> It is a very bad idea to replace an i2c xfer by a pair of i2c > send()/recv(), as, if are there any other device at the bus managed > by an independent driver, you may end by mangling i2c transfers and > eventually cause device malfunctions. For I2C, this is true and a very important detail. Yet, we are talking not I2C but SCCB here and SCCB demands a STOP between messages. So, technically, to avoid what you describe one shouldn't mix I2C and SCCB devices. I am quite aware the reality is very different, but still... My preference would be to stop acting as SCCB was I2C but give it its own set of functions so it becomes clear for everyone what protocol is used for what device. > So, IMO, the best is to push the patch you proposed that adds a > new I2C flag: > > https://patchwork.linuxtv.org/patch/49396/ Sorry, but I don't like it. This makes the I2C core code very unreadable. This is why I think SCCB should be exported to its own realm. Which may live in i2c-core-sccb.c, no need for a seperate subsystem.
Attachment:
signature.asc
Description: PGP signature
