Re: [PATCH] regmap: replace multiple BUG_ON assertions with error return code

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Sun, Dec 15, 2019 at 12:39:52PM -0600, Aditya Pakki wrote:
> Various register operations check for the validity of cache_ops
> struct and crash in case of failure. However, by returning the error
> to the callers, instead of assert, these functions can avoid the crash.
> 
> Signed-off-by: Aditya Pakki <pakki001@umn.edu>
> ---
>  drivers/base/regmap/regcache.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/base/regmap/regcache.c b/drivers/base/regmap/regcache.c
> index a93cafd7be4f..855fa25ae595 100644
> --- a/drivers/base/regmap/regcache.c
> +++ b/drivers/base/regmap/regcache.c
> @@ -238,7 +238,8 @@ int regcache_read(struct regmap *map,
>  	if (map->cache_type == REGCACHE_NONE)
>  		return -ENOSYS;
>  
> -	BUG_ON(!map->cache_ops);
> +	if (!map->cache_ops)
> +		return -EINVAL;

While it is fun to make patches like this, again, if you can prove that
this can never happen, just remove the check please.

And if it can happen, why is -EINVAL the correct error code here?

thanks,

greg k-h