9ae3a8
From 7b43b5139f8e919203d3ed20fbba6cb143fde6d7 Mon Sep 17 00:00:00 2001
9ae3a8
From: Eduardo Habkost <ehabkost@redhat.com>
9ae3a8
Date: Thu, 23 Feb 2017 14:29:32 +0100
9ae3a8
Subject: [PATCH 02/17] x86: add AVX512_4VNNIW and AVX512_4FMAPS features
9ae3a8
9ae3a8
RH-Author: Eduardo Habkost <ehabkost@redhat.com>
9ae3a8
Message-id: <20170223142945.17790-2-ehabkost@redhat.com>
9ae3a8
Patchwork-id: 74033
9ae3a8
O-Subject: [RHEL-7.4 qemu-kvm PATCH v2 01/14] x86: add AVX512_4VNNIW and AVX512_4FMAPS features
9ae3a8
Bugzilla: 1382122
9ae3a8
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>
9ae3a8
RH-Acked-by: Igor Mammedov <imammedo@redhat.com>
9ae3a8
RH-Acked-by: Miroslav Rezanina <mrezanin@redhat.com>
9ae3a8
9ae3a8
From: Luwei Kang <luwei.kang@intel.com>
9ae3a8
9ae3a8
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1382122
9ae3a8
9ae3a8
The spec can be found in Intel Software Developer Manual or in
9ae3a8
Instruction Set Extensions Programming Reference.
9ae3a8
9ae3a8
Backport notes:
9ae3a8
9ae3a8
Changes v1 -> v2:
9ae3a8
* Fixed build error, moved feat_names to a separate static array
9ae3a8
  variable
9ae3a8
* Fixed backport mistakes (I had forgotten to add
9ae3a8
  features[FEAT_7_0_EDX] initialization and filtering code)
9ae3a8
9ae3a8
Signed-off-by: Piotr Luc <piotr.luc@intel.com>
9ae3a8
Signed-off-by: Luwei Kang <luwei.kang@intel.com>
9ae3a8
Message-Id: <1477902446-5932-1-git-send-email-he.chen@linux.intel.com>
9ae3a8
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9ae3a8
(cherry picked from commit 95ea69fb46266aaa46d0c8b7f0ba8c4903dbe4e3)
9ae3a8
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
9ae3a8
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
9ae3a8
---
9ae3a8
 target-i386/cpu.c | 31 ++++++++++++++++++++++++++++++-
9ae3a8
 target-i386/cpu.h |  4 ++++
9ae3a8
 2 files changed, 34 insertions(+), 1 deletion(-)
9ae3a8
9ae3a8
diff --git a/target-i386/cpu.c b/target-i386/cpu.c
9ae3a8
index c3c8306..789e687 100644
9ae3a8
--- a/target-i386/cpu.c
9ae3a8
+++ b/target-i386/cpu.c
9ae3a8
@@ -165,6 +165,17 @@ static const char *cpuid_7_0_ecx_feature_name[] = {
9ae3a8
     NULL, NULL, NULL, NULL,
9ae3a8
 };
9ae3a8
 
9ae3a8
+static const char *cpuid_7_0_edx_feature_name[] = {
9ae3a8
+    NULL, NULL, "avx512-4vnniw", "avx512-4fmaps",
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+    NULL, NULL, NULL, NULL,
9ae3a8
+};
9ae3a8
+
9ae3a8
 static const char *cpuid_xsave_feature_name[] = {
9ae3a8
     "xsaveopt", "xsavec", "xgetbv1", NULL,
9ae3a8
     NULL, NULL, NULL, NULL,
9ae3a8
@@ -225,6 +236,12 @@ static FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
9ae3a8
         .cpuid_needs_ecx = true, .cpuid_ecx = 0,
9ae3a8
         .cpuid_reg = R_ECX,
9ae3a8
     },
9ae3a8
+    [FEAT_7_0_EDX] = {
9ae3a8
+        .feat_names = cpuid_7_0_edx_feature_name,
9ae3a8
+        .cpuid_eax = 7,
9ae3a8
+        .cpuid_needs_ecx = true, .cpuid_ecx = 0,
9ae3a8
+        .cpuid_reg = R_EDX,
9ae3a8
+    },
9ae3a8
     [FEAT_XSAVE] = {
9ae3a8
         .feat_names = cpuid_xsave_feature_name,
9ae3a8
         .cpuid_eax = 0xd,
9ae3a8
@@ -484,6 +501,7 @@ typedef struct x86_def_t {
9ae3a8
           CPUID_7_0_EBX_ERMS, CPUID_7_0_EBX_INVPCID, CPUID_7_0_EBX_RTM,
9ae3a8
           CPUID_7_0_EBX_RDSEED */
9ae3a8
 #define TCG_7_0_ECX_FEATURES 0
9ae3a8
+#define TCG_7_0_EDX_FEATURES 0
9ae3a8
 
9ae3a8
 /* built-in CPU model definitions
9ae3a8
  */
9ae3a8
@@ -1254,9 +1272,12 @@ static void kvm_cpu_fill_host(x86_def_t *x86_cpu_def)
9ae3a8
                     kvm_arch_get_supported_cpuid(s, 0x7, 0, R_EBX);
9ae3a8
         x86_cpu_def->features[FEAT_7_0_ECX] =
9ae3a8
                     kvm_arch_get_supported_cpuid(s, 0x7, 0, R_ECX);
9ae3a8
+        x86_cpu_def->features[FEAT_7_0_EDX] =
9ae3a8
+                    kvm_arch_get_supported_cpuid(s, 0x7, 0, R_EDX);
9ae3a8
     } else {
9ae3a8
         x86_cpu_def->features[FEAT_7_0_EBX] = 0;
9ae3a8
         x86_cpu_def->features[FEAT_7_0_ECX] = 0;
9ae3a8
+        x86_cpu_def->features[FEAT_7_0_EDX] = 0;
9ae3a8
     }
9ae3a8
     x86_cpu_def->features[FEAT_XSAVE] =
9ae3a8
                 kvm_arch_get_supported_cpuid(s, 0xd, 1, R_EAX);
9ae3a8
@@ -1343,6 +1364,9 @@ static int kvm_check_features_against_host(X86CPU *cpu)
9ae3a8
         {&env->features[FEAT_7_0_ECX],
9ae3a8
             &host_def.features[FEAT_7_0_ECX],
9ae3a8
             FEAT_7_0_ECX },
9ae3a8
+        {&env->features[FEAT_7_0_EDX],
9ae3a8
+            &host_def.features[FEAT_7_0_EDX],
9ae3a8
+            FEAT_7_0_EDX },
9ae3a8
         {&env->features[FEAT_XSAVE],
9ae3a8
             &host_def.features[FEAT_XSAVE],
9ae3a8
             FEAT_XSAVE },
9ae3a8
@@ -1885,6 +1909,7 @@ static void cpu_x86_parse_featurestr(X86CPU *cpu, char *features, Error **errp)
9ae3a8
     env->features[FEAT_SVM] |= plus_features[FEAT_SVM];
9ae3a8
     env->features[FEAT_7_0_EBX] |= plus_features[FEAT_7_0_EBX];
9ae3a8
     env->features[FEAT_7_0_ECX] |= plus_features[FEAT_7_0_ECX];
9ae3a8
+    env->features[FEAT_7_0_EDX] |= plus_features[FEAT_7_0_EDX];
9ae3a8
     env->features[FEAT_XSAVE] |= plus_features[FEAT_XSAVE];
9ae3a8
     env->features[FEAT_1_EDX] &= ~minus_features[FEAT_1_EDX];
9ae3a8
     env->features[FEAT_1_ECX] &= ~minus_features[FEAT_1_ECX];
9ae3a8
@@ -1895,6 +1920,7 @@ static void cpu_x86_parse_featurestr(X86CPU *cpu, char *features, Error **errp)
9ae3a8
     env->features[FEAT_SVM] &= ~minus_features[FEAT_SVM];
9ae3a8
     env->features[FEAT_7_0_EBX] &= ~minus_features[FEAT_7_0_EBX];
9ae3a8
     env->features[FEAT_7_0_ECX] &= ~minus_features[FEAT_7_0_ECX];
9ae3a8
+    env->features[FEAT_7_0_EDX] &= ~minus_features[FEAT_7_0_EDX];
9ae3a8
     env->features[FEAT_XSAVE] &= ~minus_features[FEAT_XSAVE];
9ae3a8
 
9ae3a8
 out:
9ae3a8
@@ -2032,6 +2058,7 @@ static void cpu_x86_register(X86CPU *cpu, const char *name, Error **errp)
9ae3a8
     env->features[FEAT_C000_0001_EDX] = def->features[FEAT_C000_0001_EDX];
9ae3a8
     env->features[FEAT_7_0_EBX] = def->features[FEAT_7_0_EBX];
9ae3a8
     env->features[FEAT_7_0_ECX] = def->features[FEAT_7_0_ECX];
9ae3a8
+    env->features[FEAT_7_0_EDX] = def->features[FEAT_7_0_EDX];
9ae3a8
     env->features[FEAT_XSAVE] = def->features[FEAT_XSAVE];
9ae3a8
     env->cpuid_xlevel2 = def->xlevel2;
9ae3a8
 
9ae3a8
@@ -2270,7 +2297,7 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
9ae3a8
             *eax = 0; /* Maximum ECX value for sub-leaves */
9ae3a8
             *ebx = env->features[FEAT_7_0_EBX]; /* Feature flags */
9ae3a8
             *ecx = env->features[FEAT_7_0_ECX]; /* Feature flags */
9ae3a8
-            *edx = 0; /* Reserved */
9ae3a8
+            *edx = env->features[FEAT_7_0_EDX]; /* Feature flags */
9ae3a8
         } else {
9ae3a8
             *eax = 0;
9ae3a8
             *ebx = 0;
9ae3a8
@@ -2680,6 +2707,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)
9ae3a8
         env->features[FEAT_8000_0001_ECX] &= TCG_EXT3_FEATURES;
9ae3a8
         env->features[FEAT_SVM] &= TCG_SVM_FEATURES;
9ae3a8
         env->features[FEAT_XSAVE] = 0;
9ae3a8
+        env->features[FEAT_7_0_ECX] &= TCG_7_0_ECX_FEATURES;
9ae3a8
+        env->features[FEAT_7_0_EDX] &= TCG_7_0_EDX_FEATURES;
9ae3a8
     } else {
9ae3a8
         if ((cpu->check_cpuid || cpu->enforce_cpuid)
9ae3a8
             && kvm_check_features_against_host(cpu) && cpu->enforce_cpuid) {
9ae3a8
diff --git a/target-i386/cpu.h b/target-i386/cpu.h
9ae3a8
index d541809..eec5c49 100644
9ae3a8
--- a/target-i386/cpu.h
9ae3a8
+++ b/target-i386/cpu.h
9ae3a8
@@ -401,6 +401,7 @@ typedef enum FeatureWord {
9ae3a8
     FEAT_1_ECX,         /* CPUID[1].ECX */
9ae3a8
     FEAT_7_0_EBX,       /* CPUID[EAX=7,ECX=0].EBX */
9ae3a8
     FEAT_7_0_ECX,       /* CPUID[EAX=7,ECX=0].ECX */
9ae3a8
+    FEAT_7_0_EDX,       /* CPUID[EAX=7,ECX=0].EDX */
9ae3a8
     FEAT_8000_0001_EDX, /* CPUID[8000_0001].EDX */
9ae3a8
     FEAT_8000_0001_ECX, /* CPUID[8000_0001].ECX */
9ae3a8
     FEAT_C000_0001_EDX, /* CPUID[C000_0001].EDX */
9ae3a8
@@ -580,6 +581,9 @@ typedef uint32_t FeatureWordArray[FEATURE_WORDS];
9ae3a8
 #define CPUID_7_0_ECX_OSPKE    (1U << 4)
9ae3a8
 #define CPUID_7_0_ECX_RDPID    (1U << 22)
9ae3a8
 
9ae3a8
+#define CPUID_7_0_EDX_AVX512_4VNNIW (1U << 2) /* AVX512 Neural Network Instructions */
9ae3a8
+#define CPUID_7_0_EDX_AVX512_4FMAPS (1U << 3) /* AVX512 Multiply Accumulation Single Precision */
9ae3a8
+
9ae3a8
 #define CPUID_XSAVE_XSAVEOPT   (1U << 0)
9ae3a8
 #define CPUID_XSAVE_XSAVEC     (1U << 1)
9ae3a8
 #define CPUID_XSAVE_XGETBV1    (1U << 2)
9ae3a8
-- 
9ae3a8
1.8.3.1
9ae3a8