[PATCH 0/4] perf trace: Add filter Suppport, V2

[PATCH 0/4] perf trace: Add filter Suppport, V2

[Li Zefan]

This patchset adds filter support for perf counter, so not all
profile events are recorded but only those match the filters
we set.

An example:

 #./perf record -R -f -e irq:irq_handler_entry --filter irq==18
 ^C
 # ./perf trace
            perf-4303  ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0

The syntax of filter expressions can be found in
Documentation/trace/events.txt.

Major changes from v1:

- Make filters per perf event, instead of per tracepoint.

- Use "--filter" to set a filter for a trace event, instead of
  mixing it into "-e" option.

---
 include/linux/ftrace_event.h       |   11 +-
 include/linux/perf_counter.h       |    1 +
 include/linux/perf_event.h         |    7 +
 kernel/perf_event.c                |   81 +++++++++-
 kernel/trace/trace.h               |    6 +-
 kernel/trace/trace_events_filter.c |  320 +++++++++++++++++++++---------------
 tools/perf/builtin-record.c        |   15 ++-
 tools/perf/util/parse-events.c     |   26 +++-
 tools/perf/util/parse-events.h     |    2 +
 9 files changed, 322 insertions(+), 147 deletions(-)

[PATCH 1/4] tracing/filters: Refactor subsystem filter code

[Li Zefan]

Change:
	for_each_pred
		for_each_subsystem
To:
	for_each_subsystem
		for_each_pred

This change also prepares for later patches.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 kernel/trace/trace.h               |    1 -
 kernel/trace/trace_events_filter.c |  124 +++++++++++++-----------------------
 2 files changed, 45 insertions(+), 80 deletions(-)

diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index b17e0ea..300cdae 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -705,7 +705,6 @@ struct event_filter {
 	int			n_preds;
 	struct filter_pred	**preds;
 	char			*filter_string;
-	bool			no_reset;
 };
 
 struct event_subsystem {
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index 8c194de..1d22749 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -615,14 +615,7 @@ static int init_subsystem_preds(struct event_subsystem *system)
 	return 0;
 }
 
-enum {
-	FILTER_DISABLE_ALL,
-	FILTER_INIT_NO_RESET,
-	FILTER_SKIP_NO_RESET,
-};
-
-static void filter_free_subsystem_preds(struct event_subsystem *system,
-					int flag)
+static void filter_free_subsystem_preds(struct event_subsystem *system)
 {
 	struct ftrace_event_call *call;
 
@@ -633,14 +626,6 @@ static void filter_free_subsystem_preds(struct event_subsystem *system,
 		if (strcmp(call->system, system->name) != 0)
 			continue;
 
-		if (flag == FILTER_INIT_NO_RESET) {
-			call->filter->no_reset = false;
-			continue;
-		}
-
-		if (flag == FILTER_SKIP_NO_RESET && call->filter->no_reset)
-			continue;
-
 		filter_disable_preds(call);
 		remove_filter_string(call->filter);
 	}
@@ -817,44 +802,6 @@ add_pred_fn:
 	return 0;
 }
 
-static int filter_add_subsystem_pred(struct filter_parse_state *ps,
-				     struct event_subsystem *system,
-				     struct filter_pred *pred,
-				     char *filter_string,
-				     bool dry_run)
-{
-	struct ftrace_event_call *call;
-	int err = 0;
-	bool fail = true;
-
-	list_for_each_entry(call, &ftrace_events, list) {
-
-		if (!call->define_fields)
-			continue;
-
-		if (strcmp(call->system, system->name))
-			continue;
-
-		if (call->filter->no_reset)
-			continue;
-
-		err = filter_add_pred(ps, call, pred, dry_run);
-		if (err)
-			call->filter->no_reset = true;
-		else
-			fail = false;
-
-		if (!dry_run)
-			replace_filter_string(call->filter, filter_string);
-	}
-
-	if (fail) {
-		parse_error(ps, FILT_ERR_BAD_SUBSYS_FILTER, 0);
-		return err;
-	}
-	return 0;
-}
-
 static void parse_init(struct filter_parse_state *ps,
 		       struct filter_op *ops,
 		       char *infix_string)
@@ -1208,8 +1155,7 @@ static int check_preds(struct filter_parse_state *ps)
 	return 0;
 }
 
-static int replace_preds(struct event_subsystem *system,
-			 struct ftrace_event_call *call,
+static int replace_preds(struct ftrace_event_call *call,
 			 struct filter_parse_state *ps,
 			 char *filter_string,
 			 bool dry_run)
@@ -1256,11 +1202,7 @@ static int replace_preds(struct event_subsystem *system,
 add_pred:
 		if (!pred)
 			return -ENOMEM;
-		if (call)
-			err = filter_add_pred(ps, call, pred, false);
-		else
-			err = filter_add_subsystem_pred(ps, system, pred,
-						filter_string, dry_run);
+		err = filter_add_pred(ps, call, pred, dry_run);
 		filter_free_pred(pred);
 		if (err)
 			return err;
@@ -1271,6 +1213,44 @@ add_pred:
 	return 0;
 }
 
+static int replace_system_preds(struct event_subsystem *system,
+				struct filter_parse_state *ps,
+				char *filter_string)
+{
+	struct ftrace_event_call *call;
+	int err;
+	bool fail = true;
+
+	list_for_each_entry(call, &ftrace_events, list) {
+
+		if (!call->define_fields)
+			continue;
+
+		if (strcmp(call->system, system->name) != 0)
+			continue;
+
+		/* try to see if the filter can be applied */
+		err = replace_preds(call, ps, filter_string, true);
+		if (err)
+			continue;
+
+		/* really apply the filter */
+		filter_disable_preds(call);
+		err = replace_preds(call, ps, filter_string, false);
+		if (err)
+			filter_disable_preds(call);
+		else
+			replace_filter_string(call->filter, filter_string);
+		fail = false;
+	}
+
+	if (fail) {
+		parse_error(ps, FILT_ERR_BAD_SUBSYS_FILTER, 0);
+		return err;
+	}
+	return 0;
+}
+
 int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 {
 	int err;
@@ -1305,7 +1285,7 @@ int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 		goto out;
 	}
 
-	err = replace_preds(NULL, call, ps, filter_string, false);
+	err = replace_preds(call, ps, filter_string, false);
 	if (err)
 		append_filter_err(ps, call->filter);
 
@@ -1333,7 +1313,7 @@ int apply_subsystem_event_filter(struct event_subsystem *system,
 		goto out_unlock;
 
 	if (!strcmp(strstrip(filter_string), "0")) {
-		filter_free_subsystem_preds(system, FILTER_DISABLE_ALL);
+		filter_free_subsystem_preds(system);
 		remove_filter_string(system->filter);
 		mutex_unlock(&event_mutex);
 		return 0;
@@ -1353,23 +1333,9 @@ int apply_subsystem_event_filter(struct event_subsystem *system,
 		goto out;
 	}
 
-	filter_free_subsystem_preds(system, FILTER_INIT_NO_RESET);
-
-	/* try to see the filter can be applied to which events */
-	err = replace_preds(system, NULL, ps, filter_string, true);
-	if (err) {
-		append_filter_err(ps, system->filter);
-		goto out;
-	}
-
-	filter_free_subsystem_preds(system, FILTER_SKIP_NO_RESET);
-
-	/* really apply the filter to the events */
-	err = replace_preds(system, NULL, ps, filter_string, false);
-	if (err) {
+	err = replace_system_preds(system, ps, filter_string);
+	if (err)
 		append_filter_err(ps, system->filter);
-		filter_free_subsystem_preds(system, 2);
-	}
 
 out:
 	filter_opstack_clear(ps);
-- 
1.6.3

[PATCH 2/4] tracing/filters: Use a different op for glob match

[Li Zefan]

"==" will always do a full match, and "~" will do a glob match.

In the future, we may add "=~" for regex match.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 kernel/trace/trace.h               |    2 +-
 kernel/trace/trace_events_filter.c |   59 +++++++++++++++++-------------------
 2 files changed, 29 insertions(+), 32 deletions(-)

diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 300cdae..b5f30d5 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -724,7 +724,7 @@ typedef int (*filter_pred_fn_t) (struct filter_pred *pred, void *event,
 typedef int (*regex_match_func)(char *str, struct regex *r, int len);
 
 enum regex_type {
-	MATCH_FULL,
+	MATCH_FULL = 0,
 	MATCH_FRONT_ONLY,
 	MATCH_MIDDLE_ONLY,
 	MATCH_END_ONLY,
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index 1d22749..c7e1a54 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -29,6 +29,7 @@ enum filter_op_ids
 {
 	OP_OR,
 	OP_AND,
+	OP_GLOB,
 	OP_NE,
 	OP_EQ,
 	OP_LT,
@@ -46,16 +47,17 @@ struct filter_op {
 };
 
 static struct filter_op filter_ops[] = {
-	{ OP_OR, "||", 1 },
-	{ OP_AND, "&&", 2 },
-	{ OP_NE, "!=", 4 },
-	{ OP_EQ, "==", 4 },
-	{ OP_LT, "<", 5 },
-	{ OP_LE, "<=", 5 },
-	{ OP_GT, ">", 5 },
-	{ OP_GE, ">=", 5 },
-	{ OP_NONE, "OP_NONE", 0 },
-	{ OP_OPEN_PAREN, "(", 0 },
+	{ OP_OR,	"||",		1 },
+	{ OP_AND,	"&&",		2 },
+	{ OP_GLOB,	"~",		4 },
+	{ OP_NE,	"!=",		4 },
+	{ OP_EQ,	"==",		4 },
+	{ OP_LT,	"<",		5 },
+	{ OP_LE,	"<=",		5 },
+	{ OP_GT,	">",		5 },
+	{ OP_GE,	">=",		5 },
+	{ OP_NONE,	"OP_NONE",	0 },
+	{ OP_OPEN_PAREN, "(",		0 },
 };
 
 enum {
@@ -329,22 +331,18 @@ enum regex_type filter_parse_regex(char *buff, int len, char **search, int *not)
 	return type;
 }
 
-static int filter_build_regex(struct filter_pred *pred)
+static void filter_build_regex(struct filter_pred *pred)
 {
 	struct regex *r = &pred->regex;
-	char *search, *dup;
-	enum regex_type type;
-	int not;
-
-	type = filter_parse_regex(r->pattern, r->len, &search, &not);
-	dup = kstrdup(search, GFP_KERNEL);
-	if (!dup)
-		return -ENOMEM;
-
-	strcpy(r->pattern, dup);
-	kfree(dup);
-
-	r->len = strlen(r->pattern);
+	char *search;
+	enum regex_type type = MATCH_FULL;
+	int not = 0;
+
+	if (pred->op == OP_GLOB) {
+		type = filter_parse_regex(r->pattern, r->len, &search, &not);
+		r->len = strlen(search);
+		memmove(r->pattern, search, r->len+1);
+	}
 
 	switch (type) {
 	case MATCH_FULL:
@@ -362,8 +360,6 @@ static int filter_build_regex(struct filter_pred *pred)
 	}
 
 	pred->not ^= not;
-
-	return 0;
 }
 
 /* return 1 if event matches, 0 otherwise (discard) */
@@ -676,7 +672,10 @@ static bool is_string_field(struct ftrace_event_field *field)
 
 static int is_legal_op(struct ftrace_event_field *field, int op)
 {
-	if (is_string_field(field) && (op != OP_EQ && op != OP_NE))
+	if (is_string_field(field) &&
+	    (op != OP_EQ && op != OP_NE && op != OP_GLOB))
+		return 0;
+	if (!is_string_field(field) && op == OP_GLOB)
 		return 0;
 
 	return 1;
@@ -761,15 +760,13 @@ static int filter_add_pred(struct filter_parse_state *ps,
 	}
 
 	if (is_string_field(field)) {
-		ret = filter_build_regex(pred);
-		if (ret)
-			return ret;
+		filter_build_regex(pred);
 
 		if (field->filter_type == FILTER_STATIC_STRING) {
 			fn = filter_pred_string;
 			pred->regex.field_len = field->size;
 		} else if (field->filter_type == FILTER_DYN_STRING)
-				fn = filter_pred_strloc;
+			fn = filter_pred_strloc;
 		else {
 			fn = filter_pred_pchar;
 			pred->regex.field_len = strlen(pred->regex.pattern);
-- 
1.6.3

[PATCH 3/4] tracing/profile: Add filter support

[Li Zefan]

- Add an ioctl to allocate a filter for a perf event.

- Free the filter when the associated perf event is to be freed.

- Do the filtering in perf_swevent_match().

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 include/linux/ftrace_event.h       |   11 +++-
 include/linux/perf_counter.h       |    1 +
 include/linux/perf_event.h         |    7 ++
 kernel/perf_event.c                |   81 ++++++++++++++++++--
 kernel/trace/trace.h               |    3 +-
 kernel/trace/trace_events_filter.c |  149 +++++++++++++++++++++++++++++-------
 6 files changed, 214 insertions(+), 38 deletions(-)

diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index 4ec5e67..d117704 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -144,7 +144,7 @@ extern char			*trace_profile_buf_nmi;
 #define MAX_FILTER_STR_VAL	256	/* Should handle KSYM_SYMBOL_LEN */
 
 extern void destroy_preds(struct ftrace_event_call *call);
-extern int filter_match_preds(struct ftrace_event_call *call, void *rec);
+extern int filter_match_preds(struct event_filter *filter, void *rec);
 extern int filter_current_check_discard(struct ring_buffer *buffer,
 					struct ftrace_event_call *call,
 					void *rec,
@@ -186,4 +186,13 @@ do {									\
 		__trace_printk(ip, fmt, ##args);			\
 } while (0)
 
+#ifdef CONFIG_EVENT_PROFILE
+struct perf_event;
+extern int ftrace_profile_enable(int event_id);
+extern void ftrace_profile_disable(int event_id);
+extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
+				     char *filter_str);
+extern void ftrace_profile_free_filter(struct perf_event *event);
+#endif
+
 #endif /* _LINUX_FTRACE_EVENT_H */
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 7b7fbf4..91a2b43 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -225,6 +225,7 @@ struct perf_counter_attr {
 #define PERF_COUNTER_IOC_RESET		_IO ('$', 3)
 #define PERF_COUNTER_IOC_PERIOD		_IOW('$', 4, u64)
 #define PERF_COUNTER_IOC_SET_OUTPUT	_IO ('$', 5)
+#define PERF_COUNTER_IOC_SET_FILTER	_IOW('$', 6, char *)
 
 enum perf_counter_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 2e6d95f..b7ac0f2 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -221,6 +221,7 @@ struct perf_event_attr {
 #define PERF_EVENT_IOC_RESET		_IO ('$', 3)
 #define PERF_EVENT_IOC_PERIOD		_IOW('$', 4, u64)
 #define PERF_EVENT_IOC_SET_OUTPUT	_IO ('$', 5)
+#define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
 
 enum perf_event_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,
@@ -633,7 +634,13 @@ struct perf_event {
 
 	struct pid_namespace		*ns;
 	u64				id;
+
+#ifdef CONFIG_EVENT_PROFILE
+	struct event_filter		*filter;
+	int				filter_active;
 #endif
+
+#endif /* CONFIG_PERF_EVENTS */
 };
 
 /**
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 9d0b5c6..741d45b 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -28,6 +28,7 @@
 #include <linux/anon_inodes.h>
 #include <linux/kernel_stat.h>
 #include <linux/perf_event.h>
+#include <linux/ftrace_event.h>
 
 #include <asm/irq_regs.h>
 
@@ -1658,6 +1659,8 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu)
 	return ERR_PTR(err);
 }
 
+static void perf_event_free_filter(struct perf_event *event);
+
 static void free_event_rcu(struct rcu_head *head)
 {
 	struct perf_event *event;
@@ -1665,6 +1668,7 @@ static void free_event_rcu(struct rcu_head *head)
 	event = container_of(head, struct perf_event, rcu_head);
 	if (event->ns)
 		put_pid_ns(event->ns);
+	perf_event_free_filter(event);
 	kfree(event);
 }
 
@@ -1974,7 +1978,8 @@ unlock:
 	return ret;
 }
 
-int perf_event_set_output(struct perf_event *event, int output_fd);
+static int perf_event_set_output(struct perf_event *event, int output_fd);
+static int perf_event_set_filter(struct perf_event *event, void __user *arg);
 
 static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
@@ -2002,6 +2007,9 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 	case PERF_EVENT_IOC_SET_OUTPUT:
 		return perf_event_set_output(event, arg);
 
+	case PERF_EVENT_IOC_SET_FILTER:
+		return perf_event_set_filter(event, (void __user *)arg);
+
 	default:
 		return -ENOTTY;
 	}
@@ -3806,9 +3814,14 @@ static int perf_swevent_is_counting(struct perf_event *event)
 	return 1;
 }
 
+static int perf_tp_event_match(struct perf_event *event,
+				struct perf_sample_data *data);
+
 static int perf_swevent_match(struct perf_event *event,
 				enum perf_type_id type,
-				u32 event_id, struct pt_regs *regs)
+				u32 event_id,
+				struct perf_sample_data *data,
+				struct pt_regs *regs)
 {
 	if (!perf_swevent_is_counting(event))
 		return 0;
@@ -3826,6 +3839,10 @@ static int perf_swevent_match(struct perf_event *event,
 			return 0;
 	}
 
+	if (event->attr.type == PERF_TYPE_TRACEPOINT &&
+	    !perf_tp_event_match(event, data))
+		return 0;
+
 	return 1;
 }
 
@@ -3842,7 +3859,7 @@ static void perf_swevent_ctx_event(struct perf_event_context *ctx,
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
-		if (perf_swevent_match(event, type, event_id, regs))
+		if (perf_swevent_match(event, type, event_id, data, regs))
 			perf_swevent_add(event, nr, nmi, data, regs);
 	}
 	rcu_read_unlock();
@@ -4086,6 +4103,7 @@ static const struct pmu perf_ops_task_clock = {
 };
 
 #ifdef CONFIG_EVENT_PROFILE
+
 void perf_tp_event(int event_id, u64 addr, u64 count, void *record,
 			  int entry_size)
 {
@@ -4109,8 +4127,16 @@ void perf_tp_event(int event_id, u64 addr, u64 count, void *record,
 }
 EXPORT_SYMBOL_GPL(perf_tp_event);
 
-extern int ftrace_profile_enable(int);
-extern void ftrace_profile_disable(int);
+static int perf_tp_event_match(struct perf_event *event,
+				struct perf_sample_data *data)
+{
+	void *record = data->raw->data;
+
+	if (likely(!event->filter_active) ||
+	    filter_match_preds(event->filter, record))
+		return 1;
+	return 0;
+}
 
 static void tp_perf_event_destroy(struct perf_event *event)
 {
@@ -4135,12 +4161,53 @@ static const struct pmu *tp_perf_event_init(struct perf_event *event)
 
 	return &perf_ops_generic;
 }
+
+static int perf_event_set_filter(struct perf_event *event, void __user *arg)
+{
+	char *filter_str;
+	int ret;
+
+	if (event->attr.type != PERF_TYPE_TRACEPOINT)
+		return -EINVAL;
+
+	filter_str = strndup_user(arg, PAGE_SIZE);
+	if (IS_ERR(filter_str))
+		return PTR_ERR(filter_str);
+
+	ret = ftrace_profile_set_filter(event, event->attr.config, filter_str);
+
+	kfree(filter_str);
+	return ret;
+}
+
+static void perf_event_free_filter(struct perf_event *event)
+{
+	ftrace_profile_free_filter(event);
+}
+
 #else
+
+static int perf_tp_event_match(struct perf_event *event,
+				struct perf_sample_data *data)
+{
+	return 1;
+}
+
 static const struct pmu *tp_perf_event_init(struct perf_event *event)
 {
 	return NULL;
 }
-#endif
+
+static int perf_event_set_filter(struct perf_event *event, void __user *arg)
+{
+	return -ENOENT;
+}
+
+static void perf_event_free_filter(struct perf_event *event)
+{
+}
+
+#endif /* CONFIG_EVENT_PROFILE */
 
 atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
 
@@ -4394,7 +4461,7 @@ err_size:
 	goto out;
 }
 
-int perf_event_set_output(struct perf_event *event, int output_fd)
+static int perf_event_set_output(struct perf_event *event, int output_fd)
 {
 	struct perf_event *output_event = NULL;
 	struct file *output_file = NULL;
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index b5f30d5..671bedb 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -765,7 +765,8 @@ filter_check_discard(struct ftrace_event_call *call, void *rec,
 		     struct ring_buffer *buffer,
 		     struct ring_buffer_event *event)
 {
-	if (unlikely(call->filter_active) && !filter_match_preds(call, rec)) {
+	if (unlikely(call->filter_active) &&
+	    !filter_match_preds(call->filter, rec)) {
 		ring_buffer_discard_commit(buffer, event);
 		return 1;
 	}
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index c7e1a54..a799113 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -21,6 +21,7 @@
 #include <linux/module.h>
 #include <linux/ctype.h>
 #include <linux/mutex.h>
+#include <linux/perf_event.h>
 
 #include "trace.h"
 #include "trace_output.h"
@@ -363,9 +364,8 @@ static void filter_build_regex(struct filter_pred *pred)
 }
 
 /* return 1 if event matches, 0 otherwise (discard) */
-int filter_match_preds(struct ftrace_event_call *call, void *rec)
+int filter_match_preds(struct event_filter *filter, void *rec)
 {
-	struct event_filter *filter = call->filter;
 	int match, top = 0, val1 = 0, val2 = 0;
 	int stack[MAX_FILTER_PRED];
 	struct filter_pred *pred;
@@ -538,14 +538,10 @@ static void filter_disable_preds(struct ftrace_event_call *call)
 		filter->preds[i]->fn = filter_pred_none;
 }
 
-void destroy_preds(struct ftrace_event_call *call)
+static void __free_preds(struct event_filter *filter)
 {
-	struct event_filter *filter = call->filter;
 	int i;
 
-	if (!filter)
-		return;
-
 	for (i = 0; i < MAX_FILTER_PRED; i++) {
 		if (filter->preds[i])
 			filter_free_pred(filter->preds[i]);
@@ -553,21 +549,27 @@ void destroy_preds(struct ftrace_event_call *call)
 	kfree(filter->preds);
 	kfree(filter->filter_string);
 	kfree(filter);
+}
+
+void destroy_preds(struct ftrace_event_call *call)
+{
+	if (!call->filter)
+		return;
+
+	__free_preds(call->filter);
 	call->filter = NULL;
+	call->filter_active = 0;
 }
 
-static int init_preds(struct ftrace_event_call *call)
+static struct event_filter *__alloc_preds(void)
 {
 	struct event_filter *filter;
 	struct filter_pred *pred;
 	int i;
 
-	if (call->filter)
-		return 0;
-
-	filter = call->filter = kzalloc(sizeof(*filter), GFP_KERNEL);
-	if (!call->filter)
-		return -ENOMEM;
+	filter = kzalloc(sizeof(*filter), GFP_KERNEL);
+	if (!filter)
+		return ERR_PTR(-ENOMEM);
 
 	filter->n_preds = 0;
 
@@ -583,12 +585,24 @@ static int init_preds(struct ftrace_event_call *call)
 		filter->preds[i] = pred;
 	}
 
-	return 0;
+	return filter;
 
 oom:
-	destroy_preds(call);
+	__free_preds(filter);
+	return ERR_PTR(-ENOMEM);
+}
+
+static int init_preds(struct ftrace_event_call *call)
+{
+	if (call->filter)
+		return 0;
 
-	return -ENOMEM;
+	call->filter_active = 0;
+	call->filter = __alloc_preds();
+	if (IS_ERR(call->filter))
+		return PTR_ERR(call->filter);
+
+	return 0;
 }
 
 static int init_subsystem_preds(struct event_subsystem *system)
@@ -629,10 +643,10 @@ static void filter_free_subsystem_preds(struct event_subsystem *system)
 
 static int filter_add_pred_fn(struct filter_parse_state *ps,
 			      struct ftrace_event_call *call,
+			      struct event_filter *filter,
 			      struct filter_pred *pred,
 			      filter_pred_fn_t fn)
 {
-	struct event_filter *filter = call->filter;
 	int idx, err;
 
 	if (filter->n_preds == MAX_FILTER_PRED) {
@@ -647,7 +661,6 @@ static int filter_add_pred_fn(struct filter_parse_state *ps,
 		return err;
 
 	filter->n_preds++;
-	call->filter_active = 1;
 
 	return 0;
 }
@@ -726,6 +739,7 @@ static filter_pred_fn_t select_comparison_fn(int op, int field_size,
 
 static int filter_add_pred(struct filter_parse_state *ps,
 			   struct ftrace_event_call *call,
+			   struct event_filter *filter,
 			   struct filter_pred *pred,
 			   bool dry_run)
 {
@@ -795,7 +809,7 @@ static int filter_add_pred(struct filter_parse_state *ps,
 
 add_pred_fn:
 	if (!dry_run)
-		return filter_add_pred_fn(ps, call, pred, fn);
+		return filter_add_pred_fn(ps, call, filter, pred, fn);
 	return 0;
 }
 
@@ -1153,6 +1167,7 @@ static int check_preds(struct filter_parse_state *ps)
 }
 
 static int replace_preds(struct ftrace_event_call *call,
+			 struct event_filter *filter,
 			 struct filter_parse_state *ps,
 			 char *filter_string,
 			 bool dry_run)
@@ -1199,7 +1214,7 @@ static int replace_preds(struct ftrace_event_call *call,
 add_pred:
 		if (!pred)
 			return -ENOMEM;
-		err = filter_add_pred(ps, call, pred, dry_run);
+		err = filter_add_pred(ps, call, filter, pred, dry_run);
 		filter_free_pred(pred);
 		if (err)
 			return err;
@@ -1215,6 +1230,7 @@ static int replace_system_preds(struct event_subsystem *system,
 				char *filter_string)
 {
 	struct ftrace_event_call *call;
+	struct event_filter *filter;
 	int err;
 	bool fail = true;
 
@@ -1227,17 +1243,19 @@ static int replace_system_preds(struct event_subsystem *system,
 			continue;
 
 		/* try to see if the filter can be applied */
-		err = replace_preds(call, ps, filter_string, true);
+		err = replace_preds(call, filter, ps, filter_string, true);
 		if (err)
 			continue;
 
 		/* really apply the filter */
 		filter_disable_preds(call);
-		err = replace_preds(call, ps, filter_string, false);
+		err = replace_preds(call, filter, ps, filter_string, false);
 		if (err)
 			filter_disable_preds(call);
-		else
-			replace_filter_string(call->filter, filter_string);
+		else {
+			call->filter_active = 1;
+			replace_filter_string(filter, filter_string);
+		}
 		fail = false;
 	}
 
@@ -1251,7 +1269,6 @@ static int replace_system_preds(struct event_subsystem *system,
 int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 {
 	int err;
-
 	struct filter_parse_state *ps;
 
 	mutex_lock(&event_mutex);
@@ -1282,10 +1299,11 @@ int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 		goto out;
 	}
 
-	err = replace_preds(call, ps, filter_string, false);
+	err = replace_preds(call, call->filter, ps, filter_string, false);
 	if (err)
 		append_filter_err(ps, call->filter);
-
+	else
+		call->filter_active = 1;
 out:
 	filter_opstack_clear(ps);
 	postfix_clear(ps);
@@ -1300,7 +1318,6 @@ int apply_subsystem_event_filter(struct event_subsystem *system,
 				 char *filter_string)
 {
 	int err;
-
 	struct filter_parse_state *ps;
 
 	mutex_lock(&event_mutex);
@@ -1344,3 +1361,77 @@ out_unlock:
 	return err;
 }
 
+#ifdef CONFIG_EVENT_PROFILE
+
+void ftrace_profile_free_filter(struct perf_event *event)
+{
+	if (!event->filter)
+		return;
+
+	__free_preds(event->filter);
+	event->filter = NULL;
+	event->filter_active = 0;
+}
+
+static int alloc_profile_preds(struct perf_event *event)
+{
+	if (event->filter)
+		return -EEXIST;
+
+	event->filter = __alloc_preds();
+	if (IS_ERR(event->filter))
+		return PTR_ERR(event->filter);
+
+	return 0;
+}
+
+int ftrace_profile_set_filter(struct perf_event *event, int event_id,
+			      char *filter_str)
+{
+	int err;
+	struct filter_parse_state *ps;
+	struct ftrace_event_call *call = NULL;
+
+	mutex_lock(&event_mutex);
+
+	err = -EINVAL;
+	list_for_each_entry(call, &ftrace_events, list) {
+		if (call->id == event_id)
+			break;
+	}
+	if (!call)
+		goto out;
+
+	err = alloc_profile_preds(event);
+	if (err)
+		goto out;
+
+	err = -ENOMEM;
+	ps = kzalloc(sizeof(*ps), GFP_KERNEL);
+	if (!ps)
+		goto free_preds;
+
+	parse_init(ps, filter_ops, filter_str);
+	err = filter_parse(ps);
+	if (err)
+		goto free_ps;
+
+	err = replace_preds(call, event->filter, ps, filter_str, false);
+	if (!err)
+		event->filter_active = 1;
+
+free_ps:
+	filter_opstack_clear(ps);
+	postfix_clear(ps);
+	kfree(ps);
+free_preds:
+	if (err)
+		ftrace_profile_free_filter(event);
+
+	mutex_unlock(&event_mutex);
+out:
+	return err;
+}
+
+#endif /* CONFIG_EVENT_PROFILE */
+
-- 
1.6.3

[PATCH 4/4] perf trace: Add filter Suppport

[Li Zefan]

Add a new option "--filter <filter_str>" to perf record, and
it should be right after "-e trace_point":

 #./perf record -R -f -e irq:irq_handler_entry --filter irq==18
 ^C
 # ./perf trace
            perf-4303  ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0
            init-0     ... irq_handler_entry: irq=18 handler=eth0

See Documentation/trace/events.txt for the syntax of filter
expressions.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 tools/perf/builtin-record.c    |   15 ++++++++++++++-
 tools/perf/util/parse-events.c |   26 ++++++++++++++++++++++++--
 tools/perf/util/parse-events.h |    2 ++
 3 files changed, 40 insertions(+), 3 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 59af03d..2f83c8f 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -374,9 +374,11 @@ static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int n
 
 static void create_counter(int counter, int cpu, pid_t pid)
 {
+	char *filter = filters[counter];
 	struct perf_event_attr *attr = attrs + counter;
 	struct perf_header_attr *h_attr;
 	int track = !counter; /* only the first counter needs these */
+	int ret;
 	struct {
 		u64 count;
 		u64 time_enabled;
@@ -479,7 +481,6 @@ try_again:
 		multiplex_fd = fd[nr_cpu][counter];
 
 	if (multiplex && fd[nr_cpu][counter] != multiplex_fd) {
-		int ret;
 
 		ret = ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_SET_OUTPUT, multiplex_fd);
 		assert(ret != -1);
@@ -499,6 +500,16 @@ try_again:
 		}
 	}
 
+	if (filter != NULL) {
+		ret = ioctl(fd[nr_cpu][counter],
+			    PERF_EVENT_IOC_SET_FILTER, filter);
+		if (ret) {
+			error("failed to set filter with %d (%s)\n", errno,
+			      strerror(errno));
+			exit(-1);
+		}
+	}
+
 	ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_ENABLE);
 }
 
@@ -676,6 +687,8 @@ static const struct option options[] = {
 	OPT_CALLBACK('e', "event", NULL, "event",
 		     "event selector. use 'perf list' to list available events",
 		     parse_events),
+	OPT_CALLBACK(0, "filter", NULL, "filter",
+		     "event filter", parse_filter),
 	OPT_INTEGER('p', "pid", &target_pid,
 		    "record events on existing pid"),
 	OPT_INTEGER('r', "realtime", &realtime_prio,
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 87c424d..5573290 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -8,9 +8,10 @@
 #include "cache.h"
 #include "header.h"
 
-int					nr_counters;
+int				nr_counters;
 
 struct perf_event_attr		attrs[MAX_COUNTERS];
+char				*filters[MAX_COUNTERS];
 
 struct event_symbol {
 	u8		type;
@@ -705,7 +706,6 @@ static void store_event_type(const char *orgname)
 	perf_header__push_event(id, orgname);
 }
 
-
 int parse_events(const struct option *opt __used, const char *str, int unset __used)
 {
 	struct perf_event_attr attr;
@@ -742,6 +742,28 @@ int parse_events(const struct option *opt __used, const char *str, int unset __u
 	return 0;
 }
 
+int parse_filter(const struct option *opt __used, const char *str,
+		 int unset __used)
+{
+	int i = nr_counters - 1;
+	int len = strlen(str);
+
+	if (i < 0 || attrs[i].type != PERF_TYPE_TRACEPOINT) {
+		fprintf(stderr,
+			"-F option should follow a -e tracepoint option\n");
+		return -1;
+	}
+
+	filters[i] = malloc(len + 1);
+	if (!filters[i]) {
+		fprintf(stderr, "not enough memory to hold filter string\n");
+		return -1;
+	}
+	strcpy(filters[i], str);
+
+	return 0;
+}
+
 static const char * const event_type_descriptors[] = {
 	"",
 	"Hardware event",
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index 8626a43..b8c1f64 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -17,11 +17,13 @@ extern struct tracepoint_path *tracepoint_id_to_path(u64 config);
 extern int			nr_counters;
 
 extern struct perf_event_attr attrs[MAX_COUNTERS];
+extern char *filters[MAX_COUNTERS];
 
 extern const char *event_name(int ctr);
 extern const char *__event_name(int type, u64 config);
 
 extern int parse_events(const struct option *opt, const char *str, int unset);
+extern int parse_filter(const struct option *opt, const char *str, int unset);
 
 #define EVENTS_HELP_MAX (128*1024)
 
-- 
1.6.3

Re: [PATCH 3/4] tracing/profile: Add filter support

[Peter Zijlstra]

On Tue, 2009-10-13 at 10:19 +0800, Li Zefan wrote:
> +#ifdef CONFIG_EVENT_PROFILE
> +       struct event_filter             *filter;
> +       int                             filter_active;
>  #endif

Why do you need filter_active, isn't !filter equivalent?

Re: [PATCH 3/4] tracing/profile: Add filter support

[Li Zefan]

Peter Zijlstra wrote:
> On Tue, 2009-10-13 at 10:19 +0800, Li Zefan wrote:
>> +#ifdef CONFIG_EVENT_PROFILE
>> +       struct event_filter             *filter;
>> +       int                             filter_active;
>>  #endif
> 
> Why do you need filter_active, isn't !filter equivalent?
> 

Indeed. I realized this right after I sent out this patchset. I'll
fix it. Thanks.

Re: [PATCH 3/4] tracing/profile: Add filter support

[Ingo Molnar]

* Li Zefan <lizf@cn.fujitsu.com> wrote:

> +#ifdef CONFIG_EVENT_PROFILE
> +struct perf_event;
> +extern int ftrace_profile_enable(int event_id);
> +extern void ftrace_profile_disable(int event_id);
> +extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
> +				     char *filter_str);
> +extern void ftrace_profile_free_filter(struct perf_event *event);
> +#endif

This reminds me - i think we should eliminate CONFIG_EVENT_PROFILE - 
it's an unnecessary Kconfig complication. If both PERF_EVENTS and 
EVENT_TRACING is enabled we should expose generic tracepoints.

Nor is it limited to event 'profiling', so it has become a misnomer as 
well.

	Ingo

[PATCH 1/4] tracing/filters: Refactor subsystem filter code

[Li Zefan]

Change:
	for_each_pred
		for_each_subsystem
To:
	for_each_subsystem
		for_each_pred

This change also prepares for later patches.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 kernel/trace/trace.h               |    1 -
 kernel/trace/trace_events_filter.c |  124 +++++++++++++-----------------------
 2 files changed, 45 insertions(+), 80 deletions(-)

diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index b17e0ea..300cdae 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -705,7 +705,6 @@ struct event_filter {
 	int			n_preds;
 	struct filter_pred	**preds;
 	char			*filter_string;
-	bool			no_reset;
 };
 
 struct event_subsystem {
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index 8c194de..1d22749 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -615,14 +615,7 @@ static int init_subsystem_preds(struct event_subsystem *system)
 	return 0;
 }
 
-enum {
-	FILTER_DISABLE_ALL,
-	FILTER_INIT_NO_RESET,
-	FILTER_SKIP_NO_RESET,
-};
-
-static void filter_free_subsystem_preds(struct event_subsystem *system,
-					int flag)
+static void filter_free_subsystem_preds(struct event_subsystem *system)
 {
 	struct ftrace_event_call *call;
 
@@ -633,14 +626,6 @@ static void filter_free_subsystem_preds(struct event_subsystem *system,
 		if (strcmp(call->system, system->name) != 0)
 			continue;
 
-		if (flag == FILTER_INIT_NO_RESET) {
-			call->filter->no_reset = false;
-			continue;
-		}
-
-		if (flag == FILTER_SKIP_NO_RESET && call->filter->no_reset)
-			continue;
-
 		filter_disable_preds(call);
 		remove_filter_string(call->filter);
 	}
@@ -817,44 +802,6 @@ add_pred_fn:
 	return 0;
 }
 
-static int filter_add_subsystem_pred(struct filter_parse_state *ps,
-				     struct event_subsystem *system,
-				     struct filter_pred *pred,
-				     char *filter_string,
-				     bool dry_run)
-{
-	struct ftrace_event_call *call;
-	int err = 0;
-	bool fail = true;
-
-	list_for_each_entry(call, &ftrace_events, list) {
-
-		if (!call->define_fields)
-			continue;
-
-		if (strcmp(call->system, system->name))
-			continue;
-
-		if (call->filter->no_reset)
-			continue;
-
-		err = filter_add_pred(ps, call, pred, dry_run);
-		if (err)
-			call->filter->no_reset = true;
-		else
-			fail = false;
-
-		if (!dry_run)
-			replace_filter_string(call->filter, filter_string);
-	}
-
-	if (fail) {
-		parse_error(ps, FILT_ERR_BAD_SUBSYS_FILTER, 0);
-		return err;
-	}
-	return 0;
-}
-
 static void parse_init(struct filter_parse_state *ps,
 		       struct filter_op *ops,
 		       char *infix_string)
@@ -1208,8 +1155,7 @@ static int check_preds(struct filter_parse_state *ps)
 	return 0;
 }
 
-static int replace_preds(struct event_subsystem *system,
-			 struct ftrace_event_call *call,
+static int replace_preds(struct ftrace_event_call *call,
 			 struct filter_parse_state *ps,
 			 char *filter_string,
 			 bool dry_run)
@@ -1256,11 +1202,7 @@ static int replace_preds(struct event_subsystem *system,
 add_pred:
 		if (!pred)
 			return -ENOMEM;
-		if (call)
-			err = filter_add_pred(ps, call, pred, false);
-		else
-			err = filter_add_subsystem_pred(ps, system, pred,
-						filter_string, dry_run);
+		err = filter_add_pred(ps, call, pred, dry_run);
 		filter_free_pred(pred);
 		if (err)
 			return err;
@@ -1271,6 +1213,44 @@ add_pred:
 	return 0;
 }
 
+static int replace_system_preds(struct event_subsystem *system,
+				struct filter_parse_state *ps,
+				char *filter_string)
+{
+	struct ftrace_event_call *call;
+	int err;
+	bool fail = true;
+
+	list_for_each_entry(call, &ftrace_events, list) {
+
+		if (!call->define_fields)
+			continue;
+
+		if (strcmp(call->system, system->name) != 0)
+			continue;
+
+		/* try to see if the filter can be applied */
+		err = replace_preds(call, ps, filter_string, true);
+		if (err)
+			continue;
+
+		/* really apply the filter */
+		filter_disable_preds(call);
+		err = replace_preds(call, ps, filter_string, false);
+		if (err)
+			filter_disable_preds(call);
+		else
+			replace_filter_string(call->filter, filter_string);
+		fail = false;
+	}
+
+	if (fail) {
+		parse_error(ps, FILT_ERR_BAD_SUBSYS_FILTER, 0);
+		return err;
+	}
+	return 0;
+}
+
 int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 {
 	int err;
@@ -1305,7 +1285,7 @@ int apply_event_filter(struct ftrace_event_call *call, char *filter_string)
 		goto out;
 	}
 
-	err = replace_preds(NULL, call, ps, filter_string, false);
+	err = replace_preds(call, ps, filter_string, false);
 	if (err)
 		append_filter_err(ps, call->filter);
 
@@ -1333,7 +1313,7 @@ int apply_subsystem_event_filter(struct event_subsystem *system,
 		goto out_unlock;
 
 	if (!strcmp(strstrip(filter_string), "0")) {
-		filter_free_subsystem_preds(system, FILTER_DISABLE_ALL);
+		filter_free_subsystem_preds(system);
 		remove_filter_string(system->filter);
 		mutex_unlock(&event_mutex);
 		return 0;
@@ -1353,23 +1333,9 @@ int apply_subsystem_event_filter(struct event_subsystem *system,
 		goto out;
 	}
 
-	filter_free_subsystem_preds(system, FILTER_INIT_NO_RESET);
-
-	/* try to see the filter can be applied to which events */
-	err = replace_preds(system, NULL, ps, filter_string, true);
-	if (err) {
-		append_filter_err(ps, system->filter);
-		goto out;
-	}
-
-	filter_free_subsystem_preds(system, FILTER_SKIP_NO_RESET);
-
-	/* really apply the filter to the events */
-	err = replace_preds(system, NULL, ps, filter_string, false);
-	if (err) {
+	err = replace_system_preds(system, ps, filter_string);
+	if (err)
 		append_filter_err(ps, system->filter);
-		filter_free_subsystem_preds(system, 2);
-	}
 
 out:
 	filter_opstack_clear(ps);
--
1.6.3