On Thu, Oct 18, 2018 at 01:11:53PM +0200, Petr Mladek wrote: > On Wed 2018-10-17 13:17:56, Josh Poimboeuf wrote: > > On Mon, Oct 15, 2018 at 02:37:03PM +0200, Petr Mladek wrote: > > > The definition of struct klp_func might be a bit confusing. > > > The original function is defined by name as a string. > > > The new function is defined by name as a function pointer > > > casted to unsigned long. > > > > > > This patch adds helper macros that hide the different types. > > > The functions are defined just by the name. For example: > > > > On one hand, these macros are kind of nice, because they do the function > > pointer casting for the user. > > > > On the other hand, they hide the field names, which hurts readability a > > bit. For example, it would be easy to accidentally assign the wrong > > callback function. > > I am not a big fan of the macros either. > > > > Also, it's unfortunate that these macros are needed in the first place. > > > > What if we just change new_addr (and old_addr) to be 'void *'? Then the > > macros wouldn't be as useful, and we could just get rid of them. > > Hmm, I wonder if any change make sense then. The above proposal might > just exchange one confusion with another one: > > + I would expect that a variable called addr is of the type > unsigned long > > + we would need casting when calling ftrace API > > + .new_addr = function_xxx looks a bit weird and it will > be used many times in all livepatch sources. Fair points. Instead of changing new_func to new_addr, how about we leave it alone, and instead change 'unsigned long old_addr' to 'void *old_func'? That would give us consistent naming internally, while making the external interface more sensible and cast-free. We'd still have to cast when passing to the ftrace API, but IMO it would be worth it. -- Josh
