Skip to contentSkip to conversation

target/i386: Add the immediate form MSR access instruction support

PATCH In review Xin Li (Intel) 3 participants 12 messages Jan 3, 2025 – May 29, 2025

lore
Share & tools
lore
#0 lore raw

cover target/i386: Add the immediate form MSR access instruction support

The immediate form of MSR access instructions are primarily motivated by performance, not code size: by having the MSR number in an immediate, it is available *much* earlier in the pipeline, which allows the hardware much more leeway about how a particular MSR is handled.

This new CPU feature is advertised through bit 5 of CPUID.7.1.ECX, which needs to be added as a new CPU feature word.

WRMSRNS doesn't become a required feature for FERD, and Linux has removed the dependency, as such remove the dependency from Qemu.

Xin Li (Intel) (3):
target/i386: Remove FRED dependency on WRMSRNS target/i386: Add a new CPU feature word for CPUID.7.1.ECX target/i386: Add the immediate form MSR access instruction support

 target/i386/cpu.c | 27 ++++++++++++++++++++++-----
 target/i386/cpu.h |  4 ++++
 2 files changed, 26 insertions(+), 5 deletions(-)

base-commit: 1ada452efc7d8f8bf42cd5e8a2af1b4ac9167a1f

Signature · 1 line
2.47.1
#1 lore raw

PATCH 1/3 target/i386: Remove FRED dependency on WRMSRNS

WRMSRNS doesn't become a required feature for FERD, and Linux has removed the dependency, as such remove it from Qemu.

---
 target/i386/cpu.c | 4 ----
 1 file changed, 4 deletions(-)
target/i386/cpu.c +0 -4
index 0b639848cd..8a1223acb3 100644
@@ -1771,10 +1771,6 @@ static FeatureDep feature_dependencies[] = {
         .from = { FEAT_7_1_EAX,             CPUID_7_1_EAX_LKGS },
         .to = { FEAT_7_1_EAX,               CPUID_7_1_EAX_FRED },
     },
-    {
-        .from = { FEAT_7_1_EAX,             CPUID_7_1_EAX_WRMSRNS },
-        .to = { FEAT_7_1_EAX,               CPUID_7_1_EAX_FRED },
-    },
     {
         .from = { FEAT_7_0_EBX,             CPUID_7_0_EBX_SGX },
         .to = { FEAT_7_0_ECX,               CPUID_7_0_ECX_SGX_LC },
Signature · 1 line
2.47.1
Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH v1 1/3] target/i386: Remove FRED dependency on WRMSRNS

Quoted 4 lines · Xin Li (Intel)

WRMSRNS doesn't become a required feature for FERD, and Linux has removed the dependency, as such remove it from Qemu.

Quoted 19 lines
---
  target/i386/cpu.c | 4 ----
  1 file changed, 4 deletions(-)
target/i386/cpu.c +0 -4
index 0b639848cd..8a1223acb3 100644
@@ -1771,10 +1771,6 @@ static FeatureDep feature_dependencies[] = {
          .from = { FEAT_7_1_EAX,             CPUID_7_1_EAX_LKGS },
          .to = { FEAT_7_1_EAX,               CPUID_7_1_EAX_FRED },
      },
-    {
-        .from = { FEAT_7_1_EAX,             CPUID_7_1_EAX_WRMSRNS },
-        .to = { FEAT_7_1_EAX,               CPUID_7_1_EAX_FRED },
-    },
      {
          .from = { FEAT_7_0_EBX,             CPUID_7_0_EBX_SGX },
          .to = { FEAT_7_0_ECX,               CPUID_7_0_ECX_SGX_LC },
#3 lore raw

PATCH 2/3 target/i386: Add a new CPU feature word for CPUID.7.1.ECX

In reply to Xin Li (Intel)

The immediate form of MSR access instructions will use this new CPU feature word.

---
 target/i386/cpu.c | 23 ++++++++++++++++++++++-
 target/i386/cpu.h |  1 +
 2 files changed, 23 insertions(+), 1 deletion(-)
target/i386/cpu.c +22 -1
index 8a1223acb3..2fb05879c3 100644
@@ -894,6 +894,7 @@ void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,
 
 #define TCG_7_1_EAX_FEATURES (CPUID_7_1_EAX_FZRM | CPUID_7_1_EAX_FSRS | \
           CPUID_7_1_EAX_FSRC | CPUID_7_1_EAX_CMPCCXADD)
+#define TCG_7_1_ECX_FEATURES 0
 #define TCG_7_1_EDX_FEATURES 0
 #define TCG_7_2_EDX_FEATURES 0
 #define TCG_APM_FEATURES 0
@@ -1133,6 +1134,25 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
         },
         .tcg_features = TCG_7_1_EAX_FEATURES,
     },
+    [FEAT_7_1_ECX] = {
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+        },
+        .cpuid = {
+            .eax = 7,
+            .needs_ecx = true, .ecx = 1,
+            .reg = R_ECX,
+        },
+        .tcg_features = TCG_7_1_ECX_FEATURES,
+    },
     [FEAT_7_1_EDX] = {
         .type = CPUID_FEATURE_WORD,
         .feat_names = {
@@ -6659,9 +6679,9 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
             *edx = env->features[FEAT_7_0_EDX]; /* Feature flags */
         } else if (count == 1) {
             *eax = env->features[FEAT_7_1_EAX];
+            *ecx = env->features[FEAT_7_1_ECX];
             *edx = env->features[FEAT_7_1_EDX];
             *ebx = 0;
-            *ecx = 0;
         } else if (count == 2) {
             *edx = env->features[FEAT_7_2_EDX];
             *eax = 0;
@@ -7563,6 +7583,7 @@ void x86_cpu_expand_features(X86CPU *cpu, Error **errp)
         x86_cpu_adjust_feat_level(cpu, FEAT_6_EAX);
         x86_cpu_adjust_feat_level(cpu, FEAT_7_0_ECX);
         x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EAX);
+        x86_cpu_adjust_feat_level(cpu, FEAT_7_1_ECX);
         x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EDX);
         x86_cpu_adjust_feat_level(cpu, FEAT_7_2_EDX);
         x86_cpu_adjust_feat_level(cpu, FEAT_8000_0001_EDX);
target/i386/cpu.h +1 -0
index dbd8f1ffc7..d23fa96549 100644
@@ -667,6 +667,7 @@ typedef enum FeatureWord {
     FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */
     FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */
     FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */
+    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */
     FEATURE_WORDS,
 } FeatureWord;
 
Signature · 1 line
2.47.1
#4 lore raw

Re: [PATCH v1 2/3] target/i386: Add a new CPU feature word for CPUID.7.1.ECX

Quoted 26 lines · Xin Li (Intel)

The immediate form of MSR access instructions will use this new CPU feature word.

---
  target/i386/cpu.c | 23 ++++++++++++++++++++++-
  target/i386/cpu.h |  1 +
  2 files changed, 23 insertions(+), 1 deletion(-)
target/i386/cpu.c +2 -0
index 8a1223acb3..2fb05879c3 100644
@@ -894,6 +894,7 @@ void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,
 
  #define TCG_7_1_EAX_FEATURES (CPUID_7_1_EAX_FZRM | CPUID_7_1_EAX_FSRS | \
            CPUID_7_1_EAX_FSRC | CPUID_7_1_EAX_CMPCCXADD)
+#define TCG_7_1_ECX_FEATURES 0
  #define TCG_7_1_EDX_FEATURES 0
  #define TCG_7_2_EDX_FEATURES 0
  #define TCG_APM_FEATURES 0
@@ -1133,6 +1134,25 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
          },
          .tcg_features = TCG_7_1_EAX_FEATURES,
      },
+    [FEAT_7_1_ECX] = {

target/i386/cpu.c FeatureWordInfo feature_word_info[FEATURE_WORDS] = { · line 1147

Quoted 10 lines
target/i386/cpu.c +10 -0
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,

This looks silly, and the size of feat_names[] was changed from 32 to 64. Just explicitly assign the first 32 entries with NULL doesn't make any sense after the size change.

We can just merge the next patch into this one and make it,

	.feat_names = {
	    [5] = "msr-imm",
	},
Quoted 35 lines
target/i386/cpu.c +10 -1
+        },
+        .cpuid = {
+            .eax = 7,
+            .needs_ecx = true, .ecx = 1,
+            .reg = R_ECX,
+        },
+        .tcg_features = TCG_7_1_ECX_FEATURES,
+    },
      [FEAT_7_1_EDX] = {
          .type = CPUID_FEATURE_WORD,
          .feat_names = {
@@ -6659,9 +6679,9 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
              *edx = env->features[FEAT_7_0_EDX]; /* Feature flags */
          } else if (count == 1) {
              *eax = env->features[FEAT_7_1_EAX];
+            *ecx = env->features[FEAT_7_1_ECX];
              *edx = env->features[FEAT_7_1_EDX];
              *ebx = 0;
-            *ecx = 0;
          } else if (count == 2) {
              *edx = env->features[FEAT_7_2_EDX];
              *eax = 0;
@@ -7563,6 +7583,7 @@ void x86_cpu_expand_features(X86CPU *cpu, Error **errp)
          x86_cpu_adjust_feat_level(cpu, FEAT_6_EAX);
          x86_cpu_adjust_feat_level(cpu, FEAT_7_0_ECX);
          x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EAX);
+        x86_cpu_adjust_feat_level(cpu, FEAT_7_1_ECX);
          x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EDX);
          x86_cpu_adjust_feat_level(cpu, FEAT_7_2_EDX);
          x86_cpu_adjust_feat_level(cpu, FEAT_8000_0001_EDX);
target/i386/cpu.h +0 -0
index dbd8f1ffc7..d23fa96549 100644

target/i386/cpu.h typedef enum FeatureWord { · line 670

Quoted 4 lines
target/i386/cpu.h +1 -0
@@ -667,6 +667,7 @@ typedef enum FeatureWord {
      FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */
      FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */
      FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */
+    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */

I would prefer the newly added leaf being ordered at least in the same leaf. i.e.,

@@ -661,6 +661,7 @@ typedef enum FeatureWord {
     FEAT_SGX_12_1_EAX,  /* CPUID[EAX=0x12,ECX=1].EAX (SGX 
ATTRIBUTES[31:0]) */
     FEAT_XSAVE_XSS_LO,     /* CPUID[EAX=0xd,ECX=1].ECX */
     FEAT_XSAVE_XSS_HI,     /* CPUID[EAX=0xd,ECX=1].EDX */
+    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */
     FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */
     FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */
     FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */

They are QEMU internally data, injecting a new one instead of putting it at the end is OK to me.

Quoted 3 lines
target/i386/cpu.h +0 -0
      FEATURE_WORDS,
  } FeatureWord;
 
author
#5 lore raw

Re: [PATCH v1 2/3] target/i386: Add a new CPU feature word for CPUID.7.1.ECX

Quoted 20 lines · Xiaoyao Li
Quoted 6 lines · Xin Li (Intel)
target/i386/cpu.c +1 -0
@@ -1133,6 +1134,25 @@ FeatureWordInfo 
 feature_word_info[FEATURE_WORDS] = {
           },
           .tcg_features = TCG_7_1_EAX_FEATURES,
       },
+    [FEAT_7_1_ECX] = {

target/i386/cpu.c FeatureWordInfo · line 1148

Quoted 10 lines
target/i386/cpu.c +10 -0
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,

This looks silly, and the size of feat_names[] was changed from 32 to 64. Just explicitly assign the first 32 entries with NULL doesn't make any sense after the size change.

Quoted 5 lines

We can just merge the next patch into this one and make it,

    .feat_names = {
        [5] = "msr-imm",
    },

I appreciate this format that clearly indicates which bit corresponds to which feature, and I'm inclined to proceed with the change.

On the flip side, I hope the new format won't be perceived as disrupting
the consistency of the existing codebase. If this form could get called out by maintainers, there needs a cleanup patch to change all the instances.

Quoted 22 lines
Quoted 9 lines
target/i386/cpu.h +0 -0
index dbd8f1ffc7..d23fa96549 100644
@@ -667,6 +667,7 @@ typedef enum FeatureWord {

      FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */ FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */ FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */ +    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */

I would prefer the newly added leaf being ordered at least in the same leaf. i.e.,

(quoted diff) +0 -0
@@ -661,6 +661,7 @@ typedef enum FeatureWord {
      FEAT_SGX_12_1_EAX,  /* CPUID[EAX=0x12,ECX=1].EAX (SGX 

ATTRIBUTES[31:0]) */
     FEAT_XSAVE_XSS_LO,     /* CPUID[EAX=0xd,ECX=1].ECX */ FEAT_XSAVE_XSS_HI,     /* CPUID[EAX=0xd,ECX=1].EDX */
+    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */
     FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */ FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */ FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */

Quoted 1 line

They are QEMU internally data, injecting a new one instead of putting it at the end is OK to me.

Makes sense to me. Will move FEAT_7_1_ECX, FEAT_7_1_EDX and FEAT_7_2_EDX immediate after FEAT_7_1_EAX.

Thanks!
Xin

#6 lore raw

Re: [PATCH v1 2/3] target/i386: Add a new CPU feature word for CPUID.7.1.ECX

In reply to Xiaoyao Li
Quoted 32 lines · Xiaoyao Li
Quoted 31 lines · Xin Li (Intel)

The immediate form of MSR access instructions will use this new CPU feature word.

---
  target/i386/cpu.c | 23 ++++++++++++++++++++++-
  target/i386/cpu.h |  1 +

  2 files changed, 23 insertions(+), 1 deletion(-)
target/i386/cpu.c +6 -0
index 8a1223acb3..2fb05879c3 100644
@@ -894,6 +894,7 @@ void x86_cpu_vendor_words2str(char *dst, uint32_t 
 vendor1,
   #define TCG_7_1_EAX_FEATURES (CPUID_7_1_EAX_FZRM | CPUID_7_1_EAX_FSRS | \
             CPUID_7_1_EAX_FSRC | CPUID_7_1_EAX_CMPCCXADD)
+#define TCG_7_1_ECX_FEATURES 0
   #define TCG_7_1_EDX_FEATURES 0
   #define TCG_7_2_EDX_FEATURES 0
   #define TCG_APM_FEATURES 0
@@ -1133,6 +1134,25 @@ FeatureWordInfo 
 feature_word_info[FEATURE_WORDS] = {
           },
           .tcg_features = TCG_7_1_EAX_FEATURES,
       },
+    [FEAT_7_1_ECX] = {
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
Quoted 8 lines

target/i386/cpu.c FeatureWordInfo feature_word_info[FEATURE_WORDS] = { · line 1145

Quoted 6 lines
target/i386/cpu.c +6 -0
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,

This looks silly, and the size of feat_names[] was changed from 32 to 64. Just explicitly assign the first 32 entries with NULL doesn't make any sense after the size change.

64 is just for MSR features. This is a bit silly, I agree, but it is consistent with existing feature words and ultimately it becomes more compact after just 9 features. So I'm queuing Xin's patches as they are.

Thanks for the review though! It's always appreciated even if we disagree.

Paolo

Quoted 68 lines

We can just merge the next patch into this one and make it,

    .feat_names = {
        [5] = "msr-imm",
    },

Quoted 37 lines
target/i386/cpu.c +8 -0
+        },
+        .cpuid = {
+            .eax = 7,
+            .needs_ecx = true, .ecx = 1,
+            .reg = R_ECX,
+        },
+        .tcg_features = TCG_7_1_ECX_FEATURES,
+    },
       [FEAT_7_1_EDX] = {
           .type = CPUID_FEATURE_WORD,

          .feat_names = {

target/i386/cpu.c +1 -1
@@ -6659,9 +6679,9 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t 
 index, uint32_t count,
               *edx = env->features[FEAT_7_0_EDX]; /* Feature flags */
           } else if (count == 1) {
               *eax = env->features[FEAT_7_1_EAX];
+            *ecx = env->features[FEAT_7_1_ECX];
               *edx = env->features[FEAT_7_1_EDX];
               *ebx = 0;
-            *ecx = 0;
           } else if (count == 2) {
               *edx = env->features[FEAT_7_2_EDX];

              *eax = 0;

target/i386/cpu.c +1 -0
@@ -7563,6 +7583,7 @@ void x86_cpu_expand_features(X86CPU *cpu, Error 
 **errp)
           x86_cpu_adjust_feat_level(cpu, FEAT_6_EAX);
           x86_cpu_adjust_feat_level(cpu, FEAT_7_0_ECX);
           x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EAX);
+        x86_cpu_adjust_feat_level(cpu, FEAT_7_1_ECX);
           x86_cpu_adjust_feat_level(cpu, FEAT_7_1_EDX);
           x86_cpu_adjust_feat_level(cpu, FEAT_7_2_EDX);

          x86_cpu_adjust_feat_level(cpu, FEAT_8000_0001_EDX);

target/i386/cpu.h +0 -0
index dbd8f1ffc7..d23fa96549 100644
@@ -667,6 +667,7 @@ typedef enum FeatureWord {
Quoted 4 lines

      FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */ FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */ FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */ +    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */

I would prefer the newly added leaf being ordered at least in the same leaf. i.e.,

(quoted diff) +0 -0
@@ -661,6 +661,7 @@ typedef enum FeatureWord {
      FEAT_SGX_12_1_EAX,  /* CPUID[EAX=0x12,ECX=1].EAX (SGX 

ATTRIBUTES[31:0]) */
     FEAT_XSAVE_XSS_LO,     /* CPUID[EAX=0xd,ECX=1].ECX */ FEAT_XSAVE_XSS_HI,     /* CPUID[EAX=0xd,ECX=1].EDX */
+    FEAT_7_1_ECX,       /* CPUID[EAX=7,ECX=1].ECX */
     FEAT_7_1_EDX,       /* CPUID[EAX=7,ECX=1].EDX */ FEAT_7_2_EDX,       /* CPUID[EAX=7,ECX=2].EDX */ FEAT_24_0_EBX,      /* CPUID[EAX=0x24,ECX=0].EBX */

They are QEMU internally data, injecting a new one instead of putting it at the end is OK to me.

Quoted 2 lines

      FEATURE_WORDS,
  } FeatureWord;

#7 lore raw

Re: [PATCH v1 2/3] target/i386: Add a new CPU feature word for CPUID.7.1.ECX

Quoted 42 lines · Paolo Bonzini
Quoted 32 lines · Xiaoyao Li
Quoted 31 lines · Xin Li (Intel)

The immediate form of MSR access instructions will use this new CPU feature word.

---
  target/i386/cpu.c | 23 ++++++++++++++++++++++-
  target/i386/cpu.h |  1 +

  2 files changed, 23 insertions(+), 1 deletion(-)
target/i386/cpu.c +6 -0
index 8a1223acb3..2fb05879c3 100644
@@ -894,6 +894,7 @@ void x86_cpu_vendor_words2str(char *dst, uint32_t 
 vendor1,
   #define TCG_7_1_EAX_FEATURES (CPUID_7_1_EAX_FZRM | CPUID_7_1_EAX_FSRS | \
             CPUID_7_1_EAX_FSRC | CPUID_7_1_EAX_CMPCCXADD)
+#define TCG_7_1_ECX_FEATURES 0
   #define TCG_7_1_EDX_FEATURES 0
   #define TCG_7_2_EDX_FEATURES 0
   #define TCG_APM_FEATURES 0
@@ -1133,6 +1134,25 @@ FeatureWordInfo 
 feature_word_info[FEATURE_WORDS] = {
           },
           .tcg_features = TCG_7_1_EAX_FEATURES,
       },
+    [FEAT_7_1_ECX] = {
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
Quoted 8 lines

target/i386/cpu.c FeatureWordInfo feature_word_info[FEATURE_WORDS] = { · line 1145

Quoted 6 lines
target/i386/cpu.c +6 -0
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,

This looks silly, and the size of feat_names[] was changed from 32 to 64. Just explicitly assign the first 32 entries with NULL doesn't make any sense after the size change.

Quoted 1 line

64 is just for MSR features.  This is a bit silly, I agree, but it is consistent with existing feature words and ultimately it becomes more compact after just 9 features.  So I'm queuing Xin's patches as they are.

Yes. It makes sense for this reason, especially that this leaf is general feature enumeration leaf and destined to be filled up in the future.

Quoted 1 line

Thanks for the review though!  It's always appreciated even if we disagree.

My pleasure.

author
#8 lore raw

Re: [PATCH v1 2/3] target/i386: Add a new CPU feature word for CPUID.7.1.ECX

In reply to Paolo Bonzini
Quoted 21 lines · Paolo Bonzini
Quoted 11 lines · Xiaoyao Li
Quoted 10 lines · Xin Li (Intel)
target/i386/cpu.c +5 -0
@@ -1133,6 +1134,25 @@ FeatureWordInfo 
 feature_word_info[FEATURE_WORDS] = {
           },
           .tcg_features = TCG_7_1_EAX_FEATURES,
       },
+    [FEAT_7_1_ECX] = {
+        .type = CPUID_FEATURE_WORD,
+        .feat_names = {
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
Quoted 8 lines

target/i386/cpu.c FeatureWordInfo feature_word_info[FEATURE_WORDS] = { · line 1145

Quoted 6 lines
target/i386/cpu.c +6 -0
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,
+            NULL, NULL, NULL, NULL,

This looks silly, and the size of feat_names[] was changed from 32 to 64. Just explicitly assign the first 32 entries with NULL doesn't make any sense after the size change.

Quoted 2 lines

64 is just for MSR features.  This is a bit silly, I agree, but it is consistent with existing feature words and ultimately it becomes more compact after just 9 features.  So I'm queuing Xin's patches as they are.

Paolo, thanks a lot!
Xin

#9 lore raw

PATCH 3/3 target/i386: Add the immediate form MSR access instruction support

In reply to Xin Li (Intel)

The immediate form of MSR access instructions are primarily motivated by performance, not code size: by having the MSR number in an immediate, it is available *much* earlier in the pipeline, which allows the hardware much more leeway about how a particular MSR is handled.

---
 target/i386/cpu.c | 2 +-
 target/i386/cpu.h | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
target/i386/cpu.c +1 -1
index 2fb05879c3..83fb059d09 100644
@@ -1138,7 +1138,7 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
         .type = CPUID_FEATURE_WORD,
         .feat_names = {
             NULL, NULL, NULL, NULL,
-            NULL, NULL, NULL, NULL,
+            NULL, "msr-imm", NULL, NULL,
             NULL, NULL, NULL, NULL,
             NULL, NULL, NULL, NULL,
             NULL, NULL, NULL, NULL,
target/i386/cpu.h +3 -0
index d23fa96549..df27c359d3 100644
@@ -987,6 +987,9 @@ uint64_t x86_cpu_get_supported_feature_word(X86CPU *cpu, FeatureWord w);
 /* Non-Serializing Write to Model Specific Register (WRMSRNS) */
 #define CPUID_7_1_EAX_WRMSRNS           (1U << 19)
 
+/* The immediate form of MSR access instructions */
+#define CPUID_7_1_ECX_MSR_IMM           (1U << 5)
+
 /* Do not exhibit MXCSR Configuration Dependent Timing (MCDT) behavior */
 #define CPUID_7_2_EDX_MCDT_NO           (1U << 5)
 
Signature · 1 line
2.47.1
author
#10 lore raw
In reply to Xin Li (Intel)
Quoted 7 lines · Xin Li (Intel)

The immediate form of MSR access instructions are primarily motivated by performance, not code size: by having the MSR number in an immediate, it is available *much* earlier in the pipeline, which allows the hardware much more leeway about how a particular MSR is handled.

This new CPU feature is advertised through bit 5 of CPUID.7.1.ECX, which needs to be added as a new CPU feature word.

gentle ping!

Quoted 4 lines

WRMSRNS doesn't become a required feature for FERD, and Linux has removed the dependency, as such remove the dependency from Qemu.

Maybe this should be sent out as a separate patch?

Quoted 12 lines

Xin Li (Intel) (3):
target/i386: Remove FRED dependency on WRMSRNS target/i386: Add a new CPU feature word for CPUID.7.1.ECX target/i386: Add the immediate form MSR access instruction support

  target/i386/cpu.c | 27 ++++++++++++++++++++++-----
  target/i386/cpu.h |  4 ++++
  2 files changed, 26 insertions(+), 5 deletions(-)

base-commit: 1ada452efc7d8f8bf42cd5e8a2af1b4ac9167a1f

author
#11 lore raw
Quoted 9 lines · Xin Li
Quoted 7 lines · Xin Li (Intel)

The immediate form of MSR access instructions are primarily motivated by performance, not code size: by having the MSR number in an immediate, it is available *much* earlier in the pipeline, which allows the hardware much more leeway about how a particular MSR is handled.

This new CPU feature is advertised through bit 5 of CPUID.7.1.ECX, which needs to be added as a new CPU feature word.

Quoted 21 lines

gentle ping!

Quoted 4 lines

WRMSRNS doesn't become a required feature for FERD, and Linux has removed the dependency, as such remove the dependency from Qemu.

Maybe this should be sent out as a separate patch?

Quoted 12 lines

Xin Li (Intel) (3):
   target/i386: Remove FRED dependency on WRMSRNS target/i386: Add a new CPU feature word for CPUID.7.1.ECX target/i386: Add the immediate form MSR access instruction support

  target/i386/cpu.c | 27 ++++++++++++++++++++++----- target/i386/cpu.h |  4 ++++

  2 files changed, 26 insertions(+), 5 deletions(-)

base-commit: 1ada452efc7d8f8bf42cd5e8a2af1b4ac9167a1f

Another ping :)