Skip to contentSkip to conversation

target/i386: Update EPYC CPU models for Cache property, RAS, SVM feature bits

PATCH Superseded Babu Moger 2 participants 7 messages Nov 14, 2024 – Nov 15, 2024 KVM

v4 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Babu Moger

author
#0 lore raw

v4 cover target/i386: Update EPYC CPU models for Cache property, RAS, SVM feature bits

This series addresses the following issues with EPYC CPU models.

1. Update the L1, L2, L3 cache properties to match the actual hardware (PPR). This needs to be updated on all the EPYC models.
2. RAS feature bits (SUCCOR, McaOverflowRecov).
3. Add SVM feature bits which are required in nested guests.
4. Add perfmon-v2 on Genoa.
5. Add missing feature bit fs-gs-base-ns(WRMSR to {FS,GS,KERNEL_G}S_BASE is non-serializing).

Dropped EPYC-Turin model for now. Some of the feature bits (srso-user-kernel-no, eraps, rapsize) are still work in progress in the kernel. Will post them later.

---
v4: Some of the patches in v3 are already merged. Posting the rest of the patches.
Dropped EPYC-Turin model for now. Will post them later.
Added SVM feature bit as discussed in
https://lore.kernel.org/kvm/[email protected]/
Fixed the cache property details as discussed in
https://lore.kernel.org/kvm/[email protected]/
Thanks to Maksim and Paolo for their feedback.

v3: Added SBPB, IBPB_BRTYPE, SRSO_USER_KERNEL_NO, ERAPS and RAPSIZE bits
to EPYC-Turin.

v2: Fixed couple of typos.
Added Reviewed-by tag from Zhao.
Rebased on top of 6d00c6f98256 ("Merge tag 'for-upstream' of https://repo.or.cz/qemu/kevin into staging")

v3: https://lore.kernel.org/kvm/[email protected]/
v2: https://lore.kernel.org/kvm/[email protected]/
v1: https://lore.kernel.org/qemu-devel/[email protected]/

Babu Moger (5):
target/i386: Update EPYC CPU model for Cache property, RAS, SVM
feature bits
target/i386: Update EPYC-Rome CPU model for Cache property, RAS, SVM
feature bits
target/i386: Update EPYC-Milan CPU model for Cache property, RAS, SVM
feature bits
target/i386: Add feature that indicates WRMSR to BASE reg is
non-serializing
target/i386: Update EPYC-Genoa for Cache property, perfmon-v2, RAS and
SVM feature bits

 target/i386/cpu.c | 299 +++++++++++++++++++++++++++++++++++++++++++++-
 target/i386/cpu.h |   2 +
 2 files changed, 300 insertions(+), 1 deletion(-)
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH v4 1/5 target/i386: Update EPYC CPU model for Cache property, RAS, SVM feature bits

Found that some of the cache properties are not set correctly for EPYC models.

l1d_cache.no_invd_sharing should not be true.
l1i_cache.no_invd_sharing should not be true.

L2.self_init should be true.
L2.inclusive should be true.

L3.inclusive should not be true.
L3.no_invd_sharing should be true.

Fix the cache properties.

Also add the missing RAS and SVM features bits on AMD EPYC CPU models. The SVM feature bits are used in nested guests.

succor : Software uncorrectable error containment and recovery capability.
overflow-recov : MCA overflow recovery support.
lbrv : LBR virtualization
tsc-scale : MSR based TSC rate control
vmcb-clean : VMCB clean bits
flushbyasid : Flush by ASID
pause-filter : Pause intercept filter
pfthreshold : PAUSE filter threshold
v-vmsave-vmload : Virtualized VMLOAD and VMSAVE
vgif : Virtualized GIF

---
 target/i386/cpu.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 73 insertions(+)
target/i386/cpu.c +73 -0
index 58c96eafea..a632c8030c 100644
@@ -2181,6 +2181,60 @@ static CPUCaches epyc_v4_cache_info = {
     },
 };
 
+static CPUCaches epyc_v5_cache_info = {
+    .l1d_cache = &(CPUCacheInfo) {
+        .type = DATA_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l1i_cache = &(CPUCacheInfo) {
+        .type = INSTRUCTION_CACHE,
+        .level = 1,
+        .size = 64 * KiB,
+        .line_size = 64,
+        .associativity = 4,
+        .partitions = 1,
+        .sets = 256,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l2_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 2,
+        .size = 512 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 1024,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .inclusive = true,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l3_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 3,
+        .size = 8 * MiB,
+        .line_size = 64,
+        .associativity = 16,
+        .partitions = 1,
+        .sets = 8192,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .no_invd_sharing = true,
+        .complex_indexing = false,
+        .share_level = CPU_TOPOLOGY_LEVEL_DIE,
+    },
+};
+
 static const CPUCaches epyc_rome_cache_info = {
     .l1d_cache = &(CPUCacheInfo) {
         .type = DATA_CACHE,
@@ -5053,6 +5107,25 @@ static const X86CPUDefinition builtin_x86_defs[] = {
                 },
                 .cache_info = &epyc_v4_cache_info
             },
+            {
+                .version = 5,
+                .props = (PropValue[]) {
+                    { "overflow-recov", "on" },
+                    { "succor", "on" },
+                    { "lbrv", "on" },
+                    { "tsc-scale", "on" },
+                    { "vmcb-clean", "on" },
+                    { "flushbyasid", "on" },
+                    { "pause-filter", "on" },
+                    { "pfthreshold", "on" },
+                    { "v-vmsave-vmload", "on" },
+                    { "vgif", "on" },
+                    { "model-id",
+                      "AMD EPYC-v5 Processor" },
+                    { /* end of list */ }
+                },
+                .cache_info = &epyc_v5_cache_info
+            },
             { /* end of list */ }
         }
     },
Signature · 1 line
2.34.1
author
#2 lore raw

PATCH v4 2/5 target/i386: Update EPYC-Rome CPU model for Cache property, RAS, SVM feature bits

In reply to Babu Moger

Found that some of the cache properties are not set correctly for EPYC models.

l1d_cache.no_invd_sharing should not be true.
l1i_cache.no_invd_sharing should not be true.

L2.self_init should be true.
L2.inclusive should be true.

L3.inclusive should not be true.
L3.no_invd_sharing should be true.

Fix these cache properties.

Also add the missing RAS and SVM features bits on AMD EPYC-Rome. The SVM feature bits are used in nested guests.

succor : Software uncorrectable error containment and recovery capability.
overflow-recov : MCA overflow recovery support.
lbrv : LBR virtualization
tsc-scale : MSR based TSC rate control
vmcb-clean : VMCB clean bits
flushbyasid : Flush by ASID
pause-filter : Pause intercept filter
pfthreshold : PAUSE filter threshold
v-vmsave-vmload : Virtualized VMLOAD and VMSAVE
vgif : Virtualized GIF

---
 target/i386/cpu.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 73 insertions(+)
target/i386/cpu.c +73 -0
index a632c8030c..c21b232e75 100644
@@ -2343,6 +2343,60 @@ static const CPUCaches epyc_rome_v3_cache_info = {
     },
 };
 
+static const CPUCaches epyc_rome_v5_cache_info = {
+    .l1d_cache = &(CPUCacheInfo) {
+        .type = DATA_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l1i_cache = &(CPUCacheInfo) {
+        .type = INSTRUCTION_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l2_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 2,
+        .size = 512 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 1024,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .inclusive = true,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l3_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 3,
+        .size = 16 * MiB,
+        .line_size = 64,
+        .associativity = 16,
+        .partitions = 1,
+        .sets = 16384,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .no_invd_sharing = true,
+        .complex_indexing = false,
+        .share_level = CPU_TOPOLOGY_LEVEL_DIE,
+    },
+};
+
 static const CPUCaches epyc_milan_cache_info = {
     .l1d_cache = &(CPUCacheInfo) {
         .type = DATA_CACHE,
@@ -5264,6 +5318,25 @@ static const X86CPUDefinition builtin_x86_defs[] = {
                     { /* end of list */ }
                 },
             },
+            {
+                .version = 5,
+                .props = (PropValue[]) {
+                    { "overflow-recov", "on" },
+                    { "succor", "on" },
+                    { "lbrv", "on" },
+                    { "tsc-scale", "on" },
+                    { "vmcb-clean", "on" },
+                    { "flushbyasid", "on" },
+                    { "pause-filter", "on" },
+                    { "pfthreshold", "on" },
+                    { "v-vmsave-vmload", "on" },
+                    { "vgif", "on" },
+                    { "model-id",
+                      "AMD EPYC-Rome-v5 Processor" },
+                    { /* end of list */ }
+                },
+                .cache_info = &epyc_rome_v5_cache_info
+            },
             { /* end of list */ }
         }
     },
Signature · 1 line
2.34.1
author
#3 lore raw

PATCH v4 3/5 target/i386: Update EPYC-Milan CPU model for Cache property, RAS, SVM feature bits

In reply to Babu Moger

Found that some of the cache properties are not set correctly for EPYC models. l1d_cache.no_invd_sharing should not be true. l1i_cache.no_invd_sharing should not be true.

L2.self_init should be true.
L2.inclusive should be true.

L3.inclusive should not be true.
L3.no_invd_sharing should be true.

Fix these cache properties.

Also add the missing RAS and SVM features bits on AMD EPYC-Milan model. The SVM feature bits are used in nested guests.

succor : Software uncorrectable error containment and recovery capability.
overflow-recov : MCA overflow recovery support.
lbrv : LBR virtualization
tsc-scale : MSR based TSC rate control
vmcb-clean : VMCB clean bits
flushbyasid : Flush by ASID
pause-filter : Pause intercept filter
pfthreshold : PAUSE filter threshold
v-vmsave-vmload : Virtualized VMLOAD and VMSAVE
vgif : Virtualized GIF

---
 target/i386/cpu.c | 73 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 73 insertions(+)
target/i386/cpu.c +73 -0
index c21b232e75..4a4e9b81d8 100644
@@ -2505,6 +2505,60 @@ static const CPUCaches epyc_milan_v2_cache_info = {
     },
 };
 
+static const CPUCaches epyc_milan_v3_cache_info = {
+    .l1d_cache = &(CPUCacheInfo) {
+        .type = DATA_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l1i_cache = &(CPUCacheInfo) {
+        .type = INSTRUCTION_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l2_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 2,
+        .size = 512 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 1024,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .inclusive = true,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l3_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 3,
+        .size = 32 * MiB,
+        .line_size = 64,
+        .associativity = 16,
+        .partitions = 1,
+        .sets = 32768,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .no_invd_sharing = true,
+        .complex_indexing = false,
+        .share_level = CPU_TOPOLOGY_LEVEL_DIE,
+    },
+};
+
 static const CPUCaches epyc_genoa_cache_info = {
     .l1d_cache = &(CPUCacheInfo) {
         .type = DATA_CACHE,
@@ -5412,6 +5466,25 @@ static const X86CPUDefinition builtin_x86_defs[] = {
                 },
                 .cache_info = &epyc_milan_v2_cache_info
             },
+            {
+                .version = 3,
+                .props = (PropValue[]) {
+                    { "overflow-recov", "on" },
+                    { "succor", "on" },
+                    { "lbrv", "on" },
+                    { "tsc-scale", "on" },
+                    { "vmcb-clean", "on" },
+                    { "flushbyasid", "on" },
+                    { "pause-filter", "on" },
+                    { "pfthreshold", "on" },
+                    { "v-vmsave-vmload", "on" },
+                    { "vgif", "on" },
+                    { "model-id",
+                      "AMD EPYC-Milan-v3 Processor" },
+                    { /* end of list */ }
+                },
+                .cache_info = &epyc_milan_v3_cache_info
+            },
             { /* end of list */ }
         }
     },
Signature · 1 line
2.34.1
author
#4 lore raw

PATCH v4 4/5 target/i386: Add feature that indicates WRMSR to BASE reg is non-serializing

In reply to Babu Moger

Add the CPUID bit indicates that a WRMSR to MSR_FS_BASE, MSR_GS_BASE, or MSR_KERNEL_GS_BASE is non-serializing.

CPUID_Fn80000021_EAX
Bit Feature description
1 FsGsKernelGsBaseNonSerializing.
WRMSR to FS_BASE, GS_BASE and KernelGSbase are non-serializing.

---
 target/i386/cpu.c | 2 +-
 target/i386/cpu.h | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
target/i386/cpu.c +1 -1
index 4a4e9b81d8..107ecd2bde 100644
@@ -1235,7 +1235,7 @@ FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
     [FEAT_8000_0021_EAX] = {
         .type = CPUID_FEATURE_WORD,
         .feat_names = {
-            "no-nested-data-bp", NULL, "lfence-always-serializing", NULL,
+            "no-nested-data-bp", "fs-gs-base-ns", "lfence-always-serializing", NULL,
             NULL, NULL, "null-sel-clr-base", NULL,
             "auto-ibrs", NULL, NULL, NULL,
             NULL, NULL, NULL, NULL,
target/i386/cpu.h +2 -0
index b65eedb617..12dafc9b32 100644
@@ -1032,6 +1032,8 @@ uint64_t x86_cpu_get_supported_feature_word(X86CPU *cpu, FeatureWord w);
 
 /* Processor ignores nested data breakpoints */
 #define CPUID_8000_0021_EAX_NO_NESTED_DATA_BP            (1U << 0)
+/* WRMSR to FS_BASE, GS_BASE, or KERNEL_GS_BASE is non-serializing */
+#define CPUID_8000_0021_EAX_FS_GS_BASE_NS                (1U << 1)
 /* LFENCE is always serializing */
 #define CPUID_8000_0021_EAX_LFENCE_ALWAYS_SERIALIZING    (1U << 2)
 /* Null Selector Clears Base */
Signature · 1 line
2.34.1
author
#5 lore raw

PATCH v4 5/5 target/i386: Update EPYC-Genoa for Cache property, perfmon-v2, RAS and SVM feature bits

In reply to Babu Moger

Found that some of the cache properties are not set correctly for EPYC models. l1d_cache.no_invd_sharing should not be true. l1i_cache.no_invd_sharing should not be true.

L2.self_init should be true.
L2.inclusive should be true.

L3.inclusive should not be true.
L3.no_invd_sharing should be true.

Fix these cache properties.

Also add the missing RAS and SVM features bits on AMD EPYC-Genoa model. The SVM feature bits are used in nested guests.

perfmon-v2 : Allow guests to make use of the PerfMonV2 features.
succor : Software uncorrectable error containment and recovery capability.
overflow-recov : MCA overflow recovery support.
lbrv : LBR virtualization
tsc-scale : MSR based TSC rate control
vmcb-clean : VMCB clean bits
flushbyasid : Flush by ASID
pause-filter : Pause intercept filter
pfthreshold : PAUSE filter threshold
v-vmsave-vmload: Virtualized VMLOAD and VMSAVE
vgif : Virtualized GIF
fs-gs-base-ns : WRMSR to {FS,GS,KERNEL_GS}_BASE is non-serializing

The feature details are available in APM listed below [1].
[1] AMD64 Architecture Programmer's Manual Volume 2: System Programming Publication # 24593 Revision 3.41.

---
 target/i386/cpu.c | 78 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
target/i386/cpu.c +78 -0
index 107ecd2bde..1d241fcd13 100644
@@ -2613,6 +2613,59 @@ static const CPUCaches epyc_genoa_cache_info = {
     },
 };
 
+static const CPUCaches epyc_genoa_v2_cache_info = {
+    .l1d_cache = &(CPUCacheInfo) {
+        .type = DATA_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l1i_cache = &(CPUCacheInfo) {
+        .type = INSTRUCTION_CACHE,
+        .level = 1,
+        .size = 32 * KiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 64,
+        .lines_per_tag = 1,
+        .self_init = 1,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l2_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 2,
+        .size = 1 * MiB,
+        .line_size = 64,
+        .associativity = 8,
+        .partitions = 1,
+        .sets = 2048,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .inclusive = true,
+        .share_level = CPU_TOPOLOGY_LEVEL_CORE,
+    },
+    .l3_cache = &(CPUCacheInfo) {
+        .type = UNIFIED_CACHE,
+        .level = 3,
+        .size = 32 * MiB,
+        .line_size = 64,
+        .associativity = 16,
+        .partitions = 1,
+        .sets = 32768,
+        .lines_per_tag = 1,
+        .self_init = true,
+        .no_invd_sharing = true,
+        .complex_indexing = false,
+        .share_level = CPU_TOPOLOGY_LEVEL_DIE,
+    },
+};
 /* The following VMX features are not supported by KVM and are left out in the
  * CPU definitions:
  *
@@ -5559,6 +5612,31 @@ static const X86CPUDefinition builtin_x86_defs[] = {
         .xlevel = 0x80000022,
         .model_id = "AMD EPYC-Genoa Processor",
         .cache_info = &epyc_genoa_cache_info,
+        .versions = (X86CPUVersionDefinition[]) {
+            { .version = 1 },
+            {
+                .version = 2,
+                .props = (PropValue[]) {
+                    { "overflow-recov", "on" },
+                    { "succor", "on" },
+                    { "lbrv", "on" },
+                    { "tsc-scale", "on" },
+                    { "vmcb-clean", "on" },
+                    { "flushbyasid", "on" },
+                    { "pause-filter", "on" },
+                    { "pfthreshold", "on" },
+                    { "v-vmsave-vmload", "on" },
+                    { "vgif", "on" },
+                    { "fs-gs-base-ns", "on" },
+                    { "perfmon-v2", "on" },
+                    { "model-id",
+                      "AMD EPYC-Genoa-v2 Processor" },
+                    { /* end of list */ }
+                },
+                .cache_info = &epyc_genoa_v2_cache_info
+            },
+            { /* end of list */ }
+        }
     },
 };
 
Signature · 1 line
2.34.1
#6 lore raw
In reply to Babu Moger

Hi!

Quoted 11 lines · Babu Moger

This series addresses the following issues with EPYC CPU models.

1. Update the L1, L2, L3 cache properties to match the actual hardware (PPR). This needs to be updated on all the EPYC models.
2. RAS feature bits (SUCCOR, McaOverflowRecov).
3. Add SVM feature bits which are required in nested guests.
4. Add perfmon-v2 on Genoa.
5. Add missing feature bit fs-gs-base-ns(WRMSR to {FS,GS,KERNEL_G}S_BASE is non-serializing).

Seems good for me. I've tested cache changes.
If needed, Reviewed-by: Maksim Davydov <[email protected]>

Quoted 42 lines

Dropped EPYC-Turin model for now. Some of the feature bits (srso-user-kernel-no, eraps, rapsize) are still work in progress in the kernel. Will post them later.

---
v4: Some of the patches in v3 are already merged. Posting the rest of the patches.
Dropped EPYC-Turin model for now. Will post them later.
Added SVM feature bit as discussed in
https://lore.kernel.org/kvm/[email protected]/
Fixed the cache property details as discussed in
https://lore.kernel.org/kvm/[email protected]/
Thanks to Maksim and Paolo for their feedback.

v3: Added SBPB, IBPB_BRTYPE, SRSO_USER_KERNEL_NO, ERAPS and RAPSIZE bits
to EPYC-Turin.

v2: Fixed couple of typos.
Added Reviewed-by tag from Zhao.
Rebased on top of 6d00c6f98256 ("Merge tag 'for-upstream' of https://repo.or.cz/qemu/kevin into staging")

v3: https://lore.kernel.org/kvm/[email protected]/
v2: https://lore.kernel.org/kvm/[email protected]/
v1: https://lore.kernel.org/qemu-devel/[email protected]/

Babu Moger (5):
target/i386: Update EPYC CPU model for Cache property, RAS, SVM
feature bits
target/i386: Update EPYC-Rome CPU model for Cache property, RAS, SVM
feature bits
target/i386: Update EPYC-Milan CPU model for Cache property, RAS, SVM
feature bits
target/i386: Add feature that indicates WRMSR to BASE reg is
non-serializing
target/i386: Update EPYC-Genoa for Cache property, perfmon-v2, RAS and
SVM feature bits

  target/i386/cpu.c | 299 +++++++++++++++++++++++++++++++++++++++++++++-
  target/i386/cpu.h |   2 +
  2 files changed, 300 insertions(+), 1 deletion(-)
Signature · 2 lines
Best regards,
Maksim Davydov