perf/x86: Remove perf_events_lapic_init() calling from x86_pmu_enable()
Subsystems (2)
PATCH RESEND 1/2 perf/x86: Remove perf_events_lapic_init() calling from x86_pmu_enable()
perf_events_lapic_init() helper is called to configure PMI to NMI vector and clear MASK bit simultaneously by writing APIC_LVTPC MSR. It's called firstly to initialize APIC_LVTPC MSR by init_hw_perf_events(), and the PMI handler would always to clear the MASK bit in APIC_LVTPC MSR by writing APIC_LVTPC MSR directly.
So it becomes unnecessary to call perf_events_lapic_init() again in x86_pmu_enable(), and worse x86_pmu_enable() could be called very frequently in some scenarios with very high context-switches. This would cause performance overhead which can't be ignored especially in KVM guest environment since frequent APIC_LVTPC writing would cause huge number of VM-Exits.
For example, in guest environment Geekbench score (running multiplxing perf-stat command in background) increases 1% and perf-sched benchmark increases 7% after removing perf_events_lapic_init() calling from x86_pmu_enable().
--- arch/x86/events/core.c | 1 - 1 file changed, 1 deletion(-)
arch/x86/events/core.c +0 -1
@@ -1347,7 +1347,6 @@ static void x86_pmu_enable(struct pmu *pmu)x86_pmu_start (event,PERF_EF_RELOAD ); } cpuc->n_added =0 ; -perf_events_lapic_init (); } cpuc->enabled =1 ;
base-commit: 73e931504f8e0d42978bfcda37b323dbbd1afc08
Signature · 1 line
2.40.1
PATCH RESEND 2/2 perf/x86: Typos and invalid indents fix
Fix several typos and invalid indents.
--- arch/x86/events/intel/core.c | 2 +- arch/x86/include/asm/perf_event.h | 6 +++--- 2 files changed, 4 insertions(+), 4 deletions(-)
arch/x86/events/intel/core.c +1 -1
@@ -3697,7 +3697,7 @@ static void intel_put_event_constraints(struct cpu_hw_events *cpuc,intel_put_shared_regs_event_constraints (cpuc, event);/* -* is PMU has exclusive counter restrictions, then +* If PMU has exclusive counter restrictions, then * all events are subject to and must call the * put_excl_constraints() routine */
arch/x86/include/asm/perf_event.h +3 -3
@@ -399,15 +399,15 @@ static inline bool is_topdown_idx(int idx)* * With this fake counter assigned, the guest LBR event user (such as KVM), * can program the LBR registers on its own, and we don't actually do anything -* with then in the host context. +* with them in the host context. */ -#define INTEL_PMC_IDX_FIXED_VLBR (GLOBAL_STATUS_LBRS_FROZEN_BIT) +#define INTEL_PMC_IDX_FIXED_VLBR (GLOBAL_STATUS_LBRS_FROZEN_BIT)/* * Pseudo-encoding the guest LBR event as event=0x00,umask=0x1b, * since it would claim bit 58 which is effectively Fixed26. */ -#define INTEL_FIXED_VLBR_EVENT 0x1b00 +#define INTEL_FIXED_VLBR_EVENT 0x1b00/* * Adaptive PEBS v4
Signature · 1 line
2.40.1
Peter Zijlstra
maintainerQuoted 28 lines · Dapeng Mi
perf_events_lapic_init()helper is called to configure PMI to NMI vector and clear MASK bit simultaneously by writingAPIC_LVTPCMSR. It's called firstly to initializeAPIC_LVTPCMSR byinit_hw_perf_events(), and the PMI handler would always to clear the MASK bit inAPIC_LVTPCMSR by writingAPIC_LVTPCMSR directly.So it becomes unnecessary to call
perf_events_lapic_init()again inx86_pmu_enable(), and worsex86_pmu_enable()could be called very frequently in some scenarios with very high context-switches. This would cause performance overhead which can't be ignored especially in KVM guest environment since frequentAPIC_LVTPCwriting would cause huge number of VM-Exits.For example, in guest environment Geekbench score (running multiplxing perf-stat command in background) increases 1% and perf-sched benchmark increases 7% after removing
perf_events_lapic_init()calling fromx86_pmu_enable().Signed-off-by Dapeng Mi <[email protected]>--- arch/x86/events/core.c | 1 - 1 file changed, 1 deletion(-)arch/x86/events/core.c +0
-0
arch/x86/events/core.c static void x86_pmu_enable(struct pmu *pmu) · line 1350
Quoted 5 lines
arch/x86/events/core.c +0
-1@@ -1347,7 +1347,6 @@ static void x86_pmu_enable(struct pmu *pmu)x86_pmu_start (event,PERF_EF_RELOAD ); } cpuc->n_added =0 ; -perf_events_lapic_init (); }
Stephane, I don't suppose you remember why you put that there? Afaict the above reasoning is correct and this one is entirely superfluous, but this hardware is 'funny' at the best of times and maybe I'm overlooking something.