perf counters: add support for group counters

Impact: add group counters

This patch adds the "counter groups" abstraction.

Groups of counters behave much like normal 'single' counters, with a
few semantic and behavioral extensions on top of that.

A counter group is created by creating a new counter with the open()
syscall's group-leader group_fd file descriptor parameter pointing
to another, already existing counter.

Groups of counters are scheduled in and out in one atomic group, and
they are also roundrobin-scheduled atomically.

Counters that are member of a group can also record events with an
(atomic) extended timestamp that extends to all members of the group,
if the record type is set to PERF_RECORD_GROUP.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

arch/x86/kernel/cpu/perf_counter.c

@@ -346,18 +346,22 @@ static void perf_save_and_restart(struct perf_counter *counter)
 }
 
 static void
-perf_handle_group(struct perf_counter *leader, u64 *status, u64 *overflown)
+perf_handle_group(struct perf_counter *sibling, u64 *status, u64 *overflown)
 {
-	struct perf_counter_context *ctx = leader->ctx;
-	struct perf_counter *counter;
+	struct perf_counter *counter, *group_leader = sibling->group_leader;
 	int bit;
 
-	list_for_each_entry(counter, &ctx->counters, list) {
-		if (counter->hw_event.record_type != PERF_RECORD_SIMPLE ||
-		    counter == leader)
-			continue;
+	/*
+	 * Store the counter's own timestamp first:
+	 */
+	perf_store_irq_data(sibling, sibling->hw_event.type);
+	perf_store_irq_data(sibling, atomic64_counter_read(sibling));
 
-		if (counter->active) {
+	/*
+	 * Then store sibling timestamps (if any):
+	 */
+	list_for_each_entry(counter, &group_leader->sibling_list, list_entry) {
+		if (!counter->active) {
 			/*
 			 * When counter was not in the overflow mask, we have to
 			 * read it from hardware. We read it as well, when it
@@ -371,8 +375,8 @@ perf_handle_group(struct perf_counter *leader, u64 *status, u64 *overflown)
 				perf_save_and_restart(counter);
 			}
 		}
-		perf_store_irq_data(leader, counter->hw_event.type);
-		perf_store_irq_data(leader, atomic64_counter_read(counter));
+		perf_store_irq_data(sibling, counter->hw_event.type);
+		perf_store_irq_data(sibling, atomic64_counter_read(counter));
 	}
 }
 
@@ -416,10 +420,6 @@ static void __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
 			perf_store_irq_data(counter, instruction_pointer(regs));
 			break;
 		case PERF_RECORD_GROUP:
-			perf_store_irq_data(counter,
-					    counter->hw_event.type);
-			perf_store_irq_data(counter,
-					    atomic64_counter_read(counter));
 			perf_handle_group(counter, &status, &ack);
 			break;
 		}

include/linux/perf_counter.h

@@ -117,7 +117,10 @@ struct perf_data {
  * struct perf_counter - performance counter kernel representation:
  */
 struct perf_counter {
-	struct list_head		list;
+	struct list_head		list_entry;
+	struct list_head		sibling_list;
+	struct perf_counter		*group_leader;
+
 	int				active;
 #if BITS_PER_LONG == 64
 	atomic64_t			count;
@@ -158,7 +161,8 @@ struct perf_counter_context {
 	 * Protect the list of counters:
 	 */
 	spinlock_t		lock;
-	struct list_head	counters;
+
+	struct list_head	counter_list;
 	int			nr_counters;
 	int			nr_active;
 	struct task_struct	*task;