Re: [PATCH mm] tracing: incorrect gfp_t conversion

[Joe Perches <joe@perches.com>]

On Sat, 2022-05-07 at 15:48 -0700, Andrew Morton wrote:
> On Sun, 8 May 2022 01:28:58 +0300 Vasily Averin <vvs@openvz.org> wrote:
> 
> > On 5/7/22 22:37, Andrew Morton wrote:
> > > On Sat, 7 May 2022 22:02:05 +0300 Vasily Averin <vvs@openvz.org> wrote:
> > > > +	{(__force unsigned long)GFP_KERNEL,		"GFP_KERNEL"},		\
> > > > +	{(__force unsigned long)GFP_NOFS,		"GFP_NOFS"},		\
> > > 
> > > This got all repetitive, line-wrappy and ugly :(
> > > 
> > > What do we think of something silly like this?
> > 
> > > --- a/include/trace/events/mmflags.h~tracing-incorrect-gfp_t-conversion-fix
> > > +++ a/include/trace/events/mmflags.h
> > > @@ -13,53 +13,57 @@
> > >   * Thus most bits set go first.
> > >   */
> > >  
> > > +#define FUL __force unsigned long
> > > +
> > >  #define __def_gfpflag_names						\
> > > -	{(__force unsigned long)GFP_TRANSHUGE,		"GFP_TRANSHUGE"},	\
> > > -	{(__force unsigned long)GFP_TRANSHUGE_LIGHT,	"GFP_TRANSHUGE_LIGHT"}, \
> > ...
> > > +	{(FUL)GFP_TRANSHUGE,		"GFP_TRANSHUGE"},	\
> > > +	{(FUL)GFP_TRANSHUGE_LIGHT,	"GFP_TRANSHUGE_LIGHT"}, \
> > 
> > 
> > I think it's a good idea, and I regret it was your idea and not mine.
> 
> heh
> 
> > Should I resend my patch with these changes or would you prefer 
> > to keep your patch as a separate one?
> 
> I did the below.  I'll squash them together later.

Very repetitive indeed.

Why not use another stringifying macro?

Maybe something like:

#define gfpflag_string(GFP)	\
	{(__force unsigned long)GFP, #GFP)}

#define __def_gfpflag_names			\
	gfp_flag_string(GFP_TRANSHUGE),		\
	etc...