change system file

hc

2024-10-09 244b2c5ca8b14627e4a17755e5922221e121c771

 kernel/tools/perf/builtin-sched.c
..
..
@@ -1,11 +1,12 @@
1
1
 // SPDX-License-Identifier: GPL-2.0
2
2
 #include "builtin.h"
3
3
 #include "perf.h"
4
+#include "perf-sys.h"
4
5
 
5
-#include "util/util.h"
6
+#include "util/cpumap.h"
6
7
 #include "util/evlist.h"
7
-#include "util/cache.h"
8
8
 #include "util/evsel.h"
9
+#include "util/evsel_fprintf.h"
9
10
 #include "util/symbol.h"
10
11
 #include "util/thread.h"
11
12
 #include "util/header.h"
..
..
@@ -15,16 +16,20 @@
15
16
 #include "util/thread_map.h"
16
17
 #include "util/color.h"
17
18
 #include "util/stat.h"
19
+#include "util/string2.h"
18
20
 #include "util/callchain.h"
19
21
 #include "util/time-utils.h"
20
22
 
23
+#include <subcmd/pager.h>
21
24
 #include <subcmd/parse-options.h>
22
25
 #include "util/trace-event.h"
23
26
 
24
27
 #include "util/debug.h"
28
+#include "util/event.h"
25
29
 
26
30
 #include <linux/kernel.h>
27
31
 #include <linux/log2.h>
32
+#include <linux/zalloc.h>
28
33
 #include <sys/prctl.h>
29
34
 #include <sys/resource.h>
30
35
 #include <inttypes.h>
..
..
@@ -34,15 +39,20 @@
34
39
 #include <pthread.h>
35
40
 #include <math.h>
36
41
 #include <api/fs/fs.h>
42
+#include <perf/cpumap.h>
37
43
 #include <linux/time64.h>
44
+#include <linux/err.h>
38
45
 
39
-#include "sane_ctype.h"
46
+#include <linux/ctype.h>
40
47
 
41
48
 #define PR_SET_NAME		15               /* Set process name */
42
49
 #define MAX_CPUS		4096
43
50
 #define COMM_LEN		20
44
51
 #define SYM_LEN			129
45
52
 #define MAX_PID			1024000
53
+
54
+static const char *cpu_list;
55
+static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
46
56
 
47
57
 struct sched_atom;
48
58
 
..
..
@@ -120,7 +130,8 @@
120
130
 	struct thread		*thread;
121
131
 	struct rb_node		node;
122
132
 	u64			max_lat;
123
-	u64			max_lat_at;
133
+	u64			max_lat_start;
134
+	u64			max_lat_end;
124
135
 	u64			total_lat;
125
136
 	u64			nb_atoms;
126
137
 	u64			total_runtime;
..
..
@@ -132,13 +143,13 @@
132
143
 struct perf_sched;
133
144
 
134
145
 struct trace_sched_handler {
135
-	int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel,
146
+	int (*switch_event)(struct perf_sched *sched, struct evsel *evsel,
136
147
 			    struct perf_sample *sample, struct machine *machine);
137
148
 
138
-	int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel,
149
+	int (*runtime_event)(struct perf_sched *sched, struct evsel *evsel,
139
150
 			     struct perf_sample *sample, struct machine *machine);
140
151
 
141
-	int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel,
152
+	int (*wakeup_event)(struct perf_sched *sched, struct evsel *evsel,
142
153
 			    struct perf_sample *sample, struct machine *machine);
143
154
 
144
155
 	/* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
..
..
@@ -146,7 +157,7 @@
146
157
 			  struct machine *machine);
147
158
 
148
159
 	int (*migrate_task_event)(struct perf_sched *sched,
149
-				  struct perf_evsel *evsel,
160
+				  struct evsel *evsel,
150
161
 				  struct perf_sample *sample,
151
162
 				  struct machine *machine);
152
163
 };
..
..
@@ -158,11 +169,11 @@
158
169
 	DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS);
159
170
 	int			*comp_cpus;
160
171
 	bool			 comp;
161
-	struct thread_map	*color_pids;
172
+	struct perf_thread_map *color_pids;
162
173
 	const char		*color_pids_str;
163
-	struct cpu_map		*color_cpus;
174
+	struct perf_cpu_map	*color_cpus;
164
175
 	const char		*color_cpus_str;
165
-	struct cpu_map		*cpus;
176
+	struct perf_cpu_map	*cpus;
166
177
 	const char		*cpus_str;
167
178
 };
168
179
 
..
..
@@ -213,7 +224,7 @@
213
224
 	u64		 all_runtime;
214
225
 	u64		 all_count;
215
226
 	u64		 cpu_last_switched[MAX_CPUS];
216
-	struct rb_root	 atom_root, sorted_atom_root, merged_atom_root;
227
+	struct rb_root_cached atom_root, sorted_atom_root, merged_atom_root;
217
228
 	struct list_head sort_list, cmp_pid;
218
229
 	bool force;
219
230
 	bool skip_merge;
..
..
@@ -271,7 +282,7 @@
271
282
 struct idle_thread_runtime {
272
283
 	struct thread_runtime	tr;
273
284
 	struct thread		*last_thread;
274
-	struct rb_root		sorted_root;
285
+	struct rb_root_cached	sorted_root;
275
286
 	struct callchain_root	callchain;
276
287
 	struct callchain_cursor	cursor;
277
288
 };
..
..
@@ -659,7 +670,7 @@
659
670
 	err = pthread_attr_init(&attr);
660
671
 	BUG_ON(err);
661
672
 	err = pthread_attr_setstacksize(&attr,
662
-			(size_t) max(16 * 1024, PTHREAD_STACK_MIN));
673
+			(size_t) max(16 * 1024, (int)PTHREAD_STACK_MIN));
663
674
 	BUG_ON(err);
664
675
 	err = pthread_mutex_lock(&sched->start_work_mutex);
665
676
 	BUG_ON(err);
..
..
@@ -798,11 +809,11 @@
798
809
 
799
810
 static int
800
811
 replay_wakeup_event(struct perf_sched *sched,
801
-		    struct perf_evsel *evsel, struct perf_sample *sample,
812
+		    struct evsel *evsel, struct perf_sample *sample,
802
813
 		    struct machine *machine __maybe_unused)
803
814
 {
804
-	const char *comm = perf_evsel__strval(evsel, sample, "comm");
805
-	const u32 pid	 = perf_evsel__intval(evsel, sample, "pid");
815
+	const char *comm = evsel__strval(evsel, sample, "comm");
816
+	const u32 pid	 = evsel__intval(evsel, sample, "pid");
806
817
 	struct task_desc *waker, *wakee;
807
818
 
808
819
 	if (verbose > 0) {
..
..
@@ -819,15 +830,15 @@
819
830
 }
820
831
 
821
832
 static int replay_switch_event(struct perf_sched *sched,
822
-			       struct perf_evsel *evsel,
833
+			       struct evsel *evsel,
823
834
 			       struct perf_sample *sample,
824
835
 			       struct machine *machine __maybe_unused)
825
836
 {
826
-	const char *prev_comm  = perf_evsel__strval(evsel, sample, "prev_comm"),
827
-		   *next_comm  = perf_evsel__strval(evsel, sample, "next_comm");
828
-	const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
829
-		  next_pid = perf_evsel__intval(evsel, sample, "next_pid");
830
-	const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
837
+	const char *prev_comm  = evsel__strval(evsel, sample, "prev_comm"),
838
+		   *next_comm  = evsel__strval(evsel, sample, "next_comm");
839
+	const u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
840
+		  next_pid = evsel__intval(evsel, sample, "next_pid");
841
+	const u64 prev_state = evsel__intval(evsel, sample, "prev_state");
831
842
 	struct task_desc *prev, __maybe_unused *next;
832
843
 	u64 timestamp0, timestamp = sample->time;
833
844
 	int cpu = sample->cpu;
..
..
@@ -950,10 +961,10 @@
950
961
 }
951
962
 
952
963
 static struct work_atoms *
953
-thread_atoms_search(struct rb_root *root, struct thread *thread,
964
+thread_atoms_search(struct rb_root_cached *root, struct thread *thread,
954
965
 			 struct list_head *sort_list)
955
966
 {
956
-	struct rb_node *node = root->rb_node;
967
+	struct rb_node *node = root->rb_root.rb_node;
957
968
 	struct work_atoms key = { .thread = thread };
958
969
 
959
970
 	while (node) {
..
..
@@ -976,10 +987,11 @@
976
987
 }
977
988
 
978
989
 static void
979
-__thread_latency_insert(struct rb_root *root, struct work_atoms *data,
990
+__thread_latency_insert(struct rb_root_cached *root, struct work_atoms *data,
980
991
 			 struct list_head *sort_list)
981
992
 {
982
-	struct rb_node **new = &(root->rb_node), *parent = NULL;
993
+	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
994
+	bool leftmost = true;
983
995
 
984
996
 	while (*new) {
985
997
 		struct work_atoms *this;
..
..
@@ -992,12 +1004,14 @@
992
1004
 
993
1005
 		if (cmp > 0)
994
1006
 			new = &((*new)->rb_left);
995
-		else
1007
+		else {
996
1008
 			new = &((*new)->rb_right);
1009
+			leftmost = false;
1010
+		}
997
1011
 	}
998
1012
 
999
1013
 	rb_link_node(&data->node, parent, new);
1000
-	rb_insert_color(&data->node, root);
1014
+	rb_insert_color_cached(&data->node, root, leftmost);
1001
1015
 }
1002
1016
 
1003
1017
 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
..
..
@@ -1083,19 +1097,20 @@
1083
1097
 	atoms->total_lat += delta;
1084
1098
 	if (delta > atoms->max_lat) {
1085
1099
 		atoms->max_lat = delta;
1086
-		atoms->max_lat_at = timestamp;
1100
+		atoms->max_lat_start = atom->wake_up_time;
1101
+		atoms->max_lat_end = timestamp;
1087
1102
 	}
1088
1103
 	atoms->nb_atoms++;
1089
1104
 }
1090
1105
 
1091
1106
 static int latency_switch_event(struct perf_sched *sched,
1092
-				struct perf_evsel *evsel,
1107
+				struct evsel *evsel,
1093
1108
 				struct perf_sample *sample,
1094
1109
 				struct machine *machine)
1095
1110
 {
1096
-	const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1097
-		  next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1098
-	const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
1111
+	const u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
1112
+		  next_pid = evsel__intval(evsel, sample, "next_pid");
1113
+	const u64 prev_state = evsel__intval(evsel, sample, "prev_state");
1099
1114
 	struct work_atoms *out_events, *in_events;
1100
1115
 	struct thread *sched_out, *sched_in;
1101
1116
 	u64 timestamp0, timestamp = sample->time;
..
..
@@ -1159,12 +1174,12 @@
1159
1174
 }
1160
1175
 
1161
1176
 static int latency_runtime_event(struct perf_sched *sched,
1162
-				 struct perf_evsel *evsel,
1177
+				 struct evsel *evsel,
1163
1178
 				 struct perf_sample *sample,
1164
1179
 				 struct machine *machine)
1165
1180
 {
1166
-	const u32 pid	   = perf_evsel__intval(evsel, sample, "pid");
1167
-	const u64 runtime  = perf_evsel__intval(evsel, sample, "runtime");
1181
+	const u32 pid	   = evsel__intval(evsel, sample, "pid");
1182
+	const u64 runtime  = evsel__intval(evsel, sample, "runtime");
1168
1183
 	struct thread *thread = machine__findnew_thread(machine, -1, pid);
1169
1184
 	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
1170
1185
 	u64 timestamp = sample->time;
..
..
@@ -1194,11 +1209,11 @@
1194
1209
 }
1195
1210
 
1196
1211
 static int latency_wakeup_event(struct perf_sched *sched,
1197
-				struct perf_evsel *evsel,
1212
+				struct evsel *evsel,
1198
1213
 				struct perf_sample *sample,
1199
1214
 				struct machine *machine)
1200
1215
 {
1201
-	const u32 pid	  = perf_evsel__intval(evsel, sample, "pid");
1216
+	const u32 pid	  = evsel__intval(evsel, sample, "pid");
1202
1217
 	struct work_atoms *atoms;
1203
1218
 	struct work_atom *atom;
1204
1219
 	struct thread *wakee;
..
..
@@ -1255,11 +1270,11 @@
1255
1270
 }
1256
1271
 
1257
1272
 static int latency_migrate_task_event(struct perf_sched *sched,
1258
-				      struct perf_evsel *evsel,
1273
+				      struct evsel *evsel,
1259
1274
 				      struct perf_sample *sample,
1260
1275
 				      struct machine *machine)
1261
1276
 {
1262
-	const u32 pid = perf_evsel__intval(evsel, sample, "pid");
1277
+	const u32 pid = evsel__intval(evsel, sample, "pid");
1263
1278
 	u64 timestamp = sample->time;
1264
1279
 	struct work_atoms *atoms;
1265
1280
 	struct work_atom *atom;
..
..
@@ -1309,7 +1324,7 @@
1309
1324
 	int i;
1310
1325
 	int ret;
1311
1326
 	u64 avg;
1312
-	char max_lat_at[32];
1327
+	char max_lat_start[32], max_lat_end[32];
1313
1328
 
1314
1329
 	if (!work_list->nb_atoms)
1315
1330
 		return;
..
..
@@ -1331,13 +1346,14 @@
1331
1346
 		printf(" ");
1332
1347
 
1333
1348
 	avg = work_list->total_lat / work_list->nb_atoms;
1334
-	timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at));
1349
+	timestamp__scnprintf_usec(work_list->max_lat_start, max_lat_start, sizeof(max_lat_start));
1350
+	timestamp__scnprintf_usec(work_list->max_lat_end, max_lat_end, sizeof(max_lat_end));
1335
1351
 
1336
-	printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n",
1352
+	printf("|%11.3f ms |%9" PRIu64 " | avg:%8.3f ms | max:%8.3f ms | max start: %12s s | max end: %12s s\n",
1337
1353
 	      (double)work_list->total_runtime / NSEC_PER_MSEC,
1338
1354
 		 work_list->nb_atoms, (double)avg / NSEC_PER_MSEC,
1339
1355
 		 (double)work_list->max_lat / NSEC_PER_MSEC,
1340
-		 max_lat_at);
1356
+		 max_lat_start, max_lat_end);
1341
1357
 }
1342
1358
 
1343
1359
 static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
..
..
@@ -1447,15 +1463,15 @@
1447
1463
 static void perf_sched__sort_lat(struct perf_sched *sched)
1448
1464
 {
1449
1465
 	struct rb_node *node;
1450
-	struct rb_root *root = &sched->atom_root;
1466
+	struct rb_root_cached *root = &sched->atom_root;
1451
1467
 again:
1452
1468
 	for (;;) {
1453
1469
 		struct work_atoms *data;
1454
-		node = rb_first(root);
1470
+		node = rb_first_cached(root);
1455
1471
 		if (!node)
1456
1472
 			break;
1457
1473
 
1458
-		rb_erase(node, root);
1474
+		rb_erase_cached(node, root);
1459
1475
 		data = rb_entry(node, struct work_atoms, node);
1460
1476
 		__thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
1461
1477
 	}
..
..
@@ -1466,7 +1482,7 @@
1466
1482
 }
1467
1483
 
1468
1484
 static int process_sched_wakeup_event(struct perf_tool *tool,
1469
-				      struct perf_evsel *evsel,
1485
+				      struct evsel *evsel,
1470
1486
 				      struct perf_sample *sample,
1471
1487
 				      struct machine *machine)
1472
1488
 {
..
..
@@ -1510,10 +1526,10 @@
1510
1526
 	return thread;
1511
1527
 }
1512
1528
 
1513
-static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel,
1529
+static int map_switch_event(struct perf_sched *sched, struct evsel *evsel,
1514
1530
 			    struct perf_sample *sample, struct machine *machine)
1515
1531
 {
1516
-	const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1532
+	const u32 next_pid = evsel__intval(evsel, sample, "next_pid");
1517
1533
 	struct thread *sched_in;
1518
1534
 	struct thread_runtime *tr;
1519
1535
 	int new_shortname;
..
..
@@ -1651,14 +1667,14 @@
1651
1667
 }
1652
1668
 
1653
1669
 static int process_sched_switch_event(struct perf_tool *tool,
1654
-				      struct perf_evsel *evsel,
1670
+				      struct evsel *evsel,
1655
1671
 				      struct perf_sample *sample,
1656
1672
 				      struct machine *machine)
1657
1673
 {
1658
1674
 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1659
1675
 	int this_cpu = sample->cpu, err = 0;
1660
-	u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1661
-	    next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1676
+	u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
1677
+	    next_pid = evsel__intval(evsel, sample, "next_pid");
1662
1678
 
1663
1679
 	if (sched->curr_pid[this_cpu] != (u32)-1) {
1664
1680
 		/*
..
..
@@ -1677,7 +1693,7 @@
1677
1693
 }
1678
1694
 
1679
1695
 static int process_sched_runtime_event(struct perf_tool *tool,
1680
-				       struct perf_evsel *evsel,
1696
+				       struct evsel *evsel,
1681
1697
 				       struct perf_sample *sample,
1682
1698
 				       struct machine *machine)
1683
1699
 {
..
..
@@ -1707,7 +1723,7 @@
1707
1723
 }
1708
1724
 
1709
1725
 static int process_sched_migrate_task_event(struct perf_tool *tool,
1710
-					    struct perf_evsel *evsel,
1726
+					    struct evsel *evsel,
1711
1727
 					    struct perf_sample *sample,
1712
1728
 					    struct machine *machine)
1713
1729
 {
..
..
@@ -1720,14 +1736,14 @@
1720
1736
 }
1721
1737
 
1722
1738
 typedef int (*tracepoint_handler)(struct perf_tool *tool,
1723
-				  struct perf_evsel *evsel,
1739
+				  struct evsel *evsel,
1724
1740
 				  struct perf_sample *sample,
1725
1741
 				  struct machine *machine);
1726
1742
 
1727
1743
 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused,
1728
1744
 						 union perf_event *event __maybe_unused,
1729
1745
 						 struct perf_sample *sample,
1730
-						 struct perf_evsel *evsel,
1746
+						 struct evsel *evsel,
1731
1747
 						 struct machine *machine)
1732
1748
 {
1733
1749
 	int err = 0;
..
..
@@ -1773,7 +1789,7 @@
1773
1789
 
1774
1790
 static int perf_sched__read_events(struct perf_sched *sched)
1775
1791
 {
1776
-	const struct perf_evsel_str_handler handlers[] = {
1792
+	const struct evsel_str_handler handlers[] = {
1777
1793
 		{ "sched:sched_switch",	      process_sched_switch_event, },
1778
1794
 		{ "sched:sched_stat_runtime", process_sched_runtime_event, },
1779
1795
 		{ "sched:sched_wakeup",	      process_sched_wakeup_event, },
..
..
@@ -1782,18 +1798,16 @@
1782
1798
 	};
1783
1799
 	struct perf_session *session;
1784
1800
 	struct perf_data data = {
1785
-		.file      = {
1786
-			.path = input_name,
1787
-		},
1788
-		.mode      = PERF_DATA_MODE_READ,
1789
-		.force     = sched->force,
1801
+		.path  = input_name,
1802
+		.mode  = PERF_DATA_MODE_READ,
1803
+		.force = sched->force,
1790
1804
 	};
1791
1805
 	int rc = -1;
1792
1806
 
1793
1807
 	session = perf_session__new(&data, false, &sched->tool);
1794
-	if (session == NULL) {
1795
-		pr_debug("No Memory for session\n");
1796
-		return -1;
1808
+	if (IS_ERR(session)) {
1809
+		pr_debug("Error creating perf session");
1810
+		return PTR_ERR(session);
1797
1811
 	}
1798
1812
 
1799
1813
 	symbol__init(&session->header.env);
..
..
@@ -1837,7 +1851,7 @@
1837
1851
  * returns runtime data for event, allocating memory for it the
1838
1852
  * first time it is used.
1839
1853
  */
1840
-static struct evsel_runtime *perf_evsel__get_runtime(struct perf_evsel *evsel)
1854
+static struct evsel_runtime *evsel__get_runtime(struct evsel *evsel)
1841
1855
 {
1842
1856
 	struct evsel_runtime *r = evsel->priv;
1843
1857
 
..
..
@@ -1852,10 +1866,9 @@
1852
1866
 /*
1853
1867
  * save last time event was seen per cpu
1854
1868
  */
1855
-static void perf_evsel__save_time(struct perf_evsel *evsel,
1856
-				  u64 timestamp, u32 cpu)
1869
+static void evsel__save_time(struct evsel *evsel, u64 timestamp, u32 cpu)
1857
1870
 {
1858
-	struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1871
+	struct evsel_runtime *r = evsel__get_runtime(evsel);
1859
1872
 
1860
1873
 	if (r == NULL)
1861
1874
 		return;
..
..
@@ -1879,9 +1892,9 @@
1879
1892
 }
1880
1893
 
1881
1894
 /* returns last time this event was seen on the given cpu */
1882
-static u64 perf_evsel__get_time(struct perf_evsel *evsel, u32 cpu)
1895
+static u64 evsel__get_time(struct evsel *evsel, u32 cpu)
1883
1896
 {
1884
-	struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1897
+	struct evsel_runtime *r = evsel__get_runtime(evsel);
1885
1898
 
1886
1899
 	if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu))
1887
1900
 		return 0;
..
..
@@ -1986,19 +1999,22 @@
1986
1999
 }
1987
2000
 
1988
2001
 static void timehist_print_sample(struct perf_sched *sched,
1989
-				  struct perf_evsel *evsel,
2002
+				  struct evsel *evsel,
1990
2003
 				  struct perf_sample *sample,
1991
2004
 				  struct addr_location *al,
1992
2005
 				  struct thread *thread,
1993
2006
 				  u64 t, int state)
1994
2007
 {
1995
2008
 	struct thread_runtime *tr = thread__priv(thread);
1996
-	const char *next_comm = perf_evsel__strval(evsel, sample, "next_comm");
1997
-	const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
2009
+	const char *next_comm = evsel__strval(evsel, sample, "next_comm");
2010
+	const u32 next_pid = evsel__intval(evsel, sample, "next_pid");
1998
2011
 	u32 max_cpus = sched->max_cpu + 1;
1999
2012
 	char tstr[64];
2000
2013
 	char nstr[30];
2001
2014
 	u64 wait_time;
2015
+
2016
+	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
2017
+		return;
2002
2018
 
2003
2019
 	timestamp__scnprintf_usec(t, tstr, sizeof(tstr));
2004
2020
 	printf("%15s [%04d] ", tstr, sample->cpu);
..
..
@@ -2048,7 +2064,7 @@
2048
2064
 			    EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE |
2049
2065
 			    EVSEL__PRINT_CALLCHAIN_ARROW |
2050
2066
 			    EVSEL__PRINT_SKIP_IGNORED,
2051
-			    &callchain_cursor, stdout);
2067
+			    &callchain_cursor, symbol_conf.bt_stop_list,  stdout);
2052
2068
 
2053
2069
 out:
2054
2070
 	printf("\n");
..
..
@@ -2119,18 +2135,18 @@
2119
2135
 }
2120
2136
 
2121
2137
 static bool is_idle_sample(struct perf_sample *sample,
2122
-			   struct perf_evsel *evsel)
2138
+			   struct evsel *evsel)
2123
2139
 {
2124
2140
 	/* pid 0 == swapper == idle task */
2125
-	if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0)
2126
-		return perf_evsel__intval(evsel, sample, "prev_pid") == 0;
2141
+	if (strcmp(evsel__name(evsel), "sched:sched_switch") == 0)
2142
+		return evsel__intval(evsel, sample, "prev_pid") == 0;
2127
2143
 
2128
2144
 	return sample->pid == 0;
2129
2145
 }
2130
2146
 
2131
2147
 static void save_task_callchain(struct perf_sched *sched,
2132
2148
 				struct perf_sample *sample,
2133
-				struct perf_evsel *evsel,
2149
+				struct evsel *evsel,
2134
2150
 				struct machine *machine)
2135
2151
 {
2136
2152
 	struct callchain_cursor *cursor = &callchain_cursor;
..
..
@@ -2164,7 +2180,7 @@
2164
2180
 		if (node == NULL)
2165
2181
 			break;
2166
2182
 
2167
-		sym = node->sym;
2183
+		sym = node->ms.sym;
2168
2184
 		if (sym) {
2169
2185
 			if (!strcmp(sym->name, "schedule") ||
2170
2186
 			    !strcmp(sym->name, "__schedule") ||
..
..
@@ -2284,7 +2300,7 @@
2284
2300
 static struct thread *timehist_get_thread(struct perf_sched *sched,
2285
2301
 					  struct perf_sample *sample,
2286
2302
 					  struct machine *machine,
2287
-					  struct perf_evsel *evsel)
2303
+					  struct evsel *evsel)
2288
2304
 {
2289
2305
 	struct thread *thread;
2290
2306
 
..
..
@@ -2320,7 +2336,7 @@
2320
2336
 			itr->last_thread = thread;
2321
2337
 
2322
2338
 			/* copy task callchain when entering to idle */
2323
-			if (perf_evsel__intval(evsel, sample, "next_pid") == 0)
2339
+			if (evsel__intval(evsel, sample, "next_pid") == 0)
2324
2340
 				save_idle_callchain(sched, itr, sample);
2325
2341
 		}
2326
2342
 	}
..
..
@@ -2330,7 +2346,7 @@
2330
2346
 
2331
2347
 static bool timehist_skip_sample(struct perf_sched *sched,
2332
2348
 				 struct thread *thread,
2333
-				 struct perf_evsel *evsel,
2349
+				 struct evsel *evsel,
2334
2350
 				 struct perf_sample *sample)
2335
2351
 {
2336
2352
 	bool rc = false;
..
..
@@ -2341,10 +2357,10 @@
2341
2357
 	}
2342
2358
 
2343
2359
 	if (sched->idle_hist) {
2344
-		if (strcmp(perf_evsel__name(evsel), "sched:sched_switch"))
2360
+		if (strcmp(evsel__name(evsel), "sched:sched_switch"))
2345
2361
 			rc = true;
2346
-		else if (perf_evsel__intval(evsel, sample, "prev_pid") != 0 &&
2347
-			 perf_evsel__intval(evsel, sample, "next_pid") != 0)
2362
+		else if (evsel__intval(evsel, sample, "prev_pid") != 0 &&
2363
+			 evsel__intval(evsel, sample, "next_pid") != 0)
2348
2364
 			rc = true;
2349
2365
 	}
2350
2366
 
..
..
@@ -2352,7 +2368,7 @@
2352
2368
 }
2353
2369
 
2354
2370
 static void timehist_print_wakeup_event(struct perf_sched *sched,
2355
-					struct perf_evsel *evsel,
2371
+					struct evsel *evsel,
2356
2372
 					struct perf_sample *sample,
2357
2373
 					struct machine *machine,
2358
2374
 					struct thread *awakened)
..
..
@@ -2385,9 +2401,18 @@
2385
2401
 	printf("\n");
2386
2402
 }
2387
2403
 
2404
+static int timehist_sched_wakeup_ignore(struct perf_tool *tool __maybe_unused,
2405
+					union perf_event *event __maybe_unused,
2406
+					struct evsel *evsel __maybe_unused,
2407
+					struct perf_sample *sample __maybe_unused,
2408
+					struct machine *machine __maybe_unused)
2409
+{
2410
+	return 0;
2411
+}
2412
+
2388
2413
 static int timehist_sched_wakeup_event(struct perf_tool *tool,
2389
2414
 				       union perf_event *event __maybe_unused,
2390
-				       struct perf_evsel *evsel,
2415
+				       struct evsel *evsel,
2391
2416
 				       struct perf_sample *sample,
2392
2417
 				       struct machine *machine)
2393
2418
 {
..
..
@@ -2395,7 +2420,7 @@
2395
2420
 	struct thread *thread;
2396
2421
 	struct thread_runtime *tr = NULL;
2397
2422
 	/* want pid of awakened task not pid in sample */
2398
-	const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2423
+	const u32 pid = evsel__intval(evsel, sample, "pid");
2399
2424
 
2400
2425
 	thread = machine__findnew_thread(machine, 0, pid);
2401
2426
 	if (thread == NULL)
..
..
@@ -2417,7 +2442,7 @@
2417
2442
 }
2418
2443
 
2419
2444
 static void timehist_print_migration_event(struct perf_sched *sched,
2420
-					struct perf_evsel *evsel,
2445
+					struct evsel *evsel,
2421
2446
 					struct perf_sample *sample,
2422
2447
 					struct machine *machine,
2423
2448
 					struct thread *migrated)
..
..
@@ -2431,8 +2456,8 @@
2431
2456
 		return;
2432
2457
 
2433
2458
 	max_cpus = sched->max_cpu + 1;
2434
-	ocpu = perf_evsel__intval(evsel, sample, "orig_cpu");
2435
-	dcpu = perf_evsel__intval(evsel, sample, "dest_cpu");
2459
+	ocpu = evsel__intval(evsel, sample, "orig_cpu");
2460
+	dcpu = evsel__intval(evsel, sample, "dest_cpu");
2436
2461
 
2437
2462
 	thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2438
2463
 	if (thread == NULL)
..
..
@@ -2471,7 +2496,7 @@
2471
2496
 
2472
2497
 static int timehist_migrate_task_event(struct perf_tool *tool,
2473
2498
 				       union perf_event *event __maybe_unused,
2474
-				       struct perf_evsel *evsel,
2499
+				       struct evsel *evsel,
2475
2500
 				       struct perf_sample *sample,
2476
2501
 				       struct machine *machine)
2477
2502
 {
..
..
@@ -2479,7 +2504,7 @@
2479
2504
 	struct thread *thread;
2480
2505
 	struct thread_runtime *tr = NULL;
2481
2506
 	/* want pid of migrated task not pid in sample */
2482
-	const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2507
+	const u32 pid = evsel__intval(evsel, sample, "pid");
2483
2508
 
2484
2509
 	thread = machine__findnew_thread(machine, 0, pid);
2485
2510
 	if (thread == NULL)
..
..
@@ -2499,7 +2524,7 @@
2499
2524
 
2500
2525
 static int timehist_sched_change_event(struct perf_tool *tool,
2501
2526
 				       union perf_event *event,
2502
-				       struct perf_evsel *evsel,
2527
+				       struct evsel *evsel,
2503
2528
 				       struct perf_sample *sample,
2504
2529
 				       struct machine *machine)
2505
2530
 {
..
..
@@ -2510,8 +2535,7 @@
2510
2535
 	struct thread_runtime *tr = NULL;
2511
2536
 	u64 tprev, t = sample->time;
2512
2537
 	int rc = 0;
2513
-	int state = perf_evsel__intval(evsel, sample, "prev_state");
2514
-
2538
+	int state = evsel__intval(evsel, sample, "prev_state");
2515
2539
 
2516
2540
 	if (machine__resolve(machine, &al, sample) < 0) {
2517
2541
 		pr_err("problem processing %d event. skipping it\n",
..
..
@@ -2535,7 +2559,7 @@
2535
2559
 		goto out;
2536
2560
 	}
2537
2561
 
2538
-	tprev = perf_evsel__get_time(evsel, sample->cpu);
2562
+	tprev = evsel__get_time(evsel, sample->cpu);
2539
2563
 
2540
2564
 	/*
2541
2565
 	 * If start time given:
..
..
@@ -2563,7 +2587,8 @@
2563
2587
 	}
2564
2588
 
2565
2589
 	if (!sched->idle_hist || thread->tid == 0) {
2566
-		timehist_update_runtime_stats(tr, t, tprev);
2590
+		if (!cpu_list || test_bit(sample->cpu, cpu_bitmap))
2591
+			timehist_update_runtime_stats(tr, t, tprev);
2567
2592
 
2568
2593
 		if (sched->idle_hist) {
2569
2594
 			struct idle_thread_runtime *itr = (void *)tr;
..
..
@@ -2618,14 +2643,14 @@
2618
2643
 		tr->ready_to_run = 0;
2619
2644
 	}
2620
2645
 
2621
-	perf_evsel__save_time(evsel, sample->time, sample->cpu);
2646
+	evsel__save_time(evsel, sample->time, sample->cpu);
2622
2647
 
2623
2648
 	return rc;
2624
2649
 }
2625
2650
 
2626
2651
 static int timehist_sched_switch_event(struct perf_tool *tool,
2627
2652
 			     union perf_event *event,
2628
-			     struct perf_evsel *evsel,
2653
+			     struct evsel *evsel,
2629
2654
 			     struct perf_sample *sample,
2630
2655
 			     struct machine *machine __maybe_unused)
2631
2656
 {
..
..
@@ -2641,7 +2666,7 @@
2641
2666
 
2642
2667
 	timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2643
2668
 	printf("%15s ", tstr);
2644
-	printf("lost %" PRIu64 " events on cpu %d\n", event->lost.lost, sample->cpu);
2669
+	printf("lost %" PRI_lu64 " events on cpu %d\n", event->lost.lost, sample->cpu);
2645
2670
 
2646
2671
 	return 0;
2647
2672
 }
..
..
@@ -2762,12 +2787,12 @@
2762
2787
 	return ret;
2763
2788
 }
2764
2789
 
2765
-static size_t timehist_print_idlehist_callchain(struct rb_root *root)
2790
+static size_t timehist_print_idlehist_callchain(struct rb_root_cached *root)
2766
2791
 {
2767
2792
 	size_t ret = 0;
2768
2793
 	FILE *fp = stdout;
2769
2794
 	struct callchain_node *chain;
2770
-	struct rb_node *rb_node = rb_first(root);
2795
+	struct rb_node *rb_node = rb_first_cached(root);
2771
2796
 
2772
2797
 	printf("  %16s  %8s  %s\n", "Idle time (msec)", "Count", "Callchains");
2773
2798
 	printf("  %.16s  %.8s  %.50s\n", graph_dotted_line, graph_dotted_line,
..
..
@@ -2836,6 +2861,9 @@
2836
2861
 
2837
2862
 	printf("\nIdle stats:\n");
2838
2863
 	for (i = 0; i < idle_max_cpu; ++i) {
2864
+		if (cpu_list && !test_bit(i, cpu_bitmap))
2865
+			continue;
2866
+
2839
2867
 		t = idle_threads[i];
2840
2868
 		if (!t)
2841
2869
 			continue;
..
..
@@ -2868,7 +2896,7 @@
2868
2896
 			if (itr == NULL)
2869
2897
 				continue;
2870
2898
 
2871
-			callchain_param.sort(&itr->sorted_root, &itr->callchain,
2899
+			callchain_param.sort(&itr->sorted_root.rb_root, &itr->callchain,
2872
2900
 					     0, &callchain_param);
2873
2901
 
2874
2902
 			printf("  CPU %2d:", i);
..
..
@@ -2895,14 +2923,14 @@
2895
2923
 
2896
2924
 typedef int (*sched_handler)(struct perf_tool *tool,
2897
2925
 			  union perf_event *event,
2898
-			  struct perf_evsel *evsel,
2926
+			  struct evsel *evsel,
2899
2927
 			  struct perf_sample *sample,
2900
2928
 			  struct machine *machine);
2901
2929
 
2902
2930
 static int perf_timehist__process_sample(struct perf_tool *tool,
2903
2931
 					 union perf_event *event,
2904
2932
 					 struct perf_sample *sample,
2905
-					 struct perf_evsel *evsel,
2933
+					 struct evsel *evsel,
2906
2934
 					 struct machine *machine)
2907
2935
 {
2908
2936
 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
..
..
@@ -2922,13 +2950,13 @@
2922
2950
 }
2923
2951
 
2924
2952
 static int timehist_check_attr(struct perf_sched *sched,
2925
-			       struct perf_evlist *evlist)
2953
+			       struct evlist *evlist)
2926
2954
 {
2927
-	struct perf_evsel *evsel;
2955
+	struct evsel *evsel;
2928
2956
 	struct evsel_runtime *er;
2929
2957
 
2930
-	list_for_each_entry(evsel, &evlist->entries, node) {
2931
-		er = perf_evsel__get_runtime(evsel);
2958
+	list_for_each_entry(evsel, &evlist->core.entries, core.node) {
2959
+		er = evsel__get_runtime(evsel);
2932
2960
 		if (er == NULL) {
2933
2961
 			pr_err("Failed to allocate memory for evsel runtime data\n");
2934
2962
 			return -1;
..
..
@@ -2946,24 +2974,23 @@
2946
2974
 
2947
2975
 static int perf_sched__timehist(struct perf_sched *sched)
2948
2976
 {
2949
-	const struct perf_evsel_str_handler handlers[] = {
2977
+	struct evsel_str_handler handlers[] = {
2950
2978
 		{ "sched:sched_switch",       timehist_sched_switch_event, },
2951
2979
 		{ "sched:sched_wakeup",	      timehist_sched_wakeup_event, },
2980
+		{ "sched:sched_waking",       timehist_sched_wakeup_event, },
2952
2981
 		{ "sched:sched_wakeup_new",   timehist_sched_wakeup_event, },
2953
2982
 	};
2954
-	const struct perf_evsel_str_handler migrate_handlers[] = {
2983
+	const struct evsel_str_handler migrate_handlers[] = {
2955
2984
 		{ "sched:sched_migrate_task", timehist_migrate_task_event, },
2956
2985
 	};
2957
2986
 	struct perf_data data = {
2958
-		.file      = {
2959
-			.path = input_name,
2960
-		},
2961
-		.mode      = PERF_DATA_MODE_READ,
2962
-		.force     = sched->force,
2987
+		.path  = input_name,
2988
+		.mode  = PERF_DATA_MODE_READ,
2989
+		.force = sched->force,
2963
2990
 	};
2964
2991
 
2965
2992
 	struct perf_session *session;
2966
-	struct perf_evlist *evlist;
2993
+	struct evlist *evlist;
2967
2994
 	int err = -1;
2968
2995
 
2969
2996
 	/*
..
..
@@ -2985,8 +3012,14 @@
2985
3012
 	symbol_conf.use_callchain = sched->show_callchain;
2986
3013
 
2987
3014
 	session = perf_session__new(&data, false, &sched->tool);
2988
-	if (session == NULL)
2989
-		return -ENOMEM;
3015
+	if (IS_ERR(session))
3016
+		return PTR_ERR(session);
3017
+
3018
+	if (cpu_list) {
3019
+		err = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap);
3020
+		if (err < 0)
3021
+			goto out;
3022
+	}
2990
3023
 
2991
3024
 	evlist = session->evlist;
2992
3025
 
..
..
@@ -3001,6 +3034,11 @@
3001
3034
 		goto out;
3002
3035
 
3003
3036
 	setup_pager();
3037
+
3038
+	/* prefer sched_waking if it is captured */
3039
+	if (perf_evlist__find_tracepoint_by_name(session->evlist,
3040
+						  "sched:sched_waking"))
3041
+		handlers[1].handler = timehist_sched_wakeup_ignore;
3004
3042
 
3005
3043
 	/* setup per-evsel handlers */
3006
3044
 	if (perf_session__set_tracepoints_handlers(session, handlers))
..
..
@@ -3074,11 +3112,12 @@
3074
3112
 	}
3075
3113
 }
3076
3114
 
3077
-static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data)
3115
+static void __merge_work_atoms(struct rb_root_cached *root, struct work_atoms *data)
3078
3116
 {
3079
-	struct rb_node **new = &(root->rb_node), *parent = NULL;
3117
+	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
3080
3118
 	struct work_atoms *this;
3081
3119
 	const char *comm = thread__comm_str(data->thread), *this_comm;
3120
+	bool leftmost = true;
3082
3121
 
3083
3122
 	while (*new) {
3084
3123
 		int cmp;
..
..
@@ -3092,6 +3131,7 @@
3092
3131
 			new = &((*new)->rb_left);
3093
3132
 		} else if (cmp < 0) {
3094
3133
 			new = &((*new)->rb_right);
3134
+			leftmost = false;
3095
3135
 		} else {
3096
3136
 			this->num_merged++;
3097
3137
 			this->total_runtime += data->total_runtime;
..
..
@@ -3100,7 +3140,8 @@
3100
3140
 			list_splice(&data->work_list, &this->work_list);
3101
3141
 			if (this->max_lat < data->max_lat) {
3102
3142
 				this->max_lat = data->max_lat;
3103
-				this->max_lat_at = data->max_lat_at;
3143
+				this->max_lat_start = data->max_lat_start;
3144
+				this->max_lat_end = data->max_lat_end;
3104
3145
 			}
3105
3146
 			zfree(&data);
3106
3147
 			return;
..
..
@@ -3109,7 +3150,7 @@
3109
3150
 
3110
3151
 	data->num_merged++;
3111
3152
 	rb_link_node(&data->node, parent, new);
3112
-	rb_insert_color(&data->node, root);
3153
+	rb_insert_color_cached(&data->node, root, leftmost);
3113
3154
 }
3114
3155
 
3115
3156
 static void perf_sched__merge_lat(struct perf_sched *sched)
..
..
@@ -3120,8 +3161,8 @@
3120
3161
 	if (sched->skip_merge)
3121
3162
 		return;
3122
3163
 
3123
-	while ((node = rb_first(&sched->atom_root))) {
3124
-		rb_erase(node, &sched->atom_root);
3164
+	while ((node = rb_first_cached(&sched->atom_root))) {
3165
+		rb_erase_cached(node, &sched->atom_root);
3125
3166
 		data = rb_entry(node, struct work_atoms, node);
3126
3167
 		__merge_work_atoms(&sched->merged_atom_root, data);
3127
3168
 	}
..
..
@@ -3139,11 +3180,11 @@
3139
3180
 	perf_sched__merge_lat(sched);
3140
3181
 	perf_sched__sort_lat(sched);
3141
3182
 
3142
-	printf("\n -----------------------------------------------------------------------------------------------------------------\n");
3143
-	printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms | Maximum delay at       |\n");
3144
-	printf(" -----------------------------------------------------------------------------------------------------------------\n");
3183
+	printf("\n -------------------------------------------------------------------------------------------------------------------------------------------\n");
3184
+	printf("  Task                  |   Runtime ms  | Switches | Avg delay ms    | Max delay ms    | Max delay start           | Max delay end          |\n");
3185
+	printf(" -------------------------------------------------------------------------------------------------------------------------------------------\n");
3145
3186
 
3146
-	next = rb_first(&sched->sorted_atom_root);
3187
+	next = rb_first_cached(&sched->sorted_atom_root);
3147
3188
 
3148
3189
 	while (next) {
3149
3190
 		struct work_atoms *work_list;
..
..
@@ -3168,7 +3209,7 @@
3168
3209
 
3169
3210
 static int setup_map_cpus(struct perf_sched *sched)
3170
3211
 {
3171
-	struct cpu_map *map;
3212
+	struct perf_cpu_map *map;
3172
3213
 
3173
3214
 	sched->max_cpu  = sysconf(_SC_NPROCESSORS_CONF);
3174
3215
 
..
..
@@ -3181,7 +3222,7 @@
3181
3222
 	if (!sched->map.cpus_str)
3182
3223
 		return 0;
3183
3224
 
3184
-	map = cpu_map__new(sched->map.cpus_str);
3225
+	map = perf_cpu_map__new(sched->map.cpus_str);
3185
3226
 	if (!map) {
3186
3227
 		pr_err("failed to get cpus map from %s\n", sched->map.cpus_str);
3187
3228
 		return -1;
..
..
@@ -3193,7 +3234,7 @@
3193
3234
 
3194
3235
 static int setup_color_pids(struct perf_sched *sched)
3195
3236
 {
3196
-	struct thread_map *map;
3237
+	struct perf_thread_map *map;
3197
3238
 
3198
3239
 	if (!sched->map.color_pids_str)
3199
3240
 		return 0;
..
..
@@ -3210,12 +3251,12 @@
3210
3251
 
3211
3252
 static int setup_color_cpus(struct perf_sched *sched)
3212
3253
 {
3213
-	struct cpu_map *map;
3254
+	struct perf_cpu_map *map;
3214
3255
 
3215
3256
 	if (!sched->map.color_cpus_str)
3216
3257
 		return 0;
3217
3258
 
3218
-	map = cpu_map__new(sched->map.color_cpus_str);
3259
+	map = perf_cpu_map__new(sched->map.color_cpus_str);
3219
3260
 	if (!map) {
3220
3261
 		pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str);
3221
3262
 		return -1;
..
..
@@ -3296,6 +3337,16 @@
3296
3337
 	sort_dimension__add("pid", &sched->cmp_pid);
3297
3338
 }
3298
3339
 
3340
+static bool schedstat_events_exposed(void)
3341
+{
3342
+	/*
3343
+	 * Select "sched:sched_stat_wait" event to check
3344
+	 * whether schedstat tracepoints are exposed.
3345
+	 */
3346
+	return IS_ERR(trace_event__tp_format("sched", "sched_stat_wait")) ?
3347
+		false : true;
3348
+}
3349
+
3299
3350
 static int __cmd_record(int argc, const char **argv)
3300
3351
 {
3301
3352
 	unsigned int rec_argc, i, j;
..
..
@@ -3307,17 +3358,33 @@
3307
3358
 		"-m", "1024",
3308
3359
 		"-c", "1",
3309
3360
 		"-e", "sched:sched_switch",
3310
-		"-e", "sched:sched_stat_wait",
3311
-		"-e", "sched:sched_stat_sleep",
3312
-		"-e", "sched:sched_stat_iowait",
3313
3361
 		"-e", "sched:sched_stat_runtime",
3314
3362
 		"-e", "sched:sched_process_fork",
3315
-		"-e", "sched:sched_wakeup",
3316
3363
 		"-e", "sched:sched_wakeup_new",
3317
3364
 		"-e", "sched:sched_migrate_task",
3318
3365
 	};
3319
3366
 
3320
-	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
3367
+	/*
3368
+	 * The tracepoints trace_sched_stat_{wait, sleep, iowait}
3369
+	 * are not exposed to user if CONFIG_SCHEDSTATS is not set,
3370
+	 * to prevent "perf sched record" execution failure, determine
3371
+	 * whether to record schedstat events according to actual situation.
3372
+	 */
3373
+	const char * const schedstat_args[] = {
3374
+		"-e", "sched:sched_stat_wait",
3375
+		"-e", "sched:sched_stat_sleep",
3376
+		"-e", "sched:sched_stat_iowait",
3377
+	};
3378
+	unsigned int schedstat_argc = schedstat_events_exposed() ?
3379
+		ARRAY_SIZE(schedstat_args) : 0;
3380
+
3381
+	struct tep_event *waking_event;
3382
+
3383
+	/*
3384
+	 * +2 for either "-e", "sched:sched_wakeup" or
3385
+	 * "-e", "sched:sched_waking"
3386
+	 */
3387
+	rec_argc = ARRAY_SIZE(record_args) + 2 + schedstat_argc + argc - 1;
3321
3388
 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
3322
3389
 
3323
3390
 	if (rec_argv == NULL)
..
..
@@ -3325,6 +3392,16 @@
3325
3392
 
3326
3393
 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
3327
3394
 		rec_argv[i] = strdup(record_args[i]);
3395
+
3396
+	rec_argv[i++] = "-e";
3397
+	waking_event = trace_event__tp_format("sched", "sched_waking");
3398
+	if (!IS_ERR(waking_event))
3399
+		rec_argv[i++] = strdup("sched:sched_waking");
3400
+	else
3401
+		rec_argv[i++] = strdup("sched:sched_wakeup");
3402
+
3403
+	for (j = 0; j < schedstat_argc; j++)
3404
+		rec_argv[i++] = strdup(schedstat_args[j]);
3328
3405
 
3329
3406
 	for (j = 1; j < (unsigned int)argc; j++, i++)
3330
3407
 		rec_argv[i] = argv[j];
..
..
@@ -3336,7 +3413,7 @@
3336
3413
 
3337
3414
 int cmd_sched(int argc, const char **argv)
3338
3415
 {
3339
-	const char default_sort_order[] = "avg, max, switch, runtime";
3416
+	static const char default_sort_order[] = "avg, max, switch, runtime";
3340
3417
 	struct perf_sched sched = {
3341
3418
 		.tool = {
3342
3419
 			.sample		 = perf_sched__process_tracepoint_sample,
..
..
@@ -3421,6 +3498,7 @@
3421
3498
 		   "analyze events only for given process id(s)"),
3422
3499
 	OPT_STRING('t', "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
3423
3500
 		   "analyze events only for given thread id(s)"),
3501
+	OPT_STRING('C', "cpu", &cpu_list, "cpu", "list of cpus to profile"),
3424
3502
 	OPT_PARENT(sched_options)
3425
3503
 	};
3426
3504