perf/core: Rename the perf_event_aux*() APIs to perf_event_sb*(), to separate them...
authorPeter Zijlstra <peterz@infradead.org>
Thu, 12 May 2016 15:26:46 +0000 (17:26 +0200)
committerIngo Molnar <mingo@kernel.org>
Fri, 3 Jun 2016 07:40:15 +0000 (09:40 +0200)
There are now two different things called AUX in perf, the
infrastructure to deliver the mmap/comm/task records and the
AUX part in the mmap buffer (with associated AUX_RECORD).

Since the former is internal, rename it to side-band to reduce
the confusion factor.

No change in functionality.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c

index 6615c89..f54454e 100644 (file)
@@ -5879,11 +5879,11 @@ perf_event_read_event(struct perf_event *event,
        perf_output_end(&handle);
 }
 
-typedef void (perf_event_aux_output_cb)(struct perf_event *event, void *data);
+typedef void (perf_iterate_f)(struct perf_event *event, void *data);
 
 static void
-perf_event_aux_ctx(struct perf_event_context *ctx,
-                  perf_event_aux_output_cb output,
+perf_iterate_ctx(struct perf_event_context *ctx,
+                  perf_iterate_f output,
                   void *data, bool all)
 {
        struct perf_event *event;
@@ -5900,18 +5900,7 @@ perf_event_aux_ctx(struct perf_event_context *ctx,
        }
 }
 
-static void
-perf_event_aux_task_ctx(perf_event_aux_output_cb output, void *data,
-                       struct perf_event_context *task_ctx)
-{
-       rcu_read_lock();
-       preempt_disable();
-       perf_event_aux_ctx(task_ctx, output, data, false);
-       preempt_enable();
-       rcu_read_unlock();
-}
-
-static void perf_event_sb_iterate(perf_event_aux_output_cb output, void *data)
+static void perf_iterate_sb_cpu(perf_iterate_f output, void *data)
 {
        struct pmu_event_list *pel = this_cpu_ptr(&pmu_sb_events);
        struct perf_event *event;
@@ -5925,33 +5914,40 @@ static void perf_event_sb_iterate(perf_event_aux_output_cb output, void *data)
        }
 }
 
+/*
+ * Iterate all events that need to receive side-band events.
+ *
+ * For new callers; ensure that account_pmu_sb_event() includes
+ * your event, otherwise it might not get delivered.
+ */
 static void
-perf_event_aux(perf_event_aux_output_cb output, void *data,
+perf_iterate_sb(perf_iterate_f output, void *data,
               struct perf_event_context *task_ctx)
 {
        struct perf_event_context *ctx;
        int ctxn;
 
+       rcu_read_lock();
+       preempt_disable();
+
        /*
-        * If we have task_ctx != NULL we only notify
-        * the task context itself. The task_ctx is set
-        * only for EXIT events before releasing task
+        * If we have task_ctx != NULL we only notify the task context itself.
+        * The task_ctx is set only for EXIT events before releasing task
         * context.
         */
        if (task_ctx) {
-               perf_event_aux_task_ctx(output, data, task_ctx);
-               return;
+               perf_iterate_ctx(task_ctx, output, data, false);
+               goto done;
        }
 
-       rcu_read_lock();
-       preempt_disable();
-       perf_event_sb_iterate(output, data);
+       perf_iterate_sb_cpu(output, data);
 
        for_each_task_context_nr(ctxn) {
                ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
                if (ctx)
-                       perf_event_aux_ctx(ctx, output, data, false);
+                       perf_iterate_ctx(ctx, output, data, false);
        }
+done:
        preempt_enable();
        rcu_read_unlock();
 }
@@ -6001,7 +5997,7 @@ void perf_event_exec(void)
 
                perf_event_enable_on_exec(ctxn);
 
-               perf_event_aux_ctx(ctx, perf_event_addr_filters_exec, NULL,
+               perf_iterate_ctx(ctx, perf_event_addr_filters_exec, NULL,
                                   true);
        }
        rcu_read_unlock();
@@ -6045,9 +6041,9 @@ static int __perf_pmu_output_stop(void *info)
        };
 
        rcu_read_lock();
-       perf_event_aux_ctx(&cpuctx->ctx, __perf_event_output_stop, &ro, false);
+       perf_iterate_ctx(&cpuctx->ctx, __perf_event_output_stop, &ro, false);
        if (cpuctx->task_ctx)
-               perf_event_aux_ctx(cpuctx->task_ctx, __perf_event_output_stop,
+               perf_iterate_ctx(cpuctx->task_ctx, __perf_event_output_stop,
                                   &ro, false);
        rcu_read_unlock();
 
@@ -6176,7 +6172,7 @@ static void perf_event_task(struct task_struct *task,
                },
        };
 
-       perf_event_aux(perf_event_task_output,
+       perf_iterate_sb(perf_event_task_output,
                       &task_event,
                       task_ctx);
 }
@@ -6255,7 +6251,7 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
 
        comm_event->event_id.header.size = sizeof(comm_event->event_id) + size;
 
-       perf_event_aux(perf_event_comm_output,
+       perf_iterate_sb(perf_event_comm_output,
                       comm_event,
                       NULL);
 }
@@ -6486,7 +6482,7 @@ got_name:
 
        mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size;
 
-       perf_event_aux(perf_event_mmap_output,
+       perf_iterate_sb(perf_event_mmap_output,
                       mmap_event,
                       NULL);
 
@@ -6569,7 +6565,7 @@ static void perf_addr_filters_adjust(struct vm_area_struct *vma)
                if (!ctx)
                        continue;
 
-               perf_event_aux_ctx(ctx, __perf_addr_filters_adjust, vma, true);
+               perf_iterate_ctx(ctx, __perf_addr_filters_adjust, vma, true);
        }
        rcu_read_unlock();
 }
@@ -6756,7 +6752,7 @@ static void perf_event_switch(struct task_struct *task,
                },
        };
 
-       perf_event_aux(perf_event_switch_output,
+       perf_iterate_sb(perf_event_switch_output,
                       &switch_event,
                       NULL);
 }
@@ -8654,6 +8650,13 @@ static void attach_sb_event(struct perf_event *event)
        raw_spin_unlock(&pel->lock);
 }
 
+/*
+ * We keep a list of all !task (and therefore per-cpu) events
+ * that need to receive side-band records.
+ *
+ * This avoids having to scan all the various PMU per-cpu contexts
+ * looking for them.
+ */
 static void account_pmu_sb_event(struct perf_event *event)
 {
        struct perf_event_attr *attr = &event->attr;