Re: [PATCH v9 01/10] clk: fix initial state of critical clock's parents

[James Liao]

Hi Mike,

On Fri, 2016-07-08 at 16:32 -0700, Michael Turquette wrote:
> Hi James,
> 
> Quoting James Liao (2016-07-03 20:51:48)
> > On Fri, 2016-07-01 at 18:21 -0700, Stephen Boyd wrote:
> > > (Resending to everyone)
> > > 
> > > On 06/22, Erin Lo wrote:
> > > > From: James Liao <jamesjj.liao@xxxxxxxxxxxx>
> > > > 
> > > > This patch fixed wrong state of parent clocks if they are registered
> > > > after critical clocks.
> > > > 
> > > > Signed-off-by: James Liao <jamesjj.liao@xxxxxxxxxxxx>
> > > > Signed-off-by: Erin Lo <erin.lo@xxxxxxxxxxxx>
> > > 
> > > It would be nice if you included the information about the
> > > problem from James' previous mail. This says what it does, but
> > > doesn't explain what the problem is and how it is fixing it.
> > > 
> > > > ---
> > > >  drivers/clk/clk.c | 9 ++++++++-
> > > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> > > > index d584004..e9f5f89 100644
> > > > --- a/drivers/clk/clk.c
> > > > +++ b/drivers/clk/clk.c
> > > > @@ -2388,8 +2388,15 @@ static int __clk_core_init(struct clk_core *core)
> > > >     hlist_for_each_entry_safe(orphan, tmp2, &clk_orphan_list, child_node) {
> > > >             struct clk_core *parent = __clk_init_parent(orphan);
> > > >  
> > > > -           if (parent)
> > > > +           if (parent) {
> > > >                     clk_core_reparent(orphan, parent);
> > > > +
> > > > +                   if (orphan->prepare_count)
> > > > +                           clk_core_prepare(parent);
> > > > +
> > > > +                   if (orphan->enable_count)
> > > > +                           clk_core_enable(parent);
> > > > +           }
> > > >     }
> > > 
> > > I'm pretty sure I pointed this problem out to Mike when the
> > > critical clk patches were being pushed. I can't recall what the
> > > plan was though to fix the problem. I'm pretty sure he said that
> > > clk_core_reparent() would take care of it, but obviously it is
> > > not doing that. Or perhaps it was that clk handoff should figure
> > > out that the parents of a critical clk are also on and thus keep
> > > them on.
> > 
> > Hi Mike
> > 
> > Is there any other patch to fix this issue? Or did I misuse critical
> > clock flag?
> 
> There is no fix yes. Your fix is basically correct. I was mistaken back
> when I told you and Stephen that the framework already took care of
> this.
> 
> However, instead of "open coding" this solution, I would rather re-use
> the __clk_set_parent_{before,after} helpers instead. Can you review/test
> the following patch and let me know what you think?
> 
> Thanks,
> Mike
> 
> 
> 
> From c0163b3f719b1e219b28ad425f94f9ef54a25a8f Mon Sep 17 00:00:00 2001
> From: Michael Turquette <mturquette@xxxxxxxxxxxx>
> Date: Fri, 8 Jul 2016 16:05:22 -0700
> Subject: [PATCH] clk: migrate ref counts when orphans are reunited
> 
> It's always nice to see families reunited, and this is equally true when
> talking about parent clocks and their children. However, if the orphan
> clk had a positive prepare_count or enable_count, then we would not
> migrate those counts up the parent chain correctly.
> 
> This has manifested with the recent critical clocks feature, which often
> enables clocks very early, before their parents have been registered.
> 
> Fixed by replacing the call to clk_core_reparent with calls to
> __clk_set_parent_{before,after}.
> 
> Cc: James Liao <jamesjj.liao@xxxxxxxxxxxx>
> Cc: Erin Lo <erin.lo@xxxxxxxxxxxx>
> Signed-off-by: Michael Turquette <mturquette@xxxxxxxxxxxx>
> ---
>  drivers/clk/clk.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> index 820a939fb6bb..70efe4c4e0cc 100644
> --- a/drivers/clk/clk.c
> +++ b/drivers/clk/clk.c
> @@ -2449,8 +2449,14 @@ static int __clk_core_init(struct clk_core *core)
>  	hlist_for_each_entry_safe(orphan, tmp2, &clk_orphan_list, child_node) {
>  		struct clk_core *parent = __clk_init_parent(orphan);
>  
> -		if (parent)
> -			clk_core_reparent(orphan, parent);

Is it correct to remove clk_core_reparent()? It lacks
__clk_recalc_accuracies() and __clk_recalc_rates(), so the new parent's
rate will not propagate correctly.

For example, I set vdec_sel as a critical clock. Without your patch, the
result was:

    vdecpll         0            0   338000000
       vdecpll_ck   1            1   338000000
          vdec_sel  1            1   338000000

With your patch, it became:

    vdecpll         1            1   338000000
       vdecpll_ck   1            1           0
          vdec_sel  1            1           0

The prepare_count and enable_count are correct with your patch, but the
rates of vdecpll_ck and vdec_sel become incorrect.


Best regards,

James

> +		/*
> +		 * we could call __clk_set_parent, but that would result in a
> +		 * reducant call to the .set_rate op, if it exists
> +		 */
> +		if (parent) {
> +			__clk_set_parent_before(orphan, parent);
> +			__clk_set_parent_after(orphan, parent, NULL);
> +		}
>  	}
>  
>  	/*