KVM: x86/intr: Explicitly check NMI from guest to eliminate false positives
PATCH v2 1/1 KVM: x86/intr: Explicitly check NMI from guest to eliminate false positives
From: Like Xu <[email protected]>
Explicitly checking the source of external interrupt is indeed NMI and not other types in the kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but before kvm_before_interrupt() from being incorrectly labelled as guest samples:
# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%
In the above case, perf records the samples labelled '[g]', the RIPs behind the weird samples are actually being queried by perf_instruction_pointer() after determining whether it's in GUEST state or not, and here's the issue:
If vm-exit is caused by a non-NMI interrupt (such as hrtimer_interrupt) and at least one PMU counter is enabled on host, the kvm_arch_pmi_in_guest() will remain true (KVM_HANDLING_IRQ is set) until kvm_before_interrupt().
During this window, if a PMI occurs on host (since the KVM instructions on host are being executed), the control flow, with the help of the host NMI context, will be transferred to perf/core to generate performance samples, thus perf_instruction_pointer() and perf_guest_get_ip() is called.
Since kvm_arch_pmi_in_guest() only checks if there is an interrupt, it may cause perf/core to mistakenly assume that the source RIP of the host NMI belongs to the guest world and use perf_guest_get_ip() to get the RIP of a vCPU that has already exited by a non-NMI interrupt.
Error samples are recorded and presented to the end-user via perf-report. Such false positive samples could be eliminated by explicitly determining if the exit reason is KVM_HANDLING_NMI.
Note that when vm-exit is indeed triggered by PMI and before HANDLING_NMI is cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.
---
V1 -> V2 Changelog:
- Refine commit message to cover both perf/core timer and NMI modes;
- Use in_nmi() to distinguish whether it's NMI mode or not; (Sean)
V1: https://lore.kernel.org/kvm/[email protected]/
arch/x86/include/asm/kvm_host.h | 10 +++++++++- arch/x86/kvm/x86.h | 6 ------ 2 files changed, 9 insertions(+), 7 deletions(-)
arch/x86/include/asm/kvm_host.h +9 -1
@@ -1868,8 +1868,16 @@ static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, }#endif /* CONFIG_HYPERV */ +enum kvm_intr_type { +/* Values are arbitrary, but must be non-zero. */ +KVM_HANDLING_IRQ =1 , +KVM_HANDLING_NMI , +}; + +/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ - ((vcpu) && (vcpu)->arch.handling_intr_from_guest) + ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ + (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );
arch/x86/kvm/x86.h +0 -6
@@ -431,12 +431,6 @@ static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)return kvm->arch.notify_vmexit_flags &KVM_X86_NOTIFY_VMEXIT_ENABLED ; } -enum kvm_intr_type { -/* Values are arbitrary, but must be non-zero. */ -KVM_HANDLING_IRQ =1 , -KVM_HANDLING_NMI , -}; - static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,enum kvm_intr_type intr) {
base-commit: 1ab097653e4dd8d23272d028a61352c23486fd4a
Signature · 1 line
2.43.0
Sean Christopherson
maintainerPlease don't make up random prefixes. This should really be "x86/pmu".
From Documentation/process/maintainer-kvm-x86.rst:
Shortlog
~~~~~~~~
The preferred prefix format is ``KVM: <topic>:``, where ``<topic>`` is one of::
- x86
- x86/mmu
- x86/pmu
- x86/xen
- selftests
- SVM
- nSVM
- VMX
- nVMX
...
New topics do occasionally pop up, but please start an on-list discussion if you want to propose introducing a new topic, i.e. don't go rogue.
Quoted 1 line · Sean Christopherson
Please don't make up random prefixes. This should really be "x86/pmu".
Thanks.
I'm hesitant to categorize about NMI handling into kvm/pmu scope. But you have clear idea on it and it's fine to me. Please feel free to refine commit message to fit your understanding and taste as you did.
Quoted 22 lines
From
Documentation/process/maintainer-kvm-x86.rst:Shortlog
~~~~~~~~
The preferred prefix format is ``KVM: <topic>:``, where ``<topic>`` is one of::- x86
- x86/mmu
- x86/pmu
- x86/xen
- selftests
- SVM
- nSVM
- VMX
- nVMX...
New topics do occasionally pop up, but please start an on-list discussion if you want to propose introducing a new topic, i.e. don't go rogue.
Sean Christopherson
maintainerQuoted 6 lines · Like Xu
Quoted 1 line · Sean Christopherson
Please don't make up random prefixes. This should really be "x86/pmu".
Thanks.
I'm hesitant to categorize about NMI handling into kvm/pmu scope.
Why? Literally the only thing this can affect is PMU behavior. Even if there's a bug that affects the kernel's PMU, that's still x86/pmu as far as KVM is concerned.
Quoted 90 lines · Like Xu
From: Like Xu <[email protected]>
Explicitly checking the source of external interrupt is indeed NMI and not other types in the
kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but beforekvm_before_interrupt()from being incorrectly labelled as guest samples:# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%In the above case, perf records the samples labelled '[g]', the RIPs behind the weird samples are actually being queried by
perf_instruction_pointer()after determining whether it's in GUEST state or not, and here's the issue:If vm-exit is caused by a non-NMI interrupt (such as
hrtimer_interrupt) and at least one PMU counter is enabled on host, thekvm_arch_pmi_in_guest()will remain true (KVM_HANDLING_IRQis set) untilkvm_before_interrupt().During this window, if a PMI occurs on host (since the KVM instructions on host are being executed), the control flow, with the help of the host NMI context, will be transferred to perf/core to generate performance samples, thus
perf_instruction_pointer()andperf_guest_get_ip()is called.Since
kvm_arch_pmi_in_guest()only checks if there is an interrupt, it may cause perf/core to mistakenly assume that the source RIP of the host NMI belongs to the guest world and useperf_guest_get_ip()to get the RIP of a vCPU that has already exited by a non-NMI interrupt.Error samples are recorded and presented to the end-user via perf-report. Such false positive samples could be eliminated by explicitly determining if the exit reason is
KVM_HANDLING_NMI.Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.Fixes dd60d217062f ("KVM: x86: Fix perf timer mode IP reporting")Signed-off-by Like Xu <[email protected]>---
V1 -> V2 Changelog:
- Refine commit message to cover both perf/core timer and NMI modes;
- Usein_nmi()to distinguish whether it's NMI mode or not; (Sean)
V1: https://lore.kernel.org/kvm/[email protected]/arch/x86/include/asm/kvm_host.h | 10 +++++++++- arch/x86/kvm/x86.h | 6 ------ 2 files changed, 9 insertions(+), 7 deletions(-)arch/x86/include/asm/kvm_host.h +9
-1@@ -1868,8 +1868,16 @@ static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, }#endif /* CONFIG_HYPERV */ +enum kvm_intr_type { +/* Values are arbitrary, but must be non-zero. */ +KVM_HANDLING_IRQ =1 , +KVM_HANDLING_NMI , +}; + +/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ - ((vcpu) && (vcpu)->arch.handling_intr_from_guest) + ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ + (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );arch/x86/kvm/x86.h +0
-0@@ -431,12 +431,6 @@ static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)return kvm->arch.notify_vmexit_flags &KVM_X86_NOTIFY_VMEXIT_ENABLED ; }
arch/x86/kvm/x86.h static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm) · line 436
Quoted 10 lines
arch/x86/kvm/x86.h +0
-6-enum kvm_intr_type { -/* Values are arbitrary, but must be non-zero. */ -KVM_HANDLING_IRQ =1 , -KVM_HANDLING_NMI , -}; - static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,enum kvm_intr_type intr) {
Best regards,
Maxim Levitsky
Quoted 2 lines
base-commit:
1ab097653e4dd8d23272d028a61352c23486fd4a
Hi Like,
Quoted 6 lines · Like Xu
From: Like Xu <[email protected]>
Explicitly checking the source of external interrupt is indeed NMI and not other types in the
kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but beforekvm_before_interrupt()from being incorrectly labelled as guest samples:
About the before kvm_before_interrupt() ...
Quoted 14 lines
# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%
Quoted 7 lines
In the above case, perf records the samples labelled '[g]', the RIPs behind the weird samples are actually being queried by
perf_instruction_pointer()after determining whether it's in GUEST state or not, and here's the issue:If vm-exit is caused by a non-NMI interrupt (such as
hrtimer_interrupt) and at least one PMU counter is enabled on host, thekvm_arch_pmi_in_guest()will remain true (KVM_HANDLING_IRQis set) untilkvm_before_interrupt().
... and here.
Would you mind helping why kvm_arch_pmi_in_guest() remains true before *kvm_before_interrupt()*.
According to the source code, the vcpu->arch.handling_intr_from_guest is set to non-zero only at kvm_before_interrupt(), and cleared at kvm_after_interrupt().
Or would you mean kvm_after_interrupt()?
Thank you very much!
Dongli Zhang
Quoted 75 lines
During this window, if a PMI occurs on host (since the KVM instructions on host are being executed), the control flow, with the help of the host NMI context, will be transferred to perf/core to generate performance samples, thus
perf_instruction_pointer()andperf_guest_get_ip()is called.Since
kvm_arch_pmi_in_guest()only checks if there is an interrupt, it may cause perf/core to mistakenly assume that the source RIP of the host NMI belongs to the guest world and useperf_guest_get_ip()to get the RIP of a vCPU that has already exited by a non-NMI interrupt.Error samples are recorded and presented to the end-user via perf-report. Such false positive samples could be eliminated by explicitly determining if the exit reason is
KVM_HANDLING_NMI.Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.Fixes dd60d217062f ("KVM: x86: Fix perf timer mode IP reporting")Signed-off-by Like Xu <[email protected]>---
V1 -> V2 Changelog:
- Refine commit message to cover both perf/core timer and NMI modes;
- Usein_nmi()to distinguish whether it's NMI mode or not; (Sean)
V1: https://urldefense.com/v3/__https://lore.kernel.org/kvm/[email protected]/__;!!ACWV5N9M2RV99hQ!MQ8FetD27SVKN34CS_P-K3qrhspFnpf_Mqb0McFN9y5vSUeScc5b0TlZ3ZMDvt4Cn4b3g0h9ci6EO9k3PBEQXpePrg$arch/x86/include/asm/kvm_host.h | 10 +++++++++- arch/x86/kvm/x86.h | 6 ------ 2 files changed, 9 insertions(+), 7 deletions(-)arch/x86/include/asm/kvm_host.h +9
-1@@ -1868,8 +1868,16 @@ static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, }#endif /* CONFIG_HYPERV */ +enum kvm_intr_type { +/* Values are arbitrary, but must be non-zero. */ +KVM_HANDLING_IRQ =1 , +KVM_HANDLING_NMI , +}; + +/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ - ((vcpu) && (vcpu)->arch.handling_intr_from_guest) + ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ + (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );arch/x86/kvm/x86.h +0
-6@@ -431,12 +431,6 @@ static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)return kvm->arch.notify_vmexit_flags &KVM_X86_NOTIFY_VMEXIT_ENABLED ; } -enum kvm_intr_type { -/* Values are arbitrary, but must be non-zero. */ -KVM_HANDLING_IRQ =1 , -KVM_HANDLING_NMI , -}; - static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,enum kvm_intr_type intr) {base-commit:
1ab097653e4dd8d23272d028a61352c23486fd4a
Quoted 36 lines · Dongli Zhang
Hi Like,
Quoted 6 lines · Like Xu
From: Like Xu <[email protected]>
Explicitly checking the source of external interrupt is indeed NMI and not other types in the
kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but beforekvm_before_interrupt()from being incorrectly labelled as guest samples:About the before
kvm_before_interrupt()...Quoted 14 lines
# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%Quoted 7 lines
In the above case, perf records the samples labelled '[g]', the RIPs behind the weird samples are actually being queried by
perf_instruction_pointer()after determining whether it's in GUEST state or not, and here's the issue:If vm-exit is caused by a non-NMI interrupt (such as
hrtimer_interrupt) and at least one PMU counter is enabled on host, thekvm_arch_pmi_in_guest()will remain true (KVM_HANDLING_IRQis set) untilkvm_before_interrupt().... and here.
Quoted 8 lines
Would you mind helping why
kvm_arch_pmi_in_guest()remains true before *kvm_before_interrupt()*.According to the source code, the
vcpu->arch.handling_intr_from_guestis set to non-zero only atkvm_before_interrupt(), and cleared atkvm_after_interrupt().Or would you mean
kvm_after_interrupt()?
Oops, it should refer to kvm_after_interrupt() as the code fixed. Thank you.
Quoted 80 lines
Thank you very much!
Dongli Zhang
Quoted 75 lines
During this window, if a PMI occurs on host (since the KVM instructions on host are being executed), the control flow, with the help of the host NMI context, will be transferred to perf/core to generate performance samples, thus
perf_instruction_pointer()andperf_guest_get_ip()is called.Since
kvm_arch_pmi_in_guest()only checks if there is an interrupt, it may cause perf/core to mistakenly assume that the source RIP of the host NMI belongs to the guest world and useperf_guest_get_ip()to get the RIP of a vCPU that has already exited by a non-NMI interrupt.Error samples are recorded and presented to the end-user via perf-report. Such false positive samples could be eliminated by explicitly determining if the exit reason is
KVM_HANDLING_NMI.Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.Fixes dd60d217062f ("KVM: x86: Fix perf timer mode IP reporting")Signed-off-by Like Xu <[email protected]>---
V1 -> V2 Changelog:
- Refine commit message to cover both perf/core timer and NMI modes;
- Usein_nmi()to distinguish whether it's NMI mode or not; (Sean)
V1: https://urldefense.com/v3/__https://lore.kernel.org/kvm/[email protected]/__;!!ACWV5N9M2RV99hQ!MQ8FetD27SVKN34CS_P-K3qrhspFnpf_Mqb0McFN9y5vSUeScc5b0TlZ3ZMDvt4Cn4b3g0h9ci6EO9k3PBEQXpePrg$arch/x86/include/asm/kvm_host.h | 10 +++++++++- arch/x86/kvm/x86.h | 6 ------ 2 files changed, 9 insertions(+), 7 deletions(-)arch/x86/include/asm/kvm_host.h +9
-1@@ -1868,8 +1868,16 @@ static inline int kvm_arch_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, }#endif /* CONFIG_HYPERV */ +enum kvm_intr_type { +/* Values are arbitrary, but must be non-zero. */ +KVM_HANDLING_IRQ =1 , +KVM_HANDLING_NMI , +}; + +/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ - ((vcpu) && (vcpu)->arch.handling_intr_from_guest) + ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ + (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );arch/x86/kvm/x86.h +0
-6@@ -431,12 +431,6 @@ static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)return kvm->arch.notify_vmexit_flags &KVM_X86_NOTIFY_VMEXIT_ENABLED ; } -enum kvm_intr_type { -/* Values are arbitrary, but must be non-zero. */ -KVM_HANDLING_IRQ =1 , -KVM_HANDLING_NMI , -}; - static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,enum kvm_intr_type intr) {base-commit:
1ab097653e4dd8d23272d028a61352c23486fd4a
Sean Christopherson
maintainerQuoted 42 lines · Like Xu
Quoted 39 lines · Dongli Zhang
Hi Like,
Quoted 6 lines · Like Xu
From: Like Xu <[email protected]>
Explicitly checking the source of external interrupt is indeed NMI and not other types in the
kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but beforekvm_before_interrupt()from being incorrectly labelled as guest samples:About the before
kvm_before_interrupt()...Quoted 14 lines
# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%Quoted 7 lines
In the above case, perf records the samples labelled '[g]', the RIPs behind the weird samples are actually being queried by
perf_instruction_pointer()after determining whether it's in GUEST state or not, and here's the issue:If vm-exit is caused by a non-NMI interrupt (such as
hrtimer_interrupt) and at least one PMU counter is enabled on host, thekvm_arch_pmi_in_guest()will remain true (KVM_HANDLING_IRQis set) untilkvm_before_interrupt().... and here.
Would you mind helping why
kvm_arch_pmi_in_guest()remains true before *kvm_before_interrupt()*.
Quoted 7 lines
Quoted 5 lines
According to the source code, the
vcpu->arch.handling_intr_from_guestis set to non-zero only atkvm_before_interrupt(), and cleared atkvm_after_interrupt().Or would you mean
kvm_after_interrupt()?Oops, it should refer to
kvm_after_interrupt()as the code fixed. Thank you.
No need for another version if that's the only hiccup, I can fixup when applying.
Sean Christopherson
maintainer+Oliver
Quoted 7 lines · Like Xu
Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.
Oliver has a patch[*] that he promised he would send "soon" (wink wink) to properly fix events that are configured to exclude the guest. Unless someone objects, I'm going to tweak the last part of the changelog to be:
Note that when VM-exit is indeed triggered by PMI and before HANDLING_NMI is cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core.
For events that are host-only, perf/core can and should eliminate false positives by checking event->attr.exclude_guest, i.e. events that are configured to exclude KVM guests should never fire in the guest.
Events that are configured to count host and guest are trickier, perhaps impossible to handle with 100% accuracy? And regardless of what accuracy is provided by perf/core, improving KVM's accuracy is cheap and easy, with no real downsides.
Sean Christopherson
maintainerQuoted 20 lines · Sean Christopherson
+Oliver
Quoted 7 lines · Like Xu
Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.Oliver has a patch[*] that he promised he would send "soon" (wink wink) to properly fix events that are configured to exclude the guest. Unless someone objects, I'm going to tweak the last part of the changelog to be:
Note that when VM-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core.
Quoted 8 lines
For events that are host-only, perf/core can and should eliminate false positives by checking
event->attr.exclude_guest, i.e. events that are configured to exclude KVM guests should never fire in the guest.Events that are configured to count host and guest are trickier, perhaps impossible to handle with 100% accuracy? And regardless of what accuracy is provided by perf/core, improving KVM's accuracy is cheap and easy, with no real downsides.
Never mind, this causes KUT's pmu_pebs test to fail:
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
It might be a test bug, but I have neither the time nor the inclination to investigate.
Like,
If you want any chance of your patches going anywhere but my trash folder, you need to change your upstream workflow to actually run tests. I would give most people the benefit of the doubt, e.g. assume they didn't have the requisite hardware, or didn't realize which tests would be relevant/important. But this is a recurring problem, and you have been warned, multiple times.
Quoted 53 lines · Sean Christopherson
Quoted 20 lines · Sean Christopherson
+Oliver
Quoted 7 lines · Like Xu
Note that when vm-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core. In both cases above, perf/core should correctly distinguish between real RIP sources or even need to generate two samples, belonging to host and guest separately, but that's perf/core's story for interested warriors.Oliver has a patch[*] that he promised he would send "soon" (wink wink) to properly fix events that are configured to exclude the guest. Unless someone objects, I'm going to tweak the last part of the changelog to be:
Note that when VM-exit is indeed triggered by PMI and before
HANDLING_NMIis cleared, it's also still possible that another PMI is generated on host. Also for perf/core timer mode, the false positives are still possible since that non-NMI sources of interrupts are not always being used by perf/core.Quoted 8 lines
For events that are host-only, perf/core can and should eliminate false positives by checking
event->attr.exclude_guest, i.e. events that are configured to exclude KVM guests should never fire in the guest.Events that are configured to count host and guest are trickier, perhaps impossible to handle with 100% accuracy? And regardless of what accuracy is provided by perf/core, improving KVM's accuracy is cheap and easy, with no real downsides.
Never mind, this causes KUT's
pmu_pebstest to fail:FAIL: Multiple (
0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
Quoted 2 lines
It might be a test bug, but I have neither the time nor the inclination to investigate.
For PEBS ovf case, we have "in_nmi() = 0x100000" from the core kernel and the following diff fixes the issue:
arch/x86/include/asm/kvm_host.h +1 -1
@@ -1891,7 +1891,7 @@ enum kvm_intr_type {/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ - (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI))) + (!!in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );
, does it help (tests passed on ICX) ?
Quoted 9 lines
Like,
If you want any chance of your patches going anywhere but my trash folder, you need to change your upstream workflow to actually run tests. I would give most people the benefit of the doubt, e.g. assume they didn't have the requisite hardware, or didn't realize which tests would be relevant/important. But this is a recurring problem, and you have been warned, multiple times.
Sorry, my CI resources are diverted to other downstream projects. But there's no doubt it's my fault and this behavior will be corrected.
Sean Christopherson
maintainerQuoted 45 lines · Like Xu
Quoted 24 lines · Sean Christopherson
On Tue, Feb 06, 2024, Sean Christopherson wrote:
Never mind, this causes KUT'spmu_pebstest to fail:FAIL: Multiple (
0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): GP counter 0 (0xfffffffffffe): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x2): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x4): Multiple (0x700000055): No OVF irq, none PEBS records.
FAIL: Adaptive (0x1f000008): Multiple (0x700000055): No OVF irq, none PEBS records.Quoted 2 lines
It might be a test bug, but I have neither the time nor the inclination to investigate.
For PEBS ovf case, we have "
in_nmi()=0x100000" from the core kernel and the following diff fixes the issue:arch/x86/include/asm/kvm_host.h +1
-1@@ -1891,7 +1891,7 @@ enum kvm_intr_type {/* Enable perf NMI and timer modes to work, and minimise false positives. */ #define kvm_arch_pmi_in_guest (vcpu) \ ((vcpu) && (vcpu)->arch.handling_intr_from_guest && \ - (in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI))) + (!!in_nmi() == ((vcpu)->arch.handling_intr_from_guest == KVM_HANDLING_NMI)))void __init kvm_mmu_x86_module_init (void );int kvm_mmu_vendor_module_init (void );
Quoted 1 line
, does it help (tests passed on ICX) ?
Yes, that resolves the issues I was seeing. I'll get this applied with the above squashed.
I'll also see if the tip tree folks would be open to converting the in_{nmi,hardirq,...}() macros to functions that return bools (or at least casting to bools in the macros). I can't see any reason for in_nmi() to effectively return an int since it's just a wrapper to nmi_count(), and this seems like a disaster waiting to happen.
Quoted 8 lines
Quoted 5 lines
If you want any chance of your patches going anywhere but my trash folder, you need to change your upstream workflow to actually run tests. I would give most people the benefit of the doubt, e.g. assume they didn't have the requisite hardware, or didn't realize which tests would be relevant/important. But this is a recurring problem, and you have been warned, multiple times.
Sorry, my CI resources are diverted to other downstream projects. But there's no doubt it's my fault and this behavior will be corrected.
Thank you.
Quoted 18 lines · Like Xu
Explicitly checking the source of external interrupt is indeed NMI and not other types in the
kvm_arch_pmi_in_guest(), which prevents perf-kvm false positive samples generated in perf/core NMI mode after vm-exit but beforekvm_before_interrupt()from being incorrectly labelled as guest samples:# test: perf-record + cpu-cycles:HP (which collects host-only precise samples)
# Symbol Overhead sys usr guest sys guest usr # ....................................... ........ ........ ........ ......... ......... # # Before: [g] entry_SYSCALL_64 24.63% 0.00% 0.00% 24.63% 0.00% [g] syscall_return_via_sysret 23.23% 0.00% 0.00% 23.23% 0.00% [g] files_lookup_fd_raw 6.35% 0.00% 0.00% 6.35% 0.00% # After: [k] perf_adjust_freq_unthr_context 57.23% 57.23% 0.00% 0.00% 0.00% [k] __vmx_vcpu_run 4.09% 4.09% 0.00% 0.00% 0.00% [k] vmx_update_host_rsp 3.17% 3.17% 0.00% 0.00% 0.00%
Quoted 1 line
[...]
Applied to kvm-x86 pmu, with the !!in_nmi() fixup squashed. Thanks!
[1/1] KVM: x86/intr: Explicitly check NMI from guest to eliminate false positives
https://github.com/kvm-x86/linux/commit/812d432373f6