[PATCH v2 01/14] ARM: clk: sunxi: Add driver for the H3 THS clock

[PATCH v2 01/14] ARM: clk: sunxi: Add driver for the H3 THS clock

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>

This patch adds a driver for the THS clock which is present on the
Allwinner H3.

Signed-off-by: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>
---
 Documentation/devicetree/bindings/clock/sunxi.txt |  1 +
 drivers/clk/sunxi/Makefile                        |  1 +
 drivers/clk/sunxi/clk-h3-ths.c                    | 98 +++++++++++++++++++++++
 3 files changed, 100 insertions(+)
 create mode 100644 drivers/clk/sunxi/clk-h3-ths.c

diff --git a/Documentation/devicetree/bindings/clock/sunxi.txt b/Documentation/devicetree/bindings/clock/sunxi.txt
index 8f7619d..5faae05 100644
--- a/Documentation/devicetree/bindings/clock/sunxi.txt
+++ b/Documentation/devicetree/bindings/clock/sunxi.txt
@@ -87,6 +87,7 @@ Required properties:
 	"allwinner,sun9i-a80-usb-phy-clk" - for usb phy gates + resets on A80
 	"allwinner,sun4i-a10-ve-clk" - for the Video Engine clock
 	"allwinner,sun6i-a31-display-clk" - for the display clocks
+	"allwinner,sun8i-h3-ths-clk" - for THS on H3
 
 Required properties for all clocks:
 - reg : shall be the control register address for the clock.
diff --git a/drivers/clk/sunxi/Makefile b/drivers/clk/sunxi/Makefile
index 39d2044..8e245e3 100644
--- a/drivers/clk/sunxi/Makefile
+++ b/drivers/clk/sunxi/Makefile
@@ -9,6 +9,7 @@ obj-y += clk-a10-mod1.o
 obj-y += clk-a10-pll2.o
 obj-y += clk-a10-ve.o
 obj-y += clk-a20-gmac.o
+obj-y += clk-h3-ths.o
 obj-y += clk-mod0.o
 obj-y += clk-simple-gates.o
 obj-y += clk-sun4i-display.o
diff --git a/drivers/clk/sunxi/clk-h3-ths.c b/drivers/clk/sunxi/clk-h3-ths.c
new file mode 100644
index 0000000..c1d6d32
--- /dev/null
+++ b/drivers/clk/sunxi/clk-h3-ths.c
@@ -0,0 +1,98 @@
+/*
+ * sun8i THS clock driver
+ *
+ * Copyright (C) 2015 Josef Gajdusek
+ *
+ * This software is licensed under the terms of the GNU General Public
+ * License version 2, as published by the Free Software Foundation, and
+ * may be copied, distributed, and modified under those terms.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ */
+
+#include <linux/clk-provider.h>
+#include <linux/of_address.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#define SUN8I_H3_THS_CLK_ENABLE				31
+#define SUN8I_H3_THS_CLK_DIVIDER_SHIFT		0
+#define SUN8I_H3_THS_CLK_DIVIDER_WIDTH		2
+
+static DEFINE_SPINLOCK(sun8i_h3_ths_clk_lock);
+
+static const struct clk_div_table sun8i_h3_ths_clk_table[] __initconst = {
+	{ .val = 0, .div = 1 },
+	{ .val = 1, .div = 2 },
+	{ .val = 2, .div = 4 },
+	{ .val = 3, .div = 6 },
+	{ } /* sentinel */
+};
+
+static void __init sun8i_h3_ths_clk_setup(struct device_node *node)
+{
+	struct clk *clk;
+	struct clk_gate *gate;
+	struct clk_divider *div;
+	const char *parent;
+	const char *clk_name = node->name;
+	void __iomem *reg;
+	int err;
+
+	reg = of_io_request_and_map(node, 0, of_node_full_name(node));
+
+	if (IS_ERR(reg))
+		return;
+
+	gate = kzalloc(sizeof(*gate), GFP_KERNEL);
+	if (!gate)
+		goto err_unmap;
+
+	div = kzalloc(sizeof(*gate), GFP_KERNEL);
+	if (!div)
+		goto err_gate_free;
+
+	of_property_read_string(node, "clock-output-names", &clk_name);
+	parent = of_clk_get_parent_name(node, 0);
+
+	gate->reg = reg;
+	gate->bit_idx = SUN8I_H3_THS_CLK_ENABLE;
+	gate->lock = &sun8i_h3_ths_clk_lock;
+
+	div->reg = reg;
+	div->shift = SUN8I_H3_THS_CLK_DIVIDER_SHIFT;
+	div->width = SUN8I_H3_THS_CLK_DIVIDER_WIDTH;
+	div->table = sun8i_h3_ths_clk_table;
+	div->lock = &sun8i_h3_ths_clk_lock;
+
+	clk = clk_register_composite(NULL, clk_name, &parent, 1,
+								 NULL, NULL,
+								 &div->hw, &clk_divider_ops,
+								 &gate->hw, &clk_gate_ops,
+								 CLK_SET_RATE_PARENT);
+
+	if (IS_ERR(clk))
+		goto err_div_free;
+
+	err = of_clk_add_provider(node, of_clk_src_simple_get, clk);
+	if (err)
+		goto err_unregister_clk;
+
+	return;
+
+err_unregister_clk:
+	clk_unregister(clk);
+err_gate_free:
+	kfree(gate);
+err_div_free:
+	kfree(div);
+err_unmap:
+	iounmap(reg);
+}
+
+CLK_OF_DECLARE(sun8i_h3_ths_clk, "allwinner,sun8i-h3-ths-clk",
+			   sun8i_h3_ths_clk_setup);
-- 
2.9.0

Re: [PATCH v2 01/14] ARM: clk: sunxi: Add driver for the H3 THS clock

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 673 bytes --]

Hi,

On Sat, Jun 25, 2016 at 05:44:58AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>
> 
> This patch adds a driver for the THS clock which is present on the
> Allwinner H3.
>
> Signed-off-by: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>

You might not have noticed, but we are currently rewriting the whole
clock support for the Allwinner SoCs, targetting the H3 as the first
SoC to use it.

There's already some support for the H3 THS clock, so please use that
instead.

Thanks!
Maxime

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

Re: [PATCH v2 01/14] ARM: clk: sunxi: Add driver for the H3 THS clock

[Ondřej Jirman]

Hi Maxime,

I try to base everything on the torvalds's kernel.

I did notice the patches. Is there some main git tree/branch where this
work is tracked in? I'd gladly use it.

Also there's a PLL1 rate application patch, that would need to be ported
to the new CCU code, in the case I would use it as a base for this work.
Given that the new CCU code is your work and it's fresh in your mind, do
you have suggestion how to approach it?

It is [PATCH v2 06/14] in this series.

regards,
  o.

On 25.6.2016 09:13, Maxime Ripard wrote:
> Hi,
> 
> On Sat, Jun 25, 2016 at 05:44:58AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
>> From: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>
>>
>> This patch adds a driver for the THS clock which is present on the
>> Allwinner H3.
>>
>> Signed-off-by: Josef Gajdusek <atx-MwjtXicnQwU@public.gmane.org>
> 
> You might not have noticed, but we are currently rewriting the whole
> clock support for the Allwinner SoCs, targetting the H3 as the first
> SoC to use it.
> 
> There's already some support for the H3 THS clock, so please use that
> instead.
> 
> Thanks!
> Maxime
> 

Re: [PATCH v2 01/14] ARM: clk: sunxi: Add driver for the H3 THS clock

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1235 bytes --]

Hi,

On Sat, Jun 25, 2016 at 05:23:12PM +0200, Ondřej Jirman wrote:
> Hi Maxime,
> 
> I try to base everything on the torvalds's kernel.
> 
> I did notice the patches. Is there some main git tree/branch where this
> work is tracked in? I'd gladly use it.

I just pushed it, branch sunxi/pen/clk-rework, on my github repo:
https://github.com/mripard/linux

> Also there's a PLL1 rate application patch, that would need to be ported
> to the new CCU code, in the case I would use it as a base for this work.
> Given that the new CCU code is your work and it's fresh in your mind, do
> you have suggestion how to approach it?
> 
> It is [PATCH v2 06/14] in this series.

Yes, it's still in my queue of things to review, let's discuss this on
this patch thread.

Thanks!
Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[PATCH v2 03/14] dt-bindings: document sun8i_ths - H3 thermal sensor driver

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

This patch adds the binding documentation for the
sun8i_ths driver. This is a driver for thermal sensor
found in Allwinner H3 SoC.

Signed-off-by: Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 .../devicetree/bindings/thermal/sun8i-ths.txt      | 26 ++++++++++++++++++++++
 1 file changed, 26 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/thermal/sun8i-ths.txt

diff --git a/Documentation/devicetree/bindings/thermal/sun8i-ths.txt b/Documentation/devicetree/bindings/thermal/sun8i-ths.txt
new file mode 100644
index 0000000..76859d4
--- /dev/null
+++ b/Documentation/devicetree/bindings/thermal/sun8i-ths.txt
@@ -0,0 +1,26 @@
+* Thermal sensor driver for Allwinner H3 SoC
+
+Required properties:
+- compatible : "allwinner,sun8i-h3-ths"
+- reg : Address range of the thermal sensor registers and of the calibration
+  data
+- resets : Must contain phandles to reset controls matching the entries
+  of the names
+- reset-names : Must include the name "ahb"
+- clocks : Must contain phandles to clock controls matching the entries
+  of the names
+- clock-names : Must contain "ahb" for the bus gate and "ths" for the THS
+  clock
+
+Example:
+ths: ths@01c25000 {
+	#thermal-sensor-cells = <0>;
+	compatible = "allwinner,sun8i-h3-ths";
+	reg = <0x01c25000 0x400>,
+	      <0x01c14234 0x4>;
+	interrupts = <GIC_SPI 63 IRQ_TYPE_LEVEL_HIGH>;
+	resets = <&bus_rst 136>;
+	reset-names = "ahb";
+	clocks = <&bus_gates 72>, <&ths_clk>;
+	clock-names = "ahb", "ths";
+};
-- 
2.9.0

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

Re: [PATCH v2 03/14] dt-bindings: document sun8i_ths - H3 thermal sensor driver

[Rob Herring]

On Sat, Jun 25, 2016 at 05:45:00AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> 
> This patch adds the binding documentation for the
> sun8i_ths driver. This is a driver for thermal sensor
> found in Allwinner H3 SoC.
> 
> Signed-off-by: Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> ---
>  .../devicetree/bindings/thermal/sun8i-ths.txt      | 26 ++++++++++++++++++++++
>  1 file changed, 26 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/thermal/sun8i-ths.txt

I guess the example will change for the clocks some, but otherwise:

Acked-by: Rob Herring <robh-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[PATCH v2 05/14] dt-bindings: document SY8106A regulator driver

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

This patch adds the binding documentation for the
sy8106a regulator driver.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 .../bindings/regulator/sy8106a-regulator.txt        | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/regulator/sy8106a-regulator.txt

diff --git a/Documentation/devicetree/bindings/regulator/sy8106a-regulator.txt b/Documentation/devicetree/bindings/regulator/sy8106a-regulator.txt
new file mode 100644
index 0000000..1e623a34
--- /dev/null
+++ b/Documentation/devicetree/bindings/regulator/sy8106a-regulator.txt
@@ -0,0 +1,21 @@
+SY8106A Voltage regulator
+
+Required properties:
+- compatible: Must be "silergy,sy8106a"
+- reg: I2C slave address - must be <0x65>
+
+Any property defined as part of the core regulator binding, defined in
+regulator.txt, can also be used.
+
+Example:
+
+	sy8106a {
+		compatible = "silergy,sy8106a";
+		reg = <0x65>;
+		regulator-name = "sy8106a-vdd";
+		regulator-min-microvolt = <1000000>;
+		regulator-max-microvolt = <1400000>;
+		regulator-ramp-delay = <200>;
+		regulator-boot-on;
+		regulator-always-on;
+	};
-- 
2.9.0

Re: [PATCH v2 05/14] dt-bindings: document SY8106A regulator driver

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 395 bytes --]

On Sat, Jun 25, 2016 at 05:45:02AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> 
> This patch adds the binding documentation for the
> sy8106a regulator driver.

Please submit patches using subject lines reflecting the style for the
subsystem.  This makes it easier for people to identify relevant
patches.

Re: [PATCH v2 05/14] dt-bindings: document SY8106A regulator driver

[Ondřej Jirman]

On 26.6.2016 13:27, Mark Brown wrote:
> On Sat, Jun 25, 2016 at 05:45:02AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>
>> This patch adds the binding documentation for the
>> sy8106a regulator driver.
> 
> Please submit patches using subject lines reflecting the style for the
> subsystem.  This makes it easier for people to identify relevant
> patches.
> 

Hello, I assume that means "regulator: dt-bindings: ..." in this case.
I'll keep that in mind in the future. Thank you.

Re: [PATCH v2 05/14] dt-bindings: document SY8106A regulator driver

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 630 bytes --]

On Sun, Jun 26, 2016 at 05:10:06PM +0200, Ondřej Jirman wrote:

> Hello, I assume that means "regulator: dt-bindings: ..." in this case.
> I'll keep that in mind in the future. Thank you.

Yes, or "regulator: sy8106a: " - as ever take a look at how other
commits in the same file/directory are done.

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

[PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

PLL1 on H3 requires special factors application algorithm,
when the rate is changed. This algorithm was extracted
from the arisc code that handles frequency scaling
in the BSP kernel.

This commit adds optional apply function to
struct factors_data, that can implement non-trivial
factors application method, when necessary.

Also struct clk_factors_config is extended with position
of the PLL lock flag.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 Documentation/devicetree/bindings/clock/sunxi.txt |  1 +
 drivers/clk/sunxi/clk-factors.c                   | 34 +++++------
 drivers/clk/sunxi/clk-factors.h                   | 12 ++++
 drivers/clk/sunxi/clk-sunxi.c                     | 72 ++++++++++++++++++++++-
 4 files changed, 98 insertions(+), 21 deletions(-)

diff --git a/Documentation/devicetree/bindings/clock/sunxi.txt b/Documentation/devicetree/bindings/clock/sunxi.txt
index 5faae05..774500c 100644
--- a/Documentation/devicetree/bindings/clock/sunxi.txt
+++ b/Documentation/devicetree/bindings/clock/sunxi.txt
@@ -10,6 +10,7 @@ Required properties:
 	"allwinner,sun4i-a10-pll1-clk" - for the main PLL clock and PLL4
 	"allwinner,sun6i-a31-pll1-clk" - for the main PLL clock on A31
 	"allwinner,sun8i-a23-pll1-clk" - for the main PLL clock on A23
+	"allwinner,sun8i-h3-pll1-clk" - for the main PLL clock on H3
 	"allwinner,sun4i-a10-pll3-clk" - for the video PLL clock on A10
 	"allwinner,sun9i-a80-pll4-clk" - for the peripheral PLLs on A80
 	"allwinner,sun4i-a10-pll5-clk" - for the PLL5 clock
diff --git a/drivers/clk/sunxi/clk-factors.c b/drivers/clk/sunxi/clk-factors.c
index ddefe96..7c165db 100644
--- a/drivers/clk/sunxi/clk-factors.c
+++ b/drivers/clk/sunxi/clk-factors.c
@@ -34,13 +34,6 @@
 
 #define FACTORS_MAX_PARENTS		5
 
-#define SETMASK(len, pos)		(((1U << (len)) - 1) << (pos))
-#define CLRMASK(len, pos)		(~(SETMASK(len, pos)))
-#define FACTOR_GET(bit, len, reg)	(((reg) & SETMASK(len, bit)) >> (bit))
-
-#define FACTOR_SET(bit, len, reg, val) \
-	(((reg) & CLRMASK(len, bit)) | (val << (bit)))
-
 static unsigned long clk_factors_recalc_rate(struct clk_hw *hw,
 					     unsigned long parent_rate)
 {
@@ -150,20 +143,24 @@ static int clk_factors_set_rate(struct clk_hw *hw, unsigned long rate,
 	if (factors->lock)
 		spin_lock_irqsave(factors->lock, flags);
 
-	/* Fetch the register value */
-	reg = readl(factors->reg);
+	if (factors->apply) {
+		factors->apply(factors, &req);
+	} else {
+		/* Fetch the register value */
+		reg = readl(factors->reg);
 
-	/* Set up the new factors - macros do not do anything if width is 0 */
-	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req.n);
-	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req.k);
-	reg = FACTOR_SET(config->mshift, config->mwidth, reg, req.m);
-	reg = FACTOR_SET(config->pshift, config->pwidth, reg, req.p);
+		/* Set up the new factors - macros do not do anything if width is 0 */
+		reg = FACTOR_SET(config->nshift, config->nwidth, reg, req.n);
+		reg = FACTOR_SET(config->kshift, config->kwidth, reg, req.k);
+		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req.m);
+		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req.p);
 
-	/* Apply them now */
-	writel(reg, factors->reg);
+		/* Apply them now */
+		writel(reg, factors->reg);
 
-	/* delay 500us so pll stabilizes */
-	__delay((rate >> 20) * 500 / 2);
+		/* delay 500us so pll stabilizes */
+		__delay((rate >> 20) * 500 / 2);
+	}
 
 	if (factors->lock)
 		spin_unlock_irqrestore(factors->lock, flags);
@@ -213,6 +210,7 @@ struct clk *sunxi_factors_register(struct device_node *node,
 	factors->config = data->table;
 	factors->get_factors = data->getter;
 	factors->recalc = data->recalc;
+	factors->apply = data->apply;
 	factors->lock = lock;
 
 	/* Add a gate if this factor clock can be gated */
diff --git a/drivers/clk/sunxi/clk-factors.h b/drivers/clk/sunxi/clk-factors.h
index 1e63c5b..661a45a 100644
--- a/drivers/clk/sunxi/clk-factors.h
+++ b/drivers/clk/sunxi/clk-factors.h
@@ -6,6 +6,13 @@
 
 #define SUNXI_FACTORS_NOT_APPLICABLE	(0)
 
+#define SETMASK(len, pos)		(((1U << (len)) - 1) << (pos))
+#define CLRMASK(len, pos)		(~(SETMASK(len, pos)))
+#define FACTOR_GET(bit, len, reg)	(((reg) & SETMASK(len, bit)) >> (bit))
+
+#define FACTOR_SET(bit, len, reg, val) \
+	(((reg) & CLRMASK(len, bit)) | (val << (bit)))
+
 struct clk_factors_config {
 	u8 nshift;
 	u8 nwidth;
@@ -16,6 +23,7 @@ struct clk_factors_config {
 	u8 pshift;
 	u8 pwidth;
 	u8 n_start;
+	u8 lock;
 };
 
 struct factors_request {
@@ -28,6 +36,8 @@ struct factors_request {
 	u8 p;
 };
 
+struct clk_factors;
+
 struct factors_data {
 	int enable;
 	int mux;
@@ -35,6 +45,7 @@ struct factors_data {
 	const struct clk_factors_config *table;
 	void (*getter)(struct factors_request *req);
 	void (*recalc)(struct factors_request *req);
+	void (*apply)(struct clk_factors *factors, struct factors_request *req);
 	const char *name;
 };
 
@@ -44,6 +55,7 @@ struct clk_factors {
 	const struct clk_factors_config *config;
 	void (*get_factors)(struct factors_request *req);
 	void (*recalc)(struct factors_request *req);
+	void (*apply)(struct clk_factors *factors, struct factors_request *req);
 	spinlock_t *lock;
 	/* for cleanup */
 	struct clk_mux *mux;
diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
index 838b22a..e4bb908 100644
--- a/drivers/clk/sunxi/clk-sunxi.c
+++ b/drivers/clk/sunxi/clk-sunxi.c
@@ -23,6 +23,7 @@
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/log2.h>
+#include <linux/delay.h>
 
 #include "clk-factors.h"
 
@@ -200,6 +201,56 @@ static void sun8i_a23_get_pll1_factors(struct factors_request *req)
 }
 
 /**
+ * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
+ * register using an algorithm that tries to reserve the PLL lock
+ */
+
+static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
+{
+	const struct clk_factors_config *config = factors->config;
+	u32 reg;
+
+	/* Fetch the register value */
+	reg = readl(factors->reg);
+
+	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
+		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
+
+		writel(reg, factors->reg);
+		__delay(2000);
+	}
+
+	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
+		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
+
+		writel(reg, factors->reg);
+		__delay(2000);
+	}
+
+	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
+	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
+
+	writel(reg, factors->reg);
+	__delay(20);
+
+	while (!(readl(factors->reg) & (1 << config->lock)));
+
+	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
+		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
+
+		writel(reg, factors->reg);
+		__delay(2000);
+	}
+
+	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
+		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
+
+		writel(reg, factors->reg);
+		__delay(2000);
+	}
+}
+
+/**
  * sun4i_get_pll5_factors() - calculates n, k factors for PLL5
  * PLL5 rate is calculated as follows
  * rate = parent_rate * n * (k + 1)
@@ -451,6 +502,7 @@ static const struct clk_factors_config sun8i_a23_pll1_config = {
 	.pshift = 16,
 	.pwidth = 2,
 	.n_start = 1,
+	.lock = 28
 };
 
 static const struct clk_factors_config sun4i_pll5_config = {
@@ -513,6 +565,13 @@ static const struct factors_data sun8i_a23_pll1_data __initconst = {
 	.getter = sun8i_a23_get_pll1_factors,
 };
 
+static const struct factors_data sun8i_h3_pll1_data __initconst = {
+	.enable = 31,
+	.table = &sun8i_a23_pll1_config,
+	.getter = sun8i_a23_get_pll1_factors,
+	.apply = sun8i_h3_apply_pll1_factors,
+};
+
 static const struct factors_data sun7i_a20_pll4_data __initconst = {
 	.enable = 31,
 	.table = &sun4i_pll5_config,
@@ -590,12 +649,19 @@ static void __init sun6i_pll1_clk_setup(struct device_node *node)
 CLK_OF_DECLARE(sun6i_pll1, "allwinner,sun6i-a31-pll1-clk",
 	       sun6i_pll1_clk_setup);
 
-static void __init sun8i_pll1_clk_setup(struct device_node *node)
+static void __init sun8i_a23_pll1_clk_setup(struct device_node *node)
 {
 	sunxi_factors_clk_setup(node, &sun8i_a23_pll1_data);
 }
-CLK_OF_DECLARE(sun8i_pll1, "allwinner,sun8i-a23-pll1-clk",
-	       sun8i_pll1_clk_setup);
+CLK_OF_DECLARE(sun8i_a23_pll1, "allwinner,sun8i-a23-pll1-clk",
+	       sun8i_a23_pll1_clk_setup);
+
+static void __init sun8i_h3_pll1_clk_setup(struct device_node *node)
+{
+	sunxi_factors_clk_setup(node, &sun8i_h3_pll1_data);
+}
+CLK_OF_DECLARE(sun8i_h3_pll1, "allwinner,sun8i-h3-pll1-clk",
+	       sun8i_h3_pll1_clk_setup);
 
 static void __init sun7i_pll4_clk_setup(struct device_node *node)
 {
-- 
2.9.0

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 2985 bytes --]

Hi,

On Sat, Jun 25, 2016 at 05:45:03AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> 
> PLL1 on H3 requires special factors application algorithm,
> when the rate is changed. This algorithm was extracted
> from the arisc code that handles frequency scaling
> in the BSP kernel.
> 
> This commit adds optional apply function to
> struct factors_data, that can implement non-trivial
> factors application method, when necessary.
> 
> Also struct clk_factors_config is extended with position
> of the PLL lock flag.

Have you tested the current implementation, and found that it was not
working, or did you duplicate the arisc code directly?

>  /**
> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
> + * register using an algorithm that tries to reserve the PLL lock
> + */
> +
> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
> +{
> +	const struct clk_factors_config *config = factors->config;
> +	u32 reg;
> +
> +	/* Fetch the register value */
> +	reg = readl(factors->reg);
> +
> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
> +
> +		writel(reg, factors->reg);
> +		__delay(2000);
> +	}

So there was some doubts about the fact that P was being used, or at
least that it was useful.

> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
> +
> +		writel(reg, factors->reg);
> +		__delay(2000);
> +	}
> +
> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
> +
> +	writel(reg, factors->reg);
> +	__delay(20);
> +
> +	while (!(readl(factors->reg) & (1 << config->lock)));

So, they are applying the dividers first, and then applying the
multipliers, and then wait for the PLL to stabilize.

> +
> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
> +
> +		writel(reg, factors->reg);
> +		__delay(2000);
> +	}
> +
> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
> +
> +		writel(reg, factors->reg);
> +		__delay(2000);
> +	}

However, this is kind of weird, why would you need to re-apply the
dividers? Nothing really changes. Have you tried without that part?

Since this is really specific, I guess you could simply make the
clk_ops for the nkmp clocks public, and just re-implement set_rate
using that logic.

You might also need to set an upper limit on P, since the last value
(4) is not a valid one.

I guess you could do that by adding a max field in the __ccu_div
structure.

Maxime

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

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 5649 bytes --]

Hi,

On 30.6.2016 22:40, Maxime Ripard wrote:
> Hi,
> 
> On Sat, Jun 25, 2016 at 05:45:03AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>
>> PLL1 on H3 requires special factors application algorithm,
>> when the rate is changed. This algorithm was extracted
>> from the arisc code that handles frequency scaling
>> in the BSP kernel.
>>
>> This commit adds optional apply function to
>> struct factors_data, that can implement non-trivial
>> factors application method, when necessary.
>>
>> Also struct clk_factors_config is extended with position
>> of the PLL lock flag.
> 
> Have you tested the current implementation, and found that it was not
> working, or did you duplicate the arisc code directly?

I have tested the current implementation, and it was not working. It
depended on some other factors, like the initial setup done by u-boot.
It didn't work reliably.

Then I reverse engineered arisc, in an effort to see what's the
difference, between mainline and BSP code.

>>  /**
>> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
>> + * register using an algorithm that tries to reserve the PLL lock
>> + */
>> +
>> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
>> +{
>> +	const struct clk_factors_config *config = factors->config;
>> +	u32 reg;
>> +
>> +	/* Fetch the register value */
>> +	reg = readl(factors->reg);
>> +
>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
> 
> So there was some doubts about the fact that P was being used, or at
> least that it was useful.

p is necessary to reduce frequencies below 288 MHz according to the
datasheet.

>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
>> +
>> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
>> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
>> +
>> +	writel(reg, factors->reg);
>> +	__delay(20);
>> +
>> +	while (!(readl(factors->reg) & (1 << config->lock)));
> 
> So, they are applying the dividers first, and then applying the
> multipliers, and then wait for the PLL to stabilize.

Not exactly, first we are increasing dividers if the new dividers are
higher that that what's already set. This ensures that because
application of dividers is immediate by the design of the PLL, the
application of multipliers isn't. So the VCO would still run at the same
frequency for a while gradually rising to a new value for example,
while the dividers would be reduced immediately. Leading to crash.

PLL
--------------------------
PRE DIV(f0) -> VCO(f1) -> POST DIV(f2)
   P             K,N           M

Example: (we set all factors at once, reducing dividers and multipliers
at the same time at 0ms - this should lead to no change in the output
frequency, but...)

-1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
 0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 2GHz       - boom
 1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 1.5GHz
 2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz

The current code crashes exactly at boom, you don't get any more
instructions to execute.

See.

So this patch first increases dividers (only if necessary), changes
multipliers and waits for change to happen (takes around 2000 cycles),
and then decreases dividers (only if necessary).

So we get:

-1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
 0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz   - no boom, multiplier
                                             reduced
 1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 0.75GHz
1.9ms: f0 = 24MHz, f1 = 1GHz,   f2 = 0.5GHz - we got PLL sync
 2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz   - and here we reduce divider
at last

>> +
>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
>> +
>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
> 
> However, this is kind of weird, why would you need to re-apply the
> dividers? Nothing really changes. Have you tried without that part?

See above, we either increase before PLL change, or reduce dividers
after the change. Nothing is re-applied.

> Since this is really specific, I guess you could simply make the
> clk_ops for the nkmp clocks public, and just re-implement set_rate
> using that logic.

I would argue that this may be necessary for other PLL clocks too, if
you can get out of bounds output frequency, by changing the dividers too
early or too late. So perhaps this code should be generalized for other
PLL clocks too, instead.

> 
> You might also need to set an upper limit on P, since the last value
> (4) is not a valid one.

I think, that should be done by the factors calculation function already.

> I guess you could do that by adding a max field in the __ccu_div
> structure.
> 
> Maxime
> 

regards,
  Ondrej

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Jean-Francois Moine]

On Fri, 1 Jul 2016 02:50:57 +0200
Ondřej Jirman <megous@megous.com> wrote:

> > Since this is really specific, I guess you could simply make the
> > clk_ops for the nkmp clocks public, and just re-implement set_rate
> > using that logic.
> 
> I would argue that this may be necessary for other PLL clocks too, if
> you can get out of bounds output frequency, by changing the dividers too
> early or too late. So perhaps this code should be generalized for other
> PLL clocks too, instead.

The documentation says that only the CPU and DDR PLLs can be dynamically
changed after boot.

-- 
Ken ar c'hentañ	|	      ** Breizh ha Linux atav! **
Jef		|		http://moinejf.free.fr/

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 1216 bytes --]

On 1.7.2016 07:37, Jean-Francois Moine wrote:
> On Fri, 1 Jul 2016 02:50:57 +0200
> Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
> 
>>> Since this is really specific, I guess you could simply make the
>>> clk_ops for the nkmp clocks public, and just re-implement set_rate
>>> using that logic.
>>
>> I would argue that this may be necessary for other PLL clocks too, if
>> you can get out of bounds output frequency, by changing the dividers too
>> early or too late. So perhaps this code should be generalized for other
>> PLL clocks too, instead.
> 
> The documentation says that only the CPU and DDR PLLs can be dynamically
> changed after boot.

The question is what exactly is meant by after boot. :) Anyway, if the
kernel has no business changing some other PLLs, if there's code for
changing them, should it be dropped?

regards,
  Ondrej

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Jean-Francois Moine]

On Fri, 1 Jul 2016 08:34:21 +0200
Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:

> > The documentation says that only the CPU and DDR PLLs can be dynamically
> > changed after boot.
> 
> The question is what exactly is meant by after boot. :) Anyway, if the
> kernel has no business changing some other PLLs, if there's code for
> changing them, should it be dropped?

No, because all the other PLLs may not be initialized by the U-boot
(audio, video, gpu...), and also, their rate may be changed safely by
stopping them (gate).

-- 
Ken ar c'hentañ	|	      ** Breizh ha Linux atav! **
Jef		|		http://moinejf.free.fr/
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 5795 bytes --]

On Fri, Jul 01, 2016 at 02:50:57AM +0200, Ondřej Jirman wrote:
> >>  /**
> >> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
> >> + * register using an algorithm that tries to reserve the PLL lock
> >> + */
> >> +
> >> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
> >> +{
> >> +	const struct clk_factors_config *config = factors->config;
> >> +	u32 reg;
> >> +
> >> +	/* Fetch the register value */
> >> +	reg = readl(factors->reg);
> >> +
> >> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
> >> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
> >> +
> >> +		writel(reg, factors->reg);
> >> +		__delay(2000);
> >> +	}
> > 
> > So there was some doubts about the fact that P was being used, or at
> > least that it was useful.
> 
> p is necessary to reduce frequencies below 288 MHz according to the
> datasheet.

Yes, but you could reach those frequencies without P, too, and it's
not part of any OPP provided by Allwinner.

> >> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
> >> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
> >> +
> >> +		writel(reg, factors->reg);
> >> +		__delay(2000);
> >> +	}
> >> +
> >> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
> >> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
> >> +
> >> +	writel(reg, factors->reg);
> >> +	__delay(20);
> >> +
> >> +	while (!(readl(factors->reg) & (1 << config->lock)));
> > 
> > So, they are applying the dividers first, and then applying the
> > multipliers, and then wait for the PLL to stabilize.
> 
> Not exactly, first we are increasing dividers if the new dividers are
> higher that that what's already set. This ensures that because
> application of dividers is immediate by the design of the PLL, the
> application of multipliers isn't. So the VCO would still run at the same
> frequency for a while gradually rising to a new value for example,
> while the dividers would be reduced immediately. Leading to crash.
> 
> PLL
> --------------------------
> PRE DIV(f0) -> VCO(f1) -> POST DIV(f2)
>    P             K,N           M
> 
> Example: (we set all factors at once, reducing dividers and multipliers
> at the same time at 0ms - this should lead to no change in the output
> frequency, but...)
> 
> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 2GHz       - boom
>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 1.5GHz
>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz
> 
> The current code crashes exactly at boom, you don't get any more
> instructions to execute.
> 
> See.
> 
> So this patch first increases dividers (only if necessary), changes
> multipliers and waits for change to happen (takes around 2000 cycles),
> and then decreases dividers (only if necessary).
> 
> So we get:
> 
> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz   - no boom, multiplier
>                                              reduced
>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 0.75GHz
> 1.9ms: f0 = 24MHz, f1 = 1GHz,   f2 = 0.5GHz - we got PLL sync
>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz   - and here we reduce divider
> at last

Awesome explanation, thanks!

So I guess it really all boils down to the fact that the CPU is
clocked way outside of it's operating frequency while the PLL
stabilizes, right?

If so, then yes, trying to switch to the 24MHz oscillator before
applying the factors, and then switching back when the PLL is stable
would be a nice solution.

I just checked, and all the SoCs we've had so far have that
possibility, so if it works, for now, I'd like to stick to that.

> >> +
> >> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
> >> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
> >> +
> >> +		writel(reg, factors->reg);
> >> +		__delay(2000);
> >> +	}
> >> +
> >> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
> >> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
> >> +
> >> +		writel(reg, factors->reg);
> >> +		__delay(2000);
> >> +	}
> > 
> > However, this is kind of weird, why would you need to re-apply the
> > dividers? Nothing really changes. Have you tried without that part?
> 
> See above, we either increase before PLL change, or reduce dividers
> after the change. Nothing is re-applied.
> 
> > Since this is really specific, I guess you could simply make the
> > clk_ops for the nkmp clocks public, and just re-implement set_rate
> > using that logic.
> 
> I would argue that this may be necessary for other PLL clocks too, if
> you can get out of bounds output frequency, by changing the dividers too
> early or too late. So perhaps this code should be generalized for other
> PLL clocks too, instead.

We can scale down the problem a bit. The only PLL we modify are the
CPU, audio and video ones.

The CPU should definitely be addressed, but what would be the outcome
of an out of bounds audio pll for example? Is it just taking more time
to stabilize, or would it hurt the system?

In the former case, then we can just wait. In the latter, we of course
need to come up with a solution.

Maxime


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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 9456 bytes --]

On 15.7.2016 10:53, Maxime Ripard wrote:
> On Fri, Jul 01, 2016 at 02:50:57AM +0200, Ondřej Jirman wrote:
>>>>  /**
>>>> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
>>>> + * register using an algorithm that tries to reserve the PLL lock
>>>> + */
>>>> +
>>>> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
>>>> +{
>>>> +	const struct clk_factors_config *config = factors->config;
>>>> +	u32 reg;
>>>> +
>>>> +	/* Fetch the register value */
>>>> +	reg = readl(factors->reg);
>>>> +
>>>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
>>>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>>>> +
>>>> +		writel(reg, factors->reg);
>>>> +		__delay(2000);
>>>> +	}
>>>
>>> So there was some doubts about the fact that P was being used, or at
>>> least that it was useful.
>>
>> p is necessary to reduce frequencies below 288 MHz according to the
>> datasheet.
> 
> Yes, but you could reach those frequencies without P, too, and it's
> not part of any OPP provided by Allwinner.

The arisc firmware for H3 contains table of factors for frequences from
0 to 2GHz and, P is used below 240MHz. M is never used, BTW. (other
datasheets specify M as for testing use only, whatever that means - not
H3, but it seems it's the same PLL block)


... snip ...
	{ .n = 9, .k = 0, .m = 0, .p = 2 }, // 60 => 60 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 2 }, // 66 => 66 MHz
	{ .n = 11, .k = 0, .m = 0, .p = 2 }, // 72 => 72 MHz
	{ .n = 12, .k = 0, .m = 0, .p = 2 }, // 78 => 78 MHz
	{ .n = 13, .k = 0, .m = 0, .p = 2 }, // 84 => 84 MHz
	{ .n = 14, .k = 0, .m = 0, .p = 2 }, // 90 => 90 MHz
	{ .n = 15, .k = 0, .m = 0, .p = 2 }, // 96 => 96 MHz
	{ .n = 16, .k = 0, .m = 0, .p = 2 }, // 102 => 102 MHz
	{ .n = 17, .k = 0, .m = 0, .p = 2 }, // 108 => 108 MHz
	{ .n = 18, .k = 0, .m = 0, .p = 2 }, // 114 => 114 MHz
	{ .n = 9, .k = 0, .m = 0, .p = 1 }, // 120 => 120 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 1 }, // 126 => 132 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 1 }, // 132 => 132 MHz
	{ .n = 11, .k = 0, .m = 0, .p = 1 }, // 138 => 144 MHz
	{ .n = 11, .k = 0, .m = 0, .p = 1 }, // 144 => 144 MHz
	{ .n = 12, .k = 0, .m = 0, .p = 1 }, // 150 => 156 MHz
	{ .n = 12, .k = 0, .m = 0, .p = 1 }, // 156 => 156 MHz
	{ .n = 13, .k = 0, .m = 0, .p = 1 }, // 162 => 168 MHz
	{ .n = 13, .k = 0, .m = 0, .p = 1 }, // 168 => 168 MHz
	{ .n = 14, .k = 0, .m = 0, .p = 1 }, // 174 => 180 MHz
	{ .n = 14, .k = 0, .m = 0, .p = 1 }, // 180 => 180 MHz
	{ .n = 15, .k = 0, .m = 0, .p = 1 }, // 186 => 192 MHz
	{ .n = 15, .k = 0, .m = 0, .p = 1 }, // 192 => 192 MHz
	{ .n = 16, .k = 0, .m = 0, .p = 1 }, // 198 => 204 MHz
	{ .n = 16, .k = 0, .m = 0, .p = 1 }, // 204 => 204 MHz
	{ .n = 17, .k = 0, .m = 0, .p = 1 }, // 210 => 216 MHz
	{ .n = 17, .k = 0, .m = 0, .p = 1 }, // 216 => 216 MHz
	{ .n = 18, .k = 0, .m = 0, .p = 1 }, // 222 => 228 MHz
	{ .n = 18, .k = 0, .m = 0, .p = 1 }, // 228 => 228 MHz
	{ .n = 9, .k = 0, .m = 0, .p = 0 }, // 234 => 240 MHz
	{ .n = 9, .k = 0, .m = 0, .p = 0 }, // 240 => 240 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 0 }, // 246 => 264 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 0 }, // 252 => 264 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 0 }, // 258 => 264 MHz
	{ .n = 10, .k = 0, .m = 0, .p = 0 }, // 264 => 264 MHz
	{ .n = 11, .k = 0, .m = 0, .p = 0 }, // 270 => 288 MHz
	{ .n = 11, .k = 0, .m = 0, .p = 0 }, // 276 => 288 MHz
... snip ...


>>>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
>>>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>>>> +
>>>> +		writel(reg, factors->reg);
>>>> +		__delay(2000);
>>>> +	}
>>>> +
>>>> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
>>>> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
>>>> +
>>>> +	writel(reg, factors->reg);
>>>> +	__delay(20);
>>>> +
>>>> +	while (!(readl(factors->reg) & (1 << config->lock)));
>>>
>>> So, they are applying the dividers first, and then applying the
>>> multipliers, and then wait for the PLL to stabilize.
>>
>> Not exactly, first we are increasing dividers if the new dividers are
>> higher that that what's already set. This ensures that because
>> application of dividers is immediate by the design of the PLL, the
>> application of multipliers isn't. So the VCO would still run at the same
>> frequency for a while gradually rising to a new value for example,
>> while the dividers would be reduced immediately. Leading to crash.
>>
>> PLL
>> --------------------------
>> PRE DIV(f0) -> VCO(f1) -> POST DIV(f2)
>>    P             K,N           M
>>
>> Example: (we set all factors at once, reducing dividers and multipliers
>> at the same time at 0ms - this should lead to no change in the output
>> frequency, but...)
>>
>> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 2GHz       - boom
>>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 1.5GHz
>>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz
>>
>> The current code crashes exactly at boom, you don't get any more
>> instructions to execute.
>>
>> See.
>>
>> So this patch first increases dividers (only if necessary), changes
>> multipliers and waits for change to happen (takes around 2000 cycles),
>> and then decreases dividers (only if necessary).
>>
>> So we get:
>>
>> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz   - no boom, multiplier
>>                                              reduced
>>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 0.75GHz
>> 1.9ms: f0 = 24MHz, f1 = 1GHz,   f2 = 0.5GHz - we got PLL sync
>>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz   - and here we reduce divider
>> at last
> 
> Awesome explanation, thanks!
> 
> So I guess it really all boils down to the fact that the CPU is
> clocked way outside of it's operating frequency while the PLL
> stabilizes, right?

It may be, depending on the factors before and after change. I haven't
tested what factors the mainline kernel calculates for each frequency.
The arisc never uses M, and only uses P at frequencies that would not
allow for this behavior.

I created a test program for arisc that runs a loop on the main CPU
using msgbox to send pings to the arisc CPU, and the vary the PLL1
randomly from the arisc, and haven't been able to lockup the main CPU
yet with either method.

There's also AXI clock, that depends on PLL1. Arisc firmware uses the
same method to change it while changing CPUX clock. If the clock would
rise above certain frequency, AXI divider is increased before the PLL1
change. If it would fall below certain frequency it is decreased after
the PLL1 change.

Though, aw kernel has axi divider set to 3 for all PLL1 frequencies, so
this has no effect in practice.

It's all smoke and mirrors. :D

> If so, then yes, trying to switch to the 24MHz oscillator before
> applying the factors, and then switching back when the PLL is stable
> would be a nice solution.
> 
> I just checked, and all the SoCs we've had so far have that
> possibility, so if it works, for now, I'd like to stick to that.

It would need to be tested. U-boot does the change only once, while the
kernel would be doing it all the time and between various frequencies
and PLL settings. So the issues may show up with this solution too.

>>>> +
>>>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
>>>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>>>> +
>>>> +		writel(reg, factors->reg);
>>>> +		__delay(2000);
>>>> +	}
>>>> +
>>>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
>>>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>>>> +
>>>> +		writel(reg, factors->reg);
>>>> +		__delay(2000);
>>>> +	}
>>>
>>> However, this is kind of weird, why would you need to re-apply the
>>> dividers? Nothing really changes. Have you tried without that part?
>>
>> See above, we either increase before PLL change, or reduce dividers
>> after the change. Nothing is re-applied.
>>
>>> Since this is really specific, I guess you could simply make the
>>> clk_ops for the nkmp clocks public, and just re-implement set_rate
>>> using that logic.
>>
>> I would argue that this may be necessary for other PLL clocks too, if
>> you can get out of bounds output frequency, by changing the dividers too
>> early or too late. So perhaps this code should be generalized for other
>> PLL clocks too, instead.
> 
> We can scale down the problem a bit. The only PLL we modify are the
> CPU, audio and video ones.
> 
> The CPU should definitely be addressed, but what would be the outcome
> of an out of bounds audio pll for example? Is it just taking more time
> to stabilize, or would it hurt the system?

I have no idea. Given the low frequencies, you'll probably just get some
transient audio noise.

> In the former case, then we can just wait. In the latter, we of course
> need to come up with a solution.
> 
> Maxime
> 
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Jean-Francois Moine]

On Fri, 15 Jul 2016 12:38:54 +0200
Ondřej Jirman <megous@megous.com> wrote:

> > If so, then yes, trying to switch to the 24MHz oscillator before
> > applying the factors, and then switching back when the PLL is stable
> > would be a nice solution.
> > 
> > I just checked, and all the SoCs we've had so far have that
> > possibility, so if it works, for now, I'd like to stick to that.
> 
> It would need to be tested. U-boot does the change only once, while the
> kernel would be doing it all the time and between various frequencies
> and PLL settings. So the issues may show up with this solution too.

I don't think this is a good idea: the CPU clock may be changed at any
time with the CPUFreq governor. I don't see the system moving from
1008MHz to 24MHz and then to 1200MHz when some computation is needed!

BTW, Ondřej, in my BPi M2+, I tried to change the CPU clock with your
code at kernel start time from 792MHz to 1008MHz, but the hardware
(arisc?) set an other value, and the system speed was lower than before
(the PLL-CPUx register is 0x90031521 on boot, I want to set it to
xxxx1410 and I read 0x91031f33 - sorry, I did not have a look at the
CPU SD pattern). Do you know why?

-- 
Ken ar c'hentañ	|	      ** Breizh ha Linux atav! **
Jef		|		http://moinejf.free.fr/

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 2005 bytes --]



On 15.7.2016 15:27, Jean-Francois Moine wrote:
> On Fri, 15 Jul 2016 12:38:54 +0200
> Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
> 
>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>> applying the factors, and then switching back when the PLL is stable
>>> would be a nice solution.
>>>
>>> I just checked, and all the SoCs we've had so far have that
>>> possibility, so if it works, for now, I'd like to stick to that.
>>
>> It would need to be tested. U-boot does the change only once, while the
>> kernel would be doing it all the time and between various frequencies
>> and PLL settings. So the issues may show up with this solution too.
> 
> I don't think this is a good idea: the CPU clock may be changed at any
> time with the CPUFreq governor. I don't see the system moving from
> 1008MHz to 24MHz and then to 1200MHz when some computation is needed!

PLL lock time is around 10-20us, I'd guess based on the number of loops
in the PLL lock wait loop. So unless you'll be switching frequencies
many times per second, this should be barely noticeable.

But I'd like a different solution too.

> BTW, Ondřej, in my BPi M2+, I tried to change the CPU clock with your
> code at kernel start time from 792MHz to 1008MHz, but the hardware
> (arisc?) set an other value, and the system speed was lower than before
> (the PLL-CPUx register is 0x90031521 on boot, I want to set it to
> xxxx1410 and I read 0x91031f33 - sorry, I did not have a look at the
> CPU SD pattern). Do you know why?

No idea. Arisc shouldn't do anything, unless you load some firmware into
it, and release its reset line.

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [linux-sunxi] Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Michal Suchanek]

Hello,

On 15 July 2016 at 15:48, Ondřej Jirman <megous@megous.com> wrote:
>
>
> On 15.7.2016 15:27, Jean-Francois Moine wrote:
>> On Fri, 15 Jul 2016 12:38:54 +0200
>> Ondřej Jirman <megous@megous.com> wrote:
>>
>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>> applying the factors, and then switching back when the PLL is stable
>>>> would be a nice solution.
>>>>
>>>> I just checked, and all the SoCs we've had so far have that
>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>
>>> It would need to be tested. U-boot does the change only once, while the
>>> kernel would be doing it all the time and between various frequencies
>>> and PLL settings. So the issues may show up with this solution too.
>>
>> I don't think this is a good idea: the CPU clock may be changed at any
>> time with the CPUFreq governor. I don't see the system moving from
>> 1008MHz to 24MHz and then to 1200MHz when some computation is needed!
>
> PLL lock time is around 10-20us, I'd guess based on the number of loops
> in the PLL lock wait loop. So unless you'll be switching frequencies
> many times per second, this should be barely noticeable.
>
> But I'd like a different solution too.

Do you have a patch to test this?

For me changing CPU frequency on Orange Pi One always locks up the
system. I keep running it on the u-boot setup 1.08GHz at 1.1V

Thanks

Michal

Re: [linux-sunxi] Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 1772 bytes --]

On 15.7.2016 16:22, Michal Suchanek wrote:
> Hello,
> 
> On 15 July 2016 at 15:48, Ondřej Jirman <megous@megous.com> wrote:
>>
>>
>> On 15.7.2016 15:27, Jean-Francois Moine wrote:
>>> On Fri, 15 Jul 2016 12:38:54 +0200
>>> Ondřej Jirman <megous@megous.com> wrote:
>>>
>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>>> applying the factors, and then switching back when the PLL is stable
>>>>> would be a nice solution.
>>>>>
>>>>> I just checked, and all the SoCs we've had so far have that
>>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>>
>>>> It would need to be tested. U-boot does the change only once, while the
>>>> kernel would be doing it all the time and between various frequencies
>>>> and PLL settings. So the issues may show up with this solution too.
>>>
>>> I don't think this is a good idea: the CPU clock may be changed at any
>>> time with the CPUFreq governor. I don't see the system moving from
>>> 1008MHz to 24MHz and then to 1200MHz when some computation is needed!
>>
>> PLL lock time is around 10-20us, I'd guess based on the number of loops
>> in the PLL lock wait loop. So unless you'll be switching frequencies
>> many times per second, this should be barely noticeable.
>>
>> But I'd like a different solution too.
> 
> Do you have a patch to test this?
> 
> For me changing CPU frequency on Orange Pi One always locks up the
> system. I keep running it on the u-boot setup 1.08GHz at 1.1V

Not anymore. But it's quite simple to hack it into the
drivers/clk/sunxi/clk-factors.c:clk_factors_set_rate()

You can look at u-boot arch/arm/mach-sunxi/clock_sun6i.c:clock_set_pll1
for how to change the CPU clock source.

> Thanks
> 
> Michal
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1643 bytes --]

On Fri, Jul 15, 2016 at 03:27:56PM +0200, Jean-Francois Moine wrote:
> On Fri, 15 Jul 2016 12:38:54 +0200
> Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
> 
> > > If so, then yes, trying to switch to the 24MHz oscillator before
> > > applying the factors, and then switching back when the PLL is stable
> > > would be a nice solution.
> > > 
> > > I just checked, and all the SoCs we've had so far have that
> > > possibility, so if it works, for now, I'd like to stick to that.
> > 
> > It would need to be tested. U-boot does the change only once, while the
> > kernel would be doing it all the time and between various frequencies
> > and PLL settings. So the issues may show up with this solution too.
> 
> I don't think this is a good idea: the CPU clock may be changed at any
> time with the CPUFreq governor. I don't see the system moving from
> 1008MHz to 24MHz and then to 1200MHz when some computation is needed!

It wouldn't happen that often. The sampling rate for the governor is
1000 times the latency, so, at most, 0.1% of the time would be spent
at 24MHz.

And if you're really concerned about performances, you won't enable
cpufreq anyway.

Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 6394 bytes --]

On Fri, Jul 15, 2016 at 12:38:54PM +0200, Ondřej Jirman wrote:
> On 15.7.2016 10:53, Maxime Ripard wrote:
> > On Fri, Jul 01, 2016 at 02:50:57AM +0200, Ondřej Jirman wrote:
> >>>>  /**
> >>>> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
> >>>> + * register using an algorithm that tries to reserve the PLL lock
> >>>> + */
> >>>> +
> >>>> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
> >>>> +{
> >>>> +	const struct clk_factors_config *config = factors->config;
> >>>> +	u32 reg;
> >>>> +
> >>>> +	/* Fetch the register value */
> >>>> +	reg = readl(factors->reg);
> >>>> +
> >>>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
> >>>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
> >>>> +
> >>>> +		writel(reg, factors->reg);
> >>>> +		__delay(2000);
> >>>> +	}
> >>>
> >>> So there was some doubts about the fact that P was being used, or at
> >>> least that it was useful.
> >>
> >> p is necessary to reduce frequencies below 288 MHz according to the
> >> datasheet.
> > 
> > Yes, but you could reach those frequencies without P, too, and it's
> > not part of any OPP provided by Allwinner.
> 
> The arisc firmware for H3 contains table of factors for frequences from
> 0 to 2GHz and, P is used below 240MHz. M is never used, BTW. (other
> datasheets specify M as for testing use only, whatever that means - not
> H3, but it seems it's the same PLL block)

Interesting. Which SoCs in particular?

> >>>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
> >>>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
> >>>> +
> >>>> +		writel(reg, factors->reg);
> >>>> +		__delay(2000);
> >>>> +	}
> >>>> +
> >>>> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
> >>>> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
> >>>> +
> >>>> +	writel(reg, factors->reg);
> >>>> +	__delay(20);
> >>>> +
> >>>> +	while (!(readl(factors->reg) & (1 << config->lock)));
> >>>
> >>> So, they are applying the dividers first, and then applying the
> >>> multipliers, and then wait for the PLL to stabilize.
> >>
> >> Not exactly, first we are increasing dividers if the new dividers are
> >> higher that that what's already set. This ensures that because
> >> application of dividers is immediate by the design of the PLL, the
> >> application of multipliers isn't. So the VCO would still run at the same
> >> frequency for a while gradually rising to a new value for example,
> >> while the dividers would be reduced immediately. Leading to crash.
> >>
> >> PLL
> >> --------------------------
> >> PRE DIV(f0) -> VCO(f1) -> POST DIV(f2)
> >>    P             K,N           M
> >>
> >> Example: (we set all factors at once, reducing dividers and multipliers
> >> at the same time at 0ms - this should lead to no change in the output
> >> frequency, but...)
> >>
> >> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
> >>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 2GHz       - boom
> >>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 1.5GHz
> >>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz
> >>
> >> The current code crashes exactly at boom, you don't get any more
> >> instructions to execute.
> >>
> >> See.
> >>
> >> So this patch first increases dividers (only if necessary), changes
> >> multipliers and waits for change to happen (takes around 2000 cycles),
> >> and then decreases dividers (only if necessary).
> >>
> >> So we get:
> >>
> >> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
> >>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz   - no boom, multiplier
> >>                                              reduced
> >>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 0.75GHz
> >> 1.9ms: f0 = 24MHz, f1 = 1GHz,   f2 = 0.5GHz - we got PLL sync
> >>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz   - and here we reduce divider
> >> at last
> > 
> > Awesome explanation, thanks!
> > 
> > So I guess it really all boils down to the fact that the CPU is
> > clocked way outside of it's operating frequency while the PLL
> > stabilizes, right?
> 
> It may be, depending on the factors before and after change. I haven't
> tested what factors the mainline kernel calculates for each frequency.
> The arisc never uses M, and only uses P at frequencies that would not
> allow for this behavior.
> 
> I created a test program for arisc that runs a loop on the main CPU
> using msgbox to send pings to the arisc CPU, and the vary the PLL1
> randomly from the arisc, and haven't been able to lockup the main CPU
> yet with either method.
> 
> There's also AXI clock, that depends on PLL1. Arisc firmware uses the
> same method to change it while changing CPUX clock. If the clock would
> rise above certain frequency, AXI divider is increased before the PLL1
> change. If it would fall below certain frequency it is decreased after
> the PLL1 change.

If we ever need to change that, we can always rely on a clock notifier
to adjust the divider before the change happen (and support all the
scenarios, like the clock change has been aborted).

> > If so, then yes, trying to switch to the 24MHz oscillator before
> > applying the factors, and then switching back when the PLL is stable
> > would be a nice solution.
> > 
> > I just checked, and all the SoCs we've had so far have that
> > possibility, so if it works, for now, I'd like to stick to that.
> 
> It would need to be tested. U-boot does the change only once, while the
> kernel would be doing it all the time and between various frequencies
> and PLL settings. So the issues may show up with this solution too.

That would have the benefit of being quite easy to document, not be a
huge amount of code and it would work on all the CPUs PLLs we have so
far, so still, a pretty big win. If it doesn't, of course, we don't
really have the choice.

Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 6310 bytes --]



On 21.7.2016 11:48, Maxime Ripard wrote:
> On Fri, Jul 15, 2016 at 12:38:54PM +0200, Ondřej Jirman wrote:
>> On 15.7.2016 10:53, Maxime Ripard wrote:
>>> On Fri, Jul 01, 2016 at 02:50:57AM +0200, Ondřej Jirman wrote:
>>>>>>  /**
>>>>>> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
>>>>>> + * register using an algorithm that tries to reserve the PLL lock
>>>>>> + */
>>>>>> +
>>>>>> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
>>>>>> +{
>>>>>> +	const struct clk_factors_config *config = factors->config;
>>>>>> +	u32 reg;
>>>>>> +
>>>>>> +	/* Fetch the register value */
>>>>>> +	reg = readl(factors->reg);
>>>>>> +
>>>>>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
>>>>>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>>>>>> +
>>>>>> +		writel(reg, factors->reg);
>>>>>> +		__delay(2000);
>>>>>> +	}
>>>>>
>>>>> So there was some doubts about the fact that P was being used, or at
>>>>> least that it was useful.
>>>>
>>>> p is necessary to reduce frequencies below 288 MHz according to the
>>>> datasheet.
>>>
>>> Yes, but you could reach those frequencies without P, too, and it's
>>> not part of any OPP provided by Allwinner.
>>
>> The arisc firmware for H3 contains table of factors for frequences from
>> 0 to 2GHz and, P is used below 240MHz. M is never used, BTW. (other
>> datasheets specify M as for testing use only, whatever that means - not
>> H3, but it seems it's the same PLL block)
> 
> Interesting. Which SoCs in particular?
> 
>>>>>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
>>>>>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>>>>>> +
>>>>>> +		writel(reg, factors->reg);
>>>>>> +		__delay(2000);
>>>>>> +	}
>>>>>> +
>>>>>> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
>>>>>> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
>>>>>> +
>>>>>> +	writel(reg, factors->reg);
>>>>>> +	__delay(20);
>>>>>> +
>>>>>> +	while (!(readl(factors->reg) & (1 << config->lock)));
>>>>>
>>>>> So, they are applying the dividers first, and then applying the
>>>>> multipliers, and then wait for the PLL to stabilize.
>>>>
>>>> Not exactly, first we are increasing dividers if the new dividers are
>>>> higher that that what's already set. This ensures that because
>>>> application of dividers is immediate by the design of the PLL, the
>>>> application of multipliers isn't. So the VCO would still run at the same
>>>> frequency for a while gradually rising to a new value for example,
>>>> while the dividers would be reduced immediately. Leading to crash.
>>>>
>>>> PLL
>>>> --------------------------
>>>> PRE DIV(f0) -> VCO(f1) -> POST DIV(f2)
>>>>    P             K,N           M
>>>>
>>>> Example: (we set all factors at once, reducing dividers and multipliers
>>>> at the same time at 0ms - this should lead to no change in the output
>>>> frequency, but...)
>>>>
>>>> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>>>>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 2GHz       - boom
>>>>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 1.5GHz
>>>>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz
>>>>
>>>> The current code crashes exactly at boom, you don't get any more
>>>> instructions to execute.
>>>>
>>>> See.
>>>>
>>>> So this patch first increases dividers (only if necessary), changes
>>>> multipliers and waits for change to happen (takes around 2000 cycles),
>>>> and then decreases dividers (only if necessary).
>>>>
>>>> So we get:
>>>>
>>>> -1ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz
>>>>  0ms: f0 = 24MHz, f1 = 2GHz,   f2 = 1GHz   - no boom, multiplier
>>>>                                              reduced
>>>>  1ms: f0 = 24MHz, f1 = 1.5GHz, f2 = 0.75GHz
>>>> 1.9ms: f0 = 24MHz, f1 = 1GHz,   f2 = 0.5GHz - we got PLL sync
>>>>  2ms: f0 = 24MHz, f1 = 1GHz,   f2 = 1GHz   - and here we reduce divider
>>>> at last
>>>
>>> Awesome explanation, thanks!
>>>
>>> So I guess it really all boils down to the fact that the CPU is
>>> clocked way outside of it's operating frequency while the PLL
>>> stabilizes, right?
>>
>> It may be, depending on the factors before and after change. I haven't
>> tested what factors the mainline kernel calculates for each frequency.
>> The arisc never uses M, and only uses P at frequencies that would not
>> allow for this behavior.
>>
>> I created a test program for arisc that runs a loop on the main CPU
>> using msgbox to send pings to the arisc CPU, and the vary the PLL1
>> randomly from the arisc, and haven't been able to lockup the main CPU
>> yet with either method.
>>
>> There's also AXI clock, that depends on PLL1. Arisc firmware uses the
>> same method to change it while changing CPUX clock. If the clock would
>> rise above certain frequency, AXI divider is increased before the PLL1
>> change. If it would fall below certain frequency it is decreased after
>> the PLL1 change.
> 
> If we ever need to change that, we can always rely on a clock notifier
> to adjust the divider before the change happen (and support all the
> scenarios, like the clock change has been aborted).
> 
>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>> applying the factors, and then switching back when the PLL is stable
>>> would be a nice solution.
>>>
>>> I just checked, and all the SoCs we've had so far have that
>>> possibility, so if it works, for now, I'd like to stick to that.
>>
>> It would need to be tested. U-boot does the change only once, while the
>> kernel would be doing it all the time and between various frequencies
>> and PLL settings. So the issues may show up with this solution too.
> 
> That would have the benefit of being quite easy to document, not be a
> huge amount of code and it would work on all the CPUs PLLs we have so
> far, so still, a pretty big win. If it doesn't, of course, we don't
> really have the choice.

It's probably more code though. It has to access different register from
the one that is already defined in dts, which would add a lot of code
and require dts changes. The original patch I sent is simpler than that.

regards,
  Ondrej

> Maxime
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1656 bytes --]

On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
> >>> If so, then yes, trying to switch to the 24MHz oscillator before
> >>> applying the factors, and then switching back when the PLL is stable
> >>> would be a nice solution.
> >>>
> >>> I just checked, and all the SoCs we've had so far have that
> >>> possibility, so if it works, for now, I'd like to stick to that.
> >>
> >> It would need to be tested. U-boot does the change only once, while the
> >> kernel would be doing it all the time and between various frequencies
> >> and PLL settings. So the issues may show up with this solution too.
> > 
> > That would have the benefit of being quite easy to document, not be a
> > huge amount of code and it would work on all the CPUs PLLs we have so
> > far, so still, a pretty big win. If it doesn't, of course, we don't
> > really have the choice.
> 
> It's probably more code though. It has to access different register from
> the one that is already defined in dts, which would add a lot of code
> and require dts changes. The original patch I sent is simpler than that.

Why?

You can use container_of to retrieve the parent structure of the clock
notifier, and then you get a ccu_common structure pointer, with the
CCU base address, the clock register, its lock, etc.

Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.

I don't really get why anything should be changed in the DT, or why it
would add a lot of code. Or maybe we're not talking about the same
thing?

Maxime

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

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 4274 bytes --]

Hi Maxime,

I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
to this.

On 26.7.2016 08:32, Maxime Ripard wrote:
> On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>>> applying the factors, and then switching back when the PLL is stable
>>>>> would be a nice solution.
>>>>>
>>>>> I just checked, and all the SoCs we've had so far have that
>>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>>
>>>> It would need to be tested. U-boot does the change only once, while the
>>>> kernel would be doing it all the time and between various frequencies
>>>> and PLL settings. So the issues may show up with this solution too.
>>>
>>> That would have the benefit of being quite easy to document, not be a
>>> huge amount of code and it would work on all the CPUs PLLs we have so
>>> far, so still, a pretty big win. If it doesn't, of course, we don't
>>> really have the choice.
>>
>> It's probably more code though. It has to access different register from
>> the one that is already defined in dts, which would add a lot of code
>> and require dts changes. The original patch I sent is simpler than that.
> 
> Why?
> 
> You can use container_of to retrieve the parent structure of the clock
> notifier, and then you get a ccu_common structure pointer, with the
> CCU base address, the clock register, its lock, etc.
> 
> Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
> 
> I don't really get why anything should be changed in the DT, or why it
> would add a lot of code. Or maybe we're not talking about the same
> thing?

I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
it very liberally uses divider parameters, even in situations that are
out of spec compared to the current code in the kernel.

In the current code and especially in the original vendor code, divider
parameters are used as last resort only. Presumably because, of the
inherent trouble in changing them, as I described to you in other email.

The new ccu code uses dividers often and even at very high frequencies,
which goes against the spec.

In the vendor code M is never anything else but 0, and P is used only
for frequencies below 288MHz, which matches the H3 datasheet, which says:

"The P factor only use in the condition that PLL output less than 288
MHz."

Also other datasheets of similar socs from Allwinner state that M should
not be used in production code. So it may be that they either forgot to
state it in the H3 datasheet, or it can be used. In any case, they never
use M in their code, so it may be wise to keep to that.

When I boot with the new CCU code I get this:

PLL_CPUX = 0x00001B21 : M = 2, K = 3, N = 28, P = 1, EN = 0
PLL_CPUX : freq = 1008MHz

Mathematically it works, but it is against the spec.

Also, this:

analyzing CPU 0:
  driver: cpufreq-dt
  CPUs which run at the same hardware frequency: 0 1 2 3
  CPUs which need to have their frequency coordinated by software: 0 1 2 3
  maximum transition latency: 1.74 ms
  hardware limits: 120 MHz - 1.37 GHz
  available frequency steps:  120 MHz, 240 MHz, 480 MHz, 648 MHz, 816
MHz, 960 MHz, 1.01 GHz, 1.06 GHz, 1.10 GHz, 1.15 GHz, 1.20 GHz, 1.22
GHz, 1.25 GHz, 1.30 GHz, 1.34 GHz, 1.37 GHz
  available cpufreq governors: conservative ondemand userspace powersave
performance
  current policy: frequency should be within 240 MHz and 240 MHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency: 24.0 MHz (asserted by call to hardware)


Somehow, the new CCU code sets the CPUX to 24MHz no matter what.

I'm using your pen/clk-rework branch without any other patches that were
later sent to the mailing list.

regards,
  Ondrej

> 
> Maxime
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [linux-sunxi] Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Chen-Yu Tsai]

Hi,

On Thu, Jul 28, 2016 at 7:27 PM, Ondřej Jirman <megous@megous.com> wrote:
> Hi Maxime,
>
> I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
> to this.
>
> On 26.7.2016 08:32, Maxime Ripard wrote:
>> On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
>>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>>>> applying the factors, and then switching back when the PLL is stable
>>>>>> would be a nice solution.
>>>>>>
>>>>>> I just checked, and all the SoCs we've had so far have that
>>>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>>>
>>>>> It would need to be tested. U-boot does the change only once, while the
>>>>> kernel would be doing it all the time and between various frequencies
>>>>> and PLL settings. So the issues may show up with this solution too.
>>>>
>>>> That would have the benefit of being quite easy to document, not be a
>>>> huge amount of code and it would work on all the CPUs PLLs we have so
>>>> far, so still, a pretty big win. If it doesn't, of course, we don't
>>>> really have the choice.
>>>
>>> It's probably more code though. It has to access different register from
>>> the one that is already defined in dts, which would add a lot of code
>>> and require dts changes. The original patch I sent is simpler than that.
>>
>> Why?
>>
>> You can use container_of to retrieve the parent structure of the clock
>> notifier, and then you get a ccu_common structure pointer, with the
>> CCU base address, the clock register, its lock, etc.
>>
>> Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
>>
>> I don't really get why anything should be changed in the DT, or why it
>> would add a lot of code. Or maybe we're not talking about the same
>> thing?
>
> I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
> it very liberally uses divider parameters, even in situations that are
> out of spec compared to the current code in the kernel.
>
> In the current code and especially in the original vendor code, divider
> parameters are used as last resort only. Presumably because, of the
> inherent trouble in changing them, as I described to you in other email.
>
> The new ccu code uses dividers often and even at very high frequencies,
> which goes against the spec.
>
> In the vendor code M is never anything else but 0, and P is used only
> for frequencies below 288MHz, which matches the H3 datasheet, which says:
>
> "The P factor only use in the condition that PLL output less than 288
> MHz."
>
> Also other datasheets of similar socs from Allwinner state that M should
> not be used in production code. So it may be that they either forgot to
> state it in the H3 datasheet, or it can be used. In any case, they never
> use M in their code, so it may be wise to keep to that.
>
> When I boot with the new CCU code I get this:
>
> PLL_CPUX = 0x00001B21 : M = 2, K = 3, N = 28, P = 1, EN = 0
> PLL_CPUX : freq = 1008MHz
>
> Mathematically it works, but it is against the spec.
>
> Also, this:
>
> analyzing CPU 0:
>   driver: cpufreq-dt
>   CPUs which run at the same hardware frequency: 0 1 2 3
>   CPUs which need to have their frequency coordinated by software: 0 1 2 3
>   maximum transition latency: 1.74 ms
>   hardware limits: 120 MHz - 1.37 GHz
>   available frequency steps:  120 MHz, 240 MHz, 480 MHz, 648 MHz, 816
> MHz, 960 MHz, 1.01 GHz, 1.06 GHz, 1.10 GHz, 1.15 GHz, 1.20 GHz, 1.22
> GHz, 1.25 GHz, 1.30 GHz, 1.34 GHz, 1.37 GHz
>   available cpufreq governors: conservative ondemand userspace powersave
> performance
>   current policy: frequency should be within 240 MHz and 240 MHz.
>                   The governor "performance" may decide which speed to use
>                   within this range.
>   current CPU frequency: 24.0 MHz (asserted by call to hardware)
>
>
> Somehow, the new CCU code sets the CPUX to 24MHz no matter what.
>
> I'm using your pen/clk-rework branch without any other patches that were
> later sent to the mailing list.

The H3 CPUX clock does not have the CLK_SET_RATE_PARENT flag set, which
means cpufreq's attempts to set the clk rate will, in some cases, be
ignored as the change is not propagated up to the parent PLL. In the
worst case it will select the 24 MHz oscillator. The fix is to add
CLK_SET_RATE_PARENT. And then you'll have to deal with the PLL potentially
going too high, as you stated in your other mail, and can be mitigated
with my clk notifier patch.

ChenYu

> regards,
>   Ondrej
>
>>
>> Maxime
>>
>
> --
> You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 5744 bytes --]

Hi Ondrej,

On Thu, Jul 28, 2016 at 01:27:05PM +0200, Ondřej Jirman wrote:
> Hi Maxime,
> 
> I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
> to this.

You can find it in the clock maintainers tree:
https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/log/?h=clk-sunxi-ng

> On 26.7.2016 08:32, Maxime Ripard wrote:
> > On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
> >>>>> If so, then yes, trying to switch to the 24MHz oscillator before
> >>>>> applying the factors, and then switching back when the PLL is stable
> >>>>> would be a nice solution.
> >>>>>
> >>>>> I just checked, and all the SoCs we've had so far have that
> >>>>> possibility, so if it works, for now, I'd like to stick to that.
> >>>>
> >>>> It would need to be tested. U-boot does the change only once, while the
> >>>> kernel would be doing it all the time and between various frequencies
> >>>> and PLL settings. So the issues may show up with this solution too.
> >>>
> >>> That would have the benefit of being quite easy to document, not be a
> >>> huge amount of code and it would work on all the CPUs PLLs we have so
> >>> far, so still, a pretty big win. If it doesn't, of course, we don't
> >>> really have the choice.
> >>
> >> It's probably more code though. It has to access different register from
> >> the one that is already defined in dts, which would add a lot of code
> >> and require dts changes. The original patch I sent is simpler than that.
> > 
> > Why?
> > 
> > You can use container_of to retrieve the parent structure of the clock
> > notifier, and then you get a ccu_common structure pointer, with the
> > CCU base address, the clock register, its lock, etc.
> > 
> > Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
> > 
> > I don't really get why anything should be changed in the DT, or why it
> > would add a lot of code. Or maybe we're not talking about the same
> > thing?
> 
> I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
> it very liberally uses divider parameters, even in situations that are
> out of spec compared to the current code in the kernel.
> 
> In the current code and especially in the original vendor code, divider
> parameters are used as last resort only. Presumably because, of the
> inherent trouble in changing them, as I described to you in other email.
> 
> The new ccu code uses dividers often and even at very high frequencies,
> which goes against the spec.
> 
> In the vendor code M is never anything else but 0, and P is used only
> for frequencies below 288MHz, which matches the H3 datasheet, which says:

In the vendor code, P is never used either. All the boards we had so
far don't go that low, so we cannot make any of these assumptions,
especially since the vendor code has had the bad habit of doing
something wrong and / or useless in the past.

However, this implementation is a new thing, and it was using the H3
precisely because of its early stage of support to use it as a testbed
for the more established.

If you feel like we should use a different formula to favour the
multipliers over the dividers (or want to change the class of the CPU
PLL to an NKM or something else, this is totally doable.

> "The P factor only use in the condition that PLL output less than 288
> MHz."

And the datasheet also had some issues, either misleading or wrong
comments in the past. Don't get me wrong, I'm not saying that this is
wrong, just that we should not follow it religiously, and that we
should trust more the experiments than the datasheet.

> Also other datasheets of similar socs from Allwinner state that M should
> not be used in production code.

Which ones specifically?

> So it may be that they either forgot to state it in the H3
> datasheet, or it can be used. In any case, they never use M in their
> code, so it may be wise to keep to that.
> 
> When I boot with the new CCU code I get this:
> 
> PLL_CPUX = 0x00001B21 : M = 2, K = 3, N = 28, P = 1, EN = 0
> PLL_CPUX : freq = 1008MHz
> 
> Mathematically it works, but it is against the spec.
> 
> Also, this:
> 
> analyzing CPU 0:
>   driver: cpufreq-dt
>   CPUs which run at the same hardware frequency: 0 1 2 3
>   CPUs which need to have their frequency coordinated by software: 0 1 2 3
>   maximum transition latency: 1.74 ms
>   hardware limits: 120 MHz - 1.37 GHz
>   available frequency steps:  120 MHz, 240 MHz, 480 MHz, 648 MHz, 816
> MHz, 960 MHz, 1.01 GHz, 1.06 GHz, 1.10 GHz, 1.15 GHz, 1.20 GHz, 1.22
> GHz, 1.25 GHz, 1.30 GHz, 1.34 GHz, 1.37 GHz
>   available cpufreq governors: conservative ondemand userspace powersave
> performance
>   current policy: frequency should be within 240 MHz and 240 MHz.
>                   The governor "performance" may decide which speed to use
>                   within this range.
>   current CPU frequency: 24.0 MHz (asserted by call to hardware)
> 
> 
> Somehow, the new CCU code sets the CPUX to 24MHz no matter what.
> 
> I'm using your pen/clk-rework branch without any other patches that were
> later sent to the mailing list.

It's probably because of CLK_SET_RATE_PARENT on the CPU clock missing,
as Chen-Yu suggested.

Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 6396 bytes --]

On 28.7.2016 23:00, Maxime Ripard wrote:
> Hi Ondrej,
> 
> On Thu, Jul 28, 2016 at 01:27:05PM +0200, Ondřej Jirman wrote:
>> Hi Maxime,
>>
>> I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
>> to this.
> 
> You can find it in the clock maintainers tree:
> https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/log/?h=clk-sunxi-ng
> 
>> On 26.7.2016 08:32, Maxime Ripard wrote:
>>> On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
>>>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>>>>> applying the factors, and then switching back when the PLL is stable
>>>>>>> would be a nice solution.
>>>>>>>
>>>>>>> I just checked, and all the SoCs we've had so far have that
>>>>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>>>>
>>>>>> It would need to be tested. U-boot does the change only once, while the
>>>>>> kernel would be doing it all the time and between various frequencies
>>>>>> and PLL settings. So the issues may show up with this solution too.
>>>>>
>>>>> That would have the benefit of being quite easy to document, not be a
>>>>> huge amount of code and it would work on all the CPUs PLLs we have so
>>>>> far, so still, a pretty big win. If it doesn't, of course, we don't
>>>>> really have the choice.
>>>>
>>>> It's probably more code though. It has to access different register from
>>>> the one that is already defined in dts, which would add a lot of code
>>>> and require dts changes. The original patch I sent is simpler than that.
>>>
>>> Why?
>>>
>>> You can use container_of to retrieve the parent structure of the clock
>>> notifier, and then you get a ccu_common structure pointer, with the
>>> CCU base address, the clock register, its lock, etc.
>>>
>>> Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
>>>
>>> I don't really get why anything should be changed in the DT, or why it
>>> would add a lot of code. Or maybe we're not talking about the same
>>> thing?
>>
>> I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
>> it very liberally uses divider parameters, even in situations that are
>> out of spec compared to the current code in the kernel.
>>
>> In the current code and especially in the original vendor code, divider
>> parameters are used as last resort only. Presumably because, of the
>> inherent trouble in changing them, as I described to you in other email.
>>
>> The new ccu code uses dividers often and even at very high frequencies,
>> which goes against the spec.
>>
>> In the vendor code M is never anything else but 0, and P is used only
>> for frequencies below 288MHz, which matches the H3 datasheet, which says:
> 
> In the vendor code, P is never used either. All the boards we had so
> far don't go that low, so we cannot make any of these assumptions,
> especially since the vendor code has had the bad habit of doing
> something wrong and / or useless in the past.

P is used in the arisc firmware according to the spec for the lower
frequencies.

> However, this implementation is a new thing, and it was using the H3
> precisely because of its early stage of support to use it as a testbed
> for the more established.
> 
> If you feel like we should use a different formula to favour the
> multipliers over the dividers (or want to change the class of the CPU
> PLL to an NKM or something else, this is totally doable.

I think the original formula that's currently in the mainline kernel is
better and avoids fiddling with dividers too much.

>> "The P factor only use in the condition that PLL output less than 288
>> MHz."
> 
> And the datasheet also had some issues, either misleading or wrong
> comments in the past. Don't get me wrong, I'm not saying that this is
> wrong, just that we should not follow it religiously, and that we
> should trust more the experiments than the datasheet.

I can believe that. :) Regardless, I think the reasons given for
avoiding dividers are quite reasonable. It's based on how PLL block
works, not what manual says.

>> Also other datasheets of similar socs from Allwinner state that M should
>> not be used in production code.
> 
> Which ones specifically?

A83T for example. You can search for "only for test" phrase.

https://github.com/allwinner-zh/documents/blob/master/A83T/A83T_User_Manual_v1.5.1_20150513.pdf

Those PLLs are a bit different though.


regards,
  Ondrej

>> So it may be that they either forgot to state it in the H3
>> datasheet, or it can be used. In any case, they never use M in their
>> code, so it may be wise to keep to that.
>>
>> When I boot with the new CCU code I get this:
>>
>> PLL_CPUX = 0x00001B21 : M = 2, K = 3, N = 28, P = 1, EN = 0
>> PLL_CPUX : freq = 1008MHz
>>
>> Mathematically it works, but it is against the spec.
>>
>> Also, this:
>>
>> analyzing CPU 0:
>>   driver: cpufreq-dt
>>   CPUs which run at the same hardware frequency: 0 1 2 3
>>   CPUs which need to have their frequency coordinated by software: 0 1 2 3
>>   maximum transition latency: 1.74 ms
>>   hardware limits: 120 MHz - 1.37 GHz
>>   available frequency steps:  120 MHz, 240 MHz, 480 MHz, 648 MHz, 816
>> MHz, 960 MHz, 1.01 GHz, 1.06 GHz, 1.10 GHz, 1.15 GHz, 1.20 GHz, 1.22
>> GHz, 1.25 GHz, 1.30 GHz, 1.34 GHz, 1.37 GHz
>>   available cpufreq governors: conservative ondemand userspace powersave
>> performance
>>   current policy: frequency should be within 240 MHz and 240 MHz.
>>                   The governor "performance" may decide which speed to use
>>                   within this range.
>>   current CPU frequency: 24.0 MHz (asserted by call to hardware)
>>
>>
>> Somehow, the new CCU code sets the CPUX to 24MHz no matter what.
>>
>> I'm using your pen/clk-rework branch without any other patches that were
>> later sent to the mailing list.
> 
> It's probably because of CLK_SET_RATE_PARENT on the CPU clock missing,
> as Chen-Yu suggested.
> 
> Maxime
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 6269 bytes --]

Hi,

On Fri, Jul 29, 2016 at 12:01:09AM +0200, Ondřej Jirman wrote:
> On 28.7.2016 23:00, Maxime Ripard wrote:
> > Hi Ondrej,
> > 
> > On Thu, Jul 28, 2016 at 01:27:05PM +0200, Ondřej Jirman wrote:
> >> Hi Maxime,
> >>
> >> I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
> >> to this.
> > 
> > You can find it in the clock maintainers tree:
> > https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/log/?h=clk-sunxi-ng
> > 
> >> On 26.7.2016 08:32, Maxime Ripard wrote:
> >>> On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
> >>>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
> >>>>>>> applying the factors, and then switching back when the PLL is stable
> >>>>>>> would be a nice solution.
> >>>>>>>
> >>>>>>> I just checked, and all the SoCs we've had so far have that
> >>>>>>> possibility, so if it works, for now, I'd like to stick to that.
> >>>>>>
> >>>>>> It would need to be tested. U-boot does the change only once, while the
> >>>>>> kernel would be doing it all the time and between various frequencies
> >>>>>> and PLL settings. So the issues may show up with this solution too.
> >>>>>
> >>>>> That would have the benefit of being quite easy to document, not be a
> >>>>> huge amount of code and it would work on all the CPUs PLLs we have so
> >>>>> far, so still, a pretty big win. If it doesn't, of course, we don't
> >>>>> really have the choice.
> >>>>
> >>>> It's probably more code though. It has to access different register from
> >>>> the one that is already defined in dts, which would add a lot of code
> >>>> and require dts changes. The original patch I sent is simpler than that.
> >>>
> >>> Why?
> >>>
> >>> You can use container_of to retrieve the parent structure of the clock
> >>> notifier, and then you get a ccu_common structure pointer, with the
> >>> CCU base address, the clock register, its lock, etc.
> >>>
> >>> Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
> >>>
> >>> I don't really get why anything should be changed in the DT, or why it
> >>> would add a lot of code. Or maybe we're not talking about the same
> >>> thing?
> >>
> >> I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
> >> it very liberally uses divider parameters, even in situations that are
> >> out of spec compared to the current code in the kernel.
> >>
> >> In the current code and especially in the original vendor code, divider
> >> parameters are used as last resort only. Presumably because, of the
> >> inherent trouble in changing them, as I described to you in other email.
> >>
> >> The new ccu code uses dividers often and even at very high frequencies,
> >> which goes against the spec.
> >>
> >> In the vendor code M is never anything else but 0, and P is used only
> >> for frequencies below 288MHz, which matches the H3 datasheet, which says:
> > 
> > In the vendor code, P is never used either. All the boards we had so
> > far don't go that low, so we cannot make any of these assumptions,
> > especially since the vendor code has had the bad habit of doing
> > something wrong and / or useless in the past.
> 
> P is used in the arisc firmware according to the spec for the lower
> frequencies.

Yes, but has anyone actually tested those frequencies? Judging from
the FEX files I could gather, cpufreq never actually goes lower than
480 MHz.

> > However, this implementation is a new thing, and it was using the H3
> > precisely because of its early stage of support to use it as a testbed
> > for the more established.
> > 
> > If you feel like we should use a different formula to favour the
> > multipliers over the dividers (or want to change the class of the CPU
> > PLL to an NKM or something else, this is totally doable.
> 
> I think the original formula that's currently in the mainline kernel is
> better and avoids fiddling with dividers too much.

Yeah, but the older formula is not generic at all. The whole rework
was precisely to avoid doing the whole one driver per clock that was
just becoming a nightmare to maintain, and a pain to add support for
new SoCs. That code will be used for A10's CPU and VE PLLs too for
example. And they probably have the same constraints, but with
different variations (available values of each factors for example).

> >> "The P factor only use in the condition that PLL output less than 288
> >> MHz."
> > 
> > And the datasheet also had some issues, either misleading or wrong
> > comments in the past. Don't get me wrong, I'm not saying that this is
> > wrong, just that we should not follow it religiously, and that we
> > should trust more the experiments than the datasheet.
> 
> I can believe that. :) Regardless, I think the reasons given for
> avoiding dividers are quite reasonable. It's based on how PLL block
> works, not what manual says.

Yes, indeed.

Would replacing the current factors computation function by something
like:

for (m = 1; m < max_m; m++)
    for (p = 1; p < max_p; p++)
        for (n = 1; n < max_n; n++)
	     for (k = 1; k < max_k; k++)
	         if rate == computed rate
		     break;

work for you?

That way, we will favor the multipliers over the dividers, and we can
always "blacklist" p or m (or both) by setting their maximum to 1
(this would need an extra field in _ccu_div though).

> >> Also other datasheets of similar socs from Allwinner state that M should
> >> not be used in production code.
> > 
> > Which ones specifically?
> 
> A83T for example. You can search for "only for test" phrase.
> 
> https://github.com/allwinner-zh/documents/blob/master/A83T/A83T_User_Manual_v1.5.1_20150513.pdf
> 
> Those PLLs are a bit different though.

Thanks,
Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 6892 bytes --]

Hi,

On 31.7.2016 12:31, Maxime Ripard wrote:
> Hi,
> 
> On Fri, Jul 29, 2016 at 12:01:09AM +0200, Ondřej Jirman wrote:
>> On 28.7.2016 23:00, Maxime Ripard wrote:
>>> Hi Ondrej,
>>>
>>> On Thu, Jul 28, 2016 at 01:27:05PM +0200, Ondřej Jirman wrote:
>>>> Hi Maxime,
>>>>
>>>> I don't have your sunxi-ng clock patches in my mailbox, so I'm replying
>>>> to this.
>>>
>>> You can find it in the clock maintainers tree:
>>> https://git.kernel.org/cgit/linux/kernel/git/clk/linux.git/log/?h=clk-sunxi-ng
>>>
>>>> On 26.7.2016 08:32, Maxime Ripard wrote:
>>>>> On Thu, Jul 21, 2016 at 11:52:15AM +0200, Ondřej Jirman wrote:
>>>>>>>>> If so, then yes, trying to switch to the 24MHz oscillator before
>>>>>>>>> applying the factors, and then switching back when the PLL is stable
>>>>>>>>> would be a nice solution.
>>>>>>>>>
>>>>>>>>> I just checked, and all the SoCs we've had so far have that
>>>>>>>>> possibility, so if it works, for now, I'd like to stick to that.
>>>>>>>>
>>>>>>>> It would need to be tested. U-boot does the change only once, while the
>>>>>>>> kernel would be doing it all the time and between various frequencies
>>>>>>>> and PLL settings. So the issues may show up with this solution too.
>>>>>>>
>>>>>>> That would have the benefit of being quite easy to document, not be a
>>>>>>> huge amount of code and it would work on all the CPUs PLLs we have so
>>>>>>> far, so still, a pretty big win. If it doesn't, of course, we don't
>>>>>>> really have the choice.
>>>>>>
>>>>>> It's probably more code though. It has to access different register from
>>>>>> the one that is already defined in dts, which would add a lot of code
>>>>>> and require dts changes. The original patch I sent is simpler than that.
>>>>>
>>>>> Why?
>>>>>
>>>>> You can use container_of to retrieve the parent structure of the clock
>>>>> notifier, and then you get a ccu_common structure pointer, with the
>>>>> CCU base address, the clock register, its lock, etc.
>>>>>
>>>>> Look at what is done in drivers/clk/meson/clk-cpu.c. It's like 20 LoC.
>>>>>
>>>>> I don't really get why anything should be changed in the DT, or why it
>>>>> would add a lot of code. Or maybe we're not talking about the same
>>>>> thing?
>>>>
>>>> I've looked at the new CCU code, particularly ccu_nkmp.c, and found that
>>>> it very liberally uses divider parameters, even in situations that are
>>>> out of spec compared to the current code in the kernel.
>>>>
>>>> In the current code and especially in the original vendor code, divider
>>>> parameters are used as last resort only. Presumably because, of the
>>>> inherent trouble in changing them, as I described to you in other email.
>>>>
>>>> The new ccu code uses dividers often and even at very high frequencies,
>>>> which goes against the spec.
>>>>
>>>> In the vendor code M is never anything else but 0, and P is used only
>>>> for frequencies below 288MHz, which matches the H3 datasheet, which says:
>>>
>>> In the vendor code, P is never used either. All the boards we had so
>>> far don't go that low, so we cannot make any of these assumptions,
>>> especially since the vendor code has had the bad habit of doing
>>> something wrong and / or useless in the past.
>>
>> P is used in the arisc firmware according to the spec for the lower
>> frequencies.
> 
> Yes, but has anyone actually tested those frequencies? Judging from
> the FEX files I could gather, cpufreq never actually goes lower than
> 480 MHz.
> 

I tested it. It works well down to 60MHz. You can even run on 24MHz if
you run directly from 24mhz osc. It's terribly slow, but it works ok.

I've rebased my working branch over the mainline kernel, which now
contains the sunxi-ng, and tested it. Cpufreq seems to work on orange pi
pc without any changes discussed in this thread. I didn't do any
extensive testing though. But it doesn't hang on boot or cpufreq config
changes.

You can see it here:

https://github.com/megous/linux/commits/orange-pi-4.8

>>> However, this implementation is a new thing, and it was using the H3
>>> precisely because of its early stage of support to use it as a testbed
>>> for the more established.
>>>
>>> If you feel like we should use a different formula to favour the
>>> multipliers over the dividers (or want to change the class of the CPU
>>> PLL to an NKM or something else, this is totally doable.
>>
>> I think the original formula that's currently in the mainline kernel is
>> better and avoids fiddling with dividers too much.
> 
> Yeah, but the older formula is not generic at all. The whole rework
> was precisely to avoid doing the whole one driver per clock that was
> just becoming a nightmare to maintain, and a pain to add support for
> new SoCs. That code will be used for A10's CPU and VE PLLs too for
> example. And they probably have the same constraints, but with
> different variations (available values of each factors for example).

Fair enough.

>>>> "The P factor only use in the condition that PLL output less than 288
>>>> MHz."
>>>
>>> And the datasheet also had some issues, either misleading or wrong
>>> comments in the past. Don't get me wrong, I'm not saying that this is
>>> wrong, just that we should not follow it religiously, and that we
>>> should trust more the experiments than the datasheet.
>>
>> I can believe that. :) Regardless, I think the reasons given for
>> avoiding dividers are quite reasonable. It's based on how PLL block
>> works, not what manual says.
> 
> Yes, indeed.
> 
> Would replacing the current factors computation function by something
> like:
> 
> for (m = 1; m < max_m; m++)
>     for (p = 1; p < max_p; p++)
>         for (n = 1; n < max_n; n++)
> 	     for (k = 1; k < max_k; k++)
> 	         if rate == computed rate
> 		     break;
> 
> work for you?

This would be better.

thank you and regards,
  Ondrej

> That way, we will favor the multipliers over the dividers, and we can
> always "blacklist" p or m (or both) by setting their maximum to 1
> (this would need an extra field in _ccu_div though).
> 
>>>> Also other datasheets of similar socs from Allwinner state that M should
>>>> not be used in production code.
>>>
>>> Which ones specifically?
>>
>> A83T for example. You can search for "only for test" phrase.
>>
>> https://github.com/allwinner-zh/documents/blob/master/A83T/A83T_User_Manual_v1.5.1_20150513.pdf
>>
>> Those PLLs are a bit different though.
> 
> Thanks,
> Maxime
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Changed: sunxi-ng clock code - NKMP clock implementation is wrong

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 3281 bytes --]

Hi Ondrej,

Sorry for the (very) belated reply.

On Mon, Aug 01, 2016 at 12:01:39AM +0200, Ondřej Jirman wrote:
> >>>> In the vendor code M is never anything else but 0, and P is used only
> >>>> for frequencies below 288MHz, which matches the H3 datasheet, which says:
> >>>
> >>> In the vendor code, P is never used either. All the boards we had so
> >>> far don't go that low, so we cannot make any of these assumptions,
> >>> especially since the vendor code has had the bad habit of doing
> >>> something wrong and / or useless in the past.
> >>
> >> P is used in the arisc firmware according to the spec for the lower
> >> frequencies.
> > 
> > Yes, but has anyone actually tested those frequencies? Judging from
> > the FEX files I could gather, cpufreq never actually goes lower than
> > 480 MHz.
> > 
> 
> I tested it. It works well down to 60MHz. You can even run on 24MHz if
> you run directly from 24mhz osc. It's terribly slow, but it works ok.

Ok, it's good to know (and yeah, I wouldn't expect a CPU at 24MHz to
be very fast :))

> I've rebased my working branch over the mainline kernel, which now
> contains the sunxi-ng, and tested it. Cpufreq seems to work on orange pi
> pc without any changes discussed in this thread. I didn't do any
> extensive testing though. But it doesn't hang on boot or cpufreq config
> changes.
> 
> You can see it here:
> 
> https://github.com/megous/linux/commits/orange-pi-4.8

Thanks a lot. That's good to know as well.

> >>>> "The P factor only use in the condition that PLL output less than 288
> >>>> MHz."
> >>>
> >>> And the datasheet also had some issues, either misleading or wrong
> >>> comments in the past. Don't get me wrong, I'm not saying that this is
> >>> wrong, just that we should not follow it religiously, and that we
> >>> should trust more the experiments than the datasheet.
> >>
> >> I can believe that. :) Regardless, I think the reasons given for
> >> avoiding dividers are quite reasonable. It's based on how PLL block
> >> works, not what manual says.
> > 
> > Yes, indeed.
> > 
> > Would replacing the current factors computation function by something
> > like:
> > 
> > for (m = 1; m < max_m; m++)
> >     for (p = 1; p < max_p; p++)
> >         for (n = 1; n < max_n; n++)
> > 	     for (k = 1; k < max_k; k++)
> > 	         if rate == computed rate
> > 		     break;
> > 
> > work for you?
> 
> This would be better.

Ok, I'll try to cook something up when I get the time. If you feel
like it's not done fast enough, feel free to do it.

Note that we also ended up merging for the A31 some code that reparent
CPUx temporarily while setting the PLL-CPUX rate, and it should be
somewhat generic. But of course, it's not exclusive.

Thanks again for all your tests, it's very appreciated.

Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 3626 bytes --]

On 30.6.2016 22:40, Maxime Ripard wrote:
> Hi,
> 
> On Sat, Jun 25, 2016 at 05:45:03AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>
>> PLL1 on H3 requires special factors application algorithm,
>> when the rate is changed. This algorithm was extracted
>> from the arisc code that handles frequency scaling
>> in the BSP kernel.
>>
>> This commit adds optional apply function to
>> struct factors_data, that can implement non-trivial
>> factors application method, when necessary.
>>
>> Also struct clk_factors_config is extended with position
>> of the PLL lock flag.
> 
> Have you tested the current implementation, and found that it was not
> working, or did you duplicate the arisc code directly?

Also of note is that similar code probably doesn't crash in u-boot,
because there, before changing the PLL1 clock, the cpu is switched to
24MHz osc, so it is not overclocked, even if factors align in such a way
that you'd get the behavior I described in the other email.

>>  /**
>> + * sun8i_h3_apply_pll1_factors() - applies n, k, m, p factors to the
>> + * register using an algorithm that tries to reserve the PLL lock
>> + */
>> +
>> +static void sun8i_h3_apply_pll1_factors(struct clk_factors *factors, struct factors_request *req)
>> +{
>> +	const struct clk_factors_config *config = factors->config;
>> +	u32 reg;
>> +
>> +	/* Fetch the register value */
>> +	reg = readl(factors->reg);
>> +
>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) < req->p) {
>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
> 
> So there was some doubts about the fact that P was being used, or at
> least that it was useful.
> 
>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) < req->m) {
>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
>> +
>> +	reg = FACTOR_SET(config->nshift, config->nwidth, reg, req->n);
>> +	reg = FACTOR_SET(config->kshift, config->kwidth, reg, req->k);
>> +
>> +	writel(reg, factors->reg);
>> +	__delay(20);
>> +
>> +	while (!(readl(factors->reg) & (1 << config->lock)));
> 
> So, they are applying the dividers first, and then applying the
> multipliers, and then wait for the PLL to stabilize.
> 
>> +
>> +	if (FACTOR_GET(config->mshift, config->mwidth, reg) > req->m) {
>> +		reg = FACTOR_SET(config->mshift, config->mwidth, reg, req->m);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
>> +
>> +	if (FACTOR_GET(config->pshift, config->pwidth, reg) > req->p) {
>> +		reg = FACTOR_SET(config->pshift, config->pwidth, reg, req->p);
>> +
>> +		writel(reg, factors->reg);
>> +		__delay(2000);
>> +	}
> 
> However, this is kind of weird, why would you need to re-apply the
> dividers? Nothing really changes. Have you tried without that part?
> 
> Since this is really specific, I guess you could simply make the
> clk_ops for the nkmp clocks public, and just re-implement set_rate
> using that logic.
> 
> You might also need to set an upper limit on P, since the last value
> (4) is not a valid one.
> 
> I guess you could do that by adding a max field in the __ccu_div
> structure.
> 
> Maxime
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 06/14] ARM: sun8i: clk: Add clk-factor rate application method

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1732 bytes --]

On Fri, Jul 01, 2016 at 02:53:52AM +0200, Ondřej Jirman wrote:
> On 30.6.2016 22:40, Maxime Ripard wrote:
> > Hi,
> > 
> > On Sat, Jun 25, 2016 at 05:45:03AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> >> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> >>
> >> PLL1 on H3 requires special factors application algorithm,
> >> when the rate is changed. This algorithm was extracted
> >> from the arisc code that handles frequency scaling
> >> in the BSP kernel.
> >>
> >> This commit adds optional apply function to
> >> struct factors_data, that can implement non-trivial
> >> factors application method, when necessary.
> >>
> >> Also struct clk_factors_config is extended with position
> >> of the PLL lock flag.
> > 
> > Have you tested the current implementation, and found that it was not
> > working, or did you duplicate the arisc code directly?
> 
> Also of note is that similar code probably doesn't crash in u-boot,
> because there, before changing the PLL1 clock, the cpu is switched to
> 24MHz osc, so it is not overclocked, even if factors align in such a way
> that you'd get the behavior I described in the other email.

That's also something that we can do.

See Meson's clk-cpu clock notifiers for example.

Maxime

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

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[PATCH v2 07/14] ARM: dts: sun8i: Use sun8i-h3-pll1-clk for pll1 in H3

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

PLL1 on H3 requires special factors application algorithm,
when the rate is changed. This algorithm was extracted
from the arisc code that handles frequency scaling
in the BSP kernel.

This algorithm is implemented by sun8i-h3-pll1-clk.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 arch/arm/boot/dts/sun8i-h3.dtsi | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm/boot/dts/sun8i-h3.dtsi b/arch/arm/boot/dts/sun8i-h3.dtsi
index 4a4926b..b3247f4 100644
--- a/arch/arm/boot/dts/sun8i-h3.dtsi
+++ b/arch/arm/boot/dts/sun8i-h3.dtsi
@@ -106,7 +106,7 @@
 
 		pll1: clk@01c20000 {
 			#clock-cells = <0>;
-			compatible = "allwinner,sun8i-a23-pll1-clk";
+			compatible = "allwinner,sun8i-h3-pll1-clk";
 			reg = <0x01c20000 0x4>;
 			clocks = <&osc24M>;
 			clock-output-names = "pll1";
-- 
2.9.0

[PATCH v2 08/14] ARM: dts: sun8i: Add thermal sensor node to the sun8i-h3.dtsi

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

This patch adds nodes for the thermal sensor driver and
the THS clock to the Allwinner sun8i-h3.dtsi file.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 arch/arm/boot/dts/sun8i-h3.dtsi | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/arch/arm/boot/dts/sun8i-h3.dtsi b/arch/arm/boot/dts/sun8i-h3.dtsi
index b3247f4..d3c29cc 100644
--- a/arch/arm/boot/dts/sun8i-h3.dtsi
+++ b/arch/arm/boot/dts/sun8i-h3.dtsi
@@ -77,6 +77,14 @@
 		};
 	};
 
+	thermal-zones {
+		cpu_thermal: cpu_thermal {
+			polling-delay-passive = <330>;
+			polling-delay = <1000>;
+			thermal-sensors = <&ths 0>;
+		};
+	};
+
 	timer {
 		compatible = "arm,armv7-timer";
 		interrupts = <GIC_PPI 13 (GIC_CPU_MASK_SIMPLE(4) | IRQ_TYPE_LEVEL_LOW)>,
@@ -239,6 +247,14 @@
 					     "bus_scr", "bus_ephy", "bus_dbg";
 		};
 
+		ths_clk: clk@01c20074 {
+			#clock-cells = <0>;
+			compatible = "allwinner,sun8i-h3-ths-clk";
+			reg = <0x01c20074 0x4>;
+			clocks = <&osc24M>;
+			clock-output-names = "ths";
+		};
+
 		mmc0_clk: clk@01c20088 {
 			#clock-cells = <1>;
 			compatible = "allwinner,sun4i-a10-mmc-clk";
@@ -574,6 +590,18 @@
 			interrupts = <GIC_SPI 25 IRQ_TYPE_LEVEL_HIGH>;
 		};
 
+		ths: ths@01c25000 {
+			#thermal-sensor-cells = <0>;
+			compatible = "allwinner,sun8i-h3-ths";
+			reg = <0x01c25000 0x400>,
+			      <0x01c14234 0x4>;
+			interrupts = <GIC_SPI 31 IRQ_TYPE_LEVEL_HIGH>;
+			resets = <&apb1_rst 8>;
+			reset-names = "ahb";
+			clocks = <&bus_gates 72>, <&ths_clk>;
+			clock-names = "ahb", "ths";
+		};
+
 		uart0: serial@01c28000 {
 			compatible = "snps,dw-apb-uart";
 			reg = <0x01c28000 0x400>;
-- 
2.9.0

[PATCH v2 09/14] ARM: dts: sun8i: Add cpu0 label to sun8i-h3.dtsi

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

Add label to the first cpu so that it can be referenced
from derived dts files.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
v2
- move clocks/clock-latency to sun8i-h3.dtsi  
---
 arch/arm/boot/dts/sun8i-h3.dtsi | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/arm/boot/dts/sun8i-h3.dtsi b/arch/arm/boot/dts/sun8i-h3.dtsi
index d3c29cc..56f211e 100644
--- a/arch/arm/boot/dts/sun8i-h3.dtsi
+++ b/arch/arm/boot/dts/sun8i-h3.dtsi
@@ -52,7 +52,9 @@
 		#address-cells = <1>;
 		#size-cells = <0>;
 
-		cpu@0 {
+		cpu0: cpu@0 {
+			clocks = <&cpu>;
+			clock-latency = <244144>; /* 8 32k periods */
 			compatible = "arm,cortex-a7";
 			device_type = "cpu";
 			reg = <0>;
-- 
2.9.0

[PATCH v2 10/14] ARM: dts: sun8i: Add r_twi I2C controller

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

H3 SoC contains I2C controller optionally available
on the PL0 and PL1 pins. This patch makes this controller
available.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 arch/arm/boot/dts/sun8i-h3.dtsi | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/arch/arm/boot/dts/sun8i-h3.dtsi b/arch/arm/boot/dts/sun8i-h3.dtsi
index 56f211e..e32f211 100644
--- a/arch/arm/boot/dts/sun8i-h3.dtsi
+++ b/arch/arm/boot/dts/sun8i-h3.dtsi
@@ -322,8 +322,9 @@
 			reg = <0x01f01428 0x4>;
 			#clock-cells = <1>;
 			clocks = <&apb0>;
-			clock-indices = <0>, <1>;
-			clock-output-names = "apb0_pio", "apb0_ir";
+			clock-indices = <0>, <1>, <6>;
+			clock-output-names = "apb0_pio", "apb0_ir", "apb0_i2c";
+
 		};
 
 		ir_clk: ir_clk@01f01454 {
@@ -656,6 +657,20 @@
 			status = "disabled";
 		};
 
+		r_twi: i2c@01f02400 {
+			compatible = "allwinner,sun6i-a31-i2c";
+			reg = <0x01f02400 0x400>;
+			interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
+			pinctrl-names = "default";
+			pinctrl-0 = <&r_twi_pins_a>;
+			clocks = <&apb0_gates 6>;
+			clock-frequency = <100000>;
+			resets = <&apb0_reset 6>;
+			status = "disabled";
+			#address-cells = <1>;
+			#size-cells = <0>;
+		};
+
 		gic: interrupt-controller@01c81000 {
 			compatible = "arm,cortex-a7-gic", "arm,cortex-a15-gic";
 			reg = <0x01c81000 0x1000>,
@@ -707,6 +722,13 @@
 				allwinner,drive = <SUN4I_PINCTRL_10_MA>;
 				allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;
 			};
+
+			r_twi_pins_a: r_twi@0 {
+				allwinner,pins = "PL0", "PL1";
+				allwinner,function = "s_twi";
+				allwinner,drive = <SUN4I_PINCTRL_10_MA>;
+				allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;
+			};
 		};
 	};
 };
-- 
2.9.0

Re: [PATCH v2 10/14] ARM: dts: sun8i: Add r_twi I2C controller

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 2257 bytes --]

Hi,

On Sat, Jun 25, 2016 at 05:45:07AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> 
> H3 SoC contains I2C controller optionally available
> on the PL0 and PL1 pins. This patch makes this controller
> available.
> 
> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> ---
>  arch/arm/boot/dts/sun8i-h3.dtsi | 26 ++++++++++++++++++++++++--
>  1 file changed, 24 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm/boot/dts/sun8i-h3.dtsi b/arch/arm/boot/dts/sun8i-h3.dtsi
> index 56f211e..e32f211 100644
> --- a/arch/arm/boot/dts/sun8i-h3.dtsi
> +++ b/arch/arm/boot/dts/sun8i-h3.dtsi
> @@ -322,8 +322,9 @@
>  			reg = <0x01f01428 0x4>;
>  			#clock-cells = <1>;
>  			clocks = <&apb0>;
> -			clock-indices = <0>, <1>;
> -			clock-output-names = "apb0_pio", "apb0_ir";
> +			clock-indices = <0>, <1>, <6>;
> +			clock-output-names = "apb0_pio", "apb0_ir", "apb0_i2c";
> +

The new line is not needed, and it should be part of a separate patch.

>  		};
>  
>  		ir_clk: ir_clk@01f01454 {
> @@ -656,6 +657,20 @@
>  			status = "disabled";
>  		};
>  
> +		r_twi: i2c@01f02400 {

We call it i2c everywhere, please call it that way too.

> +			compatible = "allwinner,sun6i-a31-i2c";
> +			reg = <0x01f02400 0x400>;
> +			interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
> +			pinctrl-names = "default";
> +			pinctrl-0 = <&r_twi_pins_a>;
> +			clocks = <&apb0_gates 6>;
> +			clock-frequency = <100000>;
> +			resets = <&apb0_reset 6>;
> +			status = "disabled";
> +			#address-cells = <1>;
> +			#size-cells = <0>;
> +		};
> +
>  		gic: interrupt-controller@01c81000 {
>  			compatible = "arm,cortex-a7-gic", "arm,cortex-a15-gic";
>  			reg = <0x01c81000 0x1000>,
> @@ -707,6 +722,13 @@
>  				allwinner,drive = <SUN4I_PINCTRL_10_MA>;
>  				allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;
>  			};
> +
> +			r_twi_pins_a: r_twi@0 {
> +				allwinner,pins = "PL0", "PL1";
> +				allwinner,function = "s_twi";
> +				allwinner,drive = <SUN4I_PINCTRL_10_MA>;
> +				allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;

Separate patch please.

Thanks!

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

[PATCH v2 11/14] ARM: dts: sun8i: Add sy8106a regulator to Orange Pi PC

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

Add sy8106a regulator to r_twi bus and enable the r_twi bus on
Orange Pi PC. This regulator controls the CPUX voltage.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
index daf50b9a6..79f0b49 100644
--- a/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
+++ b/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
@@ -90,6 +90,20 @@
 	};
 };
 
+&r_twi {
+	status = "okay";
+
+	vdd_cpu: regulator@65 {
+		compatible = "silergy,sy8106a";
+		reg = <0x65>;
+		regulator-min-microvolt = <1000000>;
+		regulator-max-microvolt = <1400000>;
+		regulator-ramp-delay = <200>;
+		regulator-boot-on;
+		regulator-always-on;
+	};
+};
+
 &ehci1 {
 	status = "okay";
 };
-- 
2.9.0

[PATCH v2 12/14] ARM: dts: sun8i: Setup CPU operating points for Onrage PI PC

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

Orange PI PC uses SY8106A regulator for fine grained CPUX voltage
regulation. Setup appropriate operating points for the board.
---
 arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts | 48 ++++++++++++++++++++++++++++++
 1 file changed, 48 insertions(+)

diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
index 79f0b49..1b029e9 100644
--- a/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
+++ b/arch/arm/boot/dts/sun8i-h3-orangepi-pc.dts
@@ -90,6 +90,54 @@
 	};
 };
 
+&cpu0 {
+	operating-points = <
+		/* kHz	  uV */
+		1512000	1400000
+		1440000	1400000
+		1368000	1340000
+		1344000	1340000
+		1296000	1340000
+		1248000	1300000
+		1224000	1300000
+		1200000	1300000
+		1104000	1200000
+		1008000	1140000
+		960000	1100000
+		648000	1100000
+		480000	1100000
+		240000	1100000
+		120000	1100000
+		>;
+	#cooling-cells = <2>;
+	cooling-min-level = <0>;
+	cooling-max-level = <14>;
+	cpu0-supply = <&vdd_cpu>;
+};
+
+&cpu_thermal {
+	cooling-maps {
+		map0 {
+			trip = <&cpu_alert0>;
+			cooling-device = <&cpu0 (-1) (-1)>;
+		};
+	};
+
+	trips {
+		cpu_alert0: cpu_alert0 {
+			temperature = <80000>;
+			hysteresis = <2000>;
+			type = "passive";
+		};
+
+		cpu_crit: cpu_crit {
+			temperature = <100000>;
+			hysteresis = <2000>;
+			type = "critical";
+		};
+	};
+};
+
 &r_twi {
 	status = "okay";
 
-- 
2.9.0

[PATCH v2 13/14] ARM: dts: sun8i: Add gpio-regulator used on Orange Pi One

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

Xulong Orange Pi One uses GPIO based regulator that
switches between two voltages: 1.1V and 1.3V. The
regulator is controlled from the PL6 pin.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
v2
- add missing pinctrl-names for gpio-regulator
---
 arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
index 0adf932..b1bd6b0 100644
--- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
+++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
@@ -88,6 +88,25 @@
 			gpios = <&r_pio 0 3 GPIO_ACTIVE_LOW>;
 		};
 	};
+
+	vdd_soc: gpio-regulator {
+		compatible = "regulator-gpio";
+		pinctrl-names = "default";
+		pinctrl-0 = <&vdd_reg_r_opc>;
+
+		regulator-name = "soc-vdd-supply";
+		regulator-min-microvolt = <1100000>;
+		regulator-max-microvolt = <1300000>;
+		regulator-boot-on;
+		regulator-type = "voltage";
+
+		gpios = <&r_pio 0 6 GPIO_ACTIVE_HIGH>;
+		states = <1100000 0x0
+			  1300000 0x1>;
+
+		startup-delay-us = <100000>;
+		enable-active-high;
+	};
 };
 
 &ehci1 {
@@ -131,6 +150,13 @@
 		allwinner,drive = <SUN4I_PINCTRL_10_MA>;
 		allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;
 	};
+
+	vdd_reg_r_opc: regulator_pins@0 {
+		allwinner,pins = "PL6";
+		allwinner,function = "gpio_out";
+		allwinner,drive = <SUN4I_PINCTRL_10_MA>;
+		allwinner,pull = <SUN4I_PINCTRL_NO_PULL>;
+	};
 };
 
 &uart0 {
-- 
2.9.0

Re: [PATCH v2 13/14] ARM: dts: sun8i: Add gpio-regulator used on Orange Pi One

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1489 bytes --]

On Sat, Jun 25, 2016 at 05:45:10AM +0200, megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org wrote:
> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> 
> Xulong Orange Pi One uses GPIO based regulator that
> switches between two voltages: 1.1V and 1.3V. The
> regulator is controlled from the PL6 pin.
> 
> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> ---
> v2
> - add missing pinctrl-names for gpio-regulator
> ---
>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 26 ++++++++++++++++++++++++++
>  1 file changed, 26 insertions(+)
> 
> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> index 0adf932..b1bd6b0 100644
> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> @@ -88,6 +88,25 @@
>  			gpios = <&r_pio 0 3 GPIO_ACTIVE_LOW>;
>  		};
>  	};
> +
> +	vdd_soc: gpio-regulator {
> +		compatible = "regulator-gpio";
> +		pinctrl-names = "default";
> +		pinctrl-0 = <&vdd_reg_r_opc>;
> +
> +		regulator-name = "soc-vdd-supply";
> +		regulator-min-microvolt = <1100000>;
> +		regulator-max-microvolt = <1300000>;
> +		regulator-boot-on;
> +		regulator-type = "voltage";

It should be marked as always-on.

Otherwise, if the cpufreq driver is not enabled, the regulator will be
shutdown, which is not that great :)

Maxime

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

[PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[megous-5qf/QAjKc83QT0dZR+AlfA]

From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>

Use Xulong Orange Pi One GPIO based regulator for
passive cooling and thermal management.

Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
---
 arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
index b1bd6b0..a38d871 100644
--- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
+++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
@@ -109,6 +109,45 @@
 	};
 };
 
+&cpu0 {
+	operating-points = <
+		/* kHz	  uV */
+		1296000	1300000
+		1200000	1300000
+		624000	1100000
+		480000	1100000
+		312000	1100000
+		240000	1100000
+		>;
+	#cooling-cells = <2>;
+	cooling-min-level = <0>;
+	cooling-max-level = <5>;
+	cpu0-supply = <&vdd_soc>;
+};
+
+&cpu_thermal {
+	cooling-maps {
+		map0 {
+			trip = <&cpu_alert0>;
+			cooling-device = <&cpu0 (-1) (-1)>;
+		};
+	};
+
+	trips {
+		cpu_alert0: cpu_alert0 {
+			temperature = <80000>;
+			hysteresis = <2000>;
+			type = "passive";
+		};
+
+		cpu_crit: cpu_crit {
+			temperature = <100000>;
+			hysteresis = <2000>;
+			type = "critical";
+		};
+	};
+};
+
 &ehci1 {
 	status = "okay";
 };
-- 
2.9.0

Re: [linux-sunxi] [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Michal Suchanek]

Hello,

On 25 June 2016 at 05:45,  <megous@megous.com> wrote:
> From: Ondrej Jirman <megous@megous.com>
>
> Use Xulong Orange Pi One GPIO based regulator for
> passive cooling and thermal management.
>
> Signed-off-by: Ondrej Jirman <megous@megous.com>
> ---
>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
>
> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> index b1bd6b0..a38d871 100644
> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> @@ -109,6 +109,45 @@
>         };
>  };
>
> +&cpu0 {
> +       operating-points = <
> +               /* kHz    uV */
> +               1296000 1300000
> +               1200000 1300000

First problem is that the board boots at 1008000 which is not listed
and the kernel complains.

Second problem is that the board locks up during boot with this enabled.

Do you have some suggestion for alternate configuration to test?

Thanks

Michal

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 1897 bytes --]

Hello,

On 30.6.2016 13:13, Michal Suchanek wrote:
> Hello,
> 
> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>
>> Use Xulong Orange Pi One GPIO based regulator for
>> passive cooling and thermal management.
>>
>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>> ---
>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>  1 file changed, 39 insertions(+)
>>
>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> index b1bd6b0..a38d871 100644
>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> @@ -109,6 +109,45 @@
>>         };
>>  };
>>
>> +&cpu0 {
>> +       operating-points = <
>> +               /* kHz    uV */
>> +               1296000 1300000
>> +               1200000 1300000
> 
> First problem is that the board boots at 1008000 which is not listed
> and the kernel complains.
> 
> Second problem is that the board locks up during boot with this enabled.
> 
> Do you have some suggestion for alternate configuration to test?

Just to verify, did you test with the entire series applied? (especially
the PLL1 clk application changes)

You may try dropping the highest operating point, it's probably overly
optimistic for Orange Pi One.

Is the power supply/cable you're using hard enough?

Where during the boot process does it lock up?

regards,
  Ondrej

> Thanks
> 
> Michal
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Michal Suchanek]

On 30 June 2016 at 16:19, Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
> Hello,
>
> On 30.6.2016 13:13, Michal Suchanek wrote:
>> Hello,
>>
>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>
>>> Use Xulong Orange Pi One GPIO based regulator for
>>> passive cooling and thermal management.
>>>
>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>> ---
>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>  1 file changed, 39 insertions(+)
>>>
>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> index b1bd6b0..a38d871 100644
>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> @@ -109,6 +109,45 @@
>>>         };
>>>  };
>>>
>>> +&cpu0 {
>>> +       operating-points = <
>>> +               /* kHz    uV */
>>> +               1296000 1300000
>>> +               1200000 1300000
>>
>> First problem is that the board boots at 1008000 which is not listed
>> and the kernel complains.
>>
>> Second problem is that the board locks up during boot with this enabled.
>>
>> Do you have some suggestion for alternate configuration to test?
>
> Just to verify, did you test with the entire series applied? (especially
> the PLL1 clk application changes)

Yes, I applied the whole series.

>
> You may try dropping the highest operating point, it's probably overly
> optimistic for Orange Pi One.
>
> Is the power supply/cable you're using hard enough?

I use a 7 port hub to power the board. It worked with some other small
devboards.

The cable is some random Chinese USB to power jack adaptor with an
extra adaptor to fit the Pi socket. It looks ok but I did not test
this particular combination thoroughly.

>
> Where during the boot process does it lock up?

Usually sometime around enabling cpufreq  before getty starts.
Different runs and settings give slightly different results. In
particular adding the 1008000 point seems to make it go further.

Removing all traces of the regulator, cpufreq and thermal I can boot
pretty much 100% to login prompt.

I don't think the difference between 1GHz and 1.3GHz frequency on the
core would put much additional stress on the supply but I can try with
35W PSU and some alternative cabling to be sure.

I did some more tests and it seems 1008000@1.1V is fine but switching
to 1.3V crashes the board. Even with only the first 1.3V state. Maybe
there is need for some delay somewhere for the regulator to get
stable?

Thanks

Michal

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 3456 bytes --]

On 30.6.2016 17:16, Michal Suchanek wrote:
> On 30 June 2016 at 16:19, Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>> Hello,
>>
>> On 30.6.2016 13:13, Michal Suchanek wrote:
>>> Hello,
>>>
>>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>
>>>> Use Xulong Orange Pi One GPIO based regulator for
>>>> passive cooling and thermal management.
>>>>
>>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>> ---
>>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>>  1 file changed, 39 insertions(+)
>>>>
>>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> index b1bd6b0..a38d871 100644
>>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> @@ -109,6 +109,45 @@
>>>>         };
>>>>  };
>>>>
>>>> +&cpu0 {
>>>> +       operating-points = <
>>>> +               /* kHz    uV */
>>>> +               1296000 1300000
>>>> +               1200000 1300000
>>>
>>> First problem is that the board boots at 1008000 which is not listed
>>> and the kernel complains.
>>>
>>> Second problem is that the board locks up during boot with this enabled.
>>>
>>> Do you have some suggestion for alternate configuration to test?
>>
>> Just to verify, did you test with the entire series applied? (especially
>> the PLL1 clk application changes)
> 
> Yes, I applied the whole series.
> 
>>
>> You may try dropping the highest operating point, it's probably overly
>> optimistic for Orange Pi One.
>>
>> Is the power supply/cable you're using hard enough?
> 
> I use a 7 port hub to power the board. It worked with some other small
> devboards.
> 
> The cable is some random Chinese USB to power jack adaptor with an
> extra adaptor to fit the Pi socket. It looks ok but I did not test
> this particular combination thoroughly.
> 
>>
>> Where during the boot process does it lock up?
> 
> Usually sometime around enabling cpufreq  before getty starts.
> Different runs and settings give slightly different results. In
> particular adding the 1008000 point seems to make it go further.
> 
> Removing all traces of the regulator, cpufreq and thermal I can boot
> pretty much 100% to login prompt.
> 
> I don't think the difference between 1GHz and 1.3GHz frequency on the
> core would put much additional stress on the supply but I can try with
> 35W PSU and some alternative cabling to be sure.
> 
> I did some more tests and it seems 1008000@1.1V is fine but switching
> to 1.3V crashes the board. Even with only the first 1.3V state. Maybe
> there is need for some delay somewhere for the regulator to get
> stable?

Please try adding:

  regulator-ramp-delay = <10>

to the gpio regulator.

which will allow 100ms per 1V change, which should b enough for
determining, if this is the cause.

regards,
  Ondrej

> Thanks
> 
> Michal
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Michal Suchanek]

On 30 June 2016 at 17:16, Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> On 30 June 2016 at 16:19, Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>> Hello,
>>
>> On 30.6.2016 13:13, Michal Suchanek wrote:
>>> Hello,
>>>
>>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>
>>>> Use Xulong Orange Pi One GPIO based regulator for
>>>> passive cooling and thermal management.
>>>>
>>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>> ---
>>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>>  1 file changed, 39 insertions(+)
>>>>
>>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> index b1bd6b0..a38d871 100644
>>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>> @@ -109,6 +109,45 @@
>>>>         };
>>>>  };
>>>>
>>>> +&cpu0 {
>>>> +       operating-points = <
>>>> +               /* kHz    uV */
>>>> +               1296000 1300000
>>>> +               1200000 1300000
>>>
>>> First problem is that the board boots at 1008000 which is not listed
>>> and the kernel complains.
>>>
>>> Second problem is that the board locks up during boot with this enabled.
>>>
>>> Do you have some suggestion for alternate configuration to test?
>>
>> Just to verify, did you test with the entire series applied? (especially
>> the PLL1 clk application changes)
>
> Yes, I applied the whole series.
>
>>
>> You may try dropping the highest operating point, it's probably overly
>> optimistic for Orange Pi One.
>>
>> Is the power supply/cable you're using hard enough?
>
> I use a 7 port hub to power the board. It worked with some other small
> devboards.
>
> The cable is some random Chinese USB to power jack adaptor with an
> extra adaptor to fit the Pi socket. It looks ok but I did not test
> this particular combination thoroughly.
>
>>
>> Where during the boot process does it lock up?
>
> Usually sometime around enabling cpufreq  before getty starts.
> Different runs and settings give slightly different results. In
> particular adding the 1008000 point seems to make it go further.
>
> Removing all traces of the regulator, cpufreq and thermal I can boot
> pretty much 100% to login prompt.
>
> I don't think the difference between 1GHz and 1.3GHz frequency on the
> core would put much additional stress on the supply but I can try with
> 35W PSU and some alternative cabling to be sure.
>
> I did some more tests and it seems 1008000@1.1V is fine but switching
> to 1.3V crashes the board. Even with only the first 1.3V state. Maybe
> there is need for some delay somewhere for the regulator to get
> stable?
>

Hm, the AW table shows 1008000@1.3V as supported state and it indeed
works. So the voltage switching works or does nothing. Can I measure
the regulator output somewhere? Clocking the chip higher does not
work.

I will try with another PSU later.

Thanks

Michal

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 3795 bytes --]

On 30.6.2016 17:50, Michal Suchanek wrote:
> On 30 June 2016 at 17:16, Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
>> On 30 June 2016 at 16:19, Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>> Hello,
>>>
>>> On 30.6.2016 13:13, Michal Suchanek wrote:
>>>> Hello,
>>>>
>>>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>>
>>>>> Use Xulong Orange Pi One GPIO based regulator for
>>>>> passive cooling and thermal management.
>>>>>
>>>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>> ---
>>>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>>>  1 file changed, 39 insertions(+)
>>>>>
>>>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> index b1bd6b0..a38d871 100644
>>>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> @@ -109,6 +109,45 @@
>>>>>         };
>>>>>  };
>>>>>
>>>>> +&cpu0 {
>>>>> +       operating-points = <
>>>>> +               /* kHz    uV */
>>>>> +               1296000 1300000
>>>>> +               1200000 1300000
>>>>
>>>> First problem is that the board boots at 1008000 which is not listed
>>>> and the kernel complains.
>>>>
>>>> Second problem is that the board locks up during boot with this enabled.
>>>>
>>>> Do you have some suggestion for alternate configuration to test?
>>>
>>> Just to verify, did you test with the entire series applied? (especially
>>> the PLL1 clk application changes)
>>
>> Yes, I applied the whole series.
>>
>>>
>>> You may try dropping the highest operating point, it's probably overly
>>> optimistic for Orange Pi One.
>>>
>>> Is the power supply/cable you're using hard enough?
>>
>> I use a 7 port hub to power the board. It worked with some other small
>> devboards.
>>
>> The cable is some random Chinese USB to power jack adaptor with an
>> extra adaptor to fit the Pi socket. It looks ok but I did not test
>> this particular combination thoroughly.
>>
>>>
>>> Where during the boot process does it lock up?
>>
>> Usually sometime around enabling cpufreq  before getty starts.
>> Different runs and settings give slightly different results. In
>> particular adding the 1008000 point seems to make it go further.
>>
>> Removing all traces of the regulator, cpufreq and thermal I can boot
>> pretty much 100% to login prompt.
>>
>> I don't think the difference between 1GHz and 1.3GHz frequency on the
>> core would put much additional stress on the supply but I can try with
>> 35W PSU and some alternative cabling to be sure.
>>
>> I did some more tests and it seems 1008000@1.1V is fine but switching
>> to 1.3V crashes the board. Even with only the first 1.3V state. Maybe
>> there is need for some delay somewhere for the regulator to get
>> stable?
>>
> 
> Hm, the AW table shows 1008000@1.3V as supported state and it indeed
> works. So the voltage switching works or does nothing. Can I measure
> the regulator output somewhere? Clocking the chip higher does not
> work.

Regulator output is TP1 on the bottom of the board, next ot the CSI
connector.

regards,
  Ondrej

> I will try with another PSU later.
> 
> Thanks
> 
> Michal
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Michal Suchanek]

On 30 June 2016 at 17:50, Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> On 30 June 2016 at 17:16, Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
>> On 30 June 2016 at 16:19, Ondřej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>> Hello,
>>)k, so I tried>
>>> On 30.6.2016 13:13, Michal Suchanek wrote:
>>>> Hello,
>>>>
>>>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>>
>>>>> Use Xulong Orange Pi One GPIO based regulator for
>>>>> passive cooling and thermal management.
>>>>>
>>>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>>> ---
>>>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>>>  1 file changed, 39 insertions(+)
>>>>>
>>>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> index b1bd6b0..a38d871 100644
>>>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>>>> @@ -109,6 +109,45 @@
>>>>>         };
>>>>>  };
>>>>>
>>>>> +&cpu0 {
>>>>> +       operating-points = <
>>>>> +               /* kHz    uV */
>>>>> +               1296000 1300000
>>>>> +               1200000 1300000
>>>>
>>>> First problem is that the board boots at 1008000 which is not listed
>>>> and the kernel complains.
>>>>
>>>> Second problem is that the board locks up during boot with this enabled.
>>>>
>>>> Do you have some suggestion for alternate configuration to test?
>>>
>>> Just to verify, did you test with the entire series applied? (especially
>>> the PLL1 clk application changes)
>>
>> Yes, I applied the whole series.
>>
>>>
>>> You may try dropping the highest operating point, it's probably overly
>>> optimistic for Orange Pi One.
>>>
>>> Is the power supply/cable you're using hard enough?
>>
>> I use a 7 port hub to power the board. It worked with some other small
>> devboards.
>>
>> The cable is some random Chinese USB to power jack adaptor with an
>> extra adaptor to fit the Pi socket. It looks ok but I did not test
>> this particular combination thoroughly.
>>
>>>
>>> Where during the boot process does it lock up?
>>
>> Usually sometime around enabling cpufreq  before getty starts.
>> Different runs and settings give slightly different results. In
>> particular adding the 1008000 point seems to make it go further.
>>
>> Removing all traces of the regulator, cpufreq and thermal I can boot
>> pretty much 100% to login prompt.
>>
>> I don't think the difference between 1GHz and 1.3GHz frequency on the
>> core would put much additional stress on the supply but I can try with
>> 35W PSU and some alternative cabling to be sure.
>>
>> I did some more tests and it seems 1008000@1.1V is fine but switching
>> to 1.3V crashes the board. Even with only the first 1.3V state. Maybe
>> there is need for some delay somewhere for the regulator to get
>> stable?
>>
>
> Hm, the AW table shows 1008000@1.3V as supported state and it indeed
> works. So the voltage switching works or does nothing. Can I measure
> the regulator output somewhere? Clocking the chip higher does not
> work.
>
> I will try with another PSU later.
>

Ok, so I tried. The result is the same.

A Chinese USB power meter shows voltage 5.1V which drops to like 4.95V
when the board is running. The power draw is around 170mA and
about 200mA when switch to 1.3V is attempted.
The voltage drop is not nice and is probably due to excessive cabling used
to distribute power to multiple boards.
The USB hub starts at 5.2V and drops to something like 5.1V.

When the top frequency is 1008000@1.3V and governor is performance
the board keeps running 1008000@1.1V as set up by u-boot.

Changing the top point to 1008000@1.1V the board locks up as soon as
governor is changed to conservative. So any attempt at frequency scaling
crashes even without touching the regulator.

Thanks

Michal

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Siarhei Siamashka]

On Thu, 30 Jun 2016 13:13:48 +0200
Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:

> Hello,
> 
> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
> > From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> >
> > Use Xulong Orange Pi One GPIO based regulator for
> > passive cooling and thermal management.
> >
> > Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
> > ---
> >  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
> >  1 file changed, 39 insertions(+)
> >
> > diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> > index b1bd6b0..a38d871 100644
> > --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> > +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
> > @@ -109,6 +109,45 @@
> >         };
> >  };
> >
> > +&cpu0 {
> > +       operating-points = <
> > +               /* kHz    uV */
> > +               1296000 1300000
> > +               1200000 1300000  
> 
> First problem is that the board boots at 1008000 which is not listed
> and the kernel complains.
> 
> Second problem is that the board locks up during boot with this enabled.
> 
> Do you have some suggestion for alternate configuration to test?

Maybe try the Allwinner's original DVFS table instead of these
undervolted values and see if it helps?

https://linux-sunxi.org/index.php?title=Xunlong_Orange_Pi_PC&oldid=17753#CPU_clock_speed_limit

While undervolting is tempting because it helps to decrease the SoC
temperature and avoid throttling, different units may have different
tolerances and one needs to be very careful when picking defaults
that are intended to work correctly on all boards. Some safety
headroom exists there for a reason.

If I remember correctly, some people pushed for undervolting experiments
at least twice in the past (on the Banana Pi and C.H.I.P.). In both
cases this did not end up well and had to be fixed later to solve
reliability problems.

In order to allow individual per-unit tuning, a concept of "speed
grading" may be probably introduced later. So that the board is tested
for reliability and then the speed grade rating is stored somewhere on
the non-removable storage (EEPROM, SPI flash, eFUSE, ...). Some SoC
manufacturers, such as Samsung, are already doing this with their chips.

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 3365 bytes --]


On 30.6.2016 16:23, Siarhei Siamashka wrote:
> On Thu, 30 Jun 2016 13:13:48 +0200
> Michal Suchanek <hramrach-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
> 
>> Hello,
>>
>> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>>
>>> Use Xulong Orange Pi One GPIO based regulator for
>>> passive cooling and thermal management.
>>>
>>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>> ---
>>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>>  1 file changed, 39 insertions(+)
>>>
>>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> index b1bd6b0..a38d871 100644
>>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>>> @@ -109,6 +109,45 @@
>>>         };
>>>  };
>>>
>>> +&cpu0 {
>>> +       operating-points = <
>>> +               /* kHz    uV */
>>> +               1296000 1300000
>>> +               1200000 1300000  
>>
>> First problem is that the board boots at 1008000 which is not listed
>> and the kernel complains.
>>
>> Second problem is that the board locks up during boot with this enabled.
>>
>> Do you have some suggestion for alternate configuration to test?
> 
> Maybe try the Allwinner's original DVFS table instead of these
> undervolted values and see if it helps?
> 
> https://linux-sunxi.org/index.php?title=Xunlong_Orange_Pi_PC&oldid=17753#CPU_clock_speed_limit

Thanks, I'll use these. I'm not sure what will happen if regulator is
not capable of providing the exact voltages specified in the operating
points, though. Will it do the sane thing? :)

Also LV6+ in the table is bellow specified minimum voltage in the
datasheet. But then Orange Pi works at much lower voltages for me.

Conversly LV1 is the absolute maximum. Recommended maximum is 1.4V So I
would not use that.

It might be nice to have something like that dram clock test, but for
cpux freq/voltage, and collect some statistics.

regards,
  Ondrej

> While undervolting is tempting because it helps to decrease the SoC
> temperature and avoid throttling, different units may have different
> tolerances and one needs to be very careful when picking defaults
> that are intended to work correctly on all boards. Some safety
> headroom exists there for a reason.
> 
> If I remember correctly, some people pushed for undervolting experiments
> at least twice in the past (on the Banana Pi and C.H.I.P.). In both
> cases this did not end up well and had to be fixed later to solve
> reliability problems.
> 
> In order to allow individual per-unit tuning, a concept of "speed
> grading" may be probably introduced later. So that the board is tested
> for reliability and then the speed grade rating is stored somewhere on
> the non-removable storage (EEPROM, SPI flash, eFUSE, ...). Some SoC
> manufacturers, such as Samsung, are already doing this with their chips.
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: [PATCH v2 14/14] ARM: dts: sun8i: Enable DVFS on Orange Pi One

[Ondřej Jirman]

[-- Attachment #1.1: Type: text/plain, Size: 1924 bytes --]

Hello Michal,

On 30.6.2016 13:13, Michal Suchanek wrote:
> Hello,
> 
> On 25 June 2016 at 05:45,  <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org> wrote:
>> From: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>>
>> Use Xulong Orange Pi One GPIO based regulator for
>> passive cooling and thermal management.
>>
>> Signed-off-by: Ondrej Jirman <megous-5qf/QAjKc83QT0dZR+AlfA@public.gmane.org>
>> ---
>>  arch/arm/boot/dts/sun8i-h3-orangepi-one.dts | 39 +++++++++++++++++++++++++++++
>>  1 file changed, 39 insertions(+)
>>
>> diff --git a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> index b1bd6b0..a38d871 100644
>> --- a/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> +++ b/arch/arm/boot/dts/sun8i-h3-orangepi-one.dts
>> @@ -109,6 +109,45 @@
>>         };
>>  };
>>
>> +&cpu0 {
>> +       operating-points = <
>> +               /* kHz    uV */
>> +               1296000 1300000
>> +               1200000 1300000
> 
> First problem is that the board boots at 1008000 which is not listed
> and the kernel complains.
> 
> Second problem is that the board locks up during boot with this enabled.
> 
> Do you have some suggestion for alternate configuration to test?

I've identified the problem as incorrectly set up gpio-regulator. During
boot, when it was probed, it would be initiualized with 1.1V for a
moment. This caused random non-specific lockups/crashes afterwards.

Fixed patches are in my branch on github, if you want to try again.

  https://github.com/megous/linux/commits/orange-pi-4.7

regards,
  o.

> Thanks
> 
> Michal
> 

-- 
You received this message because you are subscribed to the Google Groups "linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/Ez6ZCGd0@public.gmane.org
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]