libtracefs: Add stacktrace to tracefs_sql()

Commit Message

Steven Rostedt Jan. 17, 2023, 4:35 p.m. UTC

From: "Steven Rostedt (Google)" <[e-mail]>

Now that the kernel can pass stacktraces from the start to end of a
synthetic event, update the tracefs_sql() (and by doing so sqlhist) to
handle stacktrace.

A new keyword STACKTRACE is added, so the following will now work:

  sqlhist -e -n block_lat 'SELECT (end.TIMESTAMP_USECS - start.TIMESTAMP_USECS) AS delta,start.STACKTRACE AS stack
     FROM sched_switch AS start JOIN sched_switch AS end ON end.next_pid = start.prev_pid
     WHERE start.prev_state == 1 && start.prev_prio < 100'

Signed-off-by: Steven Rostedt (Google) <[e-mail]>
---
 Documentation/libtracefs-sql.txt | 13 +++++++++++++
 include/tracefs.h                |  2 ++
 src/tracefs-filter.c             | 10 ++++++++++
 src/tracefs-hist.c               |  3 ++-
 src/tracefs-sqlhist.c            |  5 ++++-
 5 files changed, 31 insertions(+), 2 deletions(-)

Patch

diff --git a/Documentation/libtracefs-sql.txt b/Documentation/libtracefs-sql.txt
index 6d606db..62fd771 100644
--- a/Documentation/libtracefs-sql.txt
+++ b/Documentation/libtracefs-sql.txt
@@ -162,6 +162,19 @@  select start.pid, (end.TIMESTAMP_USECS - start.TIMESTAMP_USECS) as lat from sche
    WHERE start.prio < 100 || end.prev_prio < 100
 --
 
+If the kernel supports it, you can pass around a stacktrace between events.
+
+[source, c]
+--
+select start.prev_pid as pid, (end.TIMESTAMP_USECS - start.TIMESTAMP_USECS) as delta, start.STACKTRACE as stack
+   FROM sched_switch as start JOIN sched_switch as end ON start.prev_pid = end.next_pid
+   WHERE start.prev_state == 2
+--
+
+The above will record a stacktrace when a task is in the UNINTERRUPTIBLE (blocked) state, and trigger
+the synthetic event when it is scheduled back in, recording the time delta that it was blocked for.
+It will record the stacktrace of where it was when it scheduled out along with the delta.
+
 
 KEYWORDS AS EVENT FIELDS
 ------------------------
diff --git a/include/tracefs.h b/include/tracefs.h
index 3547b5a..23b3f8c 100644
--- a/include/tracefs.h
+++ b/include/tracefs.h
@@ -553,6 +553,8 @@  int tracefs_event_verify_filter(struct tep_event *event, const char *filter,
 #define TRACEFS_TIMESTAMP "common_timestamp"
 #define TRACEFS_TIMESTAMP_USECS "common_timestamp.usecs"
 
+#define TRACEFS_STACKTRACE "stacktrace"
+
 enum tracefs_synth_handler {
 	TRACEFS_SYNTH_HANDLE_NONE	= 0,
 	TRACEFS_SYNTH_HANDLE_MATCH,
diff --git a/src/tracefs-filter.c b/src/tracefs-filter.c
index a3dd77b..068329e 100644
--- a/src/tracefs-filter.c
+++ b/src/tracefs-filter.c
@@ -41,6 +41,13 @@  static const struct tep_format_field common_comm = {
 	.size			= 16,
 };
 
+static const struct tep_format_field stacktrace = {
+	.type			= "unsigned long[]",
+	.name			= "stacktrace",
+	.size			= 4,
+	.flags			= TEP_FIELD_IS_ARRAY | TEP_FIELD_IS_DYNAMIC,
+};
+
 /*
  * This also must be able to accept fields that are OK via the histograms,
  * such as common_timestamp.
@@ -56,6 +63,9 @@  static const struct tep_format_field *get_event_field(struct tep_event *event,
 	if (!strcmp(field_name, TRACEFS_TIMESTAMP_USECS))
 		return &common_timestamp_usecs;
 
+	if (!strcmp(field_name, TRACEFS_STACKTRACE))
+		return &stacktrace;
+
 	field = tep_find_any_field(event, field_name);
 	if (!field && (!strcmp(field_name, "COMM") || !strcmp(field_name, "comm")))
 		return &common_comm;
diff --git a/src/tracefs-hist.c b/src/tracefs-hist.c
index fb6231e..1918670 100644
--- a/src/tracefs-hist.c
+++ b/src/tracefs-hist.c
@@ -915,7 +915,8 @@  static char *add_synth_field(const struct tep_format_field *field,
 	bool sign;
 
 	if (field->flags & TEP_FIELD_IS_ARRAY) {
-		str = strdup("char");
+		str = strdup(field->type);
+		str = strtok(str, "[");
 		str = append_string(str, " ", name);
 		str = append_string(str, NULL, "[");
 
diff --git a/src/tracefs-sqlhist.c b/src/tracefs-sqlhist.c
index 3f571b7..0035922 100644
--- a/src/tracefs-sqlhist.c
+++ b/src/tracefs-sqlhist.c
@@ -566,7 +566,8 @@  static int test_field_exists(struct tep_handle *tep,
 		return -1;
 
 	if (!strcmp(field_name, TRACEFS_TIMESTAMP) ||
-	    !strcmp(field->field, TRACEFS_TIMESTAMP_USECS))
+	    !strcmp(field->field, TRACEFS_TIMESTAMP_USECS) ||
+	    !strcmp(field->field, TRACEFS_STACKTRACE))
 		tfield = (void *)1L;
 	else
 		tfield = tep_find_any_field(field->event, field_name);
@@ -695,6 +696,8 @@  static int update_vars(struct tep_handle *tep,
 			field->field = store_str(sb, TRACEFS_TIMESTAMP);
 		if (!strcmp(field->field, "TIMESTAMP_USECS"))
 			field->field = store_str(sb, TRACEFS_TIMESTAMP_USECS);
+		if (!strcmp(field->field, "STACKTRACE"))
+			field->field = store_str(sb, TRACEFS_STACKTRACE);
 		if (test_field_exists(tep, sb, expr))
 			return -1;
 	}