RE: [PATCH 5/6] clk: imx8mq: Remove CLK_IS_CRITICAL flag for IMX8MQ_CLK_TMU_ROOT

[Anson Huang <anson.huang@nxp.com>]

Hi, Abel/Daniel

> On 19-07-27 09:33:10, Daniel Baluta wrote:
> > On Sat, Jul 27, 2019 at 9:19 AM Anson Huang <anson.huang@nxp.com>
> wrote:
> > >
> > > Hi, Daniel
> > >
> > > > Subject: Re: [PATCH 5/6] clk: imx8mq: Remove CLK_IS_CRITICAL flag
> > > > for IMX8MQ_CLK_TMU_ROOT
> > > >
> > > > Hi all,
> > > >
> > > > latest linux-next hangs at boot.
> > > >
> > > > commit fde50b96be821ac9673a7e00847cc4605bd88f34 (HEAD ->
> master, tag:
> > > > next-20190726, origin/master, origin/HEAD)
> > > > Author: Stephen Rothwell <sfr@canb.auug.org.au>
> > > > Date:   Fri Jul 26 15:18:02 2019 +1000
> > > >
> > > >     Add linux-next specific files for 20190726
> > > >
> > > >     Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
> > > >
> > > >
> > > > I know this is crazy but reverting commit:
> > > >
> > > > commit 431bdd1df48ee2896ea9980d9153e3aeaf0c81ef
> (refs/bisect/bad)
> > > > Author: Anson Huang <Anson.Huang@nxp.com>
> > > > Date:   Fri Jul 5 12:56:11 2019 +0800
> > > >
> > > >     clk: imx8mq: Remove CLK_IS_CRITICAL flag for
> > > > IMX8MQ_CLK_TMU_ROOT
> > > >
> > > >     IMX8MQ_CLK_TMU_ROOT is ONLY used for thermal module, the
> driver
> > > >     should manage this clock, so no need to have CLK_IS_CRITICAL flag
> > > >     set.
> > > >
> > > >
> > > >
> > > > makes the boot work again.
> > > >
> > > > Any idea?
> > >
> > > I just found if disabling SDMA1, then kernel can boot up, it does
> > > NOT make sense TMU clock is related to SDMA1, I will check with design
> and get back to you soon.
> > >
> >
> > Hi Anson,
> >
> > Applying Abel's patch:
> >
> > commit 8816c47db6a82f55bb4d64f62fd9dd3af680f0e4 (HEAD -> master)
> > Author: Abel Vesa <abel.vesa@nxp.com>
> > Date:   Tue Jun 25 12:01:56 2019 +0300
> >
> >     clk: imx8mq: Mark AHB clock as critical
> >
> >     Keep the AHB clock always on since there is no driver to control it and
> >     all the other clocks that use it as parent rely on it being always enabled.
> >
> >
> >
> > The kernel boots up again.
> >
> > It make some sense. I don't understand though why having
> > IMX8MQ_CLK_TMU_ROOT as critical also "unhangs" the kernel.
> >
> 
> OK, so this is how it works.
> 
> By removing the critical flag from TMU, the AHB doesn't stay always on.
> With my patch the AHB is marked as critical and therefore stays on.
> 
> The sdma1_clk has as parent the ipg_root which in turn has as parent the
> ahb clock. And I think what happens is some read from the sdma registers
> hangs because, for whatever reason, enabling the sdma1_clk doesn't
> propagate to enable the ahb clock. I might be wrong though.
> 

I can explain why Abel's patch can fix this issue, the AHB clock is a MUST
always ON for system bus, while it does NOT have CLK_IS_CRITICAL flag set for now,
when SDMA1 is probed, it will enable its clock, and the clk path is sdma1_clk->ipg_root->ahb,
after SDMA1 probed done, it will disable its clock since no one use it, and it will trigger
the ahb clock to be OFF, as its usecount is added by 1 when probe and decreased by 1 after
SDMA1 probe done, so usecount=0 will trigger AHB clock to be OFF.

So I think the best solution should be applying Abel's patch as you mentioned upper, the TMU
clock patch is NOT the root cause, it just triggers this issue accidently☹

But I saw Abel's AHB patch is still under discussion, so I think we need to speed it up and make 
kernel boot up work for development. AHB/IPG are always critical for i.MX SoCs.

Thanks,
Anson.