Since as a bug workaround (likely inapplicable to any 64-bit CPUs, but
it probably doesn't hurt to keep this) we clear %ecx on input anyway,
we can as well fall back to cpuid_count(). This allows getting rid of
four risky casts and makes things type-safe. The latter aspect requires
two type adjustments elsewhere. While adjusting init_intel_cacheinfo(),
convert three other local variables there as well. For the struct
cpuinfo_x86 change it is relevant to note that no 64-bit CPU comes
without CPUID support, and hence cpuid_level is never set to -1; the
comment there was simply stale.

No functional change intended, yet of course generated code isn't
identical.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
It may also be interesting to see whether this addresses issues Coverity
recently spotted (in error), in spec-ctrl.c:print_details(). Andrew
considers cpuid() being (just) a macro as a possible reason.

I'm willing to switch to an inline function, but I chose the macro route
because this way it's less code (and hence less redundancy) overall,
without any loss of functionality.

--- a/xen/arch/x86/cpu/intel_cacheinfo.c
+++ b/xen/arch/x86/cpu/intel_cacheinfo.c
@@ -172,8 +172,7 @@ void init_intel_cacheinfo(struct cpuinfo
 	    c->x86_vendor != X86_VENDOR_SHANGHAI)
 	{
 		/* supports eax=2  call */
-		int i, j, n;
-		int regs[4];
+		unsigned int i, j, n, regs[4];
 		unsigned char *dp = (unsigned char *)regs;
 		int only_trace = 0;
 
--- a/xen/arch/x86/include/asm/cpufeature.h
+++ b/xen/arch/x86/include/asm/cpufeature.h
@@ -23,7 +23,7 @@ struct cpuinfo_x86 {
     unsigned char x86_vendor;          /* CPU vendor */
     unsigned char x86_model;
     unsigned char x86_mask;
-    int cpuid_level;                   /* Maximum supported CPUID level, -1=no CPUID */
+    unsigned int cpuid_level;          /* Maximum supported CPUID level */
     unsigned int extended_cpuid_level; /* Maximum supported CPUID extended level */
     unsigned int x86_capability[NCAPINTS];
     char x86_vendor_id[16];
--- a/xen/arch/x86/include/asm/processor.h
+++ b/xen/arch/x86/include/asm/processor.h
@@ -131,13 +131,8 @@ static inline int cpu_nr_siblings(unsign
  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
  * resulting in stale register contents being returned.
  */
-#define cpuid(_op,_eax,_ebx,_ecx,_edx)          \
-    asm volatile ( "cpuid"                      \
-          : "=a" (*(int *)(_eax)),              \
-            "=b" (*(int *)(_ebx)),              \
-            "=c" (*(int *)(_ecx)),              \
-            "=d" (*(int *)(_edx))               \
-          : "0" (_op), "2" (0) )
+#define cpuid(op, eax, ebx, ecx, edx)          \
+        cpuid_count(op, 0, eax, ebx, ecx, edx)
 
 /* Some CPUID calls want 'count' to be placed in ecx */
 static inline void cpuid_count(

On 09/08/2023 1:29 pm, Jan Beulich wrote:
> Since as a bug workaround (likely inapplicable to any 64-bit CPUs, but
> it probably doesn't hurt to keep this) we clear %ecx on input anyway,
> we can as well fall back to cpuid_count(). This allows getting rid of
> four risky casts and makes things type-safe. The latter aspect requires
> two type adjustments elsewhere. While adjusting init_intel_cacheinfo(),
> convert three other local variables there as well. For the struct
> cpuinfo_x86 change it is relevant to note that no 64-bit CPU comes
> without CPUID support, and hence cpuid_level is never set to -1; the
> comment there was simply stale.
>
> No functional change intended, yet of course generated code isn't
> identical.
>
> Signed-off-by: Jan Beulich <jbeulich@suse.com>

Hmm, far less invasive than when I last tried this, but that was years ago.

> --- a/xen/arch/x86/include/asm/processor.h
> +++ b/xen/arch/x86/include/asm/processor.h
> @@ -131,13 +131,8 @@ static inline int cpu_nr_siblings(unsign
>   * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
>   * resulting in stale register contents being returned.
>   */
> -#define cpuid(_op,_eax,_ebx,_ecx,_edx)          \
> -    asm volatile ( "cpuid"                      \
> -          : "=a" (*(int *)(_eax)),              \
> -            "=b" (*(int *)(_ebx)),              \
> -            "=c" (*(int *)(_ecx)),              \
> -            "=d" (*(int *)(_edx))               \
> -          : "0" (_op), "2" (0) )
> +#define cpuid(op, eax, ebx, ecx, edx)          \
> +        cpuid_count(op, 0, eax, ebx, ecx, edx)

Can we swap op for leaf seeing as both are changing?

Preferably with that changed, Acked-by: Andrew Cooper
<andrew.cooper3@citrix.com>