[linux-sunxi] [PATCH 1/3] sunxi: A20-OLinuXino-LIME2: Fix ldo3/ldo4 in DTS

[Michael Haas]

LDO3 and LDO4 are set to regulator-always-on which causes
crashes on my A20-OLinuXino-LIME2 once axp20x-i2c.ko is loaded.

This crash is observed in combination with recent versions
of Das U-Boot starting from their commit
02cc27c74f9b884b538bcd1b93342a4c05b5d608.

commit 02cc27c74f9b884b538bcd1b93342a4c05b5d608
Date: Sat Oct 3 15:29:24 2015 +0200
sunxi: power: Change axp209 LDO3 and LDO4 default to disabled
LDO3 and LDO4 are normally either unused, or used to power csi
attached camera sensors, and as such do not need to be enabled at
boot time.
The regulator-always-on will cause both regulators to get turned on,
but the min / max constraints match the datasheet constrains / the
absolute min / max values these ldo-s can deliver, not the constraints
which the board design puts on these.
So now these ldo-s end up getting turned on at whatever voltage
is the default (which according to the datasheet is unknown),
where as the schematic says that if these get turned on (which
is not necessary) they should be run at 2.8V.
1) Does not have proper constraints for LDO3 / LDO4
2) Uses regulator-always-on; for these
I would suggest fixing both, first you can try making min = max =
2800000. And if that fixes things, which I expect it will, I would
also drop the regulator-always-on from the dts, since we really
only need to turn these on when using the csi interface in which
case it would be up to the csi driver to explicitly turn them on.

This patch implements the suggested changes. It is not enough
to set the voltage to 2800000 to avoid the crash. Hence, I have
also removed regulator-always-on.

Signed-off-by: Michael Haas <***@computerlinguist.org>
CC: Hans de Goede <***@redhat.com>
---
arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
index d5c796c..d5ff2e9 100644
--- a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
+++ b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
@@ -140,15 +140,13 @@
};

vcc_csi0: ldo3 {
- regulator-min-microvolt = <700000>;
- regulator-max-microvolt = <3500000>;
- regulator-always-on;
+ regulator-min-microvolt = <2800000>;
+ regulator-max-microvolt = <2800000>;
};

vcc_csi1: ldo4 {
- regulator-min-microvolt = <1250000>;
- regulator-max-microvolt = <3300000>;
- regulator-always-on;
+ regulator-min-microvolt = <2800000>;
+ regulator-max-microvolt = <2800000>;
};

vdd_cpu: dcdc2 {

[Michael Haas]

The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Signed-off-by: Michael Haas <***@computerlinguist.org>
---
arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts | 3 +++
1 file changed, 3 insertions(+)

diff --git a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
index d5ff2e9..d370166 100644
--- a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
+++ b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
@@ -161,6 +161,9 @@
regulator-always-on;
};
};
+ usb_power_supply: usb_power_supply {
+ compatible = "x-powers,axp202-usb-power-supply";
+ };
};
};

[Iain Paton]

Post by Michael Haas
The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Acked-by: Iain Paton <***@gmail.com>

[Maxime Ripard]

Hi,

Post by Michael Haas
The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Is there *any* reason to not use the AXP209 dtsi where it's already
defined?

Thanks,
Maxime

[Michael Haas]

Hello Maxime,

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Is there *any* reason to not use the AXP209 dtsi where it's already
defined?
Thanks,
Maxime

are you asking why usb-power-supply is not enabled by default?

It is entirely possible that there will never be USB power supply
connected to a particular board. There is also the possibility of
using just the ACIN part, for which I will be submitting a driver
soon (courtesy of Bruno Prémont).

On the other hand, the USB supply is probably so common that we could
just enable it. A board may choose to disable it or simply ignore the issue.
After all, the driver will correctly report on POWER_SUPPLY_PROP_PRESENT and
POWER_SUPPLY_PROP_ONLINE.

Michael

[Michael Haas]

Hi Maxime,

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Is there *any* reason to not use the AXP209 dtsi where it's already
defined?

do you have any preference for this being in the AXP209 dtsi? I've been
thinking about it and it makes sense to enable the power supply nodes
for all devices using the AXP209. They will just report themselves as
offline if the PMIC indicates nothing is connected.

Best,

Michael

[Maxime Ripard]

Post by Michael Haas
Hi Maxime,

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20-OLinuXino-LIME2 uses the AXP209 PMIC. This patch adds the
corresponding usb-power-supply node to the DTS.
If CONFIG_AXP20X_POWER is set, this patch enables the /sys/class/power_supply/
diirectory.

Is there *any* reason to not use the AXP209 dtsi where it's already
defined?

do you have any preference for this being in the AXP209 dtsi? I've been
thinking about it and it makes sense to enable the power supply nodes
for all devices using the AXP209. They will just report themselves as
offline if the PMIC indicates nothing is connected.

Yes, I'd prefer that a lot :)

It avoids enabling it on all the boards.

Thanks,
Maxime

[Michael Haas]

The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.

Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

Signed-off-by: Michael Haas <***@computerlinguist.org>
---
arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
index d370166..17791ef 100644
--- a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
+++ b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
@@ -179,6 +179,12 @@
};
};

+&i2c2 {
+ pinctrl-names = "default";
+ pinctrl-0 = <&i2c2_pins_a>;
+ status = "okay";
+};
+
&mmc0 {
pinctrl-names = "default";
pinctrl-0 = <&mmc0_pins_a>, <&mmc0_cd_pin_reference_design>;

[Iain Paton]

Post by Michael Haas
The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.
Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

Acked-by: Iain Paton <***@gmail.com>

This didn't go in originally simply because not everyone will have
the breakout and they may want to use those pins for other things.

[Maxime Ripard]

Hi,

Post by Michael Haas
The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.
Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

What are the other options on that pin? Is it something exclusively
dedicated to i2c on the pin headers and the documentation, or is
anyone free to use that pin for whatever he wishes?

Thanks,
Maxime

[Priit Laes]

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.
Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

Documentation for Olimex UEXT interface:
https://www.olimex.com/Products/Modules/UEXT/resources/UEXT_rev_B.pdf

Basically it *should* by default work as a board to board connector
which supports three serial communication interfaces: I2C, SPI and UART
(original document mentions RS232, but official modules work only with
3.3V logic levels).

Post by Maxime Ripard
What are the other options on that pin? Is it something exclusively
dedicated to i2c on the pin headers and the documentation, or is
anyone free to use that pin for whatever he wishes?
Thanks,
Maxime
-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

[Michael Haas]

Hello Maxime,

thank you for taking the time to review this patch set.

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.
Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

What are the other options on that pin? Is it something exclusively
dedicated to i2c on the pin headers and the documentation, or is
anyone free to use that pin for whatever he wishes?
Thanks,
Maxime

I assume you are talking about the UEXT connector, not about PB20 and
PB21. These are regular GPIO pins multiplexed with I2C, but no other
function.

Regarding UEXT: the A20-OLinuXino-LIME2-UEXT breakout board supports two
use cases here.

The first one: provide a simple adapter from the 0.05" pin headers on
the olinuxino to the more common world of 0.1" headers. In this use
case, you can plug the A20-OLinuXino-LIME2-UEXT intoany of LCD_CON or
GIO-{1,2,3} and you're good to go.

The second one: here, you want to use UEXT modules provided by Olimex.
It's a simple connector exposing I2C, SPI and UART as mentioned by
Priit. To use UEXT, you have to use GPIO-1 to get the correct pin mapping.

Regarding the official documentation provided by Olimex. The
A20-OLinuXino-Lime2_Rev_c.pdf shows the pins on GPIO-1 as PB20 and PB21.
In the schematics for the breakout board,
A20-OLinuXino-Lime2-UEXT_sch.pdf, the pins for the UEXT connector are
labeled as PB20/SCK and PB21/SDA. This acknowledges the double use.
However, PB20 and PB21 have pull-ups by default according to
A20-OLinuXino-Lime2_Rev_c.pdf,
which makes them a lot more useful as I2C than as GPIO.

What originally prompted me to submit this patch: I bought a breakout
board and a small LCD with UEXT connector together with my board. Then I
found out that it doesn't work out of the box because i2c2 is not
enabled. Since using UEXT is probably common for that particular
vendor, I believe it should be enabled by default.

Michael

[Maxime Ripard]

Hi Michael,

Post by Michael Haas
Hello Maxime,
thank you for taking the time to review this patch set.

Post by Maxime Ripard
Hi,

Post by Michael Haas
The A20 processor provides a third I2C bus on pins PB20 and PB21.
The A20-OLinuXino-LIME2 exposes this bus via its GPIO1 port.
Olimex also provide a breakout board called the
A20-OLinuXino-LIME2-UEXT. This change is required to properly
support I2C on the UEXT connector found there.

What are the other options on that pin? Is it something exclusively
dedicated to i2c on the pin headers and the documentation, or is
anyone free to use that pin for whatever he wishes?
Thanks,
Maxime

I assume you are talking about the UEXT connector, not about PB20 and
PB21. These are regular GPIO pins multiplexed with I2C, but no other
function.
Regarding UEXT: the A20-OLinuXino-LIME2-UEXT breakout board supports two
use cases here.
The first one: provide a simple adapter from the 0.05" pin headers on
the olinuxino to the more common world of 0.1" headers. In this use
case, you can plug the A20-OLinuXino-LIME2-UEXT intoany of LCD_CON or
GIO-{1,2,3} and you're good to go.
The second one: here, you want to use UEXT modules provided by Olimex.
It's a simple connector exposing I2C, SPI and UART as mentioned by
Priit. To use UEXT, you have to use GPIO-1 to get the correct pin mapping.
Regarding the official documentation provided by Olimex. The
A20-OLinuXino-Lime2_Rev_c.pdf shows the pins on GPIO-1 as PB20 and PB21.
In the schematics for the breakout board,
A20-OLinuXino-Lime2-UEXT_sch.pdf, the pins for the UEXT connector are
labeled as PB20/SCK and PB21/SDA. This acknowledges the double use.
However, PB20 and PB21 have pull-ups by default according to
A20-OLinuXino-Lime2_Rev_c.pdf,
which makes them a lot more useful as I2C than as GPIO.

Thanks for the explanation.

Post by Michael Haas
What originally prompted me to submit this patch: I bought a breakout
board and a small LCD with UEXT connector together with my board. Then I
found out that it doesn't work out of the box because i2c2 is not
enabled. Since using UEXT is probably common for that particular
vendor, I believe it should be enabled by default.

Unfortunately, if it isn't present on the main board, it shouldn't be
in the DT for that board. Otherwise, people relying on these pins and
not using the UEXT connector would be force fed a configuration that
they don't want, without any way to remove it.

That kind of adjustements is a perfect use case for the overlays
though, so I'd rather suggest writing one for the UEXT connector.

Thanks!
Maxime

[Iain Paton]

Post by Michael Haas
This patch implements the suggested changes. It is not enough
to set the voltage to 2800000 to avoid the crash. Hence, I have
also removed regulator-always-on.

NAK.

If setting both min and max to 2800000 isn't working then I'd suggest you're
just papering over the issue and not actually solving it.

It should be fairly clear that whenever those regulators are turned on by
some driver you're going to see the same issue, you've just delayed the
problem by removing the regulator-always-on item.

The real problem here is people monkeying with the regulators without
understanding what they're doing, bothering to test the results, or caring
about the consequences.

Anyway, Hans was correct in that the datasheet we have access to says that
the defaults for these regulators are undefined.
I suspect that's not the whole story though, if these regulators had random
values at powerup we'd have seen problems before now.
Most likely the values we think we know are simply incorrect plus there's
probably some element of sequencing involved - perhaps LDO4 needs turned on
first.
However I don't see sufficient detail in the datasheets to confirm any of
that, so we shouldn't assume anything.
Electrically on the lime2 LDO3 goes to VCC-PE and LDO4 to VCC-PG. Neither
groups E or G appear to be connected to anything other than gpio headers,
so this has little to do with board design, it's something internal to
the SoC that the datasheet is lacking on detail about.

One of the reasons for the regulator-always-on being in the dts was so that
the regulator didn't get touched and so didn't break anything.

All said and done, it only takes a minute with a multimeter to determine
actual defaults. Helpful to have multiple boards to gain some confidence
that they behave the same, but still basing values like this on tiny
sample sizes is probably unwise.

Looking at other A20 based boards, Cubieboards for example, shows LDO3/4
being unused and the VCC-PE/PG pins they power on the lime2 being tied
directly to 3.3v. No issues there simply because there's no way to mess
with the power to those blocks.

The patch you really want is something like the one below which works
on my boards regardless of the presence of regulator-always-on.
Verified on four boards with u-boot 2016.03 and kernel 4.5.0.

Whether you decide to remove regulator-always-on or not is a different
question, but please use 2.3v, or some other proven working value, for
ldo3. At least that way you have a fighting chance of the board
remaining functional when something wants to turn the regulator on.

It's a good bet that the same u-boot change will have implications
for other boards, if not now then at some future point.

Hans, take note, the schematic is wrong. You shouldn't assume either
the schematic or you are correct. Ill considered changes to regulators
can cause real damage to peoples boards.
I know we've discussed regulators before and that my opinion falls on
deaf ears, but worth one last try..



Author: Iain Paton <***@gmail.com>
Date: Fri Mar 25 22:07:55 2016 +0000

sunxi: A20-OLinuXino-LIME2: Update ldo3/ldo4 in DTS

Post by Michael Haas
commit 02cc27c74f9b884b538bcd1b93342a4c05b5d608
Date: Sat Oct 3 15:29:24 2015 +0200
sunxi: power: Change axp209 LDO3 and LDO4 default to disabled

causes the lime2 to lockup/crash when LDO3/4 are re-enabled by the
kernel.

The AXP209 datasheet shows the powerup values for these regulators
to be undefined and the 2.8v suggested by the lime2 schematic also
causes a lockup when the regulators are re-enabled so is obviously
also incorrect.

Empirical measurement suggests LDO3 @ 2.3v and LDO4 @ 2.8v are the
actual defaults, but is based on a small sample size.

Verified to work with u-boot 2016.03, kernel 4.5.0 on four lime2
boards.

Signed-off-by: Iain Paton <***@gmail.com>
---

diff --git a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
index d5c796c..e665d22 100644
--- a/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
+++ b/arch/arm/boot/dts/sun7i-a20-olinuxino-lime2.dts
@@ -140,14 +140,14 @@
};

vcc_csi0: ldo3 {
- regulator-min-microvolt = <700000>;
- regulator-max-microvolt = <3500000>;
+ regulator-min-microvolt = <2300000>;
+ regulator-max-microvolt = <2300000>;
regulator-always-on;
};

vcc_csi1: ldo4 {
- regulator-min-microvolt = <1250000>;
- regulator-max-microvolt = <3300000>;
+ regulator-min-microvolt = <2800000>;
+ regulator-max-microvolt = <2800000>;
regulator-always-on;
};

[Iain Paton]

Having reread the A20 datasheet yesterday, something in the back of my
mind was bothering me overnight so I took some time to check this morning.

On the lime2 schematic we see the pins labeled as follows:

F19 : VCC_CSI0
E18 : VCC_CSI1

However the A20 datasheet doesn't label them this way, instead using:

F19 : VCC-PE : Port E Power Supply
E18 : VCC-PG : Port G Power Supply

no mention at all of CSI. CSI just happens to be a function that can be
multiplexed onto ports E & G.

So lets assume for a moment we don't have a CSI device, or a CSI driver
and are not using those pins for CSI functions but are instead using
their default GPIO purpose, or in the case of PG perhaps for UART3
or UART4.

What happens when you disable the regulator? Hopefully from the above
you'll already have worked out that you lose ports E & G when you turn
their I/O power supply off and anything multiplexed onto those ports
becomes unuseable.

I haven't checked other schematics, but all of the Olimex A10/A20
boards along with Cubieboard & CubieTruck label these pins as VCC_CSIn
It would be my guess that there's a reference design being given to
manufacturers, perhaps the original Cubieboard even, that makes this
mistake and everyone else has blindly copied it.

So, unless we want to lose access to 24 GPIO pins for no real reason
I'd suggest that the u-boot patch is reverted and/or the dts is altered
such that the gpio driver claims the regulators and prevents them being
turned off.

The u-boot commit message and reasoning given on list are trivially
proved incorrect. Ports E & G are not CSI-only, losing all other
functions on these ports is not acceptable.

I've also just tested building u-boot with LDO3 & 4 voltages set to
3.3v to be the same as CubieBoard, this works for lime2.
So it's unlikely that the 2.8v & 2.3v settings from my previous patch
make sense. The crash is actually being caused by something else,
probably sequencing, due to unnecessarily turning the regulators off.
Whatever the reason, we don't have enough information in the
datasheets to know for sure so we simply shouldn't turn them off to
begin with.

The people behind this need to take responsibility and clean up their
mess.

Iain

[Michael Haas]

Post by Iain Paton
Having reread the A20 datasheet yesterday, something in the back of my
mind was bothering me overnight so I took some time to check this morning.
F19 : VCC_CSI0
E18 : VCC_CSI1
F19 : VCC-PE : Port E Power Supply
E18 : VCC-PG : Port G Power Supply
no mention at all of CSI. CSI just happens to be a function that can be
multiplexed onto ports E & G.
So lets assume for a moment we don't have a CSI device, or a CSI driver
and are not using those pins for CSI functions but are instead using
their default GPIO purpose, or in the case of PG perhaps for UART3
or UART4.
What happens when you disable the regulator? Hopefully from the above
you'll already have worked out that you lose ports E & G when you turn
their I/O power supply off and anything multiplexed onto those ports
becomes unuseable.

Hi Iain,

I took some time to go through the data sheets and your analysis is on
point.

I did raise an issue with Olimex via github [0]. Maybe they will update
their schematics.


I do have my breakout board connected to GPIO-1.. and it turns out that
mostly port G is exposed on GPIO-1. I'll be reverting Hans' patch locally
so I can toggle some pins today. I am sure he will contribute his own
thoughts on the matter.

Post by Iain Paton
I've also just tested building u-boot with LDO3 & 4 voltages set to
3.3v to be the same as CubieBoard, this works for lime2.
So it's unlikely that the 2.8v & 2.3v settings from my previous patch
make sense. The crash is actually being caused by something else,
probably sequencing, due to unnecessarily turning the regulators off.
Whatever the reason, we don't have enough information in the
datasheets to know for sure so we simply shouldn't turn them off to
begin with.

I just tried that on linux and axp20x-regulator.ko refuses to set the
voltage
on ldo4:

[ 88.914653] at24 1-0050: 2048 byte 24c16 EEPROM, writable, 16 bytes/write
[ 96.893312] axp20x-i2c 0-0034: AXP20x variant AXP209 found
[ 96.910631] axp20x-i2c 0-0034: AXP20X driver loaded
[ 96.955579] input: axp20x-pek as
/devices/platform/***@01c00000/1c2ac00.i2c/i2c-0/0-0034/axp20x-pek/input/input0
[ 96.971734] ldo4: failed to apply 3300000uV constraint(-22)
[ 96.990646] axp20x-regulator axp20x-regulator: Failed to register ldo4
[ 97.012686] axp20x-regulator: probe of axp20x-regulator failed with
error -22

Perhaps this requires another patch [1], but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Michael

[0] https://github.com/OLIMEX/OLINUXINO/issues/35
[1] https://www.spinics.net/lists/devicetree/msg118871.html

[Hans de Goede]

Hi,

Post by Michael Haas

Post by Iain Paton
Having reread the A20 datasheet yesterday, something in the back of my
mind was bothering me overnight so I took some time to check this morning.
F19 : VCC_CSI0
E18 : VCC_CSI1
F19 : VCC-PE : Port E Power Supply
E18 : VCC-PG : Port G Power Supply
no mention at all of CSI. CSI just happens to be a function that can be
multiplexed onto ports E & G.
So lets assume for a moment we don't have a CSI device, or a CSI driver
and are not using those pins for CSI functions but are instead using
their default GPIO purpose, or in the case of PG perhaps for UART3
or UART4.
What happens when you disable the regulator? Hopefully from the above
you'll already have worked out that you lose ports E & G when you turn
their I/O power supply off and anything multiplexed onto those ports
becomes unuseable.

Hi Iain,
I took some time to go through the data sheets and your analysis is on
point.
I did raise an issue with Olimex via github [0]. Maybe they will update
their schematics.
I do have my breakout board connected to GPIO-1.. and it turns out that
mostly port G is exposed on GPIO-1. I'll be reverting Hans' patch locally
so I can toggle some pins today. I am sure he will contribute his own
thoughts on the matter.

Post by Iain Paton
I've also just tested building u-boot with LDO3 & 4 voltages set to
3.3v to be the same as CubieBoard, this works for lime2.
So it's unlikely that the 2.8v & 2.3v settings from my previous patch
make sense. The crash is actually being caused by something else,
probably sequencing, due to unnecessarily turning the regulators off.
Whatever the reason, we don't have enough information in the
datasheets to know for sure so we simply shouldn't turn them off to
begin with.

I just tried that on linux and axp20x-regulator.ko refuses to set the
voltage
[ 88.914653] at24 1-0050: 2048 byte 24c16 EEPROM, writable, 16 bytes/write
[ 96.893312] axp20x-i2c 0-0034: AXP20x variant AXP209 found
[ 96.910631] axp20x-i2c 0-0034: AXP20X driver loaded
[ 96.955579] input: axp20x-pek as
[ 96.971734] ldo4: failed to apply 3300000uV constraint(-22)
[ 96.990646] axp20x-regulator axp20x-regulator: Failed to register ldo4
[ 97.012686] axp20x-regulator: probe of axp20x-regulator failed with
error -22
Perhaps this requires another patch [1],

Yes setting the constraints to 3.3v requires the patch you link for things to work.

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling ldo3/4 by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

Instead please submit a patch for configs/A20-OLinuXino-Lime2_defconfig which
configures them at 2800 mV there (per the Lime2 schematic).

It would be good to also check the Lime1 schematic if they are used in the same
way there, I would also welcome a patch for:

configs/A10-OLinuXino-Lime_defconfig
configs/A20-OLinuXino-Lime_defconfig

Regards,

Hans

[Maxime Ripard]

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling ldo3/4 by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.
Instead please submit a patch for configs/A20-OLinuXino-Lime2_defconfig which
configures them at 2800 mV there (per the Lime2 schematic).
It would be good to also check the Lime1 schematic if they are used in the same
configs/A10-OLinuXino-Lime_defconfig
configs/A20-OLinuXino-Lime_defconfig

Adding a comment to the DT summing up that whole discussion would be
great too.

Maxime

[Michael Haas]

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling ldo3/4 by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

I just checked the Banana Pi, which exposes ports from bank E on its
CON1. Admittedly,
that's the "Camera Serial Interface", but that connector is explicitly
also labelled
as supporting various GPIO pins: http://www.bananapi.org/p/product.html

On the Cubietruck, there's also CN9, exposing some GPIO ports from bank G:
https://linux-sunxi.org/Cubietech_Cubietruck

Given that there tend to be crashes on re-enabling LDO3+LDO4 and many
boards expose affected pins (not only for CSI!): wouldn't be it be the
best to
keep them enabled?

FWIW, Olimex did reply to my inquiry regarding misleading pin names:
https://github.com/OLIMEX/OLINUXINO/issues/35#issuecomment-202823804

Post by Hans de Goede
<http://dl.linux-sunxi.org/A20/a20_pad_std_v1_1.pdf> - if you inspect
the schematic you'd notice that the signals connected to E18 and E19
are named respectively CSI1-VCC and CSI0-VCC. Hence, this is where the
wrong naming of our processor pins came from. I'll add the proposal
for correction for the future hardware revision of the boards affected.
The difference between our boards and other boards come from the
simple fact that their designs use 3.3V cameras, while we followed the
original reference design and used 2.8V camera in A20-SOM-EVB (we even
used the same camera - GT2005).

Michael

[Hans de Goede]

Hi,

Post by Michael Haas

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling ldo3/4 by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

I just checked the Banana Pi, which exposes ports from bank E on its
CON1. Admittedly,
that's the "Camera Serial Interface", but that connector is explicitly
also labelled
as supporting various GPIO pins: http://www.bananapi.org/p/product.html

That connector is not usable unless you've got a really special cable,
also ldo3/4 are disabled in the kernel dts, so it does not matter what u-boot
does.

Post by Michael Haas
https://linux-sunxi.org/Cubietech_Cubietruck

The Cubietruck also has ldo3/4 disabled in the kernel dts.

Post by Michael Haas
Given that there tend to be crashes on re-enabling LDO3+LDO4 and many
boards expose affected pins (not only for CSI!): wouldn't be it be the
best to keep them enabled?

s/Many boards/Lime 2/ and we still do not fully understand what is going
on there, we just know that not enabling ldo3/4 at the u-boot level
causes issues, we do not know / understand why though.

U-boot supports 52 boards with an axp209 pmic, yet you only name a select
few which need ldo3/4 according to you, and for 2 of your examples the
kernel dts says differently.

I'm not going to enable ldo3/4 by default on all boards just because they
are needed on the Lime 2.

Regards,

Hans

[Michael Haas]

Post by Hans de Goede
Hi,

Post by Michael Haas

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling

ldo3/4

Post by Michael Haas

Post by Hans de Goede
by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

I just checked the Banana Pi, which exposes ports from bank E on its
CON1. Admittedly,
that's the "Camera Serial Interface", but that connector is

explicitly

Post by Michael Haas
also labelled

http://www.bananapi.org/p/product.html
That connector is not usable unless you've got a really special cable,
also ldo3/4 are disabled in the kernel dts, so it does not matter what u-boot
does.

Post by Michael Haas
On the Cubietruck, there's also CN9, exposing some GPIO ports from
https://linux-sunxi.org/Cubietech_Cubietruck

The Cubietruck also has ldo3/4 disabled in the kernel dts.

Post by Michael Haas
Given that there tend to be crashes on re-enabling LDO3+LDO4 and many
boards expose affected pins (not only for CSI!): wouldn't be it be

the

Post by Michael Haas
best to keep them enabled?

s/Many boards/Lime 2/ and we still do not fully understand what is going
on there, we just know that not enabling ldo3/4 at the u-boot level
causes issues, we do not know / understand why though.
U-boot supports 52 boards with an axp209 pmic, yet you only name a select
few which need ldo3/4 according to you, and for 2 of your examples the
kernel dts says differently.
I'm not going to enable ldo3/4 by default on all boards just because they
are needed on the Lime 2.
Regards,
Hans

Hi Hans,

You are making an excellent point. I will submit patches for the lime* boards as proposed in your earlier mail.

Just to ask for some clarification: you are saying the regulators are disabled in the kernel dts files.
I looked at the cubietruck dts files and the regulators are not mentioned in there. If they are not mentioned, are they turned off or do they remain in their default state, e.g the way that u-boot left them?

Stepping back from kernel/u-boot issues, would it make sense to also turn on the regulators for the cubietruck? Perhaps in u-boot and the kernel?

Michael

[Hans de Goede]

HI,

Post by Michael Haas

Post by Hans de Goede
Hi,

Post by Michael Haas

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling

ldo3/4

Post by Michael Haas

Post by Hans de Goede
by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

I just checked the Banana Pi, which exposes ports from bank E on its
CON1. Admittedly,
that's the "Camera Serial Interface", but that connector is

explicitly

Post by Michael Haas
also labelled

http://www.bananapi.org/p/product.html
That connector is not usable unless you've got a really special cable,
also ldo3/4 are disabled in the kernel dts, so it does not matter what u-boot
does.

Post by Michael Haas
On the Cubietruck, there's also CN9, exposing some GPIO ports from
https://linux-sunxi.org/Cubietech_Cubietruck

The Cubietruck also has ldo3/4 disabled in the kernel dts.

Post by Michael Haas
Given that there tend to be crashes on re-enabling LDO3+LDO4 and many
boards expose affected pins (not only for CSI!): wouldn't be it be

the

Post by Michael Haas
best to keep them enabled?

s/Many boards/Lime 2/ and we still do not fully understand what is going
on there, we just know that not enabling ldo3/4 at the u-boot level
causes issues, we do not know / understand why though.
U-boot supports 52 boards with an axp209 pmic, yet you only name a select
few which need ldo3/4 according to you, and for 2 of your examples the
kernel dts says differently.
I'm not going to enable ldo3/4 by default on all boards just because they
are needed on the Lime 2.
Regards,
Hans

Hi Hans,
You are making an excellent point. I will submit patches for the lime* boards as proposed in your earlier mail.
Just to ask for some clarification: you are saying the regulators are disabled in the kernel dts files.
I looked at the cubietruck dts files and the regulators are not mentioned in there. If they are not mentioned, are they turned off or do they remain in their default state, e.g the way that u-boot left them?

The kernel will turn off any unused regulators in a late_init call (just
before calling /sbin/init), so not listed regulators will get turned off.

Post by Michael Haas
Stepping back from kernel/u-boot issues, would it make sense to also turn on the regulators for the cubietruck? Perhaps in u-boot and the kernel?

Have you checked the schematics to see how portG is powered on the CubieTruck?

It would only make sense to turn on ldo3 or ldo4 if it is actually used
to power an exposed port.

Regards,

Hans

[Hans de Goede]

Hi,

Post by Michael Haas

Post by Hans de Goede
Hi,

Post by Michael Haas

Post by Hans de Goede

Post by Michael Haas
but unless there's a real need
for me to verify this,
I'll simply revert the u-boot patch and enjoy working GPIO.

Given the entire discussion in this thread, I agree that fixing this in u-boot
is best. But I do not see reverting the u-boot patch disabling

ldo3/4

Post by Michael Haas

Post by Hans de Goede
by default
as the solution. ldo3/4 are unused on most boards, so disabling them by default
clearly is the right thing todo IMHO.

I just checked the Banana Pi, which exposes ports from bank E on its
CON1. Admittedly,
that's the "Camera Serial Interface", but that connector is

explicitly

Post by Michael Haas
also labelled

http://www.bananapi.org/p/product.html
That connector is not usable unless you've got a really special cable,
also ldo3/4 are disabled in the kernel dts, so it does not matter what u-boot
does.

Post by Michael Haas
On the Cubietruck, there's also CN9, exposing some GPIO ports from
https://linux-sunxi.org/Cubietech_Cubietruck

The Cubietruck also has ldo3/4 disabled in the kernel dts.

Post by Michael Haas
Given that there tend to be crashes on re-enabling LDO3+LDO4 and many
boards expose affected pins (not only for CSI!): wouldn't be it be

the

Post by Michael Haas
best to keep them enabled?

s/Many boards/Lime 2/ and we still do not fully understand what is going
on there, we just know that not enabling ldo3/4 at the u-boot level
causes issues, we do not know / understand why though.
U-boot supports 52 boards with an axp209 pmic, yet you only name a select
few which need ldo3/4 according to you, and for 2 of your examples the
kernel dts says differently.
I'm not going to enable ldo3/4 by default on all boards just because they
are needed on the Lime 2.
Regards,
Hans

Hi Hans,
You are making an excellent point. I will submit patches for the lime* boards as proposed in your earlier mail.

Note I've just added a patch with the necessary changes to my tree, so there
is no need to submit a patch for this.

Regards,

Hans

[Michael Haas]

Post by Iain Paton

Post by Michael Haas
This patch implements the suggested changes. It is not enough
to set the voltage to 2800000 to avoid the crash. Hence, I have
also removed regulator-always-on.

NAK.
If setting both min and max to 2800000 isn't working then I'd suggest you're
just papering over the issue and not actually solving it.
It should be fairly clear that whenever those regulators are turned on by
some driver you're going to see the same issue, you've just delayed the
problem by removing the regulator-always-on item.
The real problem here is people monkeying with the regulators without
understanding what they're doing, bothering to test the results, or caring
about the consequences.
Anyway, Hans was correct in that the datasheet we have access to says that
the defaults for these regulators are undefined.
I suspect that's not the whole story though, if these regulators had random
values at powerup we'd have seen problems before now.
Most likely the values we think we know are simply incorrect plus there's
probably some element of sequencing involved - perhaps LDO4 needs turned on
first.
However I don't see sufficient detail in the datasheets to confirm any of
that, so we shouldn't assume anything.
Electrically on the lime2 LDO3 goes to VCC-PE and LDO4 to VCC-PG. Neither
groups E or G appear to be connected to anything other than gpio headers,
so this has little to do with board design, it's something internal to
the SoC that the datasheet is lacking on detail about.
One of the reasons for the regulator-always-on being in the dts was so that
the regulator didn't get touched and so didn't break anything.
All said and done, it only takes a minute with a multimeter to determine
actual defaults. Helpful to have multiple boards to gain some confidence
that they behave the same, but still basing values like this on tiny
sample sizes is probably unwise.
Looking at other A20 based boards, Cubieboards for example, shows LDO3/4
being unused and the VCC-PE/PG pins they power on the lime2 being tied
directly to 3.3v. No issues there simply because there's no way to mess
with the power to those blocks.
The patch you really want is something like the one below which works
on my boards regardless of the presence of regulator-always-on.
Verified on four boards with u-boot 2016.03 and kernel 4.5.0.
Whether you decide to remove regulator-always-on or not is a different
question, but please use 2.3v, or some other proven working value, for
ldo3. At least that way you have a fighting chance of the board
remaining functional when something wants to turn the regulator on.

Hi Iain,

thanks for the feedback.

I have yet to try your patch, but I have broken out the multimeter to
measure LDO3. On GPIO-2, I'm measuring 2.8V between pins 2 and 4 instead
of 2.3V you seem to measuring.
In my setup, I have simply removed the SD card from the lime2 to get
whatever defaults are used at boot.

How did you measure LDO3? Also, did you find any pins for LDO4 besides
the AXP209 itelf? Mine seems to be covered in flux.

Michael

[s***@gmail.com]

Post by Michael Haas
Hi Iain,
thanks for the feedback.
I have yet to try your patch, but I have broken out the multimeter to
measure LDO3. On GPIO-2, I'm measuring 2.8V between pins 2 and 4 instead
of 2.3V you seem to measuring.
In my setup, I have simply removed the SD card from the lime2 to get
whatever defaults are used at boot.
How did you measure LDO3? Also, did you find any pins for LDO4 besides
the AXP209 itelf? Mine seems to be covered in flux.

I measured them at C106 & C107 decoupling caps near the SoC. On my boards
they're on the bottom, in the square of caps below the SoC, corner nearest
the uSD card.

LDO3 I also measured at GPIO-2 pins 2 & 4 with a breakout board connected
as I was concerned that 2.3v at the cap was wrong.

I can easily imagine you might get different values, the datasheet
has them as undefined after all. However I'd suggest a power cycle or
use a version of u-boot from before the patches went in to set LDO3/4
to 2.8v. Soft reboot or such like probably leaves the values previously
programmed by whatever last touched them.

Even then, different batch of AXP209, or a different boot0 in the A20
could lead to different results. The danger is that we're missing
some piece of information that means there's no safe way to turn these
regulators off then back on without the lockup. Suggest that if you do
get different results from me, then it's even more likely we don't know
what's going on and something that works on my board may not on yours.

Anyway, you probably want to read my second follow up. Turning these off
is the wrong thing to do regardless. The inputs to the SoC are mis-labeled
in the schematics, disabling them disables all io pins in groups E & G
whether they're configured for CSI tasks or not.

Iain