Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20211016145939.15643-1-martin.blumenstingl@googlemail.com>
 <1j5ytuvdmw.fsf@starbuckisacylon.baylibre.com> <CAFBinCBGZi3MRqTRshyCkq8AAaqHi2NkZVV80ppZr4Lx=xWOWA@mail.gmail.com>
 <1jmtn6tu99.fsf@starbuckisacylon.baylibre.com>
In-Reply-To: <1jmtn6tu99.fsf@starbuckisacylon.baylibre.com>
From:   Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Date:   Wed, 20 Oct 2021 20:16:29 +0200
Message-ID: <CAFBinCCRWS6j=6hDo_sOBmQZw5X8L4GYP=rHQVqHd04keCrbuQ@mail.gmail.com>
Subject: Re: [PATCH] clk: meson: gxbb: Add the spread spectrum bit for MPLL0
 on GXBB
To:     Jerome Brunet <jbrunet@baylibre.com>
Cc:     linux-amlogic@lists.infradead.org,
        Neil Armstrong <narmstrong@baylibre.com>,
        linux-clk@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-kernel@vger.kernel.org,
        Christian Hewitt <christianshewitt@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Jerome,

On Mon, Oct 18, 2021 at 2:03 PM Jerome Brunet <jbrunet@baylibre.com> wrote:
[...]
> > The difference here is BIT(14). un-setting BIT(14) (documented as
> > EN_DDS0) did not change anything according to Christian's test.
> > That also means that SDM, SDM_EN and N2 have the expected values.
> > I manually did the maths:
> > (2000000000Hz * 16384) / ((16384 * 6) + 12808) = 294909640.7Hz
> > which matches what clk_summary sees:
> > 294909641Hz
>
>  ... and (2000000000Hz * 16384) / ((16384 * 6) = 333MHz which is fairly close
>  to what you get w/o flipping the bit
This is actually a great hint. So far MPLL clocks have "just worked"
for me and I didn't have to work with this.
With your explanation it makes sense that SDM_EN makes the hardware
use or ignore the SDM value.

[...]
> For example yes. I am asking check a bit more what this bit does and
> what it does not:
>  - I need confirmation whether or not it does spread spectrum. Yes this
>  needs to be observed on a SoC pin, like MCLK with a fairly low divider
>  to the averaging effect which could partially mask spread spectrum.
I did some more tests with Christian. It turns out that on GXBB
HHI_MPLL_CNTL7[15] has no impact on the rate seen by meson-clk-msr.
On the other hand, HHI_MPLL_CNTL[25] makes MPLL0 use or ignore the SDM
value (again, verified through meson-clk-msr).

>  - Get an idea what it actually does. The 2 calculations above are an
>  hint. (Spread spectrum does not change the rate mean value)
Indeed!
My conclusion is that on GXBB:
1) HHI_MPLL_CNTL[25] doesn't control the spread spectrum setting of
MPLL0 - just like you thought
2) HHI_MPLL_CNTL[25] is actually SDM_EN (and HHI_MPLL_CNTL7[15]
doesn't seem to have any impact on MPLL0's output rate)

Please let me know if there's anything else we can test.
Else I'll send a patch for making HHI_MPLL_CNTL[25] the SDM_EN bit of
MPLL0 on GXBB.


Best regards,
Martin