ARM: dts: sunxi: Fix cpufreq instabilities

PATCH Superseded Chen-Yu Tsai 3 participants 11 messages Mar 19, 2015 – Mar 26, 2015

Subsystems (2)

Device tree bindings Arm/Allwinner sunXi SoC support

lore

Share & tools

lore

Chen-Yu Tsai

author

2015-03-19 02:39 UTC #0 lore raw

cover ARM: dts: sunxi: Fix cpufreq instabilities

Hi Maxime,

Here are 2 dts fixes for sunxi cpufreq support, aimed at 4.0.

The first patch limits the Olimex A10-Lime to 960MHz, as a few people reported on IRC that it was unstable at 1008MHz.

The second patch removes the overclocking settings from sun5i/sun7i dts, so all the boards will be stable, regardless of whether regulators were added or not.

I would like to revisit this after the new OPP bindings are merged and support added to cpufreq-dt.

Regards
ChenYu

Chen-Yu Tsai (2):
ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting
ARM: dts: sunxi: Remove overclocked/overvoltaged OPP

 arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts | 14 ++++++++++++++
 arch/arm/boot/dts/sun5i-a13.dtsi               |  3 +--
 arch/arm/boot/dts/sun7i-a20.dtsi               |  3 +--
 3 files changed, 16 insertions(+), 4 deletions(-)

Signature · 1 line

2.1.4

Chen-Yu Tsai

author

2015-03-19 02:39 UTC #1 lore raw

PATCH 1/2 ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

The Olimex A10-Lime is known to be unstable when running at 1008MHz.

Signed-off-by Chen-Yu Tsai <[email protected]>

---
 arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +14 -0

index 31dc2f1c3870..16ecb8938e19 100644
@@ -74,6 +74,20 @@
 	status = "okay";
 };
 
+&cpu0 {
+	/* The A10-Lime is known to be unstable when running at 1008 MHz */
+	operating-points = <
+		/* kHz    uV */
+		960000  1400000
+		912000  1400000
+		864000  1300000
+		720000  1200000
+		528000  1100000
+		312000  1000000
+		144000  900000
+		>;
+};
+
 &ehci0 {
 	status = "okay";
 };

Signature · 1 line

2.1.4

Siarhei Siamashka

2015-03-19 06:57 UTC #2 lore raw

PATCH 1/2 ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

Quoted 22 lines · On Thu, 19 Mar 2015 10:39:32 +0800 Chen-Yu Tsai <[email protected]> wrote:

The Olimex A10-Lime is known to be unstable when running at 1008MHz.

Signed-off-by Chen-Yu Tsai <[email protected]>

---
 arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0

index 31dc2f1c3870..16ecb8938e19 100644
@@ -74,6 +74,20 @@
 	status = "okay";
 };
 
+&cpu0 {
+	/* The A10-Lime is known to be unstable when running at 1008 MHz */
+	operating-points = <
+		/* kHz    uV */
+		960000  1400000
+		912000  1400000
+		864000  1300000

arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts line 93

Quoted 10 lines

arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0

+		720000  1200000
+		528000  1100000
+		312000  1000000
+		144000  900000
+		>;
+};
+
 &ehci0 {
 	status = "okay";
 };

Thanks for the patch. At least it should make my A10-OLinuXino-LIME working without obvious failures out of the box (the U-Boot is still another story though and there is a gap during boot up when the board is running with unreliable settings, but the probability of a failure is rather low).

I should also mention that using 960MHz @1.4V does not fail, but it does not have any safety headroom either (the cyan 'sun4i_poorlime' line on the plot):

    http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png

On the other hand, my board is on the worst part of the spectrum (many other a10-lime boards do not fail even at 1008MHz), so maybe having extra safety headroom is less necessary.

An interesting question is whether the same problem may be reproducible on the Allwinner A10 devices other than A10-OLinuXino-LIME. My original problem report

https://www.mail-archive.com/linux-sunxi at googlegroups.com/msg04343.html

mentioned the A10-OLinuXino-LIME rev.A and introduced some sort of a bias by itself. At least I have seen people saying something like "my a10-lime revision is not rev.A, so it's none of my concern and I'm not going to bother running any tests". So far we have accumulated reports from 4 or 5 people having this reliability problem on their A10-OLinuXino-LIME (various revisions, not just rev.A), but not much from the other boards owners.

Anyway, this particular patch is

Tested-by Siarhei Siamashka <[email protected]>

Acked-by Siarhei Siamashka <[email protected]>

Signature · 2 lines

Best regards,
Siarhei Siamashka

Chen-Yu Tsai

author

2015-03-19 08:17 UTC #3 lore raw

PATCH 1/2 ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

Quoted 66 lines · Siarhei Siamashka

Quoted 22 lines · On Thu, 19 Mar 2015 10:39:32 +0800 Chen-Yu Tsai <[email protected]> wrote:
The Olimex A10-Lime is known to be unstable when running at 1008MHz.
Signed-off-by Chen-Yu Tsai <[email protected]>
---
 arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0
index 31dc2f1c3870..16ecb8938e19 100644
@@ -74,6 +74,20 @@
      status = "okay";
 };
 
+&cpu0 {
+     /* The A10-Lime is known to be unstable when running at 1008 MHz */
+     operating-points = <
+             /* kHz    uV */
+             960000  1400000
+             912000  1400000
+             864000  1300000
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts line 93
Quoted 10 lines
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0
+             720000  1200000
+             528000  1100000
+             312000  1000000
+             144000  900000
+             >;
+};
+
 &ehci0 {
      status = "okay";
 };
Thanks for the patch. At least it should make my A10-OLinuXino-LIME working without obvious failures out of the box (the U-Boot is still another story though and there is a gap during boot up when the board is running with unreliable settings, but the probability of a failure is rather low).
I should also mention that using 960MHz @1.4V does not fail, but it does not have any safety headroom either (the cyan 'sun4i_poorlime' line on the plot):
    http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png
On the other hand, my board is on the worst part of the spectrum (many other a10-lime boards do not fail even at 1008MHz), so maybe having extra safety headroom is less necessary.
An interesting question is whether the same problem may be reproducible on the Allwinner A10 devices other than A10-OLinuXino-LIME. My original problem report
https://www.mail-archive.com/linux-sunxi at googlegroups.com/msg04343.html
mentioned the A10-OLinuXino-LIME rev.A and introduced some sort of a bias by itself. At least I have seen people saying something like "my a10-lime revision is not rev.A, so it's none of my concern and I'm not going to bother running any tests". So far we have accumulated reports from 4 or 5 people having this reliability problem on their A10-OLinuXino-LIME (various revisions, not just rev.A), but not much from the other boards owners.

Quoted 3 lines

Anyway, this particular patch is
Tested-by Siarhei Siamashka <[email protected]>
Acked-by Siarhei Siamashka <[email protected]>

Good to hear it works. Did you test all the settings? I copied the wrong settings, from sun5i-a13.dtsi instead of sun4i-a10.dtsi.

I'll send a fixed version later.

ChenYu

Siarhei Siamashka

2015-03-19 10:28 UTC #4 lore raw

PATCH 1/2 ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

Quoted 68 lines · On Thu, 19 Mar 2015 16:17:30 +0800 Chen-Yu Tsai <[email protected]> wrote:

Quoted 66 lines · Siarhei Siamashka
Quoted 22 lines · On Thu, 19 Mar 2015 10:39:32 +0800 Chen-Yu Tsai <[email protected]> wrote:
The Olimex A10-Lime is known to be unstable when running at 1008MHz.
Signed-off-by Chen-Yu Tsai <[email protected]>
---
 arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0
index 31dc2f1c3870..16ecb8938e19 100644
@@ -74,6 +74,20 @@
      status = "okay";
 };
 
+&cpu0 {
+     /* The A10-Lime is known to be unstable when running at 1008 MHz */
+     operating-points = <
+             /* kHz    uV */
+             960000  1400000
+             912000  1400000
+             864000  1300000
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts line 93
Quoted 10 lines
arch/arm/boot/dts/sun4i-a10-olinuxino-lime.dts +7 -0
+             720000  1200000
+             528000  1100000
+             312000  1000000
+             144000  900000
+             >;
+};
+
 &ehci0 {
      status = "okay";
 };
Thanks for the patch. At least it should make my A10-OLinuXino-LIME working without obvious failures out of the box (the U-Boot is still another story though and there is a gap during boot up when the board is running with unreliable settings, but the probability of a failure is rather low).
I should also mention that using 960MHz @1.4V does not fail, but it does not have any safety headroom either (the cyan 'sun4i_poorlime' line on the plot):
    http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png
On the other hand, my board is on the worst part of the spectrum (many other a10-lime boards do not fail even at 1008MHz), so maybe having extra safety headroom is less necessary.
An interesting question is whether the same problem may be reproducible on the Allwinner A10 devices other than A10-OLinuXino-LIME. My original problem report
https://www.mail-archive.com/linux-sunxi at googlegroups.com/msg04343.html
mentioned the A10-OLinuXino-LIME rev.A and introduced some sort of a bias by itself. At least I have seen people saying something like "my a10-lime revision is not rev.A, so it's none of my concern and I'm not going to bother running any tests". So far we have accumulated reports from 4 or 5 people having this reliability problem on their A10-OLinuXino-LIME (various revisions, not just rev.A), but not much from the other boards owners.

Quoted 5 lines

Quoted 3 lines
Anyway, this particular patch is
Tested-by Siarhei Siamashka <[email protected]>
Acked-by Siarhei Siamashka <[email protected]>
Good to hear it works. Did you test all the settings?

You have nailed it. I was about to send the results of the full round of tests after running them a bit longer and revoke the initial Tested-by.

Turns out that the 312MHz and 144MHz operating points fail to work reliable and tend to deadlock:

Testing CPU 0
960 MHz ............................................................ OK 912 MHz ............................................................ OK 864 MHz ............................................................ OK 720 MHz ............................................................ OK 528 MHz ............................................................ OK 312 MHz ............................................................ OK 144 MHz Write failed: Broken pipe

Quoted 1 line

I copied the wrong settings, from sun5i-a13.dtsi instead of sun4i-a10.dtsi.

I guess, this explains the problems at lower operating points.

Quoted 1 line

I'll send a fixed version later.

Thanks.

Signature · 2 lines

Best regards,
Siarhei Siamashka

Iain Paton

2015-03-26 16:58 UTC #5 lore raw

[linux-sunxi] Re: [PATCH 1/2] ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

In reply to Siarhei Siamashka

Quoted 16 lines · Siarhei Siamashka

I should also mention that using 960MHz @1.4V does not fail, but it does not have any safety headroom either (the cyan 'sun4i_poorlime' line on the plot):
    http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png
On the other hand, my board is on the worst part of the spectrum (many other a10-lime boards do not fail even at 1008MHz), so maybe having extra safety headroom is less necessary.
An interesting question is whether the same problem may be reproducible on the Allwinner A10 devices other than A10-OLinuXino-LIME. My original problem report
https://www.mail-archive.com/linux-sunxi at googlegroups.com/msg04343.html

Quoted 7 lines

mentioned the A10-OLinuXino-LIME rev.A and introduced some sort of a bias by itself. At least I have seen people saying something like "my a10-lime revision is not rev.A, so it's none of my concern and I'm not going to bother running any tests". So far we have accumulated reports from 4 or 5 people having this reliability problem on their A10-OLinuXino-LIME (various revisions, not just rev.A), but not much from the other boards owners.

I'm somewhat sad to see 1008MHz go. I have 2x revA A10-Lime boards that are stable at 1008MHz with Maximes sunxi/for-next, a full regulator description in the dts and a very recent u-boot. They are not stable at 1056 or above however.

Can I ask what the basis is for the 960MHz setting? I don't see any instances of it in any of the a10 fex files, meaning it's likely not very well tested.

If there's interest, I'll send the patch adding regulators to the dts for the a10-lime.

Rgds,
Iain

Chen-Yu Tsai

author

2015-03-26 17:56 UTC #6 lore raw

[linux-sunxi] Re: [PATCH 1/2] ARM: dts: sun4i: a10-lime: Override and remove 1008MHz OPP setting

Hi,

Quoted 32 lines · Iain Paton

Quoted 16 lines · Siarhei Siamashka
I should also mention that using 960MHz @1.4V does not fail, but it does not have any safety headroom either (the cyan 'sun4i_poorlime' line on the plot):
    http://people.freedesktop.org/~siamashka/files/20140512/sunxi-cpufreq-plot.png
On the other hand, my board is on the worst part of the spectrum (many other a10-lime boards do not fail even at 1008MHz), so maybe having extra safety headroom is less necessary.
An interesting question is whether the same problem may be reproducible on the Allwinner A10 devices other than A10-OLinuXino-LIME. My original problem report
https://www.mail-archive.com/linux-sunxi at googlegroups.com/msg04343.html
Quoted 7 lines
mentioned the A10-OLinuXino-LIME rev.A and introduced some sort of a bias by itself. At least I have seen people saying something like "my a10-lime revision is not rev.A, so it's none of my concern and I'm not going to bother running any tests". So far we have accumulated reports from 4 or 5 people having this reliability problem on their A10-OLinuXino-LIME (various revisions, not just rev.A), but not much from the other boards owners.
I'm somewhat sad to see 1008MHz go. I have 2x revA A10-Lime boards that are stable at 1008MHz with Maximes sunxi/for-next, a full regulator description in the dts and a very recent u-boot. They are not stable at 1056 or above however.

Quoted 3 lines

Can I ask what the basis is for the 960MHz setting? I don't see any instances of it in any of the a10 fex files, meaning it's likely not very well tested.

It was copied from the wrong file. Please see v2 of the series.

Quoted 2 lines

If there's interest, I'll send the patch adding regulators to the dts for the a10-lime.

Please do.

ChenYu

Chen-Yu Tsai

author

2015-03-19 02:39 UTC #7 lore raw

PATCH 2/2 ARM: dts: sunxi: Remove overclocked/overvoltaged OPP

In reply to Chen-Yu Tsai

Without proper regulator support for individual boards, it is dangerous to have overclocked/overvoltaged OPPs in the list. Cpufreq will increase the frequency without the accompanying voltage increase, resulting in an unstable system.

Remove them for now. We can revisit them with the new version of OPP bindings, which support boost settings and frequency ranges, among other things.

Signed-off-by Chen-Yu Tsai <[email protected]>

---
 arch/arm/boot/dts/sun5i-a13.dtsi | 3 +--
 arch/arm/boot/dts/sun7i-a20.dtsi | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)

arch/arm/boot/dts/sun5i-a13.dtsi +1 -2

index 5d9ce84486bb..39264f7c7ae6 100644
@@ -140,7 +140,6 @@
 	clock-latency = <244144>; /* 8 32k periods */
 	operating-points = <
 		/* kHz    uV */
-		1104000	1500000
 		1008000 1400000
 		912000  1350000
 		864000  1300000

@@ -150,7 +149,7 @@
 		>;
 	#cooling-cells = <2>;
 	cooling-min-level = <0>;
-	cooling-max-level = <6>;
+	cooling-max-level = <5>;
 };
 
 &ehci0 {

arch/arm/boot/dts/sun7i-a20.dtsi +1 -2

index 3a8530b79f1c..fdd181792b4b 100644
@@ -105,7 +105,6 @@
 			clock-latency = <244144>; /* 8 32k periods */
 			operating-points = <
 				/* kHz    uV */
-				1008000 1450000
 				960000  1400000
 				912000  1400000
 				864000  1300000

@@ -116,7 +115,7 @@
 				>;
 			#cooling-cells = <2>;
 			cooling-min-level = <0>;
-			cooling-max-level = <7>;
+			cooling-max-level = <6>;
 		};
 
 		cpu at 1 {

Signature · 1 line

2.1.4

Siarhei Siamashka

2015-03-19 06:59 UTC #8 lore raw

PATCH 2/2 ARM: dts: sunxi: Remove overclocked/overvoltaged OPP

Quoted 9 lines · On Thu, 19 Mar 2015 10:39:33 +0800 Chen-Yu Tsai <[email protected]> wrote:

Without proper regulator support for individual boards, it is dangerous to have overclocked/overvoltaged OPPs in the list. Cpufreq will increase the frequency without the accompanying voltage increase, resulting in an unstable system.
Remove them for now. We can revisit them with the new version of OPP bindings, which support boost settings and frequency ranges, among other things.

Quoted 5 lines

Signed-off-by Chen-Yu Tsai <[email protected]>

---
 arch/arm/boot/dts/sun5i-a13.dtsi | 3 +--
 arch/arm/boot/dts/sun7i-a20.dtsi | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)

What about the sun4i-a10.dtsi file? Was it forgotten or skipped deliberately?

Signature · 2 lines

Best regards,
Siarhei Siamashka

Chen-Yu Tsai

author

2015-03-19 07:11 UTC #9 lore raw

PATCH 2/2 ARM: dts: sunxi: Remove overclocked/overvoltaged OPP

Quoted 12 lines · Siarhei Siamashka

Quoted 9 lines · On Thu, 19 Mar 2015 10:39:33 +0800 Chen-Yu Tsai <[email protected]> wrote:
Without proper regulator support for individual boards, it is dangerous to have overclocked/overvoltaged OPPs in the list. Cpufreq will increase the frequency without the accompanying voltage increase, resulting in an unstable system.
Remove them for now. We can revisit them with the new version of OPP bindings, which support boost settings and frequency ranges, among other things.

Quoted 8 lines

Quoted 5 lines
Signed-off-by Chen-Yu Tsai <[email protected]>
---
 arch/arm/boot/dts/sun5i-a13.dtsi | 3 +--
 arch/arm/boot/dts/sun7i-a20.dtsi | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)
What about the sun4i-a10.dtsi file? Was it forgotten or skipped deliberately?

sun4i-a10.dtsi does not have any overclocked/overvoltaged settings. The highest setting is 1008MHz @ 1.4V.

ChenYu

Chen-Yu Tsai

author

2015-03-19 08:14 UTC #10 lore raw

PATCH 2/2 ARM: dts: sunxi: Remove overclocked/overvoltaged OPP

Quoted 20 lines · Chen-Yu Tsai

Quoted 18 lines · Siarhei Siamashka
Quoted 14 lines · On Thu, 19 Mar 2015 10:39:33 +0800 Chen-Yu Tsai <[email protected]> wrote:
Without proper regulator support for individual boards, it is dangerous to have overclocked/overvoltaged OPPs in the list. Cpufreq will increase the frequency without the accompanying voltage increase, resulting in an unstable system.
Remove them for now. We can revisit them with the new version of OPP bindings, which support boost settings and frequency ranges, among other things.
Signed-off-by Chen-Yu Tsai <[email protected]>
---
 arch/arm/boot/dts/sun5i-a13.dtsi | 3 +--
 arch/arm/boot/dts/sun7i-a20.dtsi | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)

Quoted 5 lines

Quoted 2 lines
What about the sun4i-a10.dtsi file? Was it forgotten or skipped deliberately?
sun4i-a10.dtsi does not have any overclocked/overvoltaged settings. The highest setting is 1008MHz @ 1.4V.

Oops. Sorry. I missed it. Will send v2.

ChenYu