Hello, In media drivers, TS drivers typically hard-code their front-end (demod and tuner) init by loading the modules themselves. I feel this is not a good solution for SoCs, where the TS HW might be on the SoC, and the front-end be on the board. So we may have different front-ends for different boards, and the driver would have to hard-code all of them. Am I making sense? Here's an example of what I mean: https://elixir.bootlin.com/linux/latest/source/drivers/media/usb/dvb-usb-v2/dvbsky.c#L466 I've been working on defining the demod in DT, and having a phandle to the demod in the TSIF node. I've got everything working like I had hoped, but I have many ugly hacks. The TSIF driver needs to register the frontend, which is created in the demod driver. So I have: struct device_node *toto = of_parse_phandle(np, "demod", 0); if (!toto) panic("of_parse_phandle"); struct i2c_client *demod = of_find_i2c_device_by_node(toto); if (!demod) panic("of_find_i2c_device_by_node"); printk("\tdemod=%px\n", demod); struct dvb_frontend *get_fe(struct i2c_client *client); my_dvb_frontend = get_fe(demod); The problem is get_fe(). It needs to be a call-back, so that every demod can implement his own version. But only a few i2c_client's have a dvb_frontend to return. Could we have a generic void *get_something() callback in struct i2c_client? (Seems like the wrong place) How can I solve this conundrum? Maybe look above i2c, in struct device? Regards.
