aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorIngo Molnar <mingo@kernel.org>2015-06-09 05:46:04 -0400
committerIngo Molnar <mingo@kernel.org>2015-06-09 05:46:04 -0400
commit028c63b56795206464263fa3bc47094704c2a840 (patch)
tree5b89c379e629ed3c23f22102be66a99025c496da
parenta3d86542de8850be52e8589da22b24002941dfb7 (diff)
parentd3a7c489c7fd2463e3b2c3a2179c7be879dd9cb4 (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: User visible changes: - Fix perf.data size reporting in 'perf record' in no-buildid mode (He Kuang) Infrastructure changes: - Protect accesses the DSO rbtrees/lists with a rw lock and reference count struct dso instances (Arnaldo Carvalho de Melo) - Export dynamic symbols used by traceevent plugins (He Kuang) - Add libtrace-dynamic-list file to libtraceevent's .gitignore (He Kuang) - Refactor shadow stats code in 'perf stat', prep work for further patchkits (Jiri Olsa) Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Ingo Molnar <mingo@kernel.org>
Diffstat
-rw-r--r--tools/lib/traceevent/.gitignore1
-rw-r--r--tools/lib/traceevent/Makefile14
-rw-r--r--tools/perf/Makefile.perf14
-rw-r--r--tools/perf/builtin-record.c6
-rw-r--r--tools/perf/builtin-stat.c506
-rw-r--r--tools/perf/tests/dso-data.c4
-rw-r--r--tools/perf/tests/hists_common.c6
-rw-r--r--tools/perf/util/Build1
-rw-r--r--tools/perf/util/dso.c87
-rw-r--r--tools/perf/util/dso.h24
-rw-r--r--tools/perf/util/header.c1
-rw-r--r--tools/perf/util/machine.c58
-rw-r--r--tools/perf/util/machine.h4
-rw-r--r--tools/perf/util/map.c11
-rw-r--r--tools/perf/util/probe-event.c2
-rw-r--r--tools/perf/util/probe-finder.c2
-rw-r--r--tools/perf/util/stat-shadow.c434
-rw-r--r--tools/perf/util/stat.c35
-rw-r--r--tools/perf/util/stat.h40
-rw-r--r--tools/perf/util/symbol-elf.c2
-rw-r--r--tools/perf/util/symbol.c4
-rw-r--r--tools/perf/util/vdso.c54
22 files changed, 737 insertions, 573 deletions
diff --git a/tools/lib/traceevent/.gitignore b/tools/lib/traceevent/.gitignore
index 35f56be5a4cd..3c60335fe7be 100644
--- a/tools/lib/traceevent/.gitignore
+++ b/tools/lib/traceevent/.gitignore
@@ -1 +1,2 @@
1TRACEEVENT-CFLAGS 1TRACEEVENT-CFLAGS
2libtraceevent-dynamic-list
diff --git a/tools/lib/traceevent/Makefile b/tools/lib/traceevent/Makefile
index 84640394ebf9..6daaff652aff 100644
--- a/tools/lib/traceevent/Makefile
+++ b/tools/lib/traceevent/Makefile
@@ -23,6 +23,7 @@ endef
23# Allow setting CC and AR, or setting CROSS_COMPILE as a prefix. 23# Allow setting CC and AR, or setting CROSS_COMPILE as a prefix.
24$(call allow-override,CC,$(CROSS_COMPILE)gcc) 24$(call allow-override,CC,$(CROSS_COMPILE)gcc)
25$(call allow-override,AR,$(CROSS_COMPILE)ar) 25$(call allow-override,AR,$(CROSS_COMPILE)ar)
26$(call allow-override,NM,$(CROSS_COMPILE)nm)
26 27
27EXT = -std=gnu99 28EXT = -std=gnu99
28INSTALL = install 29INSTALL = install
@@ -157,8 +158,9 @@ PLUGINS_IN := $(PLUGINS:.so=-in.o)
157 158
158TE_IN := $(OUTPUT)libtraceevent-in.o 159TE_IN := $(OUTPUT)libtraceevent-in.o
159LIB_FILE := $(addprefix $(OUTPUT),$(LIB_FILE)) 160LIB_FILE := $(addprefix $(OUTPUT),$(LIB_FILE))
161DYNAMIC_LIST_FILE := $(OUTPUT)libtraceevent-dynamic-list
160 162
161CMD_TARGETS = $(LIB_FILE) $(PLUGINS) 163CMD_TARGETS = $(LIB_FILE) $(PLUGINS) $(DYNAMIC_LIST_FILE)
162 164
163TARGETS = $(CMD_TARGETS) 165TARGETS = $(CMD_TARGETS)
164 166
@@ -175,6 +177,9 @@ $(OUTPUT)libtraceevent.so: $(TE_IN)
175$(OUTPUT)libtraceevent.a: $(TE_IN) 177$(OUTPUT)libtraceevent.a: $(TE_IN)
176 $(QUIET_LINK)$(RM) $@; $(AR) rcs $@ $^ 178 $(QUIET_LINK)$(RM) $@; $(AR) rcs $@ $^
177 179
180$(OUTPUT)libtraceevent-dynamic-list: $(PLUGINS)
181 $(QUIET_GEN)$(call do_generate_dynamic_list_file, $(PLUGINS), $@)
182
178plugins: $(PLUGINS) 183plugins: $(PLUGINS)
179 184
180__plugin_obj = $(notdir $@) 185__plugin_obj = $(notdir $@)
@@ -244,6 +249,13 @@ define do_install_plugins
244 done 249 done
245endef 250endef
246 251
252define do_generate_dynamic_list_file
253 (echo '{'; \
254 $(NM) -u -D $1 | awk 'NF>1 {print "\t"$$2";"}' | sort -u; \
255 echo '};'; \
256 ) > $2
257endef
258
247install_lib: all_cmd install_plugins 259install_lib: all_cmd install_plugins
248 $(call QUIET_INSTALL, $(LIB_FILE)) \ 260 $(call QUIET_INSTALL, $(LIB_FILE)) \
249 $(call do_install,$(LIB_FILE),$(libdir_SQ)) 261 $(call do_install,$(LIB_FILE),$(libdir_SQ))
diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf
index 5816a3bb7e9f..b1dfcd8e93e3 100644
--- a/tools/perf/Makefile.perf
+++ b/tools/perf/Makefile.perf
@@ -173,6 +173,9 @@ endif
173LIBTRACEEVENT = $(TE_PATH)libtraceevent.a 173LIBTRACEEVENT = $(TE_PATH)libtraceevent.a
174export LIBTRACEEVENT 174export LIBTRACEEVENT
175 175
176LIBTRACEEVENT_DYNAMIC_LIST = $(TE_PATH)libtraceevent-dynamic-list
177LDFLAGS += -Xlinker --dynamic-list=$(LIBTRACEEVENT_DYNAMIC_LIST)
178
176LIBAPI = $(LIB_PATH)libapi.a 179LIBAPI = $(LIB_PATH)libapi.a
177export LIBAPI 180export LIBAPI
178 181
@@ -278,7 +281,7 @@ build := -f $(srctree)/tools/build/Makefile.build dir=. obj
278$(PERF_IN): $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)common-cmds.h FORCE 281$(PERF_IN): $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)common-cmds.h FORCE
279 $(Q)$(MAKE) $(build)=perf 282 $(Q)$(MAKE) $(build)=perf
280 283
281$(OUTPUT)perf: $(PERFLIBS) $(PERF_IN) 284$(OUTPUT)perf: $(PERFLIBS) $(PERF_IN) $(LIBTRACEEVENT_DYNAMIC_LIST)
282 $(QUIET_LINK)$(CC) $(CFLAGS) $(LDFLAGS) $(PERF_IN) $(LIBS) -o $@ 285 $(QUIET_LINK)$(CC) $(CFLAGS) $(LDFLAGS) $(PERF_IN) $(LIBS) -o $@
283 286
284$(GTK_IN): FORCE 287$(GTK_IN): FORCE
@@ -373,7 +376,13 @@ $(LIB_FILE): $(LIBPERF_IN)
373LIBTRACEEVENT_FLAGS += plugin_dir=$(plugindir_SQ) 376LIBTRACEEVENT_FLAGS += plugin_dir=$(plugindir_SQ)
374 377
375$(LIBTRACEEVENT): FORCE 378$(LIBTRACEEVENT): FORCE
376 $(Q)$(MAKE) -C $(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) O=$(OUTPUT) $(OUTPUT)libtraceevent.a plugins 379 $(Q)$(MAKE) -C $(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) O=$(OUTPUT) $(OUTPUT)libtraceevent.a
380
381libtraceevent_plugins: FORCE
382 $(Q)$(MAKE) -C $(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) O=$(OUTPUT) plugins
383
384$(LIBTRACEEVENT_DYNAMIC_LIST): libtraceevent_plugins
385 $(Q)$(MAKE) -C $(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) O=$(OUTPUT) $(OUTPUT)libtraceevent-dynamic-list
377 386
378$(LIBTRACEEVENT)-clean: 387$(LIBTRACEEVENT)-clean:
379 $(call QUIET_CLEAN, libtraceevent) 388 $(call QUIET_CLEAN, libtraceevent)
@@ -555,4 +564,5 @@ FORCE:
555.PHONY: all install clean config-clean strip install-gtk 564.PHONY: all install clean config-clean strip install-gtk
556.PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell 565.PHONY: shell_compatibility_test please_set_SHELL_PATH_to_a_more_modern_shell
557.PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE single_dep 566.PHONY: $(GIT-HEAD-PHONY) TAGS tags cscope FORCE single_dep
567.PHONY: libtraceevent_plugins
558 568
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 91aa2a3dcf19..d3731cce7c1c 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -345,12 +345,9 @@ static int process_buildids(struct record *rec)
345 struct perf_data_file *file = &rec->file; 345 struct perf_data_file *file = &rec->file;
346 struct perf_session *session = rec->session; 346 struct perf_session *session = rec->session;
347 347
348 u64 size = lseek(perf_data_file__fd(file), 0, SEEK_CUR); 348 if (file->size == 0)
349 if (size == 0)
350 return 0; 349 return 0;
351 350
352 file->size = size;
353
354 /* 351 /*
355 * During this process, it'll load kernel map and replace the 352 * During this process, it'll load kernel map and replace the
356 * dso->long_name to a real pathname it found. In this case 353 * dso->long_name to a real pathname it found. In this case
@@ -719,6 +716,7 @@ out_child:
719 716
720 if (!err && !file->is_pipe) { 717 if (!err && !file->is_pipe) {
721 rec->session->header.data_size += rec->bytes_written; 718 rec->session->header.data_size += rec->bytes_written;
719 file->size = lseek(perf_data_file__fd(file), 0, SEEK_CUR);
722 720
723 if (!rec->no_buildid) { 721 if (!rec->no_buildid) {
724 process_buildids(rec); 722 process_buildids(rec);
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index fd577f725d23..ff3d25803400 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -73,8 +73,8 @@ static void print_counter(struct perf_evsel *counter, char *prefix);
73static void print_aggr(char *prefix); 73static void print_aggr(char *prefix);
74 74
75/* Default events used for perf stat -T */ 75/* Default events used for perf stat -T */
76static const char * const transaction_attrs[] = { 76static const char *transaction_attrs = {
77 "task-clock", 77 "task-clock,"
78 "{" 78 "{"
79 "instructions," 79 "instructions,"
80 "cycles," 80 "cycles,"
@@ -86,8 +86,8 @@ static const char * const transaction_attrs[] = {
86}; 86};
87 87
88/* More limited version when the CPU does not have all events. */ 88/* More limited version when the CPU does not have all events. */
89static const char * const transaction_limited_attrs[] = { 89static const char * transaction_limited_attrs = {
90 "task-clock", 90 "task-clock,"
91 "{" 91 "{"
92 "instructions," 92 "instructions,"
93 "cycles," 93 "cycles,"
@@ -96,30 +96,12 @@ static const char * const transaction_limited_attrs[] = {
96 "}" 96 "}"
97}; 97};
98 98
99/* must match transaction_attrs and the beginning limited_attrs */
100enum {
101 T_TASK_CLOCK,
102 T_INSTRUCTIONS,
103 T_CYCLES,
104 T_CYCLES_IN_TX,
105 T_TRANSACTION_START,
106 T_ELISION_START,
107 T_CYCLES_IN_TX_CP,
108};
109
110static struct perf_evlist *evsel_list; 99static struct perf_evlist *evsel_list;
111 100
112static struct target target = { 101static struct target target = {
113 .uid = UINT_MAX, 102 .uid = UINT_MAX,
114}; 103};
115 104
116enum aggr_mode {
117 AGGR_NONE,
118 AGGR_GLOBAL,
119 AGGR_SOCKET,
120 AGGR_CORE,
121};
122
123static int run_count = 1; 105static int run_count = 1;
124static bool no_inherit = false; 106static bool no_inherit = false;
125static bool scale = true; 107static bool scale = true;
@@ -147,10 +129,6 @@ static int (*aggr_get_id)(struct cpu_map *m, int cpu);
147 129
148static volatile int done = 0; 130static volatile int done = 0;
149 131
150struct perf_stat {
151 struct stats res_stats[3];
152};
153
154static inline void diff_timespec(struct timespec *r, struct timespec *a, 132static inline void diff_timespec(struct timespec *r, struct timespec *a,
155 struct timespec *b) 133 struct timespec *b)
156{ 134{
@@ -180,6 +158,8 @@ static void perf_evsel__reset_stat_priv(struct perf_evsel *evsel)
180 158
181 for (i = 0; i < 3; i++) 159 for (i = 0; i < 3; i++)
182 init_stats(&ps->res_stats[i]); 160 init_stats(&ps->res_stats[i]);
161
162 perf_stat_evsel_id_init(evsel);
183} 163}
184 164
185static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel) 165static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel)
@@ -247,51 +227,6 @@ out_free:
247 return -1; 227 return -1;
248} 228}
249 229
250enum {
251 CTX_BIT_USER = 1 << 0,
252 CTX_BIT_KERNEL = 1 << 1,
253 CTX_BIT_HV = 1 << 2,
254 CTX_BIT_HOST = 1 << 3,
255 CTX_BIT_IDLE = 1 << 4,
256 CTX_BIT_MAX = 1 << 5,
257};
258
259#define NUM_CTX CTX_BIT_MAX
260
261static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
262static struct stats runtime_cycles_stats[NUM_CTX][MAX_NR_CPUS];
263static struct stats runtime_stalled_cycles_front_stats[NUM_CTX][MAX_NR_CPUS];
264static struct stats runtime_stalled_cycles_back_stats[NUM_CTX][MAX_NR_CPUS];
265static struct stats runtime_branches_stats[NUM_CTX][MAX_NR_CPUS];
266static struct stats runtime_cacherefs_stats[NUM_CTX][MAX_NR_CPUS];
267static struct stats runtime_l1_dcache_stats[NUM_CTX][MAX_NR_CPUS];
268static struct stats runtime_l1_icache_stats[NUM_CTX][MAX_NR_CPUS];
269static struct stats runtime_ll_cache_stats[NUM_CTX][MAX_NR_CPUS];
270static struct stats runtime_itlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
271static struct stats runtime_dtlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
272static struct stats runtime_cycles_in_tx_stats[NUM_CTX][MAX_NR_CPUS];
273static struct stats walltime_nsecs_stats;
274static struct stats runtime_transaction_stats[NUM_CTX][MAX_NR_CPUS];
275static struct stats runtime_elision_stats[NUM_CTX][MAX_NR_CPUS];
276
277static int evsel_context(struct perf_evsel *evsel)
278{
279 int ctx = 0;
280
281 if (evsel->attr.exclude_kernel)
282 ctx |= CTX_BIT_KERNEL;
283 if (evsel->attr.exclude_user)
284 ctx |= CTX_BIT_USER;
285 if (evsel->attr.exclude_hv)
286 ctx |= CTX_BIT_HV;
287 if (evsel->attr.exclude_host)
288 ctx |= CTX_BIT_HOST;
289 if (evsel->attr.exclude_idle)
290 ctx |= CTX_BIT_IDLE;
291
292 return ctx;
293}
294
295static void perf_stat__reset_stats(struct perf_evlist *evlist) 230static void perf_stat__reset_stats(struct perf_evlist *evlist)
296{ 231{
297 struct perf_evsel *evsel; 232 struct perf_evsel *evsel;
@@ -301,23 +236,7 @@ static void perf_stat__reset_stats(struct perf_evlist *evlist)
301 perf_evsel__reset_counts(evsel, perf_evsel__nr_cpus(evsel)); 236 perf_evsel__reset_counts(evsel, perf_evsel__nr_cpus(evsel));
302 } 237 }
303 238
304 memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats)); 239 perf_stat__reset_shadow_stats();
305 memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats));
306 memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats));
307 memset(runtime_stalled_cycles_back_stats, 0, sizeof(runtime_stalled_cycles_back_stats));
308 memset(runtime_branches_stats, 0, sizeof(runtime_branches_stats));
309 memset(runtime_cacherefs_stats, 0, sizeof(runtime_cacherefs_stats));
310 memset(runtime_l1_dcache_stats, 0, sizeof(runtime_l1_dcache_stats));
311 memset(runtime_l1_icache_stats, 0, sizeof(runtime_l1_icache_stats));
312 memset(runtime_ll_cache_stats, 0, sizeof(runtime_ll_cache_stats));
313 memset(runtime_itlb_cache_stats, 0, sizeof(runtime_itlb_cache_stats));
314 memset(runtime_dtlb_cache_stats, 0, sizeof(runtime_dtlb_cache_stats));
315 memset(runtime_cycles_in_tx_stats, 0,
316 sizeof(runtime_cycles_in_tx_stats));
317 memset(runtime_transaction_stats, 0,
318 sizeof(runtime_transaction_stats));
319 memset(runtime_elision_stats, 0, sizeof(runtime_elision_stats));
320 memset(&walltime_nsecs_stats, 0, sizeof(walltime_nsecs_stats));
321} 240}
322 241
323static int create_perf_stat_counter(struct perf_evsel *evsel) 242static int create_perf_stat_counter(struct perf_evsel *evsel)
@@ -354,72 +273,6 @@ static inline int nsec_counter(struct perf_evsel *evsel)
354 return 0; 273 return 0;
355} 274}
356 275
357static struct perf_evsel *nth_evsel(int n)
358{
359 static struct perf_evsel **array;
360 static int array_len;
361 struct perf_evsel *ev;
362 int j;
363
364 /* Assumes this only called when evsel_list does not change anymore. */
365 if (!array) {
366 evlist__for_each(evsel_list, ev)
367 array_len++;
368 array = malloc(array_len * sizeof(void *));
369 if (!array)
370 exit(ENOMEM);
371 j = 0;
372 evlist__for_each(evsel_list, ev)
373 array[j++] = ev;
374 }
375 if (n < array_len)
376 return array[n];
377 return NULL;
378}
379
380/*
381 * Update various tracking values we maintain to print
382 * more semantic information such as miss/hit ratios,
383 * instruction rates, etc:
384 */
385static void update_shadow_stats(struct perf_evsel *counter, u64 *count,
386 int cpu)
387{
388 int ctx = evsel_context(counter);
389
390 if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK))
391 update_stats(&runtime_nsecs_stats[cpu], count[0]);
392 else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
393 update_stats(&runtime_cycles_stats[ctx][cpu], count[0]);
394 else if (transaction_run &&
395 perf_evsel__cmp(counter, nth_evsel(T_CYCLES_IN_TX)))
396 update_stats(&runtime_transaction_stats[ctx][cpu], count[0]);
397 else if (transaction_run &&
398 perf_evsel__cmp(counter, nth_evsel(T_TRANSACTION_START)))
399 update_stats(&runtime_transaction_stats[ctx][cpu], count[0]);
400 else if (transaction_run &&
401 perf_evsel__cmp(counter, nth_evsel(T_ELISION_START)))
402 update_stats(&runtime_elision_stats[ctx][cpu], count[0]);
403 else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
404 update_stats(&runtime_stalled_cycles_front_stats[ctx][cpu], count[0]);
405 else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND))
406 update_stats(&runtime_stalled_cycles_back_stats[ctx][cpu], count[0]);
407 else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS))
408 update_stats(&runtime_branches_stats[ctx][cpu], count[0]);
409 else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES))
410 update_stats(&runtime_cacherefs_stats[ctx][cpu], count[0]);
411 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D))
412 update_stats(&runtime_l1_dcache_stats[ctx][cpu], count[0]);
413 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I))
414 update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
415 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL))
416 update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
417 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB))
418 update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count[0]);
419 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
420 update_stats(&runtime_itlb_cache_stats[ctx][cpu], count[0]);
421}
422
423static void zero_per_pkg(struct perf_evsel *counter) 276static void zero_per_pkg(struct perf_evsel *counter)
424{ 277{
425 if (counter->per_pkg_mask) 278 if (counter->per_pkg_mask)
@@ -480,7 +333,7 @@ static int read_cb(struct perf_evsel *evsel, int cpu, int thread __maybe_unused,
480 perf_counts_values__scale(count, scale, NULL); 333 perf_counts_values__scale(count, scale, NULL);
481 evsel->counts->cpu[cpu] = *count; 334 evsel->counts->cpu[cpu] = *count;
482 if (aggr_mode == AGGR_NONE) 335 if (aggr_mode == AGGR_NONE)
483 update_shadow_stats(evsel, count->values, cpu); 336 perf_stat__update_shadow_stats(evsel, count->values, cpu);
484 break; 337 break;
485 case AGGR_GLOBAL: 338 case AGGR_GLOBAL:
486 aggr->val += count->val; 339 aggr->val += count->val;
@@ -528,7 +381,7 @@ static int read_counter_aggr(struct perf_evsel *counter)
528 /* 381 /*
529 * Save the full runtime - to allow normalization during printout: 382 * Save the full runtime - to allow normalization during printout:
530 */ 383 */
531 update_shadow_stats(counter, count, 0); 384 perf_stat__update_shadow_stats(counter, count, 0);
532 385
533 return 0; 386 return 0;
534} 387}
@@ -906,200 +759,11 @@ static void nsec_printout(int id, int nr, struct perf_evsel *evsel, double avg)
906 fprintf(output, " "); 759 fprintf(output, " ");
907} 760}
908 761
909/* used for get_ratio_color() */
910enum grc_type {
911 GRC_STALLED_CYCLES_FE,
912 GRC_STALLED_CYCLES_BE,
913 GRC_CACHE_MISSES,
914 GRC_MAX_NR
915};
916
917static const char *get_ratio_color(enum grc_type type, double ratio)
918{
919 static const double grc_table[GRC_MAX_NR][3] = {
920 [GRC_STALLED_CYCLES_FE] = { 50.0, 30.0, 10.0 },
921 [GRC_STALLED_CYCLES_BE] = { 75.0, 50.0, 20.0 },
922 [GRC_CACHE_MISSES] = { 20.0, 10.0, 5.0 },
923 };
924 const char *color = PERF_COLOR_NORMAL;
925
926 if (ratio > grc_table[type][0])
927 color = PERF_COLOR_RED;
928 else if (ratio > grc_table[type][1])
929 color = PERF_COLOR_MAGENTA;
930 else if (ratio > grc_table[type][2])
931 color = PERF_COLOR_YELLOW;
932
933 return color;
934}
935
936static void print_stalled_cycles_frontend(int cpu,
937 struct perf_evsel *evsel
938 __maybe_unused, double avg)
939{
940 double total, ratio = 0.0;
941 const char *color;
942 int ctx = evsel_context(evsel);
943
944 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
945
946 if (total)
947 ratio = avg / total * 100.0;
948
949 color = get_ratio_color(GRC_STALLED_CYCLES_FE, ratio);
950
951 fprintf(output, " # ");
952 color_fprintf(output, color, "%6.2f%%", ratio);
953 fprintf(output, " frontend cycles idle ");
954}
955
956static void print_stalled_cycles_backend(int cpu,
957 struct perf_evsel *evsel
958 __maybe_unused, double avg)
959{
960 double total, ratio = 0.0;
961 const char *color;
962 int ctx = evsel_context(evsel);
963
964 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
965
966 if (total)
967 ratio = avg / total * 100.0;
968
969 color = get_ratio_color(GRC_STALLED_CYCLES_BE, ratio);
970
971 fprintf(output, " # ");
972 color_fprintf(output, color, "%6.2f%%", ratio);
973 fprintf(output, " backend cycles idle ");
974}
975
976static void print_branch_misses(int cpu,
977 struct perf_evsel *evsel __maybe_unused,
978 double avg)
979{
980 double total, ratio = 0.0;
981 const char *color;
982 int ctx = evsel_context(evsel);
983
984 total = avg_stats(&runtime_branches_stats[ctx][cpu]);
985
986 if (total)
987 ratio = avg / total * 100.0;
988
989 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
990
991 fprintf(output, " # ");
992 color_fprintf(output, color, "%6.2f%%", ratio);
993 fprintf(output, " of all branches ");
994}
995
996static void print_l1_dcache_misses(int cpu,
997 struct perf_evsel *evsel __maybe_unused,
998 double avg)
999{
1000 double total, ratio = 0.0;
1001 const char *color;
1002 int ctx = evsel_context(evsel);
1003
1004 total = avg_stats(&runtime_l1_dcache_stats[ctx][cpu]);
1005
1006 if (total)
1007 ratio = avg / total * 100.0;
1008
1009 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
1010
1011 fprintf(output, " # ");
1012 color_fprintf(output, color, "%6.2f%%", ratio);
1013 fprintf(output, " of all L1-dcache hits ");
1014}
1015
1016static void print_l1_icache_misses(int cpu,
1017 struct perf_evsel *evsel __maybe_unused,
1018 double avg)
1019{
1020 double total, ratio = 0.0;
1021 const char *color;
1022 int ctx = evsel_context(evsel);
1023
1024 total = avg_stats(&runtime_l1_icache_stats[ctx][cpu]);
1025
1026 if (total)
1027 ratio = avg / total * 100.0;
1028
1029 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
1030
1031 fprintf(output, " # ");
1032 color_fprintf(output, color, "%6.2f%%", ratio);
1033 fprintf(output, " of all L1-icache hits ");
1034}
1035
1036static void print_dtlb_cache_misses(int cpu,
1037 struct perf_evsel *evsel __maybe_unused,
1038 double avg)
1039{
1040 double total, ratio = 0.0;
1041 const char *color;
1042 int ctx = evsel_context(evsel);
1043
1044 total = avg_stats(&runtime_dtlb_cache_stats[ctx][cpu]);
1045
1046 if (total)
1047 ratio = avg / total * 100.0;
1048
1049 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
1050
1051 fprintf(output, " # ");
1052 color_fprintf(output, color, "%6.2f%%", ratio);
1053 fprintf(output, " of all dTLB cache hits ");
1054}
1055
1056static void print_itlb_cache_misses(int cpu,
1057 struct perf_evsel *evsel __maybe_unused,
1058 double avg)
1059{
1060 double total, ratio = 0.0;
1061 const char *color;
1062 int ctx = evsel_context(evsel);
1063
1064 total = avg_stats(&runtime_itlb_cache_stats[ctx][cpu]);
1065
1066 if (total)
1067 ratio = avg / total * 100.0;
1068
1069 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
1070
1071 fprintf(output, " # ");
1072 color_fprintf(output, color, "%6.2f%%", ratio);
1073 fprintf(output, " of all iTLB cache hits ");
1074}
1075
1076static void print_ll_cache_misses(int cpu,
1077 struct perf_evsel *evsel __maybe_unused,
1078 double avg)
1079{
1080 double total, ratio = 0.0;
1081 const char *color;
1082 int ctx = evsel_context(evsel);
1083
1084 total = avg_stats(&runtime_ll_cache_stats[ctx][cpu]);
1085
1086 if (total)
1087 ratio = avg / total * 100.0;
1088
1089 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
1090
1091 fprintf(output, " # ");
1092 color_fprintf(output, color, "%6.2f%%", ratio);
1093 fprintf(output, " of all LL-cache hits ");
1094}
1095
1096static void abs_printout(int id, int nr, struct perf_evsel *evsel, double avg) 762static void abs_printout(int id, int nr, struct perf_evsel *evsel, double avg)
1097{ 763{
1098 double total, ratio = 0.0, total2;
1099 double sc = evsel->scale; 764 double sc = evsel->scale;
1100 const char *fmt; 765 const char *fmt;
1101 int cpu = cpu_map__id_to_cpu(id); 766 int cpu = cpu_map__id_to_cpu(id);
1102 int ctx = evsel_context(evsel);
1103 767
1104 if (csv_output) { 768 if (csv_output) {
1105 fmt = sc != 1.0 ? "%.2f%s" : "%.0f%s"; 769 fmt = sc != 1.0 ? "%.2f%s" : "%.0f%s";
@@ -1130,138 +794,7 @@ static void abs_printout(int id, int nr, struct perf_evsel *evsel, double avg)
1130 if (csv_output || interval) 794 if (csv_output || interval)
1131 return; 795 return;
1132 796
1133 if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) { 797 perf_stat__print_shadow_stats(output, evsel, avg, cpu, aggr_mode);
1134 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
1135 if (total) {
1136 ratio = avg / total;
1137 fprintf(output, " # %5.2f insns per cycle ", ratio);
1138 } else {
1139 fprintf(output, " ");
1140 }
1141 total = avg_stats(&runtime_stalled_cycles_front_stats[ctx][cpu]);
1142 total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[ctx][cpu]));
1143
1144 if (total && avg) {
1145 ratio = total / avg;
1146 fprintf(output, "\n");
1147 if (aggr_mode == AGGR_NONE)
1148 fprintf(output, " ");
1149 fprintf(output, " # %5.2f stalled cycles per insn", ratio);
1150 }
1151
1152 } else if (perf_evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES) &&
1153 runtime_branches_stats[ctx][cpu].n != 0) {
1154 print_branch_misses(cpu, evsel, avg);
1155 } else if (
1156 evsel->attr.type == PERF_TYPE_HW_CACHE &&
1157 evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1D |
1158 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
1159 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
1160 runtime_l1_dcache_stats[ctx][cpu].n != 0) {
1161 print_l1_dcache_misses(cpu, evsel, avg);
1162 } else if (
1163 evsel->attr.type == PERF_TYPE_HW_CACHE &&
1164 evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1I |
1165 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
1166 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
1167 runtime_l1_icache_stats[ctx][cpu].n != 0) {
1168 print_l1_icache_misses(cpu, evsel, avg);
1169 } else if (
1170 evsel->attr.type == PERF_TYPE_HW_CACHE &&
1171 evsel->attr.config == ( PERF_COUNT_HW_CACHE_DTLB |
1172 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
1173 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
1174 runtime_dtlb_cache_stats[ctx][cpu].n != 0) {
1175 print_dtlb_cache_misses(cpu, evsel, avg);
1176 } else if (
1177 evsel->attr.type == PERF_TYPE_HW_CACHE &&
1178 evsel->attr.config == ( PERF_COUNT_HW_CACHE_ITLB |
1179 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
1180 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
1181 runtime_itlb_cache_stats[ctx][cpu].n != 0) {
1182 print_itlb_cache_misses(cpu, evsel, avg);
1183 } else if (
1184 evsel->attr.type == PERF_TYPE_HW_CACHE &&
1185 evsel->attr.config == ( PERF_COUNT_HW_CACHE_LL |
1186 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
1187 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
1188 runtime_ll_cache_stats[ctx][cpu].n != 0) {
1189 print_ll_cache_misses(cpu, evsel, avg);
1190 } else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES) &&
1191 runtime_cacherefs_stats[ctx][cpu].n != 0) {
1192 total = avg_stats(&runtime_cacherefs_stats[ctx][cpu]);
1193
1194 if (total)
1195 ratio = avg * 100 / total;
1196
1197 fprintf(output, " # %8.3f %% of all cache refs ", ratio);
1198
1199 } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) {
1200 print_stalled_cycles_frontend(cpu, evsel, avg);
1201 } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_BACKEND)) {
1202 print_stalled_cycles_backend(cpu, evsel, avg);
1203 } else if (perf_evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) {
1204 total = avg_stats(&runtime_nsecs_stats[cpu]);
1205
1206 if (total) {
1207 ratio = avg / total;
1208 fprintf(output, " # %8.3f GHz ", ratio);
1209 } else {
1210 fprintf(output, " ");
1211 }
1212 } else if (transaction_run &&
1213 perf_evsel__cmp(evsel, nth_evsel(T_CYCLES_IN_TX))) {
1214 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
1215 if (total)
1216 fprintf(output,
1217 " # %5.2f%% transactional cycles ",
1218 100.0 * (avg / total));
1219 } else if (transaction_run &&
1220 perf_evsel__cmp(evsel, nth_evsel(T_CYCLES_IN_TX_CP))) {
1221 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
1222 total2 = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
1223 if (total2 < avg)
1224 total2 = avg;
1225 if (total)
1226 fprintf(output,
1227 " # %5.2f%% aborted cycles ",
1228 100.0 * ((total2-avg) / total));
1229 } else if (transaction_run &&
1230 perf_evsel__cmp(evsel, nth_evsel(T_TRANSACTION_START)) &&
1231 avg > 0 &&
1232 runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
1233 total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
1234
1235 if (total)
1236 ratio = total / avg;
1237
1238 fprintf(output, " # %8.0f cycles / transaction ", ratio);
1239 } else if (transaction_run &&
1240 perf_evsel__cmp(evsel, nth_evsel(T_ELISION_START)) &&
1241 avg > 0 &&
1242 runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
1243 total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
1244
1245 if (total)
1246 ratio = total / avg;
1247
1248 fprintf(output, " # %8.0f cycles / elision ", ratio);
1249 } else if (runtime_nsecs_stats[cpu].n != 0) {
1250 char unit = 'M';
1251
1252 total = avg_stats(&runtime_nsecs_stats[cpu]);
1253
1254 if (total)
1255 ratio = 1000.0 * avg / total;
1256 if (ratio < 0.001) {
1257 ratio *= 1000;
1258 unit = 'K';
1259 }
1260
1261 fprintf(output, " # %8.3f %c/sec ", ratio, unit);
1262 } else {
1263 fprintf(output, " ");
1264 }
1265} 798}
1266 799
1267static void print_aggr(char *prefix) 800static void print_aggr(char *prefix)
@@ -1576,17 +1109,6 @@ static int perf_stat_init_aggr_mode(void)
1576 return 0; 1109 return 0;
1577} 1110}
1578 1111
1579static int setup_events(const char * const *attrs, unsigned len)
1580{
1581 unsigned i;
1582
1583 for (i = 0; i < len; i++) {
1584 if (parse_events(evsel_list, attrs[i], NULL))
1585 return -1;
1586 }
1587 return 0;
1588}
1589
1590/* 1112/*
1591 * Add default attributes, if there were no attributes specified or 1113 * Add default attributes, if there were no attributes specified or
1592 * if -d/--detailed, -d -d or -d -d -d is used: 1114 * if -d/--detailed, -d -d or -d -d -d is used:
@@ -1708,12 +1230,10 @@ static int add_default_attributes(void)
1708 int err; 1230 int err;
1709 if (pmu_have_event("cpu", "cycles-ct") && 1231 if (pmu_have_event("cpu", "cycles-ct") &&
1710 pmu_have_event("cpu", "el-start")) 1232 pmu_have_event("cpu", "el-start"))
1711 err = setup_events(transaction_attrs, 1233 err = parse_events(evsel_list, transaction_attrs, NULL);
1712 ARRAY_SIZE(transaction_attrs));
1713 else 1234 else
1714 err = setup_events(transaction_limited_attrs, 1235 err = parse_events(evsel_list, transaction_limited_attrs, NULL);
1715 ARRAY_SIZE(transaction_limited_attrs)); 1236 if (err) {
1716 if (err < 0) {
1717 fprintf(stderr, "Cannot set up transaction events\n"); 1237 fprintf(stderr, "Cannot set up transaction events\n");
1718 return -1; 1238 return -1;
1719 } 1239 }
diff --git a/tools/perf/tests/dso-data.c b/tools/perf/tests/dso-data.c
index 3e41c61bd861..a218aeaf56a0 100644
--- a/tools/perf/tests/dso-data.c
+++ b/tools/perf/tests/dso-data.c
@@ -166,7 +166,7 @@ int test__dso_data(void)
166 free(buf); 166 free(buf);
167 } 167 }
168 168
169 dso__delete(dso); 169 dso__put(dso);
170 unlink(file); 170 unlink(file);
171 return 0; 171 return 0;
172} 172}
@@ -226,7 +226,7 @@ static void dsos__delete(int cnt)
226 struct dso *dso = dsos[i]; 226 struct dso *dso = dsos[i];
227 227
228 unlink(dso->name); 228 unlink(dso->name);
229 dso__delete(dso); 229 dso__put(dso);
230 } 230 }
231 231
232 free(dsos); 232 free(dsos);
diff --git a/tools/perf/tests/hists_common.c b/tools/perf/tests/hists_common.c
index 915f60af6a0e..ce80b274b097 100644
--- a/tools/perf/tests/hists_common.c
+++ b/tools/perf/tests/hists_common.c
@@ -134,11 +134,15 @@ struct machine *setup_fake_machine(struct machines *machines)
134 134
135 sym = symbol__new(fsym->start, fsym->length, 135 sym = symbol__new(fsym->start, fsym->length,
136 STB_GLOBAL, fsym->name); 136 STB_GLOBAL, fsym->name);
137 if (sym == NULL) 137 if (sym == NULL) {
138 dso__put(dso);
138 goto out; 139 goto out;
140 }
139 141
140 symbols__insert(&dso->symbols[MAP__FUNCTION], sym); 142 symbols__insert(&dso->symbols[MAP__FUNCTION], sym);
141 } 143 }
144
145 dso__put(dso);
142 } 146 }
143 147
144 return machine; 148 return machine;
diff --git a/tools/perf/util/Build b/tools/perf/util/Build
index e4b676de2f64..586a59d46022 100644
--- a/tools/perf/util/Build
+++ b/tools/perf/util/Build
@@ -68,6 +68,7 @@ libperf-y += rblist.o
68libperf-y += intlist.o 68libperf-y += intlist.o
69libperf-y += vdso.o 69libperf-y += vdso.o
70libperf-y += stat.o 70libperf-y += stat.o
71libperf-y += stat-shadow.o
71libperf-y += record.o 72libperf-y += record.o
72libperf-y += srcline.o 73libperf-y += srcline.o
73libperf-y += data.o 74libperf-y += data.o
diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index 5ec9e892c89b..7c0c08386a1d 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -889,8 +889,8 @@ struct dso *machine__findnew_kernel(struct machine *machine, const char *name,
889 * Either one of the dso or name parameter must be non-NULL or the 889 * Either one of the dso or name parameter must be non-NULL or the
890 * function will not work. 890 * function will not work.
891 */ 891 */
892static struct dso *dso__findlink_by_longname(struct rb_root *root, 892static struct dso *__dso__findlink_by_longname(struct rb_root *root,
893 struct dso *dso, const char *name) 893 struct dso *dso, const char *name)
894{ 894{
895 struct rb_node **p = &root->rb_node; 895 struct rb_node **p = &root->rb_node;
896 struct rb_node *parent = NULL; 896 struct rb_node *parent = NULL;
@@ -937,10 +937,10 @@ static struct dso *dso__findlink_by_longname(struct rb_root *root,
937 return NULL; 937 return NULL;
938} 938}
939 939
940static inline struct dso * 940static inline struct dso *__dso__find_by_longname(struct rb_root *root,
941dso__find_by_longname(const struct rb_root *root, const char *name) 941 const char *name)
942{ 942{
943 return dso__findlink_by_longname((struct rb_root *)root, NULL, name); 943 return __dso__findlink_by_longname(root, NULL, name);
944} 944}
945 945
946void dso__set_long_name(struct dso *dso, const char *name, bool name_allocated) 946void dso__set_long_name(struct dso *dso, const char *name, bool name_allocated)
@@ -1049,6 +1049,7 @@ struct dso *dso__new(const char *name)
1049 INIT_LIST_HEAD(&dso->node); 1049 INIT_LIST_HEAD(&dso->node);
1050 INIT_LIST_HEAD(&dso->data.open_entry); 1050 INIT_LIST_HEAD(&dso->data.open_entry);
1051 pthread_mutex_init(&dso->lock, NULL); 1051 pthread_mutex_init(&dso->lock, NULL);
1052 atomic_set(&dso->refcnt, 1);
1052 } 1053 }
1053 1054
1054 return dso; 1055 return dso;
@@ -1083,6 +1084,19 @@ void dso__delete(struct dso *dso)
1083 free(dso); 1084 free(dso);
1084} 1085}
1085 1086
1087struct dso *dso__get(struct dso *dso)
1088{
1089 if (dso)
1090 atomic_inc(&dso->refcnt);
1091 return dso;
1092}
1093
1094void dso__put(struct dso *dso)
1095{
1096 if (dso && atomic_dec_and_test(&dso->refcnt))
1097 dso__delete(dso);
1098}
1099
1086void dso__set_build_id(struct dso *dso, void *build_id) 1100void dso__set_build_id(struct dso *dso, void *build_id)
1087{ 1101{
1088 memcpy(dso->build_id, build_id, sizeof(dso->build_id)); 1102 memcpy(dso->build_id, build_id, sizeof(dso->build_id));
@@ -1149,14 +1163,41 @@ bool __dsos__read_build_ids(struct list_head *head, bool with_hits)
1149 return have_build_id; 1163 return have_build_id;
1150} 1164}
1151 1165
1152void dsos__add(struct dsos *dsos, struct dso *dso) 1166void __dsos__add(struct dsos *dsos, struct dso *dso)
1153{ 1167{
1154 list_add_tail(&dso->node, &dsos->head); 1168 list_add_tail(&dso->node, &dsos->head);
1155 dso__findlink_by_longname(&dsos->root, dso, NULL); 1169 __dso__findlink_by_longname(&dsos->root, dso, NULL);
1170 /*
1171 * It is now in the linked list, grab a reference, then garbage collect
1172 * this when needing memory, by looking at LRU dso instances in the
1173 * list with atomic_read(&dso->refcnt) == 1, i.e. no references
1174 * anywhere besides the one for the list, do, under a lock for the
1175 * list: remove it from the list, then a dso__put(), that probably will
1176 * be the last and will then call dso__delete(), end of life.
1177 *
1178 * That, or at the end of the 'struct machine' lifetime, when all
1179 * 'struct dso' instances will be removed from the list, in
1180 * dsos__exit(), if they have no other reference from some other data
1181 * structure.
1182 *
1183 * E.g.: after processing a 'perf.data' file and storing references
1184 * to objects instantiated while processing events, we will have
1185 * references to the 'thread', 'map', 'dso' structs all from 'struct
1186 * hist_entry' instances, but we may not need anything not referenced,
1187 * so we might as well call machines__exit()/machines__delete() and
1188 * garbage collect it.
1189 */
1190 dso__get(dso);
1156} 1191}
1157 1192
1158struct dso *dsos__find(const struct dsos *dsos, const char *name, 1193void dsos__add(struct dsos *dsos, struct dso *dso)
1159 bool cmp_short) 1194{
1195 pthread_rwlock_wrlock(&dsos->lock);
1196 __dsos__add(dsos, dso);
1197 pthread_rwlock_unlock(&dsos->lock);
1198}
1199
1200struct dso *__dsos__find(struct dsos *dsos, const char *name, bool cmp_short)
1160{ 1201{
1161 struct dso *pos; 1202 struct dso *pos;
1162 1203
@@ -1166,15 +1207,24 @@ struct dso *dsos__find(const struct dsos *dsos, const char *name,
1166 return pos; 1207 return pos;
1167 return NULL; 1208 return NULL;
1168 } 1209 }
1169 return dso__find_by_longname(&dsos->root, name); 1210 return __dso__find_by_longname(&dsos->root, name);
1170} 1211}
1171 1212
1172struct dso *dsos__addnew(struct dsos *dsos, const char *name) 1213struct dso *dsos__find(struct dsos *dsos, const char *name, bool cmp_short)
1214{
1215 struct dso *dso;
1216 pthread_rwlock_rdlock(&dsos->lock);
1217 dso = __dsos__find(dsos, name, cmp_short);
1218 pthread_rwlock_unlock(&dsos->lock);
1219 return dso;
1220}
1221
1222struct dso *__dsos__addnew(struct dsos *dsos, const char *name)
1173{ 1223{
1174 struct dso *dso = dso__new(name); 1224 struct dso *dso = dso__new(name);
1175 1225
1176 if (dso != NULL) { 1226 if (dso != NULL) {
1177 dsos__add(dsos, dso); 1227 __dsos__add(dsos, dso);
1178 dso__set_basename(dso); 1228 dso__set_basename(dso);
1179 } 1229 }
1180 return dso; 1230 return dso;
@@ -1182,9 +1232,18 @@ struct dso *dsos__addnew(struct dsos *dsos, const char *name)
1182 1232
1183struct dso *__dsos__findnew(struct dsos *dsos, const char *name) 1233struct dso *__dsos__findnew(struct dsos *dsos, const char *name)
1184{ 1234{
1185 struct dso *dso = dsos__find(dsos, name, false); 1235 struct dso *dso = __dsos__find(dsos, name, false);
1236
1237 return dso ? dso : __dsos__addnew(dsos, name);
1238}
1186 1239
1187 return dso ? dso : dsos__addnew(dsos, name); 1240struct dso *dsos__findnew(struct dsos *dsos, const char *name)
1241{
1242 struct dso *dso;
1243 pthread_rwlock_wrlock(&dsos->lock);
1244 dso = dso__get(__dsos__findnew(dsos, name));
1245 pthread_rwlock_unlock(&dsos->lock);
1246 return dso;
1188} 1247}
1189 1248
1190size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp, 1249size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp,
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index ba2d90ed881f..2fe98bb0e95b 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -1,9 +1,11 @@
1#ifndef __PERF_DSO 1#ifndef __PERF_DSO
2#define __PERF_DSO 2#define __PERF_DSO
3 3
4#include <linux/atomic.h>
4#include <linux/types.h> 5#include <linux/types.h>
5#include <linux/rbtree.h> 6#include <linux/rbtree.h>
6#include <stdbool.h> 7#include <stdbool.h>
8#include <pthread.h>
7#include <linux/types.h> 9#include <linux/types.h>
8#include <linux/bitops.h> 10#include <linux/bitops.h>
9#include "map.h" 11#include "map.h"
@@ -124,6 +126,7 @@ struct dso_cache {
124struct dsos { 126struct dsos {
125 struct list_head head; 127 struct list_head head;
126 struct rb_root root; /* rbtree root sorted by long name */ 128 struct rb_root root; /* rbtree root sorted by long name */
129 pthread_rwlock_t lock;
127}; 130};
128 131
129struct auxtrace_cache; 132struct auxtrace_cache;
@@ -177,7 +180,7 @@ struct dso {
177 void *priv; 180 void *priv;
178 u64 db_id; 181 u64 db_id;
179 }; 182 };
180 183 atomic_t refcnt;
181 char name[0]; 184 char name[0];
182}; 185};
183 186
@@ -204,6 +207,17 @@ void dso__set_long_name(struct dso *dso, const char *name, bool name_allocated);
204 207
205int dso__name_len(const struct dso *dso); 208int dso__name_len(const struct dso *dso);
206 209
210struct dso *dso__get(struct dso *dso);
211void dso__put(struct dso *dso);
212
213static inline void __dso__zput(struct dso **dso)
214{
215 dso__put(*dso);
216 *dso = NULL;
217}
218
219#define dso__zput(dso) __dso__zput(&dso)
220
207bool dso__loaded(const struct dso *dso, enum map_type type); 221bool dso__loaded(const struct dso *dso, enum map_type type);
208 222
209bool dso__sorted_by_name(const struct dso *dso, enum map_type type); 223bool dso__sorted_by_name(const struct dso *dso, enum map_type type);
@@ -297,11 +311,13 @@ struct map *dso__new_map(const char *name);
297struct dso *machine__findnew_kernel(struct machine *machine, const char *name, 311struct dso *machine__findnew_kernel(struct machine *machine, const char *name,
298 const char *short_name, int dso_type); 312 const char *short_name, int dso_type);
299 313
314void __dsos__add(struct dsos *dsos, struct dso *dso);
300void dsos__add(struct dsos *dsos, struct dso *dso); 315void dsos__add(struct dsos *dsos, struct dso *dso);
301struct dso *dsos__addnew(struct dsos *dsos, const char *name); 316struct dso *__dsos__addnew(struct dsos *dsos, const char *name);
302struct dso *dsos__find(const struct dsos *dsos, const char *name, 317struct dso *__dsos__find(struct dsos *dsos, const char *name, bool cmp_short);
303 bool cmp_short); 318struct dso *dsos__find(struct dsos *dsos, const char *name, bool cmp_short);
304struct dso *__dsos__findnew(struct dsos *dsos, const char *name); 319struct dso *__dsos__findnew(struct dsos *dsos, const char *name);
320struct dso *dsos__findnew(struct dsos *dsos, const char *name);
305bool __dsos__read_build_ids(struct list_head *head, bool with_hits); 321bool __dsos__read_build_ids(struct list_head *head, bool with_hits);
306 322
307size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp, 323size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp,
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index ac5aaaeed7ff..21a77e7a171e 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1277,6 +1277,7 @@ static int __event_process_build_id(struct build_id_event *bev,
1277 sbuild_id); 1277 sbuild_id);
1278 pr_debug("build id event received for %s: %s\n", 1278 pr_debug("build id event received for %s: %s\n",
1279 dso->long_name, sbuild_id); 1279 dso->long_name, sbuild_id);
1280 dso__put(dso);
1280 } 1281 }
1281 1282
1282 err = 0; 1283 err = 0;
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index f15ed24a22ac..132e35765101 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -20,6 +20,7 @@ static void dsos__init(struct dsos *dsos)
20{ 20{
21 INIT_LIST_HEAD(&dsos->head); 21 INIT_LIST_HEAD(&dsos->head);
22 dsos->root = RB_ROOT; 22 dsos->root = RB_ROOT;
23 pthread_rwlock_init(&dsos->lock, NULL);
23} 24}
24 25
25int machine__init(struct machine *machine, const char *root_dir, pid_t pid) 26int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
@@ -81,15 +82,25 @@ out_delete:
81 return NULL; 82 return NULL;
82} 83}
83 84
84static void dsos__delete(struct dsos *dsos) 85static void dsos__purge(struct dsos *dsos)
85{ 86{
86 struct dso *pos, *n; 87 struct dso *pos, *n;
87 88
89 pthread_rwlock_wrlock(&dsos->lock);
90
88 list_for_each_entry_safe(pos, n, &dsos->head, node) { 91 list_for_each_entry_safe(pos, n, &dsos->head, node) {
89 RB_CLEAR_NODE(&pos->rb_node); 92 RB_CLEAR_NODE(&pos->rb_node);
90 list_del(&pos->node); 93 list_del_init(&pos->node);
91 dso__delete(pos); 94 dso__put(pos);
92 } 95 }
96
97 pthread_rwlock_unlock(&dsos->lock);
98}
99
100static void dsos__exit(struct dsos *dsos)
101{
102 dsos__purge(dsos);
103 pthread_rwlock_destroy(&dsos->lock);
93} 104}
94 105
95void machine__delete_threads(struct machine *machine) 106void machine__delete_threads(struct machine *machine)
@@ -110,7 +121,7 @@ void machine__delete_threads(struct machine *machine)
110void machine__exit(struct machine *machine) 121void machine__exit(struct machine *machine)
111{ 122{
112 map_groups__exit(&machine->kmaps); 123 map_groups__exit(&machine->kmaps);
113 dsos__delete(&machine->dsos); 124 dsos__exit(&machine->dsos);
114 machine__exit_vdso(machine); 125 machine__exit_vdso(machine);
115 zfree(&machine->root_dir); 126 zfree(&machine->root_dir);
116 zfree(&machine->current_tid); 127 zfree(&machine->current_tid);
@@ -490,17 +501,19 @@ int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
490 return 0; 501 return 0;
491} 502}
492 503
493static struct dso* 504static struct dso *machine__findnew_module_dso(struct machine *machine,
494machine__module_dso(struct machine *machine, struct kmod_path *m, 505 struct kmod_path *m,
495 const char *filename) 506 const char *filename)
496{ 507{
497 struct dso *dso; 508 struct dso *dso;
498 509
499 dso = dsos__find(&machine->dsos, m->name, true); 510 pthread_rwlock_wrlock(&machine->dsos.lock);
511
512 dso = __dsos__find(&machine->dsos, m->name, true);
500 if (!dso) { 513 if (!dso) {
501 dso = dsos__addnew(&machine->dsos, m->name); 514 dso = __dsos__addnew(&machine->dsos, m->name);
502 if (dso == NULL) 515 if (dso == NULL)
503 return NULL; 516 goto out_unlock;
504 517
505 if (machine__is_host(machine)) 518 if (machine__is_host(machine))
506 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE; 519 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
@@ -515,6 +528,9 @@ machine__module_dso(struct machine *machine, struct kmod_path *m,
515 dso__set_long_name(dso, strdup(filename), true); 528 dso__set_long_name(dso, strdup(filename), true);
516 } 529 }
517 530
531 dso__get(dso);
532out_unlock:
533 pthread_rwlock_unlock(&machine->dsos.lock);
518 return dso; 534 return dso;
519} 535}
520 536
@@ -534,8 +550,8 @@ int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
534 return 0; 550 return 0;
535} 551}
536 552
537struct map *machine__new_module(struct machine *machine, u64 start, 553struct map *machine__findnew_module_map(struct machine *machine, u64 start,
538 const char *filename) 554 const char *filename)
539{ 555{
540 struct map *map = NULL; 556 struct map *map = NULL;
541 struct dso *dso; 557 struct dso *dso;
@@ -549,7 +565,7 @@ struct map *machine__new_module(struct machine *machine, u64 start,
549 if (map) 565 if (map)
550 goto out; 566 goto out;
551 567
552 dso = machine__module_dso(machine, &m, filename); 568 dso = machine__findnew_module_dso(machine, &m, filename);
553 if (dso == NULL) 569 if (dso == NULL)
554 goto out; 570 goto out;
555 571
@@ -1017,7 +1033,7 @@ static int machine__create_module(void *arg, const char *name, u64 start)
1017 struct machine *machine = arg; 1033 struct machine *machine = arg;
1018 struct map *map; 1034 struct map *map;
1019 1035
1020 map = machine__new_module(machine, start, name); 1036 map = machine__findnew_module_map(machine, start, name);
1021 if (map == NULL) 1037 if (map == NULL)
1022 return -1; 1038 return -1;
1023 1039
@@ -1140,8 +1156,8 @@ static int machine__process_kernel_mmap_event(struct machine *machine,
1140 strlen(kmmap_prefix) - 1) == 0; 1156 strlen(kmmap_prefix) - 1) == 0;
1141 if (event->mmap.filename[0] == '/' || 1157 if (event->mmap.filename[0] == '/' ||
1142 (!is_kernel_mmap && event->mmap.filename[0] == '[')) { 1158 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1143 map = machine__new_module(machine, event->mmap.start, 1159 map = machine__findnew_module_map(machine, event->mmap.start,
1144 event->mmap.filename); 1160 event->mmap.filename);
1145 if (map == NULL) 1161 if (map == NULL)
1146 goto out_problem; 1162 goto out_problem;
1147 1163
@@ -1156,6 +1172,8 @@ static int machine__process_kernel_mmap_event(struct machine *machine,
1156 struct dso *kernel = NULL; 1172 struct dso *kernel = NULL;
1157 struct dso *dso; 1173 struct dso *dso;
1158 1174
1175 pthread_rwlock_rdlock(&machine->dsos.lock);
1176
1159 list_for_each_entry(dso, &machine->dsos.head, node) { 1177 list_for_each_entry(dso, &machine->dsos.head, node) {
1160 1178
1161 /* 1179 /*
@@ -1184,14 +1202,18 @@ static int machine__process_kernel_mmap_event(struct machine *machine,
1184 break; 1202 break;
1185 } 1203 }
1186 1204
1205 pthread_rwlock_unlock(&machine->dsos.lock);
1206
1187 if (kernel == NULL) 1207 if (kernel == NULL)
1188 kernel = machine__findnew_dso(machine, kmmap_prefix); 1208 kernel = machine__findnew_dso(machine, kmmap_prefix);
1189 if (kernel == NULL) 1209 if (kernel == NULL)
1190 goto out_problem; 1210 goto out_problem;
1191 1211
1192 kernel->kernel = kernel_type; 1212 kernel->kernel = kernel_type;
1193 if (__machine__create_kernel_maps(machine, kernel) < 0) 1213 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1214 dso__put(kernel);
1194 goto out_problem; 1215 goto out_problem;
1216 }
1195 1217
1196 if (strstr(kernel->long_name, "vmlinux")) 1218 if (strstr(kernel->long_name, "vmlinux"))
1197 dso__set_short_name(kernel, "[kernel.vmlinux]", false); 1219 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
@@ -1948,5 +1970,5 @@ int machine__get_kernel_start(struct machine *machine)
1948 1970
1949struct dso *machine__findnew_dso(struct machine *machine, const char *filename) 1971struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
1950{ 1972{
1951 return __dsos__findnew(&machine->dsos, filename); 1973 return dsos__findnew(&machine->dsos, filename);
1952} 1974}
diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
index 8e1f796fd137..ca267c41f28d 100644
--- a/tools/perf/util/machine.h
+++ b/tools/perf/util/machine.h
@@ -189,8 +189,8 @@ struct symbol *machine__find_kernel_function_by_name(struct machine *machine,
189 filter); 189 filter);
190} 190}
191 191
192struct map *machine__new_module(struct machine *machine, u64 start, 192struct map *machine__findnew_module_map(struct machine *machine, u64 start,
193 const char *filename); 193 const char *filename);
194 194
195int machine__load_kallsyms(struct machine *machine, const char *filename, 195int machine__load_kallsyms(struct machine *machine, const char *filename,
196 enum map_type type, symbol_filter_t filter); 196 enum map_type type, symbol_filter_t filter);
diff --git a/tools/perf/util/map.c b/tools/perf/util/map.c
index 365011c233a6..1241ab989cf5 100644
--- a/tools/perf/util/map.c
+++ b/tools/perf/util/map.c
@@ -132,7 +132,7 @@ void map__init(struct map *map, enum map_type type,
132 map->end = end; 132 map->end = end;
133 map->pgoff = pgoff; 133 map->pgoff = pgoff;
134 map->reloc = 0; 134 map->reloc = 0;
135 map->dso = dso; 135 map->dso = dso__get(dso);
136 map->map_ip = map__map_ip; 136 map->map_ip = map__map_ip;
137 map->unmap_ip = map__unmap_ip; 137 map->unmap_ip = map__unmap_ip;
138 RB_CLEAR_NODE(&map->rb_node); 138 RB_CLEAR_NODE(&map->rb_node);
@@ -198,6 +198,7 @@ struct map *map__new(struct machine *machine, u64 start, u64 len,
198 if (type != MAP__FUNCTION) 198 if (type != MAP__FUNCTION)
199 dso__set_loaded(dso, map->type); 199 dso__set_loaded(dso, map->type);
200 } 200 }
201 dso__put(dso);
201 } 202 }
202 return map; 203 return map;
203out_delete: 204out_delete:
@@ -224,9 +225,15 @@ struct map *map__new2(u64 start, struct dso *dso, enum map_type type)
224 return map; 225 return map;
225} 226}
226 227
227void map__delete(struct map *map) 228static void map__exit(struct map *map)
228{ 229{
229 BUG_ON(!RB_EMPTY_NODE(&map->rb_node)); 230 BUG_ON(!RB_EMPTY_NODE(&map->rb_node));
231 dso__zput(map->dso);
232}
233
234void map__delete(struct map *map)
235{
236 map__exit(map);
230 free(map); 237 free(map);
231} 238}
232 239
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index e6f215b7a052..d4cf50b91839 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -168,7 +168,7 @@ static struct map *kernel_get_module_map(const char *module)
168 168
169 /* A file path -- this is an offline module */ 169 /* A file path -- this is an offline module */
170 if (module && strchr(module, '/')) 170 if (module && strchr(module, '/'))
171 return machine__new_module(host_machine, 0, module); 171 return machine__findnew_module_map(host_machine, 0, module);
172 172
173 if (!module) 173 if (!module)
174 module = "kernel"; 174 module = "kernel";
diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
index c50da392e256..2da65a710893 100644
--- a/tools/perf/util/probe-finder.c
+++ b/tools/perf/util/probe-finder.c
@@ -130,7 +130,7 @@ struct debuginfo *debuginfo__new(const char *path)
130 continue; 130 continue;
131 dinfo = __debuginfo__new(buf); 131 dinfo = __debuginfo__new(buf);
132 } 132 }
133 dso__delete(dso); 133 dso__put(dso);
134 134
135out: 135out:
136 /* if failed to open all distro debuginfo, open given binary */ 136 /* if failed to open all distro debuginfo, open given binary */
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
new file mode 100644
index 000000000000..53e8bb7bc852
--- /dev/null
+++ b/tools/perf/util/stat-shadow.c
@@ -0,0 +1,434 @@
1#include <stdio.h>
2#include "evsel.h"
3#include "stat.h"
4#include "color.h"
5
6enum {
7 CTX_BIT_USER = 1 << 0,
8 CTX_BIT_KERNEL = 1 << 1,
9 CTX_BIT_HV = 1 << 2,
10 CTX_BIT_HOST = 1 << 3,
11 CTX_BIT_IDLE = 1 << 4,
12 CTX_BIT_MAX = 1 << 5,
13};
14
15#define NUM_CTX CTX_BIT_MAX
16
17static struct stats runtime_nsecs_stats[MAX_NR_CPUS];
18static struct stats runtime_cycles_stats[NUM_CTX][MAX_NR_CPUS];
19static struct stats runtime_stalled_cycles_front_stats[NUM_CTX][MAX_NR_CPUS];
20static struct stats runtime_stalled_cycles_back_stats[NUM_CTX][MAX_NR_CPUS];
21static struct stats runtime_branches_stats[NUM_CTX][MAX_NR_CPUS];
22static struct stats runtime_cacherefs_stats[NUM_CTX][MAX_NR_CPUS];
23static struct stats runtime_l1_dcache_stats[NUM_CTX][MAX_NR_CPUS];
24static struct stats runtime_l1_icache_stats[NUM_CTX][MAX_NR_CPUS];
25static struct stats runtime_ll_cache_stats[NUM_CTX][MAX_NR_CPUS];
26static struct stats runtime_itlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
27static struct stats runtime_dtlb_cache_stats[NUM_CTX][MAX_NR_CPUS];
28static struct stats runtime_cycles_in_tx_stats[NUM_CTX][MAX_NR_CPUS];
29static struct stats runtime_transaction_stats[NUM_CTX][MAX_NR_CPUS];
30static struct stats runtime_elision_stats[NUM_CTX][MAX_NR_CPUS];
31
32struct stats walltime_nsecs_stats;
33
34static int evsel_context(struct perf_evsel *evsel)
35{
36 int ctx = 0;
37
38 if (evsel->attr.exclude_kernel)
39 ctx |= CTX_BIT_KERNEL;
40 if (evsel->attr.exclude_user)
41 ctx |= CTX_BIT_USER;
42 if (evsel->attr.exclude_hv)
43 ctx |= CTX_BIT_HV;
44 if (evsel->attr.exclude_host)
45 ctx |= CTX_BIT_HOST;
46 if (evsel->attr.exclude_idle)
47 ctx |= CTX_BIT_IDLE;
48
49 return ctx;
50}
51
52void perf_stat__reset_shadow_stats(void)
53{
54 memset(runtime_nsecs_stats, 0, sizeof(runtime_nsecs_stats));
55 memset(runtime_cycles_stats, 0, sizeof(runtime_cycles_stats));
56 memset(runtime_stalled_cycles_front_stats, 0, sizeof(runtime_stalled_cycles_front_stats));
57 memset(runtime_stalled_cycles_back_stats, 0, sizeof(runtime_stalled_cycles_back_stats));
58 memset(runtime_branches_stats, 0, sizeof(runtime_branches_stats));
59 memset(runtime_cacherefs_stats, 0, sizeof(runtime_cacherefs_stats));
60 memset(runtime_l1_dcache_stats, 0, sizeof(runtime_l1_dcache_stats));
61 memset(runtime_l1_icache_stats, 0, sizeof(runtime_l1_icache_stats));
62 memset(runtime_ll_cache_stats, 0, sizeof(runtime_ll_cache_stats));
63 memset(runtime_itlb_cache_stats, 0, sizeof(runtime_itlb_cache_stats));
64 memset(runtime_dtlb_cache_stats, 0, sizeof(runtime_dtlb_cache_stats));
65 memset(runtime_cycles_in_tx_stats, 0,
66 sizeof(runtime_cycles_in_tx_stats));
67 memset(runtime_transaction_stats, 0,
68 sizeof(runtime_transaction_stats));
69 memset(runtime_elision_stats, 0, sizeof(runtime_elision_stats));
70 memset(&walltime_nsecs_stats, 0, sizeof(walltime_nsecs_stats));
71}
72
73/*
74 * Update various tracking values we maintain to print
75 * more semantic information such as miss/hit ratios,
76 * instruction rates, etc:
77 */
78void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
79 int cpu)
80{
81 int ctx = evsel_context(counter);
82
83 if (perf_evsel__match(counter, SOFTWARE, SW_TASK_CLOCK))
84 update_stats(&runtime_nsecs_stats[cpu], count[0]);
85 else if (perf_evsel__match(counter, HARDWARE, HW_CPU_CYCLES))
86 update_stats(&runtime_cycles_stats[ctx][cpu], count[0]);
87 else if (perf_stat_evsel__is(counter, CYCLES_IN_TX))
88 update_stats(&runtime_transaction_stats[ctx][cpu], count[0]);
89 else if (perf_stat_evsel__is(counter, TRANSACTION_START))
90 update_stats(&runtime_transaction_stats[ctx][cpu], count[0]);
91 else if (perf_stat_evsel__is(counter, ELISION_START))
92 update_stats(&runtime_elision_stats[ctx][cpu], count[0]);
93 else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_FRONTEND))
94 update_stats(&runtime_stalled_cycles_front_stats[ctx][cpu], count[0]);
95 else if (perf_evsel__match(counter, HARDWARE, HW_STALLED_CYCLES_BACKEND))
96 update_stats(&runtime_stalled_cycles_back_stats[ctx][cpu], count[0]);
97 else if (perf_evsel__match(counter, HARDWARE, HW_BRANCH_INSTRUCTIONS))
98 update_stats(&runtime_branches_stats[ctx][cpu], count[0]);
99 else if (perf_evsel__match(counter, HARDWARE, HW_CACHE_REFERENCES))
100 update_stats(&runtime_cacherefs_stats[ctx][cpu], count[0]);
101 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1D))
102 update_stats(&runtime_l1_dcache_stats[ctx][cpu], count[0]);
103 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_L1I))
104 update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
105 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_LL))
106 update_stats(&runtime_ll_cache_stats[ctx][cpu], count[0]);
107 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_DTLB))
108 update_stats(&runtime_dtlb_cache_stats[ctx][cpu], count[0]);
109 else if (perf_evsel__match(counter, HW_CACHE, HW_CACHE_ITLB))
110 update_stats(&runtime_itlb_cache_stats[ctx][cpu], count[0]);
111}
112
113/* used for get_ratio_color() */
114enum grc_type {
115 GRC_STALLED_CYCLES_FE,
116 GRC_STALLED_CYCLES_BE,
117 GRC_CACHE_MISSES,
118 GRC_MAX_NR
119};
120
121static const char *get_ratio_color(enum grc_type type, double ratio)
122{
123 static const double grc_table[GRC_MAX_NR][3] = {
124 [GRC_STALLED_CYCLES_FE] = { 50.0, 30.0, 10.0 },
125 [GRC_STALLED_CYCLES_BE] = { 75.0, 50.0, 20.0 },
126 [GRC_CACHE_MISSES] = { 20.0, 10.0, 5.0 },
127 };
128 const char *color = PERF_COLOR_NORMAL;
129
130 if (ratio > grc_table[type][0])
131 color = PERF_COLOR_RED;
132 else if (ratio > grc_table[type][1])
133 color = PERF_COLOR_MAGENTA;
134 else if (ratio > grc_table[type][2])
135 color = PERF_COLOR_YELLOW;
136
137 return color;
138}
139
140static void print_stalled_cycles_frontend(FILE *out, int cpu,
141 struct perf_evsel *evsel
142 __maybe_unused, double avg)
143{
144 double total, ratio = 0.0;
145 const char *color;
146 int ctx = evsel_context(evsel);
147
148 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
149
150 if (total)
151 ratio = avg / total * 100.0;
152
153 color = get_ratio_color(GRC_STALLED_CYCLES_FE, ratio);
154
155 fprintf(out, " # ");
156 color_fprintf(out, color, "%6.2f%%", ratio);
157 fprintf(out, " frontend cycles idle ");
158}
159
160static void print_stalled_cycles_backend(FILE *out, int cpu,
161 struct perf_evsel *evsel
162 __maybe_unused, double avg)
163{
164 double total, ratio = 0.0;
165 const char *color;
166 int ctx = evsel_context(evsel);
167
168 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
169
170 if (total)
171 ratio = avg / total * 100.0;
172
173 color = get_ratio_color(GRC_STALLED_CYCLES_BE, ratio);
174
175 fprintf(out, " # ");
176 color_fprintf(out, color, "%6.2f%%", ratio);
177 fprintf(out, " backend cycles idle ");
178}
179
180static void print_branch_misses(FILE *out, int cpu,
181 struct perf_evsel *evsel __maybe_unused,
182 double avg)
183{
184 double total, ratio = 0.0;
185 const char *color;
186 int ctx = evsel_context(evsel);
187
188 total = avg_stats(&runtime_branches_stats[ctx][cpu]);
189
190 if (total)
191 ratio = avg / total * 100.0;
192
193 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
194
195 fprintf(out, " # ");
196 color_fprintf(out, color, "%6.2f%%", ratio);
197 fprintf(out, " of all branches ");
198}
199
200static void print_l1_dcache_misses(FILE *out, int cpu,
201 struct perf_evsel *evsel __maybe_unused,
202 double avg)
203{
204 double total, ratio = 0.0;
205 const char *color;
206 int ctx = evsel_context(evsel);
207
208 total = avg_stats(&runtime_l1_dcache_stats[ctx][cpu]);
209
210 if (total)
211 ratio = avg / total * 100.0;
212
213 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
214
215 fprintf(out, " # ");
216 color_fprintf(out, color, "%6.2f%%", ratio);
217 fprintf(out, " of all L1-dcache hits ");
218}
219
220static void print_l1_icache_misses(FILE *out, int cpu,
221 struct perf_evsel *evsel __maybe_unused,
222 double avg)
223{
224 double total, ratio = 0.0;
225 const char *color;
226 int ctx = evsel_context(evsel);
227
228 total = avg_stats(&runtime_l1_icache_stats[ctx][cpu]);
229
230 if (total)
231 ratio = avg / total * 100.0;
232
233 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
234
235 fprintf(out, " # ");
236 color_fprintf(out, color, "%6.2f%%", ratio);
237 fprintf(out, " of all L1-icache hits ");
238}
239
240static void print_dtlb_cache_misses(FILE *out, int cpu,
241 struct perf_evsel *evsel __maybe_unused,
242 double avg)
243{
244 double total, ratio = 0.0;
245 const char *color;
246 int ctx = evsel_context(evsel);
247
248 total = avg_stats(&runtime_dtlb_cache_stats[ctx][cpu]);
249
250 if (total)
251 ratio = avg / total * 100.0;
252
253 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
254
255 fprintf(out, " # ");
256 color_fprintf(out, color, "%6.2f%%", ratio);
257 fprintf(out, " of all dTLB cache hits ");
258}
259
260static void print_itlb_cache_misses(FILE *out, int cpu,
261 struct perf_evsel *evsel __maybe_unused,
262 double avg)
263{
264 double total, ratio = 0.0;
265 const char *color;
266 int ctx = evsel_context(evsel);
267
268 total = avg_stats(&runtime_itlb_cache_stats[ctx][cpu]);
269
270 if (total)
271 ratio = avg / total * 100.0;
272
273 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
274
275 fprintf(out, " # ");
276 color_fprintf(out, color, "%6.2f%%", ratio);
277 fprintf(out, " of all iTLB cache hits ");
278}
279
280static void print_ll_cache_misses(FILE *out, int cpu,
281 struct perf_evsel *evsel __maybe_unused,
282 double avg)
283{
284 double total, ratio = 0.0;
285 const char *color;
286 int ctx = evsel_context(evsel);
287
288 total = avg_stats(&runtime_ll_cache_stats[ctx][cpu]);
289
290 if (total)
291 ratio = avg / total * 100.0;
292
293 color = get_ratio_color(GRC_CACHE_MISSES, ratio);
294
295 fprintf(out, " # ");
296 color_fprintf(out, color, "%6.2f%%", ratio);
297 fprintf(out, " of all LL-cache hits ");
298}
299
300void perf_stat__print_shadow_stats(FILE *out, struct perf_evsel *evsel,
301 double avg, int cpu, enum aggr_mode aggr)
302{
303 double total, ratio = 0.0, total2;
304 int ctx = evsel_context(evsel);
305
306 if (perf_evsel__match(evsel, HARDWARE, HW_INSTRUCTIONS)) {
307 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
308 if (total) {
309 ratio = avg / total;
310 fprintf(out, " # %5.2f insns per cycle ", ratio);
311 } else {
312 fprintf(out, " ");
313 }
314 total = avg_stats(&runtime_stalled_cycles_front_stats[ctx][cpu]);
315 total = max(total, avg_stats(&runtime_stalled_cycles_back_stats[ctx][cpu]));
316
317 if (total && avg) {
318 ratio = total / avg;
319 fprintf(out, "\n");
320 if (aggr == AGGR_NONE)
321 fprintf(out, " ");
322 fprintf(out, " # %5.2f stalled cycles per insn", ratio);
323 }
324
325 } else if (perf_evsel__match(evsel, HARDWARE, HW_BRANCH_MISSES) &&
326 runtime_branches_stats[ctx][cpu].n != 0) {
327 print_branch_misses(out, cpu, evsel, avg);
328 } else if (
329 evsel->attr.type == PERF_TYPE_HW_CACHE &&
330 evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1D |
331 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
332 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
333 runtime_l1_dcache_stats[ctx][cpu].n != 0) {
334 print_l1_dcache_misses(out, cpu, evsel, avg);
335 } else if (
336 evsel->attr.type == PERF_TYPE_HW_CACHE &&
337 evsel->attr.config == ( PERF_COUNT_HW_CACHE_L1I |
338 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
339 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
340 runtime_l1_icache_stats[ctx][cpu].n != 0) {
341 print_l1_icache_misses(out, cpu, evsel, avg);
342 } else if (
343 evsel->attr.type == PERF_TYPE_HW_CACHE &&
344 evsel->attr.config == ( PERF_COUNT_HW_CACHE_DTLB |
345 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
346 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
347 runtime_dtlb_cache_stats[ctx][cpu].n != 0) {
348 print_dtlb_cache_misses(out, cpu, evsel, avg);
349 } else if (
350 evsel->attr.type == PERF_TYPE_HW_CACHE &&
351 evsel->attr.config == ( PERF_COUNT_HW_CACHE_ITLB |
352 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
353 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
354 runtime_itlb_cache_stats[ctx][cpu].n != 0) {
355 print_itlb_cache_misses(out, cpu, evsel, avg);
356 } else if (
357 evsel->attr.type == PERF_TYPE_HW_CACHE &&
358 evsel->attr.config == ( PERF_COUNT_HW_CACHE_LL |
359 ((PERF_COUNT_HW_CACHE_OP_READ) << 8) |
360 ((PERF_COUNT_HW_CACHE_RESULT_MISS) << 16)) &&
361 runtime_ll_cache_stats[ctx][cpu].n != 0) {
362 print_ll_cache_misses(out, cpu, evsel, avg);
363 } else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES) &&
364 runtime_cacherefs_stats[ctx][cpu].n != 0) {
365 total = avg_stats(&runtime_cacherefs_stats[ctx][cpu]);
366
367 if (total)
368 ratio = avg * 100 / total;
369
370 fprintf(out, " # %8.3f %% of all cache refs ", ratio);
371
372 } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_FRONTEND)) {
373 print_stalled_cycles_frontend(out, cpu, evsel, avg);
374 } else if (perf_evsel__match(evsel, HARDWARE, HW_STALLED_CYCLES_BACKEND)) {
375 print_stalled_cycles_backend(out, cpu, evsel, avg);
376 } else if (perf_evsel__match(evsel, HARDWARE, HW_CPU_CYCLES)) {
377 total = avg_stats(&runtime_nsecs_stats[cpu]);
378
379 if (total) {
380 ratio = avg / total;
381 fprintf(out, " # %8.3f GHz ", ratio);
382 } else {
383 fprintf(out, " ");
384 }
385 } else if (perf_stat_evsel__is(evsel, CYCLES_IN_TX)) {
386 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
387 if (total)
388 fprintf(out,
389 " # %5.2f%% transactional cycles ",
390 100.0 * (avg / total));
391 } else if (perf_stat_evsel__is(evsel, CYCLES_IN_TX_CP)) {
392 total = avg_stats(&runtime_cycles_stats[ctx][cpu]);
393 total2 = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
394 if (total2 < avg)
395 total2 = avg;
396 if (total)
397 fprintf(out,
398 " # %5.2f%% aborted cycles ",
399 100.0 * ((total2-avg) / total));
400 } else if (perf_stat_evsel__is(evsel, TRANSACTION_START) &&
401 avg > 0 &&
402 runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
403 total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
404
405 if (total)
406 ratio = total / avg;
407
408 fprintf(out, " # %8.0f cycles / transaction ", ratio);
409 } else if (perf_stat_evsel__is(evsel, ELISION_START) &&
410 avg > 0 &&
411 runtime_cycles_in_tx_stats[ctx][cpu].n != 0) {
412 total = avg_stats(&runtime_cycles_in_tx_stats[ctx][cpu]);
413
414 if (total)
415 ratio = total / avg;
416
417 fprintf(out, " # %8.0f cycles / elision ", ratio);
418 } else if (runtime_nsecs_stats[cpu].n != 0) {
419 char unit = 'M';
420
421 total = avg_stats(&runtime_nsecs_stats[cpu]);
422
423 if (total)
424 ratio = 1000.0 * avg / total;
425 if (ratio < 0.001) {
426 ratio *= 1000;
427 unit = 'K';
428 }
429
430 fprintf(out, " # %8.3f %c/sec ", ratio, unit);
431 } else {
432 fprintf(out, " ");
433 }
434}
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 6506b3dfb605..60b92822f655 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -1,6 +1,6 @@
1#include <math.h> 1#include <math.h>
2
3#include "stat.h" 2#include "stat.h"
3#include "evsel.h"
4 4
5void update_stats(struct stats *stats, u64 val) 5void update_stats(struct stats *stats, u64 val)
6{ 6{
@@ -61,3 +61,36 @@ double rel_stddev_stats(double stddev, double avg)
61 61
62 return pct; 62 return pct;
63} 63}
64
65bool __perf_evsel_stat__is(struct perf_evsel *evsel,
66 enum perf_stat_evsel_id id)
67{
68 struct perf_stat *ps = evsel->priv;
69
70 return ps->id == id;
71}
72
73#define ID(id, name) [PERF_STAT_EVSEL_ID__##id] = #name
74static const char *id_str[PERF_STAT_EVSEL_ID__MAX] = {
75 ID(NONE, x),
76 ID(CYCLES_IN_TX, cpu/cycles-t/),
77 ID(TRANSACTION_START, cpu/tx-start/),
78 ID(ELISION_START, cpu/el-start/),
79 ID(CYCLES_IN_TX_CP, cpu/cycles-ct/),
80};
81#undef ID
82
83void perf_stat_evsel_id_init(struct perf_evsel *evsel)
84{
85 struct perf_stat *ps = evsel->priv;
86 int i;
87
88 /* ps->id is 0 hence PERF_STAT_EVSEL_ID__NONE by default */
89
90 for (i = 0; i < PERF_STAT_EVSEL_ID__MAX; i++) {
91 if (!strcmp(perf_evsel__name(evsel), id_str[i])) {
92 ps->id = i;
93 break;
94 }
95 }
96}
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index 5667fc3e39cf..615c779eb42a 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -2,6 +2,7 @@
2#define __PERF_STATS_H 2#define __PERF_STATS_H
3 3
4#include <linux/types.h> 4#include <linux/types.h>
5#include <stdio.h>
5 6
6struct stats 7struct stats
7{ 8{
@@ -9,6 +10,27 @@ struct stats
9 u64 max, min; 10 u64 max, min;
10}; 11};
11 12
13enum perf_stat_evsel_id {
14 PERF_STAT_EVSEL_ID__NONE = 0,
15 PERF_STAT_EVSEL_ID__CYCLES_IN_TX,
16 PERF_STAT_EVSEL_ID__TRANSACTION_START,
17 PERF_STAT_EVSEL_ID__ELISION_START,
18 PERF_STAT_EVSEL_ID__CYCLES_IN_TX_CP,
19 PERF_STAT_EVSEL_ID__MAX,
20};
21
22struct perf_stat {
23 struct stats res_stats[3];
24 enum perf_stat_evsel_id id;
25};
26
27enum aggr_mode {
28 AGGR_NONE,
29 AGGR_GLOBAL,
30 AGGR_SOCKET,
31 AGGR_CORE,
32};
33
12void update_stats(struct stats *stats, u64 val); 34void update_stats(struct stats *stats, u64 val);
13double avg_stats(struct stats *stats); 35double avg_stats(struct stats *stats);
14double stddev_stats(struct stats *stats); 36double stddev_stats(struct stats *stats);
@@ -22,4 +44,22 @@ static inline void init_stats(struct stats *stats)
22 stats->min = (u64) -1; 44 stats->min = (u64) -1;
23 stats->max = 0; 45 stats->max = 0;
24} 46}
47
48struct perf_evsel;
49bool __perf_evsel_stat__is(struct perf_evsel *evsel,
50 enum perf_stat_evsel_id id);
51
52#define perf_stat_evsel__is(evsel, id) \
53 __perf_evsel_stat__is(evsel, PERF_STAT_EVSEL_ID__ ## id)
54
55void perf_stat_evsel_id_init(struct perf_evsel *evsel);
56
57extern struct stats walltime_nsecs_stats;
58
59void perf_stat__reset_shadow_stats(void);
60void perf_stat__update_shadow_stats(struct perf_evsel *counter, u64 *count,
61 int cpu);
62void perf_stat__print_shadow_stats(FILE *out, struct perf_evsel *evsel,
63 double avg, int cpu, enum aggr_mode aggr);
64
25#endif 65#endif
diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index a93ba85509b2..65f7e389ae09 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -1016,7 +1016,7 @@ int dso__load_sym(struct dso *dso, struct map *map,
1016 curr_map = map__new2(start, curr_dso, 1016 curr_map = map__new2(start, curr_dso,
1017 map->type); 1017 map->type);
1018 if (curr_map == NULL) { 1018 if (curr_map == NULL) {
1019 dso__delete(curr_dso); 1019 dso__put(curr_dso);
1020 goto out_elf_end; 1020 goto out_elf_end;
1021 } 1021 }
1022 if (adjust_kernel_syms) { 1022 if (adjust_kernel_syms) {
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index a3e80d6ad70a..504f2d73b7ee 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -786,7 +786,7 @@ static int dso__split_kallsyms(struct dso *dso, struct map *map, u64 delta,
786 786
787 curr_map = map__new2(pos->start, ndso, map->type); 787 curr_map = map__new2(pos->start, ndso, map->type);
788 if (curr_map == NULL) { 788 if (curr_map == NULL) {
789 dso__delete(ndso); 789 dso__put(ndso);
790 return -1; 790 return -1;
791 } 791 }
792 792
@@ -1364,7 +1364,7 @@ static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1364 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP: 1364 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1365 /* 1365 /*
1366 * kernel modules know their symtab type - it's set when 1366 * kernel modules know their symtab type - it's set when
1367 * creating a module dso in machine__new_module(). 1367 * creating a module dso in machine__findnew_module_map().
1368 */ 1368 */
1369 return kmod && dso->symtab_type == type; 1369 return kmod && dso->symtab_type == type;
1370 1370
diff --git a/tools/perf/util/vdso.c b/tools/perf/util/vdso.c
index 2e8f6886ca72..4b89118f158d 100644
--- a/tools/perf/util/vdso.c
+++ b/tools/perf/util/vdso.c
@@ -120,14 +120,14 @@ void machine__exit_vdso(struct machine *machine)
120 zfree(&machine->vdso_info); 120 zfree(&machine->vdso_info);
121} 121}
122 122
123static struct dso *machine__addnew_vdso(struct machine *machine, const char *short_name, 123static struct dso *__machine__addnew_vdso(struct machine *machine, const char *short_name,
124 const char *long_name) 124 const char *long_name)
125{ 125{
126 struct dso *dso; 126 struct dso *dso;
127 127
128 dso = dso__new(short_name); 128 dso = dso__new(short_name);
129 if (dso != NULL) { 129 if (dso != NULL) {
130 dsos__add(&machine->dsos, dso); 130 __dsos__add(&machine->dsos, dso);
131 dso__set_long_name(dso, long_name, false); 131 dso__set_long_name(dso, long_name, false);
132 } 132 }
133 133
@@ -230,27 +230,31 @@ static const char *vdso__get_compat_file(struct vdso_file *vdso_file)
230 return vdso_file->temp_file_name; 230 return vdso_file->temp_file_name;
231} 231}
232 232
233static struct dso *vdso__findnew_compat(struct machine *machine, 233static struct dso *__machine__findnew_compat(struct machine *machine,
234 struct vdso_file *vdso_file) 234 struct vdso_file *vdso_file)
235{ 235{
236 const char *file_name; 236 const char *file_name;
237 struct dso *dso; 237 struct dso *dso;
238 238
239 dso = dsos__find(&machine->dsos, vdso_file->dso_name, true); 239 pthread_rwlock_wrlock(&machine->dsos.lock);
240 dso = __dsos__find(&machine->dsos, vdso_file->dso_name, true);
240 if (dso) 241 if (dso)
241 return dso; 242 goto out_unlock;
242 243
243 file_name = vdso__get_compat_file(vdso_file); 244 file_name = vdso__get_compat_file(vdso_file);
244 if (!file_name) 245 if (!file_name)
245 return NULL; 246 goto out_unlock;
246 247
247 return machine__addnew_vdso(machine, vdso_file->dso_name, file_name); 248 dso = __machine__addnew_vdso(machine, vdso_file->dso_name, file_name);
249out_unlock:
250 pthread_rwlock_unlock(&machine->dsos.lock);
251 return dso;
248} 252}
249 253
250static int machine__findnew_vdso_compat(struct machine *machine, 254static int __machine__findnew_vdso_compat(struct machine *machine,
251 struct thread *thread, 255 struct thread *thread,
252 struct vdso_info *vdso_info, 256 struct vdso_info *vdso_info,
253 struct dso **dso) 257 struct dso **dso)
254{ 258{
255 enum dso_type dso_type; 259 enum dso_type dso_type;
256 260
@@ -267,10 +271,10 @@ static int machine__findnew_vdso_compat(struct machine *machine,
267 271
268 switch (dso_type) { 272 switch (dso_type) {
269 case DSO__TYPE_32BIT: 273 case DSO__TYPE_32BIT:
270 *dso = vdso__findnew_compat(machine, &vdso_info->vdso32); 274 *dso = __machine__findnew_compat(machine, &vdso_info->vdso32);
271 return 1; 275 return 1;
272 case DSO__TYPE_X32BIT: 276 case DSO__TYPE_X32BIT:
273 *dso = vdso__findnew_compat(machine, &vdso_info->vdsox32); 277 *dso = __machine__findnew_compat(machine, &vdso_info->vdsox32);
274 return 1; 278 return 1;
275 case DSO__TYPE_UNKNOWN: 279 case DSO__TYPE_UNKNOWN:
276 case DSO__TYPE_64BIT: 280 case DSO__TYPE_64BIT:
@@ -285,31 +289,33 @@ struct dso *machine__findnew_vdso(struct machine *machine,
285 struct thread *thread __maybe_unused) 289 struct thread *thread __maybe_unused)
286{ 290{
287 struct vdso_info *vdso_info; 291 struct vdso_info *vdso_info;
288 struct dso *dso; 292 struct dso *dso = NULL;
289 293
294 pthread_rwlock_wrlock(&machine->dsos.lock);
290 if (!machine->vdso_info) 295 if (!machine->vdso_info)
291 machine->vdso_info = vdso_info__new(); 296 machine->vdso_info = vdso_info__new();
292 297
293 vdso_info = machine->vdso_info; 298 vdso_info = machine->vdso_info;
294 if (!vdso_info) 299 if (!vdso_info)
295 return NULL; 300 goto out_unlock;
296 301
297#if BITS_PER_LONG == 64 302#if BITS_PER_LONG == 64
298 if (machine__findnew_vdso_compat(machine, thread, vdso_info, &dso)) 303 if (__machine__findnew_vdso_compat(machine, thread, vdso_info, &dso))
299 return dso; 304 goto out_unlock;
300#endif 305#endif
301 306
302 dso = dsos__find(&machine->dsos, DSO__NAME_VDSO, true); 307 dso = __dsos__find(&machine->dsos, DSO__NAME_VDSO, true);
303 if (!dso) { 308 if (!dso) {
304 char *file; 309 char *file;
305 310
306 file = get_file(&vdso_info->vdso); 311 file = get_file(&vdso_info->vdso);
307 if (!file) 312 if (file)
308 return NULL; 313 dso = __machine__addnew_vdso(machine, DSO__NAME_VDSO, file);
309
310 dso = machine__addnew_vdso(machine, DSO__NAME_VDSO, file);
311 } 314 }
312 315
316out_unlock:
317 dso__get(dso);
318 pthread_rwlock_unlock(&machine->dsos.lock);
313 return dso; 319 return dso;
314} 320}
315 321
generated by cgit v1.2.2 (git 2.25.0) at 2025-08-25 21:04:53 -0400