Re: [PATCH v2] x86/altcall: further refine clang workaround

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Mon, Jul 29, 2024 at 02:41:23PM +0200, Jan Beulich wrote:
> On 29.07.2024 14:00, Roger Pau Monné wrote:
> > On Mon, Jul 29, 2024 at 12:47:09PM +0200, Jan Beulich wrote:
> >> On 29.07.2024 12:30, Roger Pau Monne wrote:
> >>> --- a/xen/arch/x86/include/asm/alternative.h
> >>> +++ b/xen/arch/x86/include/asm/alternative.h
> >>> @@ -183,13 +183,13 @@ extern void alternative_branches(void);
> >>>   * https://github.com/llvm/llvm-project/issues/12579
> >>>   * https://github.com/llvm/llvm-project/issues/82598
> >>>   */
> >>> -#define ALT_CALL_ARG(arg, n)                                            \
> >>> -    register union {                                                    \
> >>> -        typeof(arg) e[sizeof(long) / sizeof(arg)];                      \
> >>> -        unsigned long r;                                                \
> >>> -    } a ## n ## _ asm ( ALT_CALL_arg ## n ) = {                         \
> >>> -        .e[0] = ({ BUILD_BUG_ON(sizeof(arg) > sizeof(void *)); (arg); })\
> >>> -    }
> >>> +#define ALT_CALL_ARG(arg, n)                                             
> >>> \
> >>> +     register unsigned long a ## n ## _ asm ( ALT_CALL_arg ## n ) = ({   
> >>> \
> >>> +         unsigned long tmp = 0;                                          
> >>> \
> >>> +         *(typeof(arg) *)&tmp = (arg);                                   
> >>> \
> >>> +         BUILD_BUG_ON(sizeof(arg) > sizeof(void *));                     
> >>> \
> >>
> >> With this, even more so than before, I think the type of tmp would better
> >> be void * (or the BUILD_BUG_ON() be made use unsigned long, yet I consider
> >> that less desirable).
> > 
> > Won't using void * be uglier, as we then need to cast the last tmp
> > statement to unsigned long?
> 
> Only if we stick to using unsigned long for a ## n ## _. Afaics there's
> nothing wrong with making that void *, too.

Right, but then for consistency I would also like to make r{10,11}_
void *, and ALT_CALL_NO_ARG(), which might be too much.

My preference is likely to keep it at unsigned long, and adjust the
BUILD_BUG_ON(), unless you have a strong opinion to change it to void
* (and possibly the rest of the register variables).

Thanks, Roger.