aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorIngo Molnar <mingo@kernel.org>2013-01-31 04:20:14 -0500
committerIngo Molnar <mingo@kernel.org>2013-01-31 04:20:14 -0500
commit152fefa921535665f95840c08062844ab2f5593e (patch)
treeb6ff202ebeca4341a1332258a04403f8ce95e75a
parenta2d28d0c198b65fac28ea6212f5f8edc77b29c27 (diff)
parent5809fde040de2afa477a6c593ce2e8fd2c11d9d3 (diff)
Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo: . Fix some leaks in exit paths. . Use memdup where applicable . Remove some die() calls, allowing callers to handle exit paths gracefully. . Correct typo in tools Makefile, fix from Borislav Petkov. . Add 'perf bench numa mem' NUMA performance measurement suite, from Ingo Molnar. . Handle dynamic array's element size properly, fix from Jiri Olsa. . Fix memory leaks on evsel->counts, from Namhyung Kim. . Make numa benchmark optional, allowing the build in machines where required numa libraries are not present, fix from Peter Hurley. . Add interval printing in 'perf stat', from Stephane Eranian. . Fix compile warnings in tests/attr.c, from Sukadev Bhattiprolu. . Fix double free, pclose instead of fclose, leaks and double fclose errors found with the cppcheck tool, from Thomas Jarosch. Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Diffstat
-rw-r--r--tools/Makefile2
-rw-r--r--tools/lib/traceevent/event-parse.c39
-rw-r--r--tools/perf/Documentation/perf-stat.txt4
-rw-r--r--tools/perf/Makefile13
-rw-r--r--tools/perf/arch/common.c1
-rw-r--r--tools/perf/bench/bench.h1
-rw-r--r--tools/perf/bench/numa.c1731
-rw-r--r--tools/perf/builtin-bench.c17
-rw-r--r--tools/perf/builtin-kmem.c6
-rw-r--r--tools/perf/builtin-stat.c158
-rw-r--r--tools/perf/config/feature-tests.mak11
-rw-r--r--tools/perf/tests/attr.c5
-rw-r--r--tools/perf/tests/open-syscall-all-cpus.c1
-rw-r--r--tools/perf/tests/perf-record.c12
-rw-r--r--tools/perf/tests/vmlinux-kallsyms.c4
-rw-r--r--tools/perf/ui/browser.c2
-rw-r--r--tools/perf/util/event.c4
-rw-r--r--tools/perf/util/evsel.c31
-rw-r--r--tools/perf/util/evsel.h2
-rw-r--r--tools/perf/util/header.c25
-rw-r--r--tools/perf/util/map.c118
-rw-r--r--tools/perf/util/map.h24
-rw-r--r--tools/perf/util/sort.c7
-rw-r--r--tools/perf/util/strlist.c54
-rw-r--r--tools/perf/util/strlist.h42
25 files changed, 2154 insertions, 160 deletions
diff --git a/tools/Makefile b/tools/Makefile
index 1f9a529fe544..798fa0ef048e 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -15,7 +15,7 @@ help:
15 @echo ' x86_energy_perf_policy - Intel energy policy tool' 15 @echo ' x86_energy_perf_policy - Intel energy policy tool'
16 @echo '' 16 @echo ''
17 @echo 'You can do:' 17 @echo 'You can do:'
18 @echo ' $$ make -C tools/<tool>_install' 18 @echo ' $$ make -C tools/ <tool>_install'
19 @echo '' 19 @echo ''
20 @echo ' from the kernel command line to build and install one of' 20 @echo ' from the kernel command line to build and install one of'
21 @echo ' the tools above' 21 @echo ' the tools above'
diff --git a/tools/lib/traceevent/event-parse.c b/tools/lib/traceevent/event-parse.c
index bb8b3db0e583..82b0606dcb8a 100644
--- a/tools/lib/traceevent/event-parse.c
+++ b/tools/lib/traceevent/event-parse.c
@@ -1223,6 +1223,34 @@ static int field_is_long(struct format_field *field)
1223 return 0; 1223 return 0;
1224} 1224}
1225 1225
1226static unsigned int type_size(const char *name)
1227{
1228 /* This covers all FIELD_IS_STRING types. */
1229 static struct {
1230 const char *type;
1231 unsigned int size;
1232 } table[] = {
1233 { "u8", 1 },
1234 { "u16", 2 },
1235 { "u32", 4 },
1236 { "u64", 8 },
1237 { "s8", 1 },
1238 { "s16", 2 },
1239 { "s32", 4 },
1240 { "s64", 8 },
1241 { "char", 1 },
1242 { },
1243 };
1244 int i;
1245
1246 for (i = 0; table[i].type; i++) {
1247 if (!strcmp(table[i].type, name))
1248 return table[i].size;
1249 }
1250
1251 return 0;
1252}
1253
1226static int event_read_fields(struct event_format *event, struct format_field **fields) 1254static int event_read_fields(struct event_format *event, struct format_field **fields)
1227{ 1255{
1228 struct format_field *field = NULL; 1256 struct format_field *field = NULL;
@@ -1232,6 +1260,8 @@ static int event_read_fields(struct event_format *event, struct format_field **f
1232 int count = 0; 1260 int count = 0;
1233 1261
1234 do { 1262 do {
1263 unsigned int size_dynamic = 0;
1264
1235 type = read_token(&token); 1265 type = read_token(&token);
1236 if (type == EVENT_NEWLINE) { 1266 if (type == EVENT_NEWLINE) {
1237 free_token(token); 1267 free_token(token);
@@ -1390,6 +1420,7 @@ static int event_read_fields(struct event_format *event, struct format_field **f
1390 field->type = new_type; 1420 field->type = new_type;
1391 strcat(field->type, " "); 1421 strcat(field->type, " ");
1392 strcat(field->type, field->name); 1422 strcat(field->type, field->name);
1423 size_dynamic = type_size(field->name);
1393 free_token(field->name); 1424 free_token(field->name);
1394 strcat(field->type, brackets); 1425 strcat(field->type, brackets);
1395 field->name = token; 1426 field->name = token;
@@ -1478,10 +1509,14 @@ static int event_read_fields(struct event_format *event, struct format_field **f
1478 if (field->flags & FIELD_IS_ARRAY) { 1509 if (field->flags & FIELD_IS_ARRAY) {
1479 if (field->arraylen) 1510 if (field->arraylen)
1480 field->elementsize = field->size / field->arraylen; 1511 field->elementsize = field->size / field->arraylen;
1512 else if (field->flags & FIELD_IS_DYNAMIC)
1513 field->elementsize = size_dynamic;
1481 else if (field->flags & FIELD_IS_STRING) 1514 else if (field->flags & FIELD_IS_STRING)
1482 field->elementsize = 1; 1515 field->elementsize = 1;
1483 else 1516 else if (field->flags & FIELD_IS_LONG)
1484 field->elementsize = event->pevent->long_size; 1517 field->elementsize = event->pevent ?
1518 event->pevent->long_size :
1519 sizeof(long);
1485 } else 1520 } else
1486 field->elementsize = field->size; 1521 field->elementsize = field->size;
1487 1522
diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index cf0c3107e06e..5289da3344e9 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -114,6 +114,10 @@ with it. --append may be used here. Examples:
114 114
115perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage 115perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- make -s -j64 O=defconfig-build/ bzImage
116 116
117-I msecs::
118--interval-print msecs::
119 print count deltas every N milliseconds (minimum: 100ms)
120 example: perf stat -I 1000 -e cycles -a sleep 5
117 121
118EXAMPLES 122EXAMPLES
119-------- 123--------
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index a84021abb3fe..4b1044cbd84c 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -47,6 +47,8 @@ include config/utilities.mak
47# backtrace post unwind. 47# backtrace post unwind.
48# 48#
49# Define NO_BACKTRACE if you do not want stack backtrace debug feature 49# Define NO_BACKTRACE if you do not want stack backtrace debug feature
50#
51# Define NO_LIBNUMA if you do not want numa perf benchmark
50 52
51$(OUTPUT)PERF-VERSION-FILE: .FORCE-PERF-VERSION-FILE 53$(OUTPUT)PERF-VERSION-FILE: .FORCE-PERF-VERSION-FILE
52 @$(SHELL_PATH) util/PERF-VERSION-GEN $(OUTPUT) 54 @$(SHELL_PATH) util/PERF-VERSION-GEN $(OUTPUT)
@@ -838,6 +840,17 @@ ifndef NO_BACKTRACE
838 endif 840 endif
839endif 841endif
840 842
843ifndef NO_LIBNUMA
844 FLAGS_LIBNUMA = $(ALL_CFLAGS) $(ALL_LDFLAGS) -lnuma
845 ifneq ($(call try-cc,$(SOURCE_LIBNUMA),$(FLAGS_LIBNUMA),libnuma),y)
846 msg := $(warning No numa.h found, disables 'perf bench numa mem' benchmark, please install numa-libs-devel or libnuma-dev);
847 else
848 BASIC_CFLAGS += -DLIBNUMA_SUPPORT
849 BUILTIN_OBJS += $(OUTPUT)bench/numa.o
850 EXTLIBS += -lnuma
851 endif
852endif
853
841ifdef ASCIIDOC8 854ifdef ASCIIDOC8
842 export ASCIIDOC8 855 export ASCIIDOC8
843endif 856endif
diff --git a/tools/perf/arch/common.c b/tools/perf/arch/common.c
index 3e975cb6232e..aacef07ebf31 100644
--- a/tools/perf/arch/common.c
+++ b/tools/perf/arch/common.c
@@ -155,6 +155,7 @@ static int perf_session_env__lookup_binutils_path(struct perf_session_env *env,
155 if (lookup_path(buf)) 155 if (lookup_path(buf))
156 goto out; 156 goto out;
157 free(buf); 157 free(buf);
158 buf = NULL;
158 } 159 }
159 160
160 if (!strcmp(arch, "arm")) 161 if (!strcmp(arch, "arm"))
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index 8f89998eeaf4..a5223e6a7b43 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -1,6 +1,7 @@
1#ifndef BENCH_H 1#ifndef BENCH_H
2#define BENCH_H 2#define BENCH_H
3 3
4extern int bench_numa(int argc, const char **argv, const char *prefix);
4extern int bench_sched_messaging(int argc, const char **argv, const char *prefix); 5extern int bench_sched_messaging(int argc, const char **argv, const char *prefix);
5extern int bench_sched_pipe(int argc, const char **argv, const char *prefix); 6extern int bench_sched_pipe(int argc, const char **argv, const char *prefix);
6extern int bench_mem_memcpy(int argc, const char **argv, 7extern int bench_mem_memcpy(int argc, const char **argv,
diff --git a/tools/perf/bench/numa.c b/tools/perf/bench/numa.c
new file mode 100644
index 000000000000..30d1c3225b46
--- /dev/null
+++ b/tools/perf/bench/numa.c
@@ -0,0 +1,1731 @@
1/*
2 * numa.c
3 *
4 * numa: Simulate NUMA-sensitive workload and measure their NUMA performance
5 */
6
7#include "../perf.h"
8#include "../builtin.h"
9#include "../util/util.h"
10#include "../util/parse-options.h"
11
12#include "bench.h"
13
14#include <errno.h>
15#include <sched.h>
16#include <stdio.h>
17#include <assert.h>
18#include <malloc.h>
19#include <signal.h>
20#include <stdlib.h>
21#include <string.h>
22#include <unistd.h>
23#include <pthread.h>
24#include <sys/mman.h>
25#include <sys/time.h>
26#include <sys/wait.h>
27#include <sys/prctl.h>
28#include <sys/types.h>
29
30#include <numa.h>
31#include <numaif.h>
32
33/*
34 * Regular printout to the terminal, supressed if -q is specified:
35 */
36#define tprintf(x...) do { if (g && g->p.show_details >= 0) printf(x); } while (0)
37
38/*
39 * Debug printf:
40 */
41#define dprintf(x...) do { if (g && g->p.show_details >= 1) printf(x); } while (0)
42
43struct thread_data {
44 int curr_cpu;
45 cpu_set_t bind_cpumask;
46 int bind_node;
47 u8 *process_data;
48 int process_nr;
49 int thread_nr;
50 int task_nr;
51 unsigned int loops_done;
52 u64 val;
53 u64 runtime_ns;
54 pthread_mutex_t *process_lock;
55};
56
57/* Parameters set by options: */
58
59struct params {
60 /* Startup synchronization: */
61 bool serialize_startup;
62
63 /* Task hierarchy: */
64 int nr_proc;
65 int nr_threads;
66
67 /* Working set sizes: */
68 const char *mb_global_str;
69 const char *mb_proc_str;
70 const char *mb_proc_locked_str;
71 const char *mb_thread_str;
72
73 double mb_global;
74 double mb_proc;
75 double mb_proc_locked;
76 double mb_thread;
77
78 /* Access patterns to the working set: */
79 bool data_reads;
80 bool data_writes;
81 bool data_backwards;
82 bool data_zero_memset;
83 bool data_rand_walk;
84 u32 nr_loops;
85 u32 nr_secs;
86 u32 sleep_usecs;
87
88 /* Working set initialization: */
89 bool init_zero;
90 bool init_random;
91 bool init_cpu0;
92
93 /* Misc options: */
94 int show_details;
95 int run_all;
96 int thp;
97
98 long bytes_global;
99 long bytes_process;
100 long bytes_process_locked;
101 long bytes_thread;
102
103 int nr_tasks;
104 bool show_quiet;
105
106 bool show_convergence;
107 bool measure_convergence;
108
109 int perturb_secs;
110 int nr_cpus;
111 int nr_nodes;
112
113 /* Affinity options -C and -N: */
114 char *cpu_list_str;
115 char *node_list_str;
116};
117
118
119/* Global, read-writable area, accessible to all processes and threads: */
120
121struct global_info {
122 u8 *data;
123
124 pthread_mutex_t startup_mutex;
125 int nr_tasks_started;
126
127 pthread_mutex_t startup_done_mutex;
128
129 pthread_mutex_t start_work_mutex;
130 int nr_tasks_working;
131
132 pthread_mutex_t stop_work_mutex;
133 u64 bytes_done;
134
135 struct thread_data *threads;
136
137 /* Convergence latency measurement: */
138 bool all_converged;
139 bool stop_work;
140
141 int print_once;
142
143 struct params p;
144};
145
146static struct global_info *g = NULL;
147
148static int parse_cpus_opt(const struct option *opt, const char *arg, int unset);
149static int parse_nodes_opt(const struct option *opt, const char *arg, int unset);
150
151struct params p0;
152
153static const struct option options[] = {
154 OPT_INTEGER('p', "nr_proc" , &p0.nr_proc, "number of processes"),
155 OPT_INTEGER('t', "nr_threads" , &p0.nr_threads, "number of threads per process"),
156
157 OPT_STRING('G', "mb_global" , &p0.mb_global_str, "MB", "global memory (MBs)"),
158 OPT_STRING('P', "mb_proc" , &p0.mb_proc_str, "MB", "process memory (MBs)"),
159 OPT_STRING('L', "mb_proc_locked", &p0.mb_proc_locked_str,"MB", "process serialized/locked memory access (MBs), <= process_memory"),
160 OPT_STRING('T', "mb_thread" , &p0.mb_thread_str, "MB", "thread memory (MBs)"),
161
162 OPT_UINTEGER('l', "nr_loops" , &p0.nr_loops, "max number of loops to run"),
163 OPT_UINTEGER('s', "nr_secs" , &p0.nr_secs, "max number of seconds to run"),
164 OPT_UINTEGER('u', "usleep" , &p0.sleep_usecs, "usecs to sleep per loop iteration"),
165
166 OPT_BOOLEAN('R', "data_reads" , &p0.data_reads, "access the data via writes (can be mixed with -W)"),
167 OPT_BOOLEAN('W', "data_writes" , &p0.data_writes, "access the data via writes (can be mixed with -R)"),
168 OPT_BOOLEAN('B', "data_backwards", &p0.data_backwards, "access the data backwards as well"),
169 OPT_BOOLEAN('Z', "data_zero_memset", &p0.data_zero_memset,"access the data via glibc bzero only"),
170 OPT_BOOLEAN('r', "data_rand_walk", &p0.data_rand_walk, "access the data with random (32bit LFSR) walk"),
171
172
173 OPT_BOOLEAN('z', "init_zero" , &p0.init_zero, "bzero the initial allocations"),
174 OPT_BOOLEAN('I', "init_random" , &p0.init_random, "randomize the contents of the initial allocations"),
175 OPT_BOOLEAN('0', "init_cpu0" , &p0.init_cpu0, "do the initial allocations on CPU#0"),
176 OPT_INTEGER('x', "perturb_secs", &p0.perturb_secs, "perturb thread 0/0 every X secs, to test convergence stability"),
177
178 OPT_INCR ('d', "show_details" , &p0.show_details, "Show details"),
179 OPT_INCR ('a', "all" , &p0.run_all, "Run all tests in the suite"),
180 OPT_INTEGER('H', "thp" , &p0.thp, "MADV_NOHUGEPAGE < 0 < MADV_HUGEPAGE"),
181 OPT_BOOLEAN('c', "show_convergence", &p0.show_convergence, "show convergence details"),
182 OPT_BOOLEAN('m', "measure_convergence", &p0.measure_convergence, "measure convergence latency"),
183 OPT_BOOLEAN('q', "quiet" , &p0.show_quiet, "bzero the initial allocations"),
184 OPT_BOOLEAN('S', "serialize-startup", &p0.serialize_startup,"serialize thread startup"),
185
186 /* Special option string parsing callbacks: */
187 OPT_CALLBACK('C', "cpus", NULL, "cpu[,cpu2,...cpuN]",
188 "bind the first N tasks to these specific cpus (the rest is unbound)",
189 parse_cpus_opt),
190 OPT_CALLBACK('M', "memnodes", NULL, "node[,node2,...nodeN]",
191 "bind the first N tasks to these specific memory nodes (the rest is unbound)",
192 parse_nodes_opt),
193 OPT_END()
194};
195
196static const char * const bench_numa_usage[] = {
197 "perf bench numa <options>",
198 NULL
199};
200
201static const char * const numa_usage[] = {
202 "perf bench numa mem [<options>]",
203 NULL
204};
205
206static cpu_set_t bind_to_cpu(int target_cpu)
207{
208 cpu_set_t orig_mask, mask;
209 int ret;
210
211 ret = sched_getaffinity(0, sizeof(orig_mask), &orig_mask);
212 BUG_ON(ret);
213
214 CPU_ZERO(&mask);
215
216 if (target_cpu == -1) {
217 int cpu;
218
219 for (cpu = 0; cpu < g->p.nr_cpus; cpu++)
220 CPU_SET(cpu, &mask);
221 } else {
222 BUG_ON(target_cpu < 0 || target_cpu >= g->p.nr_cpus);
223 CPU_SET(target_cpu, &mask);
224 }
225
226 ret = sched_setaffinity(0, sizeof(mask), &mask);
227 BUG_ON(ret);
228
229 return orig_mask;
230}
231
232static cpu_set_t bind_to_node(int target_node)
233{
234 int cpus_per_node = g->p.nr_cpus/g->p.nr_nodes;
235 cpu_set_t orig_mask, mask;
236 int cpu;
237 int ret;
238
239 BUG_ON(cpus_per_node*g->p.nr_nodes != g->p.nr_cpus);
240 BUG_ON(!cpus_per_node);
241
242 ret = sched_getaffinity(0, sizeof(orig_mask), &orig_mask);
243 BUG_ON(ret);
244
245 CPU_ZERO(&mask);
246
247 if (target_node == -1) {
248 for (cpu = 0; cpu < g->p.nr_cpus; cpu++)
249 CPU_SET(cpu, &mask);
250 } else {
251 int cpu_start = (target_node + 0) * cpus_per_node;
252 int cpu_stop = (target_node + 1) * cpus_per_node;
253
254 BUG_ON(cpu_stop > g->p.nr_cpus);
255
256 for (cpu = cpu_start; cpu < cpu_stop; cpu++)
257 CPU_SET(cpu, &mask);
258 }
259
260 ret = sched_setaffinity(0, sizeof(mask), &mask);
261 BUG_ON(ret);
262
263 return orig_mask;
264}
265
266static void bind_to_cpumask(cpu_set_t mask)
267{
268 int ret;
269
270 ret = sched_setaffinity(0, sizeof(mask), &mask);
271 BUG_ON(ret);
272}
273
274static void mempol_restore(void)
275{
276 int ret;
277
278 ret = set_mempolicy(MPOL_DEFAULT, NULL, g->p.nr_nodes-1);
279
280 BUG_ON(ret);
281}
282
283static void bind_to_memnode(int node)
284{
285 unsigned long nodemask;
286 int ret;
287
288 if (node == -1)
289 return;
290
291 BUG_ON(g->p.nr_nodes > (int)sizeof(nodemask));
292 nodemask = 1L << node;
293
294 ret = set_mempolicy(MPOL_BIND, &nodemask, sizeof(nodemask)*8);
295 dprintf("binding to node %d, mask: %016lx => %d\n", node, nodemask, ret);
296
297 BUG_ON(ret);
298}
299
300#define HPSIZE (2*1024*1024)
301
302#define set_taskname(fmt...) \
303do { \
304 char name[20]; \
305 \
306 snprintf(name, 20, fmt); \
307 prctl(PR_SET_NAME, name); \
308} while (0)
309
310static u8 *alloc_data(ssize_t bytes0, int map_flags,
311 int init_zero, int init_cpu0, int thp, int init_random)
312{
313 cpu_set_t orig_mask;
314 ssize_t bytes;
315 u8 *buf;
316 int ret;
317
318 if (!bytes0)
319 return NULL;
320
321 /* Allocate and initialize all memory on CPU#0: */
322 if (init_cpu0) {
323 orig_mask = bind_to_node(0);
324 bind_to_memnode(0);
325 }
326
327 bytes = bytes0 + HPSIZE;
328
329 buf = (void *)mmap(0, bytes, PROT_READ|PROT_WRITE, MAP_ANON|map_flags, -1, 0);
330 BUG_ON(buf == (void *)-1);
331
332 if (map_flags == MAP_PRIVATE) {
333 if (thp > 0) {
334 ret = madvise(buf, bytes, MADV_HUGEPAGE);
335 if (ret && !g->print_once) {
336 g->print_once = 1;
337 printf("WARNING: Could not enable THP - do: 'echo madvise > /sys/kernel/mm/transparent_hugepage/enabled'\n");
338 }
339 }
340 if (thp < 0) {
341 ret = madvise(buf, bytes, MADV_NOHUGEPAGE);
342 if (ret && !g->print_once) {
343 g->print_once = 1;
344 printf("WARNING: Could not disable THP: run a CONFIG_TRANSPARENT_HUGEPAGE kernel?\n");
345 }
346 }
347 }
348
349 if (init_zero) {
350 bzero(buf, bytes);
351 } else {
352 /* Initialize random contents, different in each word: */
353 if (init_random) {
354 u64 *wbuf = (void *)buf;
355 long off = rand();
356 long i;
357
358 for (i = 0; i < bytes/8; i++)
359 wbuf[i] = i + off;
360 }
361 }
362
363 /* Align to 2MB boundary: */
364 buf = (void *)(((unsigned long)buf + HPSIZE-1) & ~(HPSIZE-1));
365
366 /* Restore affinity: */
367 if (init_cpu0) {
368 bind_to_cpumask(orig_mask);
369 mempol_restore();
370 }
371
372 return buf;
373}
374
375static void free_data(void *data, ssize_t bytes)
376{
377 int ret;
378
379 if (!data)
380 return;
381
382 ret = munmap(data, bytes);
383 BUG_ON(ret);
384}
385
386/*
387 * Create a shared memory buffer that can be shared between processes, zeroed:
388 */
389static void * zalloc_shared_data(ssize_t bytes)
390{
391 return alloc_data(bytes, MAP_SHARED, 1, g->p.init_cpu0, g->p.thp, g->p.init_random);
392}
393
394/*
395 * Create a shared memory buffer that can be shared between processes:
396 */
397static void * setup_shared_data(ssize_t bytes)
398{
399 return alloc_data(bytes, MAP_SHARED, 0, g->p.init_cpu0, g->p.thp, g->p.init_random);
400}
401
402/*
403 * Allocate process-local memory - this will either be shared between
404 * threads of this process, or only be accessed by this thread:
405 */
406static void * setup_private_data(ssize_t bytes)
407{
408 return alloc_data(bytes, MAP_PRIVATE, 0, g->p.init_cpu0, g->p.thp, g->p.init_random);
409}
410
411/*
412 * Return a process-shared (global) mutex:
413 */
414static void init_global_mutex(pthread_mutex_t *mutex)
415{
416 pthread_mutexattr_t attr;
417
418 pthread_mutexattr_init(&attr);
419 pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED);
420 pthread_mutex_init(mutex, &attr);
421}
422
423static int parse_cpu_list(const char *arg)
424{
425 p0.cpu_list_str = strdup(arg);
426
427 dprintf("got CPU list: {%s}\n", p0.cpu_list_str);
428
429 return 0;
430}
431
432static void parse_setup_cpu_list(void)
433{
434 struct thread_data *td;
435 char *str0, *str;
436 int t;
437
438 if (!g->p.cpu_list_str)
439 return;
440
441 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks);
442
443 str0 = str = strdup(g->p.cpu_list_str);
444 t = 0;
445
446 BUG_ON(!str);
447
448 tprintf("# binding tasks to CPUs:\n");
449 tprintf("# ");
450
451 while (true) {
452 int bind_cpu, bind_cpu_0, bind_cpu_1;
453 char *tok, *tok_end, *tok_step, *tok_len, *tok_mul;
454 int bind_len;
455 int step;
456 int mul;
457
458 tok = strsep(&str, ",");
459 if (!tok)
460 break;
461
462 tok_end = strstr(tok, "-");
463
464 dprintf("\ntoken: {%s}, end: {%s}\n", tok, tok_end);
465 if (!tok_end) {
466 /* Single CPU specified: */
467 bind_cpu_0 = bind_cpu_1 = atol(tok);
468 } else {
469 /* CPU range specified (for example: "5-11"): */
470 bind_cpu_0 = atol(tok);
471 bind_cpu_1 = atol(tok_end + 1);
472 }
473
474 step = 1;
475 tok_step = strstr(tok, "#");
476 if (tok_step) {
477 step = atol(tok_step + 1);
478 BUG_ON(step <= 0 || step >= g->p.nr_cpus);
479 }
480
481 /*
482 * Mask length.
483 * Eg: "--cpus 8_4-16#4" means: '--cpus 8_4,12_4,16_4',
484 * where the _4 means the next 4 CPUs are allowed.
485 */
486 bind_len = 1;
487 tok_len = strstr(tok, "_");
488 if (tok_len) {
489 bind_len = atol(tok_len + 1);
490 BUG_ON(bind_len <= 0 || bind_len > g->p.nr_cpus);
491 }
492
493 /* Multiplicator shortcut, "0x8" is a shortcut for: "0,0,0,0,0,0,0,0" */
494 mul = 1;
495 tok_mul = strstr(tok, "x");
496 if (tok_mul) {
497 mul = atol(tok_mul + 1);
498 BUG_ON(mul <= 0);
499 }
500
501 dprintf("CPUs: %d_%d-%d#%dx%d\n", bind_cpu_0, bind_len, bind_cpu_1, step, mul);
502
503 BUG_ON(bind_cpu_0 < 0 || bind_cpu_0 >= g->p.nr_cpus);
504 BUG_ON(bind_cpu_1 < 0 || bind_cpu_1 >= g->p.nr_cpus);
505 BUG_ON(bind_cpu_0 > bind_cpu_1);
506
507 for (bind_cpu = bind_cpu_0; bind_cpu <= bind_cpu_1; bind_cpu += step) {
508 int i;
509
510 for (i = 0; i < mul; i++) {
511 int cpu;
512
513 if (t >= g->p.nr_tasks) {
514 printf("\n# NOTE: ignoring bind CPUs starting at CPU#%d\n #", bind_cpu);
515 goto out;
516 }
517 td = g->threads + t;
518
519 if (t)
520 tprintf(",");
521 if (bind_len > 1) {
522 tprintf("%2d/%d", bind_cpu, bind_len);
523 } else {
524 tprintf("%2d", bind_cpu);
525 }
526
527 CPU_ZERO(&td->bind_cpumask);
528 for (cpu = bind_cpu; cpu < bind_cpu+bind_len; cpu++) {
529 BUG_ON(cpu < 0 || cpu >= g->p.nr_cpus);
530 CPU_SET(cpu, &td->bind_cpumask);
531 }
532 t++;
533 }
534 }
535 }
536out:
537
538 tprintf("\n");
539
540 if (t < g->p.nr_tasks)
541 printf("# NOTE: %d tasks bound, %d tasks unbound\n", t, g->p.nr_tasks - t);
542
543 free(str0);
544}
545
546static int parse_cpus_opt(const struct option *opt __maybe_unused,
547 const char *arg, int unset __maybe_unused)
548{
549 if (!arg)
550 return -1;
551
552 return parse_cpu_list(arg);
553}
554
555static int parse_node_list(const char *arg)
556{
557 p0.node_list_str = strdup(arg);
558
559 dprintf("got NODE list: {%s}\n", p0.node_list_str);
560
561 return 0;
562}
563
564static void parse_setup_node_list(void)
565{
566 struct thread_data *td;
567 char *str0, *str;
568 int t;
569
570 if (!g->p.node_list_str)
571 return;
572
573 dprintf("g->p.nr_tasks: %d\n", g->p.nr_tasks);
574
575 str0 = str = strdup(g->p.node_list_str);
576 t = 0;
577
578 BUG_ON(!str);
579
580 tprintf("# binding tasks to NODEs:\n");
581 tprintf("# ");
582
583 while (true) {
584 int bind_node, bind_node_0, bind_node_1;
585 char *tok, *tok_end, *tok_step, *tok_mul;
586 int step;
587 int mul;
588
589 tok = strsep(&str, ",");
590 if (!tok)
591 break;
592
593 tok_end = strstr(tok, "-");
594
595 dprintf("\ntoken: {%s}, end: {%s}\n", tok, tok_end);
596 if (!tok_end) {
597 /* Single NODE specified: */
598 bind_node_0 = bind_node_1 = atol(tok);
599 } else {
600 /* NODE range specified (for example: "5-11"): */
601 bind_node_0 = atol(tok);
602 bind_node_1 = atol(tok_end + 1);
603 }
604
605 step = 1;
606 tok_step = strstr(tok, "#");
607 if (tok_step) {
608 step = atol(tok_step + 1);
609 BUG_ON(step <= 0 || step >= g->p.nr_nodes);
610 }
611
612 /* Multiplicator shortcut, "0x8" is a shortcut for: "0,0,0,0,0,0,0,0" */
613 mul = 1;
614 tok_mul = strstr(tok, "x");
615 if (tok_mul) {
616 mul = atol(tok_mul + 1);
617 BUG_ON(mul <= 0);
618 }
619
620 dprintf("NODEs: %d-%d #%d\n", bind_node_0, bind_node_1, step);
621
622 BUG_ON(bind_node_0 < 0 || bind_node_0 >= g->p.nr_nodes);
623 BUG_ON(bind_node_1 < 0 || bind_node_1 >= g->p.nr_nodes);
624 BUG_ON(bind_node_0 > bind_node_1);
625
626 for (bind_node = bind_node_0; bind_node <= bind_node_1; bind_node += step) {
627 int i;
628
629 for (i = 0; i < mul; i++) {
630 if (t >= g->p.nr_tasks) {
631 printf("\n# NOTE: ignoring bind NODEs starting at NODE#%d\n", bind_node);
632 goto out;
633 }
634 td = g->threads + t;
635
636 if (!t)
637 tprintf(" %2d", bind_node);
638 else
639 tprintf(",%2d", bind_node);
640
641 td->bind_node = bind_node;
642 t++;
643 }
644 }
645 }
646out:
647
648 tprintf("\n");
649
650 if (t < g->p.nr_tasks)
651 printf("# NOTE: %d tasks mem-bound, %d tasks unbound\n", t, g->p.nr_tasks - t);
652
653 free(str0);
654}
655
656static int parse_nodes_opt(const struct option *opt __maybe_unused,
657 const char *arg, int unset __maybe_unused)
658{
659 if (!arg)
660 return -1;
661
662 return parse_node_list(arg);
663
664 return 0;
665}
666
667#define BIT(x) (1ul << x)
668
669static inline uint32_t lfsr_32(uint32_t lfsr)
670{
671 const uint32_t taps = BIT(1) | BIT(5) | BIT(6) | BIT(31);
672 return (lfsr>>1) ^ ((0x0u - (lfsr & 0x1u)) & taps);
673}
674
675/*
676 * Make sure there's real data dependency to RAM (when read
677 * accesses are enabled), so the compiler, the CPU and the
678 * kernel (KSM, zero page, etc.) cannot optimize away RAM
679 * accesses:
680 */
681static inline u64 access_data(u64 *data __attribute__((unused)), u64 val)
682{
683 if (g->p.data_reads)
684 val += *data;
685 if (g->p.data_writes)
686 *data = val + 1;
687 return val;
688}
689
690/*
691 * The worker process does two types of work, a forwards going
692 * loop and a backwards going loop.
693 *
694 * We do this so that on multiprocessor systems we do not create
695 * a 'train' of processing, with highly synchronized processes,
696 * skewing the whole benchmark.
697 */
698static u64 do_work(u8 *__data, long bytes, int nr, int nr_max, int loop, u64 val)
699{
700 long words = bytes/sizeof(u64);
701 u64 *data = (void *)__data;
702 long chunk_0, chunk_1;
703 u64 *d0, *d, *d1;
704 long off;
705 long i;
706
707 BUG_ON(!data && words);
708 BUG_ON(data && !words);
709
710 if (!data)
711 return val;
712
713 /* Very simple memset() work variant: */
714 if (g->p.data_zero_memset && !g->p.data_rand_walk) {
715 bzero(data, bytes);
716 return val;
717 }
718
719 /* Spread out by PID/TID nr and by loop nr: */
720 chunk_0 = words/nr_max;
721 chunk_1 = words/g->p.nr_loops;
722 off = nr*chunk_0 + loop*chunk_1;
723
724 while (off >= words)
725 off -= words;
726
727 if (g->p.data_rand_walk) {
728 u32 lfsr = nr + loop + val;
729 int j;
730
731 for (i = 0; i < words/1024; i++) {
732 long start, end;
733
734 lfsr = lfsr_32(lfsr);
735
736 start = lfsr % words;
737 end = min(start + 1024, words-1);
738
739 if (g->p.data_zero_memset) {
740 bzero(data + start, (end-start) * sizeof(u64));
741 } else {
742 for (j = start; j < end; j++)
743 val = access_data(data + j, val);
744 }
745 }
746 } else if (!g->p.data_backwards || (nr + loop) & 1) {
747
748 d0 = data + off;
749 d = data + off + 1;
750 d1 = data + words;
751
752 /* Process data forwards: */
753 for (;;) {
754 if (unlikely(d >= d1))
755 d = data;
756 if (unlikely(d == d0))
757 break;
758
759 val = access_data(d, val);
760
761 d++;
762 }
763 } else {
764 /* Process data backwards: */
765
766 d0 = data + off;
767 d = data + off - 1;
768 d1 = data + words;
769
770 /* Process data forwards: */
771 for (;;) {
772 if (unlikely(d < data))
773 d = data + words-1;
774 if (unlikely(d == d0))
775 break;
776
777 val = access_data(d, val);
778
779 d--;
780 }
781 }
782
783 return val;
784}
785
786static void update_curr_cpu(int task_nr, unsigned long bytes_worked)
787{
788 unsigned int cpu;
789
790 cpu = sched_getcpu();
791
792 g->threads[task_nr].curr_cpu = cpu;
793 prctl(0, bytes_worked);
794}
795
796#define MAX_NR_NODES 64
797
798/*
799 * Count the number of nodes a process's threads
800 * are spread out on.
801 *
802 * A count of 1 means that the process is compressed
803 * to a single node. A count of g->p.nr_nodes means it's
804 * spread out on the whole system.
805 */
806static int count_process_nodes(int process_nr)
807{
808 char node_present[MAX_NR_NODES] = { 0, };
809 int nodes;
810 int n, t;
811
812 for (t = 0; t < g->p.nr_threads; t++) {
813 struct thread_data *td;
814 int task_nr;
815 int node;
816
817 task_nr = process_nr*g->p.nr_threads + t;
818 td = g->threads + task_nr;
819
820 node = numa_node_of_cpu(td->curr_cpu);
821 node_present[node] = 1;
822 }
823
824 nodes = 0;
825
826 for (n = 0; n < MAX_NR_NODES; n++)
827 nodes += node_present[n];
828
829 return nodes;
830}
831
832/*
833 * Count the number of distinct process-threads a node contains.
834 *
835 * A count of 1 means that the node contains only a single
836 * process. If all nodes on the system contain at most one
837 * process then we are well-converged.
838 */
839static int count_node_processes(int node)
840{
841 int processes = 0;
842 int t, p;
843
844 for (p = 0; p < g->p.nr_proc; p++) {
845 for (t = 0; t < g->p.nr_threads; t++) {
846 struct thread_data *td;
847 int task_nr;
848 int n;
849
850 task_nr = p*g->p.nr_threads + t;
851 td = g->threads + task_nr;
852
853 n = numa_node_of_cpu(td->curr_cpu);
854 if (n == node) {
855 processes++;
856 break;
857 }
858 }
859 }
860
861 return processes;
862}
863
864static void calc_convergence_compression(int *strong)
865{
866 unsigned int nodes_min, nodes_max;
867 int p;
868
869 nodes_min = -1;
870 nodes_max = 0;
871
872 for (p = 0; p < g->p.nr_proc; p++) {
873 unsigned int nodes = count_process_nodes(p);
874
875 nodes_min = min(nodes, nodes_min);
876 nodes_max = max(nodes, nodes_max);
877 }
878
879 /* Strong convergence: all threads compress on a single node: */
880 if (nodes_min == 1 && nodes_max == 1) {
881 *strong = 1;
882 } else {
883 *strong = 0;
884 tprintf(" {%d-%d}", nodes_min, nodes_max);
885 }
886}
887
888static void calc_convergence(double runtime_ns_max, double *convergence)
889{
890 unsigned int loops_done_min, loops_done_max;
891 int process_groups;
892 int nodes[MAX_NR_NODES];
893 int distance;
894 int nr_min;
895 int nr_max;
896 int strong;
897 int sum;
898 int nr;
899 int node;
900 int cpu;
901 int t;
902
903 if (!g->p.show_convergence && !g->p.measure_convergence)
904 return;
905
906 for (node = 0; node < g->p.nr_nodes; node++)
907 nodes[node] = 0;
908
909 loops_done_min = -1;
910 loops_done_max = 0;
911
912 for (t = 0; t < g->p.nr_tasks; t++) {
913 struct thread_data *td = g->threads + t;
914 unsigned int loops_done;
915
916 cpu = td->curr_cpu;
917
918 /* Not all threads have written it yet: */
919 if (cpu < 0)
920 continue;
921
922 node = numa_node_of_cpu(cpu);
923
924 nodes[node]++;
925
926 loops_done = td->loops_done;
927 loops_done_min = min(loops_done, loops_done_min);
928 loops_done_max = max(loops_done, loops_done_max);
929 }
930
931 nr_max = 0;
932 nr_min = g->p.nr_tasks;
933 sum = 0;
934
935 for (node = 0; node < g->p.nr_nodes; node++) {
936 nr = nodes[node];
937 nr_min = min(nr, nr_min);
938 nr_max = max(nr, nr_max);
939 sum += nr;
940 }
941 BUG_ON(nr_min > nr_max);
942
943 BUG_ON(sum > g->p.nr_tasks);
944
945 if (0 && (sum < g->p.nr_tasks))
946 return;
947
948 /*
949 * Count the number of distinct process groups present
950 * on nodes - when we are converged this will decrease
951 * to g->p.nr_proc:
952 */
953 process_groups = 0;
954
955 for (node = 0; node < g->p.nr_nodes; node++) {
956 int processes = count_node_processes(node);
957
958 nr = nodes[node];
959 tprintf(" %2d/%-2d", nr, processes);
960
961 process_groups += processes;
962 }
963
964 distance = nr_max - nr_min;
965
966 tprintf(" [%2d/%-2d]", distance, process_groups);
967
968 tprintf(" l:%3d-%-3d (%3d)",
969 loops_done_min, loops_done_max, loops_done_max-loops_done_min);
970
971 if (loops_done_min && loops_done_max) {
972 double skew = 1.0 - (double)loops_done_min/loops_done_max;
973
974 tprintf(" [%4.1f%%]", skew * 100.0);
975 }
976
977 calc_convergence_compression(&strong);
978
979 if (strong && process_groups == g->p.nr_proc) {
980 if (!*convergence) {
981 *convergence = runtime_ns_max;
982 tprintf(" (%6.1fs converged)\n", *convergence/1e9);
983 if (g->p.measure_convergence) {
984 g->all_converged = true;
985 g->stop_work = true;
986 }
987 }
988 } else {
989 if (*convergence) {
990 tprintf(" (%6.1fs de-converged)", runtime_ns_max/1e9);
991 *convergence = 0;
992 }
993 tprintf("\n");
994 }
995}
996
997static void show_summary(double runtime_ns_max, int l, double *convergence)
998{
999 tprintf("\r # %5.1f%% [%.1f mins]",
1000 (double)(l+1)/g->p.nr_loops*100.0, runtime_ns_max/1e9 / 60.0);
1001
1002 calc_convergence(runtime_ns_max, convergence);
1003
1004 if (g->p.show_details >= 0)
1005 fflush(stdout);
1006}
1007
1008static void *worker_thread(void *__tdata)
1009{
1010 struct thread_data *td = __tdata;
1011 struct timeval start0, start, stop, diff;
1012 int process_nr = td->process_nr;
1013 int thread_nr = td->thread_nr;
1014 unsigned long last_perturbance;
1015 int task_nr = td->task_nr;
1016 int details = g->p.show_details;
1017 int first_task, last_task;
1018 double convergence = 0;
1019 u64 val = td->val;
1020 double runtime_ns_max;
1021 u8 *global_data;
1022 u8 *process_data;
1023 u8 *thread_data;
1024 u64 bytes_done;
1025 long work_done;
1026 u32 l;
1027
1028 bind_to_cpumask(td->bind_cpumask);
1029 bind_to_memnode(td->bind_node);
1030
1031 set_taskname("thread %d/%d", process_nr, thread_nr);
1032
1033 global_data = g->data;
1034 process_data = td->process_data;
1035 thread_data = setup_private_data(g->p.bytes_thread);
1036
1037 bytes_done = 0;
1038
1039 last_task = 0;
1040 if (process_nr == g->p.nr_proc-1 && thread_nr == g->p.nr_threads-1)
1041 last_task = 1;
1042
1043 first_task = 0;
1044 if (process_nr == 0 && thread_nr == 0)
1045 first_task = 1;
1046
1047 if (details >= 2) {
1048 printf("# thread %2d / %2d global mem: %p, process mem: %p, thread mem: %p\n",
1049 process_nr, thread_nr, global_data, process_data, thread_data);
1050 }
1051
1052 if (g->p.serialize_startup) {
1053 pthread_mutex_lock(&g->startup_mutex);
1054 g->nr_tasks_started++;
1055 pthread_mutex_unlock(&g->startup_mutex);
1056
1057 /* Here we will wait for the main process to start us all at once: */
1058 pthread_mutex_lock(&g->start_work_mutex);
1059 g->nr_tasks_working++;
1060
1061 /* Last one wake the main process: */
1062 if (g->nr_tasks_working == g->p.nr_tasks)
1063 pthread_mutex_unlock(&g->startup_done_mutex);
1064
1065 pthread_mutex_unlock(&g->start_work_mutex);
1066 }
1067
1068 gettimeofday(&start0, NULL);
1069
1070 start = stop = start0;
1071 last_perturbance = start.tv_sec;
1072
1073 for (l = 0; l < g->p.nr_loops; l++) {
1074 start = stop;
1075
1076 if (g->stop_work)
1077 break;
1078
1079 val += do_work(global_data, g->p.bytes_global, process_nr, g->p.nr_proc, l, val);
1080 val += do_work(process_data, g->p.bytes_process, thread_nr, g->p.nr_threads, l, val);
1081 val += do_work(thread_data, g->p.bytes_thread, 0, 1, l, val);
1082
1083 if (g->p.sleep_usecs) {
1084 pthread_mutex_lock(td->process_lock);
1085 usleep(g->p.sleep_usecs);
1086 pthread_mutex_unlock(td->process_lock);
1087 }
1088 /*
1089 * Amount of work to be done under a process-global lock:
1090 */
1091 if (g->p.bytes_process_locked) {
1092 pthread_mutex_lock(td->process_lock);
1093 val += do_work(process_data, g->p.bytes_process_locked, thread_nr, g->p.nr_threads, l, val);
1094 pthread_mutex_unlock(td->process_lock);
1095 }
1096
1097 work_done = g->p.bytes_global + g->p.bytes_process +
1098 g->p.bytes_process_locked + g->p.bytes_thread;
1099
1100 update_curr_cpu(task_nr, work_done);
1101 bytes_done += work_done;
1102
1103 if (details < 0 && !g->p.perturb_secs && !g->p.measure_convergence && !g->p.nr_secs)
1104 continue;
1105
1106 td->loops_done = l;
1107
1108 gettimeofday(&stop, NULL);
1109
1110 /* Check whether our max runtime timed out: */
1111 if (g->p.nr_secs) {
1112 timersub(&stop, &start0, &diff);
1113 if (diff.tv_sec >= g->p.nr_secs) {
1114 g->stop_work = true;
1115 break;
1116 }
1117 }
1118
1119 /* Update the summary at most once per second: */
1120 if (start.tv_sec == stop.tv_sec)
1121 continue;
1122
1123 /*
1124 * Perturb the first task's equilibrium every g->p.perturb_secs seconds,
1125 * by migrating to CPU#0:
1126 */
1127 if (first_task && g->p.perturb_secs && (int)(stop.tv_sec - last_perturbance) >= g->p.perturb_secs) {
1128 cpu_set_t orig_mask;
1129 int target_cpu;
1130 int this_cpu;
1131
1132 last_perturbance = stop.tv_sec;
1133
1134 /*
1135 * Depending on where we are running, move into
1136 * the other half of the system, to create some
1137 * real disturbance:
1138 */
1139 this_cpu = g->threads[task_nr].curr_cpu;
1140 if (this_cpu < g->p.nr_cpus/2)
1141 target_cpu = g->p.nr_cpus-1;
1142 else
1143 target_cpu = 0;
1144
1145 orig_mask = bind_to_cpu(target_cpu);
1146
1147 /* Here we are running on the target CPU already */
1148 if (details >= 1)
1149 printf(" (injecting perturbalance, moved to CPU#%d)\n", target_cpu);
1150
1151 bind_to_cpumask(orig_mask);
1152 }
1153
1154 if (details >= 3) {
1155 timersub(&stop, &start, &diff);
1156 runtime_ns_max = diff.tv_sec * 1000000000;
1157 runtime_ns_max += diff.tv_usec * 1000;
1158
1159 if (details >= 0) {
1160 printf(" #%2d / %2d: %14.2lf nsecs/op [val: %016lx]\n",
1161 process_nr, thread_nr, runtime_ns_max / bytes_done, val);
1162 }
1163 fflush(stdout);
1164 }
1165 if (!last_task)
1166 continue;
1167
1168 timersub(&stop, &start0, &diff);
1169 runtime_ns_max = diff.tv_sec * 1000000000ULL;
1170 runtime_ns_max += diff.tv_usec * 1000ULL;
1171
1172 show_summary(runtime_ns_max, l, &convergence);
1173 }
1174
1175 gettimeofday(&stop, NULL);
1176 timersub(&stop, &start0, &diff);
1177 td->runtime_ns = diff.tv_sec * 1000000000ULL;
1178 td->runtime_ns += diff.tv_usec * 1000ULL;
1179
1180 free_data(thread_data, g->p.bytes_thread);
1181
1182 pthread_mutex_lock(&g->stop_work_mutex);
1183 g->bytes_done += bytes_done;
1184 pthread_mutex_unlock(&g->stop_work_mutex);
1185
1186 return NULL;
1187}
1188
1189/*
1190 * A worker process starts a couple of threads:
1191 */
1192static void worker_process(int process_nr)
1193{
1194 pthread_mutex_t process_lock;
1195 struct thread_data *td;
1196 pthread_t *pthreads;
1197 u8 *process_data;
1198 int task_nr;
1199 int ret;
1200 int t;
1201
1202 pthread_mutex_init(&process_lock, NULL);
1203 set_taskname("process %d", process_nr);
1204
1205 /*
1206 * Pick up the memory policy and the CPU binding of our first thread,
1207 * so that we initialize memory accordingly:
1208 */
1209 task_nr = process_nr*g->p.nr_threads;
1210 td = g->threads + task_nr;
1211
1212 bind_to_memnode(td->bind_node);
1213 bind_to_cpumask(td->bind_cpumask);
1214
1215 pthreads = zalloc(g->p.nr_threads * sizeof(pthread_t));
1216 process_data = setup_private_data(g->p.bytes_process);
1217
1218 if (g->p.show_details >= 3) {
1219 printf(" # process %2d global mem: %p, process mem: %p\n",
1220 process_nr, g->data, process_data);
1221 }
1222
1223 for (t = 0; t < g->p.nr_threads; t++) {
1224 task_nr = process_nr*g->p.nr_threads + t;
1225 td = g->threads + task_nr;
1226
1227 td->process_data = process_data;
1228 td->process_nr = process_nr;
1229 td->thread_nr = t;
1230 td->task_nr = task_nr;
1231 td->val = rand();
1232 td->curr_cpu = -1;
1233 td->process_lock = &process_lock;
1234
1235 ret = pthread_create(pthreads + t, NULL, worker_thread, td);
1236 BUG_ON(ret);
1237 }
1238
1239 for (t = 0; t < g->p.nr_threads; t++) {
1240 ret = pthread_join(pthreads[t], NULL);
1241 BUG_ON(ret);
1242 }
1243
1244 free_data(process_data, g->p.bytes_process);
1245 free(pthreads);
1246}
1247
1248static void print_summary(void)
1249{
1250 if (g->p.show_details < 0)
1251 return;
1252
1253 printf("\n ###\n");
1254 printf(" # %d %s will execute (on %d nodes, %d CPUs):\n",
1255 g->p.nr_tasks, g->p.nr_tasks == 1 ? "task" : "tasks", g->p.nr_nodes, g->p.nr_cpus);
1256 printf(" # %5dx %5ldMB global shared mem operations\n",
1257 g->p.nr_loops, g->p.bytes_global/1024/1024);
1258 printf(" # %5dx %5ldMB process shared mem operations\n",
1259 g->p.nr_loops, g->p.bytes_process/1024/1024);
1260 printf(" # %5dx %5ldMB thread local mem operations\n",
1261 g->p.nr_loops, g->p.bytes_thread/1024/1024);
1262
1263 printf(" ###\n");
1264
1265 printf("\n ###\n"); fflush(stdout);
1266}
1267
1268static void init_thread_data(void)
1269{
1270 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks;
1271 int t;
1272
1273 g->threads = zalloc_shared_data(size);
1274
1275 for (t = 0; t < g->p.nr_tasks; t++) {
1276 struct thread_data *td = g->threads + t;
1277 int cpu;
1278
1279 /* Allow all nodes by default: */
1280 td->bind_node = -1;
1281
1282 /* Allow all CPUs by default: */
1283 CPU_ZERO(&td->bind_cpumask);
1284 for (cpu = 0; cpu < g->p.nr_cpus; cpu++)
1285 CPU_SET(cpu, &td->bind_cpumask);
1286 }
1287}
1288
1289static void deinit_thread_data(void)
1290{
1291 ssize_t size = sizeof(*g->threads)*g->p.nr_tasks;
1292
1293 free_data(g->threads, size);
1294}
1295
1296static int init(void)
1297{
1298 g = (void *)alloc_data(sizeof(*g), MAP_SHARED, 1, 0, 0 /* THP */, 0);
1299
1300 /* Copy over options: */
1301 g->p = p0;
1302
1303 g->p.nr_cpus = numa_num_configured_cpus();
1304
1305 g->p.nr_nodes = numa_max_node() + 1;
1306
1307 /* char array in count_process_nodes(): */
1308 BUG_ON(g->p.nr_nodes > MAX_NR_NODES || g->p.nr_nodes < 0);
1309
1310 if (g->p.show_quiet && !g->p.show_details)
1311 g->p.show_details = -1;
1312
1313 /* Some memory should be specified: */
1314 if (!g->p.mb_global_str && !g->p.mb_proc_str && !g->p.mb_thread_str)
1315 return -1;
1316
1317 if (g->p.mb_global_str) {
1318 g->p.mb_global = atof(g->p.mb_global_str);
1319 BUG_ON(g->p.mb_global < 0);
1320 }
1321
1322 if (g->p.mb_proc_str) {
1323 g->p.mb_proc = atof(g->p.mb_proc_str);
1324 BUG_ON(g->p.mb_proc < 0);
1325 }
1326
1327 if (g->p.mb_proc_locked_str) {
1328 g->p.mb_proc_locked = atof(g->p.mb_proc_locked_str);
1329 BUG_ON(g->p.mb_proc_locked < 0);
1330 BUG_ON(g->p.mb_proc_locked > g->p.mb_proc);
1331 }
1332
1333 if (g->p.mb_thread_str) {
1334 g->p.mb_thread = atof(g->p.mb_thread_str);
1335 BUG_ON(g->p.mb_thread < 0);
1336 }
1337
1338 BUG_ON(g->p.nr_threads <= 0);
1339 BUG_ON(g->p.nr_proc <= 0);
1340
1341 g->p.nr_tasks = g->p.nr_proc*g->p.nr_threads;
1342
1343 g->p.bytes_global = g->p.mb_global *1024L*1024L;
1344 g->p.bytes_process = g->p.mb_proc *1024L*1024L;
1345 g->p.bytes_process_locked = g->p.mb_proc_locked *1024L*1024L;
1346 g->p.bytes_thread = g->p.mb_thread *1024L*1024L;
1347
1348 g->data = setup_shared_data(g->p.bytes_global);
1349
1350 /* Startup serialization: */
1351 init_global_mutex(&g->start_work_mutex);
1352 init_global_mutex(&g->startup_mutex);
1353 init_global_mutex(&g->startup_done_mutex);
1354 init_global_mutex(&g->stop_work_mutex);
1355
1356 init_thread_data();
1357
1358 tprintf("#\n");
1359 parse_setup_cpu_list();
1360 parse_setup_node_list();
1361 tprintf("#\n");
1362
1363 print_summary();
1364
1365 return 0;
1366}
1367
1368static void deinit(void)
1369{
1370 free_data(g->data, g->p.bytes_global);
1371 g->data = NULL;
1372
1373 deinit_thread_data();
1374
1375 free_data(g, sizeof(*g));
1376 g = NULL;
1377}
1378
1379/*
1380 * Print a short or long result, depending on the verbosity setting:
1381 */
1382static void print_res(const char *name, double val,
1383 const char *txt_unit, const char *txt_short, const char *txt_long)
1384{
1385 if (!name)
1386 name = "main,";
1387
1388 if (g->p.show_quiet)
1389 printf(" %-30s %15.3f, %-15s %s\n", name, val, txt_unit, txt_short);
1390 else
1391 printf(" %14.3f %s\n", val, txt_long);
1392}
1393
1394static int __bench_numa(const char *name)
1395{
1396 struct timeval start, stop, diff;
1397 u64 runtime_ns_min, runtime_ns_sum;
1398 pid_t *pids, pid, wpid;
1399 double delta_runtime;
1400 double runtime_avg;
1401 double runtime_sec_max;
1402 double runtime_sec_min;
1403 int wait_stat;
1404 double bytes;
1405 int i, t;
1406
1407 if (init())
1408 return -1;
1409
1410 pids = zalloc(g->p.nr_proc * sizeof(*pids));
1411 pid = -1;
1412
1413 /* All threads try to acquire it, this way we can wait for them to start up: */
1414 pthread_mutex_lock(&g->start_work_mutex);
1415
1416 if (g->p.serialize_startup) {
1417 tprintf(" #\n");
1418 tprintf(" # Startup synchronization: ..."); fflush(stdout);
1419 }
1420
1421 gettimeofday(&start, NULL);
1422
1423 for (i = 0; i < g->p.nr_proc; i++) {
1424 pid = fork();
1425 dprintf(" # process %2d: PID %d\n", i, pid);
1426
1427 BUG_ON(pid < 0);
1428 if (!pid) {
1429 /* Child process: */
1430 worker_process(i);
1431
1432 exit(0);
1433 }
1434 pids[i] = pid;
1435
1436 }
1437 /* Wait for all the threads to start up: */
1438 while (g->nr_tasks_started != g->p.nr_tasks)
1439 usleep(1000);
1440
1441 BUG_ON(g->nr_tasks_started != g->p.nr_tasks);
1442
1443 if (g->p.serialize_startup) {
1444 double startup_sec;
1445
1446 pthread_mutex_lock(&g->startup_done_mutex);
1447
1448 /* This will start all threads: */
1449 pthread_mutex_unlock(&g->start_work_mutex);
1450
1451 /* This mutex is locked - the last started thread will wake us: */
1452 pthread_mutex_lock(&g->startup_done_mutex);
1453
1454 gettimeofday(&stop, NULL);
1455
1456 timersub(&stop, &start, &diff);
1457
1458 startup_sec = diff.tv_sec * 1000000000.0;
1459 startup_sec += diff.tv_usec * 1000.0;
1460 startup_sec /= 1e9;
1461
1462 tprintf(" threads initialized in %.6f seconds.\n", startup_sec);
1463 tprintf(" #\n");
1464
1465 start = stop;
1466 pthread_mutex_unlock(&g->startup_done_mutex);
1467 } else {
1468 gettimeofday(&start, NULL);
1469 }
1470
1471 /* Parent process: */
1472
1473
1474 for (i = 0; i < g->p.nr_proc; i++) {
1475 wpid = waitpid(pids[i], &wait_stat, 0);
1476 BUG_ON(wpid < 0);
1477 BUG_ON(!WIFEXITED(wait_stat));
1478
1479 }
1480
1481 runtime_ns_sum = 0;
1482 runtime_ns_min = -1LL;
1483
1484 for (t = 0; t < g->p.nr_tasks; t++) {
1485 u64 thread_runtime_ns = g->threads[t].runtime_ns;
1486
1487 runtime_ns_sum += thread_runtime_ns;
1488 runtime_ns_min = min(thread_runtime_ns, runtime_ns_min);
1489 }
1490
1491 gettimeofday(&stop, NULL);
1492 timersub(&stop, &start, &diff);
1493
1494 BUG_ON(bench_format != BENCH_FORMAT_DEFAULT);
1495
1496 tprintf("\n ###\n");
1497 tprintf("\n");
1498
1499 runtime_sec_max = diff.tv_sec * 1000000000.0;
1500 runtime_sec_max += diff.tv_usec * 1000.0;
1501 runtime_sec_max /= 1e9;
1502
1503 runtime_sec_min = runtime_ns_min/1e9;
1504
1505 bytes = g->bytes_done;
1506 runtime_avg = (double)runtime_ns_sum / g->p.nr_tasks / 1e9;
1507
1508 if (g->p.measure_convergence) {
1509 print_res(name, runtime_sec_max,
1510 "secs,", "NUMA-convergence-latency", "secs latency to NUMA-converge");
1511 }
1512
1513 print_res(name, runtime_sec_max,
1514 "secs,", "runtime-max/thread", "secs slowest (max) thread-runtime");
1515
1516 print_res(name, runtime_sec_min,
1517 "secs,", "runtime-min/thread", "secs fastest (min) thread-runtime");
1518
1519 print_res(name, runtime_avg,
1520 "secs,", "runtime-avg/thread", "secs average thread-runtime");
1521
1522 delta_runtime = (runtime_sec_max - runtime_sec_min)/2.0;
1523 print_res(name, delta_runtime / runtime_sec_max * 100.0,
1524 "%,", "spread-runtime/thread", "% difference between max/avg runtime");
1525
1526 print_res(name, bytes / g->p.nr_tasks / 1e9,
1527 "GB,", "data/thread", "GB data processed, per thread");
1528
1529 print_res(name, bytes / 1e9,
1530 "GB,", "data-total", "GB data processed, total");
1531
1532 print_res(name, runtime_sec_max * 1e9 / (bytes / g->p.nr_tasks),
1533 "nsecs,", "runtime/byte/thread","nsecs/byte/thread runtime");
1534
1535 print_res(name, bytes / g->p.nr_tasks / 1e9 / runtime_sec_max,
1536 "GB/sec,", "thread-speed", "GB/sec/thread speed");
1537
1538 print_res(name, bytes / runtime_sec_max / 1e9,
1539 "GB/sec,", "total-speed", "GB/sec total speed");
1540
1541 free(pids);
1542
1543 deinit();
1544
1545 return 0;
1546}
1547
1548#define MAX_ARGS 50
1549
1550static int command_size(const char **argv)
1551{
1552 int size = 0;
1553
1554 while (*argv) {
1555 size++;
1556 argv++;
1557 }
1558
1559 BUG_ON(size >= MAX_ARGS);
1560
1561 return size;
1562}
1563
1564static void init_params(struct params *p, const char *name, int argc, const char **argv)
1565{
1566 int i;
1567
1568 printf("\n # Running %s \"perf bench numa", name);
1569
1570 for (i = 0; i < argc; i++)
1571 printf(" %s", argv[i]);
1572
1573 printf("\"\n");
1574
1575 memset(p, 0, sizeof(*p));
1576
1577 /* Initialize nonzero defaults: */
1578
1579 p->serialize_startup = 1;
1580 p->data_reads = true;
1581 p->data_writes = true;
1582 p->data_backwards = true;
1583 p->data_rand_walk = true;
1584 p->nr_loops = -1;
1585 p->init_random = true;
1586}
1587
1588static int run_bench_numa(const char *name, const char **argv)
1589{
1590 int argc = command_size(argv);
1591
1592 init_params(&p0, name, argc, argv);
1593 argc = parse_options(argc, argv, options, bench_numa_usage, 0);
1594 if (argc)
1595 goto err;
1596
1597 if (__bench_numa(name))
1598 goto err;
1599
1600 return 0;
1601
1602err:
1603 usage_with_options(numa_usage, options);
1604 return -1;
1605}
1606
1607#define OPT_BW_RAM "-s", "20", "-zZq", "--thp", " 1", "--no-data_rand_walk"
1608#define OPT_BW_RAM_NOTHP OPT_BW_RAM, "--thp", "-1"
1609
1610#define OPT_CONV "-s", "100", "-zZ0qcm", "--thp", " 1"
1611#define OPT_CONV_NOTHP OPT_CONV, "--thp", "-1"
1612
1613#define OPT_BW "-s", "20", "-zZ0q", "--thp", " 1"
1614#define OPT_BW_NOTHP OPT_BW, "--thp", "-1"
1615
1616/*
1617 * The built-in test-suite executed by "perf bench numa -a".
1618 *
1619 * (A minimum of 4 nodes and 16 GB of RAM is recommended.)
1620 */
1621static const char *tests[][MAX_ARGS] = {
1622 /* Basic single-stream NUMA bandwidth measurements: */
1623 { "RAM-bw-local,", "mem", "-p", "1", "-t", "1", "-P", "1024",
1624 "-C" , "0", "-M", "0", OPT_BW_RAM },
1625 { "RAM-bw-local-NOTHP,",
1626 "mem", "-p", "1", "-t", "1", "-P", "1024",
1627 "-C" , "0", "-M", "0", OPT_BW_RAM_NOTHP },
1628 { "RAM-bw-remote,", "mem", "-p", "1", "-t", "1", "-P", "1024",
1629 "-C" , "0", "-M", "1", OPT_BW_RAM },
1630
1631 /* 2-stream NUMA bandwidth measurements: */
1632 { "RAM-bw-local-2x,", "mem", "-p", "2", "-t", "1", "-P", "1024",
1633 "-C", "0,2", "-M", "0x2", OPT_BW_RAM },
1634 { "RAM-bw-remote-2x,", "mem", "-p", "2", "-t", "1", "-P", "1024",
1635 "-C", "0,2", "-M", "1x2", OPT_BW_RAM },
1636
1637 /* Cross-stream NUMA bandwidth measurement: */
1638 { "RAM-bw-cross,", "mem", "-p", "2", "-t", "1", "-P", "1024",
1639 "-C", "0,8", "-M", "1,0", OPT_BW_RAM },
1640
1641 /* Convergence latency measurements: */
1642 { " 1x3-convergence,", "mem", "-p", "1", "-t", "3", "-P", "512", OPT_CONV },
1643 { " 1x4-convergence,", "mem", "-p", "1", "-t", "4", "-P", "512", OPT_CONV },
1644 { " 1x6-convergence,", "mem", "-p", "1", "-t", "6", "-P", "1020", OPT_CONV },
1645 { " 2x3-convergence,", "mem", "-p", "3", "-t", "3", "-P", "1020", OPT_CONV },
1646 { " 3x3-convergence,", "mem", "-p", "3", "-t", "3", "-P", "1020", OPT_CONV },
1647 { " 4x4-convergence,", "mem", "-p", "4", "-t", "4", "-P", "512", OPT_CONV },
1648 { " 4x4-convergence-NOTHP,",
1649 "mem", "-p", "4", "-t", "4", "-P", "512", OPT_CONV_NOTHP },
1650 { " 4x6-convergence,", "mem", "-p", "4", "-t", "6", "-P", "1020", OPT_CONV },
1651 { " 4x8-convergence,", "mem", "-p", "4", "-t", "8", "-P", "512", OPT_CONV },
1652 { " 8x4-convergence,", "mem", "-p", "8", "-t", "4", "-P", "512", OPT_CONV },
1653 { " 8x4-convergence-NOTHP,",
1654 "mem", "-p", "8", "-t", "4", "-P", "512", OPT_CONV_NOTHP },
1655 { " 3x1-convergence,", "mem", "-p", "3", "-t", "1", "-P", "512", OPT_CONV },
1656 { " 4x1-convergence,", "mem", "-p", "4", "-t", "1", "-P", "512", OPT_CONV },
1657 { " 8x1-convergence,", "mem", "-p", "8", "-t", "1", "-P", "512", OPT_CONV },
1658 { "16x1-convergence,", "mem", "-p", "16", "-t", "1", "-P", "256", OPT_CONV },
1659 { "32x1-convergence,", "mem", "-p", "32", "-t", "1", "-P", "128", OPT_CONV },
1660
1661 /* Various NUMA process/thread layout bandwidth measurements: */
1662 { " 2x1-bw-process,", "mem", "-p", "2", "-t", "1", "-P", "1024", OPT_BW },
1663 { " 3x1-bw-process,", "mem", "-p", "3", "-t", "1", "-P", "1024", OPT_BW },
1664 { " 4x1-bw-process,", "mem", "-p", "4", "-t", "1", "-P", "1024", OPT_BW },
1665 { " 8x1-bw-process,", "mem", "-p", "8", "-t", "1", "-P", " 512", OPT_BW },
1666 { " 8x1-bw-process-NOTHP,",
1667 "mem", "-p", "8", "-t", "1", "-P", " 512", OPT_BW_NOTHP },
1668 { "16x1-bw-process,", "mem", "-p", "16", "-t", "1", "-P", "256", OPT_BW },
1669
1670 { " 4x1-bw-thread,", "mem", "-p", "1", "-t", "4", "-T", "256", OPT_BW },
1671 { " 8x1-bw-thread,", "mem", "-p", "1", "-t", "8", "-T", "256", OPT_BW },
1672 { "16x1-bw-thread,", "mem", "-p", "1", "-t", "16", "-T", "128", OPT_BW },
1673 { "32x1-bw-thread,", "mem", "-p", "1", "-t", "32", "-T", "64", OPT_BW },
1674
1675 { " 2x3-bw-thread,", "mem", "-p", "2", "-t", "3", "-P", "512", OPT_BW },
1676 { " 4x4-bw-thread,", "mem", "-p", "4", "-t", "4", "-P", "512", OPT_BW },
1677 { " 4x6-bw-thread,", "mem", "-p", "4", "-t", "6", "-P", "512", OPT_BW },
1678 { " 4x8-bw-thread,", "mem", "-p", "4", "-t", "8", "-P", "512", OPT_BW },
1679 { " 4x8-bw-thread-NOTHP,",
1680 "mem", "-p", "4", "-t", "8", "-P", "512", OPT_BW_NOTHP },
1681 { " 3x3-bw-thread,", "mem", "-p", "3", "-t", "3", "-P", "512", OPT_BW },
1682 { " 5x5-bw-thread,", "mem", "-p", "5", "-t", "5", "-P", "512", OPT_BW },
1683
1684 { "2x16-bw-thread,", "mem", "-p", "2", "-t", "16", "-P", "512", OPT_BW },
1685 { "1x32-bw-thread,", "mem", "-p", "1", "-t", "32", "-P", "2048", OPT_BW },
1686
1687 { "numa02-bw,", "mem", "-p", "1", "-t", "32", "-T", "32", OPT_BW },
1688 { "numa02-bw-NOTHP,", "mem", "-p", "1", "-t", "32", "-T", "32", OPT_BW_NOTHP },
1689 { "numa01-bw-thread,", "mem", "-p", "2", "-t", "16", "-T", "192", OPT_BW },
1690 { "numa01-bw-thread-NOTHP,",
1691 "mem", "-p", "2", "-t", "16", "-T", "192", OPT_BW_NOTHP },
1692};
1693
1694static int bench_all(void)
1695{
1696 int nr = ARRAY_SIZE(tests);
1697 int ret;
1698 int i;
1699
1700 ret = system("echo ' #'; echo ' # Running test on: '$(uname -a); echo ' #'");
1701 BUG_ON(ret < 0);
1702
1703 for (i = 0; i < nr; i++) {
1704 if (run_bench_numa(tests[i][0], tests[i] + 1))
1705 return -1;
1706 }
1707
1708 printf("\n");
1709
1710 return 0;
1711}
1712
1713int bench_numa(int argc, const char **argv, const char *prefix __maybe_unused)
1714{
1715 init_params(&p0, "main,", argc, argv);
1716 argc = parse_options(argc, argv, options, bench_numa_usage, 0);
1717 if (argc)
1718 goto err;
1719
1720 if (p0.run_all)
1721 return bench_all();
1722
1723 if (__bench_numa(NULL))
1724 goto err;
1725
1726 return 0;
1727
1728err:
1729 usage_with_options(numa_usage, options);
1730 return -1;
1731}
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index afd1255a632f..77298bf892b8 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -35,6 +35,18 @@ struct bench_suite {
35/* sentinel: easy for help */ 35/* sentinel: easy for help */
36#define suite_all { "all", "Test all benchmark suites", NULL } 36#define suite_all { "all", "Test all benchmark suites", NULL }
37 37
38#ifdef LIBNUMA_SUPPORT
39static struct bench_suite numa_suites[] = {
40 { "mem",
41 "Benchmark for NUMA workloads",
42 bench_numa },
43 suite_all,
44 { NULL,
45 NULL,
46 NULL }
47};
48#endif
49
38static struct bench_suite sched_suites[] = { 50static struct bench_suite sched_suites[] = {
39 { "messaging", 51 { "messaging",
40 "Benchmark for scheduler and IPC mechanisms", 52 "Benchmark for scheduler and IPC mechanisms",
@@ -68,6 +80,11 @@ struct bench_subsys {
68}; 80};
69 81
70static struct bench_subsys subsystems[] = { 82static struct bench_subsys subsystems[] = {
83#ifdef LIBNUMA_SUPPORT
84 { "numa",
85 "NUMA scheduling and MM behavior",
86 numa_suites },
87#endif
71 { "sched", 88 { "sched",
72 "scheduler and IPC mechanism", 89 "scheduler and IPC mechanism",
73 sched_suites }, 90 sched_suites },
diff --git a/tools/perf/builtin-kmem.c b/tools/perf/builtin-kmem.c
index c746108c5d48..46878daca5cc 100644
--- a/tools/perf/builtin-kmem.c
+++ b/tools/perf/builtin-kmem.c
@@ -17,6 +17,7 @@
17#include "util/debug.h" 17#include "util/debug.h"
18 18
19#include <linux/rbtree.h> 19#include <linux/rbtree.h>
20#include <linux/string.h>
20 21
21struct alloc_stat; 22struct alloc_stat;
22typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *); 23typedef int (*sort_fn_t)(struct alloc_stat *, struct alloc_stat *);
@@ -618,12 +619,11 @@ static int sort_dimension__add(const char *tok, struct list_head *list)
618 619
619 for (i = 0; i < NUM_AVAIL_SORTS; i++) { 620 for (i = 0; i < NUM_AVAIL_SORTS; i++) {
620 if (!strcmp(avail_sorts[i]->name, tok)) { 621 if (!strcmp(avail_sorts[i]->name, tok)) {
621 sort = malloc(sizeof(*sort)); 622 sort = memdup(avail_sorts[i], sizeof(*avail_sorts[i]));
622 if (!sort) { 623 if (!sort) {
623 pr_err("%s: malloc failed\n", __func__); 624 pr_err("%s: memdup failed\n", __func__);
624 return -1; 625 return -1;
625 } 626 }
626 memcpy(sort, avail_sorts[i], sizeof(*sort));
627 list_add_tail(&sort->list, list); 627 list_add_tail(&sort->list, list);
628 return 0; 628 return 0;
629 } 629 }
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 1c2ac148a7d5..0368a1036ad6 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -65,6 +65,10 @@
65#define CNTR_NOT_SUPPORTED "<not supported>" 65#define CNTR_NOT_SUPPORTED "<not supported>"
66#define CNTR_NOT_COUNTED "<not counted>" 66#define CNTR_NOT_COUNTED "<not counted>"
67 67
68static void print_stat(int argc, const char **argv);
69static void print_counter_aggr(struct perf_evsel *counter, char *prefix);
70static void print_counter(struct perf_evsel *counter, char *prefix);
71
68static struct perf_evlist *evsel_list; 72static struct perf_evlist *evsel_list;
69 73
70static struct perf_target target = { 74static struct perf_target target = {
@@ -87,6 +91,8 @@ static FILE *output = NULL;
87static const char *pre_cmd = NULL; 91static const char *pre_cmd = NULL;
88static const char *post_cmd = NULL; 92static const char *post_cmd = NULL;
89static bool sync_run = false; 93static bool sync_run = false;
94static unsigned int interval = 0;
95static struct timespec ref_time;
90 96
91static volatile int done = 0; 97static volatile int done = 0;
92 98
@@ -94,6 +100,28 @@ struct perf_stat {
94 struct stats res_stats[3]; 100 struct stats res_stats[3];
95}; 101};
96 102
103static inline void diff_timespec(struct timespec *r, struct timespec *a,
104 struct timespec *b)
105{
106 r->tv_sec = a->tv_sec - b->tv_sec;
107 if (a->tv_nsec < b->tv_nsec) {
108 r->tv_nsec = a->tv_nsec + 1000000000L - b->tv_nsec;
109 r->tv_sec--;
110 } else {
111 r->tv_nsec = a->tv_nsec - b->tv_nsec ;
112 }
113}
114
115static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel)
116{
117 return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus;
118}
119
120static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel)
121{
122 return perf_evsel__cpus(evsel)->nr;
123}
124
97static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel) 125static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel)
98{ 126{
99 evsel->priv = zalloc(sizeof(struct perf_stat)); 127 evsel->priv = zalloc(sizeof(struct perf_stat));
@@ -106,14 +134,27 @@ static void perf_evsel__free_stat_priv(struct perf_evsel *evsel)
106 evsel->priv = NULL; 134 evsel->priv = NULL;
107} 135}
108 136
109static inline struct cpu_map *perf_evsel__cpus(struct perf_evsel *evsel) 137static int perf_evsel__alloc_prev_raw_counts(struct perf_evsel *evsel)
110{ 138{
111 return (evsel->cpus && !target.cpu_list) ? evsel->cpus : evsel_list->cpus; 139 void *addr;
140 size_t sz;
141
142 sz = sizeof(*evsel->counts) +
143 (perf_evsel__nr_cpus(evsel) * sizeof(struct perf_counts_values));
144
145 addr = zalloc(sz);
146 if (!addr)
147 return -ENOMEM;
148
149 evsel->prev_raw_counts = addr;
150
151 return 0;
112} 152}
113 153
114static inline int perf_evsel__nr_cpus(struct perf_evsel *evsel) 154static void perf_evsel__free_prev_raw_counts(struct perf_evsel *evsel)
115{ 155{
116 return perf_evsel__cpus(evsel)->nr; 156 free(evsel->prev_raw_counts);
157 evsel->prev_raw_counts = NULL;
117} 158}
118 159
119static struct stats runtime_nsecs_stats[MAX_NR_CPUS]; 160static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
@@ -245,16 +286,69 @@ static int read_counter(struct perf_evsel *counter)
245 return 0; 286 return 0;
246} 287}
247 288
289static void print_interval(void)
290{
291 static int num_print_interval;
292 struct perf_evsel *counter;
293 struct perf_stat *ps;
294 struct timespec ts, rs;
295 char prefix[64];
296
297 if (no_aggr) {
298 list_for_each_entry(counter, &evsel_list->entries, node) {
299 ps = counter->priv;
300 memset(ps->res_stats, 0, sizeof(ps->res_stats));
301 read_counter(counter);
302 }
303 } else {
304 list_for_each_entry(counter, &evsel_list->entries, node) {
305 ps = counter->priv;
306 memset(ps->res_stats, 0, sizeof(ps->res_stats));
307 read_counter_aggr(counter);
308 }
309 }
310 clock_gettime(CLOCK_MONOTONIC, &ts);
311 diff_timespec(&rs, &ts, &ref_time);
312 sprintf(prefix, "%6lu.%09lu%s", rs.tv_sec, rs.tv_nsec, csv_sep);
313
314 if (num_print_interval == 0 && !csv_output) {
315 if (no_aggr)
316 fprintf(output, "# time CPU counts events\n");
317 else
318 fprintf(output, "# time counts events\n");
319 }
320
321 if (++num_print_interval == 25)
322 num_print_interval = 0;
323
324 if (no_aggr) {
325 list_for_each_entry(counter, &evsel_list->entries, node)
326 print_counter(counter, prefix);
327 } else {
328 list_for_each_entry(counter, &evsel_list->entries, node)
329 print_counter_aggr(counter, prefix);
330 }
331}
332
248static int __run_perf_stat(int argc __maybe_unused, const char **argv) 333static int __run_perf_stat(int argc __maybe_unused, const char **argv)
249{ 334{
250 char msg[512]; 335 char msg[512];
251 unsigned long long t0, t1; 336 unsigned long long t0, t1;
252 struct perf_evsel *counter; 337 struct perf_evsel *counter;
338 struct timespec ts;
253 int status = 0; 339 int status = 0;
254 int child_ready_pipe[2], go_pipe[2]; 340 int child_ready_pipe[2], go_pipe[2];
255 const bool forks = (argc > 0); 341 const bool forks = (argc > 0);
256 char buf; 342 char buf;
257 343
344 if (interval) {
345 ts.tv_sec = interval / 1000;
346 ts.tv_nsec = (interval % 1000) * 1000000;
347 } else {
348 ts.tv_sec = 1;
349 ts.tv_nsec = 0;
350 }
351
258 if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { 352 if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
259 perror("failed to create pipes"); 353 perror("failed to create pipes");
260 return -1; 354 return -1;
@@ -347,14 +441,25 @@ static int __run_perf_stat(int argc __maybe_unused, const char **argv)
347 * Enable counters and exec the command: 441 * Enable counters and exec the command:
348 */ 442 */
349 t0 = rdclock(); 443 t0 = rdclock();
444 clock_gettime(CLOCK_MONOTONIC, &ref_time);
350 445
351 if (forks) { 446 if (forks) {
352 close(go_pipe[1]); 447 close(go_pipe[1]);
448 if (interval) {
449 while (!waitpid(child_pid, &status, WNOHANG)) {
450 nanosleep(&ts, NULL);
451 print_interval();
452 }
453 }
353 wait(&status); 454 wait(&status);
354 if (WIFSIGNALED(status)) 455 if (WIFSIGNALED(status))
355 psignal(WTERMSIG(status), argv[0]); 456 psignal(WTERMSIG(status), argv[0]);
356 } else { 457 } else {
357 while(!done) sleep(1); 458 while (!done) {
459 nanosleep(&ts, NULL);
460 if (interval)
461 print_interval();
462 }
358 } 463 }
359 464
360 t1 = rdclock(); 465 t1 = rdclock();
@@ -440,7 +545,7 @@ static void nsec_printout(int cpu, struct perf_evsel *evsel, double avg)
440 if (evsel->cgrp) 545 if (evsel->cgrp)
441 fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); 546 fprintf(output, "%s%s", csv_sep, evsel->cgrp->name);
442 547
443 if (csv_output) 548 if (csv_output || interval)
444 return; 549 return;
445 550
446 if (perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK)) 551 if (perf_evsel__match(evsel, SOFTWARE, SW_TASK_CLOCK))
@@ -654,12 +759,11 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg)
654 if (evsel->cgrp) 759 if (evsel->cgrp)
655 fprintf(output, "%s%s", csv_sep, evsel->cgrp->name); 760 fprintf(output, "%s%s", csv_sep, evsel->cgrp->name);
656 761
657 if (csv_output) 762 if (csv_output || interval)
658 return; 763 return;
659 764
660 if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) { 765 if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) {
661 total = avg_stats(&runtime_cycles_stats[cpu]); 766 total = avg_stats(&runtime_cycles_stats[cpu]);
662
663 if (total) 767 if (total)
664 ratio = avg / total; 768 ratio = avg / total;
665 769
@@ -753,12 +857,15 @@ static void abs_printout(int cpu, struct perf_evsel *evsel, double avg)
753 * Print out the results of a single counter: 857 * Print out the results of a single counter:
754 * aggregated counts in system-wide mode 858 * aggregated counts in system-wide mode
755 */ 859 */
756static void print_counter_aggr(struct perf_evsel *counter) 860static void print_counter_aggr(struct perf_evsel *counter, char *prefix)
757{ 861{
758 struct perf_stat *ps = counter->priv; 862 struct perf_stat *ps = counter->priv;
759 double avg = avg_stats(&ps->res_stats[0]); 863 double avg = avg_stats(&ps->res_stats[0]);
760 int scaled = counter->counts->scaled; 864 int scaled = counter->counts->scaled;
761 865
866 if (prefix)
867 fprintf(output, "%s", prefix);
868
762 if (scaled == -1) { 869 if (scaled == -1) {
763 fprintf(output, "%*s%s%*s", 870 fprintf(output, "%*s%s%*s",
764 csv_output ? 0 : 18, 871 csv_output ? 0 : 18,
@@ -801,7 +908,7 @@ static void print_counter_aggr(struct perf_evsel *counter)
801 * Print out the results of a single counter: 908 * Print out the results of a single counter:
802 * does not use aggregated count in system-wide 909 * does not use aggregated count in system-wide
803 */ 910 */
804static void print_counter(struct perf_evsel *counter) 911static void print_counter(struct perf_evsel *counter, char *prefix)
805{ 912{
806 u64 ena, run, val; 913 u64 ena, run, val;
807 int cpu; 914 int cpu;
@@ -810,6 +917,10 @@ static void print_counter(struct perf_evsel *counter)
810 val = counter->counts->cpu[cpu].val; 917 val = counter->counts->cpu[cpu].val;
811 ena = counter->counts->cpu[cpu].ena; 918 ena = counter->counts->cpu[cpu].ena;
812 run = counter->counts->cpu[cpu].run; 919 run = counter->counts->cpu[cpu].run;
920
921 if (prefix)
922 fprintf(output, "%s", prefix);
923
813 if (run == 0 || ena == 0) { 924 if (run == 0 || ena == 0) {
814 fprintf(output, "CPU%*d%s%*s%s%*s", 925 fprintf(output, "CPU%*d%s%*s%s%*s",
815 csv_output ? 0 : -4, 926 csv_output ? 0 : -4,
@@ -871,10 +982,10 @@ static void print_stat(int argc, const char **argv)
871 982
872 if (no_aggr) { 983 if (no_aggr) {
873 list_for_each_entry(counter, &evsel_list->entries, node) 984 list_for_each_entry(counter, &evsel_list->entries, node)
874 print_counter(counter); 985 print_counter(counter, NULL);
875 } else { 986 } else {
876 list_for_each_entry(counter, &evsel_list->entries, node) 987 list_for_each_entry(counter, &evsel_list->entries, node)
877 print_counter_aggr(counter); 988 print_counter_aggr(counter, NULL);
878 } 989 }
879 990
880 if (!csv_output) { 991 if (!csv_output) {
@@ -895,7 +1006,7 @@ static volatile int signr = -1;
895 1006
896static void skip_signal(int signo) 1007static void skip_signal(int signo)
897{ 1008{
898 if(child_pid == -1) 1009 if ((child_pid == -1) || interval)
899 done = 1; 1010 done = 1;
900 1011
901 signr = signo; 1012 signr = signo;
@@ -1115,6 +1226,8 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
1115 "command to run prior to the measured command"), 1226 "command to run prior to the measured command"),
1116 OPT_STRING(0, "post", &post_cmd, "command", 1227 OPT_STRING(0, "post", &post_cmd, "command",
1117 "command to run after to the measured command"), 1228 "command to run after to the measured command"),
1229 OPT_UINTEGER('I', "interval-print", &interval,
1230 "print counts at regular interval in ms (>= 100)"),
1118 OPT_END() 1231 OPT_END()
1119 }; 1232 };
1120 const char * const stat_usage[] = { 1233 const char * const stat_usage[] = {
@@ -1215,12 +1328,23 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
1215 usage_with_options(stat_usage, options); 1328 usage_with_options(stat_usage, options);
1216 return -1; 1329 return -1;
1217 } 1330 }
1331 if (interval && interval < 100) {
1332 pr_err("print interval must be >= 100ms\n");
1333 usage_with_options(stat_usage, options);
1334 return -1;
1335 }
1218 1336
1219 list_for_each_entry(pos, &evsel_list->entries, node) { 1337 list_for_each_entry(pos, &evsel_list->entries, node) {
1220 if (perf_evsel__alloc_stat_priv(pos) < 0 || 1338 if (perf_evsel__alloc_stat_priv(pos) < 0 ||
1221 perf_evsel__alloc_counts(pos, perf_evsel__nr_cpus(pos)) < 0) 1339 perf_evsel__alloc_counts(pos, perf_evsel__nr_cpus(pos)) < 0)
1222 goto out_free_fd; 1340 goto out_free_fd;
1223 } 1341 }
1342 if (interval) {
1343 list_for_each_entry(pos, &evsel_list->entries, node) {
1344 if (perf_evsel__alloc_prev_raw_counts(pos) < 0)
1345 goto out_free_fd;
1346 }
1347 }
1224 1348
1225 /* 1349 /*
1226 * We dont want to block the signals - that would cause 1350 * We dont want to block the signals - that would cause
@@ -1230,6 +1354,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
1230 */ 1354 */
1231 atexit(sig_atexit); 1355 atexit(sig_atexit);
1232 signal(SIGINT, skip_signal); 1356 signal(SIGINT, skip_signal);
1357 signal(SIGCHLD, skip_signal);
1233 signal(SIGALRM, skip_signal); 1358 signal(SIGALRM, skip_signal);
1234 signal(SIGABRT, skip_signal); 1359 signal(SIGABRT, skip_signal);
1235 1360
@@ -1242,11 +1367,14 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
1242 status = run_perf_stat(argc, argv); 1367 status = run_perf_stat(argc, argv);
1243 } 1368 }
1244 1369
1245 if (status != -1) 1370 if (status != -1 && !interval)
1246 print_stat(argc, argv); 1371 print_stat(argc, argv);
1247out_free_fd: 1372out_free_fd:
1248 list_for_each_entry(pos, &evsel_list->entries, node) 1373 list_for_each_entry(pos, &evsel_list->entries, node) {
1249 perf_evsel__free_stat_priv(pos); 1374 perf_evsel__free_stat_priv(pos);
1375 perf_evsel__free_counts(pos);
1376 perf_evsel__free_prev_raw_counts(pos);
1377 }
1250 perf_evlist__delete_maps(evsel_list); 1378 perf_evlist__delete_maps(evsel_list);
1251out: 1379out:
1252 perf_evlist__delete(evsel_list); 1380 perf_evlist__delete(evsel_list);
diff --git a/tools/perf/config/feature-tests.mak b/tools/perf/config/feature-tests.mak
index f5ac77485a4f..b4eabb44e381 100644
--- a/tools/perf/config/feature-tests.mak
+++ b/tools/perf/config/feature-tests.mak
@@ -225,3 +225,14 @@ int main(void)
225 return on_exit(NULL, NULL); 225 return on_exit(NULL, NULL);
226} 226}
227endef 227endef
228
229define SOURCE_LIBNUMA
230#include <numa.h>
231#include <numaif.h>
232
233int main(void)
234{
235 numa_available();
236 return 0;
237}
238endef \ No newline at end of file
diff --git a/tools/perf/tests/attr.c b/tools/perf/tests/attr.c
index f61dd3fb546b..bdcceb886f77 100644
--- a/tools/perf/tests/attr.c
+++ b/tools/perf/tests/attr.c
@@ -19,6 +19,11 @@
19 * permissions. All the event text files are stored there. 19 * permissions. All the event text files are stored there.
20 */ 20 */
21 21
22/*
23 * Powerpc needs __SANE_USERSPACE_TYPES__ before <linux/types.h> to select
24 * 'int-ll64.h' and avoid compile warnings when printing __u64 with %llu.
25 */
26#define __SANE_USERSPACE_TYPES__
22#include <stdlib.h> 27#include <stdlib.h>
23#include <stdio.h> 28#include <stdio.h>
24#include <inttypes.h> 29#include <inttypes.h>
diff --git a/tools/perf/tests/open-syscall-all-cpus.c b/tools/perf/tests/open-syscall-all-cpus.c
index 9b920a0cce79..b0657a9ccda6 100644
--- a/tools/perf/tests/open-syscall-all-cpus.c
+++ b/tools/perf/tests/open-syscall-all-cpus.c
@@ -98,6 +98,7 @@ int test__open_syscall_event_on_all_cpus(void)
98 } 98 }
99 } 99 }
100 100
101 perf_evsel__free_counts(evsel);
101out_close_fd: 102out_close_fd:
102 perf_evsel__close_fd(evsel, 1, threads->nr); 103 perf_evsel__close_fd(evsel, 1, threads->nr);
103out_evsel_delete: 104out_evsel_delete:
diff --git a/tools/perf/tests/perf-record.c b/tools/perf/tests/perf-record.c
index 6ea66cf6791b..1e8e5128d0da 100644
--- a/tools/perf/tests/perf-record.c
+++ b/tools/perf/tests/perf-record.c
@@ -96,7 +96,7 @@ int test__PERF_RECORD(void)
96 err = perf_evlist__prepare_workload(evlist, &opts, argv); 96 err = perf_evlist__prepare_workload(evlist, &opts, argv);
97 if (err < 0) { 97 if (err < 0) {
98 pr_debug("Couldn't run the workload!\n"); 98 pr_debug("Couldn't run the workload!\n");
99 goto out_delete_evlist; 99 goto out_delete_maps;
100 } 100 }
101 101
102 /* 102 /*
@@ -111,7 +111,7 @@ int test__PERF_RECORD(void)
111 err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask); 111 err = sched__get_first_possible_cpu(evlist->workload.pid, &cpu_mask);
112 if (err < 0) { 112 if (err < 0) {
113 pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno)); 113 pr_debug("sched__get_first_possible_cpu: %s\n", strerror(errno));
114 goto out_delete_evlist; 114 goto out_delete_maps;
115 } 115 }
116 116
117 cpu = err; 117 cpu = err;
@@ -121,7 +121,7 @@ int test__PERF_RECORD(void)
121 */ 121 */
122 if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) { 122 if (sched_setaffinity(evlist->workload.pid, cpu_mask_size, &cpu_mask) < 0) {
123 pr_debug("sched_setaffinity: %s\n", strerror(errno)); 123 pr_debug("sched_setaffinity: %s\n", strerror(errno));
124 goto out_delete_evlist; 124 goto out_delete_maps;
125 } 125 }
126 126
127 /* 127 /*
@@ -131,7 +131,7 @@ int test__PERF_RECORD(void)
131 err = perf_evlist__open(evlist); 131 err = perf_evlist__open(evlist);
132 if (err < 0) { 132 if (err < 0) {
133 pr_debug("perf_evlist__open: %s\n", strerror(errno)); 133 pr_debug("perf_evlist__open: %s\n", strerror(errno));
134 goto out_delete_evlist; 134 goto out_delete_maps;
135 } 135 }
136 136
137 /* 137 /*
@@ -142,7 +142,7 @@ int test__PERF_RECORD(void)
142 err = perf_evlist__mmap(evlist, opts.mmap_pages, false); 142 err = perf_evlist__mmap(evlist, opts.mmap_pages, false);
143 if (err < 0) { 143 if (err < 0) {
144 pr_debug("perf_evlist__mmap: %s\n", strerror(errno)); 144 pr_debug("perf_evlist__mmap: %s\n", strerror(errno));
145 goto out_delete_evlist; 145 goto out_delete_maps;
146 } 146 }
147 147
148 /* 148 /*
@@ -305,6 +305,8 @@ found_exit:
305 } 305 }
306out_err: 306out_err:
307 perf_evlist__munmap(evlist); 307 perf_evlist__munmap(evlist);
308out_delete_maps:
309 perf_evlist__delete_maps(evlist);
308out_delete_evlist: 310out_delete_evlist:
309 perf_evlist__delete(evlist); 311 perf_evlist__delete(evlist);
310out: 312out:
diff --git a/tools/perf/tests/vmlinux-kallsyms.c b/tools/perf/tests/vmlinux-kallsyms.c
index a1a8442829b4..7b4c4d26d1ba 100644
--- a/tools/perf/tests/vmlinux-kallsyms.c
+++ b/tools/perf/tests/vmlinux-kallsyms.c
@@ -44,7 +44,7 @@ int test__vmlinux_matches_kallsyms(void)
44 */ 44 */
45 if (machine__create_kernel_maps(&kallsyms) < 0) { 45 if (machine__create_kernel_maps(&kallsyms) < 0) {
46 pr_debug("machine__create_kernel_maps "); 46 pr_debug("machine__create_kernel_maps ");
47 return -1; 47 goto out;
48 } 48 }
49 49
50 /* 50 /*
@@ -227,5 +227,7 @@ detour:
227 map__fprintf(pos, stderr); 227 map__fprintf(pos, stderr);
228 } 228 }
229out: 229out:
230 machine__exit(&kallsyms);
231 machine__exit(&vmlinux);
230 return err; 232 return err;
231} 233}
diff --git a/tools/perf/ui/browser.c b/tools/perf/ui/browser.c
index 588bcb2d008b..809ea4632a34 100644
--- a/tools/perf/ui/browser.c
+++ b/tools/perf/ui/browser.c
@@ -273,6 +273,8 @@ void ui_browser__hide(struct ui_browser *browser __maybe_unused)
273{ 273{
274 pthread_mutex_lock(&ui__lock); 274 pthread_mutex_lock(&ui__lock);
275 ui_helpline__pop(); 275 ui_helpline__pop();
276 free(browser->helpline);
277 browser->helpline = NULL;
276 pthread_mutex_unlock(&ui__lock); 278 pthread_mutex_unlock(&ui__lock);
277} 279}
278 280
diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index 3cf2c3e0605f..5cd13d768cec 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -476,8 +476,10 @@ int perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
476 } 476 }
477 } 477 }
478 478
479 if (kallsyms__parse(filename, &args, find_symbol_cb) <= 0) 479 if (kallsyms__parse(filename, &args, find_symbol_cb) <= 0) {
480 free(event);
480 return -ENOENT; 481 return -ENOENT;
482 }
481 483
482 map = machine->vmlinux_maps[MAP__FUNCTION]; 484 map = machine->vmlinux_maps[MAP__FUNCTION];
483 size = snprintf(event->mmap.filename, sizeof(event->mmap.filename), 485 size = snprintf(event->mmap.filename, sizeof(event->mmap.filename),
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index e45332d08a58..baa26ddbcc7b 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -640,6 +640,11 @@ void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
640 } 640 }
641} 641}
642 642
643void perf_evsel__free_counts(struct perf_evsel *evsel)
644{
645 free(evsel->counts);
646}
647
643void perf_evsel__exit(struct perf_evsel *evsel) 648void perf_evsel__exit(struct perf_evsel *evsel)
644{ 649{
645 assert(list_empty(&evsel->node)); 650 assert(list_empty(&evsel->node));
@@ -659,6 +664,28 @@ void perf_evsel__delete(struct perf_evsel *evsel)
659 free(evsel); 664 free(evsel);
660} 665}
661 666
667static inline void compute_deltas(struct perf_evsel *evsel,
668 int cpu,
669 struct perf_counts_values *count)
670{
671 struct perf_counts_values tmp;
672
673 if (!evsel->prev_raw_counts)
674 return;
675
676 if (cpu == -1) {
677 tmp = evsel->prev_raw_counts->aggr;
678 evsel->prev_raw_counts->aggr = *count;
679 } else {
680 tmp = evsel->prev_raw_counts->cpu[cpu];
681 evsel->prev_raw_counts->cpu[cpu] = *count;
682 }
683
684 count->val = count->val - tmp.val;
685 count->ena = count->ena - tmp.ena;
686 count->run = count->run - tmp.run;
687}
688
662int __perf_evsel__read_on_cpu(struct perf_evsel *evsel, 689int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
663 int cpu, int thread, bool scale) 690 int cpu, int thread, bool scale)
664{ 691{
@@ -674,6 +701,8 @@ int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
674 if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0) 701 if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0)
675 return -errno; 702 return -errno;
676 703
704 compute_deltas(evsel, cpu, &count);
705
677 if (scale) { 706 if (scale) {
678 if (count.run == 0) 707 if (count.run == 0)
679 count.val = 0; 708 count.val = 0;
@@ -712,6 +741,8 @@ int __perf_evsel__read(struct perf_evsel *evsel,
712 } 741 }
713 } 742 }
714 743
744 compute_deltas(evsel, -1, aggr);
745
715 evsel->counts->scaled = 0; 746 evsel->counts->scaled = 0;
716 if (scale) { 747 if (scale) {
717 if (aggr->run == 0) { 748 if (aggr->run == 0) {
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index c68d1b82e843..cbf42322a27e 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -53,6 +53,7 @@ struct perf_evsel {
53 struct xyarray *sample_id; 53 struct xyarray *sample_id;
54 u64 *id; 54 u64 *id;
55 struct perf_counts *counts; 55 struct perf_counts *counts;
56 struct perf_counts *prev_raw_counts;
56 int idx; 57 int idx;
57 u32 ids; 58 u32 ids;
58 struct hists hists; 59 struct hists hists;
@@ -116,6 +117,7 @@ int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads);
116int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus); 117int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus);
117void perf_evsel__free_fd(struct perf_evsel *evsel); 118void perf_evsel__free_fd(struct perf_evsel *evsel);
118void perf_evsel__free_id(struct perf_evsel *evsel); 119void perf_evsel__free_id(struct perf_evsel *evsel);
120void perf_evsel__free_counts(struct perf_evsel *evsel);
119void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads); 121void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads);
120 122
121void __perf_evsel__set_sample_bit(struct perf_evsel *evsel, 123void __perf_evsel__set_sample_bit(struct perf_evsel *evsel,
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index fccd69dbbbb9..f6081cb3fca3 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -313,7 +313,8 @@ int build_id_cache__add_s(const char *sbuild_id, const char *debugdir,
313 if (is_kallsyms) { 313 if (is_kallsyms) {
314 if (symbol_conf.kptr_restrict) { 314 if (symbol_conf.kptr_restrict) {
315 pr_debug("Not caching a kptr_restrict'ed /proc/kallsyms\n"); 315 pr_debug("Not caching a kptr_restrict'ed /proc/kallsyms\n");
316 return 0; 316 err = 0;
317 goto out_free;
317 } 318 }
318 realname = (char *) name; 319 realname = (char *) name;
319 } else 320 } else
@@ -954,6 +955,7 @@ static int write_topo_node(int fd, int node)
954 } 955 }
955 956
956 fclose(fp); 957 fclose(fp);
958 fp = NULL;
957 959
958 ret = do_write(fd, &mem_total, sizeof(u64)); 960 ret = do_write(fd, &mem_total, sizeof(u64));
959 if (ret) 961 if (ret)
@@ -980,7 +982,8 @@ static int write_topo_node(int fd, int node)
980 ret = do_write_string(fd, buf); 982 ret = do_write_string(fd, buf);
981done: 983done:
982 free(buf); 984 free(buf);
983 fclose(fp); 985 if (fp)
986 fclose(fp);
984 return ret; 987 return ret;
985} 988}
986 989
@@ -2921,16 +2924,22 @@ int perf_event__process_tracing_data(union perf_event *event,
2921 session->repipe); 2924 session->repipe);
2922 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read; 2925 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
2923 2926
2924 if (readn(session->fd, buf, padding) < 0) 2927 if (readn(session->fd, buf, padding) < 0) {
2925 die("reading input file"); 2928 pr_err("%s: reading input file", __func__);
2929 return -1;
2930 }
2926 if (session->repipe) { 2931 if (session->repipe) {
2927 int retw = write(STDOUT_FILENO, buf, padding); 2932 int retw = write(STDOUT_FILENO, buf, padding);
2928 if (retw <= 0 || retw != padding) 2933 if (retw <= 0 || retw != padding) {
2929 die("repiping tracing data padding"); 2934 pr_err("%s: repiping tracing data padding", __func__);
2935 return -1;
2936 }
2930 } 2937 }
2931 2938
2932 if (size_read + padding != size) 2939 if (size_read + padding != size) {
2933 die("tracing data size mismatch"); 2940 pr_err("%s: tracing data size mismatch", __func__);
2941 return -1;
2942 }
2934 2943
2935 perf_evlist__prepare_tracepoint_events(session->evlist, 2944 perf_evlist__prepare_tracepoint_events(session->evlist,
2936 session->pevent); 2945 session->pevent);
diff --git a/tools/perf/util/map.c b/tools/perf/util/map.c
index ff94425779a2..6fcb9de62340 100644
--- a/tools/perf/util/map.c
+++ b/tools/perf/util/map.c
@@ -11,6 +11,7 @@
11#include "strlist.h" 11#include "strlist.h"
12#include "vdso.h" 12#include "vdso.h"
13#include "build-id.h" 13#include "build-id.h"
14#include <linux/string.h>
14 15
15const char *map_type__name[MAP__NR_TYPES] = { 16const char *map_type__name[MAP__NR_TYPES] = {
16 [MAP__FUNCTION] = "Functions", 17 [MAP__FUNCTION] = "Functions",
@@ -29,29 +30,29 @@ static inline int is_no_dso_memory(const char *filename)
29 !strcmp(filename, "[heap]"); 30 !strcmp(filename, "[heap]");
30} 31}
31 32
32void map__init(struct map *self, enum map_type type, 33void map__init(struct map *map, enum map_type type,
33 u64 start, u64 end, u64 pgoff, struct dso *dso) 34 u64 start, u64 end, u64 pgoff, struct dso *dso)
34{ 35{
35 self->type = type; 36 map->type = type;
36 self->start = start; 37 map->start = start;
37 self->end = end; 38 map->end = end;
38 self->pgoff = pgoff; 39 map->pgoff = pgoff;
39 self->dso = dso; 40 map->dso = dso;
40 self->map_ip = map__map_ip; 41 map->map_ip = map__map_ip;
41 self->unmap_ip = map__unmap_ip; 42 map->unmap_ip = map__unmap_ip;
42 RB_CLEAR_NODE(&self->rb_node); 43 RB_CLEAR_NODE(&map->rb_node);
43 self->groups = NULL; 44 map->groups = NULL;
44 self->referenced = false; 45 map->referenced = false;
45 self->erange_warned = false; 46 map->erange_warned = false;
46} 47}
47 48
48struct map *map__new(struct list_head *dsos__list, u64 start, u64 len, 49struct map *map__new(struct list_head *dsos__list, u64 start, u64 len,
49 u64 pgoff, u32 pid, char *filename, 50 u64 pgoff, u32 pid, char *filename,
50 enum map_type type) 51 enum map_type type)
51{ 52{
52 struct map *self = malloc(sizeof(*self)); 53 struct map *map = malloc(sizeof(*map));
53 54
54 if (self != NULL) { 55 if (map != NULL) {
55 char newfilename[PATH_MAX]; 56 char newfilename[PATH_MAX];
56 struct dso *dso; 57 struct dso *dso;
57 int anon, no_dso, vdso; 58 int anon, no_dso, vdso;
@@ -74,10 +75,10 @@ struct map *map__new(struct list_head *dsos__list, u64 start, u64 len,
74 if (dso == NULL) 75 if (dso == NULL)
75 goto out_delete; 76 goto out_delete;
76 77
77 map__init(self, type, start, start + len, pgoff, dso); 78 map__init(map, type, start, start + len, pgoff, dso);
78 79
79 if (anon || no_dso) { 80 if (anon || no_dso) {
80 self->map_ip = self->unmap_ip = identity__map_ip; 81 map->map_ip = map->unmap_ip = identity__map_ip;
81 82
82 /* 83 /*
83 * Set memory without DSO as loaded. All map__find_* 84 * Set memory without DSO as loaded. All map__find_*
@@ -85,12 +86,12 @@ struct map *map__new(struct list_head *dsos__list, u64 start, u64 len,
85 * unnecessary map__load warning. 86 * unnecessary map__load warning.
86 */ 87 */
87 if (no_dso) 88 if (no_dso)
88 dso__set_loaded(dso, self->type); 89 dso__set_loaded(dso, map->type);
89 } 90 }
90 } 91 }
91 return self; 92 return map;
92out_delete: 93out_delete:
93 free(self); 94 free(map);
94 return NULL; 95 return NULL;
95} 96}
96 97
@@ -113,48 +114,48 @@ struct map *map__new2(u64 start, struct dso *dso, enum map_type type)
113 return map; 114 return map;
114} 115}
115 116
116void map__delete(struct map *self) 117void map__delete(struct map *map)
117{ 118{
118 free(self); 119 free(map);
119} 120}
120 121
121void map__fixup_start(struct map *self) 122void map__fixup_start(struct map *map)
122{ 123{
123 struct rb_root *symbols = &self->dso->symbols[self->type]; 124 struct rb_root *symbols = &map->dso->symbols[map->type];
124 struct rb_node *nd = rb_first(symbols); 125 struct rb_node *nd = rb_first(symbols);
125 if (nd != NULL) { 126 if (nd != NULL) {
126 struct symbol *sym = rb_entry(nd, struct symbol, rb_node); 127 struct symbol *sym = rb_entry(nd, struct symbol, rb_node);
127 self->start = sym->start; 128 map->start = sym->start;
128 } 129 }
129} 130}
130 131
131void map__fixup_end(struct map *self) 132void map__fixup_end(struct map *map)
132{ 133{
133 struct rb_root *symbols = &self->dso->symbols[self->type]; 134 struct rb_root *symbols = &map->dso->symbols[map->type];
134 struct rb_node *nd = rb_last(symbols); 135 struct rb_node *nd = rb_last(symbols);
135 if (nd != NULL) { 136 if (nd != NULL) {
136 struct symbol *sym = rb_entry(nd, struct symbol, rb_node); 137 struct symbol *sym = rb_entry(nd, struct symbol, rb_node);
137 self->end = sym->end; 138 map->end = sym->end;
138 } 139 }
139} 140}
140 141
141#define DSO__DELETED "(deleted)" 142#define DSO__DELETED "(deleted)"
142 143
143int map__load(struct map *self, symbol_filter_t filter) 144int map__load(struct map *map, symbol_filter_t filter)
144{ 145{
145 const char *name = self->dso->long_name; 146 const char *name = map->dso->long_name;
146 int nr; 147 int nr;
147 148
148 if (dso__loaded(self->dso, self->type)) 149 if (dso__loaded(map->dso, map->type))
149 return 0; 150 return 0;
150 151
151 nr = dso__load(self->dso, self, filter); 152 nr = dso__load(map->dso, map, filter);
152 if (nr < 0) { 153 if (nr < 0) {
153 if (self->dso->has_build_id) { 154 if (map->dso->has_build_id) {
154 char sbuild_id[BUILD_ID_SIZE * 2 + 1]; 155 char sbuild_id[BUILD_ID_SIZE * 2 + 1];
155 156
156 build_id__sprintf(self->dso->build_id, 157 build_id__sprintf(map->dso->build_id,
157 sizeof(self->dso->build_id), 158 sizeof(map->dso->build_id),
158 sbuild_id); 159 sbuild_id);
159 pr_warning("%s with build id %s not found", 160 pr_warning("%s with build id %s not found",
160 name, sbuild_id); 161 name, sbuild_id);
@@ -184,43 +185,36 @@ int map__load(struct map *self, symbol_filter_t filter)
184 * Only applies to the kernel, as its symtabs aren't relative like the 185 * Only applies to the kernel, as its symtabs aren't relative like the
185 * module ones. 186 * module ones.
186 */ 187 */
187 if (self->dso->kernel) 188 if (map->dso->kernel)
188 map__reloc_vmlinux(self); 189 map__reloc_vmlinux(map);
189 190
190 return 0; 191 return 0;
191} 192}
192 193
193struct symbol *map__find_symbol(struct map *self, u64 addr, 194struct symbol *map__find_symbol(struct map *map, u64 addr,
194 symbol_filter_t filter) 195 symbol_filter_t filter)
195{ 196{
196 if (map__load(self, filter) < 0) 197 if (map__load(map, filter) < 0)
197 return NULL; 198 return NULL;
198 199
199 return dso__find_symbol(self->dso, self->type, addr); 200 return dso__find_symbol(map->dso, map->type, addr);
200} 201}
201 202
202struct symbol *map__find_symbol_by_name(struct map *self, const char *name, 203struct symbol *map__find_symbol_by_name(struct map *map, const char *name,
203 symbol_filter_t filter) 204 symbol_filter_t filter)
204{ 205{
205 if (map__load(self, filter) < 0) 206 if (map__load(map, filter) < 0)
206 return NULL; 207 return NULL;
207 208
208 if (!dso__sorted_by_name(self->dso, self->type)) 209 if (!dso__sorted_by_name(map->dso, map->type))
209 dso__sort_by_name(self->dso, self->type); 210 dso__sort_by_name(map->dso, map->type);
210 211
211 return dso__find_symbol_by_name(self->dso, self->type, name); 212 return dso__find_symbol_by_name(map->dso, map->type, name);
212} 213}
213 214
214struct map *map__clone(struct map *self) 215struct map *map__clone(struct map *map)
215{ 216{
216 struct map *map = malloc(sizeof(*self)); 217 return memdup(map, sizeof(*map));
217
218 if (!map)
219 return NULL;
220
221 memcpy(map, self, sizeof(*self));
222
223 return map;
224} 218}
225 219
226int map__overlap(struct map *l, struct map *r) 220int map__overlap(struct map *l, struct map *r)
@@ -237,10 +231,10 @@ int map__overlap(struct map *l, struct map *r)
237 return 0; 231 return 0;
238} 232}
239 233
240size_t map__fprintf(struct map *self, FILE *fp) 234size_t map__fprintf(struct map *map, FILE *fp)
241{ 235{
242 return fprintf(fp, " %" PRIx64 "-%" PRIx64 " %" PRIx64 " %s\n", 236 return fprintf(fp, " %" PRIx64 "-%" PRIx64 " %" PRIx64 " %s\n",
243 self->start, self->end, self->pgoff, self->dso->name); 237 map->start, map->end, map->pgoff, map->dso->name);
244} 238}
245 239
246size_t map__fprintf_dsoname(struct map *map, FILE *fp) 240size_t map__fprintf_dsoname(struct map *map, FILE *fp)
@@ -528,9 +522,9 @@ static u64 map__reloc_unmap_ip(struct map *map, u64 ip)
528 return ip - (s64)map->pgoff; 522 return ip - (s64)map->pgoff;
529} 523}
530 524
531void map__reloc_vmlinux(struct map *self) 525void map__reloc_vmlinux(struct map *map)
532{ 526{
533 struct kmap *kmap = map__kmap(self); 527 struct kmap *kmap = map__kmap(map);
534 s64 reloc; 528 s64 reloc;
535 529
536 if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->unrelocated_addr) 530 if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->unrelocated_addr)
@@ -542,9 +536,9 @@ void map__reloc_vmlinux(struct map *self)
542 if (!reloc) 536 if (!reloc)
543 return; 537 return;
544 538
545 self->map_ip = map__reloc_map_ip; 539 map->map_ip = map__reloc_map_ip;
546 self->unmap_ip = map__reloc_unmap_ip; 540 map->unmap_ip = map__reloc_unmap_ip;
547 self->pgoff = reloc; 541 map->pgoff = reloc;
548} 542}
549 543
550void maps__insert(struct rb_root *maps, struct map *map) 544void maps__insert(struct rb_root *maps, struct map *map)
@@ -567,9 +561,9 @@ void maps__insert(struct rb_root *maps, struct map *map)
567 rb_insert_color(&map->rb_node, maps); 561 rb_insert_color(&map->rb_node, maps);
568} 562}
569 563
570void maps__remove(struct rb_root *self, struct map *map) 564void maps__remove(struct rb_root *maps, struct map *map)
571{ 565{
572 rb_erase(&map->rb_node, self); 566 rb_erase(&map->rb_node, maps);
573} 567}
574 568
575struct map *maps__find(struct rb_root *maps, u64 ip) 569struct map *maps__find(struct rb_root *maps, u64 ip)
diff --git a/tools/perf/util/map.h b/tools/perf/util/map.h
index bcb39e2a6965..a887f2c9dfbb 100644
--- a/tools/perf/util/map.h
+++ b/tools/perf/util/map.h
@@ -57,9 +57,9 @@ struct map_groups {
57 struct machine *machine; 57 struct machine *machine;
58}; 58};
59 59
60static inline struct kmap *map__kmap(struct map *self) 60static inline struct kmap *map__kmap(struct map *map)
61{ 61{
62 return (struct kmap *)(self + 1); 62 return (struct kmap *)(map + 1);
63} 63}
64 64
65static inline u64 map__map_ip(struct map *map, u64 ip) 65static inline u64 map__map_ip(struct map *map, u64 ip)
@@ -85,27 +85,27 @@ struct symbol;
85 85
86typedef int (*symbol_filter_t)(struct map *map, struct symbol *sym); 86typedef int (*symbol_filter_t)(struct map *map, struct symbol *sym);
87 87
88void map__init(struct map *self, enum map_type type, 88void map__init(struct map *map, enum map_type type,
89 u64 start, u64 end, u64 pgoff, struct dso *dso); 89 u64 start, u64 end, u64 pgoff, struct dso *dso);
90struct map *map__new(struct list_head *dsos__list, u64 start, u64 len, 90struct map *map__new(struct list_head *dsos__list, u64 start, u64 len,
91 u64 pgoff, u32 pid, char *filename, 91 u64 pgoff, u32 pid, char *filename,
92 enum map_type type); 92 enum map_type type);
93struct map *map__new2(u64 start, struct dso *dso, enum map_type type); 93struct map *map__new2(u64 start, struct dso *dso, enum map_type type);
94void map__delete(struct map *self); 94void map__delete(struct map *map);
95struct map *map__clone(struct map *self); 95struct map *map__clone(struct map *map);
96int map__overlap(struct map *l, struct map *r); 96int map__overlap(struct map *l, struct map *r);
97size_t map__fprintf(struct map *self, FILE *fp); 97size_t map__fprintf(struct map *map, FILE *fp);
98size_t map__fprintf_dsoname(struct map *map, FILE *fp); 98size_t map__fprintf_dsoname(struct map *map, FILE *fp);
99 99
100int map__load(struct map *self, symbol_filter_t filter); 100int map__load(struct map *map, symbol_filter_t filter);
101struct symbol *map__find_symbol(struct map *self, 101struct symbol *map__find_symbol(struct map *map,
102 u64 addr, symbol_filter_t filter); 102 u64 addr, symbol_filter_t filter);
103struct symbol *map__find_symbol_by_name(struct map *self, const char *name, 103struct symbol *map__find_symbol_by_name(struct map *map, const char *name,
104 symbol_filter_t filter); 104 symbol_filter_t filter);
105void map__fixup_start(struct map *self); 105void map__fixup_start(struct map *map);
106void map__fixup_end(struct map *self); 106void map__fixup_end(struct map *map);
107 107
108void map__reloc_vmlinux(struct map *self); 108void map__reloc_vmlinux(struct map *map);
109 109
110size_t __map_groups__fprintf_maps(struct map_groups *mg, 110size_t __map_groups__fprintf_maps(struct map_groups *mg,
111 enum map_type type, int verbose, FILE *fp); 111 enum map_type type, int verbose, FILE *fp);
diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c
index 7ad62393aa88..83336610faa9 100644
--- a/tools/perf/util/sort.c
+++ b/tools/perf/util/sort.c
@@ -249,7 +249,7 @@ static int hist_entry__srcline_snprintf(struct hist_entry *self, char *bf,
249 size_t size, 249 size_t size,
250 unsigned int width __maybe_unused) 250 unsigned int width __maybe_unused)
251{ 251{
252 FILE *fp; 252 FILE *fp = NULL;
253 char cmd[PATH_MAX + 2], *path = self->srcline, *nl; 253 char cmd[PATH_MAX + 2], *path = self->srcline, *nl;
254 size_t line_len; 254 size_t line_len;
255 255
@@ -270,7 +270,6 @@ static int hist_entry__srcline_snprintf(struct hist_entry *self, char *bf,
270 270
271 if (getline(&path, &line_len, fp) < 0 || !line_len) 271 if (getline(&path, &line_len, fp) < 0 || !line_len)
272 goto out_ip; 272 goto out_ip;
273 fclose(fp);
274 self->srcline = strdup(path); 273 self->srcline = strdup(path);
275 if (self->srcline == NULL) 274 if (self->srcline == NULL)
276 goto out_ip; 275 goto out_ip;
@@ -280,8 +279,12 @@ static int hist_entry__srcline_snprintf(struct hist_entry *self, char *bf,
280 *nl = '\0'; 279 *nl = '\0';
281 path = self->srcline; 280 path = self->srcline;
282out_path: 281out_path:
282 if (fp)
283 pclose(fp);
283 return repsep_snprintf(bf, size, "%s", path); 284 return repsep_snprintf(bf, size, "%s", path);
284out_ip: 285out_ip:
286 if (fp)
287 pclose(fp);
285 return repsep_snprintf(bf, size, "%-#*llx", BITS_PER_LONG / 4, self->ip); 288 return repsep_snprintf(bf, size, "%-#*llx", BITS_PER_LONG / 4, self->ip);
286} 289}
287 290
diff --git a/tools/perf/util/strlist.c b/tools/perf/util/strlist.c
index 155d8b7078a7..55433aa42c8f 100644
--- a/tools/perf/util/strlist.c
+++ b/tools/perf/util/strlist.c
@@ -35,11 +35,11 @@ out_delete:
35 return NULL; 35 return NULL;
36} 36}
37 37
38static void str_node__delete(struct str_node *self, bool dupstr) 38static void str_node__delete(struct str_node *snode, bool dupstr)
39{ 39{
40 if (dupstr) 40 if (dupstr)
41 free((void *)self->s); 41 free((void *)snode->s);
42 free(self); 42 free(snode);
43} 43}
44 44
45static 45static
@@ -59,12 +59,12 @@ static int strlist__node_cmp(struct rb_node *rb_node, const void *entry)
59 return strcmp(snode->s, str); 59 return strcmp(snode->s, str);
60} 60}
61 61
62int strlist__add(struct strlist *self, const char *new_entry) 62int strlist__add(struct strlist *slist, const char *new_entry)
63{ 63{
64 return rblist__add_node(&self->rblist, new_entry); 64 return rblist__add_node(&slist->rblist, new_entry);
65} 65}
66 66
67int strlist__load(struct strlist *self, const char *filename) 67int strlist__load(struct strlist *slist, const char *filename)
68{ 68{
69 char entry[1024]; 69 char entry[1024];
70 int err; 70 int err;
@@ -80,7 +80,7 @@ int strlist__load(struct strlist *self, const char *filename)
80 continue; 80 continue;
81 entry[len - 1] = '\0'; 81 entry[len - 1] = '\0';
82 82
83 err = strlist__add(self, entry); 83 err = strlist__add(slist, entry);
84 if (err != 0) 84 if (err != 0)
85 goto out; 85 goto out;
86 } 86 }
@@ -107,56 +107,56 @@ struct str_node *strlist__find(struct strlist *slist, const char *entry)
107 return snode; 107 return snode;
108} 108}
109 109
110static int strlist__parse_list_entry(struct strlist *self, const char *s) 110static int strlist__parse_list_entry(struct strlist *slist, const char *s)
111{ 111{
112 if (strncmp(s, "file://", 7) == 0) 112 if (strncmp(s, "file://", 7) == 0)
113 return strlist__load(self, s + 7); 113 return strlist__load(slist, s + 7);
114 114
115 return strlist__add(self, s); 115 return strlist__add(slist, s);
116} 116}
117 117
118int strlist__parse_list(struct strlist *self, const char *s) 118int strlist__parse_list(struct strlist *slist, const char *s)
119{ 119{
120 char *sep; 120 char *sep;
121 int err; 121 int err;
122 122
123 while ((sep = strchr(s, ',')) != NULL) { 123 while ((sep = strchr(s, ',')) != NULL) {
124 *sep = '\0'; 124 *sep = '\0';
125 err = strlist__parse_list_entry(self, s); 125 err = strlist__parse_list_entry(slist, s);
126 *sep = ','; 126 *sep = ',';
127 if (err != 0) 127 if (err != 0)
128 return err; 128 return err;
129 s = sep + 1; 129 s = sep + 1;
130 } 130 }
131 131
132 return *s ? strlist__parse_list_entry(self, s) : 0; 132 return *s ? strlist__parse_list_entry(slist, s) : 0;
133} 133}
134 134
135struct strlist *strlist__new(bool dupstr, const char *slist) 135struct strlist *strlist__new(bool dupstr, const char *list)
136{ 136{
137 struct strlist *self = malloc(sizeof(*self)); 137 struct strlist *slist = malloc(sizeof(*slist));
138 138
139 if (self != NULL) { 139 if (slist != NULL) {
140 rblist__init(&self->rblist); 140 rblist__init(&slist->rblist);
141 self->rblist.node_cmp = strlist__node_cmp; 141 slist->rblist.node_cmp = strlist__node_cmp;
142 self->rblist.node_new = strlist__node_new; 142 slist->rblist.node_new = strlist__node_new;
143 self->rblist.node_delete = strlist__node_delete; 143 slist->rblist.node_delete = strlist__node_delete;
144 144
145 self->dupstr = dupstr; 145 slist->dupstr = dupstr;
146 if (slist && strlist__parse_list(self, slist) != 0) 146 if (slist && strlist__parse_list(slist, list) != 0)
147 goto out_error; 147 goto out_error;
148 } 148 }
149 149
150 return self; 150 return slist;
151out_error: 151out_error:
152 free(self); 152 free(slist);
153 return NULL; 153 return NULL;
154} 154}
155 155
156void strlist__delete(struct strlist *self) 156void strlist__delete(struct strlist *slist)
157{ 157{
158 if (self != NULL) 158 if (slist != NULL)
159 rblist__delete(&self->rblist); 159 rblist__delete(&slist->rblist);
160} 160}
161 161
162struct str_node *strlist__entry(const struct strlist *slist, unsigned int idx) 162struct str_node *strlist__entry(const struct strlist *slist, unsigned int idx)
diff --git a/tools/perf/util/strlist.h b/tools/perf/util/strlist.h
index dd9f922ec67c..5c7f87069d9c 100644
--- a/tools/perf/util/strlist.h
+++ b/tools/perf/util/strlist.h
@@ -17,34 +17,34 @@ struct strlist {
17}; 17};
18 18
19struct strlist *strlist__new(bool dupstr, const char *slist); 19struct strlist *strlist__new(bool dupstr, const char *slist);
20void strlist__delete(struct strlist *self); 20void strlist__delete(struct strlist *slist);
21 21
22void strlist__remove(struct strlist *self, struct str_node *sn); 22void strlist__remove(struct strlist *slist, struct str_node *sn);
23int strlist__load(struct strlist *self, const char *filename); 23int strlist__load(struct strlist *slist, const char *filename);
24int strlist__add(struct strlist *self, const char *str); 24int strlist__add(struct strlist *slist, const char *str);
25 25
26struct str_node *strlist__entry(const struct strlist *self, unsigned int idx); 26struct str_node *strlist__entry(const struct strlist *slist, unsigned int idx);
27struct str_node *strlist__find(struct strlist *self, const char *entry); 27struct str_node *strlist__find(struct strlist *slist, const char *entry);
28 28
29static inline bool strlist__has_entry(struct strlist *self, const char *entry) 29static inline bool strlist__has_entry(struct strlist *slist, const char *entry)
30{ 30{
31 return strlist__find(self, entry) != NULL; 31 return strlist__find(slist, entry) != NULL;
32} 32}
33 33
34static inline bool strlist__empty(const struct strlist *self) 34static inline bool strlist__empty(const struct strlist *slist)
35{ 35{
36 return rblist__empty(&self->rblist); 36 return rblist__empty(&slist->rblist);
37} 37}
38 38
39static inline unsigned int strlist__nr_entries(const struct strlist *self) 39static inline unsigned int strlist__nr_entries(const struct strlist *slist)
40{ 40{
41 return rblist__nr_entries(&self->rblist); 41 return rblist__nr_entries(&slist->rblist);
42} 42}
43 43
44/* For strlist iteration */ 44/* For strlist iteration */
45static inline struct str_node *strlist__first(struct strlist *self) 45static inline struct str_node *strlist__first(struct strlist *slist)
46{ 46{
47 struct rb_node *rn = rb_first(&self->rblist.entries); 47 struct rb_node *rn = rb_first(&slist->rblist.entries);
48 return rn ? rb_entry(rn, struct str_node, rb_node) : NULL; 48 return rn ? rb_entry(rn, struct str_node, rb_node) : NULL;
49} 49}
50static inline struct str_node *strlist__next(struct str_node *sn) 50static inline struct str_node *strlist__next(struct str_node *sn)
@@ -59,21 +59,21 @@ static inline struct str_node *strlist__next(struct str_node *sn)
59/** 59/**
60 * strlist_for_each - iterate over a strlist 60 * strlist_for_each - iterate over a strlist
61 * @pos: the &struct str_node to use as a loop cursor. 61 * @pos: the &struct str_node to use as a loop cursor.
62 * @self: the &struct strlist for loop. 62 * @slist: the &struct strlist for loop.
63 */ 63 */
64#define strlist__for_each(pos, self) \ 64#define strlist__for_each(pos, slist) \
65 for (pos = strlist__first(self); pos; pos = strlist__next(pos)) 65 for (pos = strlist__first(slist); pos; pos = strlist__next(pos))
66 66
67/** 67/**
68 * strlist_for_each_safe - iterate over a strlist safe against removal of 68 * strlist_for_each_safe - iterate over a strlist safe against removal of
69 * str_node 69 * str_node
70 * @pos: the &struct str_node to use as a loop cursor. 70 * @pos: the &struct str_node to use as a loop cursor.
71 * @n: another &struct str_node to use as temporary storage. 71 * @n: another &struct str_node to use as temporary storage.
72 * @self: the &struct strlist for loop. 72 * @slist: the &struct strlist for loop.
73 */ 73 */
74#define strlist__for_each_safe(pos, n, self) \ 74#define strlist__for_each_safe(pos, n, slist) \
75 for (pos = strlist__first(self), n = strlist__next(pos); pos;\ 75 for (pos = strlist__first(slist), n = strlist__next(pos); pos;\
76 pos = n, n = strlist__next(n)) 76 pos = n, n = strlist__next(n))
77 77
78int strlist__parse_list(struct strlist *self, const char *s); 78int strlist__parse_list(struct strlist *slist, const char *s);
79#endif /* __PERF_STRLIST_H */ 79#endif /* __PERF_STRLIST_H */
generated by cgit v1.2.2 (git 2.25.0) at 2025-08-12 06:51:48 -0400