perf thread-stack: Fix thread stack processing for the idle task

index 61a4286a74dc..115dc4b27a1b 100644
@@ -593,17 +593,13 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 	struct thread_stack *ts = thread->ts;
 	int err = 0;
 
-	if (ts) {
-		if (!ts->crp) {
-			/* Supersede thread_stack__event() */
-			thread_stack__free(thread);
-			thread->ts = thread_stack__new(thread, crp);
-			if (!thread->ts)
-				return -ENOMEM;
-			ts = thread->ts;
-			ts->comm = comm;
-		}
-	} else {
+	if (ts && !ts->crp) {
+		/* Supersede thread_stack__event() */
+		thread_stack__free(thread);
+		ts = NULL;
+	}
+
+	if (!ts) {
 		thread->ts = thread_stack__new(thread, crp);
 		if (!thread->ts)
 			return -ENOMEM;

index 115dc4b27a1b..068c7c8db4be 100644
@@ -449,7 +449,7 @@ static int thread_stack__pop_cp(struct thread *thread, struct thread_stack *ts,
 	return 1;
 }
 
-static int thread_stack__bottom(struct thread *thread, struct thread_stack *ts,
+static int thread_stack__bottom(struct thread_stack *ts,
 				struct perf_sample *sample,
 				struct addr_location *from_al,
 				struct addr_location *to_al, u64 ref)

@@ -474,7 +474,7 @@ static int thread_stack__bottom(struct thread *thread, struct thread_stack *ts,
 	if (!cp)
 		return -ENOMEM;
 
-	return thread_stack__push_cp(thread->ts, ip, sample->time, ref, cp,
+	return thread_stack__push_cp(ts, ip, sample->time, ref, cp,
 				     true, false);
 }
 

@@ -617,8 +617,7 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 
 	/* If the stack is empty, put the current symbol on the stack */
 	if (!ts->cnt) {
-		err = thread_stack__bottom(thread, ts, sample, from_al, to_al,
-					   ref);
+		err = thread_stack__bottom(ts, sample, from_al, to_al, ref);
 		if (err)
 			return err;
 	}

index 068c7c8db4be..e13127755293 100644
@@ -111,9 +111,16 @@ static struct thread_stack *thread_stack__new(struct thread *thread,
 		ts->kernel_start = 1ULL << 63;
 	ts->crp = crp;
 
+	thread->ts = ts;
+
 	return ts;
 }
 
+static inline struct thread_stack *thread_stack__ts(struct thread *thread)
+{
+	return thread ? thread->ts : NULL;
+}
+
 static int thread_stack__push(struct thread_stack *ts, u64 ret_addr,
 			      bool trace_end)
 {

@@ -226,8 +233,10 @@ static int __thread_stack__flush(struct thread *thread, struct thread_stack *ts)
 
 int thread_stack__flush(struct thread *thread)
 {
-	if (thread->ts)
-		return __thread_stack__flush(thread, thread->ts);
+	struct thread_stack *ts = thread->ts;
+
+	if (ts)
+		return __thread_stack__flush(thread, ts);
 
 	return 0;
 }

@@ -235,16 +244,18 @@ int thread_stack__flush(struct thread *thread)
 int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
 			u64 to_ip, u16 insn_len, u64 trace_nr)
 {
+	struct thread_stack *ts = thread_stack__ts(thread);
+
 	if (!thread)
 		return -EINVAL;
 
-	if (!thread->ts) {
-		thread->ts = thread_stack__new(thread, NULL);
-		if (!thread->ts) {
+	if (!ts) {
+		ts = thread_stack__new(thread, NULL);
+		if (!ts) {
 			pr_warning("Out of memory: no thread stack\n");
 			return -ENOMEM;
 		}
-		thread->ts->trace_nr = trace_nr;
+		ts->trace_nr = trace_nr;
 	}
 
 	/*

@@ -252,14 +263,14 @@ int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
 	 * the stack might be completely invalid.  Better to report nothing than
 	 * to report something misleading, so flush the stack.
 	 */
-	if (trace_nr != thread->ts->trace_nr) {
-		if (thread->ts->trace_nr)
-			__thread_stack__flush(thread, thread->ts);
-		thread->ts->trace_nr = trace_nr;
+	if (trace_nr != ts->trace_nr) {
+		if (ts->trace_nr)
+			__thread_stack__flush(thread, ts);
+		ts->trace_nr = trace_nr;
 	}
 
 	/* Stop here if thread_stack__process() is in use */
-	if (thread->ts->crp)
+	if (ts->crp)
 		return 0;
 
 	if (flags & PERF_IP_FLAG_CALL) {

@@ -270,7 +281,7 @@ int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
 		ret_addr = from_ip + insn_len;
 		if (ret_addr == to_ip)
 			return 0; /* Zero-length calls are excluded */
-		return thread_stack__push(thread->ts, ret_addr,
+		return thread_stack__push(ts, ret_addr,
 					  flags & PERF_IP_FLAG_TRACE_END);
 	} else if (flags & PERF_IP_FLAG_TRACE_BEGIN) {
 		/*

@@ -280,10 +291,10 @@ int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
 		 * address, so try to pop that. Also, do not expect a call made
 		 * when the trace ended, to return, so pop that.
 		 */
-		thread_stack__pop(thread->ts, to_ip);
-		thread_stack__pop_trace_end(thread->ts);
+		thread_stack__pop(ts, to_ip);
+		thread_stack__pop_trace_end(ts);
 	} else if ((flags & PERF_IP_FLAG_RETURN) && from_ip) {
-		thread_stack__pop(thread->ts, to_ip);
+		thread_stack__pop(ts, to_ip);
 	}
 
 	return 0;

@@ -291,21 +302,25 @@ int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
 
 void thread_stack__set_trace_nr(struct thread *thread, u64 trace_nr)
 {
-	if (!thread || !thread->ts)
+	struct thread_stack *ts = thread_stack__ts(thread);
+
+	if (!ts)
 		return;
 
-	if (trace_nr != thread->ts->trace_nr) {
-		if (thread->ts->trace_nr)
-			__thread_stack__flush(thread, thread->ts);
-		thread->ts->trace_nr = trace_nr;
+	if (trace_nr != ts->trace_nr) {
+		if (ts->trace_nr)
+			__thread_stack__flush(thread, ts);
+		ts->trace_nr = trace_nr;
 	}
 }
 
 void thread_stack__free(struct thread *thread)
 {
-	if (thread->ts) {
-		__thread_stack__flush(thread, thread->ts);
-		zfree(&thread->ts->stack);
+	struct thread_stack *ts = thread->ts;
+
+	if (ts) {
+		__thread_stack__flush(thread, ts);
+		zfree(&ts->stack);
 		zfree(&thread->ts);
 	}
 }

@@ -318,6 +333,7 @@ static inline u64 callchain_context(u64 ip, u64 kernel_start)
 void thread_stack__sample(struct thread *thread, struct ip_callchain *chain,
 			  size_t sz, u64 ip, u64 kernel_start)
 {
+	struct thread_stack *ts = thread_stack__ts(thread);
 	u64 context = callchain_context(ip, kernel_start);
 	u64 last_context;
 	size_t i, j;

@@ -330,15 +346,15 @@ void thread_stack__sample(struct thread *thread, struct ip_callchain *chain,
 	chain->ips[0] = context;
 	chain->ips[1] = ip;
 
-	if (!thread || !thread->ts) {
+	if (!ts) {
 		chain->nr = 2;
 		return;
 	}
 
 	last_context = context;
 
-	for (i = 2, j = 1; i < sz && j <= thread->ts->cnt; i++, j++) {
-		ip = thread->ts->stack[thread->ts->cnt - j].ret_addr;
+	for (i = 2, j = 1; i < sz && j <= ts->cnt; i++, j++) {
+		ip = ts->stack[ts->cnt - j].ret_addr;
 		context = callchain_context(ip, kernel_start);
 		if (context != last_context) {
 			if (i >= sz - 1)

@@ -590,7 +606,7 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 			  struct addr_location *to_al, u64 ref,
 			  struct call_return_processor *crp)
 {
-	struct thread_stack *ts = thread->ts;
+	struct thread_stack *ts = thread_stack__ts(thread);
 	int err = 0;
 
 	if (ts && !ts->crp) {

@@ -600,10 +616,9 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 	}
 
 	if (!ts) {
-		thread->ts = thread_stack__new(thread, crp);
-		if (!thread->ts)
+		ts = thread_stack__new(thread, crp);
+		if (!ts)
 			return -ENOMEM;
-		ts = thread->ts;
 		ts->comm = comm;
 	}
 

@@ -668,7 +683,9 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 
 size_t thread_stack__depth(struct thread *thread)
 {
-	if (!thread->ts)
+	struct thread_stack *ts = thread_stack__ts(thread);
+
+	if (!ts)
 		return 0;
-	return thread->ts->cnt;
+	return ts->cnt;
 }

index e13127755293..e59f6fa4c67d 100644
@@ -60,6 +60,7 @@ struct thread_stack_entry {
  * @last_time: last timestamp
  * @crp: call/return processor
  * @comm: current comm
+ * @arr_sz: size of array if this is the first element of an array
  */
 struct thread_stack {
 	struct thread_stack_entry *stack;

@@ -71,6 +72,7 @@ struct thread_stack {
 	u64 last_time;
 	struct call_return_processor *crp;
 	struct comm *comm;
+	unsigned int arr_sz;
 };
 
 static int thread_stack__grow(struct thread_stack *ts)

@@ -100,6 +102,8 @@ static struct thread_stack *thread_stack__new(struct thread *thread,
 	if (!ts)
 		return NULL;
 
+	ts->arr_sz = 1;
+
 	if (thread_stack__grow(ts)) {
 		free(ts);
 		return NULL;

@@ -234,11 +238,19 @@ static int __thread_stack__flush(struct thread *thread, struct thread_stack *ts)
 int thread_stack__flush(struct thread *thread)
 {
 	struct thread_stack *ts = thread->ts;
+	unsigned int pos;
+	int err = 0;
 
-	if (ts)
-		return __thread_stack__flush(thread, ts);
+	if (ts) {
+		for (pos = 0; pos < ts->arr_sz; pos++) {
+			int ret = __thread_stack__flush(thread, ts + pos);
 
-	return 0;
+			if (ret)
+				err = ret;
+		}
+	}
+
+	return err;
 }
 
 int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,

@@ -314,13 +326,29 @@ void thread_stack__set_trace_nr(struct thread *thread, u64 trace_nr)
 	}
 }
 
+static void __thread_stack__free(struct thread *thread, struct thread_stack *ts)
+{
+	__thread_stack__flush(thread, ts);
+	zfree(&ts->stack);
+}
+
+static void thread_stack__reset(struct thread *thread, struct thread_stack *ts)
+{
+	unsigned int arr_sz = ts->arr_sz;
+
+	__thread_stack__free(thread, ts);
+	memset(ts, 0, sizeof(*ts));
+	ts->arr_sz = arr_sz;
+}
+
 void thread_stack__free(struct thread *thread)
 {
 	struct thread_stack *ts = thread->ts;
+	unsigned int pos;
 
 	if (ts) {
-		__thread_stack__flush(thread, ts);
-		zfree(&ts->stack);
+		for (pos = 0; pos < ts->arr_sz; pos++)
+			__thread_stack__free(thread, ts + pos);
 		zfree(&thread->ts);
 	}
 }

@@ -611,7 +639,7 @@ int thread_stack__process(struct thread *thread, struct comm *comm,
 
 	if (ts && !ts->crp) {
 		/* Supersede thread_stack__event() */
-		thread_stack__free(thread);
+		thread_stack__reset(thread, ts);
 		ts = NULL;
 	}
 

index e59f6fa4c67d..4340381d54c2 100644
@@ -93,6 +93,24 @@ static int thread_stack__grow(struct thread_stack *ts)
 	return 0;
 }
 
+static int thread_stack__init(struct thread_stack *ts, struct thread *thread,
+			      struct call_return_processor *crp)
+{
+	int err;
+
+	err = thread_stack__grow(ts);
+	if (err)
+		return err;
+
+	if (thread->mg && thread->mg->machine)
+		ts->kernel_start = machine__kernel_start(thread->mg->machine);
+	else
+		ts->kernel_start = 1ULL << 63;
+	ts->crp = crp;
+
+	return 0;
+}
+
 static struct thread_stack *thread_stack__new(struct thread *thread,
 					      struct call_return_processor *crp)
 {

@@ -104,17 +122,11 @@ static struct thread_stack *thread_stack__new(struct thread *thread,
 
 	ts->arr_sz = 1;
 
-	if (thread_stack__grow(ts)) {
+	if (thread_stack__init(ts, thread, crp)) {
 		free(ts);
 		return NULL;
 	}
 
-	if (thread->mg && thread->mg->machine)
-		ts->kernel_start = machine__kernel_start(thread->mg->machine);
-	else
-		ts->kernel_start = 1ULL << 63;
-	ts->crp = crp;
-
 	thread->ts = ts;
 
 	return ts;

index 4340381d54c2..a896d89fe5f7 100644
@@ -114,20 +114,26 @@ static int thread_stack__init(struct thread_stack *ts, struct thread *thread,
 static struct thread_stack *thread_stack__new(struct thread *thread,
 					      struct call_return_processor *crp)
 {
-	struct thread_stack *ts;
-
-	ts = zalloc(sizeof(struct thread_stack));
-	if (!ts)
-		return NULL;
-
-	ts->arr_sz = 1;
-
-	if (thread_stack__init(ts, thread, crp)) {
-		free(ts);
-		return NULL;
+	struct thread_stack *ts = thread->ts, *new_ts;
+	unsigned int old_sz = ts ? ts->arr_sz : 0;
+	unsigned int new_sz = 1;
+
+	if (!ts || new_sz > old_sz) {
+		new_ts = calloc(new_sz, sizeof(*ts));
+		if (!new_ts)
+			return NULL;
+		if (ts)
+			memcpy(new_ts, ts, old_sz * sizeof(*ts));
+		new_ts->arr_sz = new_sz;
+		if (thread->ts)
+			zfree(&thread->ts);
+		thread->ts = new_ts;
+		ts = new_ts;
 	}
 
-	thread->ts = ts;
+	if (!ts->stack &&
+	    thread_stack__init(ts, thread, crp))
+		return NULL;
 
 	return ts;
 }

index 78a067777144..5456c84c7dd1 100644
@@ -1527,6 +1527,13 @@ struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
 	return machine__findnew_thread(&session->machines.host, -1, pid);
 }
 
+/*
+ * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
+ * So here a single thread is created for that, but actually there is a separate
+ * idle task per cpu, so there should be one 'struct thread' per cpu, but there
+ * is only 1. That causes problems for some tools, requiring workarounds. For
+ * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
+ */
 int perf_session__register_idle_thread(struct perf_session *session)
 {
 	struct thread *thread;