1perf-script-python(1)
2====================
3
4NAME
5----
6perf-script-python - Process trace data with a Python script
7
8SYNOPSIS
9--------
10[verse]
11'perf script' [-s [Python]:script[.py] ]
12
13DESCRIPTION
14-----------
15
16This perf script option is used to process perf script data using perf's
17built-in Python interpreter.  It reads and processes the input file and
18displays the results of the trace analysis implemented in the given
19Python script, if any.
20
21A QUICK EXAMPLE
22---------------
23
24This section shows the process, start to finish, of creating a working
25Python script that aggregates and extracts useful information from a
26raw perf script stream.  You can avoid reading the rest of this
27document if an example is enough for you; the rest of the document
28provides more details on each step and lists the library functions
29available to script writers.
30
31This example actually details the steps that were used to create the
32'syscall-counts' script you see when you list the available perf script
33scripts via 'perf script -l'.  As such, this script also shows how to
34integrate your script into the list of general-purpose 'perf script'
35scripts listed by that command.
36
37The syscall-counts script is a simple script, but demonstrates all the
38basic ideas necessary to create a useful script.  Here's an example
39of its output (syscall names are not yet supported, they will appear
40as numbers):
41
42----
43syscall events:
44
45event                                          count
46----------------------------------------  -----------
47sys_write                                     455067
48sys_getdents                                    4072
49sys_close                                       3037
50sys_swapoff                                     1769
51sys_read                                         923
52sys_sched_setparam                               826
53sys_open                                         331
54sys_newfstat                                     326
55sys_mmap                                         217
56sys_munmap                                       216
57sys_futex                                        141
58sys_select                                       102
59sys_poll                                          84
60sys_setitimer                                     12
61sys_writev                                         8
6215                                                 8
63sys_lseek                                          7
64sys_rt_sigprocmask                                 6
65sys_wait4                                          3
66sys_ioctl                                          3
67sys_set_robust_list                                1
68sys_exit                                           1
6956                                                 1
70sys_access                                         1
71----
72
73Basically our task is to keep a per-syscall tally that gets updated
74every time a system call occurs in the system.  Our script will do
75that, but first we need to record the data that will be processed by
76that script.  Theoretically, there are a couple of ways we could do
77that:
78
79- we could enable every event under the tracing/events/syscalls
80  directory, but this is over 600 syscalls, well beyond the number
81  allowable by perf.  These individual syscall events will however be
82  useful if we want to later use the guidance we get from the
83  general-purpose scripts to drill down and get more detail about
84  individual syscalls of interest.
85
86- we can enable the sys_enter and/or sys_exit syscalls found under
87  tracing/events/raw_syscalls.  These are called for all syscalls; the
88  'id' field can be used to distinguish between individual syscall
89  numbers.
90
91For this script, we only need to know that a syscall was entered; we
92don't care how it exited, so we'll use 'perf record' to record only
93the sys_enter events:
94
95----
96# perf record -a -e raw_syscalls:sys_enter
97
98^C[ perf record: Woken up 1 times to write data ]
99[ perf record: Captured and wrote 56.545 MB perf.data (~2470503 samples) ]
100----
101
102The options basically say to collect data for every syscall event
103system-wide and multiplex the per-cpu output into a single stream.
104That single stream will be recorded in a file in the current directory
105called perf.data.
106
107Once we have a perf.data file containing our data, we can use the -g
108'perf script' option to generate a Python script that will contain a
109callback handler for each event type found in the perf.data trace
110stream (for more details, see the STARTER SCRIPTS section).
111
112----
113# perf script -g python
114generated Python script: perf-script.py
115
116The output file created also in the current directory is named
117perf-script.py.  Here's the file in its entirety:
118
119# perf script event handlers, generated by perf script -g python
120# Licensed under the terms of the GNU GPL License version 2
121
122# The common_* event handler fields are the most useful fields common to
123# all events.  They don't necessarily correspond to the 'common_*' fields
124# in the format files.  Those fields not available as handler params can
125# be retrieved using Python functions of the form common_*(context).
126# See the perf-script-python Documentation for the list of available functions.
127
128import os
129import sys
130
131sys.path.append(os.environ['PERF_EXEC_PATH'] + \
132	'/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
133
134from perf_trace_context import *
135from Core import *
136
137def trace_begin():
138	print "in trace_begin"
139
140def trace_end():
141	print "in trace_end"
142
143def raw_syscalls__sys_enter(event_name, context, common_cpu,
144	common_secs, common_nsecs, common_pid, common_comm,
145	id, args):
146		print_header(event_name, common_cpu, common_secs, common_nsecs,
147			common_pid, common_comm)
148
149		print "id=%d, args=%s\n" % \
150		(id, args),
151
152def trace_unhandled(event_name, context, event_fields_dict):
153		print ' '.join(['%s=%s'%(k,str(v))for k,v in sorted(event_fields_dict.items())])
154
155def print_header(event_name, cpu, secs, nsecs, pid, comm):
156	print "%-20s %5u %05u.%09u %8u %-20s " % \
157	(event_name, cpu, secs, nsecs, pid, comm),
158----
159
160At the top is a comment block followed by some import statements and a
161path append which every perf script script should include.
162
163Following that are a couple generated functions, trace_begin() and
164trace_end(), which are called at the beginning and the end of the
165script respectively (for more details, see the SCRIPT_LAYOUT section
166below).
167
168Following those are the 'event handler' functions generated one for
169every event in the 'perf record' output.  The handler functions take
170the form subsystem__event_name, and contain named parameters, one for
171each field in the event; in this case, there's only one event,
172raw_syscalls__sys_enter().  (see the EVENT HANDLERS section below for
173more info on event handlers).
174
175The final couple of functions are, like the begin and end functions,
176generated for every script.  The first, trace_unhandled(), is called
177every time the script finds an event in the perf.data file that
178doesn't correspond to any event handler in the script.  This could
179mean either that the record step recorded event types that it wasn't
180really interested in, or the script was run against a trace file that
181doesn't correspond to the script.
182
183The script generated by -g option simply prints a line for each
184event found in the trace stream i.e. it basically just dumps the event
185and its parameter values to stdout.  The print_header() function is
186simply a utility function used for that purpose.  Let's rename the
187script and run it to see the default output:
188
189----
190# mv perf-script.py syscall-counts.py
191# perf script -s syscall-counts.py
192
193raw_syscalls__sys_enter     1 00840.847582083     7506 perf                  id=1, args=
194raw_syscalls__sys_enter     1 00840.847595764     7506 perf                  id=1, args=
195raw_syscalls__sys_enter     1 00840.847620860     7506 perf                  id=1, args=
196raw_syscalls__sys_enter     1 00840.847710478     6533 npviewer.bin          id=78, args=
197raw_syscalls__sys_enter     1 00840.847719204     6533 npviewer.bin          id=142, args=
198raw_syscalls__sys_enter     1 00840.847755445     6533 npviewer.bin          id=3, args=
199raw_syscalls__sys_enter     1 00840.847775601     6533 npviewer.bin          id=3, args=
200raw_syscalls__sys_enter     1 00840.847781820     6533 npviewer.bin          id=3, args=
201.
202.
203.
204----
205
206Of course, for this script, we're not interested in printing every
207trace event, but rather aggregating it in a useful way.  So we'll get
208rid of everything to do with printing as well as the trace_begin() and
209trace_unhandled() functions, which we won't be using.  That leaves us
210with this minimalistic skeleton:
211
212----
213import os
214import sys
215
216sys.path.append(os.environ['PERF_EXEC_PATH'] + \
217	'/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
218
219from perf_trace_context import *
220from Core import *
221
222def trace_end():
223	print "in trace_end"
224
225def raw_syscalls__sys_enter(event_name, context, common_cpu,
226	common_secs, common_nsecs, common_pid, common_comm,
227	id, args):
228----
229
230In trace_end(), we'll simply print the results, but first we need to
231generate some results to print.  To do that we need to have our
232sys_enter() handler do the necessary tallying until all events have
233been counted.  A hash table indexed by syscall id is a good way to
234store that information; every time the sys_enter() handler is called,
235we simply increment a count associated with that hash entry indexed by
236that syscall id:
237
238----
239  syscalls = autodict()
240
241  try:
242    syscalls[id] += 1
243  except TypeError:
244    syscalls[id] = 1
245----
246
247The syscalls 'autodict' object is a special kind of Python dictionary
248(implemented in Core.py) that implements Perl's 'autovivifying' hashes
249in Python i.e. with autovivifying hashes, you can assign nested hash
250values without having to go to the trouble of creating intermediate
251levels if they don't exist e.g syscalls[comm][pid][id] = 1 will create
252the intermediate hash levels and finally assign the value 1 to the
253hash entry for 'id' (because the value being assigned isn't a hash
254object itself, the initial value is assigned in the TypeError
255exception.  Well, there may be a better way to do this in Python but
256that's what works for now).
257
258Putting that code into the raw_syscalls__sys_enter() handler, we
259effectively end up with a single-level dictionary keyed on syscall id
260and having the counts we've tallied as values.
261
262The print_syscall_totals() function iterates over the entries in the
263dictionary and displays a line for each entry containing the syscall
264name (the dictionary keys contain the syscall ids, which are passed to
265the Util function syscall_name(), which translates the raw syscall
266numbers to the corresponding syscall name strings).  The output is
267displayed after all the events in the trace have been processed, by
268calling the print_syscall_totals() function from the trace_end()
269handler called at the end of script processing.
270
271The final script producing the output shown above is shown in its
272entirety below (syscall_name() helper is not yet available, you can
273only deal with id's for now):
274
275----
276import os
277import sys
278
279sys.path.append(os.environ['PERF_EXEC_PATH'] + \
280	'/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
281
282from perf_trace_context import *
283from Core import *
284from Util import *
285
286syscalls = autodict()
287
288def trace_end():
289	print_syscall_totals()
290
291def raw_syscalls__sys_enter(event_name, context, common_cpu,
292	common_secs, common_nsecs, common_pid, common_comm,
293	id, args):
294	try:
295		syscalls[id] += 1
296	except TypeError:
297		syscalls[id] = 1
298
299def print_syscall_totals():
300    if for_comm is not None:
301	    print "\nsyscall events for %s:\n\n" % (for_comm),
302    else:
303	    print "\nsyscall events:\n\n",
304
305    print "%-40s  %10s\n" % ("event", "count"),
306    print "%-40s  %10s\n" % ("----------------------------------------", \
307                                 "-----------"),
308
309    for id, val in sorted(syscalls.iteritems(), key = lambda(k, v): (v, k), \
310				  reverse = True):
311	    print "%-40s  %10d\n" % (syscall_name(id), val),
312----
313
314The script can be run just as before:
315
316  # perf script -s syscall-counts.py
317
318So those are the essential steps in writing and running a script.  The
319process can be generalized to any tracepoint or set of tracepoints
320you're interested in - basically find the tracepoint(s) you're
321interested in by looking at the list of available events shown by
322'perf list' and/or look in /sys/kernel/debug/tracing/events/ for
323detailed event and field info, record the corresponding trace data
324using 'perf record', passing it the list of interesting events,
325generate a skeleton script using 'perf script -g python' and modify the
326code to aggregate and display it for your particular needs.
327
328After you've done that you may end up with a general-purpose script
329that you want to keep around and have available for future use.  By
330writing a couple of very simple shell scripts and putting them in the
331right place, you can have your script listed alongside the other
332scripts listed by the 'perf script -l' command e.g.:
333
334----
335# perf script -l
336List of available trace scripts:
337  wakeup-latency                       system-wide min/max/avg wakeup latency
338  rw-by-file <comm>                    r/w activity for a program, by file
339  rw-by-pid                            system-wide r/w activity
340----
341
342A nice side effect of doing this is that you also then capture the
343probably lengthy 'perf record' command needed to record the events for
344the script.
345
346To have the script appear as a 'built-in' script, you write two simple
347scripts, one for recording and one for 'reporting'.
348
349The 'record' script is a shell script with the same base name as your
350script, but with -record appended.  The shell script should be put
351into the perf/scripts/python/bin directory in the kernel source tree.
352In that script, you write the 'perf record' command-line needed for
353your script:
354
355----
356# cat kernel-source/tools/perf/scripts/python/bin/syscall-counts-record
357
358#!/bin/bash
359perf record -a -e raw_syscalls:sys_enter
360----
361
362The 'report' script is also a shell script with the same base name as
363your script, but with -report appended.  It should also be located in
364the perf/scripts/python/bin directory.  In that script, you write the
365'perf script -s' command-line needed for running your script:
366
367----
368# cat kernel-source/tools/perf/scripts/python/bin/syscall-counts-report
369
370#!/bin/bash
371# description: system-wide syscall counts
372perf script -s ~/libexec/perf-core/scripts/python/syscall-counts.py
373----
374
375Note that the location of the Python script given in the shell script
376is in the libexec/perf-core/scripts/python directory - this is where
377the script will be copied by 'make install' when you install perf.
378For the installation to install your script there, your script needs
379to be located in the perf/scripts/python directory in the kernel
380source tree:
381
382----
383# ls -al kernel-source/tools/perf/scripts/python
384total 32
385drwxr-xr-x 4 trz trz 4096 2010-01-26 22:30 .
386drwxr-xr-x 4 trz trz 4096 2010-01-26 22:29 ..
387drwxr-xr-x 2 trz trz 4096 2010-01-26 22:29 bin
388-rw-r--r-- 1 trz trz 2548 2010-01-26 22:29 check-perf-script.py
389drwxr-xr-x 3 trz trz 4096 2010-01-26 22:49 Perf-Trace-Util
390-rw-r--r-- 1 trz trz 1462 2010-01-26 22:30 syscall-counts.py
391----
392
393Once you've done that (don't forget to do a new 'make install',
394otherwise your script won't show up at run-time), 'perf script -l'
395should show a new entry for your script:
396
397----
398# perf script -l
399List of available trace scripts:
400  wakeup-latency                       system-wide min/max/avg wakeup latency
401  rw-by-file <comm>                    r/w activity for a program, by file
402  rw-by-pid                            system-wide r/w activity
403  syscall-counts                       system-wide syscall counts
404----
405
406You can now perform the record step via 'perf script record':
407
408  # perf script record syscall-counts
409
410and display the output using 'perf script report':
411
412  # perf script report syscall-counts
413
414STARTER SCRIPTS
415---------------
416
417You can quickly get started writing a script for a particular set of
418trace data by generating a skeleton script using 'perf script -g
419python' in the same directory as an existing perf.data trace file.
420That will generate a starter script containing a handler for each of
421the event types in the trace file; it simply prints every available
422field for each event in the trace file.
423
424You can also look at the existing scripts in
425~/libexec/perf-core/scripts/python for typical examples showing how to
426do basic things like aggregate event data, print results, etc.  Also,
427the check-perf-script.py script, while not interesting for its results,
428attempts to exercise all of the main scripting features.
429
430EVENT HANDLERS
431--------------
432
433When perf script is invoked using a trace script, a user-defined
434'handler function' is called for each event in the trace.  If there's
435no handler function defined for a given event type, the event is
436ignored (or passed to a 'trace_unhandled' function, see below) and the
437next event is processed.
438
439Most of the event's field values are passed as arguments to the
440handler function; some of the less common ones aren't - those are
441available as calls back into the perf executable (see below).
442
443As an example, the following perf record command can be used to record
444all sched_wakeup events in the system:
445
446 # perf record -a -e sched:sched_wakeup
447
448Traces meant to be processed using a script should be recorded with
449the above option: -a to enable system-wide collection.
450
451The format file for the sched_wakep event defines the following fields
452(see /sys/kernel/debug/tracing/events/sched/sched_wakeup/format):
453
454----
455 format:
456        field:unsigned short common_type;
457        field:unsigned char common_flags;
458        field:unsigned char common_preempt_count;
459        field:int common_pid;
460
461        field:char comm[TASK_COMM_LEN];
462        field:pid_t pid;
463        field:int prio;
464        field:int success;
465        field:int target_cpu;
466----
467
468The handler function for this event would be defined as:
469
470----
471def sched__sched_wakeup(event_name, context, common_cpu, common_secs,
472       common_nsecs, common_pid, common_comm,
473       comm, pid, prio, success, target_cpu):
474       pass
475----
476
477The handler function takes the form subsystem__event_name.
478
479The common_* arguments in the handler's argument list are the set of
480arguments passed to all event handlers; some of the fields correspond
481to the common_* fields in the format file, but some are synthesized,
482and some of the common_* fields aren't common enough to to be passed
483to every event as arguments but are available as library functions.
484
485Here's a brief description of each of the invariant event args:
486
487 event_name 	  	    the name of the event as text
488 context		    an opaque 'cookie' used in calls back into perf
489 common_cpu		    the cpu the event occurred on
490 common_secs		    the secs portion of the event timestamp
491 common_nsecs		    the nsecs portion of the event timestamp
492 common_pid		    the pid of the current task
493 common_comm		    the name of the current process
494
495All of the remaining fields in the event's format file have
496counterparts as handler function arguments of the same name, as can be
497seen in the example above.
498
499The above provides the basics needed to directly access every field of
500every event in a trace, which covers 90% of what you need to know to
501write a useful trace script.  The sections below cover the rest.
502
503SCRIPT LAYOUT
504-------------
505
506Every perf script Python script should start by setting up a Python
507module search path and 'import'ing a few support modules (see module
508descriptions below):
509
510----
511 import os
512 import sys
513
514 sys.path.append(os.environ['PERF_EXEC_PATH'] + \
515	      '/scripts/python/Perf-Trace-Util/lib/Perf/Trace')
516
517 from perf_trace_context import *
518 from Core import *
519----
520
521The rest of the script can contain handler functions and support
522functions in any order.
523
524Aside from the event handler functions discussed above, every script
525can implement a set of optional functions:
526
527*trace_begin*, if defined, is called before any event is processed and
528gives scripts a chance to do setup tasks:
529
530----
531def trace_begin():
532    pass
533----
534
535*trace_end*, if defined, is called after all events have been
536 processed and gives scripts a chance to do end-of-script tasks, such
537 as display results:
538
539----
540def trace_end():
541    pass
542----
543
544*trace_unhandled*, if defined, is called after for any event that
545 doesn't have a handler explicitly defined for it.  The standard set
546 of common arguments are passed into it:
547
548----
549def trace_unhandled(event_name, context, event_fields_dict):
550    pass
551----
552
553The remaining sections provide descriptions of each of the available
554built-in perf script Python modules and their associated functions.
555
556AVAILABLE MODULES AND FUNCTIONS
557-------------------------------
558
559The following sections describe the functions and variables available
560via the various perf script Python modules.  To use the functions and
561variables from the given module, add the corresponding 'from XXXX
562import' line to your perf script script.
563
564Core.py Module
565~~~~~~~~~~~~~~
566
567These functions provide some essential functions to user scripts.
568
569The *flag_str* and *symbol_str* functions provide human-readable
570strings for flag and symbolic fields.  These correspond to the strings
571and values parsed from the 'print fmt' fields of the event format
572files:
573
574  flag_str(event_name, field_name, field_value) - returns the string representation corresponding to field_value for the flag field field_name of event event_name
575  symbol_str(event_name, field_name, field_value) - returns the string representation corresponding to field_value for the symbolic field field_name of event event_name
576
577The *autodict* function returns a special kind of Python
578dictionary that implements Perl's 'autovivifying' hashes in Python
579i.e. with autovivifying hashes, you can assign nested hash values
580without having to go to the trouble of creating intermediate levels if
581they don't exist.
582
583  autodict() - returns an autovivifying dictionary instance
584
585
586perf_trace_context Module
587~~~~~~~~~~~~~~~~~~~~~~~~~
588
589Some of the 'common' fields in the event format file aren't all that
590common, but need to be made accessible to user scripts nonetheless.
591
592perf_trace_context defines a set of functions that can be used to
593access this data in the context of the current event.  Each of these
594functions expects a context variable, which is the same as the
595context variable passed into every event handler as the second
596argument.
597
598 common_pc(context) - returns common_preempt count for the current event
599 common_flags(context) - returns common_flags for the current event
600 common_lock_depth(context) - returns common_lock_depth for the current event
601
602Util.py Module
603~~~~~~~~~~~~~~
604
605Various utility functions for use with perf script:
606
607  nsecs(secs, nsecs) - returns total nsecs given secs/nsecs pair
608  nsecs_secs(nsecs) - returns whole secs portion given nsecs
609  nsecs_nsecs(nsecs) - returns nsecs remainder given nsecs
610  nsecs_str(nsecs) - returns printable string in the form secs.nsecs
611  avg(total, n) - returns average given a sum and a total number of values
612
613SUPPORTED FIELDS
614----------------
615
616Currently supported fields:
617
618ev_name, comm, pid, tid, cpu, ip, time, period, phys_addr, addr,
619symbol, dso, time_enabled, time_running, values, callchain,
620brstack, brstacksym, datasrc, datasrc_decode, iregs, uregs,
621weight, transaction, raw_buf, attr.
622
623Some fields have sub items:
624
625brstack:
626    from, to, from_dsoname, to_dsoname, mispred,
627    predicted, in_tx, abort, cycles.
628
629brstacksym:
630    items: from, to, pred, in_tx, abort (converted string)
631
632For example,
633We can use this code to print brstack "from", "to", "cycles".
634
635if 'brstack' in dict:
636	for entry in dict['brstack']:
637		print "from %s, to %s, cycles %s" % (entry["from"], entry["to"], entry["cycles"])
638
639SEE ALSO
640--------
641linkperf:perf-script[1]
642