On Fri, Sep 30, 2022 at 5:04 PM AngeloGioacchino Del Regno
<angelogioacchino.delregno@xxxxxxxxxxxxx> wrote:
>
> Il 30/09/22 11:02, Chen-Yu Tsai ha scritto:
> > On Fri, Sep 30, 2022 at 4:58 PM AngeloGioacchino Del Regno
> > <angelogioacchino.delregno@xxxxxxxxxxxxx> wrote:
> >>
> >> Il 30/09/22 10:44, Chen-Yu Tsai ha scritto:
> >>> On Fri, Sep 30, 2022 at 4:29 PM AngeloGioacchino Del Regno
> >>> <angelogioacchino.delregno@xxxxxxxxxxxxx> wrote:
> >>>>
> >>>> Il 30/09/22 07:59, MandyJH Liu (劉人僖) ha scritto:
> >>>>> On Tue, 2022-09-27 at 12:11 +0200, AngeloGioacchino Del Regno wrote:
> >>>>>> These PLLs are conflicting with GPU rates that can be generated by
> >>>>>> the GPU-dedicated MFGPLL and would require a special clock handler
> >>>>>> to be used, for very little and ignorable power consumption benefits.
> >>>>>> Also, we're in any case unable to set the rate of these PLLs to
> >>>>>> something else that is sensible for this task, so simply drop them:
> >>>>>> this will make the GPU to be clocked exclusively from MFGPLL for
> >>>>>> "fast" rates, while still achieving the right "safe" rate during
> >>>>>> PLL frequency locking.
> >>>>>>
> >>>>>> Signed-off-by: AngeloGioacchino Del Regno <
> >>>>>> angelogioacchino.delregno@xxxxxxxxxxxxx>
> >>>>>> Reviewed-by: Chen-Yu Tsai <wenst@xxxxxxxxxxxx>
> >>>>>> ---
> >>>>>> drivers/clk/mediatek/clk-mt8195-topckgen.c | 9 ++++++---
> >>>>>> 1 file changed, 6 insertions(+), 3 deletions(-)
> >>>>>>
> >>>>>> diff --git a/drivers/clk/mediatek/clk-mt8195-topckgen.c
> >>>>>> b/drivers/clk/mediatek/clk-mt8195-topckgen.c
> >>>>>> index 4dde23bece66..8cbab5ca2e58 100644
> >>>>>> --- a/drivers/clk/mediatek/clk-mt8195-topckgen.c
> >>>>>> +++ b/drivers/clk/mediatek/clk-mt8195-topckgen.c
> >>>>>> @@ -298,11 +298,14 @@ static const char * const ipu_if_parents[] = {
> >>>>>> "mmpll_d4"
> >>>>>> };
> >>>>>>
> >>>>>> +/*
> >>>>>> + * MFG can be also parented to "univpll_d6" and "univpll_d7":
> >>>>>> + * these have been removed from the parents list to let us
> >>>>>> + * achieve GPU DVFS without any special clock handlers.
> >>>>>> + */
> >>>>>> static const char * const mfg_parents[] = {
> >>>>>> "clk26m",
> >>>>>> - "mainpll_d5_d2",
> >>>>>> - "univpll_d6",
> >>>>>> - "univpll_d7"
> >>>>>> + "mainpll_d5_d2"
> >>>>>> };
> >>>>>>
> >>>>>> static const char * const camtg_parents[] = {
> >>>>> There might be a problem here. Since the univpll_d6 and univpll_d7 are
> >>>>> available parents in hardware design and they can be selected other
> >>>>> than kernel stage, like bootloader, the clk tree listed in clk_summary
> >>>>> cannot show the real parent-child relationship in such case.
> >>>>
> >>>> I agree about that, but the clock framework will change the parent to
> >>>> the "best parent" in that case... this was done to avoid writing complicated
> >>>> custom clock ops just for that one.
> >>>>
> >>>> This issue is present only on MT8195, so it can be safely solved this way,
> >>>> at least for now.
> >>>>
> >>>> Should this become a thing on another couple SoCs, it'll then make sense
> >>>> to write custom clock ops just for the MFG.
> >>>
> >>> Would CLK_SET_RATE_NO_REPARENT on the fast mux coupled with forcing
> >>> the clk tree to a state that we like (mfgpll->fast_mux->gate) work?
> >>
> >> I'm not sure that it would, and then this would mean that we'd have to add
> >> assigned-clock-parents to the devicetree and the day we will introduce the
> >> "complicated custom clock ops" for that, we'll most probably have to change
> >> the devicetree as well... which is something that I'm a bit reluctant to do
> >> as a kernel upgrade doesn't automatically mean that you upgrade the DT with
> >> it to get the "new full functionality".
> >
> > You can also do it by doing clk_set_parent() in the clock driver after the
> > clocks are registered, or just write to the register before the clock is
> > registered.
> >
>
> I honestly don't like doing that - but I can try if that works and, if it does,
> I can send a commit with a Fixes tag later, perhaps?
Sounds good to me.
FWIW, I think it's OK for drivers to reinitialize hardware to a known
state that it can work with. As long as it doesn't break the system
while doing so.
ChenYu
>
> > We do the latter in some of the sunxi-ng drivers, though IIRC it was to
> > force a certain divider on what we expose as a fixed divider clock.
> >
> > ChenYu
> >
> >> Introducing the new clock ops for the mfg mux is something that will happen
> >> for sure, but if we don't get new SoCs with a similar "issue", I don't feel
> >> confident to write them, as I fear these won't be as flexible as needed and
> >> will eventually need a rewrite; that's why I want to wait to get the same
> >> situation on "something new".
> >>
> >> In my opinion, it is safe to keep this change as it is, even though I do
> >> understand the shown concerns about the eventual unability to show the tree
> >> relationship in case the bootloader chooses to initialize the mfg mux with
> >> a univpll parent.
> >>
> >> Regards,
> >> Angelo
> >>
>
>
>
