diff mbox series

arm64: dts: rockchip: fix rk3399-puma eMMC HS400 signal integrity

Message ID 20220119134948.1444965-1-quentin.schulz@theobroma-systems.com (mailing list archive)
State New, archived
Headers show
Series arm64: dts: rockchip: fix rk3399-puma eMMC HS400 signal integrity | expand

Commit Message

Quentin Schulz Jan. 19, 2022, 1:49 p.m. UTC 
From: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>

There are signal integrity issues running the eMMC at 200MHz on Puma
RK3399-Q7.

Similar to the work-around found for RK3399 Gru boards, lowering the
frequency to 100MHz made the eMMC much more stable, so let's lower the
frequency to 100MHz.

It might be possible to run at 150MHz as on RK3399 Gru boards but only
100MHz was extensively tested.

Cc: Quentin Schulz <foss+kernel@0leil.net>
Signed-off-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
Signed-off-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
---

Note/RFC: as opposed to gru DTSI, max-frequency is used here instead of
assigned-clock-rates.

AFAIU, max-frequency applies to the SD bus rate, while
assigned-clock-rates applies to the clock fed to the SD host controller
inside the SoC. max-frequency does not interact with the clock subsystem
AFAICT. assigned-clock-rates sets the clock rate at probe, regardless of
eMMC tuning.
Technically, the Arasan SDHC IP supports silicon-hardcoded clock
multiplier so I think setting assigned-clock-rates as a way of rate
limiting the eMMC block is incorrect and max-frequency should be used
instead (as done in this patch). Otherwise the SDHC could still potentially
derive a 200MHz clock from a lower rate clock thanks to its multiplier.

 arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi | 6 ++++++
 1 file changed, 6 insertions(+)

Comments

Peter Geis Jan. 20, 2022, 3:06 p.m. UTC | #1 
On Wed, Jan 19, 2022 at 8:52 AM <quentin.schulz@theobroma-systems.com> wrote:
>
> From: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
>
> There are signal integrity issues running the eMMC at 200MHz on Puma
> RK3399-Q7.
>
> Similar to the work-around found for RK3399 Gru boards, lowering the
> frequency to 100MHz made the eMMC much more stable, so let's lower the
> frequency to 100MHz.
>
> It might be possible to run at 150MHz as on RK3399 Gru boards but only
> 100MHz was extensively tested.
>
> Cc: Quentin Schulz <foss+kernel@0leil.net>
> Signed-off-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
> Signed-off-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
> ---
>
> Note/RFC: as opposed to gru DTSI, max-frequency is used here instead of
> assigned-clock-rates.
>
> AFAIU, max-frequency applies to the SD bus rate, while
> assigned-clock-rates applies to the clock fed to the SD host controller
> inside the SoC. max-frequency does not interact with the clock subsystem
> AFAICT. assigned-clock-rates sets the clock rate at probe, regardless of
> eMMC tuning.
> Technically, the Arasan SDHC IP supports silicon-hardcoded clock
> multiplier so I think setting assigned-clock-rates as a way of rate
> limiting the eMMC block is incorrect and max-frequency should be used
> instead (as done in this patch). Otherwise the SDHC could still potentially
> derive a 200MHz clock from a lower rate clock thanks to its multiplier.
>
>  arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi | 6 ++++++
>  1 file changed, 6 insertions(+)
>
> diff --git a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> index fb67db4619ea..a6108578aae0 100644
> --- a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> +++ b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> @@ -425,6 +425,12 @@ vcc5v0_host_en: vcc5v0-host-en {
>  };
>
>  &sdhci {
> +       /*
> +        * Signal integrity isn't great at 200MHz but 100MHz has proven stable
> +        * enough.
> +        */
> +       max-frequency = <100000000>;
> +
>         bus-width = <8>;
>         mmc-hs400-1_8v;
>         mmc-hs400-enhanced-strobe;

I don't have these boards nor the schematics handy for them, but
wouldn't it be better to simply switch to mmc-hs200-1_8v?
Other rk3399 boards don't have issues with hs200 at full 200mhz, and
as I understand it hs400-es exhibits stability issues on most rk3399
boards.

Feel free to disregard my comments if you've already tested this.

> --
> 2.34.1
>
>
> _______________________________________________
> Linux-rockchip mailing list
> Linux-rockchip@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-rockchip
Quentin Schulz Jan. 20, 2022, 4:03 p.m. UTC | #2 
Hi Peter,

On 1/20/22 16:06, Peter Geis wrote:
> On Wed, Jan 19, 2022 at 8:52 AM <quentin.schulz@theobroma-systems.com> wrote:
>>
>> From: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
>>
>> There are signal integrity issues running the eMMC at 200MHz on Puma
>> RK3399-Q7.
>>
>> Similar to the work-around found for RK3399 Gru boards, lowering the
>> frequency to 100MHz made the eMMC much more stable, so let's lower the
>> frequency to 100MHz.
>>
>> It might be possible to run at 150MHz as on RK3399 Gru boards but only
>> 100MHz was extensively tested.
>>
>> Cc: Quentin Schulz <foss+kernel@0leil.net>
>> Signed-off-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
>> Signed-off-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
>> ---
>>
>> Note/RFC: as opposed to gru DTSI, max-frequency is used here instead of
>> assigned-clock-rates.
>>
>> AFAIU, max-frequency applies to the SD bus rate, while
>> assigned-clock-rates applies to the clock fed to the SD host controller
>> inside the SoC. max-frequency does not interact with the clock subsystem
>> AFAICT. assigned-clock-rates sets the clock rate at probe, regardless of
>> eMMC tuning.
>> Technically, the Arasan SDHC IP supports silicon-hardcoded clock
>> multiplier so I think setting assigned-clock-rates as a way of rate
>> limiting the eMMC block is incorrect and max-frequency should be used
>> instead (as done in this patch). Otherwise the SDHC could still potentially
>> derive a 200MHz clock from a lower rate clock thanks to its multiplier.
>>
>>   arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>> index fb67db4619ea..a6108578aae0 100644
>> --- a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>> +++ b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>> @@ -425,6 +425,12 @@ vcc5v0_host_en: vcc5v0-host-en {
>>   };
>>
>>   &sdhci {
>> +       /*
>> +        * Signal integrity isn't great at 200MHz but 100MHz has proven stable
>> +        * enough.
>> +        */
>> +       max-frequency = <100000000>;
>> +
>>          bus-width = <8>;
>>          mmc-hs400-1_8v;
>>          mmc-hs400-enhanced-strobe;
> 
> I don't have these boards nor the schematics handy for them, but
> wouldn't it be better to simply switch to mmc-hs200-1_8v?

Thanks for the suggestion, I just tested, by removing mmc-hs400-1_8v; 
and mmc-hs400-enhanced-strobe; but keeping the clock rate at 200MHz and 
adding mmc-hs200-1_8v; failed a basic dd test.

Cheers,
Quentin

> Other rk3399 boards don't have issues with hs200 at full 200mhz, and
> as I understand it hs400-es exhibits stability issues on most rk3399
> boards.
> 
> Feel free to disregard my comments if you've already tested this.
> 
>> --
>> 2.34.1
>>
>>
>> _______________________________________________
>> Linux-rockchip mailing list
>> Linux-rockchip@lists.infradead.org
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.infradead.org_mailman_listinfo_linux-2Drockchip&d=DwIBaQ&c=_sEr5x9kUWhuk4_nFwjJtA&r=LYjLexDn7rXIzVmkNPvw5ymA1XTSqHGq8yBP6m6qZZ4njZguQhZhkI_-172IIy1t&m=OTusi9XARCbef1Ba5t1gv4_LqRkOaxFY3UtvYky6NVLpeiziqqmdOJ8-R5WXWFNr&s=xBnqhBu20rxvu_V-5EhXyA_yNxcJpa4Dq6HO2Ow3Gk8&e=
Peter Geis Jan. 20, 2022, 5:14 p.m. UTC | #3 
On Thu, Jan 20, 2022 at 11:03 AM Quentin Schulz
<quentin.schulz@theobroma-systems.com> wrote:
>
> Hi Peter,
>
> On 1/20/22 16:06, Peter Geis wrote:
> > On Wed, Jan 19, 2022 at 8:52 AM <quentin.schulz@theobroma-systems.com> wrote:
> >>
> >> From: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
> >>
> >> There are signal integrity issues running the eMMC at 200MHz on Puma
> >> RK3399-Q7.
> >>
> >> Similar to the work-around found for RK3399 Gru boards, lowering the
> >> frequency to 100MHz made the eMMC much more stable, so let's lower the
> >> frequency to 100MHz.
> >>
> >> It might be possible to run at 150MHz as on RK3399 Gru boards but only
> >> 100MHz was extensively tested.
> >>
> >> Cc: Quentin Schulz <foss+kernel@0leil.net>
> >> Signed-off-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
> >> Signed-off-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
> >> ---
> >>
> >> Note/RFC: as opposed to gru DTSI, max-frequency is used here instead of
> >> assigned-clock-rates.
> >>
> >> AFAIU, max-frequency applies to the SD bus rate, while
> >> assigned-clock-rates applies to the clock fed to the SD host controller
> >> inside the SoC. max-frequency does not interact with the clock subsystem
> >> AFAICT. assigned-clock-rates sets the clock rate at probe, regardless of
> >> eMMC tuning.
> >> Technically, the Arasan SDHC IP supports silicon-hardcoded clock
> >> multiplier so I think setting assigned-clock-rates as a way of rate
> >> limiting the eMMC block is incorrect and max-frequency should be used
> >> instead (as done in this patch). Otherwise the SDHC could still potentially
> >> derive a 200MHz clock from a lower rate clock thanks to its multiplier.
> >>
> >>   arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi | 6 ++++++
> >>   1 file changed, 6 insertions(+)
> >>
> >> diff --git a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> >> index fb67db4619ea..a6108578aae0 100644
> >> --- a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> >> +++ b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
> >> @@ -425,6 +425,12 @@ vcc5v0_host_en: vcc5v0-host-en {
> >>   };
> >>
> >>   &sdhci {
> >> +       /*
> >> +        * Signal integrity isn't great at 200MHz but 100MHz has proven stable
> >> +        * enough.
> >> +        */
> >> +       max-frequency = <100000000>;
> >> +
> >>          bus-width = <8>;
> >>          mmc-hs400-1_8v;
> >>          mmc-hs400-enhanced-strobe;
> >
> > I don't have these boards nor the schematics handy for them, but
> > wouldn't it be better to simply switch to mmc-hs200-1_8v?
>
> Thanks for the suggestion, I just tested, by removing mmc-hs400-1_8v;
> and mmc-hs400-enhanced-strobe; but keeping the clock rate at 200MHz and
> adding mmc-hs200-1_8v; failed a basic dd test.

Quite understandable.
I more meant it may permit you to reach a higher clock rate than
100mhz and thus possibly a higher data rate, for example perhaps
150mhz would be reachable.
For example I would do a bandwidth test for hs400-es at 100mhz, then
test at hs200 and find the maximum stable clock rate.
If 200mhz is stable enough to probe and /sys/kernel/debug/mmc<x>/ios
shows you are in 8bit hs200, you can adjust the clock rate through
/sys/kernel/debug/mmc<x>/clock without needing to reboot.
It will speed up testing.

If hs200 permits a higher overall data rate than hs400-es, it would be
worth making that switch.

Always,
Peter

>
> Cheers,
> Quentin
>
> > Other rk3399 boards don't have issues with hs200 at full 200mhz, and
> > as I understand it hs400-es exhibits stability issues on most rk3399
> > boards.
> >
> > Feel free to disregard my comments if you've already tested this.
> >
> >> --
> >> 2.34.1
> >>
> >>
> >> _______________________________________________
> >> Linux-rockchip mailing list
> >> Linux-rockchip@lists.infradead.org
> >> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.infradead.org_mailman_listinfo_linux-2Drockchip&d=DwIBaQ&c=_sEr5x9kUWhuk4_nFwjJtA&r=LYjLexDn7rXIzVmkNPvw5ymA1XTSqHGq8yBP6m6qZZ4njZguQhZhkI_-172IIy1t&m=OTusi9XARCbef1Ba5t1gv4_LqRkOaxFY3UtvYky6NVLpeiziqqmdOJ8-R5WXWFNr&s=xBnqhBu20rxvu_V-5EhXyA_yNxcJpa4Dq6HO2Ow3Gk8&e=
Quentin Schulz Jan. 24, 2022, 11:17 a.m. UTC | #4 
Hi Peter,

On 1/20/22 18:14, Peter Geis wrote:
> On Thu, Jan 20, 2022 at 11:03 AM Quentin Schulz
> <quentin.schulz@theobroma-systems.com> wrote:
>>
>> Hi Peter,
>>
>> On 1/20/22 16:06, Peter Geis wrote:
>>> On Wed, Jan 19, 2022 at 8:52 AM <quentin.schulz@theobroma-systems.com> wrote:
>>>>
>>>> From: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
>>>>
>>>> There are signal integrity issues running the eMMC at 200MHz on Puma
>>>> RK3399-Q7.
>>>>
>>>> Similar to the work-around found for RK3399 Gru boards, lowering the
>>>> frequency to 100MHz made the eMMC much more stable, so let's lower the
>>>> frequency to 100MHz.
>>>>
>>>> It might be possible to run at 150MHz as on RK3399 Gru boards but only
>>>> 100MHz was extensively tested.
>>>>
>>>> Cc: Quentin Schulz <foss+kernel@0leil.net>
>>>> Signed-off-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
>>>> Signed-off-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
>>>> ---
>>>>
>>>> Note/RFC: as opposed to gru DTSI, max-frequency is used here instead of
>>>> assigned-clock-rates.
>>>>
>>>> AFAIU, max-frequency applies to the SD bus rate, while
>>>> assigned-clock-rates applies to the clock fed to the SD host controller
>>>> inside the SoC. max-frequency does not interact with the clock subsystem
>>>> AFAICT. assigned-clock-rates sets the clock rate at probe, regardless of
>>>> eMMC tuning.
>>>> Technically, the Arasan SDHC IP supports silicon-hardcoded clock
>>>> multiplier so I think setting assigned-clock-rates as a way of rate
>>>> limiting the eMMC block is incorrect and max-frequency should be used
>>>> instead (as done in this patch). Otherwise the SDHC could still potentially
>>>> derive a 200MHz clock from a lower rate clock thanks to its multiplier.
>>>>
>>>>    arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi | 6 ++++++
>>>>    1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>>>> index fb67db4619ea..a6108578aae0 100644
>>>> --- a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>>>> +++ b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
>>>> @@ -425,6 +425,12 @@ vcc5v0_host_en: vcc5v0-host-en {
>>>>    };
>>>>
>>>>    &sdhci {
>>>> +       /*
>>>> +        * Signal integrity isn't great at 200MHz but 100MHz has proven stable
>>>> +        * enough.
>>>> +        */
>>>> +       max-frequency = <100000000>;
>>>> +
>>>>           bus-width = <8>;
>>>>           mmc-hs400-1_8v;
>>>>           mmc-hs400-enhanced-strobe;
>>>
>>> I don't have these boards nor the schematics handy for them, but
>>> wouldn't it be better to simply switch to mmc-hs200-1_8v?
>>
>> Thanks for the suggestion, I just tested, by removing mmc-hs400-1_8v;
>> and mmc-hs400-enhanced-strobe; but keeping the clock rate at 200MHz and
>> adding mmc-hs200-1_8v; failed a basic dd test.
> 
> Quite understandable.
> I more meant it may permit you to reach a higher clock rate than
> 100mhz and thus possibly a higher data rate, for example perhaps
> 150mhz would be reachable.
> For example I would do a bandwidth test for hs400-es at 100mhz, then
> test at hs200 and find the maximum stable clock rate.
> If 200mhz is stable enough to probe and /sys/kernel/debug/mmc<x>/ios
> shows you are in 8bit hs200, you can adjust the clock rate through
> /sys/kernel/debug/mmc<x>/clock without needing to reboot.
> It will speed up testing.
> 
> If hs200 permits a higher overall data rate than hs400-es, it would be
> worth making that switch.
> 

Got the HW dept to look into this and they said that since the main 
difference between HS200 and HS400-ES is that for the same clock rate, 
twice the data rate is possible and that the signal integrity issue we 
have is on the clock line, anything that impacts HS400-ES wrt clock rate 
will impact HS200 too. Therefore HS200 should practically be half data 
rate on our board and no improvement should be observed.

Lemme know if there's something we misunderstood or check more thoroughly.

Thanks!
Quentin

> Always,
> Peter
> 
>>
>> Cheers,
>> Quentin
>>
>>> Other rk3399 boards don't have issues with hs200 at full 200mhz, and
>>> as I understand it hs400-es exhibits stability issues on most rk3399
>>> boards.
>>>
>>> Feel free to disregard my comments if you've already tested this.
>>>
>>>> --
>>>> 2.34.1
>>>>
>>>>
>>>> _______________________________________________
>>>> Linux-rockchip mailing list
>>>> Linux-rockchip@lists.infradead.org
>>>> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.infradead.org_mailman_listinfo_linux-2Drockchip&d=DwIBaQ&c=_sEr5x9kUWhuk4_nFwjJtA&r=LYjLexDn7rXIzVmkNPvw5ymA1XTSqHGq8yBP6m6qZZ4njZguQhZhkI_-172IIy1t&m=OTusi9XARCbef1Ba5t1gv4_LqRkOaxFY3UtvYky6NVLpeiziqqmdOJ8-R5WXWFNr&s=xBnqhBu20rxvu_V-5EhXyA_yNxcJpa4Dq6HO2Ow3Gk8&e=
Heiko Stübner Jan. 29, 2022, 5:57 p.m. UTC | #5 
On Wed, 19 Jan 2022 14:49:48 +0100, quentin.schulz@theobroma-systems.com wrote:
> There are signal integrity issues running the eMMC at 200MHz on Puma
> RK3399-Q7.
> 
> Similar to the work-around found for RK3399 Gru boards, lowering the
> frequency to 100MHz made the eMMC much more stable, so let's lower the
> frequency to 100MHz.
> 
> [...]

Applied, thanks!

[1/1] arm64: dts: rockchip: fix rk3399-puma eMMC HS400 signal integrity
      commit: 7623e16af6bac131305292eb99cea988adbf0484

Best regards,
diff mbox series

Patch

diff --git a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
index fb67db4619ea..a6108578aae0 100644
--- a/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
+++ b/arch/arm64/boot/dts/rockchip/rk3399-puma.dtsi
@@ -425,6 +425,12 @@  vcc5v0_host_en: vcc5v0-host-en {
 };
 
 &sdhci {
+	/*
+	 * Signal integrity isn't great at 200MHz but 100MHz has proven stable
+	 * enough.
+	 */
+	max-frequency = <100000000>;
+
 	bus-width = <8>;
 	mmc-hs400-1_8v;
 	mmc-hs400-enhanced-strobe;