Re: [PATCH v2 4/5] tc90522: add driver for Toshiba TC90522 quad demodulator

[Akihiro TSUKADA <tskd08@gmail.com>]

Moi!

> Yes, using the I2C binding way provides a better decoupling than using the
> legacy way. The current dvb_attach() macros are hacks that were created
> by the time where the I2C standard bind didn't work with DVB.

I understand. I converted my code to use i2c binding model,
but I'm uncertain about a few things.

1. How to load the modules of i2c driver?
Currently I use request_module()/module_put()
like an example (ddbrige-core.c) from Antti does,
but I'd prefer implicit module loading/ref-counting
like in dvb_attach() if it exists.


2. Is there a standard way to pass around dvb_frontend*, i2c_client*,
regmap* between bridge(dvb_adapter) and demod/tuner drivers?
Currently I use config structure for the purpose, which is set to
dev.platform_data (via i2c_board_info/i2c_new_device()) or
dev.driver_data (via i2c_{get,set}_clientdata()),
but using config as both IN/OUT looks a bit hacky.

3. Should I also use RegMap API for register access?
I tried using it but gave up,
because it does not fit well to one of my use-case,
where (only) reads must be done via 0xfb register, like
   READ(reg, buf, len) -> [addr/w, 0xfb, reg], [addr/r, buf[0]...],
   WRITE(reg, buf, len) -> [addr/w, reg, buf[0]...],
and regmap looked to me overkill for 8bit-reg, 8bit-val cases
and did not simplify the code.
so I'd like to go without RegMap if possible,
since I'm already puzzled enough by I2C binding, regmap, clock source,
as well as dvb-core, PCI ;)

regards,
Akihiro