perf trace: Add option to analyze events in a file versus live
authorDavid Ahern <dsahern@gmail.com>
Thu, 29 Aug 2013 04:29:52 +0000 (22:29 -0600)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 29 Aug 2013 20:42:34 +0000 (17:42 -0300)
Allows capture of raw_syscall:* events and analyzed at a later time.

v2: change -i option from inherit to input name for consistency with
    other perf commands

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1377750593-48046-3-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/Documentation/perf-trace.txt
tools/perf/builtin-trace.c

index fe19811faf90ac00c61adca199ed4137a8a482fd..daccd2c0a48fe9c933cf50c870f68aa9bfe253a6 100644 (file)
@@ -74,6 +74,10 @@ the thread executes on the designated CPUs. Default is to monitor all CPUs.
 --sched:
        Accrue thread runtime and provide a summary at the end of the session.
 
+-i
+--input
+       Process events from a given perf data file.
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-script[1]
index 88387c565687a961a35e55a22b6b8c29fc08d82b..2a6ebe1848027805bcf427fb3ac6fb482ac861db 100644 (file)
@@ -4,6 +4,7 @@
 #include "util/debug.h"
 #include "util/evlist.h"
 #include "util/machine.h"
+#include "util/session.h"
 #include "util/thread.h"
 #include "util/parse-options.h"
 #include "util/strlist.h"
@@ -652,6 +653,36 @@ out_dump:
        return 0;
 }
 
+static int trace__process_sample(struct perf_tool *tool,
+                                union perf_event *event __maybe_unused,
+                                struct perf_sample *sample,
+                                struct perf_evsel *evsel,
+                                struct machine *machine __maybe_unused)
+{
+       struct trace *trace = container_of(tool, struct trace, tool);
+       int err = 0;
+
+       tracepoint_handler handler = evsel->handler.func;
+
+       if (trace->base_time == 0)
+               trace->base_time = sample->time;
+
+       if (handler)
+               handler(trace, evsel, sample);
+
+       return err;
+}
+
+static bool
+perf_session__has_tp(struct perf_session *session, const char *name)
+{
+       struct perf_evsel *evsel;
+
+       evsel = perf_evlist__find_tracepoint_by_name(session->evlist, name);
+
+       return evsel != NULL;
+}
+
 static int trace__run(struct trace *trace, int argc, const char **argv)
 {
        struct perf_evlist *evlist = perf_evlist__new();
@@ -791,6 +822,65 @@ out:
        return err;
 }
 
+static int trace__replay(struct trace *trace)
+{
+       const struct perf_evsel_str_handler handlers[] = {
+               { "raw_syscalls:sys_enter",  trace__sys_enter, },
+               { "raw_syscalls:sys_exit",   trace__sys_exit, },
+       };
+
+       struct perf_session *session;
+       int err = -1;
+
+       trace->tool.sample        = trace__process_sample;
+       trace->tool.mmap          = perf_event__process_mmap;
+       trace->tool.comm          = perf_event__process_comm;
+       trace->tool.exit          = perf_event__process_exit;
+       trace->tool.fork          = perf_event__process_fork;
+       trace->tool.attr          = perf_event__process_attr;
+       trace->tool.tracing_data = perf_event__process_tracing_data;
+       trace->tool.build_id      = perf_event__process_build_id;
+
+       trace->tool.ordered_samples = true;
+       trace->tool.ordering_requires_timestamps = true;
+
+       /* add tid to output */
+       trace->multiple_threads = true;
+
+       if (symbol__init() < 0)
+               return -1;
+
+       session = perf_session__new(input_name, O_RDONLY, 0, false,
+                                   &trace->tool);
+       if (session == NULL)
+               return -ENOMEM;
+
+       err = perf_session__set_tracepoints_handlers(session, handlers);
+       if (err)
+               goto out;
+
+       if (!perf_session__has_tp(session, "raw_syscalls:sys_enter")) {
+               pr_err("Data file does not have raw_syscalls:sys_enter events\n");
+               goto out;
+       }
+
+       if (!perf_session__has_tp(session, "raw_syscalls:sys_exit")) {
+               pr_err("Data file does not have raw_syscalls:sys_exit events\n");
+               goto out;
+       }
+
+       setup_pager();
+
+       err = perf_session__process_events(session, &trace->tool);
+       if (err)
+               pr_err("Failed to process events, error %d", err);
+
+out:
+       perf_session__delete(session);
+
+       return err;
+}
+
 static size_t trace__fprintf_threads_header(FILE *fp)
 {
        size_t printed;
@@ -892,6 +982,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
        OPT_STRING('e', "expr", &ev_qualifier_str, "expr",
                    "list of events to trace"),
        OPT_STRING('o', "output", &output_name, "file", "output file name"),
+       OPT_STRING('i', "input", &input_name, "file", "Analyze events in file"),
        OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
                    "trace events on existing process id"),
        OPT_STRING('t', "tid", &trace.opts.target.tid, "tid",
@@ -900,7 +991,7 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
                    "system-wide collection from all CPUs"),
        OPT_STRING('C', "cpu", &trace.opts.target.cpu_list, "cpu",
                    "list of cpus to monitor"),
-       OPT_BOOLEAN('i', "no-inherit", &trace.opts.no_inherit,
+       OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
                    "child tasks do not inherit counters"),
        OPT_UINTEGER('m', "mmap-pages", &trace.opts.mmap_pages,
                     "number of mmap data pages"),
@@ -958,7 +1049,10 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
        if (!argc && perf_target__none(&trace.opts.target))
                trace.opts.target.system_wide = true;
 
-       err = trace__run(&trace, argc, argv);
+       if (input_name)
+               err = trace__replay(&trace);
+       else
+               err = trace__run(&trace, argc, argv);
 
        if (trace.sched && !err)
                trace__fprintf_thread_summary(&trace, trace.output);