Re: [PATCH] mips: lib: uncached: fix uninitialized variable 'sp'

[Maciej W. Rozycki]

On Fri, 27 Nov 2020, Anders Roxell wrote:

> When building mips tinyconfig with clang the following warning show up:
> 
> /tmp/arch/mips/lib/uncached.c:40:18: note: initialize the variable 'sp' to silence this warning
>         register long sp __asm__("$sp");
>                         ^
>                          = 0
> 
> Rework to make an explicit inline move.

 It's not uninitialised, because by definition $sp has been initialised.  
This code wouldn't have worked for the last 15 years otherwise.
 
 It is however non-standard usage (which may not have been clarified at 
the time this piece was made; I recall extensive discussions around this 
feature over the years):

"6.47.5.2 Specifying Registers for Local Variables
.................................................
[...]

"The only supported use for this feature is to specify registers for
input and output operands when calling Extended 'asm' (*note Extended
Asm::).  [...]"

so the change description should refer to that as the rationale rather 
than an implementation-specific misleading warning (although that can be 
quoted for a reference of course).

> diff --git a/arch/mips/lib/uncached.c b/arch/mips/lib/uncached.c
> index 09d5deea747f..21a4b94a0558 100644
> --- a/arch/mips/lib/uncached.c
> +++ b/arch/mips/lib/uncached.c
> @@ -37,10 +37,11 @@
>   */
>  unsigned long run_uncached(void *func)
>  {
> -	register long sp __asm__("$sp");
>  	register long ret __asm__("$2");
>  	long lfunc = (long)func, ufunc;
>  	long usp;
> +	long sp;
> +	asm ("move %0, $sp" : "=r" (sp));

 Missing newline between the variable block and code here; you may well 
swap it with the empty line below you have left.  Also use `__asm__' for 
consistency, and remove the extraneous space ahead of the parenthesis as 
we don't use them in Linux.  Using a tab to separate the mnemonic from 
operands is also preferable, i.e. "move\t%0, $sp" in this case (using 
actual tabs works better with multi-line assembly, but this is not one).

 I think this should be OK once the issues I mentioned here have been 
addressed.

  Maciej