[PATCH 0/3] arm: tegra: implement NVEC driver using tegra i2c.

Hi,

NVEC driver contains code to manage tegra i2c controller in slave mode.
I2C slave support was implemented in linux kernel. The goal of this
patch serie is to implement I2C slave mode in tegra drived and rework
NVEC driver to use it.

Patches are based on i2c for-next.

Patch 1 imeplents slave mode for tegra I2C controller. This patch
was checked on tegra 2 device (Toshiba AC100) only. Please review
carefully.

Patch 2 reworks NVEC driver itself. I kept code close to original.

Patch 3 fixes device tree and documentation.


Thanks in advance

Andrey Danin (3):
  i2c: tegra: implement slave mode
  staging/nvec: reimplement on top of tegra i2c driver
  dt: paz00: define nvec as child of i2c bus

 .../devicetree/bindings/nvec/nvidia,nvec.txt       |  19 +-
 arch/arm/boot/dts/tegra20-paz00.dts                |  22 +-
 drivers/i2c/busses/i2c-tegra.c                     | 131 +++++++
 drivers/staging/nvec/nvec.c                        | 379 +++++++--------------
 drivers/staging/nvec/nvec.h                        |  17 +-
 5 files changed, 264 insertions(+), 304 deletions(-)

-- 
1.9.1

NVEC driver was reimplemented to use tegra i2c. Use common i2c bindings
for NVEC node.

Signed-off-by: Andrey Danin <redacted>
---
 .../devicetree/bindings/nvec/nvidia,nvec.txt       | 19 ++-----------------
 arch/arm/boot/dts/tegra20-paz00.dts                | 22 +++++++++-------------
 2 files changed, 11 insertions(+), 30 deletions(-)

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt
index 5ae601e..d82c125 100644
--- a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt
+++ b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller
 
 Required properties:
 - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts b/arch/arm/boot/dts/tegra20-paz00.dts
index ed7e100..65e247b 100644
--- a/arch/arm/boot/dts/tegra20-paz00.dts
+++ b/arch/arm/boot/dts/tegra20-paz00.dts

@@ -288,20 +288,16 @@
 		clock-frequency = <100000>;
 	};
 
-	nvec at 7000c500 {
-		compatible = "nvidia,nvec";
-		reg = <0x7000c500 0x100>;
-		interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-		#address-cells = <1>;
-		#size-cells = <0>;
+	i2c at 7000c500 {
+		status = "okay";
 		clock-frequency = <80000>;
-		request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-		slave-addr = <138>;
-		clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-		         <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-		clock-names = "div-clk", "fast-clk";
-		resets = <&tegra_car 67>;
-		reset-names = "i2c";
+
+		nvec: nvec at 45 {
+			compatible = "nvidia,nvec";
+			request-gpios = <&gpio TEGRA_GPIO(V, 2)
+				GPIO_ACTIVE_HIGH>;
+			reg = <0x45>;
+		};
 	};
 
 	i2c at 7000d000 {

-- 
1.9.1

Am Donnerstag, 29. Januar 2015, 10:20:22 schrieb Andrey Danin:

based on the earlier discussion with Wolfram, this should be "nvidia,nvec-
slave" to distunguish it from a possible nvec master driver.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150129/8e419aec/attachment-0001.sig>

On 01/29/2015 12:20 AM, Andrey Danin wrote:

NVEC driver was reimplemented to use tegra i2c. Use common i2c bindings
for NVEC node.

The changes to this file make more sense either as a standalone patch 
1/4, or as part of the driver changes.

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I 
think you should introduce a new compatible value to describe the 
external NVEC chip.

-	nvec at 7000c500 {
-		compatible = "nvidia,nvec";
-		reg = <0x7000c500 0x100>;
-		interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-		#address-cells = <1>;
-		#size-cells = <0>;
+	i2c at 7000c500 {
+		status = "okay";
  		clock-frequency = <80000>;
-		request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-		slave-addr = <138>;
-		clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-		         <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-		clock-names = "div-clk", "fast-clk";
-		resets = <&tegra_car 67>;
-		reset-names = "i2c";
+
+		nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this 
child device is a slave that is implemented by the host SoC rather than 
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec 
only calls I2C APIs suitable for internal slaves rather than external 
slaves? Even so though, I think the distinction needs to be clearly 
marked in the DT so that any generic code outside the NVEC driver that 
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with cell 
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C driver 
would need to support #address-cells=<1> for backwards-compatibility.

+			compatible = "nvidia,nvec";
+			request-gpios = <&gpio TEGRA_GPIO(V, 2)
+				GPIO_ACTIVE_HIGH>;
+			reg = <0x45>;

The order is typically compatible, reg, other properties.

+		};
  	};

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

On 01/29/2015 12:20 AM, Andrey Danin wrote:

quoted

NVEC driver was reimplemented to use tegra i2c. Use common i2c bindings
for NVEC node.

quoted

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

The changes to this file make more sense either as a standalone patch
1/4, or as part of the driver changes.

quoted

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller

  Required properties:
  - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I
think you should introduce a new compatible value to describe the
external NVEC chip.

I changed compatible value to nvec-slave in v2.

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

quoted

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";
          clock-frequency = <80000>;
-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C driver
would need to support #address-cells=<1> for backwards-compatibility.

Driver (nvec in this case) can decide what mode should it use according 
to compatible value. Is it not enough ?

quoted

+            compatible = "nvidia,nvec";
+            request-gpios = <&gpio TEGRA_GPIO(V, 2)
+                GPIO_ACTIVE_HIGH>;
+            reg = <0x45>;

The order is typically compatible, reg, other properties.

Ok, thanks.

quoted

+        };
      };

On 03/31/2015 12:40 AM, Andrey Danin wrote:

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

quoted

On 01/29/2015 12:20 AM, Andrey Danin wrote:

quoted

NVEC driver was reimplemented to use tegra i2c. Use common i2c bindings
for NVEC node.

quoted

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

The changes to this file make more sense either as a standalone patch
1/4, or as part of the driver changes.

quoted

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller

  Required properties:
  - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I
think you should introduce a new compatible value to describe the
external NVEC chip.

I changed compatible value to nvec-slave in v2.

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

quoted

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";
          clock-frequency = <80000>;
-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C driver
would need to support #address-cells=<1> for backwards-compatibility.

Driver (nvec in this case) can decide what mode should it use according
to compatible value. Is it not enough ?

No, I don't think so.

The I2C binding model is that each child of an I2C controller represents 
a device attached to the bus. which SW will communicate with using the 
I2C controller as master and the device as a slave. If there's no 
explicit representation of child-vs-slave in the DT, how does the I2C 
core know whether a particular node is intended to be accessed as a 
master or slave?

In other words, without an explicit "communicate with this device" or 
"implement this device as a slave" flag, how could DT contain:

i2c-controller {
     ...
     master at 1a {
         compatible = "foo,device";
         reg = <0x1a 1>;
     };
     slave at 1a {
         compatible = "foo,device-slave";
         reg = <0x1a 1>;
     };
};

where:

- "foo,device" means: instantiate a driver to communicate with a device 
of this type.

- "foo,device-slave" means: instantiate a driver to act as this I2C device.

Sure it's possible for the drivers for those two nodes to simply use the 
I2C subsystem's master or slave APIs, but I suspect DT content would 
confuse the I2C core into thinking that two I2C devices with the same 
address had been represented in DT, and the I2C core would refuse to 
instantiate one of them. The solution here is for the reg value to 
encode a "master" vs. "slave" flag, so the I2C core can allow both a 
master and a slave for each address.

I'm pretty sure this is the nth time I've explained this.

On 31.03.2015 17:09, Stephen Warren wrote:

On 03/31/2015 12:40 AM, Andrey Danin wrote:

quoted

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

quoted

On 01/29/2015 12:20 AM, Andrey Danin wrote:

quoted

NVEC driver was reimplemented to use tegra i2c. Use common i2c bindings
for NVEC node.

quoted

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

The changes to this file make more sense either as a standalone patch
1/4, or as part of the driver changes.

quoted

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller

  Required properties:
  - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I
think you should introduce a new compatible value to describe the
external NVEC chip.

I changed compatible value to nvec-slave in v2.

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

quoted

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";
          clock-frequency = <80000>;
-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C driver
would need to support #address-cells=<1> for backwards-compatibility.

Driver (nvec in this case) can decide what mode should it use according
to compatible value. Is it not enough ?

No, I don't think so.

The I2C binding model is that each child of an I2C controller represents
a device attached to the bus. which SW will communicate with using the
I2C controller as master and the device as a slave. If there's no
explicit representation of child-vs-slave in the DT, how does the I2C
core know whether a particular node is intended to be accessed as a
master or slave?

Device driver registers itself via slave API. Bus driver calls 
appropriate callback function when needed.
If device driver decides to access hardware via master API, then it can 
do it.

Am I missing something ?

In other words, without an explicit "communicate with this device" or
"implement this device as a slave" flag, how could DT contain:

i2c-controller {
     ...
     master at 1a {
         compatible = "foo,device";
         reg = <0x1a 1>;
     };
     slave at 1a {
         compatible = "foo,device-slave";
         reg = <0x1a 1>;
     };
};

where:

- "foo,device" means: instantiate a driver to communicate with a device
of this type.

- "foo,device-slave" means: instantiate a driver to act as this I2C device.

Sure it's possible for the drivers for those two nodes to simply use the
I2C subsystem's master or slave APIs, but I suspect DT content would
confuse the I2C core into thinking that two I2C devices with the same
address had been represented in DT, and the I2C core would refuse to
instantiate one of them. The solution here is for the reg value to
encode a "master" vs. "slave" flag, so the I2C core can allow both a
master and a slave for each address.

If there is one device, then it must be one node. If there is two 
devices then it looks incorrect to me to have two devices with the same 
address. Does I2C allow two devices with same address ?

I can imagine this:
- we have hardware with I2C device. This device can act as master or as 
slave
- we have device driver, that can work in one, other or both modes.

If we want to force master or slave mode, we can use flags (for combined 
mode we can use two nodes, but it looks weird).
If we want to let driver decide (preferred mode, arbitration, something 
else), we can use current rules.

I'm pretty sure this is the nth time I've explained this.

Sorry. I don't understand why you still suggest to use flags. We can use 
existing infrastructure in this case. There is already similar case in 
arch/arm/boot/dts/r8a7790-lager.dts (see i2c1 and eeprom).

Do we *really* need this extra rules at this moment ?

Added Wolfram Sang and linux-i2c ML

On 31.03.2015 18:46, Andrey Danin wrote:

On 31.03.2015 17:09, Stephen Warren wrote:

quoted

On 03/31/2015 12:40 AM, Andrey Danin wrote:

quoted

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

quoted

On 01/29/2015 12:20 AM, Andrey Danin wrote:

quoted

NVEC driver was reimplemented to use tegra i2c. Use common i2c
bindings
for NVEC node.

quoted

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

The changes to this file make more sense either as a standalone patch
1/4, or as part of the driver changes.

quoted

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller

  Required properties:
  - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I
think you should introduce a new compatible value to describe the
external NVEC chip.

I changed compatible value to nvec-slave in v2.

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

quoted

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";
          clock-frequency = <80000>;
-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with
cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C
driver
would need to support #address-cells=<1> for backwards-compatibility.

Driver (nvec in this case) can decide what mode should it use according
to compatible value. Is it not enough ?

No, I don't think so.

The I2C binding model is that each child of an I2C controller represents
a device attached to the bus. which SW will communicate with using the
I2C controller as master and the device as a slave. If there's no
explicit representation of child-vs-slave in the DT, how does the I2C
core know whether a particular node is intended to be accessed as a
master or slave?

Device driver registers itself via slave API. Bus driver calls
appropriate callback function when needed.
If device driver decides to access hardware via master API, then it can
do it.

Am I missing something ?

quoted

In other words, without an explicit "communicate with this device" or
"implement this device as a slave" flag, how could DT contain:

i2c-controller {
     ...
     master at 1a {
         compatible = "foo,device";
         reg = <0x1a 1>;
     };
     slave at 1a {
         compatible = "foo,device-slave";
         reg = <0x1a 1>;
     };
};

where:

- "foo,device" means: instantiate a driver to communicate with a device
of this type.

- "foo,device-slave" means: instantiate a driver to act as this I2C
device.

Sure it's possible for the drivers for those two nodes to simply use the
I2C subsystem's master or slave APIs, but I suspect DT content would
confuse the I2C core into thinking that two I2C devices with the same
address had been represented in DT, and the I2C core would refuse to
instantiate one of them. The solution here is for the reg value to
encode a "master" vs. "slave" flag, so the I2C core can allow both a
master and a slave for each address.

If there is one device, then it must be one node. If there is two
devices then it looks incorrect to me to have two devices with the same
address. Does I2C allow two devices with same address ?

I can imagine this:
- we have hardware with I2C device. This device can act as master or as
slave
- we have device driver, that can work in one, other or both modes.

If we want to force master or slave mode, we can use flags (for combined
mode we can use two nodes, but it looks weird).
If we want to let driver decide (preferred mode, arbitration, something
else), we can use current rules.

quoted

I'm pretty sure this is the nth time I've explained this.

Sorry. I don't understand why you still suggest to use flags. We can use
existing infrastructure in this case. There is already similar case in
arch/arm/boot/dts/r8a7790-lager.dts (see i2c1 and eeprom).

Do we *really* need this extra rules at this moment ?

On 03/31/2015 09:46 AM, Andrey Danin wrote:

On 31.03.2015 17:09, Stephen Warren wrote:

quoted

On 03/31/2015 12:40 AM, Andrey Danin wrote:

quoted

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

quoted

On 01/29/2015 12:20 AM, Andrey Danin wrote:

quoted

NVEC driver was reimplemented to use tegra i2c. Use common i2c
bindings
for NVEC node.

quoted

diff --git a/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

b/Documentation/devicetree/bindings/nvec/nvidia,nvec.txt

The changes to this file make more sense either as a standalone patch
1/4, or as part of the driver changes.

quoted

@@ -2,20 +2,5 @@ NVIDIA compliant embedded controller

  Required properties:
  - compatible : should be "nvidia,nvec".
-- reg : the iomem of the i2c slave controller
-- interrupts : the interrupt line of the i2c slave controller
-- clock-frequency : the frequency of the i2c bus
-- gpios : the gpio used for ec request
-- slave-addr: the i2c address of the slave controller
-- clocks : Must contain an entry for each entry in clock-names.
-  See ../clocks/clock-bindings.txt for details.
-- clock-names : Must include the following entries:
-  Tegra20/Tegra30:
-  - div-clk
-  - fast-clk
-  Tegra114:
-  - div-clk
-- resets : Must contain an entry for each entry in reset-names.
-  See ../reset/reset.txt for details.
-- reset-names : Must include the following entries:
-  - i2c
+- request-gpios : the gpio used for ec request
+- reg: the i2c address of the slave controller

This change breaks ABI.

Instead of modifying the definition of the existing compatible value, I
think you should introduce a new compatible value to describe the
external NVEC chip.

I changed compatible value to nvec-slave in v2.

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

quoted

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";
          clock-frequency = <80000>;
-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with
cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C
driver
would need to support #address-cells=<1> for backwards-compatibility.

Driver (nvec in this case) can decide what mode should it use according
to compatible value. Is it not enough ?

No, I don't think so.

The I2C binding model is that each child of an I2C controller represents
a device attached to the bus. which SW will communicate with using the
I2C controller as master and the device as a slave. If there's no
explicit representation of child-vs-slave in the DT, how does the I2C
core know whether a particular node is intended to be accessed as a
master or slave?

Device driver registers itself via slave API. Bus driver calls
appropriate callback function when needed.
If device driver decides to access hardware via master API, then it can
do it.

Am I missing something ?

quoted

In other words, without an explicit "communicate with this device" or
"implement this device as a slave" flag, how could DT contain:

i2c-controller {
     ...
     master at 1a {
         compatible = "foo,device";
         reg = <0x1a 1>;
     };
     slave at 1a {
         compatible = "foo,device-slave";
         reg = <0x1a 1>;
     };
};

where:

- "foo,device" means: instantiate a driver to communicate with a device
of this type.

- "foo,device-slave" means: instantiate a driver to act as this I2C
device.

Sure it's possible for the drivers for those two nodes to simply use the
I2C subsystem's master or slave APIs, but I suspect DT content would
confuse the I2C core into thinking that two I2C devices with the same
address had been represented in DT, and the I2C core would refuse to
instantiate one of them. The solution here is for the reg value to
encode a "master" vs. "slave" flag, so the I2C core can allow both a
master and a slave for each address.

If there is one device, then it must be one node. If there is two
devices then it looks incorrect to me to have two devices with the same
address. Does I2C allow two devices with same address ?

One of the nodes is to indicate that the kernel should implement the 
slave mode device and one is to indicate that the kernel should 
implement the master mode device. Those two devices/nodes have 
completely different semantics, so while they share the I2C bus address 
they don't represent the same thing.

Admittedly it would be uncommon to do this, since it'd be using the I2C 
bus in loopback mode. However, I don't see why we should set out to 
prevent that.

I can imagine this:
- we have hardware with I2C device. This device can act as master or as
slave
- we have device driver, that can work in one, other or both modes.

If we want to force master or slave mode, we can use flags (for combined
mode we can use two nodes, but it looks weird).
If we want to let driver decide (preferred mode, arbitration, something
else), we can use current rules.

quoted

I'm pretty sure this is the nth time I've explained this.

Sorry. I don't understand why you still suggest to use flags. We can use
existing infrastructure in this case. There is already similar case in
arch/arm/boot/dts/r8a7790-lager.dts (see i2c1 and eeprom).

Do we *really* need this extra rules at this moment ?

Am Mittwoch, 1. April 2015, 11:28:32 schrieb Stephen Warren:

On 03/31/2015 09:46 AM, Andrey Danin wrote:

quoted

On 31.03.2015 17:09, Stephen Warren wrote:

quoted

On 03/31/2015 12:40 AM, Andrey Danin wrote:

quoted

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

[ snipped old patch parts ]

quoted

quoted

quoted

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";

          clock-frequency = <80000>;

-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with
cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C
driver
would need to support #address-cells=<1> for backwards-compatibility.

Stephen, we haven't used your suggestion because Wolfram disliked the idea in 
e.g. http://lkml.iu.edu/hypermail/linux/kernel/1409.1/03446.html

quoted

quoted

quoted

Driver (nvec in this case) can decide what mode should it use according
to compatible value. Is it not enough ?

No, I don't think so.

The I2C binding model is that each child of an I2C controller represents
a device attached to the bus. which SW will communicate with using the
I2C controller as master and the device as a slave. If there's no
explicit representation of child-vs-slave in the DT, how does the I2C
core know whether a particular node is intended to be accessed as a
master or slave?

Device driver registers itself via slave API. Bus driver calls
appropriate callback function when needed.
If device driver decides to access hardware via master API, then it can
do it.

Am I missing something ?

quoted

In other words, without an explicit "communicate with this device" or
"implement this device as a slave" flag, how could DT contain:

i2c-controller {

     ...
     master at 1a {
     
         compatible = "foo,device";
         reg = <0x1a 1>;
     
     };
     slave at 1a {
     
         compatible = "foo,device-slave";
         reg = <0x1a 1>;
     
     };

};

where:

- "foo,device" means: instantiate a driver to communicate with a device
of this type.

- "foo,device-slave" means: instantiate a driver to act as this I2C
device.

Sure it's possible for the drivers for those two nodes to simply use the
I2C subsystem's master or slave APIs, but I suspect DT content would
confuse the I2C core into thinking that two I2C devices with the same
address had been represented in DT, and the I2C core would refuse to
instantiate one of them. The solution here is for the reg value to
encode a "master" vs. "slave" flag, so the I2C core can allow both a
master and a slave for each address.

If there is one device, then it must be one node. If there is two
devices then it looks incorrect to me to have two devices with the same
address. Does I2C allow two devices with same address ?

One of the nodes is to indicate that the kernel should implement the
slave mode device and one is to indicate that the kernel should
implement the master mode device. Those two devices/nodes have
completely different semantics, so while they share the I2C bus address
they don't represent the same thing.

Admittedly it would be uncommon to do this, since it'd be using the I2C
bus in loopback mode. However, I don't see why we should set out to
prevent that.

We are sitting between the chairs currently. I hope Wolfram can further 
comment on this.

Having a generic loopback slave driver which just echos all messages it 
received back to the master (on the same controller or a different one) would 
be nice IMHO. 

quoted

I can imagine this:
- we have hardware with I2C device. This device can act as master or as
slave
- we have device driver, that can work in one, other or both modes.

If we want to force master or slave mode, we can use flags (for combined
mode we can use two nodes, but it looks weird).
If we want to let driver decide (preferred mode, arbitration, something
else), we can use current rules.

quoted

I'm pretty sure this is the nth time I've explained this.

Sorry. I don't understand why you still suggest to use flags. We can use
existing infrastructure in this case. There is already similar case in
arch/arm/boot/dts/r8a7790-lager.dts (see i2c1 and eeprom).

Do we *really* need this extra rules at this moment ?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150402/a61f8d04/attachment.sig>

On 04/02/2015 03:37 AM, Marc Dietrich wrote:

Am Mittwoch, 1. April 2015, 11:28:32 schrieb Stephen Warren:

quoted

On 03/31/2015 09:46 AM, Andrey Danin wrote:

quoted

On 31.03.2015 17:09, Stephen Warren wrote:

quoted

On 03/31/2015 12:40 AM, Andrey Danin wrote:

quoted

Hi,

Thanks for the review.

On 03.02.2015 0:20, Stephen Warren wrote:

[ snipped old patch parts ]

quoted

quoted

quoted

quoted

quoted

quoted

diff --git a/arch/arm/boot/dts/tegra20-paz00.dts

b/arch/arm/boot/dts/tegra20-paz00.dts

-    nvec at 7000c500 {
-        compatible = "nvidia,nvec";
-        reg = <0x7000c500 0x100>;
-        interrupts = <GIC_SPI 92 IRQ_TYPE_LEVEL_HIGH>;
-        #address-cells = <1>;
-        #size-cells = <0>;
+    i2c at 7000c500 {
+        status = "okay";

           clock-frequency = <80000>;

-        request-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
-        slave-addr = <138>;
-        clocks = <&tegra_car TEGRA20_CLK_I2C3>,
-                 <&tegra_car TEGRA20_CLK_PLL_P_OUT3>;
-        clock-names = "div-clk", "fast-clk";
-        resets = <&tegra_car 67>;
-        reset-names = "i2c";
+
+        nvec: nvec at 45 {

This doesn't feel correct. There's nothing here to indicate that this
child device is a slave that is implemented by the host SoC rather than
something external attached to the I2C bus.

Perhaps you can get away with this, since the driver for nvidia,nvec
only calls I2C APIs suitable for internal slaves rather than external
slaves? Even so though, I think the distinction needs to be clearly
marked in the DT so that any generic code outside the NVEC driver that
parses the DT can determine the difference.

I would recommend the I2C controller having #address-cells=<2> with
cell
0 being 0==master,1==slave, cell 1 being the I2C address. The I2C
driver
would need to support #address-cells=<1> for backwards-compatibility.

Stephen, we haven't used your suggestion because Wolfram disliked the idea in
e.g. http://lkml.iu.edu/hypermail/linux/kernel/1409.1/03446.html

As you said in the response you linked to, the objection is invalid 
since it won't break any DTs. The driver for a node is responsible for 
defining the meaning of its own reg properties. It should be pretty 
trivial to allow the Tegra I2C controller driver (or indeed any driver 
at all) to handle either #address-cells=<1> (the current setting) or 
#address-cells=<2> (a new value which enables adding a new flag cell) or 
even #address-cells=<1> with some of the upper bits of the reg value 
used as flags (which would default to 0 in all current DTs, so e.g. 
using the MSB==1 as a slave flag), all at run-time with complete 
backwards-compatibility.

Initialization code is based on NVEC driver.

There is a HW bug in AP20 that was also mentioned in kernel sources
for Toshiba AC100.

Signed-off-by: Andrey Danin <redacted>
---
 drivers/i2c/busses/i2c-tegra.c | 131 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 131 insertions(+)

diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
index 28b87e6..cfc4e67 100644
--- a/drivers/i2c/busses/i2c-tegra.c
+++ b/drivers/i2c/busses/i2c-tegra.c

@@ -42,8 +42,15 @@
 #define I2C_SL_CNFG				0x020
 #define I2C_SL_CNFG_NACK			(1<<1)
 #define I2C_SL_CNFG_NEWSL			(1<<2)
+#define I2C_SL_RCVD				0x024
+#define I2C_SL_STATUS				0x028
+#define I2C_SL_ST_IRQ				(1<<3)
+#define I2C_SL_ST_END_TRANS			(1<<4)
+#define I2C_SL_ST_RCVD				(1<<2)
+#define I2C_SL_ST_RNW				(1<<1)
 #define I2C_SL_ADDR1				0x02c
 #define I2C_SL_ADDR2				0x030
+#define I2C_SL_DELAY_COUNT			0x03c
 #define I2C_TX_FIFO				0x050
 #define I2C_RX_FIFO				0x054
 #define I2C_PACKET_TRANSFER_STATUS		0x058

@@ -125,6 +132,8 @@ enum msg_end_type {
  * @clk_divisor_std_fast_mode: Clock divisor in standard/fast mode. It is
  *		applicable if there is no fast clock source i.e. single clock
  *		source.
+ * @slave_read_start_delay: Workaround for AP20 I2C Slave Controller bug. Delay
+ *              before writing data byte into register I2C_SL_RCVD.
  */
 
 struct tegra_i2c_hw_feature {

@@ -133,6 +142,7 @@ struct tegra_i2c_hw_feature {
 	bool has_single_clk_source;
 	int clk_divisor_hs_mode;
 	int clk_divisor_std_fast_mode;
+	int slave_read_start_delay;
 };
 
 /**

@@ -173,6 +183,7 @@ struct tegra_i2c_dev {
 	int msg_read;
 	u32 bus_clk_rate;
 	bool is_suspended;
+	struct i2c_client *slave;
 };
 
 static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val, unsigned long reg)

@@ -398,6 +409,12 @@ static inline int tegra_i2c_clock_enable(struct tegra_i2c_dev *i2c_dev)
 
 static inline void tegra_i2c_clock_disable(struct tegra_i2c_dev *i2c_dev)
 {
+	if (i2c_dev->slave) {
+		dev_warn(i2c_dev->dev,
+			"i2c slave is registered, don't disable a clock\n");
+		return;
+	}
+
 	clk_disable(i2c_dev->div_clk);
 	if (!i2c_dev->hw->has_single_clk_source)
 		clk_disable(i2c_dev->fast_clk);

@@ -459,12 +476,84 @@ static int tegra_i2c_init(struct tegra_i2c_dev *i2c_dev)
 	return err;
 }
 
+static inline int is_ready(unsigned long status)
+{
+	return status & I2C_SL_ST_IRQ;
+}
+
+static inline int is_write(unsigned long status)
+{
+	return (status & I2C_SL_ST_RNW) == 0;
+}
+
+static inline int is_read(unsigned long status)
+{
+	return !is_write(status);
+}
+
+static inline int is_trans_start(unsigned long status)
+{
+	return status & I2C_SL_ST_RCVD;
+}
+
+static inline int is_trans_end(unsigned long status)
+{
+	return status & I2C_SL_ST_END_TRANS;
+}
+
+static bool tegra_i2c_slave_isr(int irq, struct tegra_i2c_dev *i2c_dev)
+{
+	unsigned long status;
+	u8 value;
+
+	if (!i2c_dev->slave || !i2c_dev->slave->slave_cb)
+		return false;
+
+	status = i2c_readl(i2c_dev, I2C_SL_STATUS);
+	if (!is_ready(status))
+		return false;
+
+	/* master sent stop */
+	if (is_trans_end(status)) {
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_STOP, NULL);
+		if (!is_trans_start(status))
+			return true;
+	}
+
+	/* i2c master sends data to us */
+	if (is_write(status)) {
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_WRITE_START,
+				NULL);
+		value = i2c_readl(i2c_dev, I2C_SL_RCVD);
+		if (is_trans_start(status))
+			i2c_writel(i2c_dev, 0, I2C_SL_RCVD);
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_WRITE_END,
+				&value);
+	}
+
+	/* i2c master reads data from us */
+	if (is_read(status)) {
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_READ_START,
+				&value);
+		if (is_trans_start(status)
+				&& i2c_dev->hw->slave_read_start_delay)
+			udelay(i2c_dev->hw->slave_read_start_delay);
+		i2c_writel(i2c_dev, value, I2C_SL_RCVD);
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_READ_END, NULL);
+	}
+
+	return true;
+}
+
 static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
 {
 	u32 status;
 	const u32 status_err = I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST;
 	struct tegra_i2c_dev *i2c_dev = dev_id;
 
+	if (tegra_i2c_slave_isr(irq, i2c_dev))
+		return IRQ_HANDLED;
+
 	status = i2c_readl(i2c_dev, I2C_INT_STATUS);
 
 	if (status == 0) {

@@ -660,9 +749,48 @@ static u32 tegra_i2c_func(struct i2c_adapter *adap)
 	return ret;
 }
 
+static int tegra_reg_slave(struct i2c_client *slave)
+{
+	struct tegra_i2c_dev *i2c_dev = i2c_get_adapdata(slave->adapter);
+
+	if (i2c_dev->slave)
+		return -EBUSY;
+
+	i2c_dev->slave = slave;
+
+	tegra_i2c_clock_enable(i2c_dev);
+
+	reset_control_assert(i2c_dev->rst);
+	udelay(2);
+	reset_control_deassert(i2c_dev->rst);
+
+	i2c_writel(i2c_dev, I2C_SL_CNFG_NEWSL, I2C_SL_CNFG);
+	i2c_writel(i2c_dev, 0x1E, I2C_SL_DELAY_COUNT);
+
+	i2c_writel(i2c_dev, slave->addr, I2C_SL_ADDR1);
+	i2c_writel(i2c_dev, 0, I2C_SL_ADDR2);
+
+	return 0;
+}
+
+static int tegra_unreg_slave(struct i2c_client *slave)
+{
+	struct tegra_i2c_dev *i2c_dev = i2c_get_adapdata(slave->adapter);
+
+	WARN_ON(!i2c_dev->slave);
+
+	i2c_writel(i2c_dev, 0, I2C_SL_CNFG);
+
+	i2c_dev->slave = NULL;
+
+	return 0;
+}
+
 static const struct i2c_algorithm tegra_i2c_algo = {
 	.master_xfer	= tegra_i2c_xfer,
 	.functionality	= tegra_i2c_func,
+	.reg_slave		= tegra_reg_slave,
+	.unreg_slave	= tegra_unreg_slave,
 };
 
 static const struct tegra_i2c_hw_feature tegra20_i2c_hw = {

@@ -671,6 +799,7 @@ static const struct tegra_i2c_hw_feature tegra20_i2c_hw = {
 	.has_single_clk_source = false,
 	.clk_divisor_hs_mode = 3,
 	.clk_divisor_std_fast_mode = 0,
+	.slave_read_start_delay = 8,
 };
 
 static const struct tegra_i2c_hw_feature tegra30_i2c_hw = {

@@ -679,6 +808,7 @@ static const struct tegra_i2c_hw_feature tegra30_i2c_hw = {
 	.has_single_clk_source = false,
 	.clk_divisor_hs_mode = 3,
 	.clk_divisor_std_fast_mode = 0,
+	.slave_read_start_delay = 0,
 };
 
 static const struct tegra_i2c_hw_feature tegra114_i2c_hw = {

@@ -687,6 +817,7 @@ static const struct tegra_i2c_hw_feature tegra114_i2c_hw = {
 	.has_single_clk_source = true,
 	.clk_divisor_hs_mode = 1,
 	.clk_divisor_std_fast_mode = 0x19,
+	.slave_read_start_delay = 0,
 };
 
 /* Match table for of_platform binding */

-- 
1.9.1

Hi Andrey,

first, thanks for accepting the challenge once more ;-)

The driver depends on I2C_SLAVE now, so you need to add this to Kconfig. The 
amount of code (and additional overhead) is pretty small, so I think it's ok 
to always enable it. Otherwise we would need lots of ifdefs.

Am Donnerstag, 29. Januar 2015, 10:20:20 schrieb Andrey Danin:

is this really required? What are the callers of clock_disable? I think it 
would be ok to make master or slave operation exclusive for now. We have no 
way to test this anyway. Maybe some flag which blocks master operation.

is_slave_irq?

+
+static inline int is_write(unsigned long status)
+{
+	return (status & I2C_SL_ST_RNW) == 0;
+}
+
+static inline int is_read(unsigned long status)
+{
+	return !is_write(status);
+}
+
+static inline int is_trans_start(unsigned long status)
+{
+	return status & I2C_SL_ST_RCVD;
+}
+
+static inline int is_trans_end(unsigned long status)
+{
+	return status & I2C_SL_ST_END_TRANS;
+}

Following the rest of the files coding style, I think this helpers can be 
moved to the caller itself. 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150129/ba8ae458/attachment.sig>

Hi,

Initialization code is based on NVEC driver.

There is a HW bug in AP20 that was also mentioned in kernel sources
for Toshiba AC100.

Signed-off-by: Andrey Danin <redacted>

Cool, thanks for the converison. While I usually like to only get the
patches which I need to handle, please CC me to all patches next time. I
am interested what changes were needed for the user of the slave
framework, too.

+static bool tegra_i2c_slave_isr(int irq, struct tegra_i2c_dev *i2c_dev)
+{
+	unsigned long status;
+	u8 value;
+
+	if (!i2c_dev->slave || !i2c_dev->slave->slave_cb)
+		return false;

Can this happen?

+	/* i2c master sends data to us */
+	if (is_write(status)) {
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_WRITE_START,
+				NULL);

Can this HW create an interrupt once the address detection + RW bit are
received? Or only if a complete write has been received?

+static int tegra_reg_slave(struct i2c_client *slave)
+{
+	struct tegra_i2c_dev *i2c_dev = i2c_get_adapdata(slave->adapter);
+
+	if (i2c_dev->slave)
+		return -EBUSY;
+
+	i2c_dev->slave = slave;
+
+	tegra_i2c_clock_enable(i2c_dev);
+
+	reset_control_assert(i2c_dev->rst);
+	udelay(2);
+	reset_control_deassert(i2c_dev->rst);

Why do you need a reset when a slave gets registered?

+
+	i2c_writel(i2c_dev, I2C_SL_CNFG_NEWSL, I2C_SL_CNFG);
+	i2c_writel(i2c_dev, 0x1E, I2C_SL_DELAY_COUNT);

What does this magic number mean?

+
+	i2c_writel(i2c_dev, slave->addr, I2C_SL_ADDR1);
+	i2c_writel(i2c_dev, 0, I2C_SL_ADDR2);

Handling 10 bit addresses?

+
+	return 0;
+}
+

No need to init to 0 IMO.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20150129/84b5c7aa/attachment.sig>

Hi,

Sorry for long delay.

And thanks for the quick review. It helped a lot!

On 29.01.2015 14:41, Wolfram Sang wrote:

Hi,

quoted

Initialization code is based on NVEC driver.

There is a HW bug in AP20 that was also mentioned in kernel sources
for Toshiba AC100.

Signed-off-by: Andrey Danin <redacted>

Cool, thanks for the converison. While I usually like to only get the
patches which I need to handle, please CC me to all patches next time. I
am interested what changes were needed for the user of the slave
framework, too.

Done. I sent v2 yesterday evening.

quoted

+static bool tegra_i2c_slave_isr(int irq, struct tegra_i2c_dev *i2c_dev)
+{
+	unsigned long status;
+	u8 value;
+
+	if (!i2c_dev->slave || !i2c_dev->slave->slave_cb)
+		return false;

Can this happen?

Yes. I call slave ISR without any conditions from main ISR routine.

quoted

+	/* i2c master sends data to us */
+	if (is_write(status)) {
+		i2c_slave_event(i2c_dev->slave, I2C_SLAVE_REQ_WRITE_START,
+				NULL);

Can this HW create an interrupt once the address detection + RW bit are
received? Or only if a complete write has been received?

Tegra I2C generates one interrupt per byte (address or data) and one 
interrupt for stop bit.

quoted

+static int tegra_reg_slave(struct i2c_client *slave)
+{
+	struct tegra_i2c_dev *i2c_dev = i2c_get_adapdata(slave->adapter);
+
+	if (i2c_dev->slave)
+		return -EBUSY;
+
+	i2c_dev->slave = slave;
+
+	tegra_i2c_clock_enable(i2c_dev);
+
+	reset_control_assert(i2c_dev->rst);
+	udelay(2);
+	reset_control_deassert(i2c_dev->rst);

Why do you need a reset when a slave gets registered?

I copied this code from nvec driver. Reset is done during I2C controller 
initialization. This reset is not needed. Thanks for pointing.

quoted

+
+	i2c_writel(i2c_dev, I2C_SL_CNFG_NEWSL, I2C_SL_CNFG);
+	i2c_writel(i2c_dev, 0x1E, I2C_SL_DELAY_COUNT);

What does this magic number mean?

It's a default value. I created a constant for it.

quoted

+
+	i2c_writel(i2c_dev, slave->addr, I2C_SL_ADDR1);
+	i2c_writel(i2c_dev, 0, I2C_SL_ADDR2);

Handling 10 bit addresses?

In v2.

quoted

+
+	return 0;
+}
+

quoted

  static const struct tegra_i2c_hw_feature tegra30_i2c_hw = {

@@ -679,6 +808,7 @@ static const struct tegra_i2c_hw_feature tegra30_i2c_hw = {
  	.has_single_clk_source = false,
  	.clk_divisor_hs_mode = 3,
  	.clk_divisor_std_fast_mode = 0,
+	.slave_read_start_delay = 0,

No need to init to 0 IMO.

Ok.