[PATCH v2 10/20] tools/x86/kcpuid: Use <cpuid.h> intrinsics
From: Ahmed S. Darwish
Date: Wed Mar 12 2025 - 10:39:59 EST
Use the __cpuid_count() intrinsic, provided by GCC and LLVM, instead of
rolling a manual version. Both of the kernel's minimum required GCC
version (5.1) and LLVM version (13.0.1) supports it, and it is heavily
used across standard Linux user-space tooling.
This also makes the CPUID call sites more readable.
Signed-off-by: Ahmed S. Darwish <darwi@xxxxxxxxxxxxx>
---
tools/arch/x86/kcpuid/kcpuid.c | 37 ++++++++++++++--------------------
1 file changed, 15 insertions(+), 22 deletions(-)
diff --git a/tools/arch/x86/kcpuid/kcpuid.c b/tools/arch/x86/kcpuid/kcpuid.c
index 79deb506b349..0dbd93ab652a 100644
--- a/tools/arch/x86/kcpuid/kcpuid.c
+++ b/tools/arch/x86/kcpuid/kcpuid.c
@@ -1,6 +1,7 @@
// SPDX-License-Identifier: GPL-2.0
#define _GNU_SOURCE
+#include <cpuid.h>
#include <err.h>
#include <getopt.h>
#include <stdbool.h>
@@ -86,16 +87,16 @@ static u32 user_index = 0xFFFFFFFF;
static u32 user_sub = 0xFFFFFFFF;
static int flines;
-static inline void cpuid(u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
-{
- /* ecx is often an input as well as an output. */
- asm volatile("cpuid"
- : "=a" (*eax),
- "=b" (*ebx),
- "=c" (*ecx),
- "=d" (*edx)
- : "0" (*eax), "2" (*ecx));
-}
+/*
+ * Force using <cpuid.h> __cpuid_count() instead of __cpuid(). The
+ * latter leaves ECX uninitialized, which can break CPUID queries.
+ */
+
+#define cpuid(leaf, a, b, c, d) \
+ __cpuid_count(leaf, 0, a, b, c, d)
+
+#define cpuid_count(leaf, subleaf, a, b, c, d) \
+ __cpuid_count(leaf, subleaf, a, b, c, d)
static inline bool has_subleafs(u32 f)
{
@@ -195,12 +196,7 @@ struct cpuid_range *setup_cpuid_range(u32 input_eax)
u32 max_func, idx_func;
u32 eax, ebx, ecx, edx;
- eax = input_eax;
- ebx = ecx = edx = 0;
- cpuid(&eax, &ebx, &ecx, &edx);
-
- max_func = eax;
- idx_func = (max_func & 0xffff) + 1;
+ cpuid(input_eax, max_func, ebx, ecx, edx);
range = malloc(sizeof(struct cpuid_range));
if (!range)
@@ -211,6 +207,7 @@ struct cpuid_range *setup_cpuid_range(u32 input_eax)
else
range->is_ext = false;
+ idx_func = (max_func & 0xffff) + 1;
range->funcs = malloc(sizeof(struct cpuid_func) * idx_func);
if (!range->funcs)
err(EXIT_FAILURE, NULL);
@@ -222,9 +219,7 @@ struct cpuid_range *setup_cpuid_range(u32 input_eax)
u32 max_subleaf = MAX_SUBLEAF_NUM;
bool allzero;
- eax = f;
- ecx = 0;
- cpuid(&eax, &ebx, &ecx, &edx);
+ cpuid(f, eax, ebx, ecx, edx);
allzero = cpuid_store(range, f, 0, eax, ebx, ecx, edx);
if (allzero)
@@ -251,9 +246,7 @@ struct cpuid_range *setup_cpuid_range(u32 input_eax)
max_subleaf = 5;
for (u32 subleaf = 1; subleaf < max_subleaf; subleaf++) {
- eax = f;
- ecx = subleaf;
- cpuid(&eax, &ebx, &ecx, &edx);
+ cpuid_count(f, subleaf, eax, ebx, ecx, edx);
allzero = cpuid_store(range, f, subleaf, eax, ebx, ecx, edx);
if (allzero)
--
2.48.1