Skip to contentSkip to conversation

trace-cmd: Fixes for reading older trace-cmd trace.dat files

PATCH New Steven Rostedt 1 participant 3 messages Oct 25, 2019

lore
Share & tools
lore
#0 lore raw

cover trace-cmd: Fixes for reading older trace-cmd trace.dat files

Fix two bugs that caused reading an trace.dat file created by an older trace-cmd not to print properly.

Steven Rostedt (VMware) (2):
libtraceevent: Remove extra '\n' in print_event_time()
trace-cmd: Use usecs unless specified not to

----

 lib/trace-cmd/trace-input.c  | 12 ++++++++----
 lib/traceevent/event-parse.c |  2 +-
 2 files changed, 9 insertions(+), 5 deletions(-)
#1 lore raw

PATCH 1/2 libtraceevent: Remove extra \n in print_event_time()

From: "Steven Rostedt (VMware)" <[email protected]>

If the precesion of print_event_time() is zero or greater than the timestamp, it uses a different format. But that format had an extra new line at the end, and caused the output to not look right:

cpus=2

           sleep-3946  [001]111264306005
: function:             inotify_inode_queue_event
           sleep-3946  [001]111264307158
: function:             __fsnotify_parent
           sleep-3946  [001]111264307637
: function:             inotify_dentry_parent_queue_event
           sleep-3946  [001]111264307989
: function:             fsnotify
           sleep-3946  [001]111264308401

: function: audit_syscall_exit

Fixes 150d479b623a ("libtraceevent, perf tools: Changes in tep_print_event_* APIs")
---
 lib/traceevent/event-parse.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/traceevent/event-parse.c +1 -1
index eb84fbb49e4d..0a49eb936b34 100644
@@ -5528,7 +5528,7 @@ static void print_event_time(struct tep_handle *tep, struct trace_seq *s,
 	if (p10 > 1 && p10 < time)
 		trace_seq_printf(s, "%5llu.%0*llu", time / p10, prec, time % p10);
 	else
-		trace_seq_printf(s, "%12llu\n", time);
+		trace_seq_printf(s, "%12llu", time);
 }
 
 struct print_event_type {
Signature · 1 line
2.23.0
#2 lore raw

PATCH 2/2 trace-cmd: Use usecs unless specified not to

In reply to Steven Rostedt

From: "Steven Rostedt (VMware)" <[email protected]>

Old trace-cmd data files did not have an option to store the clock that was used. It was always assumed that the clock would be in usecs. If there's no clock in the options of the trace.dat file, then use usecs by default to stay compatible with older versions of trace-cmd trace.dat files.

Fixes 150d479b623a ("libtraceevent, perf tools: Changes in tep_print_event_* APIs")
---
 lib/trace-cmd/trace-input.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
lib/trace-cmd/trace-input.c +8 -4
index 43bdf6ace33b..ef8f0762a535 100644
@@ -2311,6 +2311,9 @@ static int handle_options(struct tracecmd_input *handle)
 	char *buf;
 	int cpus;
 
+	/* By default, use usecs, unless told otherwise */
+	handle->flags |= TRACECMD_FL_IN_USECS;
+
 	for (;;) {
 		if (do_read_check(handle, &option, 2))
 			return -1;
@@ -2586,10 +2589,11 @@ static void extract_trace_clock(struct tracecmd_input *handle, char *line)
 	if (!clock)
 		return;
 
-	if (!strcmp(clock, "local") || !strcmp(clock, "global")
-	    || !strcmp(clock, "uptime") || !strcmp(clock, "perf")
-	    || !strncmp(clock, "mono", 4))
-		handle->flags |= TRACECMD_FL_IN_USECS;
+	/* Clear usecs if not one of the specified clocks */
+	if (strcmp(clock, "local") && strcmp(clock, "global") &&
+	    strcmp(clock, "uptime") && strcmp(clock, "perf") &&
+	    strncmp(clock, "mono", 4))
+		handle->flags &= ~TRACECMD_FL_IN_USECS;
 }
 
 void tracecmd_parse_trace_clock(struct tracecmd_input *handle,
Signature · 1 line
2.23.0