Re: [PATCH 5/6] arm64: dts: Add ipq6018 SoC and CP01 board support

From: Christian Lamparter
Date: Thu Jun 20 2019 - 11:37:44 EST

Hello Sricharan,

On Wednesday, June 19, 2019 4:42:11 PM CEST Sricharan R wrote:
> On 6/15/2019 2:11 AM, Christian Lamparter wrote:
> > On Wednesday, June 12, 2019 11:48:48 AM CEST Sricharan R wrote:
> >> Hi Christian,
> >>
> >> On 6/10/2019 5:45 PM, Christian Lamparter wrote:
> >>> On Monday, June 10, 2019 12:09:56 PM CEST Sricharan R wrote:
> >>>> Hi Christian,
> >>>>
> >>>> On 6/6/2019 2:11 AM, Christian Lamparter wrote:
> >>>>> On Wed, Jun 5, 2019 at 7:16 PM Sricharan R <sricharan@xxxxxxxxxxxxxx> wrote:
> >>>>>>
> >>>>>> Add initial device tree support for the Qualcomm IPQ6018 SoC and
> >>>>>> CP01 evaluation board.
> >>>>>>
> >>>>>> Signed-off-by: Sricharan R <sricharan@xxxxxxxxxxxxxx>
> >>>>>> Signed-off-by: Abhishek Sahu <absahu@xxxxxxxxxxxxxx>
> >>>>>> --- /dev/null
> >>>>>> +++ b/arch/arm64/boot/dts/qcom/ipq6018.dtsi
> >>>>>>
> >>>>>> + clocks {
> >>>>>> + sleep_clk: sleep_clk {
> >>>>>> + compatible = "fixed-clock";
> >>>>>> + clock-frequency = <32000>;
> >>>>>> + #clock-cells = <0>;
> >>>>>> + };
> >>>>>> +
> >>>>> Recently-ish, we ran into an issue with the clock-frequency of the sleep_clk
> >>>>> on older IPQ40XX (and IPQ806x) on the OpenWrt Github and ML.
> >>>>> From what I know, the external "32KHz" crystals have 32768 Hz, but the QSDK
> >>>>> declares them at 32000 Hz. Since you probably have access to the BOM and
> >>>>> datasheets. Can you please confirm what's the real clock frequency for
> >>>>> the IPQ6018.
> >>>>> (And maybe also for the sleep_clk of the IPQ4018 as well?).
> >>>>>
> >>>>
> >>>> What exactly is the issue that you faced ?
> >>>> Looking in to the docs, it is <32000> only on ipq6018 and ipq40xx as well.
> >>>
> >>> We need just a confirmation.
> >>>
> >>> Then again, Currently the qcom-ipq4019.dtsi is using 32768 Hz.
> >>>
> >>> | sleep_clk: sleep_clk {
> >>> | compatible = "fixed-clock";
> >>> | clock-frequency = <32768>;
> >>> | #clock-cells = <0>;
> >>> | };
> >>>
> >>> <https://github.com/torvalds/linux/blob/master/arch/arm/boot/dts/qcom-ipq4019.dtsi#L144>
> >>>
> >>> Which makes sense, because all previous Qualcomm Atheros MIPS and the
> >>> future IPQ8072 SoCs have been either using or deriving a 32768 Hz clock.
> >>>
> >>> For example: The AR9344 derives the clock from the 25MHz/40MHz external
> >>> oscillator. This is explained in "8.16.9 Derived RTC Clock (DERIVED_RTC_CLK)".
> >>> Which mentions that the "32KHz" clock interval is 30.5 usec / 30.48 usec
> >>> depending whenever the external reference crystal has 40MHz or 25MHz.
> >>> (1/30.5usec = 32.7868852 kilohertz!). The QCA9558 datasheet says the same
> >>> in "10.19.11 Derived RTC Clock".
> >>>
> >>> For IPQ8072: I point to the post by Sven Eckelmann on the OpenWrt ML:
> >>> <http://lists.infradead.org/pipermail/openwrt-devel/2019-May/017131.html>
> >>> "I was only able to verify for IPQ8072 that it had a 32.768 KHz
> >>> sleep clock."
> >>>
> >>> So this is pretty much "why there is an issue", it's confusing.
> >>> Is possible can you please look if there are (fixed) divisors values
> >>> listed in the documentation or the registers and bits that the values
> >>> are stored in? Because then we could just calculate it.
> >>>
> >>
> >> Really sorry for the confusion. So looking little more, SLEEP_CLK is derived
> >> from an external 38.4MHZ crystal, it is 32.768 KHZ.
> > That's really valuable information to have. Thank you!
> >
> >> Somehow the clk freq plan etc seems to mention them only as .032 MHZ and misses
> >> out. That means i will correct the patch for 32768 and probably the
> >> ipq8074.dtsi as well
> >
> > Ok, there's one more issue that Paul found (at least with the IPQ4019),
> > https://patchwork.ozlabs.org/patch/1099482
> >
> > it seems that the "sleep_clk" node in the qcom-ipq4019.dtsi is not used by
> > the gcc-ipq4019.c clk driver. this causes both wifi rtc_clks and the usb sleep
> > clks to dangle in the /sys/kernel/debug/clk/clk_summary (from a RT-AC58U)
> >
> > clock enable_cnt prepare_cnt rate accuracy phase
> > ----------------------------------------------------------------------------------------
> > xo 9 9 48000000 0 0
> > [...]
> > sleep_clk 1 1 32768 0 0
> > gcc_wcss5g_rtc_clk 1 1 0 0 0
> > gcc_wcss2g_rtc_clk 1 1 0 0 0
> > gcc_usb3_sleep_clk 1 1 0 0 0
> > gcc_usb2_sleep_clk 1 1 0 0 0
> >
> > with his patch the /sys/kernel/debug/clk/clk_summary looks "better"
> >
> > (something like this:)
> >
> > clock enable_cnt prepare_cnt rate accuracy phase
> > ----------------------------------------------------------------------------------------
> > xo 9 9 48000000 0 0
> > [...]
> > gcc_sleep_clk_src 5 5 32000 0 0
> > gcc_wcss5g_rtc_clk 1 1 32000 0 0
> > gcc_wcss2g_rtc_clk 1 1 32000 0 0
> > gcc_usb3_sleep_clk 1 1 32000 0 0
> > gcc_usb2_sleep_clk 1 1 32000 0 0
> >
> > but judging from your comment "SLEEP_CLK is derived from an
> > external 38.4MHZ crystal" the gcc_sleep_clk_src / sleep_clk
> > should have xo as the parent. so the ideal output should be:
> >
> > clock enable_cnt prepare_cnt rate accuracy phase
> > ----------------------------------------------------------------------------------------
> > xo 10 10 48000000 0 0
> > [...]
> > gcc_sleep_clk 5 5 32768 0 0
> > gcc_wcss5g_rtc_clk 1 1 32768 0 0
> > gcc_wcss2g_rtc_clk 1 1 32768 0 0
> > gcc_usb3_sleep_clk 1 1 32768 0 0
> > gcc_usb2_sleep_clk 1 1 32768 0 0
> >
> > or am I missing/skipping over something important?
> >
>
> Sorry for the delayed response. So what i said above (32768 clk) looks
> like true only for ipq8074. For ipq4019, looks like 32000.
>
> That means, there is still some thing unclear. I am checking for precise
> information from HW team for ipq4019/8074/6018. Please hang on, will
> update you asap.

Thank you for looking this up! I'll definitely stick around for the final
verdict.

Also, I think the "xo" clk of your IPQ6018 dts should get the
"always-on;" property (any maybe sleep_clk as well?).

Paul discovered that the QSDK had this extra commit
<https://lore.kernel.org/patchwork/patch/1089385/>
(Maybe the changeid can help you look it up internally)

For IPQ4019, this enables the high resolution with a 1ns resolution
instead of 10ms.

(echo q > /proc/sysrq-trigger can be used to check this just look for
the "resolution" value before and after.)

Cheers,
Christian