Skip to contentSkip to conversation

perf trace: get rid of the hard-coded paths in the report scripts

PATCH In review Tom Zanussi 3 participants 7 messages Aug 12, 2010 – Aug 14, 2010 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf trace: get rid of the hard-coded paths in the report scripts

The perf trace report shell scripts hard-code the exec path of the scripts into their command-lines, which doesn't work if perf has been installed somewhere else.

Instead, perf trace should create the paths at run-time. This patch does that and removes the hard-coded paths from all the report scripts.

v2 changes: The first version inadvertantly caused scripts run from outside the perf exec path to fail e.g. 'perf trace -s test.py'. The fix is to try the script name without the exec path first, then the version using the exec path, which restores the expected behavior.

---
 tools/perf/builtin-trace.c                         |   22 ++++++++++++++++---
 tools/perf/scripts/perl/bin/failed-syscalls-report |    2 +-
 tools/perf/scripts/perl/bin/rw-by-file-report      |    2 +-
 tools/perf/scripts/perl/bin/rw-by-pid-report       |    2 +-
 tools/perf/scripts/perl/bin/rwtop-report           |    2 +-
 tools/perf/scripts/perl/bin/wakeup-latency-report  |    2 +-
 tools/perf/scripts/perl/bin/workqueue-stats-report |    2 +-
 .../python/bin/failed-syscalls-by-pid-report       |    2 +-
 .../perf/scripts/python/bin/sched-migration-report |    2 +-
 tools/perf/scripts/python/bin/sctop-report         |    2 +-
 .../python/bin/syscall-counts-by-pid-report        |    2 +-
 .../perf/scripts/python/bin/syscall-counts-report  |    3 +-
 12 files changed, 30 insertions(+), 15 deletions(-)
tools/perf/builtin-trace.c +18 -4
index 40a6a29..88a1883 100644
@@ -573,6 +573,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
 	const char *suffix = NULL;
 	const char **__argv;
 	char *script_path;
+	struct stat perf_stat;
 	int i, err;
 
 	if (argc >= 2 && strncmp(argv[1], "rec", strlen("rec")) == 0) {
@@ -689,8 +690,6 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
 		return -EINVAL;
 
 	if (generate_script_lang) {
-		struct stat perf_stat;
-
 		int input = open(input_name, O_RDONLY);
 		if (input < 0) {
 			perror("failed to open file");
@@ -719,10 +718,25 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
 	}
 
 	if (script_name) {
-		err = scripting_ops->start_script(script_name, argc, argv);
+		char script_exec_path[MAXPATHLEN];
+
+		snprintf(script_exec_path, MAXPATHLEN, "%s", script_name);
+		err = stat(script_exec_path, &perf_stat);
+		if (err < 0) {
+			snprintf(script_exec_path, MAXPATHLEN, "%s/scripts/%s",
+				 perf_exec_path(), script_name);
+			err = stat(script_exec_path, &perf_stat);
+			if (err < 0)
+				goto out;
+		}
+
+		err = scripting_ops->start_script(script_exec_path,
+						  argc, argv);
 		if (err)
 			goto out;
-		pr_debug("perf trace started with script %s\n\n", script_name);
+
+		pr_debug("perf trace started with script %s\n\n",
+			 script_exec_path);
 	}
 
 	err = __cmd_trace(session);
tools/perf/scripts/perl/bin/failed-syscalls-report +1 -1
index e3a5e55..47cf0b7 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/failed-syscalls.pl $comm
+perf trace $@ -s perl/failed-syscalls.pl $comm
tools/perf/scripts/perl/bin/rw-by-file-report +1 -1
index d83070b..407ec70 100644
@@ -7,7 +7,7 @@ if [ $# -lt 1 ] ; then
 fi
 comm=$1
 shift
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rw-by-file.pl $comm
+perf trace $@ -s perl/rw-by-file.pl $comm
 
 
 
tools/perf/scripts/perl/bin/rw-by-pid-report +1 -1
index 7ef4698..0c129e1 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: system-wide r/w activity
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rw-by-pid.pl
+perf trace $@ -s perl/rw-by-pid.pl
 
 
 
tools/perf/scripts/perl/bin/rwtop-report +1 -1
index 93e698c..5bac5bb 100644
@@ -17,7 +17,7 @@ if [ "$n_args" -gt 0 ] ; then
     interval=$1
     shift
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rwtop.pl $interval
+perf trace $@ -s perl/rwtop.pl $interval
 
 
 
tools/perf/scripts/perl/bin/wakeup-latency-report +1 -1
index a0d898f..dcfdd05 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: system-wide min/max/avg wakeup latency
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/wakeup-latency.pl
+perf trace $@ -s perl/wakeup-latency.pl
 
 
 
tools/perf/scripts/perl/bin/workqueue-stats-report +1 -1
index 3508113..6ce9247 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: workqueue stats (ins/exe/create/destroy)
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/workqueue-stats.pl
+perf trace $@ -s perl/workqueue-stats.pl
 
 
 
tools/perf/scripts/python/bin/failed-syscalls-by-pid-report +1 -1
index 3029354..b39182f 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/failed-syscalls-by-pid.py $comm
+perf trace $@ -s python/failed-syscalls-by-pid.py $comm
tools/perf/scripts/python/bin/sched-migration-report +1 -1
index 61d05f7..8299b69 100644
@@ -1,3 +1,3 @@
 #!/bin/bash
 # description: sched migration overview
-perf trace $@ -s ~/libexec/perf-core/scripts/python/sched-migration.py
+perf trace $@ -s python/sched-migration.py
tools/perf/scripts/python/bin/sctop-report +1 -1
index b01c842..9fe812c 100644
@@ -21,4 +21,4 @@ elif [ "$n_args" -gt 0 ] ; then
     interval=$1
     shift
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/sctop.py $comm $interval
+perf trace $@ -s python/sctop.py $comm $interval
tools/perf/scripts/python/bin/syscall-counts-by-pid-report +1 -1
index 9e9d8dd..ec9d4d9 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/syscall-counts-by-pid.py $comm
+perf trace $@ -s python/syscall-counts-by-pid.py $comm
tools/perf/scripts/python/bin/syscall-counts-report +2 -1
index dc076b6..9ed7854 100644
@@ -1,3 +1,4 @@
+
 #!/bin/bash
 # description: system-wide syscall counts
 # args: [comm]
@@ -7,4 +8,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/syscall-counts.py $comm
+perf trace $@ -s python/syscall-counts.py $comm
Signature · 1 line
1.6.5.rc2
#1 lore raw

Em Thu, Aug 12, 2010 at 12:59:18PM -0500, Tom Zanussi escreveu:

Quoted 7 lines

The perf trace report shell scripts hard-code the exec path of the scripts into their command-lines, which doesn't work if perf has been installed somewhere else.

Instead, perf trace should create the paths at run-time. This patch does that and removes the hard-coded paths from all the report scripts.

Quoted 7 lines

v2 changes: The first version inadvertantly caused scripts run from outside the perf exec path to fail e.g. 'perf trace -s test.py'. The fix is to try the script name without the exec path first, then the version using the exec path, which restores the expected behavior.

Stephane,

Lemme know when you tried this patch so that I can merge it with
a Tested-by: you tag, ok?

- Arnaldo

#2 lore raw
Quoted 13 lines · Arnaldo Carvalho de Melo

Em Thu, Aug 12, 2010 at 12:59:18PM -0500, Tom Zanussi escreveu:

Quoted 12 lines

The perf trace report shell scripts hard-code the exec path of the scripts into their command-lines, which doesn't work if perf has been installed somewhere else.

Instead, perf trace should create the paths at run-time.  This patch does that and removes the hard-coded paths from all the report scripts.

v2 changes: The first version inadvertantly caused scripts run from outside the perf exec path to fail e.g. 'perf trace -s test.py'.  The fix is to try the script name without the exec path first, then the version using the exec path, which restores the expected behavior.

Quoted 8 lines
Quoted 2 lines

Stephane,

       Lemme know when you tried this patch so that I can merge it with a Tested-by: you tag, ok?

Will get back to you tomorrow morning (your time).

#3 lore raw
In reply to Tom Zanussi
Quoted 51 lines · Tom Zanussi

The perf trace report shell scripts hard-code the exec path of the scripts into their command-lines, which doesn't work if perf has been installed somewhere else.

Instead, perf trace should create the paths at run-time.  This patch does that and removes the hard-coded paths from all the report scripts.

v2 changes: The first version inadvertantly caused scripts run from outside the perf exec path to fail e.g. 'perf trace -s test.py'.  The fix is to try the script name without the exec path first, then the version using the exec path, which restores the expected behavior.

---
 tools/perf/builtin-trace.c                         |   22 ++++++++++++++++--- tools/perf/scripts/perl/bin/failed-syscalls-report |    2 +- tools/perf/scripts/perl/bin/rw-by-file-report      |    2 +- tools/perf/scripts/perl/bin/rw-by-pid-report       |    2 +- tools/perf/scripts/perl/bin/rwtop-report           |    2 +- tools/perf/scripts/perl/bin/wakeup-latency-report  |    2 +- tools/perf/scripts/perl/bin/workqueue-stats-report |    2 +- .../python/bin/failed-syscalls-by-pid-report       |    2 +- .../perf/scripts/python/bin/sched-migration-report |    2 +- tools/perf/scripts/python/bin/sctop-report         |    2 +- .../python/bin/syscall-counts-by-pid-report        |    2 +- .../perf/scripts/python/bin/syscall-counts-report  |    3 +-

 12 files changed, 30 insertions(+), 15 deletions(-)
tools/perf/builtin-trace.c +1 -2
index 40a6a29..88a1883 100644
@@ -573,6 +573,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
        const char *suffix = NULL;
        const char **__argv;
        char *script_path;
+       struct stat perf_stat;
        int i, err;
 
        if (argc >= 2 && strncmp(argv[1], "rec", strlen("rec")) == 0) {
@@ -689,8 +690,6 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
                return -EINVAL;
 
        if (generate_script_lang) {
-               struct stat perf_stat;
-
                int input = open(input_name, O_RDONLY);
                if (input < 0) {
                        perror("failed to open file");

tools/perf/builtin-trace.c int cmd_trace(int argc, const char **argv, const char *prefix __used) · line 724

Quoted 8 lines
tools/perf/builtin-trace.c +4 -1
@@ -719,10 +718,25 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
        }
 
        if (script_name) {
-               err = scripting_ops->start_script(script_name, argc, argv);
+               char script_exec_path[MAXPATHLEN];
+
+               snprintf(script_exec_path, MAXPATHLEN, "%s", script_name);
+               err = stat(script_exec_path, &perf_stat);

Why not simply use access() instead of stat() here?

tools/perf/builtin-trace.c int cmd_trace(int argc, const char **argv, const char *prefix __used) · line 728

Quoted 4 lines
tools/perf/builtin-trace.c +4 -0
+               if (err < 0) {
+                       snprintf(script_exec_path, MAXPATHLEN, "%s/scripts/%s",
+                                perf_exec_path(), script_name);
+                       err = stat(script_exec_path, &perf_stat);

And here?

#4 lore raw
Quoted 54 lines · Stephane Eranian
Quoted 53 lines · Tom Zanussi

The perf trace report shell scripts hard-code the exec path of the scripts into their command-lines, which doesn't work if perf has been installed somewhere else.

Instead, perf trace should create the paths at run-time.  This patch does that and removes the hard-coded paths from all the report scripts.

v2 changes: The first version inadvertantly caused scripts run from outside the perf exec path to fail e.g. 'perf trace -s test.py'.  The fix is to try the script name without the exec path first, then the version using the exec path, which restores the expected behavior.

---
 tools/perf/builtin-trace.c                         |   22 ++++++++++++++++--- tools/perf/scripts/perl/bin/failed-syscalls-report |    2 +- tools/perf/scripts/perl/bin/rw-by-file-report      |    2 +- tools/perf/scripts/perl/bin/rw-by-pid-report       |    2 +- tools/perf/scripts/perl/bin/rwtop-report           |    2 +- tools/perf/scripts/perl/bin/wakeup-latency-report  |    2 +- tools/perf/scripts/perl/bin/workqueue-stats-report |    2 +- .../python/bin/failed-syscalls-by-pid-report       |    2 +- .../perf/scripts/python/bin/sched-migration-report |    2 +- tools/perf/scripts/python/bin/sctop-report         |    2 +- .../python/bin/syscall-counts-by-pid-report        |    2 +- .../perf/scripts/python/bin/syscall-counts-report  |    3 +-

 12 files changed, 30 insertions(+), 15 deletions(-)
tools/perf/builtin-trace.c +1 -2
index 40a6a29..88a1883 100644
@@ -573,6 +573,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
        const char *suffix = NULL;
        const char **__argv;
        char *script_path;
+       struct stat perf_stat;
        int i, err;
 
        if (argc >= 2 && strncmp(argv[1], "rec", strlen("rec")) == 0) {
@@ -689,8 +690,6 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
                return -EINVAL;
 
        if (generate_script_lang) {
-               struct stat perf_stat;
-
                int input = open(input_name, O_RDONLY);
                if (input < 0) {
                        perror("failed to open file");
@@ -719,10 +718,25 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
        }
 
Quoted 8 lines

tools/perf/builtin-trace.c int cmd_trace(int argc, const char **argv, const char *prefix __used) · line 724

Quoted 6 lines
tools/perf/builtin-trace.c +4 -1
        if (script_name) {
-               err = scripting_ops->start_script(script_name, argc, argv);
+               char script_exec_path[MAXPATHLEN];
+
+               snprintf(script_exec_path, MAXPATHLEN, "%s", script_name);
+               err = stat(script_exec_path, &perf_stat);

Why not simply use access() instead of stat() here?

Yeah, that would probably be better - if this patch works for you, I can submit a follow-on patch to do that along with some other small patches I'm hoping to get to over the weekend...

Thanks,

Tom

#5 lore raw
In reply to Tom Zanussi

Tom,

Here is version that works for me. Besides the stat() to access() change, I fixed the pathname you were building, it was missing the language name. The way I solved that is by leveraging the scripting_ops->name field, but for that I had to make sure it would match the subdir name, i.e., all lower-case. I also add an error message when the script is not found, so users understand what went wrong without turning on any sort of debugging.

tools/perf/builtin-trace.c +21 -2
index 40a6a29..db1d950 100644
@@ -719,10 +719,29 @@ int cmd_trace(int argc, const char **argv, const char *prefix __used)
 	}
 
 	if (script_name) {
-		err = scripting_ops->start_script(script_name, argc, argv);
+		char script_exec_path[MAXPATHLEN];
+
+		snprintf(script_exec_path, MAXPATHLEN, "%s", script_name);
+
+		if (access(script_exec_path, R_OK)) {
+
+			snprintf(script_exec_path, MAXPATHLEN, "%s/scripts/%s/%s",
+				 perf_exec_path(), scripting_ops->name, script_name);
+
+			err = access(script_exec_path, R_OK);
+			if (err) {
+				fprintf(stderr, "script %s not found\n", script_name);
+				goto out;
+			}
+		}
+
+		err = scripting_ops->start_script(script_exec_path,
+						  argc, argv);
 		if (err)
 			goto out;
-		pr_debug("perf trace started with script %s\n\n", script_name);
+
+		pr_debug("perf trace started with script %s\n\n",
+			 script_exec_path);
 	}
 
 	err = __cmd_trace(session);
tools/perf/scripts/perl/bin/failed-syscalls-report +1 -1
index e3a5e55..47cf0b7 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/failed-syscalls.pl $comm
+perf trace $@ -s perl/failed-syscalls.pl $comm
tools/perf/scripts/perl/bin/rw-by-file-report +1 -1
index d83070b..407ec70 100644
@@ -7,7 +7,7 @@ if [ $# -lt 1 ] ; then
 fi
 comm=$1
 shift
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rw-by-file.pl $comm
+perf trace $@ -s perl/rw-by-file.pl $comm
 
 
 
tools/perf/scripts/perl/bin/rw-by-pid-report +1 -1
index 7ef4698..0c129e1 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: system-wide r/w activity
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rw-by-pid.pl
+perf trace $@ -s perl/rw-by-pid.pl
 
 
 
tools/perf/scripts/perl/bin/rwtop-report +1 -1
index 93e698c..5bac5bb 100644
@@ -17,7 +17,7 @@ if [ "$n_args" -gt 0 ] ; then
     interval=$1
     shift
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/rwtop.pl $interval
+perf trace $@ -s perl/rwtop.pl $interval
 
 
 
tools/perf/scripts/perl/bin/wakeup-latency-report +1 -1
index a0d898f..dcfdd05 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: system-wide min/max/avg wakeup latency
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/wakeup-latency.pl
+perf trace $@ -s perl/wakeup-latency.pl
 
 
 
tools/perf/scripts/perl/bin/workqueue-stats-report +1 -1
index 3508113..6ce9247 100644
@@ -1,6 +1,6 @@
 #!/bin/bash
 # description: workqueue stats (ins/exe/create/destroy)
-perf trace $@ -s ~/libexec/perf-core/scripts/perl/workqueue-stats.pl
+perf trace $@ -s perl/workqueue-stats.pl
 
 
 
tools/perf/scripts/python/bin/failed-syscalls-by-pid-report +1 -1
index 3029354..b39182f 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/failed-syscalls-by-pid.py $comm
+perf trace $@ -s python/failed-syscalls-by-pid.py $comm
tools/perf/scripts/python/bin/sched-migration-report +1 -1
index 61d05f7..8299b69 100644
@@ -1,3 +1,3 @@
 #!/bin/bash
 # description: sched migration overview
-perf trace $@ -s ~/libexec/perf-core/scripts/python/sched-migration.py
+perf trace $@ -s python/sched-migration.py
tools/perf/scripts/python/bin/sctop-report +1 -1
index b01c842..9fe812c 100644
@@ -21,4 +21,4 @@ elif [ "$n_args" -gt 0 ] ; then
     interval=$1
     shift
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/sctop.py $comm $interval
+perf trace $@ -s python/sctop.py $comm $interval
tools/perf/scripts/python/bin/syscall-counts-by-pid-report +1 -1
index 9e9d8dd..ec9d4d9 100644
@@ -7,4 +7,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/syscall-counts-by-pid.py $comm
+perf trace $@ -s python/syscall-counts-by-pid.py $comm
tools/perf/scripts/python/bin/syscall-counts-report +2 -1
index dc076b6..9ed7854 100644
@@ -1,3 +1,4 @@
+
 #!/bin/bash
 # description: system-wide syscall counts
 # args: [comm]
@@ -7,4 +8,4 @@ if [ $# -gt 0 ] ; then
 	shift
     fi
 fi
-perf trace $@ -s ~/libexec/perf-core/scripts/python/syscall-counts.py $comm
+perf trace $@ -s python/syscall-counts.py $comm
tools/perf/util/scripting-engines/trace-event-perl.c +1 -1
index b059dc5..97e8c6d 100644
@@ -557,7 +557,7 @@ static int perl_generate_script(const char *outfile)
 }
 
 struct scripting_ops perl_scripting_ops = {
-	.name = "Perl",
+	.name = "perl",
 	.start_script = perl_start_script,
 	.stop_script = perl_stop_script,
 	.process_event = perl_process_event,
tools/perf/util/scripting-engines/trace-event-python.c +1 -1
index 33a6325..49655a8 100644
@@ -586,7 +586,7 @@ static int python_generate_script(const char *outfile)
 }
 
 struct scripting_ops python_scripting_ops = {
-	.name = "Python",
+	.name = "python",
 	.start_script = python_start_script,
 	.stop_script = python_stop_script,
 	.process_event = python_process_event,
#6 lore raw

Hi Stephane,

Quoted 9 lines · Stephane Eranian

Tom,

Here is version that works for me. Besides the stat() to access() change, I fixed the pathname you were building, it was missing the language name. The way I solved that is by leveraging the scripting_ops->name field, but for that I had to make sure it would match the subdir name, i.e., all lower-case. I also add an error message when the script is not found, so users understand what went wrong without turning on any sort of debugging.

Hmm, I'm not sure why the previous version didn't work for you, since that version changed all the scripts to the form:

python/sctop.py

which applied to the format string there, "%s/scripts/%s"

gives

..../scripts/python/sctop.py

so it should have worked.

With your version, I get:

root@tropicana:~# perf trace sctop
script python/sctop.py not found

because the script names still have the form python/sctop.py but are applied to the new format string "%s/scripts/%s/%s" which gives:

..../scripts/python/python/sctop.py

After removing the python/ prefix from all the scripts, everything works fine:

root@tropicana:~# perf trace sctop
good output...
root@tropicana:~# perf record -c 1 -e raw_syscalls:sys_enter -a -- sleep
1
[ perf record: Woken up 0 times to write data ]
[ perf record: Captured and wrote 22.546 MB perf.data (~985061
samples) ]
root@tropicana:~# perf trace -g python
generated Python script: perf-trace.py
root@tropicana:~# perf trace -s perf-trace.py > out
good output...

Anyway, for all the reasons you stated and the fact that it gets rid of everything but the bare script name, I think your version is a nice improvement - I'll repost it along with the small script changes mentioned shortly.

Tom