dienstag, dem 09.04.2024 um 14:22 +0100 schrieb Vitor Soares:
> > > Hi Lucas,
> > >
> > > Thanks for your feedback.
> > >
> > > On Tue, 2024-04-09 at 11:13 +0200, Lucas Stach wrote:
> > > > Hi Vitor,
> > > >
> > > > Am Dienstag, dem 09.04.2024 um 09:58 +0100 schrieb Vitor
> > > > Soares:
> > > > > From: Vitor Soares <vitor.soares@xxxxxxxxxxx>
> > > > >
> > > > > The pgc_vpu_* nodes miss the reference to the power domain
> > > > > parent,
> > > > > leading the system to hang during the resume.
> > > > >
> > > > This change is not correct. The vpumix domain is controlled
> > > > through
> > > > the
> > > > imx8mm-vpu-blk-ctrl and must not be directly triggered by the
> > > > child
> > > > domains in order to guarantee proper power sequencing.
> > > >
> > > > If the sequencing is incorrect for resume, it needs to be fixed
> > > > in
> > > > the
> > > > blk-ctrl driver. I'll happily assist if you have any questions
> > > > about
> > > > this intricate mix between GPC and blk-ctrl hardware/drivers.
> > >  
> > > I'm new into the topic, so I tried to follow same approach as in
> > > imx8mp
> > > DT.
> > >
> > That's a good hint, the 8MP VPU GPC node additions missed my radar.
> > The
> > direct dependency there between the GPC domains is equally wrong.
> >
> > > I also checked the imx8mq DT and it only have one domain for the
> > > VPU in the GPC. It seem blk-ctrl also dependes on pgc_vpu_* to
> > > work
> > > properly.
> > >
> > > The blk-ctrl driver hangs on imx8m_blk_ctrl_power_on() when
> > > access
> > > the
> > > ip registers for the soft reset. I tried to power-up the before
> > > the
> > > soft reset, but it didn't work.
> > >
> > The runtime_pm_get_sync() at the start of that function should
> > ensure
> > that bus GPC domain aka vpumix is powered up. Can you check if that
> > is
> > happening?
>
> I checked bc->bus_power_dev->power.runtime_status and it is
> RPM_ACTIVE.
>
> Am I looking to on the right thing? It is RPM_ACTIVE event before
> runtime_pm_get_sync().

During the probe I can see that
bus_power_dev->power.runtime_status = RPM_SUSPENDED and then vpumix is
powered up on GPC driver.

On resume routine I can't see this flow. bus_power_dev-
>power.runtime_status = RPM_ACTIVE and vpumix end up not being powered-
up.

I checked the suspend flow and the GPC tries to poweroff vpumix.


Best regards,
Vitor Soares

>
>
> >
> > Regards,
> > Lucas
> >
> > > Do you have an idea how we can address this within blk-ctrl?
> > >
> > > Best regards,
> > > Vitor
>