Skip to contentSkip to conversation

perf tools: Assorted fixes and cleanups for namespace events

PATCH Applied Namhyung Kim 3 participants 15 messages May 22, 2019 – May 30, 2019 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

cover perf tools: Assorted fixes and cleanups for namespace events

Hello,

These are some missing pieces that I found during the code-reading.

Thanks,
Namhyung

Namhyung Kim (3):
perf tools: Protect reading thread's namespace perf tools: Add missing swap ops for namespace events perf top: Enable --namespaces option

 tools/perf/Documentation/perf-top.txt |  5 +++++
 tools/perf/builtin-top.c              |  5 +++++
 tools/perf/util/session.c             | 21 +++++++++++++++++++++
 tools/perf/util/thread.c              | 15 +++++++++++++--
 4 files changed, 44 insertions(+), 2 deletions(-)
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
author maintainer
#1 lore raw

PATCH 1/3 perf tools: Protect reading thread's namespace

It seems that the current code lacks holding the namespace lock in thread__namespaces(). Otherwise it can see inconsistent results.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/thread.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
tools/perf/util/thread.c +13 -2
index 403045a2bbea..b413ba5b9835 100644
@@ -133,7 +133,7 @@ void thread__put(struct thread *thread)
 	}
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+static struct namespaces *__thread__namespaces(const struct thread *thread)
 {
 	if (list_empty(&thread->namespaces_list))
 		return NULL;

@@ -141,10 +141,21 @@ struct namespaces *thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
 
+struct namespaces *thread__namespaces(const struct thread *thread)
+{
+	struct namespaces *ns;
+
+	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	ns = __thread__namespaces(thread);
+	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+
+	return ns;
+}
+
 static int __thread__set_namespaces(struct thread *thread, u64 timestamp,
 				    struct namespaces_event *event)
 {
-	struct namespaces *new, *curr = thread__namespaces(thread);
+	struct namespaces *new, *curr = __thread__namespaces(thread);
 
 	new = namespaces__new(event);
 	if (!new)
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
#2 lore raw

Re: [PATCH 1/3] perf tools: Protect reading thread's namespace

Em Wed, May 22, 2019 at 02:32:48PM +0900, Namhyung Kim escreveu:

Quoted 23 lines

It seems that the current code lacks holding the namespace lock in thread__namespaces(). Otherwise it can see inconsistent results.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/thread.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
tools/perf/util/thread.c +1 -1
index 403045a2bbea..b413ba5b9835 100644
@@ -133,7 +133,7 @@ void thread__put(struct thread *thread)
 	}
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+static struct namespaces *__thread__namespaces(const struct thread *thread)
 {
 	if (list_empty(&thread->namespaces_list))
 		return NULL;

@@ -141,10 +141,21 @@ struct namespaces *thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);

tools/perf/util/thread.c struct namespaces *thread__namespaces(const struct thread *thread) · line 151

Quoted 10 lines
tools/perf/util/thread.c +8 -0
 }
 
+struct namespaces *thread__namespaces(const struct thread *thread)
+{
+	struct namespaces *ns;
+
+	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	ns = __thread__namespaces(thread);
+	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+

Humm, so we need to change thread__namespaces() to remove that const instead of throwing it away with that cast, right?

- Arnaldo

Quoted 13 lines
tools/perf/util/thread.c +4 -1
+	return ns;
+}
+
 static int __thread__set_namespaces(struct thread *thread, u64 timestamp,
 				    struct namespaces_event *event)
 {
-	struct namespaces *new, *curr = thread__namespaces(thread);
+	struct namespaces *new, *curr = __thread__namespaces(thread);
 
 	new = namespaces__new(event);
 	if (!new)
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
Signature · 1 line
- Arnaldo
author maintainer
#3 lore raw

Re: [PATCH 1/3] perf tools: Protect reading thread's namespace

Hi Arnaldo,

Quoted 26 lines · Arnaldo Carvalho de Melo

Em Wed, May 22, 2019 at 02:32:48PM +0900, Namhyung Kim escreveu:

Quoted 25 lines

It seems that the current code lacks holding the namespace lock in thread__namespaces(). Otherwise it can see inconsistent results.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/thread.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
tools/perf/util/thread.c +1 -1
index 403045a2bbea..b413ba5b9835 100644
@@ -133,7 +133,7 @@ void thread__put(struct thread *thread)
 	}
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+static struct namespaces *__thread__namespaces(const struct thread *thread)
 {
 	if (list_empty(&thread->namespaces_list))
 		return NULL;

@@ -141,10 +141,21 @@ struct namespaces *thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
Quoted 11 lines

tools/perf/util/thread.c struct namespaces *thread__namespaces(const struct thread *thread) · line 151

Quoted 8 lines
tools/perf/util/thread.c +8 -0
+struct namespaces *thread__namespaces(const struct thread *thread)
+{
+	struct namespaces *ns;
+
+	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	ns = __thread__namespaces(thread);
+	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+

Humm, so we need to change thread__namespaces() to remove that const instead of throwing it away with that cast, right?

Yes, that's possible too. Note that thread__comm_str() has the same issue as well.

Thanks,
Namhyung

author maintainer
#4 lore raw

[PATCH 4/3] perf tools: Remove const from thread read accessors

The namespaces and comm fields of a thread are protected by rwsem and require write access for it. So it ended up using a cast to remove the const qualifier. Let's get rid of the const then.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/hist.c   |  2 +-
 tools/perf/util/thread.c | 12 ++++++------
 tools/perf/util/thread.h |  4 ++--
 3 files changed, 9 insertions(+), 9 deletions(-)
tools/perf/util/hist.c +1 -1
index 7ace7a10054d..fb3271fd420c 100644
@@ -2561,7 +2561,7 @@ int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh
 	char unit;
 	int printed;
 	const struct dso *dso = hists->dso_filter;
-	const struct thread *thread = hists->thread_filter;
+	struct thread *thread = hists->thread_filter;
 	int socket_id = hists->socket_filter;
 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
 	u64 nr_events = hists->stats.total_period;
tools/perf/util/thread.c +6 -6
index b413ba5b9835..aab7807d445f 100644
@@ -141,13 +141,13 @@ static struct namespaces *__thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+struct namespaces *thread__namespaces(struct thread *thread)
 {
 	struct namespaces *ns;
 
-	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	down_read(&thread->namespaces_lock);
 	ns = __thread__namespaces(thread);
-	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+	up_read(&thread->namespaces_lock);
 
 	return ns;
 }

@@ -271,13 +271,13 @@ static const char *__thread__comm_str(const struct thread *thread)
 	return comm__str(comm);
 }
 
-const char *thread__comm_str(const struct thread *thread)
+const char *thread__comm_str(struct thread *thread)
 {
 	const char *str;
 
-	down_read((struct rw_semaphore *)&thread->comm_lock);
+	down_read(&thread->comm_lock);
 	str = __thread__comm_str(thread);
-	up_read((struct rw_semaphore *)&thread->comm_lock);
+	up_read(&thread->comm_lock);
 
 	return str;
 }
tools/perf/util/thread.h +2 -2
index cf8375c017a0..e97ef6977eb9 100644
@@ -76,7 +76,7 @@ static inline void thread__exited(struct thread *thread)
 	thread->dead = true;
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread);
+struct namespaces *thread__namespaces(struct thread *thread);
 int thread__set_namespaces(struct thread *thread, u64 timestamp,
 			   struct namespaces_event *event);
 

@@ -93,7 +93,7 @@ int thread__set_comm_from_proc(struct thread *thread);
 int thread__comm_len(struct thread *thread);
 struct comm *thread__comm(const struct thread *thread);
 struct comm *thread__exec_comm(const struct thread *thread);
-const char *thread__comm_str(const struct thread *thread);
+const char *thread__comm_str(struct thread *thread);
 int thread__insert_map(struct thread *thread, struct map *map);
 int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp, bool do_maps_clone);
 size_t thread__fprintf(struct thread *thread, FILE *fp);
Signature · 1 line
2.22.0.rc1.257.g3120a18244-goog
Applied
#5 lore raw

Re: [PATCH 4/3] perf tools: Remove const from thread read accessors

Em Mon, May 27, 2019 at 03:11:49PM +0900, Namhyung Kim escreveu:

Quoted 3 lines

The namespaces and comm fields of a thread are protected by rwsem and require write access for it. So it ended up using a cast to remove the const qualifier. Let's get rid of the const then.

Thanks, applied 1/3 and 4/3.

- Arnaldo

Quoted 82 lines
Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/hist.c   |  2 +-
 tools/perf/util/thread.c | 12 ++++++------
 tools/perf/util/thread.h |  4 ++--
 3 files changed, 9 insertions(+), 9 deletions(-)
tools/perf/util/hist.c +1 -1
index 7ace7a10054d..fb3271fd420c 100644
@@ -2561,7 +2561,7 @@ int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh
 	char unit;
 	int printed;
 	const struct dso *dso = hists->dso_filter;
-	const struct thread *thread = hists->thread_filter;
+	struct thread *thread = hists->thread_filter;
 	int socket_id = hists->socket_filter;
 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
 	u64 nr_events = hists->stats.total_period;
tools/perf/util/thread.c +6 -6
index b413ba5b9835..aab7807d445f 100644
@@ -141,13 +141,13 @@ static struct namespaces *__thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+struct namespaces *thread__namespaces(struct thread *thread)
 {
 	struct namespaces *ns;
 
-	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	down_read(&thread->namespaces_lock);
 	ns = __thread__namespaces(thread);
-	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+	up_read(&thread->namespaces_lock);
 
 	return ns;
 }

@@ -271,13 +271,13 @@ static const char *__thread__comm_str(const struct thread *thread)
 	return comm__str(comm);
 }
 
-const char *thread__comm_str(const struct thread *thread)
+const char *thread__comm_str(struct thread *thread)
 {
 	const char *str;
 
-	down_read((struct rw_semaphore *)&thread->comm_lock);
+	down_read(&thread->comm_lock);
 	str = __thread__comm_str(thread);
-	up_read((struct rw_semaphore *)&thread->comm_lock);
+	up_read(&thread->comm_lock);
 
 	return str;
 }
tools/perf/util/thread.h +2 -2
index cf8375c017a0..e97ef6977eb9 100644
@@ -76,7 +76,7 @@ static inline void thread__exited(struct thread *thread)
 	thread->dead = true;
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread);
+struct namespaces *thread__namespaces(struct thread *thread);
 int thread__set_namespaces(struct thread *thread, u64 timestamp,
 			   struct namespaces_event *event);
 

@@ -93,7 +93,7 @@ int thread__set_comm_from_proc(struct thread *thread);
 int thread__comm_len(struct thread *thread);
 struct comm *thread__comm(const struct thread *thread);
 struct comm *thread__exec_comm(const struct thread *thread);
-const char *thread__comm_str(const struct thread *thread);
+const char *thread__comm_str(struct thread *thread);
 int thread__insert_map(struct thread *thread, struct map *map);
 int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp, bool do_maps_clone);
 size_t thread__fprintf(struct thread *thread, FILE *fp);
Signature · 1 line
2.22.0.rc1.257.g3120a18244-goog
Signature · 1 line
- Arnaldo
merge bot
#6 lore raw

[tip:perf/core] perf tools: Remove const from thread read accessors

In reply to Namhyung Kim
perf tools: Remove const from thread read accessors
Commit-ID:  7cb10a08df98e643b87d4bc8422e50e9c43b5c60
Gitweb:     https://git.kernel.org/tip/7cb10a08df98e643b87d4bc8422e50e9c43b5c60
Author:     Namhyung Kim <[email protected]>
AuthorDate: Mon, 27 May 2019 15:11:49 +0900
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 28 May 2019 18:37:43 -0300

perf tools: Remove const from thread read accessors

The namespaces and comm fields of a thread are protected by rwsem and require write access for it. So it ended up using a cast to remove the const qualifier. Let's get rid of the const then.

Signed-off-by Namhyung Kim <[email protected]>
Suggested-by Arnaldo Carvalho de Melo <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Krister Johansen <[email protected]>
Link http://lkml.kernel.org/r/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/hist.c   |  2 +-
 tools/perf/util/thread.c | 12 ++++++------
 tools/perf/util/thread.h |  4 ++--
 3 files changed, 9 insertions(+), 9 deletions(-)
tools/perf/util/hist.c +1 -1
index 7ace7a10054d..fb3271fd420c 100644
@@ -2561,7 +2561,7 @@ int __hists__scnprintf_title(struct hists *hists, char *bf, size_t size, bool sh
 	char unit;
 	int printed;
 	const struct dso *dso = hists->dso_filter;
-	const struct thread *thread = hists->thread_filter;
+	struct thread *thread = hists->thread_filter;
 	int socket_id = hists->socket_filter;
 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
 	u64 nr_events = hists->stats.total_period;
tools/perf/util/thread.c +6 -6
index b413ba5b9835..aab7807d445f 100644
@@ -141,13 +141,13 @@ static struct namespaces *__thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+struct namespaces *thread__namespaces(struct thread *thread)
 {
 	struct namespaces *ns;
 
-	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	down_read(&thread->namespaces_lock);
 	ns = __thread__namespaces(thread);
-	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+	up_read(&thread->namespaces_lock);
 
 	return ns;
 }

@@ -271,13 +271,13 @@ static const char *__thread__comm_str(const struct thread *thread)
 	return comm__str(comm);
 }
 
-const char *thread__comm_str(const struct thread *thread)
+const char *thread__comm_str(struct thread *thread)
 {
 	const char *str;
 
-	down_read((struct rw_semaphore *)&thread->comm_lock);
+	down_read(&thread->comm_lock);
 	str = __thread__comm_str(thread);
-	up_read((struct rw_semaphore *)&thread->comm_lock);
+	up_read(&thread->comm_lock);
 
 	return str;
 }
tools/perf/util/thread.h +2 -2
index cf8375c017a0..e97ef6977eb9 100644
@@ -76,7 +76,7 @@ static inline void thread__exited(struct thread *thread)
 	thread->dead = true;
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread);
+struct namespaces *thread__namespaces(struct thread *thread);
 int thread__set_namespaces(struct thread *thread, u64 timestamp,
 			   struct namespaces_event *event);
 

@@ -93,7 +93,7 @@ int thread__set_comm_from_proc(struct thread *thread);
 int thread__comm_len(struct thread *thread);
 struct comm *thread__comm(const struct thread *thread);
 struct comm *thread__exec_comm(const struct thread *thread);
-const char *thread__comm_str(const struct thread *thread);
+const char *thread__comm_str(struct thread *thread);
 int thread__insert_map(struct thread *thread, struct map *map);
 int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp, bool do_maps_clone);
 size_t thread__fprintf(struct thread *thread, FILE *fp);
merge bot
#7 lore raw

[tip:perf/urgent] perf namespace: Protect reading thread's namespace

In reply to Namhyung Kim
perf namespace: Protect reading thread's namespace
Commit-ID:  6584140ba9e6762dd7ec73795243289b914f31f9
Gitweb:     https://git.kernel.org/tip/6584140ba9e6762dd7ec73795243289b914f31f9
Author:     Namhyung Kim <[email protected]>
AuthorDate: Wed, 22 May 2019 14:32:48 +0900
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 28 May 2019 09:52:23 -0300

perf namespace: Protect reading thread's namespace

It seems that the current code lacks holding the namespace lock in thread__namespaces(). Otherwise it can see inconsistent results.

Signed-off-by Namhyung Kim <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Krister Johansen <[email protected]>
Link http://lkml.kernel.org/r/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/thread.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
tools/perf/util/thread.c +13 -2
index 403045a2bbea..b413ba5b9835 100644
@@ -133,7 +133,7 @@ void thread__put(struct thread *thread)
 	}
 }
 
-struct namespaces *thread__namespaces(const struct thread *thread)
+static struct namespaces *__thread__namespaces(const struct thread *thread)
 {
 	if (list_empty(&thread->namespaces_list))
 		return NULL;

@@ -141,10 +141,21 @@ struct namespaces *thread__namespaces(const struct thread *thread)
 	return list_first_entry(&thread->namespaces_list, struct namespaces, list);
 }
 
+struct namespaces *thread__namespaces(const struct thread *thread)
+{
+	struct namespaces *ns;
+
+	down_read((struct rw_semaphore *)&thread->namespaces_lock);
+	ns = __thread__namespaces(thread);
+	up_read((struct rw_semaphore *)&thread->namespaces_lock);
+
+	return ns;
+}
+
 static int __thread__set_namespaces(struct thread *thread, u64 timestamp,
 				    struct namespaces_event *event)
 {
-	struct namespaces *new, *curr = thread__namespaces(thread);
+	struct namespaces *new, *curr = __thread__namespaces(thread);
 
 	new = namespaces__new(event);
 	if (!new)
author maintainer
#8 lore raw

PATCH 2/3 perf tools: Add missing swap ops for namespace events

In reply to Namhyung Kim

In case it's recorded from other arch.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/session.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
tools/perf/util/session.c +21 -0
index 2310a1752983..54cf163347f7 100644
@@ -647,6 +647,26 @@ static void perf_event__throttle_swap(union perf_event *event,
 		swap_sample_id_all(event, &event->throttle + 1);
 }
 
+static void perf_event__namespaces_swap(union perf_event *event,
+					bool sample_id_all)
+{
+	u64 i;
+
+	event->namespaces.pid		= bswap_32(event->namespaces.pid);
+	event->namespaces.tid		= bswap_32(event->namespaces.tid);
+	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
+
+	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
+		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
+
+		ns->dev = bswap_64(ns->dev);
+		ns->ino = bswap_64(ns->ino);
+	}
+
+	if (sample_id_all)
+		swap_sample_id_all(event, &event->namespaces.link_info[i]);
+}
+
 static u8 revbyte(u8 b)
 {
 	int rev = (b >> 4) | ((b & 0xf) << 4);

@@ -887,6 +907,7 @@ static perf_event__swap_op perf_event__swap_ops[] = {
 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
+	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
#9 lore raw

Re: [PATCH 2/3] perf tools: Add missing swap ops for namespace events

Em Wed, May 22, 2019 at 02:32:49PM +0900, Namhyung Kim escreveu:

Quoted 1 line

In case it's recorded from other arch.

Applied and added:

Fixes f3b3614a284d ("perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info")

So that the stable guys can pick it.

- Arnaldo

Quoted 46 lines
Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/util/session.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
tools/perf/util/session.c +21 -0
index 2310a1752983..54cf163347f7 100644
@@ -647,6 +647,26 @@ static void perf_event__throttle_swap(union perf_event *event,
 		swap_sample_id_all(event, &event->throttle + 1);
 }
 
+static void perf_event__namespaces_swap(union perf_event *event,
+					bool sample_id_all)
+{
+	u64 i;
+
+	event->namespaces.pid		= bswap_32(event->namespaces.pid);
+	event->namespaces.tid		= bswap_32(event->namespaces.tid);
+	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
+
+	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
+		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
+
+		ns->dev = bswap_64(ns->dev);
+		ns->ino = bswap_64(ns->ino);
+	}
+
+	if (sample_id_all)
+		swap_sample_id_all(event, &event->namespaces.link_info[i]);
+}
+
 static u8 revbyte(u8 b)
 {
 	int rev = (b >> 4) | ((b & 0xf) << 4);

@@ -887,6 +907,7 @@ static perf_event__swap_op perf_event__swap_ops[] = {
 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
+	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
Signature · 1 line
- Arnaldo
merge bot
#10 lore raw

[tip:perf/urgent] perf session: Add missing swap ops for namespace events

In reply to Namhyung Kim
perf session: Add missing swap ops for namespace events
Commit-ID:  acd244b84b80d53fa2cee98659b55d3f09b4f5a7
Gitweb:     https://git.kernel.org/tip/acd244b84b80d53fa2cee98659b55d3f09b4f5a7
Author:     Namhyung Kim <[email protected]>
AuthorDate: Wed, 22 May 2019 14:32:49 +0900
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 28 May 2019 09:52:23 -0300

perf session: Add missing swap ops for namespace events

In case it's recorded in a different arch.

Signed-off-by Namhyung Kim <[email protected]>
Cc Hari Bathini <[email protected]> <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Krister Johansen <[email protected]>
Fixes f3b3614a284d ("perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info")
Link http://lkml.kernel.org/r/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/util/session.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
tools/perf/util/session.c +21 -0
index 2310a1752983..54cf163347f7 100644
@@ -647,6 +647,26 @@ static void perf_event__throttle_swap(union perf_event *event,
 		swap_sample_id_all(event, &event->throttle + 1);
 }
 
+static void perf_event__namespaces_swap(union perf_event *event,
+					bool sample_id_all)
+{
+	u64 i;
+
+	event->namespaces.pid		= bswap_32(event->namespaces.pid);
+	event->namespaces.tid		= bswap_32(event->namespaces.tid);
+	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
+
+	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
+		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
+
+		ns->dev = bswap_64(ns->dev);
+		ns->ino = bswap_64(ns->ino);
+	}
+
+	if (sample_id_all)
+		swap_sample_id_all(event, &event->namespaces.link_info[i]);
+}
+
 static u8 revbyte(u8 b)
 {
 	int rev = (b >> 4) | ((b & 0xf) << 4);

@@ -887,6 +907,7 @@ static perf_event__swap_op perf_event__swap_ops[] = {
 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
+	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
author maintainer
#11 lore raw

PATCH 3/3 perf top: Enable --namespaces option

In reply to Namhyung Kim

Since perf record already have the option, let's have it for perf top as well.

Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/Documentation/perf-top.txt | 5 +++++
 tools/perf/builtin-top.c              | 5 +++++
 2 files changed, 10 insertions(+)
tools/perf/Documentation/perf-top.txt +5 -0
index 44d89fb9c788..cfea87c6f38e 100644
@@ -262,6 +262,11 @@ Default is to monitor all CPUS.
 	The number of threads to run when synthesizing events for existing processes.
 	By default, the number of threads equals to the number of online CPUs.
 
+--namespaces::
+	Record events of type PERF_RECORD_NAMESPACES and display it with the
+	'cgroup_id' sort key.
+
+
 INTERACTIVE PROMPTING KEYS
 --------------------------
tools/perf/builtin-top.c +5 -0
index 3648ef576996..6651377fd762 100644
@@ -1208,6 +1208,9 @@ static int __cmd_top(struct perf_top *top)
 
 	init_process_thread(top);
 
+	if (opts->record_namespaces)
+		top->tool.namespace_events = true;
+
 	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
 						&top->session->machines.host,
 						&top->record_opts);

@@ -1500,6 +1503,8 @@ int cmd_top(int argc, const char **argv)
 	OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
 	OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
 			"number of thread to run event synthesize"),
+	OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
+		    "Record namespaces events"),
 	OPT_END()
 	};
 	struct perf_evlist *sb_evlist = NULL;
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
#12 lore raw

Re: [PATCH 3/3] perf top: Enable --namespaces option

Em Wed, May 22, 2019 at 02:32:50PM +0900, Namhyung Kim escreveu:

Quoted 2 lines

Since perf record already have the option, let's have it for perf top as well.

I'm applying, but I wonder if this shouldn't be the default...

- Arnaldo

Quoted 47 lines
Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/Documentation/perf-top.txt | 5 +++++
 tools/perf/builtin-top.c              | 5 +++++
 2 files changed, 10 insertions(+)
tools/perf/Documentation/perf-top.txt +5 -0
index 44d89fb9c788..cfea87c6f38e 100644
@@ -262,6 +262,11 @@ Default is to monitor all CPUS.
 	The number of threads to run when synthesizing events for existing processes.
 	By default, the number of threads equals to the number of online CPUs.
 
+--namespaces::
+	Record events of type PERF_RECORD_NAMESPACES and display it with the
+	'cgroup_id' sort key.
+
+
 INTERACTIVE PROMPTING KEYS
 --------------------------
tools/perf/builtin-top.c +5 -0
index 3648ef576996..6651377fd762 100644
@@ -1208,6 +1208,9 @@ static int __cmd_top(struct perf_top *top)
 
 	init_process_thread(top);
 
+	if (opts->record_namespaces)
+		top->tool.namespace_events = true;
+
 	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
 						&top->session->machines.host,
 						&top->record_opts);

@@ -1500,6 +1503,8 @@ int cmd_top(int argc, const char **argv)
 	OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
 	OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
 			"number of thread to run event synthesize"),
+	OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
+		    "Record namespaces events"),
 	OPT_END()
 	};
 	struct perf_evlist *sb_evlist = NULL;
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
Signature · 1 line
- Arnaldo
author maintainer
#13 lore raw

Re: [PATCH 3/3] perf top: Enable --namespaces option

Quoted 5 lines · Arnaldo Carvalho de Melo

Em Wed, May 22, 2019 at 02:32:50PM +0900, Namhyung Kim escreveu:

Quoted 2 lines

Since perf record already have the option, let's have it for perf top as well.

I'm applying, but I wonder if this shouldn't be the default...

Not sure, it'll only add a bit of overhead to task and/or namespace creation.

Thanks,
Namhyung

Quoted 54 lines

- Arnaldo

Quoted 47 lines
Signed-off-by Namhyung Kim <[email protected]>
---
 tools/perf/Documentation/perf-top.txt | 5 +++++
 tools/perf/builtin-top.c              | 5 +++++
 2 files changed, 10 insertions(+)
tools/perf/Documentation/perf-top.txt +5 -0
index 44d89fb9c788..cfea87c6f38e 100644
@@ -262,6 +262,11 @@ Default is to monitor all CPUS.
 	The number of threads to run when synthesizing events for existing processes.
 	By default, the number of threads equals to the number of online CPUs.
 
+--namespaces::
+	Record events of type PERF_RECORD_NAMESPACES and display it with the
+	'cgroup_id' sort key.
+
+
 INTERACTIVE PROMPTING KEYS
 --------------------------
tools/perf/builtin-top.c +5 -0
index 3648ef576996..6651377fd762 100644
@@ -1208,6 +1208,9 @@ static int __cmd_top(struct perf_top *top)
 
 	init_process_thread(top);
 
+	if (opts->record_namespaces)
+		top->tool.namespace_events = true;
+
 	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
 						&top->session->machines.host,
 						&top->record_opts);

@@ -1500,6 +1503,8 @@ int cmd_top(int argc, const char **argv)
 	OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
 	OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
 			"number of thread to run event synthesize"),
+	OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
+		    "Record namespaces events"),
 	OPT_END()
 	};
 	struct perf_evlist *sb_evlist = NULL;
Signature · 1 line
2.21.0.1020.gf2820cf01a-goog
Signature · 1 line
- Arnaldo
merge bot
#14 lore raw

[tip:perf/core] perf top: Add --namespaces option

In reply to Namhyung Kim
perf top: Add --namespaces option
Commit-ID:  a0c0a4ac021b017e385d0328541ccfebeef165fc
Gitweb:     https://git.kernel.org/tip/a0c0a4ac021b017e385d0328541ccfebeef165fc
Author:     Namhyung Kim <[email protected]>
AuthorDate: Wed, 22 May 2019 14:32:50 +0900
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 28 May 2019 18:37:43 -0300

perf top: Add --namespaces option

Since 'perf record' already have this option, let's have it for 'perf top' as well.

Signed-off-by Namhyung Kim <[email protected]>
Cc Hari Bathini <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Krister Johansen <[email protected]>
Link http://lkml.kernel.org/r/[email protected] open here
Signed-off-by Arnaldo Carvalho de Melo <[email protected]>
---
 tools/perf/Documentation/perf-top.txt | 5 +++++
 tools/perf/builtin-top.c              | 5 +++++
 2 files changed, 10 insertions(+)
tools/perf/Documentation/perf-top.txt +5 -0
index 44d89fb9c788..cfea87c6f38e 100644
@@ -262,6 +262,11 @@ Default is to monitor all CPUS.
 	The number of threads to run when synthesizing events for existing processes.
 	By default, the number of threads equals to the number of online CPUs.
 
+--namespaces::
+	Record events of type PERF_RECORD_NAMESPACES and display it with the
+	'cgroup_id' sort key.
+
+
 INTERACTIVE PROMPTING KEYS
 --------------------------
tools/perf/builtin-top.c +5 -0
index fbbb0da43abb..31d78d874fc7 100644
@@ -1208,6 +1208,9 @@ static int __cmd_top(struct perf_top *top)
 
 	init_process_thread(top);
 
+	if (opts->record_namespaces)
+		top->tool.namespace_events = true;
+
 	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
 						&top->session->machines.host,
 						&top->record_opts);

@@ -1500,6 +1503,8 @@ int cmd_top(int argc, const char **argv)
 	OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
 	OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
 			"number of thread to run event synthesize"),
+	OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
+		    "Record namespaces events"),
 	OPT_END()
 	};
 	struct perf_evlist *sb_evlist = NULL;