Skip to contentSkip to conversation

sunxi: set up PLL1 on sun6i+ without use dividers

PATCH New Icenowy Zheng 4 participants 10 messages Apr 9, 2017 – Apr 11, 2017

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

---

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

 arch/arm/mach-sunxi/clock_sun6i.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
arch/arm/mach-sunxi/clock_sun6i.c +3 -4
index 4762fbf0c3..ce4291b30e 100644
@@ -98,11 +98,10 @@ void clock_set_pll1(unsigned int clk)
 	int k = 1;
 	int m = 1;
 
-	if (clk > 1152000000) {
-		k = 2;
-	} else if (clk > 768000000) {
+	if (clk >= 1368000000) {
 		k = 3;
-		m = 2;
+	} else if (clk >= 768000000) {
+		k = 2;
 	}
 
 	/* Switch to 24MHz clock while changing PLL1 */
Signature · 1 line
2.12.2
#1 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

Hello,

Quoted 3 lines · Icenowy Zheng

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

Isn't adding a "From:" header line the way to reflect that ?

Signature · 1 line
Vincent Legoll
#2 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

In reply to Icenowy Zheng
Quoted 13 lines · Icenowy Zheng

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

Quoted 6 lines

---

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170410/bc7bb7c8/attachment.sig>
#3 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

Quoted 16 lines · 在 2017-04-10 14:59,Maxime Ripard 写道:
Quoted 15 lines · Icenowy Zheng

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add
a Signed-off-by tag here to his commit. So I take this code and added my
Signed-off-by.

Quoted 8 lines
Quoted 6 lines

---

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

Armbian has included this patch since several months ago, and it's proven that without it DVFS will be not stable. (And it proved at least it won't make the system more unstable). See [1].

And for Ondrej's research, please see [2].

I also verified it on my own board.

P.S. it's also the behavior of the clk-sun8iw7_tbl.c, in which the M part of PLL_CPUX is always 0.

[1]
https://github.com/igorpecovnik/lib/blob/master/patch/u-boot/u-boot-sunxi/h3-Fix-PLL1-setup-to-never-use-dividers.patch
[2] https://github.com/megous/h3-firmware

Quoted 2 lines

Maxime

#4 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

Quoted 22 lines · icenowy at aosc.io
Quoted 21 lines · 在 2017-04-10 14:59,Maxime Ripard 写道:
Quoted 20 lines · Icenowy Zheng

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add
a Signed-off-by tag here to his commit. So I take this code and added my
Signed-off-by.

---

This is a critical patch, and should be added to 2017.05.

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

Armbian has included this patch since several months ago, and it's proven that without it DVFS will be not stable. (And it proved at least it won't make the system more unstable). See [1].

Then, this should have been your commit log.

Thanks!
Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170411/7d450592/attachment.sig>
#5 lore raw
In reply to Maxime Ripard

Hi Maxime,

Maxime Ripard píše v Po 10. 04. 2017 v 08:59 +0200:

Quoted 14 lines · Icenowy Zheng
Quoted 13 lines

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

Quoted 8 lines
Quoted 6 lines

---

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

We already discussed this here: https://patchwork.kernel.org/patch/9446
365/

regards,
  o.j.

Quoted 7 lines

Maxime

Signature · 3 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170410/ac718063/attachment.sig>

#6 lore raw
Quoted 20 lines · 在 2017-04-10 18:06,Ondřej Jirman 写道:

Hi Maxime,

Maxime Ripard píše v Po 10. 04. 2017 v 08:59 +0200:

Quoted 17 lines · Icenowy Zheng
Quoted 16 lines

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

---

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

We already discussed this here: https://patchwork.kernel.org/patch/9446
365/

P.S. as Chen-Yu dig out, it's not the correct work to restrict factors, but we need to gate the PLL_CPUX when adjusting. (Of course the mux should
also be switched to osc24M when PLL_CPUX is gated)

I tried this on your h3-firmware code, and it succeeded with even P overused
(for any frequency) (and M is also free in this situation).

Quoted 11 lines

regards,
  o.j.

Quoted 7 lines

Maxime

--
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

#7 lore raw

icenowy at aosc.io píše v Po 10. 04. 2017 v 18:15 +0800:

Quoted 30 lines · 在 2017-04-10 18:06,Ondřej Jirman 写道:
Quoted 20 lines

Hi Maxime,

Maxime Ripard píše v Po 10. 04. 2017 v 08:59 +0200:

Quoted 17 lines · Icenowy Zheng
Quoted 16 lines

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

---

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

We already discussed this here: https://patchwork.kernel.org/patch/9446
365/

Quoted 8 lines

P.S. as Chen-Yu dig out, it's not the correct work to restrict factors, but we need to gate the PLL_CPUX when adjusting. (Of course the mux should
also be switched to osc24M when PLL_CPUX is gated)

I tried this on your h3-firmware code, and it succeeded with even P overused
(for any frequency) (and M is also free in this situation).

Interesting, I was only switching the source of the CPUX clock, and not gating the CPUX_PLL for the test. I guess the kernel still doesn't gate CPUX_PLL, so this whole situation should be fixed by adding the gating of PLL during the chnage and that would be it?

Sounds like the cleanest solution so far.

regards,
o.j.

Quoted 11 lines
Quoted 11 lines

regards,
  o.j.

Quoted 7 lines

Maxime

--
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170410/63542b44/attachment.sig>

#8 lore raw
In reply to Ondřej Jirman
Quoted 20 lines · Ondřej Jirman

Hi Maxime,

Maxime Ripard píše v Po 10. 04. 2017 v 08:59 +0200:

Quoted 17 lines · Icenowy Zheng
Quoted 16 lines

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

---

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

This doesn't mean anything. How has this been verified?

We already discussed this here: https://patchwork.kernel.org/patch/9446
365/

Yes, I know. I wasn't discussing the fact that this was the right thing to do or not, just that the wording didn't allow anyone to get an idea of *how* to reproduce that test on their own board.

Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170411/c0df8c43/attachment.sig>
#9 lore raw

PATCH 1/1 sunxi: set up PLL1 on sun6i+ without use dividers

In reply to Icenowy Zheng

Hi Icenowy,

Icenowy Zheng píše v Po 10. 04. 2017 v 00:19 +0800:

Quoted 6 lines

According to the researching result of Ondrej Jirman, the factor M of PLL1 shouldn't be used and the factor P should be used only if the intended frequency is lower than 288MHz. This is proven by the clk-sun8iw7_tbl.c in the BSP source code -- in there the M value is always 0 and the maximum frequency that P is not 0 is 224MHz.

Quoted 6 lines

As P is ignored on sun6i, it's not currently used. This patch removed the usage of M.

This patch is an original work by Ondrej Jirman, however, he didn't add a Signed-off-by tag here to his commit. So I take this code and added my Signed-off-by.

Well, here it is if that helps:

Though this might be somewhat difficult change to make. It works perfectly when combined with kernel that will not use dividers either. So it's ok with BSP kernel as you noted, but it might cause trouble (lockup on boot) with the mainline kernel unless the kernel is patched too to not use dividers. (it does now)

Current kernel patches are here: https://megous.com/dl/h3-cpufreq/

I have only tested that it breaks if kernel is patched and u-boot is not. I haven't tried to verify that it works the other way round, which is important for the mainlining of these changes.

That's also the reason it's been sitting in my trees for a long time and not being pushed out. It's not a problem when you control both u- boot and the kernel, but that will not be a case when mainlining. Though the current kernel will be broken anyway, when the kernel starts to enable cpufreq/DVFS, so it might not matter either way.

TLDR:
- if we patch the kernel first without patching u-boot there will be lockups
- if we patch u-boot first without kernel, that's not tested yet

regards,
o.j.

Quoted 32 lines

---

This is a critical patch, and should be added to 2017.05.

It has been verified by the Armbian.

 arch/arm/mach-sunxi/clock_sun6i.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
arch/arm/mach-sunxi/clock_sun6i.c +3 -4
index 4762fbf0c3..ce4291b30e 100644
@@ -98,11 +98,10 @@ void clock_set_pll1(unsigned int clk)
 	int k = 1;
 	int m = 1;
 
-	if (clk > 1152000000) {
-		k = 2;
-	} else if (clk > 768000000) {
+	if (clk >= 1368000000) {
 		k = 3;
-		m = 2;
+	} else if (clk >= 768000000) {
+		k = 2;
 	}
 
 	/* Switch to 24MHz clock while changing PLL1 */
Signature · 1 line
2.12.2

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20170410/8819ebef/attachment.sig>