Skip to contentSkip to conversation

ftrace: fix perf-tracepoint OOPS

PATCH In review Peter Zijlstra 5 participants 11 messages Aug 5, 2009 – Aug 7, 2009

Subsystems (3)
perf events Tracing ftrace
lore
Share & tools
lore

Peter Zijlstra

author maintainer
#0 lore raw

PATCH 1/1 ftrace: fix perf-tracepoint OOPS

Not all tracepoints are created equal, in specific the ftrace tracepoints are created with TRACE_EVENT_FORMAT() which does not generate the needed bits to tie them into perf counters.

For those events, don't create the 'id' file and fail ->profile_enable when their ID is specified through other means.

Reported-by Chris Mason <[email protected]>
Signed-off-by Peter Zijlstra <[email protected]>
---
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
kernel/trace/trace_event_profile.c +1 -1
index 5b5895a..11ba5bb 100644
@@ -14,7 +14,7 @@ int ftrace_profile_enable(int event_id)
 
 	mutex_lock(&event_mutex);
 	list_for_each_entry(event, &ftrace_events, list) {
-		if (event->id == event_id) {
+		if (event->id == event_id && event->profile_enable) {
 			ret = event->profile_enable(event);
 			break;
 		}
kernel/trace/trace_events.c +1 -1
index 90cdec5..e2b301d 100644
@@ -971,7 +971,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
 
-	if (call->id)
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);
author maintainer
#1 lore raw

[tip:perfcounters/urgent] ftrace: Fix perf-tracepoint OOPS

Commit-ID:  5fe31b14cc8eccc06311c3cc5674420a66431372
Gitweb:     http://git.kernel.org/tip/5fe31b14cc8eccc06311c3cc5674420a66431372
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Wed, 5 Aug 2009 20:41:04 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 5 Aug 2009 20:45:52 +0200

ftrace: Fix perf-tracepoint OOPS

Not all tracepoints are created equal, in specific the ftrace tracepoints are created with TRACE_EVENT_FORMAT() which does not generate the needed bits to tie them into perf counters.

For those events, don't create the 'id' file and fail ->profile_enable when their ID is specified through other means.

Reported-by Chris Mason <[email protected]>
Signed-off-by Peter Zijlstra <[email protected]>
Cc Steven Rostedt <[email protected]>

LKML-Reference: <1249497664.5890.4.camel@laptop>

Signed-off-by Ingo Molnar <[email protected]>
---
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
kernel/trace/trace_event_profile.c +1 -1
index 5b5895a..11ba5bb 100644
@@ -14,7 +14,7 @@ int ftrace_profile_enable(int event_id)
 
 	mutex_lock(&event_mutex);
 	list_for_each_entry(event, &ftrace_events, list) {
-		if (event->id == event_id) {
+		if (event->id == event_id && event->profile_enable) {
 			ret = event->profile_enable(event);
 			break;
 		}
kernel/trace/trace_events.c +1 -1
index 23d2972..e75276a 100644
@@ -940,7 +940,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
 
-	if (call->id)
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);
maintainer Acked-by on v1 1/1
#2 lore raw
In reply to Peter Zijlstra
Quoted 9 lines · Peter Zijlstra

Not all tracepoints are created equal, in specific the ftrace tracepoints are created with TRACE_EVENT_FORMAT() which does not generate the needed bits to tie them into perf counters.

For those events, don't create the 'id' file and fail ->profile_enable when their ID is specified through other means.

Reported-by Chris Mason <[email protected]>
Signed-off-by Peter Zijlstra <[email protected]>
Acked-by Steven Rostedt <[email protected]>

-- Steve

Quoted 34 lines
---
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
kernel/trace/trace_event_profile.c +1 -1
index 5b5895a..11ba5bb 100644
@@ -14,7 +14,7 @@ int ftrace_profile_enable(int event_id)
 
 	mutex_lock(&event_mutex);
 	list_for_each_entry(event, &ftrace_events, list) {
-		if (event->id == event_id) {
+		if (event->id == event_id && event->profile_enable) {
 			ret = event->profile_enable(event);
 			break;
 		}
kernel/trace/trace_events.c +1 -1
index 90cdec5..e2b301d 100644
@@ -971,7 +971,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
 
-	if (call->id)
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);
author maintainer
#3 lore raw

PATCH 1/1 perf: Don't list tracepoints without an id

In reply to Peter Zijlstra

Stop perf list from displaying tracepoints without an id file.

Signed-off-by Peter Zijlstra <[email protected]>
---
 tools/perf/util/parse-events.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
tools/perf/util/parse-events.c +10 -0
index 7bdad8d..d00636b 100644
@@ -568,6 +568,7 @@ static void print_tracepoint_events(void)
 	struct dirent *sys_next, *evt_next, sys_dirent, evt_dirent;
 	struct stat st;
 	char evt_path[MAXPATHLEN];
+	int fd;
 
 	if (valid_debugfs_mount(debugfs_path))
 		return;

@@ -582,6 +583,15 @@ static void print_tracepoint_events(void)
 			goto cleanup;
 		for_each_event(sys_dirent, evt_dir, evt_dirent, evt_next,
 								evt_path, st) {
+
+			snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id",
+					debugfs_path, 
+					sys_dirent.d_name, evt_dirent.d_name);
+			fd = open(evt_path, O_RDONLY);
+			if (fd < 0)
+				continue;
+			close(fd);
+
 			snprintf(evt_path, MAXPATHLEN, "%s:%s",
 				 sys_dirent.d_name, evt_dirent.d_name);
 			fprintf(stderr, "  %-40s [%s]\n", evt_path,
#4 lore raw

Re: [PATCH] perf: Don't list tracepoints without an id

Quoted 27 lines · Peter Zijlstra

Stop perf list from displaying tracepoints without an id file.

Signed-off-by Peter Zijlstra <[email protected]>
---
 tools/perf/util/parse-events.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
tools/perf/util/parse-events.c +3 -0
index 7bdad8d..d00636b 100644
@@ -568,6 +568,7 @@ static void print_tracepoint_events(void)
 	struct dirent *sys_next, *evt_next, sys_dirent, evt_dirent;
 	struct stat st;
 	char evt_path[MAXPATHLEN];
+	int fd;
 
 	if (valid_debugfs_mount(debugfs_path))
 		return;

@@ -582,6 +583,15 @@ static void print_tracepoint_events(void)
 			goto cleanup;
 		for_each_event(sys_dirent, evt_dir, evt_dirent, evt_next,
 								evt_path, st) {
+
+			snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id",

tools/perf/util/parse-events.c static void print_tracepoint_events(void) · line 597

Quoted 10 lines
tools/perf/util/parse-events.c +7 -0
+					debugfs_path, 
+					sys_dirent.d_name, evt_dirent.d_name);
+			fd = open(evt_path, O_RDONLY);
+			if (fd < 0)
+				continue;
+			close(fd);
+
 			snprintf(evt_path, MAXPATHLEN, "%s:%s",
 				 sys_dirent.d_name, evt_dirent.d_name);
 			fprintf(stderr, "  %-40s [%s]\n", evt_path,

hmmm, might be nicer to add this check to the 'for_each_event()' macro instead...

thanks,

-Jason

author maintainer
#5 lore raw

PATCH 1/1 perf: Don't list tracepoints without an id

Stop perf list from displaying tracepoints without an id file.

Signed-off-by Peter Zijlstra <[email protected]>
---
 tools/perf/util/parse-events.c |   18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)
tools/perf/util/parse-events.c +17 -1
Index: linux-2.6/tools/perf/util/parse-events.c
===================================================================
@@ -121,13 +121,29 @@ static unsigned long hw_cache_stat[C(MAX
 	   (strcmp(sys_dirent.d_name, ".")) &&				       \
 	   (strcmp(sys_dirent.d_name, "..")))
 
+static int tp_event_has_id(struct dirent *sys_dir, struct dirent *evt_dir)
+{
+	char evt_path[MAXPATHLEN];
+	int fd;
+
+	snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path,
+			sys_dir->d_name, evt_dir->d_name);
+	fd = open(evt_path, O_RDONLY);
+	if (fd < 0)
+		return -EINVAL;
+	close(fd);
+
+	return 0;
+}
+
 #define for_each_event(sys_dirent, evt_dir, evt_dirent, evt_next, file, st)    \
 	while (!readdir_r(evt_dir, &evt_dirent, &evt_next) && evt_next)        \
 	if (snprintf(file, MAXPATHLEN, "%s/%s/%s", debugfs_path,	       \
 		     sys_dirent.d_name, evt_dirent.d_name) &&		       \
 	   (!stat(file, &st)) && (S_ISDIR(st.st_mode)) &&		       \
 	   (strcmp(evt_dirent.d_name, ".")) &&				       \
-	   (strcmp(evt_dirent.d_name, "..")))
+	   (strcmp(evt_dirent.d_name, "..")) &&				       \
+	   (!tp_event_has_id(&sys_dirent, &evt_dirent)))
 
 #define MAX_EVENT_LENGTH 30
Acked-by on v1 1/1
#6 lore raw

Re: [PATCH -v2] perf: Don't list tracepoints without an id

Quoted 32 lines · Peter Zijlstra

Stop perf list from displaying tracepoints without an id file.

Signed-off-by Peter Zijlstra <[email protected]>
---
 tools/perf/util/parse-events.c |   18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

Index: linux-2.6/tools/perf/util/parse-events.c
===================================================================

tools/perf/util/parse-events.c +15 -0
@@ -121,13 +121,29 @@ static unsigned long hw_cache_stat[C(MAX
 	   (strcmp(sys_dirent.d_name, ".")) &&				       \
 	   (strcmp(sys_dirent.d_name, "..")))
 
+static int tp_event_has_id(struct dirent *sys_dir, struct dirent *evt_dir)
+{
+	char evt_path[MAXPATHLEN];
+	int fd;
+
+	snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path,
+			sys_dir->d_name, evt_dir->d_name);
+	fd = open(evt_path, O_RDONLY);
+	if (fd < 0)
+		return -EINVAL;
+	close(fd);
+
+	return 0;
+}
+
 #define for_each_event(sys_dirent, evt_dir, evt_dirent, evt_next, file, st)    \
 	while (!readdir_r(evt_dir, &evt_dirent, &evt_next) && evt_next)        \

tools/perf/util/parse-events.c static unsigned long hw_cache_stat[C(MAX · line 149

Quoted 10 lines
tools/perf/util/parse-events.c +2 -1
 	if (snprintf(file, MAXPATHLEN, "%s/%s/%s", debugfs_path,	       \
 		     sys_dirent.d_name, evt_dirent.d_name) &&		       \
 	   (!stat(file, &st)) && (S_ISDIR(st.st_mode)) &&		       \
 	   (strcmp(evt_dirent.d_name, ".")) &&				       \
-	   (strcmp(evt_dirent.d_name, "..")))
+	   (strcmp(evt_dirent.d_name, "..")) &&				       \
+	   (!tp_event_has_id(&sys_dirent, &evt_dirent)))
 
 #define MAX_EVENT_LENGTH 30

looks good to me.

Acked-by Jason Baron <[email protected]>

thanks,

-Jason

author maintainer
#7 lore raw

[tip:perfcounters/urgent] perf: Don't list tracepoints without an id

In reply to Peter Zijlstra
Commit-ID:  79e372825f775fabd2219c9c45fbdff93c24cd3f
Gitweb:     http://git.kernel.org/tip/79e372825f775fabd2219c9c45fbdff93c24cd3f
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Thu, 6 Aug 2009 16:48:54 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 7 Aug 2009 13:09:10 +0200

perf: Don't list tracepoints without an id

Stop perf list from displaying tracepoints without an id file, those are special tracepoints that are not interfaced to perfcounters.

Signed-off-by Peter Zijlstra <[email protected]>
Acked-by Jason Baron <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Chris Mason <[email protected]>

LKML-Reference: <1249570134.32113.483.camel@twins>

Signed-off-by Ingo Molnar <[email protected]>
---
 tools/perf/util/parse-events.c |   18 +++++++++++++++++-
 1 files changed, 17 insertions(+), 1 deletions(-)
tools/perf/util/parse-events.c +17 -1
index f77407b..4858d83 100644
@@ -121,13 +121,29 @@ static unsigned long hw_cache_stat[C(MAX)] = {
 	   (strcmp(sys_dirent.d_name, ".")) &&				       \
 	   (strcmp(sys_dirent.d_name, "..")))
 
+static int tp_event_has_id(struct dirent *sys_dir, struct dirent *evt_dir)
+{
+	char evt_path[MAXPATHLEN];
+	int fd;
+
+	snprintf(evt_path, MAXPATHLEN, "%s/%s/%s/id", debugfs_path,
+			sys_dir->d_name, evt_dir->d_name);
+	fd = open(evt_path, O_RDONLY);
+	if (fd < 0)
+		return -EINVAL;
+	close(fd);
+
+	return 0;
+}
+
 #define for_each_event(sys_dirent, evt_dir, evt_dirent, evt_next, file, st)    \
 	while (!readdir_r(evt_dir, &evt_dirent, &evt_next) && evt_next)        \
 	if (snprintf(file, MAXPATHLEN, "%s/%s/%s", debugfs_path,	       \
 		     sys_dirent.d_name, evt_dirent.d_name) &&		       \
 	   (!stat(file, &st)) && (S_ISDIR(st.st_mode)) &&		       \
 	   (strcmp(evt_dirent.d_name, ".")) &&				       \
-	   (strcmp(evt_dirent.d_name, "..")))
+	   (strcmp(evt_dirent.d_name, "..")) &&				       \
+	   (!tp_event_has_id(&sys_dirent, &evt_dirent)))
 
 #define MAX_EVENT_LENGTH 30
author maintainer
#8 lore raw

[tip:perfcounters/urgent] ftrace: Fix perf-tracepoint OOPS

In reply to Peter Zijlstra
Commit-ID:  af6af30c0fcd77e621638e53ef8b176bca8bd3b4
Gitweb:     http://git.kernel.org/tip/af6af30c0fcd77e621638e53ef8b176bca8bd3b4
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Wed, 5 Aug 2009 20:41:04 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 6 Aug 2009 06:26:09 +0200

ftrace: Fix perf-tracepoint OOPS

Not all tracepoints are created equal, in specific the ftrace tracepoints are created with TRACE_EVENT_FORMAT() which does not generate the needed bits to tie them into perf counters.

For those events, don't create the 'id' file and fail ->profile_enable when their ID is specified through other means.

Reported-by Chris Mason <[email protected]>
Signed-off-by Peter Zijlstra <[email protected]>
Cc Steven Rostedt <[email protected]>

LKML-Reference: <1249497664.5890.4.camel@laptop>

[ v2: fix build error in the !CONFIG_EVENT_PROFILE case ]
Signed-off-by Ingo Molnar <[email protected]>
---
 include/linux/ftrace_event.h       |    8 +++-----
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 3 files changed, 5 insertions(+), 7 deletions(-)
include/linux/ftrace_event.h +3 -5
index 5c093ff..d7cd193 100644
@@ -119,11 +119,9 @@ struct ftrace_event_call {
 	void			*filter;
 	void			*mod;
 
-#ifdef CONFIG_EVENT_PROFILE
-	atomic_t	profile_count;
-	int		(*profile_enable)(struct ftrace_event_call *);
-	void		(*profile_disable)(struct ftrace_event_call *);
-#endif
+	atomic_t		profile_count;
+	int			(*profile_enable)(struct ftrace_event_call *);
+	void			(*profile_disable)(struct ftrace_event_call *);
 };
 
 #define MAX_FILTER_PRED		32
kernel/trace/trace_event_profile.c +1 -1
index 5b5895a..11ba5bb 100644
@@ -14,7 +14,7 @@ int ftrace_profile_enable(int event_id)
 
 	mutex_lock(&event_mutex);
 	list_for_each_entry(event, &ftrace_events, list) {
-		if (event->id == event_id) {
+		if (event->id == event_id && event->profile_enable) {
 			ret = event->profile_enable(event);
 			break;
 		}
kernel/trace/trace_events.c +1 -1
index 23d2972..e75276a 100644
@@ -940,7 +940,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
 
-	if (call->id)
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);
#9 lore raw

Re: [tip:perfcounters/urgent] ftrace: Fix perf-tracepoint OOPS

Quoted 18 lines
[ v2: fix build error in the !CONFIG_EVENT_PROFILE case ]
Signed-off-by Ingo Molnar <[email protected]>
---
 include/linux/ftrace_event.h       |    8 +++-----
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 3 files changed, 5 insertions(+), 7 deletions(-)
include/linux/ftrace_event.h +0 -0
index 5c093ff..d7cd193 100644
@@ -119,11 +119,9 @@ struct ftrace_event_call {
 	void			*filter;
 	void			*mod;

include/linux/ftrace_event.h struct ftrace_event_call { · line 125

Quoted 10 lines
include/linux/ftrace_event.h +3 -5
 
-#ifdef CONFIG_EVENT_PROFILE
-	atomic_t	profile_count;
-	int		(*profile_enable)(struct ftrace_event_call *);
-	void		(*profile_disable)(struct ftrace_event_call *);
-#endif
+	atomic_t		profile_count;
+	int			(*profile_enable)(struct ftrace_event_call *);
+	void			(*profile_disable)(struct ftrace_event_call *);
 };

Ah, I would have added ifdefs around the below bit.

kernel/trace/trace_events.c event_create_dir(struct ftrace_event_call *call, struct dentry *d_events, · line 943

Quoted 9 lines
kernel/trace/trace_events.c +0 -1
index 23d2972..e75276a 100644
@@ -940,7 +940,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
 
-	if (call->id)

#ifdef CONFIG_EVENT_PROFILE

kernel/trace/trace_events.c event_create_dir(struct ftrace_event_call *call, struct dentry *d_events, · line 945

Quoted 3 lines
kernel/trace/trace_events.c +1 -0
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);

#endif

Like that, but I guess this works too ;-)

#10 lore raw

Re: [tip:perfcounters/urgent] ftrace: Fix perf-tracepoint OOPS

Quoted 39 lines · Peter Zijlstra
Quoted 18 lines
[ v2: fix build error in the !CONFIG_EVENT_PROFILE case ]
Signed-off-by Ingo Molnar <[email protected]>
---
 include/linux/ftrace_event.h       |    8 +++-----
 kernel/trace/trace_event_profile.c |    2 +-
 kernel/trace/trace_events.c        |    2 +-
 3 files changed, 5 insertions(+), 7 deletions(-)
include/linux/ftrace_event.h +0 -0
index 5c093ff..d7cd193 100644
@@ -119,11 +119,9 @@ struct ftrace_event_call {
 	void			*filter;
 	void			*mod;

include/linux/ftrace_event.h struct ftrace_event_call { · line 125

Quoted 10 lines
include/linux/ftrace_event.h +3 -5
 
-#ifdef CONFIG_EVENT_PROFILE
-	atomic_t	profile_count;
-	int		(*profile_enable)(struct ftrace_event_call *);
-	void		(*profile_disable)(struct ftrace_event_call *);
-#endif
+	atomic_t		profile_count;
+	int			(*profile_enable)(struct ftrace_event_call *);
+	void			(*profile_disable)(struct ftrace_event_call *);
 };

Ah, I would have added ifdefs around the below bit.

Quoted 8 lines
kernel/trace/trace_events.c +0 -0
index 23d2972..e75276a 100644
@@ -940,7 +940,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
 		entry = trace_create_file("enable", 0644, call->dir, call,
 					  enable);
Quoted 8 lines

kernel/trace/trace_events.c event_create_dir(struct ftrace_event_call *call, struct dentry *d_events, · line 943

Quoted 1 line
kernel/trace/trace_events.c +0 -1
-	if (call->id)

#ifdef CONFIG_EVENT_PROFILE

kernel/trace/trace_events.c event_create_dir(struct ftrace_event_call *call, struct dentry *d_events, · line 945

Quoted 3 lines
kernel/trace/trace_events.c +1 -0
+	if (call->id && call->profile_enable)
 		entry = trace_create_file("id", 0444, call->dir, call,
 					  id);

#endif

Like that, but I guess this works too ;-)

i think CONFIG_EVENT_PROFILE should go away - it's clearly a core functionality of ftrace. Whenever we hit a Kconfig induced build breakage we should first consider reducing the Kconfig complexity a bit.

Ingo