On 14.08.2024 13:14, Wolfram Sang wrote: > Hi Heiner, > > thanks for tackling this! > >> +static int i2c_lock_addr(struct i2c_adapter *adap, unsigned short addr, >> + unsigned short flags) > > What about just using 'struct i2c_client *client' here as an argument. > It has all we need and it currently seems unlikely that we need to call > it from somewhere else where we need this seperation. > >> + if (!(flags & I2C_CLIENT_TEN) && !(flags & I2C_CLIENT_SLAVE) && > > From a pedantic point of view, I don't see a reason for not handling > those two cases above. I hate to be pedantic because 10-bit mode is > practically unused (and I am tempted to remove support for it once in a > while because it makes other solutions clumsy). And the other one is > super unlikely to happen because the backends do not autoload. However, > it is theoretically possible if someone loads a devicetree overlay and > initiates via sysfs at the same time. I liked the solution with the > bitfield and atomic access, but maybe a linked list is better? > Wrt 10 bit addresses: I didn't find a single dts(i) with a 10bit i2c device. Because you said that you're tempted to remove 10 bit support: Could the device tree part be a starting point? > Happy hacking, > > Wolfram > Heiner
