[PATCH v9 0/8] perf tools: Support overwritable ring buffer

[PATCH v9 0/8] perf tools: Support overwritable ring buffer

[Wang Nan]

This patch set enables daemonized perf recording by utilizing
overwritable backward ring buffer. With this feature one can
put perf background, and dump ring buffer records by a SIGUSR2
when he/she find something unusual. For example, following
command record system calls, schedule events and samples on cpu cycles
continously:

 # perf record -g -e cycles -e raw_syscalls:*/call-graph=no/ \
                  -e sched:sched_switch/call-graph=no/ \
                  --switch-output --overwrite -a

Then by sending SIGUSR2 to perf when lagging is happen, we get multiple
perf.data output, each of them correspond a abnormal event, and the data
size is reasonable:

 # ls -l ./perf.data*
 -rw------- 1 root root 5122165 May 13 23:51 ./perf.data.2016051323511683
 -rw------- 1 root root 5135093 May 13 23:51 ./perf.data.2016051323512107
 -rw------- 1 root root 5135213 May 13 23:51 ./perf.data.2016051323512215
 -rw------- 1 root root 5135157 May 13 23:51 ./perf.data.2016051323512387

v1 -> v2: Totally redesign: drop the principle of 'channal', use
          auxiliary evlist instead. Fix missing documentation.

v2 -> v3: Rename perf_evlist__toggle_paused() to perf_evlist__pause/resume.

v3 -> v4: Update commit message to describe auxiliary evlist more clearly.

v4 -> v5: Reorder commits, ensure '--overwrite' works right after perf
          support the option.
          Add test cases for auxiliary evlist.
          Avoid bug if main evlist is empty.

v5 -> v6: Improve filter pollfd related code.

v6 -> v7: Rebase to newest perf/core.

v7 -> v8: Unmap mmaps from parent and children in
          perf_evlist__munmap_filtered(), hide more detail of aux evlist.
          Add --tail-synthesize, do synthesize at the end of perf.data.

v8 -> v9: Beautify code of test case, make patch set more granular,
          improve documentation.

Wang Nan (8):
  perf evlist: Introduce aux evlist
  perf tests: Add testcase for auxiliary evlist
  perf record: Extract perf_evlist__mmap_ex() processing to a new
    function
  perf record: Introduce rec->overwrite_evlist for overwritable events
  perf record: Toggle overwrite ring buffer for reading
  perf tools: Enable overwrite settings
  perf tools: Don't warn about out of order event if write_backward is
    used
  perf tools: Add --tail-synthesize option

 tools/perf/Documentation/perf-record.txt |  22 +++
 tools/perf/builtin-record.c              | 309 +++++++++++++++++++++++++++----
 tools/perf/perf.h                        |   2 +
 tools/perf/tests/backward-ring-buffer.c  |  89 +++++++--
 tools/perf/util/evlist.c                 |  49 +++--
 tools/perf/util/evlist.h                 |  20 ++
 tools/perf/util/evsel.c                  |  12 ++
 tools/perf/util/evsel.h                  |   2 +
 tools/perf/util/parse-events.c           |  20 +-
 tools/perf/util/parse-events.h           |   2 +
 tools/perf/util/parse-events.l           |   2 +
 tools/perf/util/session.c                |  22 ++-
 12 files changed, 479 insertions(+), 72 deletions(-)

-- 
1.8.3.4

[PATCH v9 1/8] perf evlist: Introduce aux evlist

[Wang Nan]

An auxiliary evlist is created by perf_evlist__new_aux() using an
existing evlist as its parent. An auxiliary evlist can have its own
'struct perf_mmap', but can't have any other data. User should use its
parent instead when accessing other data.

Auxiliary evlists are containers of 'struct perf_mmap'. It is introduced
to allow its parent evlist to map different events into separated mmaps.

Following commits create an auxiliary evlist for overwritable
events, because overwritable events need a read only and backwards ring
buffer, which is different from normal events.

To achieve this goal, this patch carefully changes 'evlist' to
'evlist->parent' in all functions in the path of 'perf_evlist__mmap_ex',
except 'evlist->mmap' related operations, to make sure all evlist
modifications (like pollfd and event id hash tables) goes to original
evlist.

A 'evlist->parent' pointer is added to 'struct perf_evlist' and points to
the evlist itself for normal evlists.

Children of one evlist are linked into it so one can find all children
from its parent.

To avoid potential complexity, forbid creating aux evlist from another
aux evlist.

Improve perf_evlist__munmap_filtered(), so when recording, if an event
is terminated, unmap mmaps, from parent and children.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/util/evlist.c | 49 +++++++++++++++++++++++++++++++++++++-----------
 tools/perf/util/evlist.h | 12 ++++++++++++
 2 files changed, 50 insertions(+), 11 deletions(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 1b918aa..b03cb3c 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -41,10 +41,12 @@ void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus,
 	for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i)
 		INIT_HLIST_HEAD(&evlist->heads[i]);
 	INIT_LIST_HEAD(&evlist->entries);
+	INIT_LIST_HEAD(&evlist->children);
 	perf_evlist__set_maps(evlist, cpus, threads);
 	fdarray__init(&evlist->pollfd, 64);
 	evlist->workload.pid = -1;
 	evlist->backward = false;
+	evlist->parent = evlist;
 }
 
 struct perf_evlist *perf_evlist__new(void)
@@ -487,13 +489,17 @@ static void perf_evlist__munmap_filtered(struct fdarray *fda, int fd,
 					 void *arg __maybe_unused)
 {
 	struct perf_evlist *evlist = container_of(fda, struct perf_evlist, pollfd);
+	struct perf_evlist *child;
 
 	perf_evlist__mmap_put(evlist, fda->priv[fd].idx);
+	list_for_each_entry(child, &evlist->children, list)
+		perf_evlist__mmap_put(child, fda->priv[fd].idx);
+
 }
 
 int perf_evlist__filter_pollfd(struct perf_evlist *evlist, short revents_and_mask)
 {
-	return fdarray__filter(&evlist->pollfd, revents_and_mask,
+	return fdarray__filter(&evlist->parent->pollfd, revents_and_mask,
 			       perf_evlist__munmap_filtered, NULL);
 }
 
@@ -1012,7 +1018,7 @@ static int perf_evlist__mmap_per_evsel(struct perf_evlist *evlist, int idx,
 	struct perf_evsel *evsel;
 	int revent;
 
-	evlist__for_each(evlist, evsel) {
+	evlist__for_each(evlist->parent, evsel) {
 		int fd;
 
 		if (evsel->overwrite != (evlist->overwrite && evlist->backward))
@@ -1044,16 +1050,16 @@ static int perf_evlist__mmap_per_evsel(struct perf_evlist *evlist, int idx,
 		 * Therefore don't add it for polling.
 		 */
 		if (!evsel->system_wide &&
-		    __perf_evlist__add_pollfd(evlist, fd, idx, revent) < 0) {
+		    __perf_evlist__add_pollfd(evlist->parent, fd, idx, revent) < 0) {
 			perf_evlist__mmap_put(evlist, idx);
 			return -1;
 		}
 
 		if (evsel->attr.read_format & PERF_FORMAT_ID) {
-			if (perf_evlist__id_add_fd(evlist, evsel, cpu, thread,
+			if (perf_evlist__id_add_fd(evlist->parent, evsel, cpu, thread,
 						   fd) < 0)
 				return -1;
-			perf_evlist__set_sid_idx(evlist, evsel, idx, cpu,
+			perf_evlist__set_sid_idx(evlist->parent, evsel, idx, cpu,
 						 thread);
 		}
 	}
@@ -1094,13 +1100,13 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist,
 					struct mmap_params *mp)
 {
 	int thread;
-	int nr_threads = thread_map__nr(evlist->threads);
+	int nr_threads = thread_map__nr(evlist->parent->threads);
 
 	pr_debug2("perf event ring buffer mmapped per thread\n");
 	for (thread = 0; thread < nr_threads; thread++) {
 		int output = -1;
 
-		auxtrace_mmap_params__set_idx(&mp->auxtrace_mp, evlist, thread,
+		auxtrace_mmap_params__set_idx(&mp->auxtrace_mp, evlist->parent, thread,
 					      false);
 
 		if (perf_evlist__mmap_per_evsel(evlist, thread, mp, 0, thread,
@@ -1239,8 +1245,8 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 			 bool auxtrace_overwrite)
 {
 	struct perf_evsel *evsel;
-	const struct cpu_map *cpus = evlist->cpus;
-	const struct thread_map *threads = evlist->threads;
+	const struct cpu_map *cpus = evlist->parent->cpus;
+	const struct thread_map *threads = evlist->parent->threads;
 	struct mmap_params mp = {
 		.prot = PROT_READ | (overwrite ? 0 : PROT_WRITE),
 	};
@@ -1248,7 +1254,7 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 	if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0)
 		return -ENOMEM;
 
-	if (evlist->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist) < 0)
+	if (evlist->parent->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist->parent) < 0)
 		return -ENOMEM;
 
 	evlist->overwrite = overwrite;
@@ -1259,7 +1265,7 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 	auxtrace_mmap_params__init(&mp.auxtrace_mp, evlist->mmap_len,
 				   auxtrace_pages, auxtrace_overwrite);
 
-	evlist__for_each(evlist, evsel) {
+	evlist__for_each(evlist->parent, evsel) {
 		if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
 		    evsel->sample_id == NULL &&
 		    perf_evsel__alloc_id(evsel, cpu_map__nr(cpus), threads->nr) < 0)
@@ -1916,3 +1922,24 @@ perf_evlist__find_evsel_by_str(struct perf_evlist *evlist,
 
 	return NULL;
 }
+
+struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *parent)
+{
+	struct perf_evlist *evlist;
+
+	if (perf_evlist__is_aux(parent)) {
+		pr_err("Internal error: create aux evlist from another aux evlist\n");
+		return NULL;
+	}
+
+	evlist = zalloc(sizeof(*evlist));
+	if (!evlist)
+		return NULL;
+
+	perf_evlist__init(evlist, parent->cpus, parent->threads);
+	evlist->parent = parent->parent;
+	INIT_LIST_HEAD(&evlist->list);
+	list_add(&evlist->list, &parent->children);
+
+	return evlist;
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 68cb136..5b50692 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -37,6 +37,10 @@ struct perf_mmap {
 
 struct perf_evlist {
 	struct list_head entries;
+	union {
+		struct list_head children;
+		struct list_head list;
+	};
 	struct hlist_head heads[PERF_EVLIST__HLIST_SIZE];
 	int		 nr_entries;
 	int		 nr_groups;
@@ -60,8 +64,14 @@ struct perf_evlist {
 	struct perf_evsel *selected;
 	struct events_stats stats;
 	struct perf_env	*env;
+	struct perf_evlist *parent;
 };
 
+static inline bool perf_evlist__is_aux(struct perf_evlist *evlist)
+{
+	return evlist->parent != evlist;
+}
+
 struct perf_evsel_str_handler {
 	const char *name;
 	void	   *handler;
@@ -70,6 +80,8 @@ struct perf_evsel_str_handler {
 struct perf_evlist *perf_evlist__new(void);
 struct perf_evlist *perf_evlist__new_default(void);
 struct perf_evlist *perf_evlist__new_dummy(void);
+struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *);
+
 void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus,
 		       struct thread_map *threads);
 void perf_evlist__exit(struct perf_evlist *evlist);
-- 
1.8.3.4

Re: [PATCH v9 1/8] perf evlist: Introduce aux evlist

[Nilay Vaish]

On 22 June 2016 at 04:08, Wang Nan <wangnan0@huawei.com> wrote:

> @@ -1916,3 +1922,24 @@ perf_evlist__find_evsel_by_str(struct perf_evlist *evlist,
>
>         return NULL;
>  }
> +
> +struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *parent)
> +{
> +       struct perf_evlist *evlist;
> +
> +       if (perf_evlist__is_aux(parent)) {
> +               pr_err("Internal error: create aux evlist from another aux evlist\n");
> +               return NULL;
> +       }
> +
> +       evlist = zalloc(sizeof(*evlist));
> +       if (!evlist)
> +               return NULL;
> +
> +       perf_evlist__init(evlist, parent->cpus, parent->threads);
> +       evlist->parent = parent->parent;

A very minor suggestion.  I think evlist->parent  should be set to
'parent' and not 'parent->parent'.  I agree the two values are equal,
but setting to parent->parent just does not seem right.

--
Nilay

Re: [PATCH v9 1/8] perf evlist: Introduce aux evlist

[Arnaldo Carvalho de Melo]

Em Wed, Jun 22, 2016 at 08:17:02AM -0500, Nilay Vaish escreveu:
> On 22 June 2016 at 04:08, Wang Nan <wangnan0@huawei.com> wrote:
> > +struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *parent)
> > +{
> > +       struct perf_evlist *evlist;
> > +
> > +       if (perf_evlist__is_aux(parent)) {
> > +               pr_err("Internal error: create aux evlist from another aux evlist\n");
> > +               return NULL;
> > +       }
> > +
> > +       evlist = zalloc(sizeof(*evlist));
> > +       if (!evlist)
> > +               return NULL;
> > +
> > +       perf_evlist__init(evlist, parent->cpus, parent->threads);
> > +       evlist->parent = parent->parent;
> 
> A very minor suggestion.  I think evlist->parent  should be set to
> 'parent' and not 'parent->parent'.  I agree the two values are equal,
> but setting to parent->parent just does not seem right.

I felt like that, thought I was missing something, which is always a bad
feeling when processing a patch... So, Wang, does that have some value
we are not seeing?

I thought about the possibility of adding an aux2 to an aux1 evlist and
that making aux2 have the same parent as aux1, but that is checked on
that pr_err() test...

- Arnaldo

Re: [PATCH v9 1/8] perf evlist: Introduce aux evlist

[Wangnan (F)]

On 2016/6/22 21:40, Arnaldo Carvalho de Melo wrote:
> Em Wed, Jun 22, 2016 at 08:17:02AM -0500, Nilay Vaish escreveu:
>> On 22 June 2016 at 04:08, Wang Nan <wangnan0@huawei.com> wrote:
>>> +struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *parent)
>>> +{
>>> +       struct perf_evlist *evlist;
>>> +
>>> +       if (perf_evlist__is_aux(parent)) {
>>> +               pr_err("Internal error: create aux evlist from another aux evlist\n");
>>> +               return NULL;
>>> +       }
>>> +
>>> +       evlist = zalloc(sizeof(*evlist));
>>> +       if (!evlist)
>>> +               return NULL;
>>> +
>>> +       perf_evlist__init(evlist, parent->cpus, parent->threads);
>>> +       evlist->parent = parent->parent;
>> A very minor suggestion.  I think evlist->parent  should be set to
>> 'parent' and not 'parent->parent'.  I agree the two values are equal,
>> but setting to parent->parent just does not seem right.
> I felt like that, thought I was missing something, which is always a bad
> feeling when processing a patch... So, Wang, does that have some value
> we are not seeing?
>
> I thought about the possibility of adding an aux2 to an aux1 evlist and
> that making aux2 have the same parent as aux1, but that is checked on
> that pr_err() test...

The pr_err() test is added yesterday to emphasize we don't accept
grandchildren. Before we have this test, setting parent to parent->parent
is a defensive programming. And I think it is clever :)

Thank you.

[PATCH v9.1] perf evlist: Introduce aux evlist

[Wang Nan]

An auxiliary evlist is created by perf_evlist__new_aux() using an
existing evlist as its parent. An auxiliary evlist can have its own
'struct perf_mmap', but can't have any other data. User should use its
parent instead when accessing other data.

Auxiliary evlists are containers of 'struct perf_mmap'. It is introduced
to allow its parent evlist to map different events into separated mmaps.

Following commits create an auxiliary evlist for overwritable
events, because overwritable events need a read only and backwards ring
buffer, which is different from normal events.

To achieve this goal, this patch carefully changes 'evlist' to
'evlist->parent' in all functions in the path of 'perf_evlist__mmap_ex',
except 'evlist->mmap' related operations, to make sure all evlist
modifications (like pollfd and event id hash tables) goes to original
evlist.

A 'evlist->parent' pointer is added to 'struct perf_evlist' and points to
the evlist itself for normal evlists.

Children of one evlist are linked into it so one can find all children
from its parent.

To avoid potential complexity, forbid creating aux evlist from another
aux evlist.

Improve perf_evlist__munmap_filtered(), so when recording, if an event
is terminated, unmap mmaps, from parent and children.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---

Minor code change: in perf_evlist__new_aux, set evlist->parent to parent,
not parent->parent.

Because we have already forbid creating an aux evlist from another aux
evlist, we don't need to use parent->parent trick to avoid grandchildren.

Thanks Nilay and Arnaldo.

---
 tools/perf/util/evlist.c | 49 +++++++++++++++++++++++++++++++++++++-----------
 tools/perf/util/evlist.h | 12 ++++++++++++
 2 files changed, 50 insertions(+), 11 deletions(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 1b918aa..5b924ae8 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -41,10 +41,12 @@ void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus,
 	for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i)
 		INIT_HLIST_HEAD(&evlist->heads[i]);
 	INIT_LIST_HEAD(&evlist->entries);
+	INIT_LIST_HEAD(&evlist->children);
 	perf_evlist__set_maps(evlist, cpus, threads);
 	fdarray__init(&evlist->pollfd, 64);
 	evlist->workload.pid = -1;
 	evlist->backward = false;
+	evlist->parent = evlist;
 }
 
 struct perf_evlist *perf_evlist__new(void)
@@ -487,13 +489,17 @@ static void perf_evlist__munmap_filtered(struct fdarray *fda, int fd,
 					 void *arg __maybe_unused)
 {
 	struct perf_evlist *evlist = container_of(fda, struct perf_evlist, pollfd);
+	struct perf_evlist *child;
 
 	perf_evlist__mmap_put(evlist, fda->priv[fd].idx);
+	list_for_each_entry(child, &evlist->children, list)
+		perf_evlist__mmap_put(child, fda->priv[fd].idx);
+
 }
 
 int perf_evlist__filter_pollfd(struct perf_evlist *evlist, short revents_and_mask)
 {
-	return fdarray__filter(&evlist->pollfd, revents_and_mask,
+	return fdarray__filter(&evlist->parent->pollfd, revents_and_mask,
 			       perf_evlist__munmap_filtered, NULL);
 }
 
@@ -1012,7 +1018,7 @@ static int perf_evlist__mmap_per_evsel(struct perf_evlist *evlist, int idx,
 	struct perf_evsel *evsel;
 	int revent;
 
-	evlist__for_each(evlist, evsel) {
+	evlist__for_each(evlist->parent, evsel) {
 		int fd;
 
 		if (evsel->overwrite != (evlist->overwrite && evlist->backward))
@@ -1044,16 +1050,16 @@ static int perf_evlist__mmap_per_evsel(struct perf_evlist *evlist, int idx,
 		 * Therefore don't add it for polling.
 		 */
 		if (!evsel->system_wide &&
-		    __perf_evlist__add_pollfd(evlist, fd, idx, revent) < 0) {
+		    __perf_evlist__add_pollfd(evlist->parent, fd, idx, revent) < 0) {
 			perf_evlist__mmap_put(evlist, idx);
 			return -1;
 		}
 
 		if (evsel->attr.read_format & PERF_FORMAT_ID) {
-			if (perf_evlist__id_add_fd(evlist, evsel, cpu, thread,
+			if (perf_evlist__id_add_fd(evlist->parent, evsel, cpu, thread,
 						   fd) < 0)
 				return -1;
-			perf_evlist__set_sid_idx(evlist, evsel, idx, cpu,
+			perf_evlist__set_sid_idx(evlist->parent, evsel, idx, cpu,
 						 thread);
 		}
 	}
@@ -1094,13 +1100,13 @@ static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist,
 					struct mmap_params *mp)
 {
 	int thread;
-	int nr_threads = thread_map__nr(evlist->threads);
+	int nr_threads = thread_map__nr(evlist->parent->threads);
 
 	pr_debug2("perf event ring buffer mmapped per thread\n");
 	for (thread = 0; thread < nr_threads; thread++) {
 		int output = -1;
 
-		auxtrace_mmap_params__set_idx(&mp->auxtrace_mp, evlist, thread,
+		auxtrace_mmap_params__set_idx(&mp->auxtrace_mp, evlist->parent, thread,
 					      false);
 
 		if (perf_evlist__mmap_per_evsel(evlist, thread, mp, 0, thread,
@@ -1239,8 +1245,8 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 			 bool auxtrace_overwrite)
 {
 	struct perf_evsel *evsel;
-	const struct cpu_map *cpus = evlist->cpus;
-	const struct thread_map *threads = evlist->threads;
+	const struct cpu_map *cpus = evlist->parent->cpus;
+	const struct thread_map *threads = evlist->parent->threads;
 	struct mmap_params mp = {
 		.prot = PROT_READ | (overwrite ? 0 : PROT_WRITE),
 	};
@@ -1248,7 +1254,7 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 	if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0)
 		return -ENOMEM;
 
-	if (evlist->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist) < 0)
+	if (evlist->parent->pollfd.entries == NULL && perf_evlist__alloc_pollfd(evlist->parent) < 0)
 		return -ENOMEM;
 
 	evlist->overwrite = overwrite;
@@ -1259,7 +1265,7 @@ int perf_evlist__mmap_ex(struct perf_evlist *evlist, unsigned int pages,
 	auxtrace_mmap_params__init(&mp.auxtrace_mp, evlist->mmap_len,
 				   auxtrace_pages, auxtrace_overwrite);
 
-	evlist__for_each(evlist, evsel) {
+	evlist__for_each(evlist->parent, evsel) {
 		if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
 		    evsel->sample_id == NULL &&
 		    perf_evsel__alloc_id(evsel, cpu_map__nr(cpus), threads->nr) < 0)
@@ -1916,3 +1922,24 @@ perf_evlist__find_evsel_by_str(struct perf_evlist *evlist,
 
 	return NULL;
 }
+
+struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *parent)
+{
+	struct perf_evlist *evlist;
+
+	if (perf_evlist__is_aux(parent)) {
+		pr_err("Internal error: create aux evlist from another aux evlist\n");
+		return NULL;
+	}
+
+	evlist = zalloc(sizeof(*evlist));
+	if (!evlist)
+		return NULL;
+
+	perf_evlist__init(evlist, parent->cpus, parent->threads);
+	evlist->parent = parent;
+	INIT_LIST_HEAD(&evlist->list);
+	list_add(&evlist->list, &parent->children);
+
+	return evlist;
+}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 68cb136..5b50692 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -37,6 +37,10 @@ struct perf_mmap {
 
 struct perf_evlist {
 	struct list_head entries;
+	union {
+		struct list_head children;
+		struct list_head list;
+	};
 	struct hlist_head heads[PERF_EVLIST__HLIST_SIZE];
 	int		 nr_entries;
 	int		 nr_groups;
@@ -60,8 +64,14 @@ struct perf_evlist {
 	struct perf_evsel *selected;
 	struct events_stats stats;
 	struct perf_env	*env;
+	struct perf_evlist *parent;
 };
 
+static inline bool perf_evlist__is_aux(struct perf_evlist *evlist)
+{
+	return evlist->parent != evlist;
+}
+
 struct perf_evsel_str_handler {
 	const char *name;
 	void	   *handler;
@@ -70,6 +80,8 @@ struct perf_evsel_str_handler {
 struct perf_evlist *perf_evlist__new(void);
 struct perf_evlist *perf_evlist__new_default(void);
 struct perf_evlist *perf_evlist__new_dummy(void);
+struct perf_evlist *perf_evlist__new_aux(struct perf_evlist *);
+
 void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus,
 		       struct thread_map *threads);
 void perf_evlist__exit(struct perf_evlist *evlist);
-- 
1.8.3.4

[PATCH v9 2/8] perf tests: Add testcase for auxiliary evlist

[Wang Nan]

Improve test backward-ring-buffer, trace both enter and exit event of
prctl() syscall, utilize auxiliary evlist to mmap enter and exit event
into separated mmaps.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
---
 tools/perf/tests/backward-ring-buffer.c | 90 ++++++++++++++++++++++++++-------
 tools/perf/util/evlist.h                |  8 +++
 2 files changed, 81 insertions(+), 17 deletions(-)

diff --git a/tools/perf/tests/backward-ring-buffer.c b/tools/perf/tests/backward-ring-buffer.c
index d9ba991..0b100b8 100644
--- a/tools/perf/tests/backward-ring-buffer.c
+++ b/tools/perf/tests/backward-ring-buffer.c
@@ -26,12 +26,21 @@ static void testcase(void)
 static int count_samples(struct perf_evlist *evlist, int *sample_count,
 			 int *comm_count)
 {
-	int i;
+	int i, dummy;
+
+	if (!comm_count)
+		comm_count = &dummy;
+	if (!sample_count)
+		sample_count = &dummy;
 
 	for (i = 0; i < evlist->nr_mmaps; i++) {
 		union perf_event *event;
 
-		perf_evlist__mmap_read_catchup(evlist, i);
+		/*
+		 * Before calling count_samples(), ring buffers in backward
+		 * evlist should have catched up with newest record
+		 * using perf_evlist__mmap_read_catchup_all().
+		 */
 		while ((event = perf_evlist__mmap_read_backward(evlist, i)) != NULL) {
 			const u32 type = event->header.type;
 
@@ -51,34 +60,54 @@ static int count_samples(struct perf_evlist *evlist, int *sample_count,
 	return TEST_OK;
 }
 
-static int do_test(struct perf_evlist *evlist, int mmap_pages,
-		   int *sample_count, int *comm_count)
+static int do_test(struct perf_evlist *evlist,
+		   struct perf_evlist *aux_evlist,
+		   int mmap_pages,
+		   int *enter_sample_count,
+		   int *exit_sample_count,
+		   int *comm_count)
 {
 	int err;
 	char sbuf[STRERR_BUFSIZE];
 
-	err = perf_evlist__mmap(evlist, mmap_pages, true);
+	err = perf_evlist__mmap(evlist, mmap_pages, false);
 	if (err < 0) {
 		pr_debug("perf_evlist__mmap: %s\n",
 			 strerror_r(errno, sbuf, sizeof(sbuf)));
 		return TEST_FAIL;
 	}
 
+	err = perf_evlist__mmap(aux_evlist, mmap_pages, true);
+	if (err < 0) {
+		pr_debug("perf_evlist__mmap for aux_evlist: %s\n",
+			 strerror_r(errno, sbuf, sizeof(sbuf)));
+		return TEST_FAIL;
+	}
+
 	perf_evlist__enable(evlist);
 	testcase();
 	perf_evlist__disable(evlist);
 
-	err = count_samples(evlist, sample_count, comm_count);
+	perf_evlist__mmap_read_catchup_all(aux_evlist);
+	err = count_samples(aux_evlist, exit_sample_count, comm_count);
+	if (err)
+		goto errout;
+	err = count_samples(evlist, enter_sample_count, NULL);
+	if (err)
+		goto errout;
+errout:
 	perf_evlist__munmap(evlist);
+	perf_evlist__munmap(aux_evlist);
 	return err;
 }
 
 
 int test__backward_ring_buffer(int subtest __maybe_unused)
 {
-	int ret = TEST_SKIP, err, sample_count = 0, comm_count = 0;
+	int ret = TEST_SKIP, err;
+	int enter_sample_count = 0, exit_sample_count = 0, comm_count = 0;
 	char pid[16], sbuf[STRERR_BUFSIZE];
-	struct perf_evlist *evlist;
+	struct perf_evlist *evlist, *aux_evlist = NULL;
 	struct perf_evsel *evsel __maybe_unused;
 	struct parse_events_error parse_error;
 	struct record_opts opts = {
@@ -115,11 +144,22 @@ int test__backward_ring_buffer(int subtest __maybe_unused)
 		goto out_delete_evlist;
 	}
 
-	perf_evlist__config(evlist, &opts, NULL);
+	/*
+	 * Set backward bit, ring buffer should be writing from end. Record
+	 * it in aux evlist
+	 */
+	perf_evlist__last(evlist)->overwrite = true;
+	perf_evlist__last(evlist)->attr.write_backward = 1;
 
-	/* Set backward bit, ring buffer should be writing from end */
-	evlist__for_each(evlist, evsel)
-		evsel->attr.write_backward = 1;
+	err = parse_events(evlist, "syscalls:sys_exit_prctl", &parse_error);
+	if (err) {
+		pr_debug("Failed to parse tracepoint event, try use root\n");
+		ret = TEST_SKIP;
+		goto out_delete_evlist;
+	}
+	/* Don't set backward bit for exit event. Record it in main evlist */
+
+	perf_evlist__config(evlist, &opts, NULL);
 
 	err = perf_evlist__open(evlist);
 	if (err < 0) {
@@ -128,24 +168,40 @@ int test__backward_ring_buffer(int subtest __maybe_unused)
 		goto out_delete_evlist;
 	}
 
+	aux_evlist = perf_evlist__new_aux(evlist);
+	if (!aux_evlist) {
+		pr_debug("perf_evlist__new_aux failed\n");
+		goto out_delete_evlist;
+	}
+	aux_evlist->backward = true;
+
 	ret = TEST_FAIL;
-	err = do_test(evlist, opts.mmap_pages, &sample_count,
+	err = do_test(evlist, aux_evlist, opts.mmap_pages,
+		      &enter_sample_count, &exit_sample_count,
 		      &comm_count);
 	if (err != TEST_OK)
 		goto out_delete_evlist;
 
-	if ((sample_count != NR_ITERS) || (comm_count != NR_ITERS)) {
-		pr_err("Unexpected counter: sample_count=%d, comm_count=%d\n",
-		       sample_count, comm_count);
+	if (enter_sample_count != exit_sample_count) {
+		pr_err("Unexpected counter: enter_sample_count=%d, exit_sample_count=%d\n",
+		       enter_sample_count, exit_sample_count);
+		goto out_delete_evlist;
+	}
+
+	if ((exit_sample_count != NR_ITERS) || (comm_count != NR_ITERS)) {
+		pr_err("Unexpected counter: exit_sample_count=%d, comm_count=%d\n",
+		       exit_sample_count, comm_count);
 		goto out_delete_evlist;
 	}
 
-	err = do_test(evlist, 1, &sample_count, &comm_count);
+	err = do_test(evlist, aux_evlist, 1, NULL, NULL, NULL);
 	if (err != TEST_OK)
 		goto out_delete_evlist;
 
 	ret = TEST_OK;
 out_delete_evlist:
+	if (aux_evlist)
+		perf_evlist__delete(aux_evlist);
 	perf_evlist__delete(evlist);
 	return ret;
 }
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 5b50692..51d333b 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -149,6 +149,14 @@ union perf_event *perf_evlist__mmap_read_backward(struct perf_evlist *evlist,
 						  int idx);
 void perf_evlist__mmap_read_catchup(struct perf_evlist *evlist, int idx);
 
+static inline void perf_evlist__mmap_read_catchup_all(struct perf_evlist *evlist)
+{
+	int i;
+
+	for (i = 0; i < evlist->nr_mmaps; i++)
+		perf_evlist__mmap_read_catchup(evlist, i);
+}
+
 void perf_evlist__mmap_consume(struct perf_evlist *evlist, int idx);
 
 int perf_evlist__pause(struct perf_evlist *evlist);
-- 
1.8.3.4

[PATCH v9 3/8] perf record: Extract perf_evlist__mmap_ex() processing to a new function

[Wang Nan]

Create record__mmap() and record__mmap_evlist() to wrap
perf_evlist__mmap_ex() and its error processing.

Following commits will introdule multiple evlists for 'perf record'.
record__mmap() and record__mmap_evlist() will be improvemented to
create mmap for each evlist.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
---
 tools/perf/builtin-record.c | 56 +++++++++++++++++++++++++++++----------------
 1 file changed, 36 insertions(+), 20 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index b1304eb..7998ec3 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -341,6 +341,40 @@ int auxtrace_record__snapshot_start(struct auxtrace_record *itr __maybe_unused)
 
 #endif
 
+static int record__mmap_evlist(struct record *rec,
+			       struct perf_evlist *evlist)
+{
+	struct record_opts *opts = &rec->opts;
+	char msg[512];
+
+	if (perf_evlist__mmap_ex(evlist, opts->mmap_pages, false,
+				 opts->auxtrace_mmap_pages,
+				 opts->auxtrace_snapshot_mode) < 0) {
+		if (errno == EPERM) {
+			pr_err("Permission error mapping pages.\n"
+			       "Consider increasing "
+			       "/proc/sys/kernel/perf_event_mlock_kb,\n"
+			       "or try again with a smaller value of -m/--mmap_pages.\n"
+			       "(current value: %u,%u)\n",
+			       opts->mmap_pages, opts->auxtrace_mmap_pages);
+			return -errno;
+		} else {
+			pr_err("failed to mmap with %d (%s)\n", errno,
+				strerror_r(errno, msg, sizeof(msg)));
+			if (errno)
+				return -errno;
+			else
+				return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+static int record__mmap(struct record *rec)
+{
+	return record__mmap_evlist(rec, rec->evlist);
+}
+
 static int record__open(struct record *rec)
 {
 	char msg[512];
@@ -377,27 +411,9 @@ try_again:
 		goto out;
 	}
 
-	if (perf_evlist__mmap_ex(evlist, opts->mmap_pages, false,
-				 opts->auxtrace_mmap_pages,
-				 opts->auxtrace_snapshot_mode) < 0) {
-		if (errno == EPERM) {
-			pr_err("Permission error mapping pages.\n"
-			       "Consider increasing "
-			       "/proc/sys/kernel/perf_event_mlock_kb,\n"
-			       "or try again with a smaller value of -m/--mmap_pages.\n"
-			       "(current value: %u,%u)\n",
-			       opts->mmap_pages, opts->auxtrace_mmap_pages);
-			rc = -errno;
-		} else {
-			pr_err("failed to mmap with %d (%s)\n", errno,
-				strerror_r(errno, msg, sizeof(msg)));
-			if (errno)
-				rc = -errno;
-			else
-				rc = -EINVAL;
-		}
+	rc = record__mmap(rec);
+	if (rc)
 		goto out;
-	}
 
 	session->evlist = evlist;
 	perf_session__set_id_hdr_size(session);
-- 
1.8.3.4

[PATCH v9 4/8] perf record: Introduce rec->overwrite_evlist for overwritable events

[Wang Nan]

Create an auxiliary evlist for overwritable events.

Before mmap, build this evlist and set 'overwrite' and 'backward'
attribute. Since perf_evlist__mmap_ex() only maps events when
evsel->overwrite matches evlist's corresponding attributes, with
these two evlists an event goes to either rec->evlist or
rec->overwrite_evlist.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/builtin-record.c | 81 ++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 76 insertions(+), 5 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 7998ec3..4a77bf3 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -50,6 +50,7 @@ struct record {
 	struct perf_data_file	file;
 	struct auxtrace_record	*itr;
 	struct perf_evlist	*evlist;
+	struct perf_evlist	*overwrite_evlist;
 	struct perf_session	*session;
 	const char		*progname;
 	int			realtime_prio;
@@ -341,13 +342,41 @@ int auxtrace_record__snapshot_start(struct auxtrace_record *itr __maybe_unused)
 
 #endif
 
+static int record__create_overwrite_evlist(struct record *rec)
+{
+	struct perf_evlist *evlist = rec->evlist;
+	struct perf_evsel *pos;
+
+	evlist__for_each(evlist, pos) {
+		if (!pos->overwrite)
+			continue;
+
+		if (!rec->overwrite_evlist) {
+			rec->overwrite_evlist = perf_evlist__new_aux(evlist);
+			if (rec->overwrite_evlist) {
+				rec->overwrite_evlist->backward = true;
+				rec->overwrite_evlist->overwrite = true;
+				return 0;
+			} else
+				return -ENOMEM;
+		}
+	}
+	return 0;
+}
+
 static int record__mmap_evlist(struct record *rec,
-			       struct perf_evlist *evlist)
+			       struct perf_evlist *evlist,
+			       bool overwrite)
 {
 	struct record_opts *opts = &rec->opts;
 	char msg[512];
 
-	if (perf_evlist__mmap_ex(evlist, opts->mmap_pages, false,
+	/*
+	 * Don't use evlist->overwrite because it is logically an
+	 * internal attribute and is set by perf_evlist__mmap_ex().
+	 * Avoid circular dependency.
+	 */
+	if (perf_evlist__mmap_ex(evlist, opts->mmap_pages, overwrite,
 				 opts->auxtrace_mmap_pages,
 				 opts->auxtrace_snapshot_mode) < 0) {
 		if (errno == EPERM) {
@@ -372,7 +401,23 @@ static int record__mmap_evlist(struct record *rec,
 
 static int record__mmap(struct record *rec)
 {
-	return record__mmap_evlist(rec, rec->evlist);
+	int err;
+
+	err = record__create_overwrite_evlist(rec);
+	if (err)
+		return err;
+
+	err = record__mmap_evlist(rec, rec->evlist, false);
+	if (err)
+		return err;
+
+	if (!rec->overwrite_evlist)
+		return 0;
+
+	err = record__mmap_evlist(rec, rec->overwrite_evlist, true);
+	if (err)
+		return err;
+	return 0;
 }
 
 static int record__open(struct record *rec)
@@ -387,6 +432,13 @@ static int record__open(struct record *rec)
 	perf_evlist__config(evlist, opts, &callchain_param);
 
 	evlist__for_each(evlist, pos) {
+		if (pos->overwrite) {
+			if (!pos->attr.write_backward) {
+				ui__warning("Unable to read from overwrite ring buffer\n\n");
+				rc = -ENOSYS;
+				goto out;
+			}
+		}
 try_again:
 		if (perf_evsel__open(pos, pos->cpus, pos->threads) < 0) {
 			if (perf_evsel__fallback(pos, errno, msg, sizeof(msg))) {
@@ -671,10 +723,26 @@ perf_event__synth_time_conv(const struct perf_event_mmap_page *pc __maybe_unused
 	return 0;
 }
 
+static const struct perf_event_mmap_page *
+perf_evlist__pick_pc(struct perf_evlist *evlist)
+{
+	if (evlist && evlist->mmap && evlist->mmap[0].base)
+		return evlist->mmap[0].base;
+	return NULL;
+}
+
 static const struct perf_event_mmap_page *record__pick_pc(struct record *rec)
 {
-	if (rec->evlist && rec->evlist->mmap && rec->evlist->mmap[0].base)
-		return rec->evlist->mmap[0].base;
+	const struct perf_event_mmap_page *pc;
+
+	/* Change it to a loop if a new aux evlist is added */
+	pc = perf_evlist__pick_pc(rec->evlist);
+	if (pc)
+		return pc;
+	pc = perf_evlist__pick_pc(rec->overwrite_evlist);
+	if (pc)
+		return pc;
+
 	return NULL;
 }
 
@@ -1285,6 +1353,7 @@ static struct record record = {
 		.mmap2		= perf_event__process_mmap2,
 		.ordered_events	= true,
 	},
+	.overwrite_evlist = NULL,
 };
 
 const char record_callchain_help[] = CALLCHAIN_RECORD_HELP
@@ -1588,6 +1657,8 @@ int cmd_record(int argc, const char **argv, const char *prefix __maybe_unused)
 	err = __cmd_record(&record, argc, argv);
 out_symbol_exit:
 	perf_evlist__delete(rec->evlist);
+	if (rec->overwrite_evlist)
+		perf_evlist__delete(rec->overwrite_evlist);
 	symbol__exit();
 	auxtrace_record__free(rec->itr);
 	return err;
-- 
1.8.3.4

[PATCH v9 5/8] perf record: Toggle overwrite ring buffer for reading

[Wang Nan]

overwrite_evt_state is introduced to reflect the state of overwritable
ring buffers. It is a state machine with 3 states:

 RUNNING --(1)--> DATA_PENDING --(2)--> EMPTY
    ^                  ^                 |
    |                  |___(disallow)___/|
    |                                    |
     \_________________(3)______________/

 RUNNING      : Overwritable ring buffers are recording
 DATA_PENDING : We are required to collect overwritable ring buffers
 EMPTY        : We have collected data from those ring buffers.

 (1): Pause ring buffers for reading
 (2): Read from ring buffers
 (3): Resume ring buffers for recording

We can't avoid this complexity. Since we deliberately drop records from
overwritable ring buffer, there's no mean for us to check remaining from
ring buffer itself (by checking head and old pointers). Therefore, we
need DATA_PENDING and EMPTY state to help us recording what we have done
to the ring buffer.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: He Kuang <hekuang@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/builtin-record.c | 146 +++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 136 insertions(+), 10 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 4a77bf3..019151a 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -42,6 +42,28 @@
 #include <sys/mman.h>
 #include <asm/bug.h>
 
+/*
+ * State machine of overwrite_evt_state:
+ *
+ * RUNNING --(1)--> DATA_PENDING --(2)--> EMPTY
+ *    ^                  ^                 |
+ *    |                  |___(disallow)___/|
+ *    |                                    |
+ *     \_________________(3)______________/
+ *
+ * RUNNING      : Overwritable ring buffers are recording
+ * DATA_PENDING : We are required to collect overwritable ring buffers
+ * EMPTY        : We have collected data from those ring buffers.
+ *
+ * (1): Pause ring buffers for reading
+ * (2): Read from ring buffers
+ * (3): Resume ring buffers for recording
+ */
+enum overwrite_evt_state {
+	OVERWRITE_EVT_RUNNING,
+	OVERWRITE_EVT_DATA_PENDING,
+	OVERWRITE_EVT_EMPTY,
+};
 
 struct record {
 	struct perf_tool	tool;
@@ -61,6 +83,7 @@ struct record {
 	bool			buildid_all;
 	bool			timestamp_filename;
 	bool			switch_output;
+	enum overwrite_evt_state overwrite_evt_state;
 	unsigned long long	samples;
 };
 
@@ -132,9 +155,9 @@ rb_find_range(struct perf_evlist *evlist,
 	return backward_rb_find_range(data, mask, head, start, end);
 }
 
-static int record__mmap_read(struct record *rec, int idx)
+static int record__mmap_read(struct record *rec, struct perf_evlist *evlist, int idx)
 {
-	struct perf_mmap *md = &rec->evlist->mmap[idx];
+	struct perf_mmap *md = &evlist->mmap[idx];
 	u64 head = perf_mmap__read_head(md);
 	u64 old = md->prev;
 	u64 end = head, start = old;
@@ -143,7 +166,7 @@ static int record__mmap_read(struct record *rec, int idx)
 	void *buf;
 	int rc = 0;
 
-	if (rb_find_range(rec->evlist, data, md->mask, head,
+	if (rb_find_range(evlist, data, md->mask, head,
 			  old, &start, &end))
 		return -1;
 
@@ -157,7 +180,7 @@ static int record__mmap_read(struct record *rec, int idx)
 		WARN_ONCE(1, "failed to keep up with mmap data. (warn only once)\n");
 
 		md->prev = head;
-		perf_evlist__mmap_consume(rec->evlist, idx);
+		perf_evlist__mmap_consume(evlist, idx);
 		return 0;
 	}
 
@@ -182,7 +205,7 @@ static int record__mmap_read(struct record *rec, int idx)
 	}
 
 	md->prev = head;
-	perf_evlist__mmap_consume(rec->evlist, idx);
+	perf_evlist__mmap_consume(evlist, idx);
 out:
 	return rc;
 }
@@ -469,6 +492,7 @@ try_again:
 
 	session->evlist = evlist;
 	perf_session__set_id_hdr_size(session);
+	rec->overwrite_evt_state = OVERWRITE_EVT_RUNNING;
 out:
 	return rc;
 }
@@ -549,17 +573,72 @@ static struct perf_event_header finished_round_event = {
 	.type = PERF_RECORD_FINISHED_ROUND,
 };
 
-static int record__mmap_read_all(struct record *rec)
+static void
+record__toggle_overwrite_evsels(struct record *rec,
+				enum overwrite_evt_state state)
+{
+	struct perf_evlist *evlist = rec->overwrite_evlist;
+	enum overwrite_evt_state old_state = rec->overwrite_evt_state;
+	enum action {
+		NONE,
+		PAUSE,
+		RESUME,
+	} action = NONE;
+
+	switch (old_state) {
+	case OVERWRITE_EVT_RUNNING:
+		if (state != OVERWRITE_EVT_RUNNING)
+			action = PAUSE;
+		break;
+	case OVERWRITE_EVT_DATA_PENDING:
+		if (state == OVERWRITE_EVT_RUNNING)
+			action = RESUME;
+		break;
+	case OVERWRITE_EVT_EMPTY:
+		if (state == OVERWRITE_EVT_RUNNING)
+			action = RESUME;
+		if (state == OVERWRITE_EVT_DATA_PENDING)
+			state = OVERWRITE_EVT_EMPTY;
+		break;
+	default:
+		WARN_ONCE(1, "Shouldn't get there\n");
+	}
+
+	rec->overwrite_evt_state = state;
+
+	if (action == NONE)
+		return;
+
+	if (!evlist)
+		return;
+
+	switch (action) {
+	case PAUSE:
+		perf_evlist__pause(evlist);
+		break;
+	case RESUME:
+		perf_evlist__resume(evlist);
+		break;
+	case NONE:
+	default:
+		break;
+	}
+}
+
+static int __record__mmap_read_evlist(struct record *rec, struct perf_evlist *evlist)
 {
 	u64 bytes_written = rec->bytes_written;
 	int i;
 	int rc = 0;
 
-	for (i = 0; i < rec->evlist->nr_mmaps; i++) {
-		struct auxtrace_mmap *mm = &rec->evlist->mmap[i].auxtrace_mmap;
+	if (!evlist)
+		return 0;
+
+	for (i = 0; i < evlist->nr_mmaps; i++) {
+		struct auxtrace_mmap *mm = &evlist->mmap[i].auxtrace_mmap;
 
-		if (rec->evlist->mmap[i].base) {
-			if (record__mmap_read(rec, i) != 0) {
+		if (evlist->mmap[i].base) {
+			if (record__mmap_read(rec, evlist, i) != 0) {
 				rc = -1;
 				goto out;
 			}
@@ -583,6 +662,23 @@ out:
 	return rc;
 }
 
+static int record__mmap_read_all(struct record *rec)
+{
+	int err;
+
+	err = __record__mmap_read_evlist(rec, rec->evlist);
+	if (err)
+		return err;
+
+	if (rec->overwrite_evt_state == OVERWRITE_EVT_DATA_PENDING) {
+		err = __record__mmap_read_evlist(rec, rec->overwrite_evlist);
+		if (err)
+			return err;
+		record__toggle_overwrite_evsels(rec, OVERWRITE_EVT_EMPTY);
+	}
+	return 0;
+}
+
 static void record__init_features(struct record *rec)
 {
 	struct perf_session *session = rec->session;
@@ -979,6 +1075,17 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 	for (;;) {
 		unsigned long long hits = rec->samples;
 
+		/*
+		 * rec->overwrite_evt_state is possible to be
+		 * OVERWRITE_EVT_EMPTY here: when done == true and
+		 * hits != rec->samples after previous reading.
+		 *
+		 * record__toggle_overwrite_evsels ensure we never
+		 * convert OVERWRITE_EVT_EMPTY to OVERWRITE_EVT_DATA_PENDING.
+		 */
+		if (trigger_is_hit(&switch_output_trigger) || done || draining)
+			record__toggle_overwrite_evsels(rec, OVERWRITE_EVT_DATA_PENDING);
+
 		if (record__mmap_read_all(rec) < 0) {
 			trigger_error(&auxtrace_snapshot_trigger);
 			trigger_error(&switch_output_trigger);
@@ -998,8 +1105,27 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 		}
 
 		if (trigger_is_hit(&switch_output_trigger)) {
+			/*
+			 * If switch_output_trigger is hit, the data in
+			 * overwritable ring buffer should have been collected,
+			 * so overwrite_evt_state should be set to
+			 * OVERWRITE_EVT_EMPTY.
+			 *
+			 * If SIGUSR2 raise after or during record__mmap_read_all(),
+			 * record__mmap_read_all() didn't collect data from
+			 * overwritable ring buffer. Read again.
+			 */
+			if (rec->overwrite_evt_state == OVERWRITE_EVT_RUNNING)
+				continue;
 			trigger_ready(&switch_output_trigger);
 
+			/*
+			 * Reenable events in overwrite ring buffer after
+			 * record__mmap_read_all(): we should have collected
+			 * data from it.
+			 */
+			record__toggle_overwrite_evsels(rec, OVERWRITE_EVT_RUNNING);
+
 			if (!quiet)
 				fprintf(stderr, "[ perf record: dump data: Woken up %ld times ]\n",
 					waking);
-- 
1.8.3.4

Re: [PATCH v9 5/8] perf record: Toggle overwrite ring buffer for reading

[Nilay Vaish]

On 22 June 2016 at 04:08, Wang Nan <wangnan0@huawei.com> wrote:
> @@ -549,17 +573,72 @@ static struct perf_event_header finished_round_event = {
>         .type = PERF_RECORD_FINISHED_ROUND,
>  };
>
> -static int record__mmap_read_all(struct record *rec)
> +static void
> +record__toggle_overwrite_evsels(struct record *rec,
> +                               enum overwrite_evt_state state)
> +{
> +       struct perf_evlist *evlist = rec->overwrite_evlist;
> +       enum overwrite_evt_state old_state = rec->overwrite_evt_state;
> +       enum action {
> +               NONE,
> +               PAUSE,
> +               RESUME,
> +       } action = NONE;
> +
> +       switch (old_state) {
> +       case OVERWRITE_EVT_RUNNING:
> +               if (state != OVERWRITE_EVT_RUNNING)
> +                       action = PAUSE;
> +               break;
> +       case OVERWRITE_EVT_DATA_PENDING:
> +               if (state == OVERWRITE_EVT_RUNNING)
> +                       action = RESUME;
> +               break;
> +       case OVERWRITE_EVT_EMPTY:
> +               if (state == OVERWRITE_EVT_RUNNING)
> +                       action = RESUME;
> +               if (state == OVERWRITE_EVT_DATA_PENDING)
> +                       state = OVERWRITE_EVT_EMPTY;

else if (state == OVERWRITE_EVT_DATA_PENDING)

> +               break;
> +       default:
> +               WARN_ONCE(1, "Shouldn't get there\n");
> +       }
> +
> +       rec->overwrite_evt_state = state;
> +
> +       if (action == NONE)
> +               return;

I think the above two lines are not required.  The switch below should
be enough.

> +
> +       if (!evlist)
> +               return;
> +
> +       switch (action) {
> +       case PAUSE:
> +               perf_evlist__pause(evlist);
> +               break;
> +       case RESUME:
> +               perf_evlist__resume(evlist);
> +               break;
> +       case NONE:
> +       default:
> +               break;
> +       }
> +}
> +

--
Nilay

Re: [PATCH v9 5/8] perf record: Toggle overwrite ring buffer for reading

[Wangnan (F)]

On 2016/6/22 22:33, Nilay Vaish wrote:
> On 22 June 2016 at 04:08, Wang Nan <wangnan0@huawei.com> wrote:
>> @@ -549,17 +573,72 @@ static struct perf_event_header finished_round_event = {
>>          .type = PERF_RECORD_FINISHED_ROUND,
>>   };
>>
>> -static int record__mmap_read_all(struct record *rec)
>> +static void
>> +record__toggle_overwrite_evsels(struct record *rec,
>> +                               enum overwrite_evt_state state)
>> +{
>> +       struct perf_evlist *evlist = rec->overwrite_evlist;
>> +       enum overwrite_evt_state old_state = rec->overwrite_evt_state;
>> +       enum action {
>> +               NONE,
>> +               PAUSE,
>> +               RESUME,
>> +       } action = NONE;
>> +
>> +       switch (old_state) {
>> +       case OVERWRITE_EVT_RUNNING:
>> +               if (state != OVERWRITE_EVT_RUNNING)
>> +                       action = PAUSE;
>> +               break;
>> +       case OVERWRITE_EVT_DATA_PENDING:
>> +               if (state == OVERWRITE_EVT_RUNNING)
>> +                       action = RESUME;
>> +               break;
>> +       case OVERWRITE_EVT_EMPTY:
>> +               if (state == OVERWRITE_EVT_RUNNING)
>> +                       action = RESUME;
>> +               if (state == OVERWRITE_EVT_DATA_PENDING)
>> +                       state = OVERWRITE_EVT_EMPTY;
> else if (state == OVERWRITE_EVT_DATA_PENDING)

You are right, but I believe compiler makes identical binaries even
without 'else'. With no 'else' these two 'if' are aligned.

I'll recheck this patch.

>> +               break;
>> +       default:
>> +               WARN_ONCE(1, "Shouldn't get there\n");
>> +       }
>> +
>> +       rec->overwrite_evt_state = state;
>> +
>> +       if (action == NONE)
>> +               return;
> I think the above two lines are not required.  The switch below should
> be enough.
>
>> +
>> +       if (!evlist)
>> +               return;
>> +
>> +       switch (action) {
>> +       case PAUSE:
>> +               perf_evlist__pause(evlist);
>> +               break;
>> +       case RESUME:
>> +               perf_evlist__resume(evlist);
>> +               break;
>> +       case NONE:
>> +       default:
>> +               break;
>> +       }
>> +}
>> +
> --
> Nilay

[PATCH v9 6/8] perf tools: Enable overwrite settings

[Wang Nan]

This patch allows following config terms and option:

Globally setting events to overwrite;

 # perf record --overwrite ...

Set specific events to be overwrite or no-overwrite.

 # perf record --event cycles/overwrite/ ...
 # perf record --event cycles/no-overwrite/ ...

Add missing config terms and update config term array size because the
longest string length is changed.

For overwritable events, automatically select attr.write_backward since
perf requires it to be backward for reading.

Test result:
 # perf record --overwrite -e syscalls:*enter_nanosleep* usleep 1
 [ perf record: Woken up 2 times to write data ]
 [ perf record: Captured and wrote 0.011 MB perf.data (1 samples) ]
 # perf evlist -v
 syscalls:sys_enter_nanosleep: type: 2, size: 112, config: 0x134, { sample_period, sample_freq }: 1, sample_type: IP|TID|TIME|CPU|PERIOD|RAW, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, write_backward: 1
 # Tip: use 'perf evlist --trace-fields' to show fields for tracepoint events

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: He Kuang <hekuang@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/Documentation/perf-record.txt | 14 ++++++++++++++
 tools/perf/builtin-record.c              |  1 +
 tools/perf/perf.h                        |  1 +
 tools/perf/tests/backward-ring-buffer.c  | 15 ++++++---------
 tools/perf/util/evsel.c                  | 12 ++++++++++++
 tools/perf/util/evsel.h                  |  2 ++
 tools/perf/util/parse-events.c           | 20 ++++++++++++++++++--
 tools/perf/util/parse-events.h           |  2 ++
 tools/perf/util/parse-events.l           |  2 ++
 9 files changed, 58 insertions(+), 11 deletions(-)

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index 5b46b1d..384c630 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -367,6 +367,20 @@ options.
 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
 in config file is set to true.
 
+--overwrite::
+Makes all events use an overwritable ring buffer. An overwritable ring
+buffer works like a flight recorder: when it gets full, the kernel will
+overwrite the oldest records, that thus will never make it to the
+perf.data file.
+
+When '--overwrite' and '--switch-output' are used perf records and drops
+events until it receives a signal, meaning that something unusual was
+detected that warrants taking a snapshot of the most current events,
+those fitting in the ring buffer at that moment.
+
+'overwrite' attribute can also be set or canceled for an event using
+config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 019151a..b8c4a64 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1523,6 +1523,7 @@ struct option __record_options[] = {
 	OPT_BOOLEAN_SET('i', "no-inherit", &record.opts.no_inherit,
 			&record.opts.no_inherit_set,
 			"child tasks do not inherit counters"),
+	OPT_BOOLEAN(0, "overwrite", &record.opts.overwrite, "use overwrite mode"),
 	OPT_UINTEGER('F', "freq", &record.opts.user_freq, "profile at this frequency"),
 	OPT_CALLBACK('m', "mmap-pages", &record.opts, "pages[,pages]",
 		     "number of mmap data pages and AUX area tracing mmap pages",
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index cd8f1b1..608b42b 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -59,6 +59,7 @@ struct record_opts {
 	bool	     record_switch_events;
 	bool	     all_kernel;
 	bool	     all_user;
+	bool	     overwrite;
 	unsigned int freq;
 	unsigned int mmap_pages;
 	unsigned int auxtrace_mmap_pages;
diff --git a/tools/perf/tests/backward-ring-buffer.c b/tools/perf/tests/backward-ring-buffer.c
index 0b100b8..926e30e 100644
--- a/tools/perf/tests/backward-ring-buffer.c
+++ b/tools/perf/tests/backward-ring-buffer.c
@@ -137,27 +137,24 @@ int test__backward_ring_buffer(int subtest __maybe_unused)
 	}
 
 	bzero(&parse_error, sizeof(parse_error));
-	err = parse_events(evlist, "syscalls:sys_enter_prctl", &parse_error);
+	/*
+	 * Set backward bit, ring buffer should be writing from end. Record
+	 * it in aux evlist
+	 */
+	err = parse_events(evlist, "syscalls:sys_enter_prctl/overwrite/", &parse_error);
 	if (err) {
 		pr_debug("Failed to parse tracepoint event, try use root\n");
 		ret = TEST_SKIP;
 		goto out_delete_evlist;
 	}
 
-	/*
-	 * Set backward bit, ring buffer should be writing from end. Record
-	 * it in aux evlist
-	 */
-	perf_evlist__last(evlist)->overwrite = true;
-	perf_evlist__last(evlist)->attr.write_backward = 1;
-
+	/* Don't set backward bit for exit event. Record it in main evlist */
 	err = parse_events(evlist, "syscalls:sys_exit_prctl", &parse_error);
 	if (err) {
 		pr_debug("Failed to parse tracepoint event, try use root\n");
 		ret = TEST_SKIP;
 		goto out_delete_evlist;
 	}
-	/* Don't set backward bit for exit event. Record it in main evlist */
 
 	perf_evlist__config(evlist, &opts, NULL);
 
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 1d8f2bb..3629520 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -677,11 +677,22 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			 */
 			attr->inherit = term->val.inherit ? 1 : 0;
 			break;
+		case PERF_EVSEL__CONFIG_TERM_OVERWRITE:
+			evsel->overwrite = term->val.overwrite ? 1 : 0;
+			break;
 		default:
 			break;
 		}
 	}
 
+	/*
+	 * Set backward after config term processing because it is
+	 * possible to set overwrite globally, without config
+	 * terms.
+	 */
+	if (evsel->overwrite)
+		attr->write_backward = 1;
+
 	/* User explicitly set per-event callgraph, clear the old setting and reset. */
 	if ((callgraph_buf != NULL) || (dump_size > 0) || max_stack) {
 		if (max_stack) {
@@ -758,6 +769,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 
 	attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1;
 	attr->inherit	    = !opts->no_inherit;
+	evsel->overwrite    = opts->overwrite;
 
 	perf_evsel__set_sample_bit(evsel, IP);
 	perf_evsel__set_sample_bit(evsel, TID);
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 828ddd1..55ff958 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -45,6 +45,7 @@ enum {
 	PERF_EVSEL__CONFIG_TERM_STACK_USER,
 	PERF_EVSEL__CONFIG_TERM_INHERIT,
 	PERF_EVSEL__CONFIG_TERM_MAX_STACK,
+	PERF_EVSEL__CONFIG_TERM_OVERWRITE,
 	PERF_EVSEL__CONFIG_TERM_MAX,
 };
 
@@ -59,6 +60,7 @@ struct perf_evsel_config_term {
 		u64	stack_user;
 		int	max_stack;
 		bool	inherit;
+		bool	overwrite;
 	} val;
 };
 
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index d15e335..b641f75 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -901,6 +901,8 @@ static const char *config_term_names[__PARSE_EVENTS__TERM_TYPE_NR] = {
 	[PARSE_EVENTS__TERM_TYPE_NOINHERIT]		= "no-inherit",
 	[PARSE_EVENTS__TERM_TYPE_INHERIT]		= "inherit",
 	[PARSE_EVENTS__TERM_TYPE_MAX_STACK]		= "max-stack",
+	[PARSE_EVENTS__TERM_TYPE_OVERWRITE]		= "overwrite",
+	[PARSE_EVENTS__TERM_TYPE_NOOVERWRITE]		= "no-overwrite",
 };
 
 static bool config_term_shrinked;
@@ -993,6 +995,12 @@ do {									   \
 	case PARSE_EVENTS__TERM_TYPE_NOINHERIT:
 		CHECK_TYPE_VAL(NUM);
 		break;
+	case PARSE_EVENTS__TERM_TYPE_OVERWRITE:
+		CHECK_TYPE_VAL(NUM);
+		break;
+	case PARSE_EVENTS__TERM_TYPE_NOOVERWRITE:
+		CHECK_TYPE_VAL(NUM);
+		break;
 	case PARSE_EVENTS__TERM_TYPE_NAME:
 		CHECK_TYPE_VAL(STR);
 		break;
@@ -1045,6 +1053,8 @@ static int config_term_tracepoint(struct perf_event_attr *attr,
 	case PARSE_EVENTS__TERM_TYPE_INHERIT:
 	case PARSE_EVENTS__TERM_TYPE_NOINHERIT:
 	case PARSE_EVENTS__TERM_TYPE_MAX_STACK:
+	case PARSE_EVENTS__TERM_TYPE_OVERWRITE:
+	case PARSE_EVENTS__TERM_TYPE_NOOVERWRITE:
 		return config_term_common(attr, term, err);
 	default:
 		if (err) {
@@ -1117,6 +1127,12 @@ do {								\
 		case PARSE_EVENTS__TERM_TYPE_MAX_STACK:
 			ADD_CONFIG_TERM(MAX_STACK, max_stack, term->val.num);
 			break;
+		case PARSE_EVENTS__TERM_TYPE_OVERWRITE:
+			ADD_CONFIG_TERM(OVERWRITE, overwrite, term->val.num ? 1 : 0);
+			break;
+		case PARSE_EVENTS__TERM_TYPE_NOOVERWRITE:
+			ADD_CONFIG_TERM(OVERWRITE, overwrite, term->val.num ? 0 : 1);
+			break;
 		default:
 			break;
 		}
@@ -2330,9 +2346,9 @@ static void config_terms_list(char *buf, size_t buf_sz)
 char *parse_events_formats_error_string(char *additional_terms)
 {
 	char *str;
-	/* "branch_type" is the longest name */
+	/* "no-overwrite" is the longest name */
 	char static_terms[__PARSE_EVENTS__TERM_TYPE_NR *
-			  (sizeof("branch_type") - 1)];
+			  (sizeof("no-overwrite") - 1)];
 
 	config_terms_list(static_terms, sizeof(static_terms));
 	/* valid terms */
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index 46c05cc..1b04d82 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -69,6 +69,8 @@ enum {
 	PARSE_EVENTS__TERM_TYPE_NOINHERIT,
 	PARSE_EVENTS__TERM_TYPE_INHERIT,
 	PARSE_EVENTS__TERM_TYPE_MAX_STACK,
+	PARSE_EVENTS__TERM_TYPE_NOOVERWRITE,
+	PARSE_EVENTS__TERM_TYPE_OVERWRITE,
 	__PARSE_EVENTS__TERM_TYPE_NR,
 };
 
diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l
index 3c15b33..7a25194 100644
--- a/tools/perf/util/parse-events.l
+++ b/tools/perf/util/parse-events.l
@@ -202,6 +202,8 @@ stack-size		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_STACKSIZE); }
 max-stack		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_MAX_STACK); }
 inherit			{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_INHERIT); }
 no-inherit		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOINHERIT); }
+overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_OVERWRITE); }
+no-overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOOVERWRITE); }
 ,			{ return ','; }
 "/"			{ BEGIN(INITIAL); return '/'; }
 {name_minus}		{ return str(yyscanner, PE_NAME); }
-- 
1.8.3.4

[PATCH v9 7/8] perf tools: Don't warn about out of order event if write_backward is used

[Wang Nan]

If write_backward attribute is set, records are written into kernel
ring buffer from end to beginning, but read from beginning to end.
To avoid 'XX out of order events recorded' warning message (timestamps
of records is in reverse order when using write_backward), suppress the
warning message if write_backward is selected by at lease one event.

Result:

Before this patch:
 # perf record -m 1 -e raw_syscalls:sys_exit/overwrite/ \
                    -e raw_syscalls:sys_enter \
                    dd if=/dev/zero of=/dev/null count=300
 300+0 records in
 300+0 records out
 153600 bytes (154 kB) copied, 0.000601617 s, 255 MB/s
 [ perf record: Woken up 5 times to write data ]
 Warning:
 40 out of order events recorded.
 [ perf record: Captured and wrote 0.096 MB perf.data (696 samples) ]

After this patch:
 # perf record -m 1 -e raw_syscalls:sys_exit/overwrite/ \
                    -e raw_syscalls:sys_enter \
                    dd if=/dev/zero of=/dev/null count=300
 300+0 records in
 300+0 records out
 153600 bytes (154 kB) copied, 0.000644873 s, 238 MB/s
 [ perf record: Woken up 5 times to write data ]
 [ perf record: Captured and wrote 0.096 MB perf.data (696 samples) ]

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: He Kuang <hekuang@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/util/session.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index dfedf09..561441e 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -1497,10 +1497,27 @@ int perf_session__register_idle_thread(struct perf_session *session)
 	return err;
 }
 
+static void
+perf_session__warn_order(const struct perf_session *session)
+{
+	const struct ordered_events *oe = &session->ordered_events;
+	struct perf_evsel *evsel;
+	bool should_warn = true;
+
+	evlist__for_each(session->evlist, evsel) {
+		if (evsel->attr.write_backward)
+			should_warn = false;
+	}
+
+	if (!should_warn)
+		return;
+	if (oe->nr_unordered_events != 0)
+		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
+}
+
 static void perf_session__warn_about_errors(const struct perf_session *session)
 {
 	const struct events_stats *stats = &session->evlist->stats;
-	const struct ordered_events *oe = &session->ordered_events;
 
 	if (session->tool->lost == perf_event__process_lost &&
 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
@@ -1557,8 +1574,7 @@ static void perf_session__warn_about_errors(const struct perf_session *session)
 			    stats->nr_unprocessable_samples);
 	}
 
-	if (oe->nr_unordered_events != 0)
-		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
+	perf_session__warn_order(session);
 
 	events_stats__auxtrace_error_warn(stats);
 
-- 
1.8.3.4

[PATCH v9 8/8] perf tools: Add --tail-synthesize option

[Wang Nan]

When working with overwritable ring buffer there's a inconvenience
problem: if perf dumps data after a long period after it starts, non-sample
events may lost, which makes following 'perf report' unable to identify
proc name and mmap layout. For example:

 # perf record -m 4 -e raw_syscalls:* -g --overwrite --switch-output \
        dd if=/dev/zero of=/dev/null

send SIGUSR2 after dd runs long enough. The resuling perf.data lost
correct comm and mmap events:

 # perf script -i perf.data.2016061522374354
 perf 24478 [004] 2581325.601789:  raw_syscalls:sys_exit: NR 0 = 512
 ^^^^
 Should be 'dd'
                   27b2e8 syscall_slow_exit_work+0xfe2000e3 (/lib/modules/4.6.0-rc3+/build/vmlinux)
                   203cc7 do_syscall_64+0xfe200117 (/lib/modules/4.6.0-rc3+/build/vmlinux)
                   b18d83 return_from_SYSCALL_64+0xfe200000 (/lib/modules/4.6.0-rc3+/build/vmlinux)
             7f47c417edf0 [unknown] ([unknown])
             ^^^^^^^^^^^^
             Fail to unwind

This patch provides a '--tail-synthesize' option, allows perf to collect
system status when finalizing output file. In resuling output file, the
non-sample events reflect system status when dumping data.

After this patch:
 # perf record -m 4 -e raw_syscalls:* -g --overwrite --switch-output --tail-synthesize \
        dd if=/dev/zero of=/dev/null

 # perf script -i perf.data.2016061600544998
 dd 27364 [004] 2583244.994464: raw_syscalls:sys_enter: NR 1 (1, ...
 ^^
 Correct comm
                   203a18 syscall_trace_enter_phase2+0xfe2001a8 ([kernel.kallsyms])
                   203aa5 syscall_trace_enter+0xfe200055 ([kernel.kallsyms])
                   203caa do_syscall_64+0xfe2000fa ([kernel.kallsyms])
                   b18d83 return_from_SYSCALL_64+0xfe200000 ([kernel.kallsyms])
                    d8e50 __GI___libc_write+0xffff01d9639f4010 (/tmp/oxygen_root-w00229757/lib64/libc-2.18.so)
                    ^^^^^
                    Correct unwind

This option doesn't aim to solve this problem completely. If a process
terminates before SIGUSR2, we still lost its COMM and MMAP events. For
example, we can't unwind correctly from the final perf.data we get from
the previous example, because when perf collects the final output file
(when we press C-c), 'dd' has been terminated so its
'/proc/<pid>/mmap' becomes empty. However, this is a cheaper choice. To
completely solve this problem we need to continously output non-sample
events. To satisify the requirement of daemonization, we need to merge
them periodically. It is possible but requires much more code and cycles.

Automatically select --tail-synthesize when --overwrite is provided.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 tools/perf/Documentation/perf-record.txt |  8 ++++++++
 tools/perf/builtin-record.c              | 31 +++++++++++++++++++++++++------
 tools/perf/perf.h                        |  1 +
 3 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index 384c630..69966ab 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -367,6 +367,12 @@ options.
 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
 in config file is set to true.
 
+--tail-synthesize::
+Instead of collecting non-sample events (for example, fork, comm, mmap) at
+the beginning of record, collect them during finalizing an output file.
+The collected non-sample events reflects the status of the system when
+record is finished.
+
 --overwrite::
 Makes all events use an overwritable ring buffer. An overwritable ring
 buffer works like a flight recorder: when it gets full, the kernel will
@@ -381,6 +387,8 @@ those fitting in the ring buffer at that moment.
 'overwrite' attribute can also be set or canceled for an event using
 config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
 
+Implies --tail-synthesize.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index b8c4a64..337ca69 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -725,13 +725,16 @@ record__finish_output(struct record *rec)
 	return;
 }
 
-static int record__synthesize_workload(struct record *rec)
+static int record__synthesize_workload(struct record *rec, bool tail)
 {
 	struct {
 		struct thread_map map;
 		struct thread_map_data map_data;
 	} thread_map;
 
+	if (rec->opts.tail_synthesize != tail)
+		return 0;
+
 	thread_map.map.nr = 1;
 	thread_map.map.map[0].pid = rec->evlist->workload.pid;
 	thread_map.map.map[0].comm = NULL;
@@ -742,7 +745,7 @@ static int record__synthesize_workload(struct record *rec)
 						 rec->opts.proc_map_timeout);
 }
 
-static int record__synthesize(struct record *rec);
+static int record__synthesize(struct record *rec, bool tail);
 
 static int
 record__switch_output(struct record *rec, bool at_exit)
@@ -753,6 +756,10 @@ record__switch_output(struct record *rec, bool at_exit)
 	/* Same Size:      "2015122520103046"*/
 	char timestamp[] = "InvalidTimestamp";
 
+	record__synthesize(rec, true);
+	if (target__none(&rec->opts.target))
+		record__synthesize_workload(rec, true);
+
 	rec->samples = 0;
 	record__finish_output(rec);
 	err = fetch_current_timestamp(timestamp, sizeof(timestamp));
@@ -775,7 +782,7 @@ record__switch_output(struct record *rec, bool at_exit)
 
 	/* Output tracking events */
 	if (!at_exit) {
-		record__synthesize(rec);
+		record__synthesize(rec, false);
 
 		/*
 		 * In 'perf record --switch-output' without -a,
@@ -787,7 +794,7 @@ record__switch_output(struct record *rec, bool at_exit)
 		 * perf_event__synthesize_thread_map() for those events.
 		 */
 		if (target__none(&rec->opts.target))
-			record__synthesize_workload(rec);
+			record__synthesize_workload(rec, false);
 	}
 	return fd;
 }
@@ -842,7 +849,7 @@ static const struct perf_event_mmap_page *record__pick_pc(struct record *rec)
 	return NULL;
 }
 
-static int record__synthesize(struct record *rec)
+static int record__synthesize(struct record *rec, bool tail)
 {
 	struct perf_session *session = rec->session;
 	struct machine *machine = &session->machines.host;
@@ -852,6 +859,9 @@ static int record__synthesize(struct record *rec)
 	int fd = perf_data_file__fd(file);
 	int err = 0;
 
+	if (rec->opts.tail_synthesize != tail)
+		return 0;
+
 	if (file->is_pipe) {
 		err = perf_event__synthesize_attrs(tool, session,
 						   process_synthesized_event);
@@ -1015,7 +1025,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 
 	machine = &session->machines.host;
 
-	err = record__synthesize(rec);
+	err = record__synthesize(rec, false);
 	if (err < 0)
 		goto out_child;
 
@@ -1180,6 +1190,9 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 	if (!quiet)
 		fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);
 
+	if (target__none(&rec->opts.target))
+		record__synthesize_workload(rec, true);
+
 out_child:
 	if (forks) {
 		int exit_status;
@@ -1198,6 +1211,7 @@ out_child:
 	} else
 		status = err;
 
+	record__synthesize(rec, true);
 	/* this will be recalculated during process_buildids() */
 	rec->samples = 0;
 
@@ -1523,6 +1537,8 @@ struct option __record_options[] = {
 	OPT_BOOLEAN_SET('i', "no-inherit", &record.opts.no_inherit,
 			&record.opts.no_inherit_set,
 			"child tasks do not inherit counters"),
+	OPT_BOOLEAN(0, "tail-synthesize", &record.opts.tail_synthesize,
+		    "synthesize non-sample events at the end of output"),
 	OPT_BOOLEAN(0, "overwrite", &record.opts.overwrite, "use overwrite mode"),
 	OPT_UINTEGER('F', "freq", &record.opts.user_freq, "profile at this frequency"),
 	OPT_CALLBACK('m', "mmap-pages", &record.opts, "pages[,pages]",
@@ -1734,6 +1750,9 @@ int cmd_record(int argc, const char **argv, const char *prefix __maybe_unused)
 		}
 	}
 
+	if (record.opts.overwrite)
+		record.opts.tail_synthesize = true;
+
 	if (rec->evlist->nr_entries == 0 &&
 	    perf_evlist__add_default(rec->evlist) < 0) {
 		pr_err("Not enough memory for event selector list\n");
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 608b42b..a7e0f14 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -59,6 +59,7 @@ struct record_opts {
 	bool	     record_switch_events;
 	bool	     all_kernel;
 	bool	     all_user;
+	bool	     tail_synthesize;
 	bool	     overwrite;
 	unsigned int freq;
 	unsigned int mmap_pages;
-- 
1.8.3.4