On Mon, Mar 02, 2020 at 05:34:19PM +0100, Linus Walleij wrote: > On Mon, Mar 2, 2020 at 10:36 AM Russell King - ARM Linux admin > <linux@xxxxxxxxxxxxxxx> wrote: > > On Mon, Mar 02, 2020 at 09:39:59AM +0100, Uwe Kleine-König wrote: > > > > - Sometimes it is useful to make use of a GPIO and a dedicated function > > > on the same pin in a driver (e.g. an i2c driver might need to switch > > > to gpio to do a bus recovery). The automatic pinmuxing then has > > > strange side effect because you have to remux the pins after > > > requesting the GPIOs even if you didn't drive the pins as GPIO and > > > there is a short time where the pin function isn't the dedicated > > > one. > > > > It's worse than that for the i2c driver. The pins are muxed to the i2c > > function when the driver binds. When the i2c driver claims the GPIOs > > corresponding with those pins, they get switched to GPIO mode behind > > the back of pinctrl. You then have to _explicitly_ switch pinctrl to > > GPIO mode and back to I2C mode to get them back to I2C mode. > > That's especially annoying. I would consider adding a specific > consumer flag for GPIOs used this way, in additon to > GPIOD_ASIS, something like GPIOD_ASIS_NOMUX > (thinking of better names). It's very annoying, and I believe something I did point out in my email about it when I discovered it towards the end of last year. Having a way to avoid the muxing would be a very good idea, as there are cases where we really should not be taking the I2C pins away from the controller during driver initialisation. In the case of i2c-pxa, when the pins are taken away, the controller sees the disconnected SCL and SDA lines go low, and it can assume that the bus is busy as a result, or worse see it as a START condition if SDA goes low while SCL is high. -- RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up According to speedtest.net: 11.9Mbps down 500kbps up