Create one perf structure to rule them all

* perf.h (stap_perf_probe): Define one structure for all our needs.
* perf.c (_stp_perf_init): Use the new struct, return errors directly.
  (_stp_perf_exit): Use the new struct.
* tapset-perfmon.cxx (perf_derived_probe_group::emit_module_decls): Use
  the stap_perf_probe definition from perf.h now.
  (perf_derived_probe_group::emit_module_init): Adapt.
  (perf_derived_probe_group::emit_module_exit): Adapt.

3 files changed, 38 insertions, 77 deletions

diff --git a/runtime/perf.c b/runtime/perf.c
index 12bf220a..bf31cdaa 100644
--- a/runtime/perf.c
+++ b/runtime/perf.c
@@ -22,84 +22,59 @@
 
 /** Initialize performance sampling
  * Call this during probe initialization to set up performance event sampling
- * for all online cpus.  Returns ERR_PTR on error.
+ * for all online cpus.  Returns non-zero on error.
  *
- * @param attr description of event to sample
- * @param callback function to call when perf event overflows
- * @param pp associated probe point
- * @param ph probe handler
+ * @param stp Handle for the event to be registered.
  */
-static Perf *_stp_perf_init (struct perf_event_attr *attr,
-			     perf_overflow_handler_t callback,
-			     const char *pp, void (*ph) (struct context *) )
+static long _stp_perf_init (struct stap_perf_probe *stp)
 {
-	long rc = -EINVAL;
 	int cpu;
-	Perf *pe;
-
-	pe = (Perf *) _stp_kmalloc (sizeof(Perf));
-	if (pe == NULL)
-		return ERR_PTR(-ENOMEM);
 
 	/* allocate space for the event descriptor for each cpu */
-	pe->pd = (perfcpu *) _stp_alloc_percpu (sizeof(perfcpu));
-	if (pe->pd == NULL) {
-		rc = -ENOMEM;
-		goto exit1;
+	stp->events = _stp_alloc_percpu (sizeof(struct perf_event*));
+	if (stp->events == NULL) {
+		return -ENOMEM;
 	}
 
 	/* initialize event on each processor */
 	stp_for_each_cpu(cpu) {
-		perfcpu *pd = per_cpu_ptr (pe->pd, cpu);
-		struct perf_event **event = &(pd->event);
+		struct perf_event **event = per_cpu_ptr (stp->events, cpu);
 		if (cpu_is_offline(cpu)) {
-		 	*event = NULL;
+			*event = NULL;
 			continue;
 		}
-		*event = perf_event_create_kernel_counter(attr, cpu, -1,
-							  callback);
+		*event = perf_event_create_kernel_counter(&stp->attr,
+							  cpu, -1,
+							  stp->callback);
 
 		if (IS_ERR(*event)) {
-			rc = PTR_ERR(*event);
+			long rc = PTR_ERR(*event);
 			*event = NULL;
-			goto exit2;
+			_stp_perf_del(stp);
+			return rc;
 		}
-		pd->pp = pp;
-		pd->ph = ph;
-	}
-	return pe;
-
-exit2:
-	stp_for_each_cpu(cpu) {
-		perfcpu *pd = per_cpu_ptr (pe->pd, cpu);
-		struct perf_event **event = &(pd->event);
-		if (*event) perf_event_release_kernel(*event);
 	}
-	_stp_free_percpu(pe->pd);
-exit1:
-	_stp_kfree(pe);
-	return ERR_PTR(rc);
+	return 0;
 }
 
 /** Delete performance event.
  * Call this to shutdown performance event sampling
  *
- * @param pe
+ * @param stp Handle for the event to be unregistered.
  */
-static void _stp_perf_del (Perf *pe)
+static void _stp_perf_del (struct stap_perf_probe *stp)
 {
-	if (pe) {
+	if (stp && stp->events) {
 		int cpu;
 		/* shut down performance event sampling */
 		stp_for_each_cpu(cpu) {
-			perfcpu *pd = per_cpu_ptr (pe->pd, cpu);
-			struct perf_event **event = &(pd->event);
+			struct perf_event **event = per_cpu_ptr (stp->events, cpu);
 			if (*event) {
 				perf_event_release_kernel(*event);
 			}
 		}
-		_stp_free_percpu (pe->pd);
-		_stp_kfree (pe);
+		_stp_free_percpu (stp->events);
+		stp->events = NULL;
 	}
 }
 
diff --git a/runtime/perf.h b/runtime/perf.h
index 19f89fe7..bdc79ade 100644
--- a/runtime/perf.h
+++ b/runtime/perf.h
@@ -16,22 +16,18 @@
  * @brief Header file for performance monitoring hardware support
  */
 
-typedef struct {
-	struct perf_event *event;
+struct stap_perf_probe {
+	struct perf_event_attr attr;
+	perf_overflow_handler_t callback;
 	const char *pp;
-	void (*ph) (struct context *);
-} perfcpu;
+	void (*ph) (struct context*);
 
-typedef struct {
 	/* per-cpu data. allocated with _stp_alloc_percpu() */
-	perfcpu *pd;
-} Perf;
+	struct perf_event **events;
+};
 
-static Perf *_stp_perf_init (struct perf_event_attr *attr,
-			     perf_overflow_handler_t callback,
-			     const char *pp,
-			     void (*ph) (struct context *) );
+static long _stp_perf_init (struct stap_perf_probe *stp);
 
-static void _stp_perf_del (Perf *pe);
+static void _stp_perf_del (struct stap_perf_probe *stp);
 
 #endif /* _PERF_H_ */
diff --git a/tapset-perfmon.cxx b/tapset-perfmon.cxx
index 9078dfd9..98bc8fe6 100644
--- a/tapset-perfmon.cxx
+++ b/tapset-perfmon.cxx
@@ -94,13 +94,8 @@ perf_derived_probe_group::emit_module_decls (systemtap_session& s)
   s.op->newline();
 
   /* data structures */
-  s.op->newline() << "static struct stap_perf_probe {";
-  s.op->newline(1) << "struct perf_event_attr attr;";
-  s.op->newline() << "perf_overflow_handler_t cb;";
-  s.op->newline() << "const char *pp;";
-  s.op->newline() << "void (*ph) (struct context*);";
-  s.op->newline() << "Perf *perf;";
-  s.op->newline(-1) << "} stap_perf_probes [" << probes.size() << "] = {";
+  s.op->newline() << "static struct stap_perf_probe stap_perf_probes ["
+                  << probes.size() << "] = {";
   s.op->indent(1);
   for (unsigned i=0; i < probes.size(); i++)
     {
@@ -109,7 +104,7 @@ perf_derived_probe_group::emit_module_decls (systemtap_session& s)
                        << ".type=" << probes[i]->event_type << "ULL, "
                        << ".config=" << probes[i]->event_config << "ULL, "
                        << "{ .sample_period=" << probes[i]->interval << "ULL }},";
-      s.op->newline() << ".cb=enter_perf_probe_" << i << ", ";
+      s.op->newline() << ".callback=enter_perf_probe_" << i << ", ";
       s.op->newline() << ".pp=" << lex_cast_qstring (*probes[i]->sole_location()) << ",";
       s.op->newline() << ".ph=" << probes[i]->name << ",";
       s.op->newline(-1) << "},";
@@ -150,17 +145,14 @@ perf_derived_probe_group::emit_module_init (systemtap_session& s)
 
   s.op->newline() << "for (i=0; i<" << probes.size() << "; i++) {";
   s.op->newline(1) << "struct stap_perf_probe* stp = & stap_perf_probes [i];";
-  s.op->newline() << "stp->perf = _stp_perf_init(&stp->attr, stp->cb, stp->pp, stp->ph);";
-  s.op->newline() << "if (IS_ERR(stp->perf)) {";
-  s.op->newline(1) << "rc = PTR_ERR(stp->perf);";
-  s.op->newline() << "stp->perf = NULL;";
-  s.op->newline() << "probe_point = stp->pp;";
+  s.op->newline() << "rc = _stp_perf_init(stp);";
+  s.op->newline() << "if (rc) {";
+  s.op->newline(1) << "probe_point = stp->pp;";
   s.op->newline() << "for (j=0; j<i; j++) {";
-  s.op->newline(1) << "stp = & stap_perf_probes [j];";
-  s.op->newline() << "_stp_perf_del(stp->perf);";
-  s.op->newline() << "stp->perf = NULL;";
+  s.op->newline(1) << "_stp_perf_del(& stap_perf_probes [j]);";
   s.op->newline(-1) << "}"; // for unwind loop
   s.op->newline(-1) << "}"; // if-error
+  s.op->newline() << "break;";
   s.op->newline(-1) << "}"; // for loop
 }
 
@@ -171,9 +163,7 @@ perf_derived_probe_group::emit_module_exit (systemtap_session& s)
   if (probes.empty()) return;
 
   s.op->newline() << "for (i=0; i<" << probes.size() << "; i++) {";
-  s.op->newline(1) << "struct stap_perf_probe* stp = & stap_perf_probes [i];";
-  s.op->newline() << "_stp_perf_del(stp->perf);";
-  s.op->newline() << "stp->perf = NULL;";
+  s.op->newline(1) << "_stp_perf_del(& stap_perf_probes [i]);";
   s.op->newline(-1) << "}"; // for loop
 }