[I2C] informations + advice about messages handling

[anish singh]

On Fri, May 24, 2013 at 1:14 PM, Jean Delvare <khali at linux-fr.org> wrote:
> Hi Anish, Mylène,
>
> On Fri, 24 May 2013 12:52:40 +0530, anish singh wrote:
>> On Fri, May 24, 2013 at 12:41 PM, Mylene Josserand
>> <Mylene.Josserand at navocap.com> wrote:
>> > I have read that this function "i2c_smbus_write_byte_data" uses
>> > "i2c_smbus_xfer" which uses "i2c_lock_adapter".
>> > In this function, there is a mutex so I thought that it will handle it
>> > but it says "Get exclusive access to an I2C bus segment". What is
>> > exactly an I2C segment ? Is it the device we are talking about ? Or is
>> > it the use of the i2c bus ?
>> Don't know what you are referring here.
>
> In the most simple case, your I2C bus has a single segment so "segment"
> or "bus" mean the same thing.
>
> It only starts mattering when I2C bus multiplexing comes into play.
> Then your bus is split into segments, with one segment (trunk) between
> the master (controller) and the multiplexer, and one or more segments
> (branches) hanging off the multiplexer.
>
> Take look at
> https://i2c.wiki.kernel.org/index.php/I2C_bus_multiplexing
> for a sample topology.
>
> But anyway the comment in front of i2c_lock_adapter() is somewhat
> misleading. If you read the code you'll see that it always locks the
> whole bus tree, because it uses the root segment's mutex.
>
>> > I will certainly have to create an i2c driver and I would like to know
>> > if this "collision" handling (if it is handled) is done in old kernel
>> > (2.6.32) or is it handled only in new kernel versions ?
>> AFAIK collision handling and detection was not supported till now
>> in linux kernel until recently but I think this patch is doing that.
>> I may be wrong but I didn't see collision handling in earlier linux
>> kernels.
>> http://thread.gmane.org/gmane.linux.kernel/1410276
>
> This is for a specific case. The general case is handled by the
> per-adapter mutex for years now. 2.6.32 should be just fine in this
> respect.
I may be mis-understanding here but when two client wants to
communicate with same master and at the same
time how does that happen?
In the thread I mentioned client pulls the line low indicating that
it wants to own the bus and waits for some time and then checks
if the line is still low or not.If it is still low then it owns the bus
and start transaction and if clients see otherwise it releases the
bus and tries again after some time.How is that handled in the
previous kernel?
>
>> > If you have any documentation on how the i2c messages are handled in
>> > case of different devices uses, it will help me a lot ! I will search in
>> > the kernel documentation but there is many files about i2c.
>> > And if you know a good i2c driver that I can use as an example to design
>> > mine, it will be great !
>
> Best is to look at a driver for a device which is similar in
> functionality to yours.
>
> --
> Jean Delvare