perf_counter: Optimize perf_swcounter_event()

Similar to tracepoints, use an enable variable to reduce
overhead when unused.

Only look for a counter of a particular event type when we know
there is at least one in the system.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 89698d8aba5c1dc1f92dbc6cbf6addd8f96790f4..e7213e46cf9c5a635c7496e7f3eae26bf048ce3d 100644 (file)
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -669,7 +669,16 @@ static inline int is_software_counter(struct perf_counter *counter)
                (counter->attr.type != PERF_TYPE_HW_CACHE);
 }
 
-extern void perf_swcounter_event(u32, u64, int, struct pt_regs *, u64);
+extern atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX];
+
+extern void __perf_swcounter_event(u32, u64, int, struct pt_regs *, u64);
+
+static inline void
+perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
+{
+       if (atomic_read(&perf_swcounter_enabled[event]))
+               __perf_swcounter_event(event, nr, nmi, regs, addr);
+}
 
 extern void __perf_counter_mmap(struct vm_area_struct *vma);
 

diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 1a933a221ea4a299922f4db98d62de86c7e5ea0b..7515c769542816ad2f34165e46c586b23550c82b 100644 (file)
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -3317,8 +3317,8 @@ out:
        put_cpu_var(perf_cpu_context);
 }
 
-void
-perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
+void __perf_swcounter_event(u32 event, u64 nr, int nmi,
+                           struct pt_regs *regs, u64 addr)
 {
        struct perf_sample_data data = {
                .regs = regs,
@@ -3509,9 +3509,19 @@ static const struct pmu *tp_perf_counter_init(struct perf_counter *counter)
 }
 #endif
 
+atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX];
+
+static void sw_perf_counter_destroy(struct perf_counter *counter)
+{
+       u64 event = counter->attr.config;
+
+       atomic_dec(&perf_swcounter_enabled[event]);
+}
+
 static const struct pmu *sw_perf_counter_init(struct perf_counter *counter)
 {
        const struct pmu *pmu = NULL;
+       u64 event = counter->attr.config;
 
        /*
         * Software counters (currently) can't in general distinguish
@@ -3520,7 +3530,7 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter)
         * to be kernel events, and page faults are never hypervisor
         * events.
         */
-       switch (counter->attr.config) {
+       switch (event) {
        case PERF_COUNT_SW_CPU_CLOCK:
                pmu = &perf_ops_cpu_clock;
 
@@ -3541,6 +3551,8 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter)
        case PERF_COUNT_SW_PAGE_FAULTS_MAJ:
        case PERF_COUNT_SW_CONTEXT_SWITCHES:
        case PERF_COUNT_SW_CPU_MIGRATIONS:
+               atomic_inc(&perf_swcounter_enabled[event]);
+               counter->destroy = sw_perf_counter_destroy;
                pmu = &perf_ops_generic;
                break;
        }