perf evsel: Fix use of inherit

perf stat doesn't mmap and its perfectly fine for it to use task-bound
counters with inheritance.

So set the attr.inherit on the caller and leave the syscall itself to
validate it.

When the mmap fails perf_evlist__mmap will just emit a warning if this
is the failure reason.

Reported-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110414170121.GC3229@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 17d1dcb3c66771cb1d7603cff6a5ed981d44ca69..416538248a4bb04eaf2e28a1a2f58f062e6372fc 100644 (file)
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -163,6 +163,7 @@ static void config_attr(struct perf_evsel *evsel, struct perf_evlist *evlist)
        struct perf_event_attr *attr = &evsel->attr;
        int track = !evsel->idx; /* only the first counter needs these */
 
+       attr->inherit           = !no_inherit;
        attr->read_format       = PERF_FORMAT_TOTAL_TIME_ENABLED |
                                  PERF_FORMAT_TOTAL_TIME_RUNNING |
                                  PERF_FORMAT_ID;
@@ -251,6 +252,9 @@ static void open_counters(struct perf_evlist *evlist)
 {
        struct perf_evsel *pos;
 
+       if (evlist->cpus->map[0] < 0)
+               no_inherit = true;
+
        list_for_each_entry(pos, &evlist->entries, node) {
                struct perf_event_attr *attr = &pos->attr;
                /*
@@ -271,8 +275,7 @@ static void open_counters(struct perf_evlist *evlist)
 retry_sample_id:
                attr->sample_id_all = sample_id_all_avail ? 1 : 0;
 try_again:
-               if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group,
-                                    !no_inherit) < 0) {
+               if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group) < 0) {
                        int err = errno;
 
                        if (err == EPERM || err == EACCES) {

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index e2109f9b43eb3ed99a5239805087277b229739d3..03f0e45f1479c2fab31de2624d69d95458886157 100644 (file)
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -167,16 +167,17 @@ static int create_perf_stat_counter(struct perf_evsel *evsel)
                attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
                                    PERF_FORMAT_TOTAL_TIME_RUNNING;
 
+       attr->inherit = !no_inherit;
+
        if (system_wide)
-               return perf_evsel__open_per_cpu(evsel, evsel_list->cpus, false, false);
+               return perf_evsel__open_per_cpu(evsel, evsel_list->cpus, false);
 
-       attr->inherit = !no_inherit;
        if (target_pid == -1 && target_tid == -1) {
                attr->disabled = 1;
                attr->enable_on_exec = 1;
        }
 
-       return perf_evsel__open_per_thread(evsel, evsel_list->threads, false, false);
+       return perf_evsel__open_per_thread(evsel, evsel_list->threads, false);
 }
 
 /*

diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c
index 1b2106c58f660c83a858636df0862c458d127eb8..11e3c84583622bac20ed799cfab29e7066c6d6eb 100644 (file)
--- a/tools/perf/builtin-test.c
+++ b/tools/perf/builtin-test.c
@@ -290,7 +290,7 @@ static int test__open_syscall_event(void)
                goto out_thread_map_delete;
        }
 
-       if (perf_evsel__open_per_thread(evsel, threads, false, false) < 0) {
+       if (perf_evsel__open_per_thread(evsel, threads, false) < 0) {
                pr_debug("failed to open counter: %s, "
                         "tweak /proc/sys/kernel/perf_event_paranoid?\n",
                         strerror(errno));
@@ -303,7 +303,7 @@ static int test__open_syscall_event(void)
        }
 
        if (perf_evsel__read_on_cpu(evsel, 0, 0) < 0) {
-               pr_debug("perf_evsel__open_read_on_cpu\n");
+               pr_debug("perf_evsel__read_on_cpu\n");
                goto out_close_fd;
        }
 
@@ -365,7 +365,7 @@ static int test__open_syscall_event_on_all_cpus(void)
                goto out_thread_map_delete;
        }
 
-       if (perf_evsel__open(evsel, cpus, threads, false, false) < 0) {
+       if (perf_evsel__open(evsel, cpus, threads, false) < 0) {
                pr_debug("failed to open counter: %s, "
                         "tweak /proc/sys/kernel/perf_event_paranoid?\n",
                         strerror(errno));
@@ -418,7 +418,7 @@ static int test__open_syscall_event_on_all_cpus(void)
                        continue;
 
                if (perf_evsel__read_on_cpu(evsel, cpu, 0) < 0) {
-                       pr_debug("perf_evsel__open_read_on_cpu\n");
+                       pr_debug("perf_evsel__read_on_cpu\n");
                        err = -1;
                        break;
                }
@@ -529,7 +529,7 @@ static int test__basic_mmap(void)
 
                perf_evlist__add(evlist, evsels[i]);
 
-               if (perf_evsel__open(evsels[i], cpus, threads, false, false) < 0) {
+               if (perf_evsel__open(evsels[i], cpus, threads, false) < 0) {
                        pr_debug("failed to open counter: %s, "
                                 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
                                 strerror(errno));

diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index fc1273e976c58f4c25d6321abf578392ba1126ad..7e3d6e310bf839f09e5d6afed39c8c1530a13807 100644 (file)
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -845,9 +845,10 @@ static void start_counters(struct perf_evlist *evlist)
                }
 
                attr->mmap = 1;
+               attr->inherit = inherit;
 try_again:
                if (perf_evsel__open(counter, top.evlist->cpus,
-                                    top.evlist->threads, group, inherit) < 0) {
+                                    top.evlist->threads, group) < 0) {
                        int err = errno;
 
                        if (err == EPERM || err == EACCES) {

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index d852cefa20def5ff48665e7ef44bc842093f6f49..45da8d186b492d4f82512862aef867acd4649f09 100644 (file)
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -12,6 +12,7 @@
 #include "evlist.h"
 #include "evsel.h"
 #include "util.h"
+#include "debug.h"
 
 #include <sys/mman.h>
 
@@ -250,15 +251,19 @@ int perf_evlist__alloc_mmap(struct perf_evlist *evlist)
        return evlist->mmap != NULL ? 0 : -ENOMEM;
 }
 
-static int __perf_evlist__mmap(struct perf_evlist *evlist, int cpu, int prot,
-                              int mask, int fd)
+static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *evsel,
+                              int cpu, int prot, int mask, int fd)
 {
        evlist->mmap[cpu].prev = 0;
        evlist->mmap[cpu].mask = mask;
        evlist->mmap[cpu].base = mmap(NULL, evlist->mmap_len, prot,
                                      MAP_SHARED, fd, 0);
-       if (evlist->mmap[cpu].base == MAP_FAILED)
+       if (evlist->mmap[cpu].base == MAP_FAILED) {
+               if (evlist->cpus->map[cpu] == -1 && evsel->attr.inherit)
+                       ui__warning("Inherit is not allowed on per-task "
+                                   "events using mmap.\n");
                return -1;
+       }
 
        perf_evlist__add_pollfd(evlist, fd);
        return 0;
@@ -312,7 +317,8 @@ int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite)
                                        if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT,
                                                  FD(first_evsel, cpu, 0)) != 0)
                                                goto out_unmap;
-                               } else if (__perf_evlist__mmap(evlist, cpu, prot, mask, fd) < 0)
+                               } else if (__perf_evlist__mmap(evlist, evsel, cpu,
+                                                              prot, mask, fd) < 0)
                                        goto out_unmap;
 
                                if ((evsel->attr.read_format & PERF_FORMAT_ID) &&

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 662596afd7f1d49fccaeb8c95ee7c8c64748b8c9..d6fd59beb860739aea40e1b4052f6eda1a741996 100644 (file)
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -175,7 +175,7 @@ int __perf_evsel__read(struct perf_evsel *evsel,
 }
 
 static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-                             struct thread_map *threads, bool group, bool inherit)
+                             struct thread_map *threads, bool group)
 {
        int cpu, thread;
        unsigned long flags = 0;
@@ -192,19 +192,6 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 
        for (cpu = 0; cpu < cpus->nr; cpu++) {
                int group_fd = -1;
-               /*
-                * Don't allow mmap() of inherited per-task counters. This
-                * would create a performance issue due to all children writing
-                * to the same buffer.
-                *
-                * FIXME:
-                * Proper fix is not to pass 'inherit' to perf_evsel__open*,
-                * but a 'flags' parameter, with 'group' folded there as well,
-                * then introduce a PERF_O_{MMAP,GROUP,INHERIT} enum, and if
-                * O_MMAP is set, emit a warning if cpu < 0 and O_INHERIT is
-                * set. Lets go for the minimal fix first tho.
-                */
-               evsel->attr.inherit = (cpus->map[cpu] >= 0) && inherit;
 
                for (thread = 0; thread < threads->nr; thread++) {
 
@@ -253,7 +240,7 @@ static struct {
 };
 
 int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-                    struct thread_map *threads, bool group, bool inherit)
+                    struct thread_map *threads, bool group)
 {
        if (cpus == NULL) {
                /* Work around old compiler warnings about strict aliasing */
@@ -263,19 +250,19 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
        if (threads == NULL)
                threads = &empty_thread_map.map;
 
-       return __perf_evsel__open(evsel, cpus, threads, group, inherit);
+       return __perf_evsel__open(evsel, cpus, threads, group);
 }
 
 int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-                            struct cpu_map *cpus, bool group, bool inherit)
+                            struct cpu_map *cpus, bool group)
 {
-       return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group, inherit);
+       return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group);
 }
 
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
-                               struct thread_map *threads, bool group, bool inherit)
+                               struct thread_map *threads, bool group)
 {
-       return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, inherit);
+       return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group);
 }
 
 static int perf_event__parse_id_sample(const union perf_event *event, u64 type,

diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 6710ab538342fe1d5970c812423908ddec3ee5a7..f79bb2c09a6cbea510e31d260a1fabb778e13950 100644 (file)
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -81,11 +81,11 @@ void perf_evsel__free_id(struct perf_evsel *evsel);
 void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads);
 
 int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-                            struct cpu_map *cpus, bool group, bool inherit);
+                            struct cpu_map *cpus, bool group);
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
-                               struct thread_map *threads, bool group, bool inherit);
+                               struct thread_map *threads, bool group);
 int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-                    struct thread_map *threads, bool group, bool inherit);
+                    struct thread_map *threads, bool group);
 
 #define perf_evsel__match(evsel, t, c)         \
        (evsel->attr.type == PERF_TYPE_##t &&   \

diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c
index a9f2d7e1204d1655b14b99cafb06d08d5046d8a6..f5e38451fdc505e329ac77415e9e8aba5fbe7ebb 100644 (file)
--- a/tools/perf/util/python.c
+++ b/tools/perf/util/python.c
@@ -498,11 +498,11 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
        struct cpu_map *cpus = NULL;
        struct thread_map *threads = NULL;
        PyObject *pcpus = NULL, *pthreads = NULL;
-       int group = 0, overwrite = 0;
-       static char *kwlist[] = {"cpus", "threads", "group", "overwrite", NULL, NULL};
+       int group = 0, inherit = 0;
+       static char *kwlist[] = {"cpus", "threads", "group", "inherit", NULL, NULL};
 
        if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|OOii", kwlist,
-                                        &pcpus, &pthreads, &group, &overwrite))
+                                        &pcpus, &pthreads, &group, &inherit))
                return NULL;
 
        if (pthreads != NULL)
@@ -511,7 +511,8 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
        if (pcpus != NULL)
                cpus = ((struct pyrf_cpu_map *)pcpus)->cpus;
 
-       if (perf_evsel__open(evsel, cpus, threads, group, overwrite) < 0) {
+       evsel->attr.inherit = inherit;
+       if (perf_evsel__open(evsel, cpus, threads, group) < 0) {
                PyErr_SetFromErrno(PyExc_OSError);
                return NULL;
        }