Re: [PATCH RFC v2] m68k: remove get_fs()/set_fs()

[Christoph Hellwig <hch@lst.de>]

On Wed, Jul 07, 2021 at 08:01:33PM -0700, Linus Torvalds wrote:
> The '#' turns the argument into a string, but it was already
> _supposed_ to be a string. But no, the problem is that it turns the
> macro name MOVES into the _string_ "MOVES".
> 
> Which happens to compile just fine, because "moves" is a real
> instruction. But it's actually _meant_ to be a macro that expands to
> either the string "moves" or "move".
> 
> So what happens is that at least in Christoph's version, I think the
> code _always_ uses "MOVES", even in configurations where the macro
> MOVES should have just become "move".
> 
> So it all builds fine, looks fine to the assembler, but it uses the
> wrong instruction.
> 
> Macro expansion with the '#' character and other macros used as
> arguments is something people need to be very careful with.  It's why
> we have a whole header file for just the "stringify" operation, see
> <linux/stringify.h>
> 
> But in this case, it shouldn't have used '#' at all, since the
> argument was already a string, and never needed to be turned into a
> string by the pre-processor.

Yes, that problem exists, but just removing the # causes "inst" to go
into the cpp output.  Let me brush up my cpp-foo.