perf: Multiple task contexts

Provide the infrastructure for multiple task contexts.

A more flexible approach would have resulted in more pointer chases
in the scheduling hot-paths. This approach has the limitation of a
static number of task contexts.

Since I expect most external PMUs to be system wide, or at least node
wide (as per the intel uncore unit) they won't actually need a task
context.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

3 files changed, 239 insertions, 106 deletions

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 22155ef3b362..9ecfd856ce6e 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -572,6 +572,7 @@ struct pmu {
 
         int * __percpu                  pmu_disable_count;
         struct perf_cpu_context * __percpu pmu_cpu_context;
+        int                             task_ctx_nr;
 
         /*
          * Fully disable/enable this PMU, can be used to protect from the PMI
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 1e2a6db2d7dd..89d6023c6f82 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1160,6 +1160,12 @@ struct sched_rt_entity {
 
 struct rcu_node;
 
+enum perf_event_task_context {
+        perf_invalid_context = -1,
+        perf_hw_context = 0,
+        perf_nr_task_contexts,
+};
+
 struct task_struct {
         volatile long state;    /* -1 unrunnable, 0 runnable, >0 stopped */
         void *stack;
@@ -1431,7 +1437,7 @@ struct task_struct {
         struct futex_pi_state *pi_state_cache;
 #endif
 #ifdef CONFIG_PERF_EVENTS
-        struct perf_event_context *perf_event_ctxp;
+        struct perf_event_context *perf_event_ctxp[perf_nr_task_contexts];
         struct mutex perf_event_mutex;
         struct list_head perf_event_list;
 #endif
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 13d98d756347..7223ea875861 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -148,13 +148,13 @@ static u64 primary_event_id(struct perf_event *event)
  * the context could get moved to another task.
  */
 static struct perf_event_context *
-perf_lock_task_context(struct task_struct *task, unsigned long *flags)
+perf_lock_task_context(struct task_struct *task, int ctxn, unsigned long *flags)
 {
         struct perf_event_context *ctx;
 
         rcu_read_lock();
 retry:
-        ctx = rcu_dereference(task->perf_event_ctxp);
+        ctx = rcu_dereference(task->perf_event_ctxp[ctxn]);
         if (ctx) {
                 /*
                  * If this context is a clone of another, it might
@@ -167,7 +167,7 @@ retry:
                  * can't get swapped on us any more.
                  */
                 raw_spin_lock_irqsave(&ctx->lock, *flags);
-                if (ctx != rcu_dereference(task->perf_event_ctxp)) {
+                if (ctx != rcu_dereference(task->perf_event_ctxp[ctxn])) {
                         raw_spin_unlock_irqrestore(&ctx->lock, *flags);
                         goto retry;
                 }
@@ -186,12 +186,13 @@ retry:
  * can't get swapped to another task.  This also increments its
  * reference count so that the context can't get freed.
  */
-static struct perf_event_context *perf_pin_task_context(struct task_struct *task)
+static struct perf_event_context *
+perf_pin_task_context(struct task_struct *task, int ctxn)
 {
         struct perf_event_context *ctx;
         unsigned long flags;
 
-        ctx = perf_lock_task_context(task, &flags);
+        ctx = perf_lock_task_context(task, ctxn, &flags);
         if (ctx) {
                 ++ctx->pin_count;
                 raw_spin_unlock_irqrestore(&ctx->lock, flags);
@@ -1179,28 +1180,15 @@ static void perf_event_sync_stat(struct perf_event_context *ctx,
         }
 }
 
-/*
- * Called from scheduler to remove the events of the current task,
- * with interrupts disabled.
- *
- * We stop each event and update the event value in event->count.
- *
- * This does not protect us against NMI, but disable()
- * sets the disabled bit in the control field of event _before_
- * accessing the event control register. If a NMI hits, then it will
- * not restart the event.
- */
-void perf_event_task_sched_out(struct task_struct *task,
-                                 struct task_struct *next)
+void perf_event_context_sched_out(struct task_struct *task, int ctxn,
+                                  struct task_struct *next)
 {
-        struct perf_event_context *ctx = task->perf_event_ctxp;
+        struct perf_event_context *ctx = task->perf_event_ctxp[ctxn];
         struct perf_event_context *next_ctx;
         struct perf_event_context *parent;
         struct perf_cpu_context *cpuctx;
         int do_switch = 1;
 
-        perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0);
-
         if (likely(!ctx))
                 return;
 
@@ -1210,7 +1198,7 @@ void perf_event_task_sched_out(struct task_struct *task,
 
         rcu_read_lock();
         parent = rcu_dereference(ctx->parent_ctx);
-        next_ctx = next->perf_event_ctxp;
+        next_ctx = next->perf_event_ctxp[ctxn];
         if (parent && next_ctx &&
             rcu_dereference(next_ctx->parent_ctx) == parent) {
                 /*
@@ -1229,8 +1217,8 @@ void perf_event_task_sched_out(struct task_struct *task,
                          * XXX do we need a memory barrier of sorts
                          * wrt to rcu_dereference() of perf_event_ctxp
                          */
-                        task->perf_event_ctxp = next_ctx;
-                        next->perf_event_ctxp = ctx;
+                        task->perf_event_ctxp[ctxn] = next_ctx;
+                        next->perf_event_ctxp[ctxn] = ctx;
                         ctx->task = next;
                         next_ctx->task = task;
                         do_switch = 0;
@@ -1248,6 +1236,31 @@ void perf_event_task_sched_out(struct task_struct *task,
         }
 }
 
+#define for_each_task_context_nr(ctxn)                                  \
+        for ((ctxn) = 0; (ctxn) < perf_nr_task_contexts; (ctxn)++)
+
+/*
+ * Called from scheduler to remove the events of the current task,
+ * with interrupts disabled.
+ *
+ * We stop each event and update the event value in event->count.
+ *
+ * This does not protect us against NMI, but disable()
+ * sets the disabled bit in the control field of event _before_
+ * accessing the event control register. If a NMI hits, then it will
+ * not restart the event.
+ */
+void perf_event_task_sched_out(struct task_struct *task,
+                               struct task_struct *next)
+{
+        int ctxn;
+
+        perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0);
+
+        for_each_task_context_nr(ctxn)
+                perf_event_context_sched_out(task, ctxn, next);
+}
+
 static void task_ctx_sched_out(struct perf_event_context *ctx,
                                enum event_type_t event_type)
 {
@@ -1366,38 +1379,23 @@ static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,