14:45 <rfs613> CounterPillow: the one other thing about regmap is that it lets you "intercept" all the read/writes, which can be useful both for debuggin, and also in situations where hardware has some exotic errata you need to work around.

14:48 <broonie> Yeah, the trace/diagnostic stuff is useful (there's also things like being able to see the regmap in debugfs), and it avoids open coding the I/O stuff (which can improve interoperability with weird controllers). In general it's implementing a bunch of helpful stuff that you'd probably never open code in an individual driver.

14:50 <broonie> If your control bus is something like I2C or SPI the overheads are negligable.

15:02 <CounterPillow> For now I went with cacheless regmap (the bus is mmio) because I do like the regmap layer catching me trying to write a register I didn't declare as being writable

15:22 <broonie> Amusingly the MMIO regmap was added precisely because someone wanted the caching (the IP mostly did circular DMA when running and regcache_sync() makes resume super simple).