Re: [RFC/PATCH 0/3] OMAP3 I2C/SCCB support

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Jean-Philippe,

On Tuesday 26 June 2012 16:49:35 jean-philippe francois wrote:
> 2012/6/26 Laurent Pinchart <laurent.pinchart@ideasonboard.com>:
> > Hi everybody,
> > 
> > While trying to use an OV7725 camera sensor with an OMAP3 system I ran
> > into a couple of issues related to the sensor communication protocol.
> > Instead of using the obiquitous I2C protocol, Omnivision invited the SCCB
> > communication bus, very similar to I2C but different enough not to be
> > compatible.
> I have been using omnivision sensor without being aware of this issue,
> and without any i2c reliability issue either.
> 
> Here is typical code I use :
> 
> /* This function is used to read value from register for i2c client */
> static int ov2710_read(struct i2c_client *client, unsigned short reg,
> unsigned char *val)
> {
>     int err = 0;
>     unsigned char outbuf[] = {(reg>>8)&0xff, reg&0xff};

According to the SCCB specification (or at least the version I've found) 
register addresses are 8-bit. The OV2710 might just be I2C-compatible.

>     unsigned char inbuf[1];
>     struct i2c_msg msg[] = {
>         { .addr = client->addr, .flags = 0, .buf = outbuf, .len = 2 },
>         { .addr = client->addr, .flags = I2C_M_RD, .buf = inbuf, .len = 1 },
> };
>     err = i2c_transfer(client->adapter, msg, 2);
>     /* error handling and pass by ref handling */
>     ....
> }

The ov772x driver uses i2c_smbus_write_byte_data() and 
i2c_smbus_read_byte_data(). The later calls

i2c_smbus_xfer(client->adapter, client->addr, client->flags,
               I2C_SMBUS_READ, I2C_SMBUS_BYTE_DATA, &data);

which calls i2c_smbus_xfer_emulated() for hosts that don't support SMBus 
transfers natively, and that's pretty much equivalent to your above code 
(except for the 8/16 bit register address).

It might be a good idea to implement i2c_smbus_*-like functions for 16-bit 
register addresses in the I2C core, they could be reused across drivers.

> Is the point of this patch to avoid writing such functions again and again
> in every ov driver ?

No, but that's a good idea as well :-)

> What is solved in practice by this patch that is not solved by a regular i2c
> transfer ?

The above code will not ignore acks/nacks and will not generate a stop 
condition between the two messages. The SCCB specification states that 
acks/nacks must be ignored and that a stop condition must be generated. The 
OV7725 handles acks/nacks correctly, but chokes if no stop condition is 
generated. The point of this patch set is to fix both problems (although the 
acks/nacks issue might be theoretical only).

-- 
Regards,

Laurent Pinchart