I am fine with taking a patch to get rid of __constant_cpu_to_XXX (and converting to the same cpu_to_XXX with the "__constant") in fs/cifs (assuming that that is still recommended). On Thu, Mar 14, 2019 at 2:39 AM Sergey Senozhatsky <sergey.senozhatsky.work@xxxxxxxxx> wrote: > > On (03/14/19 02:19), Steve French wrote: > > All of those uses of __constant_cpu_to_le32 apparently (at least > > according to checkpatch) should be changed (someday) to cpu_to_le32 > > but I didn't research why the change from __constant_cpu_to_le32 > > ---> cpu_to_le32 > > Probably historic reasons. > > Looking at linux 2.4.21 > > /* > * Allow constant folding > */ > #if defined(__GNUC__) && (__GNUC__ >= 2) && defined(__OPTIMIZE__) > # define __swahw32(x) \ > (__builtin_constant_p((__u32)(x)) ? \ > ___swahw32((x)) : \ > __fswahw32((x))) > > > My assumption would be that __GNUC__ < 2 did no support > __builtin_constant_p? > > > > If it has benefit - and checkpatch is right (it warned about > > __constant_cpu_to_le32 being no longer preferred) ... perhaps would be > > worth a followup patch to clean the rest of them up? If you have any > > context on why kernel code has moved away from using the older format > > of __constant_cpu_to_.... would be useful to know if any benefit to > > the change > > Right, that's what I'm going to do - send out patches and update the rest > of __constant_cpu_to_XX users; so, eventually, __constant_cpu_to_XX > can be removed. > > -ss -- Thanks, Steve
