| Message ID | 20220714120330.1410308-1-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | kvm: stats: tell userspace which values are boolean | expand |
On 7/14/22 14:03, Paolo Bonzini wrote: > Some of the statistics values exported by KVM are always only 0 or 1. > It can be useful to export this fact to userspace so that it can track > them specially (for example by polling the value every now and then to > compute a % of time spent in a specific state). > > Therefore, add "boolean value" as a new "unit". While it is not exactly > a unit, it walks and quacks like one. In particular, using the type > would be wrong because boolean values could be instantaneous or peak > values (e.g. "is the rmap allocated?") or even two-bucket histograms > (e.g. "number of posted vs. non-posted interrupt injections"). > > Suggested-by: Amneesh Singh <natto@weirdnatto.in> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > Documentation/virt/kvm/api.rst | 6 ++++++ > arch/x86/kvm/x86.c | 2 +- > include/linux/kvm_host.h | 11 ++++++++++- > include/uapi/linux/kvm.h | 1 + > 4 files changed, 18 insertions(+), 2 deletions(-) > > diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst > index 11e00a46c610..48bf6e49a7de 100644 > --- a/Documentation/virt/kvm/api.rst > +++ b/Documentation/virt/kvm/api.rst > @@ -5657,6 +5657,7 @@ by a string of size ``name_size``. > #define KVM_STATS_UNIT_BYTES (0x1 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_SECONDS (0x2 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_CYCLES (0x3 << KVM_STATS_UNIT_SHIFT) > + #define KVM_STATS_UNIT_BOOLEAN (0x4 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_MAX KVM_STATS_UNIT_CYCLES > > #define KVM_STATS_BASE_SHIFT 8 [...] > diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h > index 5088bd9f1922..811897dadcae 100644 > --- a/include/uapi/linux/kvm.h > +++ b/include/uapi/linux/kvm.h > @@ -2083,6 +2083,7 @@ struct kvm_stats_header { > #define KVM_STATS_UNIT_BYTES (0x1 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_SECONDS (0x2 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_CYCLES (0x3 << KVM_STATS_UNIT_SHIFT) > +#define KVM_STATS_UNIT_BOOLEAN (0x4 << KVM_STATS_UNIT_SHIFT) > #define KVM_STATS_UNIT_MAX KVM_STATS_UNIT_CYCLES Shouldn't KVM_STATS_UNIT_MAX be KVM_STATS_UNIT_BOOLEAN then? The selftest has: TEST_ASSERT((pdesc->flags & KVM_STATS_UNIT_MASK) <= KVM_STATS_UNIT_MAX, "Unknown KVM stats unit"); > > #define KVM_STATS_BASE_SHIFT 8
diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst index 11e00a46c610..48bf6e49a7de 100644 --- a/Documentation/virt/kvm/api.rst +++ b/Documentation/virt/kvm/api.rst @@ -5657,6 +5657,7 @@ by a string of size ``name_size``. #define KVM_STATS_UNIT_BYTES (0x1 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_SECONDS (0x2 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_CYCLES (0x3 << KVM_STATS_UNIT_SHIFT) + #define KVM_STATS_UNIT_BOOLEAN (0x4 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_MAX KVM_STATS_UNIT_CYCLES #define KVM_STATS_BASE_SHIFT 8 @@ -5724,6 +5725,11 @@ Bits 4-7 of ``flags`` encode the unit: It indicates that the statistics data is used to measure time or latency. * ``KVM_STATS_UNIT_CYCLES`` It indicates that the statistics data is used to measure CPU clock cycles. + * ``KVM_STATS_UNIT_BOOLEAN`` + It indicates that the statistic will always be either 0 or 1. Boolean + statistics of "peak" type will never go back from 1 to 0. Boolean + statistics can be linear histograms (with two buckets) but not logarithmic + histograms. Bits 8-11 of ``flags``, together with ``exponent``, encode the scale of the unit: diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 26d0cac32f73..0c3e85e8fce9 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -298,7 +298,7 @@ const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = { STATS_DESC_COUNTER(VCPU, directed_yield_successful), STATS_DESC_COUNTER(VCPU, preemption_reported), STATS_DESC_COUNTER(VCPU, preemption_other), - STATS_DESC_ICOUNTER(VCPU, guest_mode) + STATS_DESC_IBOOLEAN(VCPU, guest_mode) }; const struct kvm_stats_header kvm_vcpu_stats_header = { diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index 83cf7fd842e0..90a45ef7203b 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -1822,6 +1822,15 @@ struct _kvm_stats_desc { STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_NONE, \ KVM_STATS_BASE_POW10, 0) +/* Instantaneous boolean value, read only */ +#define STATS_DESC_IBOOLEAN(SCOPE, name) \ + STATS_DESC_INSTANT(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \ + KVM_STATS_BASE_POW10, 0) +/* Peak (sticky) boolean value, read/write */ +#define STATS_DESC_PBOOLEAN(SCOPE, name) \ + STATS_DESC_PEAK(SCOPE, name, KVM_STATS_UNIT_BOOLEAN, \ + KVM_STATS_BASE_POW10, 0) + /* Cumulative time in nanosecond */ #define STATS_DESC_TIME_NSEC(SCOPE, name) \ STATS_DESC_CUMULATIVE(SCOPE, name, KVM_STATS_UNIT_SECONDS, \ @@ -1853,7 +1862,7 @@ struct _kvm_stats_desc { HALT_POLL_HIST_COUNT), \ STATS_DESC_LOGHIST_TIME_NSEC(VCPU_GENERIC, halt_wait_hist, \ HALT_POLL_HIST_COUNT), \ - STATS_DESC_ICOUNTER(VCPU_GENERIC, blocking) + STATS_DESC_IBOOLEAN(VCPU_GENERIC, blocking) extern struct dentry *kvm_debugfs_dir; diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h index 5088bd9f1922..811897dadcae 100644 --- a/include/uapi/linux/kvm.h +++ b/include/uapi/linux/kvm.h @@ -2083,6 +2083,7 @@ struct kvm_stats_header { #define KVM_STATS_UNIT_BYTES (0x1 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_SECONDS (0x2 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_CYCLES (0x3 << KVM_STATS_UNIT_SHIFT) +#define KVM_STATS_UNIT_BOOLEAN (0x4 << KVM_STATS_UNIT_SHIFT) #define KVM_STATS_UNIT_MAX KVM_STATS_UNIT_CYCLES #define KVM_STATS_BASE_SHIFT 8
Some of the statistics values exported by KVM are always only 0 or 1. It can be useful to export this fact to userspace so that it can track them specially (for example by polling the value every now and then to compute a % of time spent in a specific state). Therefore, add "boolean value" as a new "unit". While it is not exactly a unit, it walks and quacks like one. In particular, using the type would be wrong because boolean values could be instantaneous or peak values (e.g. "is the rmap allocated?") or even two-bucket histograms (e.g. "number of posted vs. non-posted interrupt injections"). Suggested-by: Amneesh Singh <natto@weirdnatto.in> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- Documentation/virt/kvm/api.rst | 6 ++++++ arch/x86/kvm/x86.c | 2 +- include/linux/kvm_host.h | 11 ++++++++++- include/uapi/linux/kvm.h | 1 + 4 files changed, 18 insertions(+), 2 deletions(-)