On 13-Dec-16 14:11, Rob Herring wrote:

Again, please don't top post. And your line wrapping is messed up.
IOW, you can't use Outlook.

On Tue, Dec 13, 2016 at 4:50 AM, Luis de Oliveira
[off-list ref] wrote:

quoted

The controller for i2c-designware cannot be slave/master at the same time and it has to be enabled knowing beforehand if we want it to be slave or master by something outside of the controller itself.

I as looking and I see the use of this I2C_OWN_SLAVE_ADDRESS with the "linux,slave-24c02" slave interface  but I am not seeing how it will help me identify a selected i2c-designware block as a "slave" device before instantiation. I'm sorry if I'm not understanding properly.
I use the "linux,slave-24c02" to instantiate the i2c-designware as a slave with an address so I can do write/read operations, it is how I tested it.

Something like this:

of_for_each_child_node(child) {
  of_property_read_u32(child, "reg", &reg);
  if (reg & I2C_OWN_SLAVE_ADDRESS))
    im_a_slave = true;
}

...rather than testing "mode" is equal to "slave".

Rob

Hi Rob, Andy,

I'm struggling to implement your suggestion @Rob. I checked the
tegra124-jetson-tk1.dts that uses that approach but I have some doubts.

My DT is as follows

	i2c@0x2000 {
                        compatible = "snps,designware-i2c";
                        reg = <0x2000 0x100>;
                        clock-frequency = <400000>;
                        clocks = <&i2cclk>;
                        interrupts = <0>;

I could add something like this:

			eeprom@64 {
				compatible = "linux,slave-24c02";
				reg = <(I2C_OWN_SLAVE_ADDRESS | 0x64)>;
			}

But I think this is different form what I was doing before. I have two questions:

- This way the I2C controller is identified as a slave controller or just the
subnode eeprom?
- This way looks like my slave address will be fixed

In the previous Patch v3 submission @Andy suggested a property that selects mode
that I did and it's working. And you @Rob suggested to do it a common property.
It is implemented in the DT like:

			mode = "slave";

So before I do this changes can you please agree both if you still think this is
the best approach?


Thank you both for your time,
Luis

quoted

Luis

-----Original Message-----
From: Rob Herring [mailto:robh@kernel.org]
Sent: Monday, December 12, 2016 23:16
To: Luis de Oliveira <redacted>
Cc: wsa@the-dreams.de; mark.rutland@arm.com; jarkko.nikula@linux.intel.com; andriy.shevchenko@linux.intel.com; mika.westerberg@linux.intel.com; linux-i2c@vger.kernel.org; devicetree@vger.kernel.org; linux-kernel@vger.kernel.org; Ramiro.Oliveira@synopsys.com; Joao.Pinto@synopsys.com; CARLOS.PALMINHA@synopsys.com
Subject: Re: [PATCH v4 3/5] i2c: designware: Add slave definitions

On Mon, Dec 12, 2016 at 12:35 PM, Luis de Oliveira [off-list ref] wrote:

quoted

Hi all,

Please don't top post.

quoted

The slave address could be set by the I2C slave backend so I can't use it to setup the controller.
A boolean property was my initial approach then Andy and Wolfram Sang suggested the use of compatible strings and later It was suggested to use a property to select mode but I can do it again if it's the best way.
Can you please tell me where should it be documented?

bindings/i2c/i2c.txt.

Actually, looking at this some more, we already have a way to describe the controller being a slave device with the I2C_OWN_SLAVE_ADDRESS flag in the reg property. We should just need a helper to read reg property of each child and check for the bit set.

Rob