Quoting Jeffrey Hugo (2019-12-19 06:20:23)
> On Wed, Dec 18, 2019 at 11:00 PM Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
> >
> > Quoting Jeffrey Hugo (2019-12-17 09:12:05)
> > > diff --git a/drivers/clk/qcom/gcc-msm8998.c b/drivers/clk/qcom/gcc-msm8998.c
> > > index df1d7056436c..26cc1458ce4a 100644
> > > --- a/drivers/clk/qcom/gcc-msm8998.c
> > > +++ b/drivers/clk/qcom/gcc-msm8998.c
> > > @@ -2044,6 +2044,7 @@ static struct clk_branch gcc_gpu_cfg_ahb_clk = {
> > > .hw.init = &(struct clk_init_data){
> > > .name = "gcc_gpu_cfg_ahb_clk",
> > > .ops = &clk_branch2_ops,
> > > + .flags = CLK_IS_CRITICAL, /* to access gpucc */
> >
> > Can we not do the thing that Bjorn did to turn on ahb clks with runtime
> > PM for clk controllers that need them? See 892df0191b29 ("clk: qcom: Add
> > QCS404 TuringCC").
> >
>
> Interesting. I didn't think of that solution, nor was I aware of that
> change. Let me have a look. Thanks for the tip.
The other option is to just always turn the clk on and leave it enabled
forever. I believe Bjorn had to use runtime PM because the clk would
turn off when the subsystem was reset. Maybe in the GPU case that isn't
true so we can just turn on the AHB and not try anything else. If that
works then I'd prefer that so we can save on code/data size.
