Skip to contentSkip to conversation

i386: Minor fixes of building CPUIDs

PATCH Superseded Xiaoyao Li 2 participants 7 messages Jun 13, 2023 – Jun 27, 2023 KVM

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Xiaoyao Li

author
#0 lore raw

v2 cover i386: Minor fixes of building CPUIDs

This v2 adds patch 3 to fix the build of CPUID leaf 7.

The issue that fixed by Patch 1 looks fatal though it doesn't appear on KVM because KVM always searches with assending order and hit with the correct cpuid leaf 0.

Patch 2 removes the wrong constraint on CPUID leaf 1f.

Changes in v2:
- Add Patch 3;
- rebase to latest master branch
v1: https://lore.kernel.org/qemu-devel/[email protected]/

Xiaoyao Li (3):
i386/cpuid: Decrease cpuid_i when skipping CPUID leaf 1F i386/cpuid: Remove subleaf constraint on CPUID leaf 1F i386/cpuid: Move leaf 7 to correct group

 target/i386/kvm/kvm.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

base-commit: fdd0df5340a8ebc8de88078387ebc85c5af7b40f

Signature · 1 line
2.34.1
author
#1 lore raw

PATCH v2 1/3 i386/cpuid: Decrease cpuid_i when skipping CPUID leaf 1F

Decrease array index cpuid_i when CPUID leaf 1F is skipped, otherwise it will get an all zero'ed CPUID entry with leaf 0 and subleaf 0. It conflicts with correct leaf 0.

---
 target/i386/kvm/kvm.c | 1 +
 1 file changed, 1 insertion(+)
target/i386/kvm/kvm.c +1 -0
index de531842f6b1..afa97799d89a 100644
@@ -1956,6 +1956,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
         }
         case 0x1f:
             if (env->nr_dies < 2) {
+                cpuid_i--;
                 break;
             }
             /* fallthrough */
Signature · 1 line
2.34.1
Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH v2 1/3] i386/cpuid: Decrease cpuid_i when skipping CPUID leaf 1F

Quoted 3 lines · Xiaoyao Li

Decrease array index cpuid_i when CPUID leaf 1F is skipped, otherwise it will get an all zero'ed CPUID entry with leaf 0 and subleaf 0. It conflicts with correct leaf 0.

Maybe change the commit log like this:

Exiting code misses a decrement of cpuid_i when skip left 0x1F, so there's a blank CPUID

entry(with all fields stuffed 0s) left in the CPUID array.  Fix the issue to avoid the blank slot.

Quoted 18 lines
---
  target/i386/kvm/kvm.c | 1 +
  1 file changed, 1 insertion(+)
target/i386/kvm/kvm.c +1 -0
index de531842f6b1..afa97799d89a 100644
@@ -1956,6 +1956,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
          }
          case 0x1f:
              if (env->nr_dies < 2) {
+                cpuid_i--;
                  break;
              }
              /* fallthrough */
author
#3 lore raw

PATCH v2 2/3 i386/cpuid: Remove subleaf constraint on CPUID leaf 1F

In reply to Xiaoyao Li

No such constraint that subleaf index needs to be less than 64.

---
 target/i386/kvm/kvm.c | 4 ----
 1 file changed, 4 deletions(-)
target/i386/kvm/kvm.c +0 -4
index afa97799d89a..d7e235ce35a6 100644
@@ -1968,10 +1968,6 @@ int kvm_arch_init_vcpu(CPUState *cs)
                     break;
                 }
 
-                if (i == 0x1f && j == 64) {
-                    break;
-                }
-
                 c->function = i;
                 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
                 c->index = j;
Signature · 1 line
2.34.1
Reviewed-by on v2 2/3
#4 lore raw

Re: [PATCH v2 2/3] i386/cpuid: Remove subleaf constraint on CPUID leaf 1F

Quoted 12 lines · Xiaoyao Li

No such constraint that subleaf index needs to be less than 64.

---
  target/i386/kvm/kvm.c | 4 ----
  1 file changed, 4 deletions(-)
target/i386/kvm/kvm.c +0 -0
index afa97799d89a..d7e235ce35a6 100644

target/i386/kvm/kvm.c int kvm_arch_init_vcpu(CPUState *cs) · line 1973

Quoted 10 lines
target/i386/kvm/kvm.c +0 -4
@@ -1968,10 +1968,6 @@ int kvm_arch_init_vcpu(CPUState *cs)
                      break;
                  }
 
-                if (i == 0x1f && j == 64) {
-                    break;
-                }
-
                  c->function = i;
                  c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
                  c->index = j;
author
#5 lore raw

PATCH v2 3/3 i386/cpuid: Move leaf 7 to correct group

In reply to Xiaoyao Li

CPUID leaf 7 was grouped together with SGX leaf 0x12 by commit b9edbadefb9e ("i386: Propagate SGX CPUID sub-leafs to KVM") by mistake.

SGX leaf 0x12 has its specific logic to check if subleaf (starting from 2) is valid or not by checking the bit 0:3 of corresponding EAX is 1 or not.

Leaf 7 follows the logic that EAX of subleaf 0 enumerates the maximum valid subleaf.

Fixes b9edbadefb9e ("i386: Propagate SGX CPUID sub-leafs to KVM")
---
 target/i386/kvm/kvm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/i386/kvm/kvm.c +1 -1
index d7e235ce35a6..86aab9ca4ba2 100644
@@ -1993,7 +1993,6 @@ int kvm_arch_init_vcpu(CPUState *cs)
                 c = &cpuid_data.entries[cpuid_i++];
             }
             break;
-        case 0x7:
         case 0x12:
             for (j = 0; ; j++) {
                 c->function = i;
@@ -2013,6 +2012,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
                 c = &cpuid_data.entries[cpuid_i++];
             }
             break;
+        case 0x7:
         case 0x14:
         case 0x1d:
         case 0x1e: {
Signature · 1 line
2.34.1
Reviewed-by on v2 3/3
#6 lore raw

Re: [PATCH v2 3/3] i386/cpuid: Move leaf 7 to correct group

Quoted 29 lines · Xiaoyao Li

CPUID leaf 7 was grouped together with SGX leaf 0x12 by commit b9edbadefb9e ("i386: Propagate SGX CPUID sub-leafs to KVM") by mistake.

SGX leaf 0x12 has its specific logic to check if subleaf (starting from 2) is valid or not by checking the bit 0:3 of corresponding EAX is 1 or not.

Leaf 7 follows the logic that EAX of subleaf 0 enumerates the maximum valid subleaf.

Fixes b9edbadefb9e ("i386: Propagate SGX CPUID sub-leafs to KVM")
---
  target/i386/kvm/kvm.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
target/i386/kvm/kvm.c +0 -1
index d7e235ce35a6..86aab9ca4ba2 100644
@@ -1993,7 +1993,6 @@ int kvm_arch_init_vcpu(CPUState *cs)
                  c = &cpuid_data.entries[cpuid_i++];
              }
              break;
-        case 0x7:
          case 0x12:
              for (j = 0; ; j++) {
                  c->function = i;

target/i386/kvm/kvm.c int kvm_arch_init_vcpu(CPUState *cs) · line 2018

Quoted 7 lines
target/i386/kvm/kvm.c +1 -0
@@ -2013,6 +2012,7 @@ int kvm_arch_init_vcpu(CPUState *cs)
                  c = &cpuid_data.entries[cpuid_i++];
              }
              break;
+        case 0x7:
          case 0x14:
          case 0x1d:
          case 0x1e: {