Skip to contentSkip to conversation

perf tools: fixes and tweaks

PATCH In review Adrian Hunter 7 participants 33 messages Nov 1, 2013 – Nov 15, 2013 1 message not in the archive perf events

lore
Share & tools
lore

Adrian Hunter

author reviewer
#0 lore raw

v3 cover perf tools: fixes and tweaks

Hi

Here are some fixes and tweaks (version 3) for perf tools.

Changes in V3:
perf evsel: Always use perf_evsel__set_sample_bit()
New patch
perf evsel: Add missing overflow check
New patch
perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION
New patch
perf evsel: Add missing PERF_SAMPLE_TRANSACTION
New patch

Patches dropped because they have been applied:
perf tools: Fix non-debug build
perf inject: Do not repipe attributes to a perf.data file

Changes in V2:

perf tools: Fix non-debug build
New patch
perf evsel: Add a debug print if perf_event_open fails
Unchanged
perf script: Make perf_script a local variable
Split from "perf script: Set up output options for in-stream attributes"
perf script: Set up output options for in-stream attributes
Split out "perf script: Make perf_script a local variable"
perf inject: Do not repipe attributes to a perf.data file
Unchanged
perf tools: Fix 32-bit cross build
Pass only EXTRA_CFLAGS
perf tools: Fix libunwind build and feature detection for 32-bit build
Add Jiri's Ack
perf evlist: Add a debug print if event buffer mmap fails
Add errno
perf tools: Allow non-matching sample types
Suppress compatible sample types for trace tool
perf sched: Make struct perf_sched sched a local variable
New patch
perf sched: Fix optimized build time
New patch
perf tools: Do not accept parse_tag_value() overflow
New patch
perf tools: Validate that mmap_pages is not too big
New patch

Patches dropped because they have been applied:
perf evsel: Add missing 'mmap2' from debug print perf record: Improve write_output error message perf evsel: Add missing decrement in id sample parsing perf session: Add missing sample flush for piped events perf session: Add missing members to perf_event__attr_swap() perf evlist: Fix 32-bit build error
perf tools: Fix test_on_exit for 32-bit build perf tools: Fix bench/numa.c for 32-bit build perf tools: fix perf_evlist__mmap comments
perf tools: factor out duplicated evlist mmap code perf script: print addr by default for BTS

Adrian Hunter (11):
perf evsel: Add a debug print if perf_event_open fails perf script: Set up output options for in-stream attributes perf tools: Fix 32-bit cross build
perf tools: Fix libunwind build and feature detection for 32-bit build perf evlist: Add a debug print if event buffer mmap fails perf record: Add an option to force per-cpu mmaps
perf evsel: Always use perf_evsel__set_sample_bit()
perf evsel: Add missing overflow check
perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION perf evsel: Add missing PERF_SAMPLE_TRANSACTION
perf tools: Allow non-matching sample types

 tools/perf/Documentation/perf-record.txt  |  6 +++
 tools/perf/Makefile.perf                  |  2 +-
 tools/perf/builtin-record.c               |  2 +
 tools/perf/builtin-script.c               | 64 +++++++++++++++++++++++--------
 tools/perf/builtin-trace.c                |  1 +
 tools/perf/config/Makefile                |  8 ++--
 tools/perf/config/feature-checks/Makefile |  6 +--
 tools/perf/perf.h                         |  1 +
 tools/perf/tests/sample-parsing.c         |  7 +++-
 tools/perf/util/event.h                   | 16 ++++++++
 tools/perf/util/evlist.c                  | 31 ++++++++++++++-
 tools/perf/util/evlist.h                  |  1 +
 tools/perf/util/evsel.c                   | 21 +++++++---
 tools/perf/util/record.c                  |  5 ++-
 tools/perf/util/target.h                  |  1 +
 15 files changed, 141 insertions(+), 31 deletions(-)

Regards
Adrian

author reviewer
#1 lore raw

PATCH v3 1/11 perf evsel: Add a debug print if perf_event_open fails

There is a debug print (at verbose level 2) for each call to perf_event_open. Add another debug print if the call fails, and print the error number.

---
 tools/perf/util/evsel.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/evsel.c +2 -0
index ec0cc1e..e09c7e6 100644
@@ -1052,6 +1052,8 @@ retry_open:
 								     group_fd, flags);
 			if (FD(evsel, cpu, thread) < 0) {
 				err = -errno;
+				pr_debug2("perf_event_open failed, error %d\n",
+					  err);
 				goto try_fallback;
 			}
 			set_rlimit = NO_CHANGE;
Signature · 1 line
1.7.11.7
#2 lore raw

[tip:perf/core] perf evsel: Add a debug print if perf_event_open fails

perf evsel: Add a debug print if perf_event_open fails
Commit-ID:  f852fd621ca19f557f2e3d05900366be7c7afb83
Gitweb:     http://git.kernel.org/tip/f852fd621ca19f557f2e3d05900366be7c7afb83
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:29 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:18:17 -0300

perf evsel: Add a debug print if perf_event_open fails

There is a debug print (at verbose level 2) for each call to perf_event_open. Add another debug print if the call fails, and print the error number.

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evsel.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/evsel.c +2 -0
index 3a334f0..f0e65de 100644
@@ -1051,6 +1051,8 @@ retry_open:
 								     group_fd, flags);
 			if (FD(evsel, cpu, thread) < 0) {
 				err = -errno;
+				pr_debug2("perf_event_open failed, error %d\n",
+					  err);
 				goto try_fallback;
 			}
 			set_rlimit = NO_CHANGE;
author reviewer
#3 lore raw

PATCH v3 2/11 perf script: Set up output options for in-stream attributes

In reply to Adrian Hunter

Attributes (struct perf_event_attr) are recorded separately in the perf.data file. perf script uses them to set up output options. However attributes can also be in the event stream, for example when the input is a pipe (i.e. live mode). This patch makes perf script process in-stream attributes in the same way as on-file attributes.

Here is an example:

Before this patch:

$ perf record uname | perf script
Linux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ]
:4220  4220 [-01] 2933367.838906: cycles:

:4220 4220 [-01] 2933367.838910: cycles:

:4220 4220 [-01] 2933367.838912: cycles:

:4220 4220 [-01] 2933367.838914: cycles:

:4220 4220 [-01] 2933367.838916: cycles:

:4220 4220 [-01] 2933367.838918: cycles:

uname 4220 [-01] 2933367.838938: cycles:

uname 4220 [-01] 2933367.839207: cycles:

After this patch:

$ perf record uname | perf script
Linux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ]
           :4582  4582 2933425.707724: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707728: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707730: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707732: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707734: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707736: cycles:  ffffffff81309a24 memcpy ([kernel.kallsyms])
           uname  4582 2933425.707760: cycles:  ffffffff8109c1c7 enqueue_task_fair ([kernel.kallsyms])
           uname  4582 2933425.707978: cycles:  ffffffff81308457 clear_page_c ([kernel.kallsyms])
---
 tools/perf/builtin-script.c | 64 +++++++++++++++++++++++++++++++++------------
 1 file changed, 48 insertions(+), 16 deletions(-)
tools/perf/builtin-script.c +48 -16
index 0ae88c2..471fe0e 100644
@@ -229,6 +229,24 @@ static int perf_evsel__check_attr(struct perf_evsel *evsel,
 	return 0;
 }
 
+static void set_print_ip_opts(struct perf_event_attr *attr)
+{
+	unsigned int type = attr->type;
+
+	output[type].print_ip_opts = 0;
+	if (PRINT_FIELD(IP))
+		output[type].print_ip_opts |= PRINT_IP_OPT_IP;
+
+	if (PRINT_FIELD(SYM))
+		output[type].print_ip_opts |= PRINT_IP_OPT_SYM;
+
+	if (PRINT_FIELD(DSO))
+		output[type].print_ip_opts |= PRINT_IP_OPT_DSO;
+
+	if (PRINT_FIELD(SYMOFFSET))
+		output[type].print_ip_opts |= PRINT_IP_OPT_SYMOFFSET;
+}
+
 /*
  * verify all user requested events exist and the samples
  * have the expected data
@@ -237,7 +255,6 @@ static int perf_session__check_output_opt(struct perf_session *session)
 {
 	int j;
 	struct perf_evsel *evsel;
-	struct perf_event_attr *attr;
 
 	for (j = 0; j < PERF_TYPE_MAX; ++j) {
 		evsel = perf_session__find_first_evtype(session, j);
@@ -260,20 +277,7 @@ static int perf_session__check_output_opt(struct perf_session *session)
 		if (evsel == NULL)
 			continue;
 
-		attr = &evsel->attr;
-
-		output[j].print_ip_opts = 0;
-		if (PRINT_FIELD(IP))
-			output[j].print_ip_opts |= PRINT_IP_OPT_IP;
-
-		if (PRINT_FIELD(SYM))
-			output[j].print_ip_opts |= PRINT_IP_OPT_SYM;
-
-		if (PRINT_FIELD(DSO))
-			output[j].print_ip_opts |= PRINT_IP_OPT_DSO;
-
-		if (PRINT_FIELD(SYMOFFSET))
-			output[j].print_ip_opts |= PRINT_IP_OPT_SYMOFFSET;
+		set_print_ip_opts(&evsel->attr);
 	}
 
 	return 0;
@@ -547,6 +551,34 @@ struct perf_script {
 	struct perf_session	*session;
 };
 
+static int process_attr(struct perf_tool *tool, union perf_event *event,
+			struct perf_evlist **pevlist)
+{
+	struct perf_script *scr = container_of(tool, struct perf_script, tool);
+	struct perf_evlist *evlist;
+	struct perf_evsel *evsel, *pos;
+	int err;
+
+	err = perf_event__process_attr(tool, event, pevlist);
+	if (err)
+		return err;
+
+	evlist = *pevlist;
+	evsel = perf_evlist__last(*pevlist);
+
+	if (evsel->attr.type >= PERF_TYPE_MAX)
+		return 0;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		if (pos->attr.type == evsel->attr.type && pos != evsel)
+			return 0;
+	}
+
+	set_print_ip_opts(&evsel->attr);
+
+	return perf_evsel__check_attr(evsel, scr->session);
+}
+
 static void sig_handler(int sig __maybe_unused)
 {
 	session_done = 1;
@@ -1272,7 +1304,7 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 			.comm		 = perf_event__process_comm,
 			.exit		 = perf_event__process_exit,
 			.fork		 = perf_event__process_fork,
-			.attr		 = perf_event__process_attr,
+			.attr		 = process_attr,
 			.tracing_data	 = perf_event__process_tracing_data,
 			.build_id	 = perf_event__process_build_id,
 			.ordered_samples = true,
Signature · 1 line
1.7.11.7
#4 lore raw

[tip:perf/core] perf script: Set up output options for in-stream attributes

perf script: Set up output options for in-stream attributes
Commit-ID:  7ea95727af571d592c9d6aa7627690d44b114a2d
Gitweb:     http://git.kernel.org/tip/7ea95727af571d592c9d6aa7627690d44b114a2d
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:30 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:19:26 -0300

perf script: Set up output options for in-stream attributes

Attributes (struct perf_event_attr) are recorded separately in the perf.data file. perf script uses them to set up output options. However attributes can also be in the event stream, for example when the input is a pipe (i.e. live mode). This patch makes perf script process in-stream attributes in the same way as on-file attributes.

Here is an example:

Before this patch:

$ perf record uname | perf script
Linux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ]
:4220  4220 [-01] 2933367.838906: cycles:

:4220 4220 [-01] 2933367.838910: cycles:

:4220 4220 [-01] 2933367.838912: cycles:

:4220 4220 [-01] 2933367.838914: cycles:

:4220 4220 [-01] 2933367.838916: cycles:

:4220 4220 [-01] 2933367.838918: cycles:

uname 4220 [-01] 2933367.838938: cycles:

uname 4220 [-01] 2933367.839207: cycles:

After this patch:

$ perf record uname | perf script
Linux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ]
           :4582  4582 2933425.707724: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707728: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707730: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707732: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707734: cycles:  ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms])
           :4582  4582 2933425.707736: cycles:  ffffffff81309a24 memcpy ([kernel.kallsyms])
           uname  4582 2933425.707760: cycles:  ffffffff8109c1c7 enqueue_task_fair ([kernel.kallsyms])
           uname  4582 2933425.707978: cycles:  ffffffff81308457 clear_page_c ([kernel.kallsyms])
Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/builtin-script.c | 64 +++++++++++++++++++++++++++++++++------------
 1 file changed, 48 insertions(+), 16 deletions(-)
tools/perf/builtin-script.c +48 -16
index b866cc8..baf1798 100644
@@ -229,6 +229,24 @@ static int perf_evsel__check_attr(struct perf_evsel *evsel,
 	return 0;
 }
 
+static void set_print_ip_opts(struct perf_event_attr *attr)
+{
+	unsigned int type = attr->type;
+
+	output[type].print_ip_opts = 0;
+	if (PRINT_FIELD(IP))
+		output[type].print_ip_opts |= PRINT_IP_OPT_IP;
+
+	if (PRINT_FIELD(SYM))
+		output[type].print_ip_opts |= PRINT_IP_OPT_SYM;
+
+	if (PRINT_FIELD(DSO))
+		output[type].print_ip_opts |= PRINT_IP_OPT_DSO;
+
+	if (PRINT_FIELD(SYMOFFSET))
+		output[type].print_ip_opts |= PRINT_IP_OPT_SYMOFFSET;
+}
+
 /*
  * verify all user requested events exist and the samples
  * have the expected data
@@ -237,7 +255,6 @@ static int perf_session__check_output_opt(struct perf_session *session)
 {
 	int j;
 	struct perf_evsel *evsel;
-	struct perf_event_attr *attr;
 
 	for (j = 0; j < PERF_TYPE_MAX; ++j) {
 		evsel = perf_session__find_first_evtype(session, j);
@@ -260,20 +277,7 @@ static int perf_session__check_output_opt(struct perf_session *session)
 		if (evsel == NULL)
 			continue;
 
-		attr = &evsel->attr;
-
-		output[j].print_ip_opts = 0;
-		if (PRINT_FIELD(IP))
-			output[j].print_ip_opts |= PRINT_IP_OPT_IP;
-
-		if (PRINT_FIELD(SYM))
-			output[j].print_ip_opts |= PRINT_IP_OPT_SYM;
-
-		if (PRINT_FIELD(DSO))
-			output[j].print_ip_opts |= PRINT_IP_OPT_DSO;
-
-		if (PRINT_FIELD(SYMOFFSET))
-			output[j].print_ip_opts |= PRINT_IP_OPT_SYMOFFSET;
+		set_print_ip_opts(&evsel->attr);
 	}
 
 	return 0;
@@ -547,6 +551,34 @@ struct perf_script {
 	struct perf_session	*session;
 };
 
+static int process_attr(struct perf_tool *tool, union perf_event *event,
+			struct perf_evlist **pevlist)
+{
+	struct perf_script *scr = container_of(tool, struct perf_script, tool);
+	struct perf_evlist *evlist;
+	struct perf_evsel *evsel, *pos;
+	int err;
+
+	err = perf_event__process_attr(tool, event, pevlist);
+	if (err)
+		return err;
+
+	evlist = *pevlist;
+	evsel = perf_evlist__last(*pevlist);
+
+	if (evsel->attr.type >= PERF_TYPE_MAX)
+		return 0;
+
+	list_for_each_entry(pos, &evlist->entries, node) {
+		if (pos->attr.type == evsel->attr.type && pos != evsel)
+			return 0;
+	}
+
+	set_print_ip_opts(&evsel->attr);
+
+	return perf_evsel__check_attr(evsel, scr->session);
+}
+
 static void sig_handler(int sig __maybe_unused)
 {
 	session_done = 1;
@@ -1272,7 +1304,7 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 			.comm		 = perf_event__process_comm,
 			.exit		 = perf_event__process_exit,
 			.fork		 = perf_event__process_fork,
-			.attr		 = perf_event__process_attr,
+			.attr		 = process_attr,
 			.tracing_data	 = perf_event__process_tracing_data,
 			.build_id	 = perf_event__process_build_id,
 			.ordered_samples = true,
author reviewer
#5 lore raw

PATCH v3 3/11 perf tools: Fix 32-bit cross build

In reply to Adrian Hunter

Setting EXTRA_CFLAGS=-m32 did not work because it
was not passed around.

---
 tools/perf/Makefile.perf                  | 2 +-
 tools/perf/config/Makefile                | 4 ++--
 tools/perf/config/feature-checks/Makefile | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)
tools/perf/Makefile.perf +1 -1
index bc7cfa1..f1cb389 100644
@@ -709,7 +709,7 @@ $(LIB_FILE): $(LIB_OBJS)
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
 $(LIBTRACEEVENT): $(TE_SOURCES)
-	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(QUIET_SUBDIR1) O=$(OUTPUT) libtraceevent.a
+	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" libtraceevent.a
 
 $(LIBTRACEEVENT)-clean:
 	$(call QUIET_CLEAN, libtraceevent)
tools/perf/config/Makefile +2 -2
index 543aa95..2f1d7d7 100644
@@ -96,7 +96,7 @@ endif
 
 feature_check = $(eval $(feature_check_code))
 define feature_check_code
-  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
+  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
 endef
 
 feature_set = $(eval $(feature_set_code))
@@ -173,7 +173,7 @@ ifeq ($(feature-all), 1)
   #
   $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_set,$(feat)))
 else
-  $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) LDFLAGS=$(LDFLAGS) -i -j -C config/feature-checks $(CORE_FEATURE_TESTS) >/dev/null 2>&1)
+  $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -i -j -C config/feature-checks $(CORE_FEATURE_TESTS) >/dev/null 2>&1)
   $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_check,$(feat)))
 endif
 
tools/perf/config/feature-checks/Makefile +1 -1
index 452b67c..353c00c 100644
@@ -31,7 +31,7 @@ CC := $(CC) -MD
 
 all: $(FILES)
 
-BUILD = $(CC) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
+BUILD = $(CC) $(CFLAGS) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
 
 ###############################
 
Signature · 1 line
1.7.11.7
#6 lore raw

[tip:perf/core] perf tools: Fix 32-bit cross build

perf tools: Fix 32-bit cross build
Commit-ID:  28e962b9d79f496f214d7fc8ffd1a3f2a67e9090
Gitweb:     http://git.kernel.org/tip/28e962b9d79f496f214d7fc8ffd1a3f2a67e9090
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:31 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:20:45 -0300

perf tools: Fix 32-bit cross build

Setting EXTRA_CFLAGS=-m32 did not work because it was not passed around.

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/Makefile.perf                  | 2 +-
 tools/perf/config/Makefile                | 4 ++--
 tools/perf/config/feature-checks/Makefile | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)
tools/perf/Makefile.perf +1 -1
index cb52bdb..5b86390 100644
@@ -711,7 +711,7 @@ $(LIB_FILE): $(LIB_OBJS)
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
 $(LIBTRACEEVENT): $(TE_SOURCES)
-	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(QUIET_SUBDIR1) O=$(OUTPUT) libtraceevent.a
+	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" libtraceevent.a
 
 $(LIBTRACEEVENT)-clean:
 	$(call QUIET_CLEAN, libtraceevent)
tools/perf/config/Makefile +2 -2
index 543aa95..2f1d7d7 100644
@@ -96,7 +96,7 @@ endif
 
 feature_check = $(eval $(feature_check_code))
 define feature_check_code
-  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
+  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
 endef
 
 feature_set = $(eval $(feature_set_code))
@@ -173,7 +173,7 @@ ifeq ($(feature-all), 1)
   #
   $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_set,$(feat)))
 else
-  $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) LDFLAGS=$(LDFLAGS) -i -j -C config/feature-checks $(CORE_FEATURE_TESTS) >/dev/null 2>&1)
+  $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -i -j -C config/feature-checks $(CORE_FEATURE_TESTS) >/dev/null 2>&1)
   $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_check,$(feat)))
 endif
 
tools/perf/config/feature-checks/Makefile +1 -1
index 452b67c..353c00c 100644
@@ -31,7 +31,7 @@ CC := $(CC) -MD
 
 all: $(FILES)
 
-BUILD = $(CC) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
+BUILD = $(CC) $(CFLAGS) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
 
 ###############################
 
author reviewer
#7 lore raw

PATCH v3 4/11 perf tools: Fix libunwind build and feature detection for 32-bit build

In reply to Adrian Hunter

Use -lunwind-x86 instead of -lunwind-x86_64 for
32-bit build.

---
 tools/perf/config/Makefile                | 6 ++++--
 tools/perf/config/feature-checks/Makefile | 4 ++--
 2 files changed, 6 insertions(+), 4 deletions(-)
tools/perf/config/Makefile +4 -2
index 2f1d7d7..ffb5f55 100644
@@ -25,9 +25,11 @@ ifeq ($(ARCH),x86_64)
     RAW_ARCH := x86_64
     CFLAGS += -DHAVE_ARCH_X86_64_SUPPORT
     ARCH_INCLUDE = ../../arch/x86/lib/memcpy_64.S ../../arch/x86/lib/memset_64.S
+    LIBUNWIND_LIBS = -lunwind -lunwind-x86_64
+  else
+    LIBUNWIND_LIBS = -lunwind -lunwind-x86
   endif
   NO_PERF_REGS := 0
-  LIBUNWIND_LIBS = -lunwind -lunwind-x86_64
 endif
 
 ifeq ($(NO_PERF_REGS),0)
@@ -96,7 +98,7 @@ endif
 
 feature_check = $(eval $(feature_check_code))
 define feature_check_code
-  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
+  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) LIBUNWIND_LIBS="$(LIBUNWIND_LIBS)" -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
 endef
 
 feature_set = $(eval $(feature_set_code))
tools/perf/config/feature-checks/Makefile +2 -2
index 353c00c..d37d58d 100644
@@ -36,7 +36,7 @@ BUILD = $(CC) $(CFLAGS) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
 ###############################
 
 test-all:
-	$(BUILD) -Werror -fstack-protector -fstack-protector-all -O2 -Werror -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma -lunwind -lunwind-x86_64 -lelf -laudit -I/usr/include/slang -lslang $(shell pkg-config --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl
+	$(BUILD) -Werror -fstack-protector -fstack-protector-all -O2 -Werror -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma $(LIBUNWIND_LIBS) -lelf -laudit -I/usr/include/slang -lslang $(shell pkg-config --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl
 
 test-hello:
 	$(BUILD)
@@ -72,7 +72,7 @@ test-libnuma:
 	$(BUILD) -lnuma
 
 test-libunwind:
-	$(BUILD) -lunwind -lunwind-x86_64 -lelf
+	$(BUILD) $(LIBUNWIND_LIBS) -lelf
 
 test-libaudit:
 	$(BUILD) -laudit
Signature · 1 line
1.7.11.7
#8 lore raw

[tip:perf/core] perf tools: Fix libunwind build and feature detection for 32-bit build

perf tools: Fix libunwind build and feature detection for 32-bit build
Commit-ID:  8a0c4c2843d3b72e23c3c12079b8d5c3ae99f3e3
Gitweb:     http://git.kernel.org/tip/8a0c4c2843d3b72e23c3c12079b8d5c3ae99f3e3
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:32 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:21:18 -0300

perf tools: Fix libunwind build and feature detection for 32-bit build

Use -lunwind-x86 instead of -lunwind-x86_64 for 32-bit build.

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/config/Makefile                | 6 ++++--
 tools/perf/config/feature-checks/Makefile | 4 ++--
 2 files changed, 6 insertions(+), 4 deletions(-)
tools/perf/config/Makefile +4 -2
index 2f1d7d7..ffb5f55 100644
@@ -25,9 +25,11 @@ ifeq ($(ARCH),x86_64)
     RAW_ARCH := x86_64
     CFLAGS += -DHAVE_ARCH_X86_64_SUPPORT
     ARCH_INCLUDE = ../../arch/x86/lib/memcpy_64.S ../../arch/x86/lib/memset_64.S
+    LIBUNWIND_LIBS = -lunwind -lunwind-x86_64
+  else
+    LIBUNWIND_LIBS = -lunwind -lunwind-x86
   endif
   NO_PERF_REGS := 0
-  LIBUNWIND_LIBS = -lunwind -lunwind-x86_64
 endif
 
 ifeq ($(NO_PERF_REGS),0)
@@ -96,7 +98,7 @@ endif
 
 feature_check = $(eval $(feature_check_code))
 define feature_check_code
-  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
+  feature-$(1) := $(shell $(MAKE) OUTPUT=$(OUTPUT_FEATURES) CFLAGS="$(EXTRA_CFLAGS)" LDFLAGS=$(LDFLAGS) LIBUNWIND_LIBS="$(LIBUNWIND_LIBS)" -C config/feature-checks test-$1 >/dev/null 2>/dev/null && echo 1 || echo 0)
 endef
 
 feature_set = $(eval $(feature_set_code))
tools/perf/config/feature-checks/Makefile +2 -2
index 353c00c..d37d58d 100644
@@ -36,7 +36,7 @@ BUILD = $(CC) $(CFLAGS) $(LDFLAGS) -o $(OUTPUT)$@ [email protected]
 ###############################
 
 test-all:
-	$(BUILD) -Werror -fstack-protector -fstack-protector-all -O2 -Werror -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma -lunwind -lunwind-x86_64 -lelf -laudit -I/usr/include/slang -lslang $(shell pkg-config --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl
+	$(BUILD) -Werror -fstack-protector -fstack-protector-all -O2 -Werror -D_FORTIFY_SOURCE=2 -ldw -lelf -lnuma $(LIBUNWIND_LIBS) -lelf -laudit -I/usr/include/slang -lslang $(shell pkg-config --libs --cflags gtk+-2.0 2>/dev/null) $(FLAGS_PERL_EMBED) $(FLAGS_PYTHON_EMBED) -DPACKAGE='"perf"' -lbfd -ldl
 
 test-hello:
 	$(BUILD)
@@ -72,7 +72,7 @@ test-libnuma:
 	$(BUILD) -lnuma
 
 test-libunwind:
-	$(BUILD) -lunwind -lunwind-x86_64 -lelf
+	$(BUILD) $(LIBUNWIND_LIBS) -lelf
 
 test-libaudit:
 	$(BUILD) -laudit
author reviewer
#9 lore raw

PATCH v3 5/11 perf evlist: Add a debug print if event buffer mmap fails

In reply to Adrian Hunter

Add a debug print if mmap of the perf event
ring buffer fails.

---
 tools/perf/util/evlist.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/evlist.c +2 -0
index 0582f67..1c173cc 100644
@@ -607,6 +607,8 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist,
 	evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot,
 				      MAP_SHARED, fd, 0);
 	if (evlist->mmap[idx].base == MAP_FAILED) {
+		pr_debug2("failed to mmap perf event ring buffer, error %d\n",
+			  errno);
 		evlist->mmap[idx].base = NULL;
 		return -1;
 	}
Signature · 1 line
1.7.11.7
#10 lore raw

[tip:perf/core] perf evlist: Add a debug print if event buffer mmap fails

perf evlist: Add a debug print if event buffer mmap fails
Commit-ID:  026359658aecd348bc5c4a136a26f204b169103b
Gitweb:     http://git.kernel.org/tip/026359658aecd348bc5c4a136a26f204b169103b
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:33 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:21:41 -0300

perf evlist: Add a debug print if event buffer mmap fails

Add a debug print if mmap of the perf event ring buffer fails.

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evlist.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/util/evlist.c +2 -0
index 0582f67..1c173cc 100644
@@ -607,6 +607,8 @@ static int __perf_evlist__mmap(struct perf_evlist *evlist,
 	evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot,
 				      MAP_SHARED, fd, 0);
 	if (evlist->mmap[idx].base == MAP_FAILED) {
+		pr_debug2("failed to mmap perf event ring buffer, error %d\n",
+			  errno);
 		evlist->mmap[idx].base = NULL;
 		return -1;
 	}
author reviewer
#11 lore raw

PATCH v3 6/11 perf record: Add an option to force per-cpu mmaps

In reply to Adrian Hunter

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

---
 tools/perf/Documentation/perf-record.txt | 6 ++++++
 tools/perf/builtin-record.c              | 2 ++
 tools/perf/util/evlist.c                 | 4 +++-
 tools/perf/util/evsel.c                  | 4 ++--
 tools/perf/util/target.h                 | 1 +
 5 files changed, 14 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +6 -0
index f10ab63..2ea6685 100644
@@ -189,6 +189,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
tools/perf/builtin-record.c +2 -0
index ab8d15e..4c0657f 100644
@@ -860,6 +860,8 @@ const struct option record_options[] = {
 		    "sample by weight (on special events only)"),
 	OPT_BOOLEAN(0, "transaction", &record.opts.sample_transaction,
 		    "sample transaction flags (special events only)"),
+	OPT_BOOLEAN(0, "force-per-cpu", &record.opts.target.force_per_cpu,
+		    "force the use of per-cpu mmaps"),
 	OPT_END()
 };
 
tools/perf/util/evlist.c +3 -1
index 1c173cc..b3183d5 100644
@@ -805,7 +805,9 @@ int perf_evlist__create_maps(struct perf_evlist *evlist,
 	if (evlist->threads == NULL)
 		return -1;
 
-	if (perf_target__has_task(target))
+	if (target->force_per_cpu)
+		evlist->cpus = cpu_map__new(target->cpu_list);
+	else if (perf_target__has_task(target))
 		evlist->cpus = cpu_map__dummy_new();
 	else if (!perf_target__has_cpu(target) && !target->uses_mmap)
 		evlist->cpus = cpu_map__dummy_new();
tools/perf/util/evsel.c +2 -2
index e09c7e6..f1839b0 100644
@@ -645,7 +645,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 		}
 	}
 
-	if (perf_target__has_cpu(&opts->target))
+	if (perf_target__has_cpu(&opts->target) || opts->target.force_per_cpu)
 		perf_evsel__set_sample_bit(evsel, CPU);
 
 	if (opts->period)
@@ -653,7 +653,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 
 	if (!perf_missing_features.sample_id_all &&
 	    (opts->sample_time || !opts->no_inherit ||
-	     perf_target__has_cpu(&opts->target)))
+	     perf_target__has_cpu(&opts->target) || opts->target.force_per_cpu))
 		perf_evsel__set_sample_bit(evsel, TIME);
 
 	if (opts->raw_samples) {
tools/perf/util/target.h +1 -0
index a4be857..6d7efbd 100644
@@ -12,6 +12,7 @@ struct perf_target {
 	uid_t	     uid;
 	bool	     system_wide;
 	bool	     uses_mmap;
+	bool	     force_per_cpu;
 };
 
 enum perf_target_errno {
Signature · 1 line
1.7.11.7

A message that is not in the archive (sent off-list or not yet received): [email protected]

author reviewer
#13 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Quoted 31 lines · Jiri Olsa
Quoted 18 lines · Adrian Hunter

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

---
 tools/perf/Documentation/perf-record.txt | 6 ++++++
 tools/perf/builtin-record.c              | 2 ++
 tools/perf/util/evlist.c                 | 4 +++-
 tools/perf/util/evsel.c                  | 4 ++--
 tools/perf/util/target.h                 | 1 +
 5 files changed, 14 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +0 -0
index f10ab63..2ea6685 100644

tools/perf/Documentation/perf-record.txt abort events and some memory events in precise mode on modern Intel CPUs. · line 196

Quoted 8 lines
tools/perf/Documentation/perf-record.txt +5 -0
@@ -189,6 +189,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.

I recently sent out patch that actually force perf cpu mmaps for -p,-t,-u
http://marc.info/?l=linux-kernel&m=138332119912433&w=2

Quoted 3 lines

Is there a reason why would you want to keep single mmap (in record command) and cut yourself from inherited events?

Not sure I understand the question.

perf supports having a single context for a thread. That is a feature. You seem to be removing perf tools support for it.

I image a case where the user has hundreds of CPUs but just wants to record one thread. Currently -t does that. i.e. one file descriptor and one mmap saving megabytes of memory.

Another advantage of per-thread mmaps is that you do not need to sample time (nor cpu), because the events are recorded in order.

I was adding a feature. Users can choose per-cpu mmaps if they want.

#14 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Quoted 45 lines · Adrian Hunter
Quoted 31 lines · Jiri Olsa
Quoted 18 lines · Adrian Hunter

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

---
 tools/perf/Documentation/perf-record.txt | 6 ++++++
 tools/perf/builtin-record.c              | 2 ++
 tools/perf/util/evlist.c                 | 4 +++-
 tools/perf/util/evsel.c                  | 4 ++--
 tools/perf/util/target.h                 | 1 +
 5 files changed, 14 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +0 -0
index f10ab63..2ea6685 100644

tools/perf/Documentation/perf-record.txt abort events and some memory events in precise mode on modern Intel CPUs. · line 196

Quoted 8 lines
tools/perf/Documentation/perf-record.txt +5 -0
@@ -189,6 +189,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.

I recently sent out patch that actually force perf cpu mmaps for -p,-t,-u
http://marc.info/?l=linux-kernel&m=138332119912433&w=2

Quoted 3 lines

Is there a reason why would you want to keep single mmap (in record command) and cut yourself from inherited events?

Not sure I understand the question.

perf supports having a single context for a thread. That is a feature. You seem to be removing perf tools support for it.

I image a case where the user has hundreds of CPUs but just wants to record one thread. Currently -t does that. i.e. one file descriptor and one mmap saving megabytes of memory.

Quoted 6 lines

Another advantage of per-thread mmaps is that you do not need to sample time (nor cpu), because the events are recorded in order.

I was adding a feature. Users can choose per-cpu mmaps if they want.

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

jirka

author reviewer
#15 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Quoted 50 lines · Jiri Olsa
Quoted 49 lines · Adrian Hunter
Quoted 31 lines · Jiri Olsa
Quoted 18 lines · Adrian Hunter

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

---
 tools/perf/Documentation/perf-record.txt | 6 ++++++
 tools/perf/builtin-record.c              | 2 ++
 tools/perf/util/evlist.c                 | 4 +++-
 tools/perf/util/evsel.c                  | 4 ++--
 tools/perf/util/target.h                 | 1 +
 5 files changed, 14 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +0 -0
index f10ab63..2ea6685 100644

tools/perf/Documentation/perf-record.txt abort events and some memory events in precise mode on modern Intel CPUs. · line 196

Quoted 8 lines
tools/perf/Documentation/perf-record.txt +5 -0
@@ -189,6 +189,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.

I recently sent out patch that actually force perf cpu mmaps for -p,-t,-u
http://marc.info/?l=linux-kernel&m=138332119912433&w=2

Quoted 3 lines

Is there a reason why would you want to keep single mmap (in record command) and cut yourself from inherited events?

Not sure I understand the question.

perf supports having a single context for a thread. That is a feature. You seem to be removing perf tools support for it.

I image a case where the user has hundreds of CPUs but just wants to record one thread. Currently -t does that. i.e. one file descriptor and one mmap saving megabytes of memory.

Another advantage of per-thread mmaps is that you do not need to sample time (nor cpu), because the events are recorded in order.

Quoted 6 lines
Quoted 2 lines

I was adding a feature. Users can choose per-cpu mmaps if they want.

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

How about this:

From: Adrian Hunter <[email protected]>

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

---
 tools/perf/Documentation/perf-record.txt | 15 +++++++++++++++
 tools/perf/builtin-record.c              |  2 ++
 tools/perf/util/evlist.c                 |  4 +++-
 tools/perf/util/evsel.c                  |  4 ++--
 tools/perf/util/target.h                 |  1 +
 5 files changed, 23 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +15 -0
index 052f7c4..5d5cbed 100644
@@ -53,14 +53,20 @@ OPTIONS
 -p::
 --pid=::
 	Record events on existing process ID (comma separated list).
+	Note this causes the use of per-thread mmaps which prevents
+	inheritance. This can be overridden by --force-per-cpu.
 
 -t::
 --tid=::
         Record events on existing thread ID (comma separated list).
+	Note this causes the use of per-thread mmaps which prevents
+	inheritance. This can be overridden by --force-per-cpu.
 
 -u::
 --uid=::
         Record events in threads owned by uid. Name or number.
+	Note this causes the use of per-thread mmaps which prevents
+	inheritance. This can be overridden by --force-per-cpu.
 
 -r::
 --realtime=::
@@ -81,6 +87,9 @@ OPTIONS
 -i::
 --no-inherit::
 	Child tasks do not inherit counters.
+	Note that inheritance is not possible when events are recorded
+	with per-thread mmaps e.g. -p, -t or -u options.
+
 -F::
 --freq=::
 	Profile at this frequency.
@@ -201,6 +210,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
tools/perf/builtin-record.c +2 -0
index 8b45fce..6b6f5d9 100644
@@ -889,6 +889,8 @@ const struct option record_options[] = {
 		    "sample by weight (on special events only)"),
 	OPT_BOOLEAN(0, "transaction", &record.opts.sample_transaction,
 		    "sample transaction flags (special events only)"),
+	OPT_BOOLEAN(0, "force-per-cpu", &record.opts.target.force_per_cpu,
+		    "force the use of per-cpu mmaps"),
 	OPT_END()
 };
 
tools/perf/util/evlist.c +3 -1
index 1c173cc..b3183d5 100644
@@ -805,7 +805,9 @@ int perf_evlist__create_maps(struct perf_evlist *evlist,
 	if (evlist->threads == NULL)
 		return -1;
 
-	if (perf_target__has_task(target))
+	if (target->force_per_cpu)
+		evlist->cpus = cpu_map__new(target->cpu_list);
+	else if (perf_target__has_task(target))
 		evlist->cpus = cpu_map__dummy_new();
 	else if (!perf_target__has_cpu(target) && !target->uses_mmap)
 		evlist->cpus = cpu_map__dummy_new();
tools/perf/util/evsel.c +2 -2
index 5280820..080cc55 100644
@@ -645,7 +645,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 		}
 	}
 
-	if (perf_target__has_cpu(&opts->target))
+	if (perf_target__has_cpu(&opts->target) || opts->target.force_per_cpu)
 		perf_evsel__set_sample_bit(evsel, CPU);
 
 	if (opts->period)
@@ -653,7 +653,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 
 	if (!perf_missing_features.sample_id_all &&
 	    (opts->sample_time || !opts->no_inherit ||
-	     perf_target__has_cpu(&opts->target)))
+	     perf_target__has_cpu(&opts->target) || opts->target.force_per_cpu))
 		perf_evsel__set_sample_bit(evsel, TIME);
 
 	if (opts->raw_samples) {
tools/perf/util/target.h +1 -0
index a4be857..6d7efbd 100644
@@ -12,6 +12,7 @@ struct perf_target {
 	uid_t	     uid;
 	bool	     system_wide;
 	bool	     uses_mmap;
+	bool	     force_per_cpu;
 };
 
 enum perf_target_errno {
Signature · 1 line
1.7.11.7
Acked-by
#16 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

On Tue, Nov 05, 2013 at 03:09:30PM +0200, Adrian Hunter wrote:

SNIP

Quoted 13 lines
Quoted 10 lines
Quoted 6 lines

Another advantage of per-thread mmaps is that you do not need to sample time (nor cpu), because the events are recorded in order.

I was adding a feature. Users can choose per-cpu mmaps if they want.

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

How about this:

Quoted 7 lines

From: Adrian Hunter <[email protected]>

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

seems ok, thanks

jirka

#17 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Em Tue, Nov 05, 2013 at 02:30:24PM +0100, Jiri Olsa escreveu:

Quoted 9 lines · Adrian Hunter
Quoted 8 lines
Quoted 3 lines

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

How about this:

From: Adrian Hunter <[email protected]>

Quoted 7 lines
Quoted 5 lines

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

seems ok, thanks

Much better, but may I ask that the explanation in the man pages be present in the changeset commit as well? So that by just looking at 'git log' we can more readily get the whole picture, i.e. scalability versus wanting to have inherit.

- Arnaldo

Quoted 3 lines

jirka

#18 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

PeterZ,

Can I have your Acked-by for this one? I guess now the goal is
achieved, no?

- Arnaldo

Em Tue, Nov 05, 2013 at 11:25:45AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 28 lines

Em Tue, Nov 05, 2013 at 02:30:24PM +0100, Jiri Olsa escreveu:

Quoted 9 lines · Adrian Hunter
Quoted 8 lines
Quoted 3 lines

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

How about this:

From: Adrian Hunter <[email protected]>

Quoted 7 lines
Quoted 5 lines

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

seems ok, thanks

Much better, but may I ask that the explanation in the man pages be present in the changeset commit as well? So that by just looking at 'git log' we can more readily get the whole picture, i.e. scalability versus wanting to have inherit.

- Arnaldo

Quoted 3 lines

jirka

author reviewer
#19 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Quoted 4 lines · Arnaldo Carvalho de Melo

PeterZ,

Can I have your Acked-by for this one? I guess now the goal is
achieved, no?

Ping

Quoted 30 lines

- Arnaldo

Em Tue, Nov 05, 2013 at 11:25:45AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 25 lines

Em Tue, Nov 05, 2013 at 02:30:24PM +0100, Jiri Olsa escreveu:

Quoted 6 lines · Adrian Hunter
Quoted 5 lines
Quoted 3 lines

right, I haven't considered the hundreds CPU machine.. I just saw that it disables inherited events in my test ;-) maybe we could mentioned that somewhere, because it's not clear

How about this:

Quoted 8 lines
Quoted 7 lines

From: Adrian Hunter <[email protected]>

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created. Add an option to override that and force per-cpu mmaps.

seems ok, thanks

Much better, but may I ask that the explanation in the man pages be present in the changeset commit as well? So that by just looking at 'git log' we can more readily get the whole picture, i.e. scalability versus wanting to have inherit.

- Arnaldo

Quoted 3 lines

jirka

maintainer Acked-by
#20 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

In reply to Arnaldo Carvalho de Melo
Quoted 4 lines · Arnaldo Carvalho de Melo

PeterZ,

Can I have your Acked-by for this one? I guess now the goal is
achieved, no?

So this option allows -t/-p/-u to create one buffer per cpu and attach all the various thread/process/user tasks' their counters to that one buffer?

As opposed to the current state where each such counter would have its own buffer.

If this is what the patch does, then yes, although I would prefer a slightly clearer Changelog.

#21 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Em Fri, Nov 08, 2013 at 09:40:01AM +0100, Peter Zijlstra escreveu:

Quoted 8 lines · Arnaldo Carvalho de Melo
Quoted 2 lines

Can I have your Acked-by for this one? I guess now the goal is
achieved, no?

So this option allows -t/-p/-u to create one buffer per cpu and attach all the various thread/process/user tasks' their counters to that one buffer?

Quoted 5 lines

As opposed to the current state where each such counter would have its own buffer.

If this is what the patch does, then yes, although I would prefer a slightly clearer Changelog.

/me too

I will re-read the patch to check that it matches what you think it does and will update the changelog to make it clearer.

Quoted 1 line

Thanks!

- Arnaldo

#22 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

In reply to Peter Zijlstra
Quoted 10 lines · Peter Zijlstra
Quoted 4 lines · Arnaldo Carvalho de Melo

PeterZ,

Can I have your Acked-by for this one? I guess now the goal is
achieved, no?

So this option allows -t/-p/-u to create one buffer per cpu and attach all the various thread/process/user tasks' their counters to that one buffer?

Quoted 7 lines

As opposed to the current state where each such counter would have its own buffer.

If this is what the patch does, then yes, although I would prefer a slightly clearer Changelog.

Is there any reason why we wouldn't want to make this the default behavior?

That way we could also lose the somewhat suboptimal 'force' naming:
there's nothing forced really, we simply switch to another ring-buffer setup ...

Thanks,

Ingo

#23 lore raw

Re: [PATCH V3 06/11] perf record: Add an option to force per-cpu mmaps

Ingo Molnar [[email protected]] wrote:

| 
| * Peter Zijlstra <[email protected]> wrote:
| 
| > On Tue, Nov 05, 2013 at 02:31:52PM -0300, Arnaldo Carvalho de Melo wrote:
| > > PeterZ,
| > > 
| > > 	Can I have your Acked-by for this one? I guess now the goal is
| > > achieved, no?

Being able to profile children with the --pid is a big plus.

| > 
| > So this option allows -t/-p/-u to create one buffer per cpu and attach
| > all the various thread/process/user tasks' their counters to that one
| > buffer?
| > 
| > As opposed to the current state where each such counter would have its
| > own buffer.
| > 
| > If this is what the patch does, then yes, although I would prefer a
| > slightly clearer Changelog.
| > 
| > Acked-by: Peter Zijlstra <[email protected]>
| 
| Is there any reason why we wouldn't want to make this the default 
| behavior?
| 
| That way we could also lose the somewhat suboptimal 'force' naming: 
| there's nothing forced really, we simply switch to another ring-buffer 
| setup ...

It would be also good if the man page added a comment on when a user would want one ring-buffer setup over the other.

If the main benefit is to have the children profiled when tasks are specified, how about changing the option to --inherit (-I) ?

Or for consistency with 'perf record <application>', have --pid profile children by default and let users specify --no-inherit with --pid if they don't want children profiled.

Sukadev

#24 lore raw

[tip:perf/urgent] perf record: Add an option to force per-cpu mmaps

In reply to Adrian Hunter
perf record: Add an option to force per-cpu mmaps
Commit-ID:  539e6bb71e350541105e67e3d6c31392d9da25ef
Gitweb:     http://git.kernel.org/tip/539e6bb71e350541105e67e3d6c31392d9da25ef
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:34 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Thu, 14 Nov 2013 16:10:27 -0300

perf record: Add an option to force per-cpu mmaps

By default, when tasks are specified (i.e. -p, -t or -u options) per-thread mmaps are created.

Add an option to override that and force per-cpu mmaps.

Further comments by peterz:

So this option allows -t/-p/-u to create one buffer per cpu and attach all the various thread/process/user tasks' their counters to that one buffer?

As opposed to the current state where each such counter would have its own buffer.

---
 tools/perf/Documentation/perf-record.txt | 6 ++++++
 tools/perf/builtin-record.c              | 2 ++
 tools/perf/util/evlist.c                 | 4 +++-
 tools/perf/util/evsel.c                  | 4 ++--
 tools/perf/util/target.h                 | 1 +
 5 files changed, 14 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +6 -0
index 052f7c4..43b42c4 100644
@@ -201,6 +201,12 @@ abort events and some memory events in precise mode on modern Intel CPUs.
 --transaction::
 Record transaction flags for transaction related events.
 
+--force-per-cpu::
+Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
+-t or -u options) per-thread mmaps are created.  This option overrides that and
+forces per-cpu mmaps.  A side-effect of that is that inheritance is
+automatically enabled.  Add the -i option also to disable inheritance.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
tools/perf/builtin-record.c +2 -0
index 4d644fe..7c8020a 100644
@@ -888,6 +888,8 @@ const struct option record_options[] = {
 		    "sample by weight (on special events only)"),
 	OPT_BOOLEAN(0, "transaction", &record.opts.sample_transaction,
 		    "sample transaction flags (special events only)"),
+	OPT_BOOLEAN(0, "force-per-cpu", &record.opts.target.force_per_cpu,
+		    "force the use of per-cpu mmaps"),
 	OPT_END()
 };
 
tools/perf/util/evlist.c +3 -1
index 5ce2ace..bbc746a 100644
@@ -819,7 +819,9 @@ int perf_evlist__create_maps(struct perf_evlist *evlist, struct target *target)
 	if (evlist->threads == NULL)
 		return -1;
 
-	if (target__has_task(target))
+	if (target->force_per_cpu)
+		evlist->cpus = cpu_map__new(target->cpu_list);
+	else if (target__has_task(target))
 		evlist->cpus = cpu_map__dummy_new();
 	else if (!target__has_cpu(target) && !target->uses_mmap)
 		evlist->cpus = cpu_map__dummy_new();
tools/perf/util/evsel.c +2 -2
index 18f7c18..46dd4c2 100644
@@ -645,7 +645,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 		}
 	}
 
-	if (target__has_cpu(&opts->target))
+	if (target__has_cpu(&opts->target) || opts->target.force_per_cpu)
 		perf_evsel__set_sample_bit(evsel, CPU);
 
 	if (opts->period)
@@ -653,7 +653,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 
 	if (!perf_missing_features.sample_id_all &&
 	    (opts->sample_time || !opts->no_inherit ||
-	     target__has_cpu(&opts->target)))
+	     target__has_cpu(&opts->target) || opts->target.force_per_cpu))
 		perf_evsel__set_sample_bit(evsel, TIME);
 
 	if (opts->raw_samples) {
tools/perf/util/target.h +1 -0
index 89bab71..2d0c506 100644
@@ -12,6 +12,7 @@ struct target {
 	uid_t	     uid;
 	bool	     system_wide;
 	bool	     uses_mmap;
+	bool	     force_per_cpu;
 };
 
 enum target_errno {
author reviewer
#25 lore raw

PATCH v3 7/11 perf evsel: Always use perf_evsel__set_sample_bit()

In reply to Adrian Hunter

Always use perf_evsel__set_sample_bit() rather than just setting the bit.

---
 tools/perf/util/evsel.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
tools/perf/util/evsel.c +3 -3
index f1839b0..f0e263c 100644
@@ -663,7 +663,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 	}
 
 	if (opts->sample_address)
-		attr->sample_type	|= PERF_SAMPLE_DATA_SRC;
+		perf_evsel__set_sample_bit(evsel, DATA_SRC);
 
 	if (opts->no_delay) {
 		attr->watermark = 0;
@@ -675,14 +675,14 @@ void perf_evsel__config(struct perf_evsel *evsel,
 	}
 
 	if (opts->sample_weight)
-		attr->sample_type	|= PERF_SAMPLE_WEIGHT;
+		perf_evsel__set_sample_bit(evsel, WEIGHT);
 
 	attr->mmap  = track;
 	attr->mmap2 = track && !perf_missing_features.mmap2;
 	attr->comm  = track;
 
 	if (opts->sample_transaction)
-		attr->sample_type	|= PERF_SAMPLE_TRANSACTION;
+		perf_evsel__set_sample_bit(evsel, TRANSACTION);
 
 	/*
 	 * XXX see the function comment above
Signature · 1 line
1.7.11.7
#26 lore raw

[tip:perf/core] perf evsel: Always use perf_evsel__set_sample_bit ()

perf evsel: Always use perf_evsel__set_sample_bit()
Commit-ID:  1e7ed5ec54e3998bda6ea625599a0644404cb421
Gitweb:     http://git.kernel.org/tip/1e7ed5ec54e3998bda6ea625599a0644404cb421
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:35 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:42:28 -0300

perf evsel: Always use perf_evsel__set_sample_bit()

Always use perf_evsel__set_sample_bit() rather than just setting the bit.

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
[ Cope with 3090ffb "perf: Disable PERF_RECORD_MMAP2 support" ]
---
 tools/perf/util/evsel.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
tools/perf/util/evsel.c +3 -3
index f0e65de..47bbf03 100644
@@ -663,7 +663,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
 	}
 
 	if (opts->sample_address)
-		attr->sample_type	|= PERF_SAMPLE_DATA_SRC;
+		perf_evsel__set_sample_bit(evsel, DATA_SRC);
 
 	if (opts->no_delay) {
 		attr->watermark = 0;
@@ -675,13 +675,13 @@ void perf_evsel__config(struct perf_evsel *evsel,
 	}
 
 	if (opts->sample_weight)
-		attr->sample_type	|= PERF_SAMPLE_WEIGHT;
+		perf_evsel__set_sample_bit(evsel, WEIGHT);
 
 	attr->mmap  = track;
 	attr->comm  = track;
 
 	if (opts->sample_transaction)
-		attr->sample_type	|= PERF_SAMPLE_TRANSACTION;
+		perf_evsel__set_sample_bit(evsel, TRANSACTION);
 
 	/*
 	 * XXX see the function comment above
author reviewer
#27 lore raw

PATCH v3 8/11 perf evsel: Add missing overflow check

In reply to Adrian Hunter

Add missing overflow check in
perf_evsel__parse_sample()

---
 tools/perf/util/evsel.c | 1 +
 1 file changed, 1 insertion(+)
tools/perf/util/evsel.c +1 -0
index f0e263c..dc2b2dc 100644
@@ -1482,6 +1482,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 
 	data->transaction = 0;
 	if (type & PERF_SAMPLE_TRANSACTION) {
+		OVERFLOW_CHECK_u64(array);
 		data->transaction = *array;
 		array++;
 	}
Signature · 1 line
1.7.11.7
#28 lore raw

[tip:perf/core] perf evsel: Add missing overflow check for TRANSACTION

perf evsel: Add missing overflow check for TRANSACTION
Commit-ID:  87b955247d71975460774435241be3aa05218a7b
Gitweb:     http://git.kernel.org/tip/87b955247d71975460774435241be3aa05218a7b
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:36 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:44:01 -0300

perf evsel: Add missing overflow check for TRANSACTION

Add missing overflow check for PERF_SAMPLE_TRANSACTION in perf_evsel__parse_sample().

Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evsel.c | 1 +
 1 file changed, 1 insertion(+)
tools/perf/util/evsel.c +1 -0
index 47bbf03..b121717 100644
@@ -1481,6 +1481,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 
 	data->transaction = 0;
 	if (type & PERF_SAMPLE_TRANSACTION) {
+		OVERFLOW_CHECK_u64(array);
 		data->transaction = *array;
 		array++;
 	}
author reviewer
#29 lore raw

PATCH v3 9/11 perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION

In reply to Adrian Hunter

In fact the "sample parsing" test does not automatically check new sample type bits - they must be added to the comparison logic. Doing that shows that the test fails because the functions perf_event__synthesize_sample() and perf_event__sample_event_size() have not been updated with PERF_SAMPLE_TRANSACTION either.

---
 tools/perf/tests/sample-parsing.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/perf/tests/sample-parsing.c +6 -1
index 61c9da2..1b67720 100644
@@ -121,6 +121,9 @@ static bool samples_same(const struct perf_sample *s1,
 	if (type & PERF_SAMPLE_DATA_SRC)
 		COMP(data_src);
 
+	if (type & PERF_SAMPLE_TRANSACTION)
+		COMP(transaction);
+
 	return true;
 }
 
@@ -165,6 +168,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
 		.cpu		= 110,
 		.raw_size	= sizeof(raw_data),
 		.data_src	= 111,
+		.transaction	= 112,
 		.raw_data	= (void *)raw_data,
 		.callchain	= &callchain.callchain,
 		.branch_stack	= &branch_stack.branch_stack,
@@ -273,7 +277,8 @@ int test__sample_parsing(void)
 
 	/*
 	 * Fail the test if it has not been updated when new sample format bits
-	 * were added.
+	 * were added.  Please actually update the test rather than just change
+	 * the condition below.
 	 */
 	if (PERF_SAMPLE_MAX > PERF_SAMPLE_TRANSACTION << 1) {
 		pr_debug("sample format has changed, some new PERF_SAMPLE_ bit was introduced - test needs updating\n");
Signature · 1 line
1.7.11.7
#30 lore raw

[tip:perf/core] perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION

perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION
Commit-ID:  091a4ef5a94d46d26a05f0c32d2f64800ed91306
Gitweb:     http://git.kernel.org/tip/091a4ef5a94d46d26a05f0c32d2f64800ed91306
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:37 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:47:24 -0300

perf test: Update "sample parsing" test for PERF_SAMPLE_TRANSACTION

In fact the "sample parsing" test does not automatically check new sample type bits - they must be added to the comparison logic.

Doing that shows that the test fails because the functions perf_event__synthesize_sample() and perf_event__sample_event_size() have not been updated with PERF_SAMPLE_TRANSACTION either.

Cc Andi Kleen <[email protected]>
Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/tests/sample-parsing.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/perf/tests/sample-parsing.c +6 -1
index 61c9da2..1b67720 100644
@@ -121,6 +121,9 @@ static bool samples_same(const struct perf_sample *s1,
 	if (type & PERF_SAMPLE_DATA_SRC)
 		COMP(data_src);
 
+	if (type & PERF_SAMPLE_TRANSACTION)
+		COMP(transaction);
+
 	return true;
 }
 
@@ -165,6 +168,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
 		.cpu		= 110,
 		.raw_size	= sizeof(raw_data),
 		.data_src	= 111,
+		.transaction	= 112,
 		.raw_data	= (void *)raw_data,
 		.callchain	= &callchain.callchain,
 		.branch_stack	= &branch_stack.branch_stack,
@@ -273,7 +277,8 @@ int test__sample_parsing(void)
 
 	/*
 	 * Fail the test if it has not been updated when new sample format bits
-	 * were added.
+	 * were added.  Please actually update the test rather than just change
+	 * the condition below.
 	 */
 	if (PERF_SAMPLE_MAX > PERF_SAMPLE_TRANSACTION << 1) {
 		pr_debug("sample format has changed, some new PERF_SAMPLE_ bit was introduced - test needs updating\n");
author reviewer
#31 lore raw

PATCH v3 10/11 perf evsel: Add missing PERF_SAMPLE_TRANSACTION

In reply to Adrian Hunter

Add missing PERF_SAMPLE_TRANSACTION to
perf_event__synthesize_sample() and
perf_event__sample_event_size().

This makes the "sample parsing" test pass.

---
 tools/perf/util/evsel.c | 8 ++++++++
 1 file changed, 8 insertions(+)
tools/perf/util/evsel.c +8 -0
index dc2b2dc..7802bde 100644
@@ -1579,6 +1579,9 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_DATA_SRC)
 		result += sizeof(u64);
 
+	if (type & PERF_SAMPLE_TRANSACTION)
+		result += sizeof(u64);
+
 	return result;
 }
 
@@ -1752,6 +1755,11 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 		array++;
 	}
 
+	if (type & PERF_SAMPLE_TRANSACTION) {
+		*array = sample->transaction;
+		array++;
+	}
+
 	return 0;
 }
 
Signature · 1 line
1.7.11.7
#32 lore raw

[tip:perf/core] perf evsel: Synthesize PERF_SAMPLE_TRANSACTION

perf evsel: Synthesize PERF_SAMPLE_TRANSACTION
Commit-ID:  42d88910c717ba21089251d0ca559abfef0df22d
Gitweb:     http://git.kernel.org/tip/42d88910c717ba21089251d0ca559abfef0df22d
Author:     Adrian Hunter <[email protected]>
AuthorDate: Fri, 1 Nov 2013 15:51:38 +0200
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Mon, 4 Nov 2013 12:49:36 -0300

perf evsel: Synthesize PERF_SAMPLE_TRANSACTION

Add missing PERF_SAMPLE_TRANSACTION to perf_event__synthesize_sample() and perf_event__sample_event_size().

This makes the "sample parsing" test pass.

Cc Andi Kleen <[email protected]>
Cc David Ahern <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 tools/perf/util/evsel.c | 8 ++++++++
 1 file changed, 8 insertions(+)
tools/perf/util/evsel.c +8 -0
index b121717..5280820 100644
@@ -1578,6 +1578,9 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_DATA_SRC)
 		result += sizeof(u64);
 
+	if (type & PERF_SAMPLE_TRANSACTION)
+		result += sizeof(u64);
+
 	return result;
 }
 
@@ -1751,6 +1754,11 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 		array++;
 	}
 
+	if (type & PERF_SAMPLE_TRANSACTION) {
+		*array = sample->transaction;
+		array++;
+	}
+
 	return 0;
 }
 
author reviewer
#33 lore raw

PATCH v3 11/11 perf tools: Allow non-matching sample types

In reply to Adrian Hunter

For kernels that do not support PERF_SAMPLE_IDENTIFIER, sample types need not be identical to determine the sample id from the event. Only the position of the sample id needs to be the same.

Compatible sample types are ones in which the bits defined by PERF_COMPAT_MASK are the same. 'perf_evlist__config()' forces sample types to be compatible on that basis.

---
 tools/perf/builtin-trace.c |  1 +
 tools/perf/perf.h          |  1 +
 tools/perf/util/event.h    | 16 ++++++++++++++++
 tools/perf/util/evlist.c   | 25 +++++++++++++++++++++++++
 tools/perf/util/evlist.h   |  1 +
 tools/perf/util/record.c   |  5 ++++-
 6 files changed, 48 insertions(+), 1 deletion(-)
tools/perf/builtin-trace.c +1 -0
index dc3da65..08a34e3 100644
@@ -2062,6 +2062,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 			.user_interval = ULLONG_MAX,
 			.no_delay      = true,
 			.mmap_pages    = 1024,
+			.incompatible_sample_types = true,
 		},
 		.output = stdout,
 		.show_comm = true,
tools/perf/perf.h +1 -0
index f61c230..aeecdf7 100644
@@ -233,6 +233,7 @@ struct perf_record_opts {
 	u64	     user_interval;
 	u16	     stack_dump_size;
 	bool	     sample_transaction;
+	bool         incompatible_sample_types;
 };
 
 #endif
tools/perf/util/event.h +16 -0
index 752709c..ca0689c 100644
@@ -78,6 +78,22 @@ struct throttle_event {
 /* perf sample has 16 bits size limit */
 #define PERF_SAMPLE_MAX_SIZE (1 << 16)
 
+/*
+ * Events have compatible sample types if the following bits all have the same
+ * value.  This is because the order of sample members is fixed.  For sample
+ * events the order is: PERF_SAMPLE_IP, PERF_SAMPLE_TID, PERF_SAMPLE_TIME,
+ * PERF_SAMPLE_ADDR, PERF_SAMPLE_ID.  For non-sample events the sample members
+ * are accessed in reverse order.  The order is: PERF_SAMPLE_ID,
+ * PERF_SAMPLE_STREAM_ID, PERF_SAMPLE_CPU.  PERF_SAMPLE_IDENTIFIER is added for
+ * completeness but it should not be used with PERF_SAMPLE_ID.  Sample types
+ * that include PERF_SAMPLE_IDENTIFIER are always compatible.
+ */
+#define PERF_COMPAT_MASK				\
+	(PERF_SAMPLE_IP   | PERF_SAMPLE_TID       |	\
+	 PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR      |	\
+	 PERF_SAMPLE_ID   | PERF_SAMPLE_STREAM_ID |	\
+	 PERF_SAMPLE_CPU  | PERF_SAMPLE_IDENTIFIER)
+
 struct sample_event {
 	struct perf_event_header        header;
 	u64 array[];
tools/perf/util/evlist.c +25 -0
index b3183d5..c893113 100644
@@ -87,6 +87,31 @@ static void perf_evlist__update_id_pos(struct perf_evlist *evlist)
 	perf_evlist__set_id_pos(evlist);
 }
 
+/**
+ * perf_evlist__make_sample_types_compatible - make sample types compatible.
+ * @evlist: selected event list
+ *
+ * Events with compatible sample types all have the same id_pos and is_pos.
+ * This can be achieved by matching the bits of PERF_COMPAT_MASK.
+ */
+void perf_evlist__make_sample_types_compatible(struct perf_evlist *evlist)
+{
+	struct perf_evsel *evsel;
+	u64 compat = 0;
+
+	list_for_each_entry(evsel, &evlist->entries, node)
+		compat |= evsel->attr.sample_type & PERF_COMPAT_MASK;
+
+	list_for_each_entry(evsel, &evlist->entries, node) {
+		evsel->attr.sample_type |= compat;
+		evsel->sample_size =
+			__perf_evsel__sample_size(evsel->attr.sample_type);
+		perf_evsel__calc_id_pos(evsel);
+	}
+
+	perf_evlist__set_id_pos(evlist);
+}
+
 static void perf_evlist__purge(struct perf_evlist *evlist)
 {
 	struct perf_evsel *pos, *n;
tools/perf/util/evlist.h +1 -0
index 6e8acc9..d0d8a39 100644
@@ -96,6 +96,7 @@ int perf_evlist__open(struct perf_evlist *evlist);
 void perf_evlist__close(struct perf_evlist *evlist);
 
 void perf_evlist__set_id_pos(struct perf_evlist *evlist);
+void perf_evlist__make_sample_types_compatible(struct perf_evlist *evlist);
 bool perf_can_sample_identifier(void);
 void perf_evlist__config(struct perf_evlist *evlist,
 			 struct perf_record_opts *opts);
tools/perf/util/record.c +4 -1
index 18d73aa..1eb1290 100644
@@ -104,5 +104,8 @@ void perf_evlist__config(struct perf_evlist *evlist,
 			perf_evsel__set_sample_id(evsel, use_sample_identifier);
 	}
 
-	perf_evlist__set_id_pos(evlist);
+	if (use_sample_identifier || opts->incompatible_sample_types)
+		perf_evlist__set_id_pos(evlist);
+	else
+		perf_evlist__make_sample_types_compatible(evlist);
 }
Signature · 1 line
1.7.11.7