Skip to contentSkip to conversation

kvm: x86: limit the maximum number of vPMU fixed counters to 3

PATCH Applied Like Xu 3 participants 5 messages Jun 24, 2020 – Jul 8, 2020

Subsystems (3)
lore
Share & tools
lore

Like Xu

author
#0 lore raw

PATCH 1/1 kvm: x86: limit the maximum number of vPMU fixed counters to 3

Some new Intel platforms (such as TGL) already have the fourth fixed counter TOPDOWN.SLOTS, but it has not been fully enabled on KVM and the host.

Therefore, we limit edx.split.num_counters_fixed to 3, so that it does not break the kvm-unit-tests PMU test case and bad-handled userspace.

---
 arch/x86/kvm/cpuid.c | 2 +-
 arch/x86/kvm/pmu.h   | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/x86/kvm/cpuid.c +1 -1
index 8a294f9747aa..0a2c6d2b4650 100644
@@ -604,7 +604,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
 		eax.split.bit_width = cap.bit_width_gp;
 		eax.split.mask_length = cap.events_mask_len;
 
-		edx.split.num_counters_fixed = cap.num_counters_fixed;
+		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
 		edx.split.bit_width_fixed = cap.bit_width_fixed;
 		edx.split.reserved = 0;
 
arch/x86/kvm/pmu.h +2 -0
index ab85eed8a6cc..067fef51760c 100644
@@ -15,6 +15,8 @@
 #define VMWARE_BACKDOOR_PMC_REAL_TIME		0x10001
 #define VMWARE_BACKDOOR_PMC_APPARENT_TIME	0x10002
 
+#define MAX_FIXED_COUNTERS	3
+
 struct kvm_event_hw_type_mapping {
 	u8 eventsel;
 	u8 unit_mask;
Signature · 1 line
2.21.3
#1 lore raw

Kindly ping.

I think we may need this patch, as we limit the maximum vPMU version to 2:
    eax.split.version_id = min(cap.version, 2);

Thanks,
Like Xu

Quoted 40 lines · Like Xu

Some new Intel platforms (such as TGL) already have the fourth fixed counter TOPDOWN.SLOTS, but it has not been fully enabled on KVM and the host.

Therefore, we limit edx.split.num_counters_fixed to 3, so that it does not break the kvm-unit-tests PMU test case and bad-handled userspace.

---
  arch/x86/kvm/cpuid.c | 2 +-
  arch/x86/kvm/pmu.h   | 2 ++
  2 files changed, 3 insertions(+), 1 deletion(-)
arch/x86/kvm/cpuid.c +1 -1
index 8a294f9747aa..0a2c6d2b4650 100644
@@ -604,7 +604,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
  		eax.split.bit_width = cap.bit_width_gp;
  		eax.split.mask_length = cap.events_mask_len;
 
-		edx.split.num_counters_fixed = cap.num_counters_fixed;
+		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
  		edx.split.bit_width_fixed = cap.bit_width_fixed;
  		edx.split.reserved = 0;
 
arch/x86/kvm/pmu.h +2 -0
index ab85eed8a6cc..067fef51760c 100644
@@ -15,6 +15,8 @@
  #define VMWARE_BACKDOOR_PMC_REAL_TIME		0x10001
  #define VMWARE_BACKDOOR_PMC_APPARENT_TIME	0x10002
 
+#define MAX_FIXED_COUNTERS	3
+
  struct kvm_event_hw_type_mapping {
  	u8 eventsel;
  	u8 unit_mask;
maintainer
#2 lore raw
Quoted 4 lines · Like

Kindly ping.

I think we may need this patch, as we limit the maximum vPMU version to 2:
    eax.split.version_id = min(cap.version, 2);

I don't think this is a problem. Are you planning to add support for the fourth counter?

Paolo

#3 lore raw
Quoted 7 lines · Paolo Bonzini
Quoted 4 lines · Like

Kindly ping.

I think we may need this patch, as we limit the maximum vPMU version to 2:
    eax.split.version_id = min(cap.version, 2);

I don't think this is a problem. Are you planning to add support for the fourth counter?

Yes, we plan to provide this support on the KVM after fully enabling the fourth counter (and an accompanying special counter) on the host perf side.

This may require one or two kernel cycles, so I have to prevent it from being exposed to non-linux guest with this fix. Thanks for your support.

Thanks,
Like Xu

Quoted 3 lines

Paolo

maintainer Applied
#4 lore raw
In reply to Like Xu
Quoted 33 lines · Like Xu

Some new Intel platforms (such as TGL) already have the fourth fixed counter TOPDOWN.SLOTS, but it has not been fully enabled on KVM and the host.

Therefore, we limit edx.split.num_counters_fixed to 3, so that it does not break the kvm-unit-tests PMU test case and bad-handled userspace.

---
 arch/x86/kvm/cpuid.c | 2 +-
 arch/x86/kvm/pmu.h   | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
arch/x86/kvm/cpuid.c +1 -1
index 8a294f9747aa..0a2c6d2b4650 100644
@@ -604,7 +604,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
 		eax.split.bit_width = cap.bit_width_gp;
 		eax.split.mask_length = cap.events_mask_len;
 
-		edx.split.num_counters_fixed = cap.num_counters_fixed;
+		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
 		edx.split.bit_width_fixed = cap.bit_width_fixed;
 		edx.split.reserved = 0;
 
arch/x86/kvm/pmu.h +0 -0
index ab85eed8a6cc..067fef51760c 100644

arch/x86/kvm/pmu.h line 22

Quoted 8 lines
arch/x86/kvm/pmu.h +2 -0
@@ -15,6 +15,8 @@
 #define VMWARE_BACKDOOR_PMC_REAL_TIME		0x10001
 #define VMWARE_BACKDOOR_PMC_APPARENT_TIME	0x10002
 
+#define MAX_FIXED_COUNTERS	3
+
 struct kvm_event_hw_type_mapping {
 	u8 eventsel;
 	u8 unit_mask;

Queued, thanks.

Paolo