Hi Arnd,

> The gxp_i2c_slave_irq_handler() is hidden in an #ifdef, but the
> caller uses an IS_ENABLED() check:
> 
> drivers/i2c/busses/i2c-gxp.c: In function 'gxp_i2c_irq_handler':
> drivers/i2c/busses/i2c-gxp.c:467:29: error: implicit declaration of function 'gxp_i2c_slave_irq_handler'; did you mean 'gxp_i2c_irq_handler'? [-Werror=implicit-function-declaration]
> 
> It has to consistently use one method or the other to avoid warnings,
> so move to IS_ENABLED() here for readability and build coverage, and
> move the #ifdef in linux/i2c.h to allow building it as dead code.

Can't we have a solution which modifies this driver only (maybe by
defining an empty irq handler for the non-IS_ENABLED part?)? Doesn't
feel good to touch i2c.h only because of this...

> -#if IS_ENABLED(CONFIG_I2C_SLAVE)
>  enum i2c_slave_event {
>  	I2C_SLAVE_READ_REQUESTED,
>  	I2C_SLAVE_WRITE_REQUESTED,
> @@ -396,9 +395,10 @@ enum i2c_slave_event {
>  
>  int i2c_slave_register(struct i2c_client *client, i2c_slave_cb_t slave_cb);
>  int i2c_slave_unregister(struct i2c_client *client);

... especially with moving these two prototypes out of the protected
block. The functions themselves are also protected by the same symbol
via the Makefile. I'd rather get a build error right away than a linker
error later if a driver misses to select I2C_SLAVE. Or do I miss
something?

> -bool i2c_detect_slave_mode(struct device *dev);
>  int i2c_slave_event(struct i2c_client *client,
>  		    enum i2c_slave_event event, u8 *val);
> +#if IS_ENABLED(CONFIG_I2C_SLAVE)
> +bool i2c_detect_slave_mode(struct device *dev);
>  #else
>  static inline bool i2c_detect_slave_mode(struct device *dev) { return false; }
>  #endif

All the best,

   Wolfram