[1/6] x86: move zmm exclusion list into CPU feature flag

Commit Message

Eric Biggers Nov. 25, 2024, 4:11 a.m. UTC

From: Eric Biggers <[e-mail]>

Lift zmm_exclusion_list in aesni-intel_glue.c into the x86 CPU setup
code, and add a new x86 CPU feature flag X86_FEATURE_PREFER_YMM that is
set when the CPU is on this list.

This allows other code in arch/x86/, such as the CRC library code, to
apply the same exclusion list when deciding whether to execute 256-bit
or 512-bit optimized functions.

Note that full AVX512 support including zmm registers is still exposed
to userspace and is still supported for in-kernel use.  This flag just
indicates whether in-kernel code should prefer to use ymm registers.

Signed-off-by: Eric Biggers <[e-mail]>
---
 arch/x86/crypto/aesni-intel_glue.c | 22 +---------------------
 arch/x86/include/asm/cpufeatures.h |  1 +
 arch/x86/kernel/cpu/intel.c        | 22 ++++++++++++++++++++++
 3 files changed, 24 insertions(+), 21 deletions(-)

Patch

diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index fbf43482e1f5e..8e648abfb5ab8 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -1534,30 +1534,10 @@  DEFINE_GCM_ALGS(vaes_avx10_256, FLAG_AVX10_256,
 DEFINE_GCM_ALGS(vaes_avx10_512, FLAG_AVX10_512,
 		"generic-gcm-vaes-avx10_512", "rfc4106-gcm-vaes-avx10_512",
 		AES_GCM_KEY_AVX10_SIZE, 800);
 #endif /* CONFIG_AS_VAES && CONFIG_AS_VPCLMULQDQ */
 
-/*
- * This is a list of CPU models that are known to suffer from downclocking when
- * zmm registers (512-bit vectors) are used.  On these CPUs, the AES mode
- * implementations with zmm registers won't be used by default.  Implementations
- * with ymm registers (256-bit vectors) will be used by default instead.
- */
-static const struct x86_cpu_id zmm_exclusion_list[] = {
-	X86_MATCH_VFM(INTEL_SKYLAKE_X,		0),
-	X86_MATCH_VFM(INTEL_ICELAKE_X,		0),
-	X86_MATCH_VFM(INTEL_ICELAKE_D,		0),
-	X86_MATCH_VFM(INTEL_ICELAKE,		0),
-	X86_MATCH_VFM(INTEL_ICELAKE_L,		0),
-	X86_MATCH_VFM(INTEL_ICELAKE_NNPI,	0),
-	X86_MATCH_VFM(INTEL_TIGERLAKE_L,	0),
-	X86_MATCH_VFM(INTEL_TIGERLAKE,		0),
-	/* Allow Rocket Lake and later, and Sapphire Rapids and later. */
-	/* Also allow AMD CPUs (starting with Zen 4, the first with AVX-512). */
-	{},
-};
-
 static int __init register_avx_algs(void)
 {
 	int err;
 
 	if (!boot_cpu_has(X86_FEATURE_AVX))
@@ -1598,11 +1578,11 @@  static int __init register_avx_algs(void)
 					 ARRAY_SIZE(aes_gcm_algs_vaes_avx10_256),
 					 aes_gcm_simdalgs_vaes_avx10_256);
 	if (err)
 		return err;
 
-	if (x86_match_cpu(zmm_exclusion_list)) {
+	if (boot_cpu_has(X86_FEATURE_PREFER_YMM)) {
 		int i;
 
 		aes_xts_alg_vaes_avx10_512.base.cra_priority = 1;
 		for (i = 0; i < ARRAY_SIZE(aes_gcm_algs_vaes_avx10_512); i++)
 			aes_gcm_algs_vaes_avx10_512[i].base.cra_priority = 1;
diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h
index 17b6590748c00..948bfa25ccc7b 100644
--- a/arch/x86/include/asm/cpufeatures.h
+++ b/arch/x86/include/asm/cpufeatures.h
@@ -477,10 +477,11 @@ 
 #define X86_FEATURE_CLEAR_BHB_HW	(21*32+ 3) /* BHI_DIS_S HW control enabled */
 #define X86_FEATURE_CLEAR_BHB_LOOP_ON_VMEXIT (21*32+ 4) /* Clear branch history at vmexit using SW loop */
 #define X86_FEATURE_AMD_FAST_CPPC	(21*32 + 5) /* Fast CPPC */
 #define X86_FEATURE_AMD_HETEROGENEOUS_CORES (21*32 + 6) /* Heterogeneous Core Topology */
 #define X86_FEATURE_AMD_WORKLOAD_CLASS	(21*32 + 7) /* Workload Classification */
+#define X86_FEATURE_PREFER_YMM		(21*32 + 8) /* Avoid zmm registers due to downclocking */
 
 /*
  * BUG word(s)
  */
 #define X86_BUG(x)			(NCAPINTS*32 + (x))
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index d1de300af1737..0beb44c4ac026 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -519,10 +519,29 @@  static void init_intel_misc_features(struct cpuinfo_x86 *c)
 
 	msr = this_cpu_read(msr_misc_features_shadow);
 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
 }
 
+/*
+ * This is a list of Intel CPUs that are known to suffer from downclocking when
+ * zmm registers (512-bit vectors) are used.  On these CPUs, when the kernel
+ * executes SIMD-optimized code such as cryptography functions or CRCs, it
+ * should prefer 256-bit (ymm) code to 512-bit (zmm) code.
+ */
+static const struct x86_cpu_id zmm_exclusion_list[] = {
+	X86_MATCH_VFM(INTEL_SKYLAKE_X,		0),
+	X86_MATCH_VFM(INTEL_ICELAKE_X,		0),
+	X86_MATCH_VFM(INTEL_ICELAKE_D,		0),
+	X86_MATCH_VFM(INTEL_ICELAKE,		0),
+	X86_MATCH_VFM(INTEL_ICELAKE_L,		0),
+	X86_MATCH_VFM(INTEL_ICELAKE_NNPI,	0),
+	X86_MATCH_VFM(INTEL_TIGERLAKE_L,	0),
+	X86_MATCH_VFM(INTEL_TIGERLAKE,		0),
+	/* Allow Rocket Lake and later, and Sapphire Rapids and later. */
+	{},
+};
+
 static void init_intel(struct cpuinfo_x86 *c)
 {
 	early_init_intel(c);
 
 	intel_workarounds(c);
@@ -602,10 +621,13 @@  static void init_intel(struct cpuinfo_x86 *c)
 		set_cpu_cap(c, X86_FEATURE_P4);
 	if (c->x86 == 6)
 		set_cpu_cap(c, X86_FEATURE_P3);
 #endif
 
+	if (x86_match_cpu(zmm_exclusion_list))
+		set_cpu_cap(c, X86_FEATURE_PREFER_YMM);
+
 	/* Work around errata */
 	srat_detect_node(c);
 
 	init_ia32_feat_ctl(c);