linux/tools/perf/builtin-trace.c

311 lines
7.9 KiB
C
Raw Normal View History

perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
#include "builtin.h"
#include "util/evlist.h"
#include "util/parse-options.h"
#include "util/thread_map.h"
#include "event-parse.h"
#include <libaudit.h>
#include <stdlib.h>
static struct syscall_fmt {
const char *name;
const char *alias;
perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
bool errmsg;
bool timeout;
} syscall_fmts[] = {
{ .name = "arch_prctl", .errmsg = true, .alias = "prctl", },
{ .name = "fstat", .errmsg = true, .alias = "newfstat", },
{ .name = "fstatat", .errmsg = true, .alias = "newfstatat", },
{ .name = "futex", .errmsg = true, },
{ .name = "poll", .errmsg = true, .timeout = true, },
{ .name = "ppoll", .errmsg = true, .timeout = true, },
{ .name = "read", .errmsg = true, },
{ .name = "recvfrom", .errmsg = true, },
{ .name = "select", .errmsg = true, .timeout = true, },
{ .name = "stat", .errmsg = true, .alias = "newstat", },
perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
};
static int syscall_fmt__cmp(const void *name, const void *fmtp)
{
const struct syscall_fmt *fmt = fmtp;
return strcmp(name, fmt->name);
}
static struct syscall_fmt *syscall_fmt__find(const char *name)
{
const int nmemb = ARRAY_SIZE(syscall_fmts);
return bsearch(name, syscall_fmts, nmemb, sizeof(struct syscall_fmt), syscall_fmt__cmp);
}
struct syscall {
struct event_format *tp_format;
const char *name;
struct syscall_fmt *fmt;
};
struct trace {
int audit_machine;
struct {
int max;
struct syscall *table;
} syscalls;
struct perf_record_opts opts;
};
static int trace__read_syscall_info(struct trace *trace, int id)
{
char tp_name[128];
struct syscall *sc;
if (id > trace->syscalls.max) {
struct syscall *nsyscalls = realloc(trace->syscalls.table, (id + 1) * sizeof(*sc));
if (nsyscalls == NULL)
return -1;
if (trace->syscalls.max != -1) {
memset(nsyscalls + trace->syscalls.max + 1, 0,
(id - trace->syscalls.max) * sizeof(*sc));
} else {
memset(nsyscalls, 0, (id + 1) * sizeof(*sc));
}
trace->syscalls.table = nsyscalls;
trace->syscalls.max = id;
}
sc = trace->syscalls.table + id;
sc->name = audit_syscall_to_name(id, trace->audit_machine);
if (sc->name == NULL)
return -1;
sc->fmt = syscall_fmt__find(sc->name);
perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
sc->tp_format = event_format__new("syscalls", tp_name);
if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
sc->tp_format = event_format__new("syscalls", tp_name);
}
perf trace: New tool Initially should look loosely like the venerable 'strace' tool, but using the infrastructure in the perf tools to allow tracing extra targets: [acme@sandy linux]$ perf trace --hell Error: unknown option `hell' usage: perf trace <PID> -p, --pid <pid> trace events on existing process id --tid <tid> trace events on existing thread id --all-cpus system-wide collection from all CPUs --cpu <cpu> list of cpus to monitor --no-inherit child tasks do not inherit counters --mmap-pages <n> number of mmap data pages --uid <user> user to profile [acme@sandy linux]$ Those should have the same semantics as when using with 'perf record'. It gets stuck sometimes, but hey, it works sometimes too! In time it should support perf.data based workloads, i.e. it should have a: -o filename Command line option that will produce a perf.data file that can then be used with 'perf trace' or any of the other perf tools (script, report, etc). It will also eventually have the set of functionalities described in the previous 'trace' prototype by Thomas Gleixner: "Announcing a new utility: 'trace'" http://lwn.net/Articles/415728/ Also planned is to have some of the features suggested in the comments of that LWN article. Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/n/tip-v9x3q9rv4caxtox7wtjpchq5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-09-26 23:05:56 +00:00
return sc->tp_format != NULL ? 0 : -1;
}
static size_t syscall__fprintf_args(struct syscall *sc, unsigned long *args, FILE *fp)
{
int i = 0;
size_t printed = 0;
if (sc->tp_format != NULL) {
struct format_field *field;
for (field = sc->tp_format->format.fields->next; field; field = field->next) {
printed += fprintf(fp, "%s%s: %ld", printed ? ", " : "",
field->name, args[i++]);
}
} else {
while (i < 6) {
printed += fprintf(fp, "%sarg%d: %ld", printed ? ", " : "", i, args[i]);
++i;
}
}
return printed;
}
static int trace__run(struct trace *trace)
{
struct perf_evlist *evlist = perf_evlist__new(NULL, NULL);
struct perf_evsel *evsel, *evsel_enter, *evsel_exit;
int err = -1, i, nr_events = 0, before;
if (evlist == NULL) {
printf("Not enough memory to run!\n");
goto out;
}
evsel_enter = perf_evsel__newtp("raw_syscalls", "sys_enter", 0);
if (evsel_enter == NULL) {
printf("Couldn't read the raw_syscalls:sys_enter tracepoint information!\n");
goto out_delete_evlist;
}
perf_evlist__add(evlist, evsel_enter);
evsel_exit = perf_evsel__newtp("raw_syscalls", "sys_exit", 1);
if (evsel_exit == NULL) {
printf("Couldn't read the raw_syscalls:sys_exit tracepoint information!\n");
goto out_delete_evlist;
}
perf_evlist__add(evlist, evsel_exit);
err = perf_evlist__create_maps(evlist, &trace->opts.target);
if (err < 0) {
printf("Problems parsing the target to trace, check your options!\n");
goto out_delete_evlist;
}
perf_evlist__config_attrs(evlist, &trace->opts);
err = perf_evlist__open(evlist);
if (err < 0) {
printf("Couldn't create the events: %s\n", strerror(errno));
goto out_delete_evlist;
}
err = perf_evlist__mmap(evlist, UINT_MAX, false);
if (err < 0) {
printf("Couldn't mmap the events: %s\n", strerror(errno));
goto out_delete_evlist;
}
perf_evlist__enable(evlist);
again:
before = nr_events;
for (i = 0; i < evlist->nr_mmaps; i++) {
union perf_event *event;
while ((event = perf_evlist__mmap_read(evlist, i)) != NULL) {
const u32 type = event->header.type;
struct syscall *sc;
struct perf_sample sample;
int id;
++nr_events;
switch (type) {
case PERF_RECORD_SAMPLE:
break;
case PERF_RECORD_LOST:
printf("LOST %" PRIu64 " events!\n", event->lost.lost);
continue;
default:
printf("Unexpected %s event, skipping...\n",
perf_event__name(type));
continue;
}
err = perf_evlist__parse_sample(evlist, event, &sample);
if (err) {
printf("Can't parse sample, err = %d, skipping...\n", err);
continue;
}
evsel = perf_evlist__id2evsel(evlist, sample.id);
if (evsel == NULL) {
printf("Unknown tp ID %" PRIu64 ", skipping...\n", sample.id);
continue;
}
id = perf_evsel__intval(evsel, &sample, "id");
if (id < 0) {
printf("Invalid syscall %d id, skipping...\n", id);
continue;
}
if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL) &&
trace__read_syscall_info(trace, id))
continue;
if ((id > trace->syscalls.max || trace->syscalls.table[id].name == NULL))
continue;
sc = &trace->syscalls.table[id];
if (evlist->threads->map[0] == -1 || evlist->threads->nr > 1)
printf("%d ", sample.tid);
if (evsel == evsel_enter) {
void *args = perf_evsel__rawptr(evsel, &sample, "args");
printf("%s(", sc->name);
syscall__fprintf_args(sc, args, stdout);
} else if (evsel == evsel_exit) {
int ret = perf_evsel__intval(evsel, &sample, "ret");
if (ret < 0 && sc->fmt && sc->fmt->errmsg) {
char bf[256];
const char *emsg = strerror_r(-ret, bf, sizeof(bf)),
*e = audit_errno_to_name(-ret);
printf(") = -1 %s %s", e, emsg);
} else if (ret == 0 && sc->fmt && sc->fmt->timeout)
printf(") = 0 Timeout");
else
printf(") = %d", ret);
putchar('\n');
}
}
}
if (nr_events == before)
poll(evlist->pollfd, evlist->nr_fds, -1);
goto again;
out_delete_evlist:
perf_evlist__delete(evlist);
out:
return err;
}
int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
{
const char * const trace_usage[] = {
"perf trace [<options>]",
NULL
};
struct trace trace = {
.audit_machine = audit_detect_machine(),
.syscalls = {
. max = -1,
},
.opts = {
.target = {
.uid = UINT_MAX,
.uses_mmap = true,
},
.user_freq = UINT_MAX,
.user_interval = ULLONG_MAX,
.no_delay = true,
.mmap_pages = 1024,
},
};
const struct option trace_options[] = {
OPT_STRING('p', "pid", &trace.opts.target.pid, "pid",
"trace events on existing process id"),
OPT_STRING(0, "tid", &trace.opts.target.tid, "tid",
"trace events on existing thread id"),
OPT_BOOLEAN(0, "all-cpus", &trace.opts.target.system_wide,
"system-wide collection from all CPUs"),
OPT_STRING(0, "cpu", &trace.opts.target.cpu_list, "cpu",
"list of cpus to monitor"),
OPT_BOOLEAN(0, "no-inherit", &trace.opts.no_inherit,
"child tasks do not inherit counters"),
OPT_UINTEGER(0, "mmap-pages", &trace.opts.mmap_pages,
"number of mmap data pages"),
OPT_STRING(0, "uid", &trace.opts.target.uid_str, "user",
"user to profile"),
OPT_END()
};
int err;
argc = parse_options(argc, argv, trace_options, trace_usage, 0);
if (argc)
usage_with_options(trace_usage, trace_options);
err = perf_target__parse_uid(&trace.opts.target);
if (err) {
char bf[BUFSIZ];
perf_target__strerror(&trace.opts.target, err, bf, sizeof(bf));
printf("%s", bf);
return err;
}
return trace__run(&trace);
}