Merge branch 'tracing/hw-branch-tracing' into tracing/core

Merge reason: this topic is ready for upstream now. It passed
              Oleg's review and Andrew had no further mm/*
              objections/observations either.

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

8 files changed, 214 insertions, 110 deletions

diff --git a/kernel/Makefile b/kernel/Makefile
index 42423665660..a35eee3436d 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -93,6 +93,7 @@ obj-$(CONFIG_LATENCYTOP) += latencytop.o
 obj-$(CONFIG_HAVE_GENERIC_DMA_COHERENT) += dma-coherent.o
 obj-$(CONFIG_FUNCTION_TRACER) += trace/
 obj-$(CONFIG_TRACING) += trace/
+obj-$(CONFIG_X86_DS) += trace/
 obj-$(CONFIG_SMP) += sched_cpupri.o
 obj-$(CONFIG_SLOW_WORK) += slow-work.o
 
diff --git a/kernel/fork.c b/kernel/fork.c
index 085f73ebcea..711468f3db2 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1088,8 +1088,8 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 #ifdef CONFIG_DEBUG_MUTEXES
 	p->blocked_on = NULL; /* not blocked yet */
 #endif
-	if (unlikely(current->ptrace))
-		ptrace_fork(p, clone_flags);
+
+	p->bts = NULL;
 
 	/* Perform scheduler related setup. Assign this task to a CPU. */
 	sched_fork(p, clone_flags);
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index 0692ab5a0d6..e950805f863 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -25,16 +25,6 @@
 
 
 /*
- * Initialize a new task whose father had been ptraced.
- *
- * Called from copy_process().
- */
-void ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
-	arch_ptrace_fork(child, clone_flags);
-}
-
-/*
  * ptrace a task: make the debugger its new parent and
  * move it to the ptrace list.
  *
diff --git a/kernel/sched.c b/kernel/sched.c
index 14a19b17674..6530a27052f 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2011,6 +2011,49 @@ migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req)
 }
 
 /*
+ * wait_task_context_switch -	wait for a thread to complete at least one
+ *				context switch.
+ *
+ * @p must not be current.
+ */
+void wait_task_context_switch(struct task_struct *p)
+{
+	unsigned long nvcsw, nivcsw, flags;
+	int running;
+	struct rq *rq;
+
+	nvcsw	= p->nvcsw;
+	nivcsw	= p->nivcsw;
+	for (;;) {
+		/*
+		 * The runqueue is assigned before the actual context
+		 * switch. We need to take the runqueue lock.
+		 *
+		 * We could check initially without the lock but it is
+		 * very likely that we need to take the lock in every
+		 * iteration.
+		 */
+		rq = task_rq_lock(p, &flags);
+		running = task_running(rq, p);
+		task_rq_unlock(rq, &flags);
+
+		if (likely(!running))
+			break;
+		/*
+		 * The switch count is incremented before the actual
+		 * context switch. We thus wait for two switches to be
+		 * sure at least one completed.
+		 */
+		if ((p->nvcsw - nvcsw) > 1)
+			break;
+		if ((p->nivcsw - nivcsw) > 1)
+			break;
+
+		cpu_relax();
+	}
+}
+
+/*
  * wait_task_inactive - wait for a thread to unschedule.
  *
  * If @match_state is nonzero, it's the @p->state value just checked and
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 7c34cbfff96..06b85850fab 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -15,12 +15,17 @@ ifdef CONFIG_TRACING_BRANCHES
 KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
 endif
 
+#
+# Make the trace clocks available generally: it's infrastructure
+# relied on by ptrace for example:
+#
+obj-y += trace_clock.o
+
 obj-$(CONFIG_FUNCTION_TRACER) += libftrace.o
 obj-$(CONFIG_RING_BUFFER) += ring_buffer.o
 obj-$(CONFIG_RING_BUFFER_BENCHMARK) += ring_buffer_benchmark.o
 
 obj-$(CONFIG_TRACING) += trace.o
-obj-$(CONFIG_TRACING) += trace_clock.o
 obj-$(CONFIG_TRACING) += trace_output.o
 obj-$(CONFIG_TRACING) += trace_stat.o
 obj-$(CONFIG_TRACING) += trace_printk.o
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index ba25793ffe6..6e735d4771f 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -538,6 +538,8 @@ extern int trace_selftest_startup_sysprof(struct tracer *trace,
 					       struct trace_array *tr);
 extern int trace_selftest_startup_branch(struct tracer *trace,
 					 struct trace_array *tr);
+extern int trace_selftest_startup_hw_branches(struct tracer *trace,
+					      struct trace_array *tr);
 #endif /* CONFIG_FTRACE_STARTUP_TEST */
 
 extern void *head_page(struct trace_array_cpu *data);
diff --git a/kernel/trace/trace_hw_branches.c b/kernel/trace/trace_hw_branches.c
index 8683d50a753..ca7d7c4d0c2 100644
--- a/kernel/trace/trace_hw_branches.c
+++ b/kernel/trace/trace_hw_branches.c
@@ -1,10 +1,9 @@
 /*
- * h/w branch tracer for x86 based on bts
+ * h/w branch tracer for x86 based on BTS
  *
  * Copyright (C) 2008-2009 Intel Corporation.
  * Markus Metzger <markus.t.metzger@gmail.com>, 2008-2009
  */
-#include <linux/spinlock.h>
 #include <linux/kallsyms.h>
 #include <linux/debugfs.h>
 #include <linux/ftrace.h>
@@ -15,110 +14,119 @@
 
 #include <asm/ds.h>
 
-#include "trace.h"
 #include "trace_output.h"
+#include "trace.h"
 
 
-#define SIZEOF_BTS (1 << 13)
+#define BTS_BUFFER_SIZE (1 << 13)
 
-/*
- * The tracer lock protects the below per-cpu tracer array.
- * It needs to be held to:
- * - start tracing on all cpus
- * - stop tracing on all cpus
- * - start tracing on a single hotplug cpu
- * - stop tracing on a single hotplug cpu
- * - read the trace from all cpus
- * - read the trace from a single cpu
- */
-static DEFINE_SPINLOCK(bts_tracer_lock);
 static DEFINE_PER_CPU(struct bts_tracer *, tracer);
-static DEFINE_PER_CPU(unsigned char[SIZEOF_BTS], buffer);
+static DEFINE_PER_CPU(unsigned char[BTS_BUFFER_SIZE], buffer);
 
 #define this_tracer per_cpu(tracer, smp_processor_id())
-#define this_buffer per_cpu(buffer, smp_processor_id())
 
-static int __read_mostly trace_hw_branches_enabled;
+static int trace_hw_branches_enabled __read_mostly;
+static int trace_hw_branches_suspended __read_mostly;
 static struct trace_array *hw_branch_trace __read_mostly;
 
 
-/*
- * Start tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_start_cpu(void *arg)
+static void bts_trace_init_cpu(int cpu)
 {
-	if (this_tracer)
-		ds_release_bts(this_tracer);
-
-	this_tracer =
-		ds_request_bts(/* task = */ NULL, this_buffer, SIZEOF_BTS,
-			       /* ovfl = */ NULL, /* th = */ (size_t)-1,
-			       BTS_KERNEL);
-	if (IS_ERR(this_tracer)) {
-		this_tracer = NULL;
-		return;
-	}
+	per_cpu(tracer, cpu) =
+		ds_request_bts_cpu(cpu, per_cpu(buffer, cpu), BTS_BUFFER_SIZE,
+				   NULL, (size_t)-1, BTS_KERNEL);
+
+	if (IS_ERR(per_cpu(tracer, cpu)))
+		per_cpu(tracer, cpu) = NULL;
 }
 
-static void bts_trace_start(struct trace_array *tr)
+static int bts_trace_init(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
+
+	hw_branch_trace = tr;
+	trace_hw_branches_enabled = 0;
 
-	on_each_cpu(bts_trace_start_cpu, NULL, 1);
-	trace_hw_branches_enabled = 1;
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		bts_trace_init_cpu(cpu);
 
-	spin_unlock(&bts_tracer_lock);
+		if (likely(per_cpu(tracer, cpu)))
+			trace_hw_branches_enabled = 1;
+	}
+	trace_hw_branches_suspended = 0;
+	put_online_cpus();
+
+	/* If we could not enable tracing on a single cpu, we fail. */
+	return trace_hw_branches_enabled ? 0 : -EOPNOTSUPP;
 }
 
-/*
- * Stop tracing on the current cpu.
- * The argument is ignored.
- *
- * pre: bts_tracer_lock must be locked.
- */
-static void bts_trace_stop_cpu(void *arg)
+static void bts_trace_reset(struct trace_array *tr)
 {
-	if (this_tracer) {
-		ds_release_bts(this_tracer);
-		this_tracer = NULL;
+	int cpu;
+
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
 	}
+	trace_hw_branches_enabled = 0;
+	trace_hw_branches_suspended = 0;
+	put_online_cpus();
 }
 
-static void bts_trace_stop(struct trace_array *tr)
+static void bts_trace_start(struct trace_array *tr)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
-	trace_hw_branches_enabled = 0;
-	on_each_cpu(bts_trace_stop_cpu, NULL, 1);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
+	trace_hw_branches_suspended = 0;
+	put_online_cpus();
+}
 
-	spin_unlock(&bts_tracer_lock);
+static void bts_trace_stop(struct trace_array *tr)
+{
+	int cpu;
+
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
+	trace_hw_branches_suspended = 1;
+	put_online_cpus();
 }
 
 static int __cpuinit bts_hotcpu_handler(struct notifier_block *nfb,
 				     unsigned long action, void *hcpu)
 {
-	unsigned int cpu = (unsigned long)hcpu;
-
-	spin_lock(&bts_tracer_lock);
-
-	if (!trace_hw_branches_enabled)
-		goto out;
+	int cpu = (long)hcpu;
 
 	switch (action) {
 	case CPU_ONLINE:
 	case CPU_DOWN_FAILED:
-		smp_call_function_single(cpu, bts_trace_start_cpu, NULL, 1);
+		/* The notification is sent with interrupts enabled. */
+		if (trace_hw_branches_enabled) {
+			bts_trace_init_cpu(cpu);
+
+			if (trace_hw_branches_suspended &&
+			    likely(per_cpu(tracer, cpu)))
+				ds_suspend_bts(per_cpu(tracer, cpu));
+		}
 		break;
+
 	case CPU_DOWN_PREPARE:
-		smp_call_function_single(cpu, bts_trace_stop_cpu, NULL, 1);
-		break;
+		/* The notification is sent with interrupts enabled. */
+		if (likely(per_cpu(tracer, cpu))) {
+			ds_release_bts(per_cpu(tracer, cpu));
+			per_cpu(tracer, cpu) = NULL;
+		}
 	}
 
- out:
-	spin_unlock(&bts_tracer_lock);
 	return NOTIFY_DONE;
 }
 
@@ -126,20 +134,6 @@ static struct notifier_block bts_hotcpu_notifier __cpuinitdata = {
 	.notifier_call = bts_hotcpu_handler
 };
 
-static int bts_trace_init(struct trace_array *tr)
-{
-	hw_branch_trace = tr;
-
-	bts_trace_start(tr);
-
-	return 0;
-}
-
-static void bts_trace_reset(struct trace_array *tr)
-{
-	bts_trace_stop(tr);
-}
-
 static void bts_trace_print_header(struct seq_file *m)
 {
 	seq_puts(m, "# CPU#        TO  <-  FROM\n");
@@ -147,10 +141,10 @@ static void bts_trace_print_header(struct seq_file *m)
 
 static enum print_line_t bts_trace_print_line(struct trace_iterator *iter)
 {
+	unsigned long symflags = TRACE_ITER_SYM_OFFSET;
 	struct trace_entry *entry = iter->ent;
 	struct trace_seq *seq = &iter->seq;
 	struct hw_branch_entry *it;
-	unsigned long symflags = TRACE_ITER_SYM_OFFSET;
 
 	trace_assign_type(it, entry);
 
@@ -226,11 +220,11 @@ static void trace_bts_at(const struct bts_trace *trace, void *at)
 /*
  * Collect the trace on the current cpu and write it into the ftrace buffer.
  *
- * pre: bts_tracer_lock must be locked
+ * pre: tracing must be suspended on the current cpu
  */
 static void trace_bts_cpu(void *arg)
 {
-	struct trace_array *tr = (struct trace_array *) arg;
+	struct trace_array *tr = (struct trace_array *)arg;
 	const struct bts_trace *trace;
 	unsigned char *at;
 
@@ -243,10 +237,9 @@ static void trace_bts_cpu(void *arg)
 	if (unlikely(!this_tracer))
 		return;
 
-	ds_suspend_bts(this_tracer);
 	trace = ds_read_bts(this_tracer);
 	if (!trace)
-		goto out;
+		return;
 
 	for (at = trace->ds.top; (void *)at < trace->ds.end;
 	     at += trace->ds.size)
@@ -255,18 +248,27 @@ static void trace_bts_cpu(void *arg)
 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
 	     at += trace->ds.size)
 		trace_bts_at(trace, at);
-
-out:
-	ds_resume_bts(this_tracer);
 }
 
 static void trace_bts_prepare(struct trace_iterator *iter)
 {
-	spin_lock(&bts_tracer_lock);
+	int cpu;
 
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_suspend_bts(per_cpu(tracer, cpu));
+	/*
+	 * We need to collect the trace on the respective cpu since ftrace
+	 * implicitly adds the record for the current cpu.
+	 * Once that is more flexible, we could collect the data from any cpu.
+	 */
 	on_each_cpu(trace_bts_cpu, iter->tr, 1);
 
-	spin_unlock(&bts_tracer_lock);
+	for_each_online_cpu(cpu)
+		if (likely(per_cpu(tracer, cpu)))
+			ds_resume_bts(per_cpu(tracer, cpu));
+	put_online_cpus();
 }
 
 static void trace_bts_close(struct trace_iterator *iter)
@@ -276,11 +278,11 @@ static void trace_bts_close(struct trace_iterator *iter)
 
 void trace_hw_branch_oops(void)
 {
-	spin_lock(&bts_tracer_lock);
-
-	trace_bts_cpu(hw_branch_trace);
-
-	spin_unlock(&bts_tracer_lock);
+	if (this_tracer) {
+		ds_suspend_bts_noirq(this_tracer);
+		trace_bts_cpu(hw_branch_trace);
+		ds_resume_bts_noirq(this_tracer);
+	}
 }
 
 struct tracer bts_tracer __read_mostly =
@@ -293,7 +295,10 @@ struct tracer bts_tracer __read_mostly =
 	.start		= bts_trace_start,
 	.stop		= bts_trace_stop,
 	.open		= trace_bts_prepare,
-	.close		= trace_bts_close
+	.close		= trace_bts_close,
+#ifdef CONFIG_FTRACE_SELFTEST
+	.selftest	= trace_selftest_startup_hw_branches,
+#endif /* CONFIG_FTRACE_SELFTEST */
 };
 
 __init static int init_bts_trace(void)
diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
index 08f4eb2763d..00dd6485bdd 100644
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -16,6 +16,7 @@ static inline int trace_valid_entry(struct trace_entry *entry)
 	case TRACE_BRANCH:
 	case TRACE_GRAPH_ENT:
 	case TRACE_GRAPH_RET:
+	case TRACE_HW_BRANCHES:
 		return 1;
 	}
 	return 0;
@@ -188,6 +189,7 @@ int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
 #else
 # define trace_selftest_startup_dynamic_tracing(trace, tr, func) ({ 0; })
 #endif /* CONFIG_DYNAMIC_FTRACE */
+
 /*
  * Simple verification test of ftrace function tracer.
  * Enable ftrace, sleep 1/10 second, and then read the trace
@@ -749,3 +751,59 @@ trace_selftest_startup_branch(struct tracer *trace, struct trace_array *tr)
 	return ret;
 }
 #endif /* CONFIG_BRANCH_TRACER */
+
+#ifdef CONFIG_HW_BRANCH_TRACER
+int
+trace_selftest_startup_hw_branches(struct tracer *trace,
+				   struct trace_array *tr)
+{
+	struct trace_iterator *iter;
+	struct tracer tracer;
+	unsigned long count;
+	int ret;
+
+	if (!trace->open) {
+		printk(KERN_CONT "missing open function...");
+		return -1;
+	}
+
+	ret = tracer_init(trace, tr);
+	if (ret) {
+		warn_failed_init_tracer(trace, ret);
+		return ret;
+	}
+
+	/*
+	 * The hw-branch tracer needs to collect the trace from the various
+	 * cpu trace buffers - before tracing is stopped.
+	 */
+	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		return -ENOMEM;
+
+	memcpy(&tracer, trace, sizeof(tracer));
+
+	iter->trace = &tracer;
+	iter->tr = tr;
+	iter->pos = -1;
+	mutex_init(&iter->mutex);
+
+	trace->open(iter);
+
+	mutex_destroy(&iter->mutex);
+	kfree(iter);
+
+	tracing_stop();
+
+	ret = trace_test_buffer(tr, &count);
+	trace->reset(tr);
+	tracing_start();
+
+	if (!ret && !count) {
+		printk(KERN_CONT "no entries found..");
+		ret = -1;
+	}
+
+	return ret;
+}
+#endif /* CONFIG_HW_BRANCH_TRACER */