[GIT PULL 0/9] perf/core improvements and fixes

[GIT PULL 0/9] perf/core improvements and fixes

[Arnaldo Carvalho de Melo]

Hi Ingo,

	This is a signed tag, please lemme know if everything went well.

	The --uid feature works for root, we still need to sort out that
paranoia with some threads owned by a user that prevents 'perf --uid
non-root-user' to work for 'non-root-user'.

	http://git.kernel.org/?p=linux/kernel/git/acme/linux.git;a=tag;h=ce9600c4c664ce7f97e8aa5e756b0b4ea5b017c7
looks ok to me, need just to improve on the commit log message, I'll get
used to it :-)

- Arnaldo

The following changes since commit 172d1b0b73256551f100fc00c69e356d047103f5:

  perf tools: Fix compile error on x86_64 Ubuntu (2012-01-08 13:34:55 -0200)

are available in the git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux perf-core-for-mingo

for you to fetch changes up to f8f4b2872295dca88339ec0c403b2217b1197353:

  perf tools: Fix strlen() bug in perf_event__synthesize_event_type() (2012-01-24 20:31:34 -0200)

----------------------------------------------------------------
perf/core improvements and fixes

----------------------------------------------------------------
Arnaldo Carvalho de Melo (2):
      perf tools: Add fprintf methods for thread_map and cpu_map classes
      perf tools: Introduce per user view

David Daney (1):
      perf tools: Fix broken build by defining _GNU_SOURCE in Makefile

Jan Beulich (4):
      perf bench: Make "default" memcpy() selection actually use glibc's implementation
      perf bench: Also allow measuring alternative memcpy implementations
      perf bench: Also allow measuring memset()
      perf bench: Allow passing an iteration count to "bench mem mem{cpy,set}"

Srikar Dronamraju (1):
      perf probe: Usability fixes

Stephane Eranian (1):
      perf tools: Fix strlen() bug in perf_event__synthesize_event_type()

 tools/perf/Documentation/perf-record.txt     |    4 +
 tools/perf/Documentation/perf-top.txt        |    4 +
 tools/perf/Makefile                          |   11 +-
 tools/perf/bench/bench.h                     |    1 +
 tools/perf/bench/mem-memcpy-x86-64-asm-def.h |    8 +
 tools/perf/bench/mem-memcpy-x86-64-asm.S     |    6 +-
 tools/perf/bench/mem-memcpy.c                |   11 +-
 tools/perf/bench/mem-memset-arch.h           |   12 +
 tools/perf/bench/mem-memset-x86-64-asm-def.h |   12 +
 tools/perf/bench/mem-memset-x86-64-asm.S     |    6 +
 tools/perf/bench/mem-memset.c                |  298 ++++++++++++++++++++++++++
 tools/perf/builtin-bench.c                   |    3 +
 tools/perf/builtin-probe.c                   |    2 -
 tools/perf/builtin-record.c                  |   12 +-
 tools/perf/builtin-stat.c                    |    2 +-
 tools/perf/builtin-test.c                    |    8 +-
 tools/perf/builtin-top.c                     |   22 ++-
 tools/perf/perf.h                            |    1 +
 tools/perf/util/cpumap.c                     |   11 +
 tools/perf/util/cpumap.h                     |    4 +
 tools/perf/util/evlist.c                     |    6 +-
 tools/perf/util/evlist.h                     |    2 +-
 tools/perf/util/header.c                     |    2 +-
 tools/perf/util/hist.h                       |    1 +
 tools/perf/util/include/asm/dwarf2.h         |    4 +-
 tools/perf/util/probe-event.c                |    8 +-
 tools/perf/util/python.c                     |   10 +-
 tools/perf/util/symbol.c                     |    1 -
 tools/perf/util/thread_map.c                 |  109 +++++++++-
 tools/perf/util/thread_map.h                 |    7 +-
 tools/perf/util/top.c                        |    3 +
 tools/perf/util/top.h                        |    2 +
 tools/perf/util/trace-event-parse.c          |    3 +-
 tools/perf/util/ui/browsers/hists.c          |    5 +-
 tools/perf/util/ui/helpline.c                |    1 -
 tools/perf/util/usage.c                      |   39 ++++
 tools/perf/util/util.h                       |    3 +-
 37 files changed, 597 insertions(+), 47 deletions(-)
 create mode 100644 tools/perf/bench/mem-memset-arch.h
 create mode 100644 tools/perf/bench/mem-memset-x86-64-asm-def.h
 create mode 100644 tools/perf/bench/mem-memset-x86-64-asm.S
 create mode 100644 tools/perf/bench/mem-memset.c

[PATCH 1/9] perf tools: Add fprintf methods for thread_map and cpu_map classes

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

For helping with debugging.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-m06n4rp7pwr6dlzwoq89cl69@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/cpumap.c     |   11 +++++++++++
 tools/perf/util/cpumap.h     |    4 ++++
 tools/perf/util/thread_map.c |   11 +++++++++++
 tools/perf/util/thread_map.h |    4 ++++
 4 files changed, 30 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c
index 6893eec..adc72f0 100644
--- a/tools/perf/util/cpumap.c
+++ b/tools/perf/util/cpumap.c
@@ -166,6 +166,17 @@ out:
 	return cpus;
 }
 
+size_t cpu_map__fprintf(struct cpu_map *map, FILE *fp)
+{
+	int i;
+	size_t printed = fprintf(fp, "%d cpu%s: ",
+				 map->nr, map->nr > 1 ? "s" : "");
+	for (i = 0; i < map->nr; ++i)
+		printed += fprintf(fp, "%s%d", i ? ", " : "", map->map[i]);
+
+	return printed + fprintf(fp, "\n");
+}
+
 struct cpu_map *cpu_map__dummy_new(void)
 {
 	struct cpu_map *cpus = malloc(sizeof(*cpus) + sizeof(int));
diff --git a/tools/perf/util/cpumap.h b/tools/perf/util/cpumap.h
index 072c0a3..c415185 100644
--- a/tools/perf/util/cpumap.h
+++ b/tools/perf/util/cpumap.h
@@ -1,6 +1,8 @@
 #ifndef __PERF_CPUMAP_H
 #define __PERF_CPUMAP_H
 
+#include <stdio.h>
+
 struct cpu_map {
 	int nr;
 	int map[];
@@ -10,4 +12,6 @@ struct cpu_map *cpu_map__new(const char *cpu_list);
 struct cpu_map *cpu_map__dummy_new(void);
 void cpu_map__delete(struct cpu_map *map);
 
+size_t cpu_map__fprintf(struct cpu_map *map, FILE *fp);
+
 #endif /* __PERF_CPUMAP_H */
diff --git a/tools/perf/util/thread_map.c b/tools/perf/util/thread_map.c
index a5df131..894d52f 100644
--- a/tools/perf/util/thread_map.c
+++ b/tools/perf/util/thread_map.c
@@ -62,3 +62,14 @@ void thread_map__delete(struct thread_map *threads)
 {
 	free(threads);
 }
+
+size_t thread_map__fprintf(struct thread_map *threads, FILE *fp)
+{
+	int i;
+	size_t printed = fprintf(fp, "%d thread%s: ",
+				 threads->nr, threads->nr > 1 ? "s" : "");
+	for (i = 0; i < threads->nr; ++i)
+		printed += fprintf(fp, "%s%d", i ? ", " : "", threads->map[i]);
+
+	return printed + fprintf(fp, "\n");
+}
diff --git a/tools/perf/util/thread_map.h b/tools/perf/util/thread_map.h
index 3cb9073..736ab4a26 100644
--- a/tools/perf/util/thread_map.h
+++ b/tools/perf/util/thread_map.h
@@ -2,6 +2,7 @@
 #define __PERF_THREAD_MAP_H
 
 #include <sys/types.h>
+#include <stdio.h>
 
 struct thread_map {
 	int nr;
@@ -12,4 +13,7 @@ struct thread_map *thread_map__new_by_pid(pid_t pid);
 struct thread_map *thread_map__new_by_tid(pid_t tid);
 struct thread_map *thread_map__new(pid_t pid, pid_t tid);
 void thread_map__delete(struct thread_map *threads);
+
+size_t thread_map__fprintf(struct thread_map *threads, FILE *fp);
+
 #endif	/* __PERF_THREAD_MAP_H */
-- 
1.7.9.rc2.1.g69204

[PATCH 2/9] perf tools: Introduce per user view

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

The new --uid command line option will show only the tasks for a given
user, using the proc interface to figure out the existing tasks.

Kernel work is needed to close races at startup, but this should already
be useful in many use cases.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-bdnspm000gw2l984a2t53o8z@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Documentation/perf-record.txt |    4 +
 tools/perf/Documentation/perf-top.txt    |    4 +
 tools/perf/builtin-record.c              |   12 +++-
 tools/perf/builtin-stat.c                |    2 +-
 tools/perf/builtin-test.c                |    8 +-
 tools/perf/builtin-top.c                 |   22 ++++++-
 tools/perf/perf.h                        |    1 +
 tools/perf/util/evlist.c                 |    6 +-
 tools/perf/util/evlist.h                 |    2 +-
 tools/perf/util/hist.h                   |    1 +
 tools/perf/util/python.c                 |   10 ++--
 tools/perf/util/thread_map.c             |   98 +++++++++++++++++++++++++++++-
 tools/perf/util/thread_map.h             |    3 +-
 tools/perf/util/top.c                    |    3 +
 tools/perf/util/top.h                    |    2 +
 tools/perf/util/ui/browsers/hists.c      |    3 +
 tools/perf/util/usage.c                  |   39 ++++++++++++
 tools/perf/util/util.h                   |    2 +
 18 files changed, 200 insertions(+), 22 deletions(-)

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index 2937f7e..ff9a66e 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -58,6 +58,10 @@ OPTIONS
 --tid=::
         Record events on existing thread ID.
 
+-u::
+--uid=::
+        Record events in threads owned by uid. Name or number.
+
 -r::
 --realtime=::
 	Collect data with this RT SCHED_FIFO priority.
diff --git a/tools/perf/Documentation/perf-top.txt b/tools/perf/Documentation/perf-top.txt
index b1a5bbb..ab1454e 100644
--- a/tools/perf/Documentation/perf-top.txt
+++ b/tools/perf/Documentation/perf-top.txt
@@ -78,6 +78,10 @@ Default is to monitor all CPUS.
 --tid=<tid>::
         Profile events on existing thread ID.
 
+-u::
+--uid=::
+        Record events in threads owned by uid. Name or number.
+
 -r <priority>::
 --realtime=<priority>::
 	Collect data with this RT SCHED_FIFO priority.
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0abfb18..32870ee 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -44,6 +44,7 @@ struct perf_record {
 	struct perf_evlist	*evlist;
 	struct perf_session	*session;
 	const char		*progname;
+	const char		*uid_str;
 	int			output;
 	unsigned int		page_size;
 	int			realtime_prio;
@@ -727,6 +728,7 @@ const struct option record_options[] = {
 	OPT_CALLBACK('G', "cgroup", &record.evlist, "name",
 		     "monitor event in cgroup name only",
 		     parse_cgroups),
+	OPT_STRING('u', "uid", &record.uid_str, "user", "user to profile"),
 	OPT_END()
 };
 
@@ -748,7 +750,7 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 	argc = parse_options(argc, argv, record_options, record_usage,
 			    PARSE_OPT_STOP_AT_NON_OPTION);
 	if (!argc && rec->opts.target_pid == -1 && rec->opts.target_tid == -1 &&
-		!rec->opts.system_wide && !rec->opts.cpu_list)
+		!rec->opts.system_wide && !rec->opts.cpu_list && !rec->uid_str)
 		usage_with_options(record_usage, record_options);
 
 	if (rec->force && rec->append_file) {
@@ -788,11 +790,17 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 		goto out_symbol_exit;
 	}
 
+	rec->opts.uid = parse_target_uid(rec->uid_str, rec->opts.target_tid,
+					 rec->opts.target_pid);
+	if (rec->uid_str != NULL && rec->opts.uid == UINT_MAX - 1)
+		goto out_free_fd;
+
 	if (rec->opts.target_pid != -1)
 		rec->opts.target_tid = rec->opts.target_pid;
 
 	if (perf_evlist__create_maps(evsel_list, rec->opts.target_pid,
-				     rec->opts.target_tid, rec->opts.cpu_list) < 0)
+				     rec->opts.target_tid, rec->opts.uid,
+				     rec->opts.cpu_list) < 0)
 		usage_with_options(record_usage, record_options);
 
 	list_for_each_entry(pos, &evsel_list->entries, node) {
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index f5d2a63..459b862 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1201,7 +1201,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 	if (target_pid != -1)
 		target_tid = target_pid;
 
-	evsel_list->threads = thread_map__new(target_pid, target_tid);
+	evsel_list->threads = thread_map__new(target_pid, target_tid, UINT_MAX);
 	if (evsel_list->threads == NULL) {
 		pr_err("Problems finding threads of monitor\n");
 		usage_with_options(stat_usage, options);
diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c
index 3854e86..3ce709e 100644
--- a/tools/perf/builtin-test.c
+++ b/tools/perf/builtin-test.c
@@ -276,7 +276,7 @@ static int test__open_syscall_event(void)
 		return -1;
 	}
 
-	threads = thread_map__new(-1, getpid());
+	threads = thread_map__new(-1, getpid(), UINT_MAX);
 	if (threads == NULL) {
 		pr_debug("thread_map__new\n");
 		return -1;
@@ -342,7 +342,7 @@ static int test__open_syscall_event_on_all_cpus(void)
 		return -1;
 	}
 
-	threads = thread_map__new(-1, getpid());
+	threads = thread_map__new(-1, getpid(), UINT_MAX);
 	if (threads == NULL) {
 		pr_debug("thread_map__new\n");
 		return -1;
@@ -490,7 +490,7 @@ static int test__basic_mmap(void)
 		expected_nr_events[i] = random() % 257;
 	}
 
-	threads = thread_map__new(-1, getpid());
+	threads = thread_map__new(-1, getpid(), UINT_MAX);
 	if (threads == NULL) {
 		pr_debug("thread_map__new\n");
 		return -1;
@@ -1054,7 +1054,7 @@ static int test__PERF_RECORD(void)
 	 * we're monitoring, the one forked there.
 	 */
 	err = perf_evlist__create_maps(evlist, opts.target_pid,
-				       opts.target_tid, opts.cpu_list);
+				       opts.target_tid, UINT_MAX, opts.cpu_list);
 	if (err < 0) {
 		pr_debug("Not enough memory to create thread/cpu maps\n");
 		goto out_delete_evlist;
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 8f80df8..e8b033c 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -64,7 +64,6 @@
 #include <linux/unistd.h>
 #include <linux/types.h>
 
-
 void get_term_dimensions(struct winsize *ws)
 {
 	char *s = getenv("LINES");
@@ -537,10 +536,20 @@ static void perf_top__sort_new_samples(void *arg)
 
 static void *display_thread_tui(void *arg)
 {
+	struct perf_evsel *pos;
 	struct perf_top *top = arg;
 	const char *help = "For a higher level overview, try: perf top --sort comm,dso";
 
 	perf_top__sort_new_samples(top);
+
+	/*
+	 * Initialize the uid_filter_str, in the future the TUI will allow
+	 * Zooming in/out UIDs. For now juse use whatever the user passed
+	 * via --uid.
+	 */
+	list_for_each_entry(pos, &top->evlist->entries, node)
+		pos->hists.uid_filter_str = top->uid_str;
+
 	perf_evlist__tui_browse_hists(top->evlist, help,
 				      perf_top__sort_new_samples,
 				      top, top->delay_secs);
@@ -949,7 +958,7 @@ static int __cmd_top(struct perf_top *top)
 	if (ret)
 		goto out_delete;
 
-	if (top->target_tid != -1)
+	if (top->target_tid != -1 || top->uid != UINT_MAX)
 		perf_event__synthesize_thread_map(&top->tool, top->evlist->threads,
 						  perf_event__process,
 						  &top->session->host_machine);
@@ -1089,6 +1098,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 		.delay_secs	     = 2,
 		.target_pid	     = -1,
 		.target_tid	     = -1,
+		.uid		     = UINT_MAX,
 		.freq		     = 1000, /* 1 KHz */
 		.sample_id_all_avail = true,
 		.mmap_pages	     = 128,
@@ -1162,6 +1172,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 		    "Display raw encoding of assembly instructions (default)"),
 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
+	OPT_STRING('u', "uid", &top.uid_str, "user", "user to profile"),
 	OPT_END()
 	};
 
@@ -1187,6 +1198,10 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 
 	setup_browser(false);
 
+	top.uid = parse_target_uid(top.uid_str, top.target_tid, top.target_pid);
+	if (top.uid_str != NULL && top.uid == UINT_MAX - 1)
+		goto out_delete_evlist;
+
 	/* CPU and PID are mutually exclusive */
 	if (top.target_tid > 0 && top.cpu_list) {
 		printf("WARNING: PID switch overriding CPU\n");
@@ -1198,7 +1213,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 		top.target_tid = top.target_pid;
 
 	if (perf_evlist__create_maps(top.evlist, top.target_pid,
-				     top.target_tid, top.cpu_list) < 0)
+				     top.target_tid, top.uid, top.cpu_list) < 0)
 		usage_with_options(top_usage, options);
 
 	if (!top.evlist->nr_entries &&
@@ -1262,6 +1277,7 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 
 	status = __cmd_top(&top);
 
+out_delete_evlist:
 	perf_evlist__delete(top.evlist);
 
 	return status;
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 64f8bee..92af168 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -188,6 +188,7 @@ void pthread__unblock_sigwinch(void);
 struct perf_record_opts {
 	pid_t	     target_pid;
 	pid_t	     target_tid;
+	uid_t	     uid;
 	bool	     call_graph;
 	bool	     group;
 	bool	     inherit_stat;
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 3f16e08..a6d50e3 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -594,14 +594,14 @@ int perf_evlist__mmap(struct perf_evlist *evlist, unsigned int pages,
 }
 
 int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid,
-			     pid_t target_tid, const char *cpu_list)
+			     pid_t target_tid, uid_t uid, const char *cpu_list)
 {
-	evlist->threads = thread_map__new(target_pid, target_tid);
+	evlist->threads = thread_map__new(target_pid, target_tid, uid);
 
 	if (evlist->threads == NULL)
 		return -1;
 
-	if (cpu_list == NULL && target_tid != -1)
+	if (uid != UINT_MAX || (cpu_list == NULL && target_tid != -1))
 		evlist->cpus = cpu_map__dummy_new();
 	else
 		evlist->cpus = cpu_map__new(cpu_list);
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 8922aee..9c51660 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -107,7 +107,7 @@ static inline void perf_evlist__set_maps(struct perf_evlist *evlist,
 }
 
 int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid,
-			     pid_t target_tid, const char *cpu_list);
+			     pid_t tid, uid_t uid, const char *cpu_list);
 void perf_evlist__delete_maps(struct perf_evlist *evlist);
 int perf_evlist__set_filters(struct perf_evlist *evlist);
 
diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h
index f55f0a8d..0d48613 100644
--- a/tools/perf/util/hist.h
+++ b/tools/perf/util/hist.h
@@ -55,6 +55,7 @@ struct hists {
 	u64			nr_entries;
 	const struct thread	*thread_filter;
 	const struct dso	*dso_filter;
+	const char		*uid_filter_str;
 	pthread_mutex_t		lock;
 	struct events_stats	stats;
 	u64			event_stream;
diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c
index 9dd47a4..e03b58a 100644
--- a/tools/perf/util/python.c
+++ b/tools/perf/util/python.c
@@ -425,14 +425,14 @@ struct pyrf_thread_map {
 static int pyrf_thread_map__init(struct pyrf_thread_map *pthreads,
 				 PyObject *args, PyObject *kwargs)
 {
-	static char *kwlist[] = { "pid", "tid", NULL };
-	int pid = -1, tid = -1;
+	static char *kwlist[] = { "pid", "tid", "uid", NULL };
+	int pid = -1, tid = -1, uid = UINT_MAX;
 
-	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|ii",
-					 kwlist, &pid, &tid))
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|iii",
+					 kwlist, &pid, &tid, &uid))
 		return -1;
 
-	pthreads->threads = thread_map__new(pid, tid);
+	pthreads->threads = thread_map__new(pid, tid, uid);
 	if (pthreads->threads == NULL)
 		return -1;
 	return 0;
diff --git a/tools/perf/util/thread_map.c b/tools/perf/util/thread_map.c
index 894d52f..3d4b6c5 100644
--- a/tools/perf/util/thread_map.c
+++ b/tools/perf/util/thread_map.c
@@ -1,6 +1,11 @@
 #include <dirent.h>
+#include <limits.h>
+#include <stdbool.h>
 #include <stdlib.h>
 #include <stdio.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
 #include "thread_map.h"
 
 /* Skip "." and ".." directories */
@@ -23,7 +28,7 @@ struct thread_map *thread_map__new_by_pid(pid_t pid)
 	sprintf(name, "/proc/%d/task", pid);
 	items = scandir(name, &namelist, filter, NULL);
 	if (items <= 0)
-                return NULL;
+		return NULL;
 
 	threads = malloc(sizeof(*threads) + sizeof(pid_t) * items);
 	if (threads != NULL) {
@@ -51,10 +56,99 @@ struct thread_map *thread_map__new_by_tid(pid_t tid)
 	return threads;
 }
 
-struct thread_map *thread_map__new(pid_t pid, pid_t tid)
+struct thread_map *thread_map__new_by_uid(uid_t uid)
+{
+	DIR *proc;
+	int max_threads = 32, items, i;
+	char path[256];
+	struct dirent dirent, *next, **namelist = NULL;
+	struct thread_map *threads = malloc(sizeof(*threads) +
+					    max_threads * sizeof(pid_t));
+	if (threads == NULL)
+		goto out;
+
+	proc = opendir("/proc");
+	if (proc == NULL)
+		goto out_free_threads;
+
+	threads->nr = 0;
+
+	while (!readdir_r(proc, &dirent, &next) && next) {
+		char *end;
+		bool grow = false;
+		struct stat st;
+		pid_t pid = strtol(dirent.d_name, &end, 10);
+
+		if (*end) /* only interested in proper numerical dirents */
+			continue;
+
+		snprintf(path, sizeof(path), "/proc/%s", dirent.d_name);
+
+		if (stat(path, &st) != 0)
+			continue;
+
+		if (st.st_uid != uid)
+			continue;
+
+		snprintf(path, sizeof(path), "/proc/%d/task", pid);
+		items = scandir(path, &namelist, filter, NULL);
+		if (items <= 0)
+			goto out_free_closedir;
+
+		while (threads->nr + items >= max_threads) {
+			max_threads *= 2;
+			grow = true;
+		}
+
+		if (grow) {
+			struct thread_map *tmp;
+
+			tmp = realloc(threads, (sizeof(*threads) +
+						max_threads * sizeof(pid_t)));
+			if (tmp == NULL)
+				goto out_free_namelist;
+
+			threads = tmp;
+		}
+
+		for (i = 0; i < items; i++)
+			threads->map[threads->nr + i] = atoi(namelist[i]->d_name);
+
+		for (i = 0; i < items; i++)
+			free(namelist[i]);
+		free(namelist);
+
+		threads->nr += items;
+	}
+
+out_closedir:
+	closedir(proc);
+out:
+	return threads;
+
+out_free_threads:
+	free(threads);
+	return NULL;
+
+out_free_namelist:
+	for (i = 0; i < items; i++)
+		free(namelist[i]);
+	free(namelist);
+
+out_free_closedir:
+	free(threads);
+	threads = NULL;
+	goto out_closedir;
+}
+
+struct thread_map *thread_map__new(pid_t pid, pid_t tid, uid_t uid)
 {
 	if (pid != -1)
 		return thread_map__new_by_pid(pid);
+
+	if (tid == -1 && uid != UINT_MAX)
+		return thread_map__new_by_uid(uid);
+
 	return thread_map__new_by_tid(tid);
 }
 
diff --git a/tools/perf/util/thread_map.h b/tools/perf/util/thread_map.h
index 736ab4a26..c75ddba 100644
--- a/tools/perf/util/thread_map.h
+++ b/tools/perf/util/thread_map.h
@@ -11,7 +11,8 @@ struct thread_map {
 
 struct thread_map *thread_map__new_by_pid(pid_t pid);
 struct thread_map *thread_map__new_by_tid(pid_t tid);
-struct thread_map *thread_map__new(pid_t pid, pid_t tid);
+struct thread_map *thread_map__new_by_uid(uid_t uid);
+struct thread_map *thread_map__new(pid_t pid, pid_t tid, uid_t uid);
 void thread_map__delete(struct thread_map *threads);
 
 size_t thread_map__fprintf(struct thread_map *threads, FILE *fp);
diff --git a/tools/perf/util/top.c b/tools/perf/util/top.c
index 500471d..e4370ca 100644
--- a/tools/perf/util/top.c
+++ b/tools/perf/util/top.c
@@ -75,6 +75,9 @@ size_t perf_top__header_snprintf(struct perf_top *top, char *bf, size_t size)
 	else if (top->target_tid != -1)
 		ret += SNPRINTF(bf + ret, size - ret, " (target_tid: %d",
 				top->target_tid);
+	else if (top->uid_str != NULL)
+		ret += SNPRINTF(bf + ret, size - ret, " (uid: %s",
+				top->uid_str);
 	else
 		ret += SNPRINTF(bf + ret, size - ret, " (all");
 
diff --git a/tools/perf/util/top.h b/tools/perf/util/top.h
index a248f3c..def3e53 100644
--- a/tools/perf/util/top.h
+++ b/tools/perf/util/top.h
@@ -24,6 +24,7 @@ struct perf_top {
 	int		   print_entries, count_filter, delay_secs;
 	int		   freq;
 	pid_t		   target_pid, target_tid;
+	uid_t		   uid;
 	bool		   hide_kernel_symbols, hide_user_symbols, zero;
 	bool		   system_wide;
 	bool		   use_tui, use_stdio;
@@ -45,6 +46,7 @@ struct perf_top {
 	int		   realtime_prio;
 	int		   sym_pcnt_filter;
 	const char	   *sym_filter;
+	const char	   *uid_str;
 };
 
 size_t perf_top__header_snprintf(struct perf_top *top, char *bf, size_t size);
diff --git a/tools/perf/util/ui/browsers/hists.c b/tools/perf/util/ui/browsers/hists.c
index 1212a38..7b6669d 100644
--- a/tools/perf/util/ui/browsers/hists.c
+++ b/tools/perf/util/ui/browsers/hists.c
@@ -841,6 +841,9 @@ static int hists__browser_title(struct hists *self, char *bf, size_t size,
 	nr_events = convert_unit(nr_events, &unit);
 	printed = snprintf(bf, size, "Events: %lu%c %s", nr_events, unit, ev_name);
 
+	if (self->uid_filter_str)
+		printed += snprintf(bf + printed, size - printed,
+				    ", UID: %s", self->uid_filter_str);
 	if (thread)
 		printed += snprintf(bf + printed, size - printed,
 				    ", Thread: %s(%d)",
diff --git a/tools/perf/util/usage.c b/tools/perf/util/usage.c
index d76d1c0..d0c0139 100644
--- a/tools/perf/util/usage.c
+++ b/tools/perf/util/usage.c
@@ -7,6 +7,7 @@
  * Copyright (C) Linus Torvalds, 2005
  */
 #include "util.h"
+#include "debug.h"
 
 static void report(const char *prefix, const char *err, va_list params)
 {
@@ -81,3 +82,41 @@ void warning(const char *warn, ...)
 	warn_routine(warn, params);
 	va_end(params);
 }
+
+uid_t parse_target_uid(const char *str, pid_t tid, pid_t pid)
+{
+	struct passwd pwd, *result;
+	char buf[1024];
+
+	if (str == NULL)
+		return UINT_MAX;
+
+	/* CPU and PID are mutually exclusive */
+	if (tid > 0 || pid > 0) {
+		ui__warning("PID/TID switch overriding UID\n");
+		sleep(1);
+		return UINT_MAX;
+	}
+
+	getpwnam_r(str, &pwd, buf, sizeof(buf), &result);
+
+	if (result == NULL) {
+		char *endptr;
+		int uid = strtol(str, &endptr, 10);
+
+		if (*endptr != '\0') {
+			ui__error("Invalid user %s\n", str);
+			return UINT_MAX - 1;
+		}
+
+		getpwuid_r(uid, &pwd, buf, sizeof(buf), &result);
+
+		if (result == NULL) {
+			ui__error("Problems obtaining information for user %s\n",
+				  str);
+			return UINT_MAX - 1;
+		}
+	}
+
+	return result->pw_uid;
+}
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index b9c530c..061dbf8 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -246,6 +246,8 @@ struct perf_event_attr;
 
 void event_attr_init(struct perf_event_attr *attr);
 
+uid_t parse_target_uid(const char *str, pid_t tid, pid_t pid);
+
 #define _STR(x) #x
 #define STR(x) _STR(x)
 
-- 
1.7.9.rc2.1.g69204

[PATCH 3/9] perf bench: Make "default" memcpy() selection actually use glibc's implementation

[Arnaldo Carvalho de Melo]

From: Jan Beulich <JBeulich@suse.com>

Since arch/x86/lib/memcpy_64.S implements not only __memcpy, but also
memcpy, without further precautions this function will get chose by the
static linker for resolving all references, and hence the "default"
measurement didn't really measure anything else than the
"x86-64-unrolled" one.

Fix this by renaming (through the pre-processor) the conflicting symbol.

On my Westmere system, the glibc variant turns out to require about 4%
less instructions, but 15% more cycles for the default 1Mb block size
measured.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4F16D6FD020000780006D72F@nat28.tlf.novell.com
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/bench/mem-memcpy-x86-64-asm.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/tools/perf/bench/mem-memcpy-x86-64-asm.S b/tools/perf/bench/mem-memcpy-x86-64-asm.S
index a57b66e..384b607 100644
--- a/tools/perf/bench/mem-memcpy-x86-64-asm.S
+++ b/tools/perf/bench/mem-memcpy-x86-64-asm.S
@@ -1,2 +1,2 @@
-
+#define memcpy MEMCPY /* don't hide glibc's memcpy() */
 #include "../../../arch/x86/lib/memcpy_64.S"
-- 
1.7.9.rc2.1.g69204

[PATCH 4/9] perf bench: Also allow measuring alternative memcpy implementations

[Arnaldo Carvalho de Melo]

From: Jan Beulich <JBeulich@suse.com>

Intended to be able to support the current selection of the preferred
memcpy() implementation, this patch adds the ability to also measure the
two alternative implementations, again by way of using some
pre-processsor replacement.

While on my Westmere system this proves that the movsb based variant is
worse than the movsq based one (since the ERMS feature isn't there), it
also shows that here for the default as well as small sizes the unrolled
variant outperforms the movsq one.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4F16D728020000780006D732@nat28.tlf.novell.com
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/bench/mem-memcpy-x86-64-asm-def.h |    8 ++++++++
 tools/perf/bench/mem-memcpy-x86-64-asm.S     |    4 ++++
 2 files changed, 12 insertions(+), 0 deletions(-)

diff --git a/tools/perf/bench/mem-memcpy-x86-64-asm-def.h b/tools/perf/bench/mem-memcpy-x86-64-asm-def.h
index d588b87..d66ab79 100644
--- a/tools/perf/bench/mem-memcpy-x86-64-asm-def.h
+++ b/tools/perf/bench/mem-memcpy-x86-64-asm-def.h
@@ -2,3 +2,11 @@
 MEMCPY_FN(__memcpy,
 	"x86-64-unrolled",
 	"unrolled memcpy() in arch/x86/lib/memcpy_64.S")
+
+MEMCPY_FN(memcpy_c,
+	"x86-64-movsq",
+	"movsq-based memcpy() in arch/x86/lib/memcpy_64.S")
+
+MEMCPY_FN(memcpy_c_e,
+	"x86-64-movsb",
+	"movsb-based memcpy() in arch/x86/lib/memcpy_64.S")
diff --git a/tools/perf/bench/mem-memcpy-x86-64-asm.S b/tools/perf/bench/mem-memcpy-x86-64-asm.S
index 384b607..a20780b 100644
--- a/tools/perf/bench/mem-memcpy-x86-64-asm.S
+++ b/tools/perf/bench/mem-memcpy-x86-64-asm.S
@@ -1,2 +1,6 @@
 #define memcpy MEMCPY /* don't hide glibc's memcpy() */
+#define altinstr_replacement text
+#define globl p2align 4; .globl
+#define Lmemcpy_c globl memcpy_c; memcpy_c
+#define Lmemcpy_c_e globl memcpy_c_e; memcpy_c_e
 #include "../../../arch/x86/lib/memcpy_64.S"
-- 
1.7.9.rc2.1.g69204

[PATCH 5/9] perf bench: Also allow measuring memset()

[Arnaldo Carvalho de Melo]

From: Jan Beulich <jbeulich@suse.com>

This simply clones the respective memcpy() implementation.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/4F16D743020000780006D735@nat28.tlf.novell.com
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Makefile                          |    4 +-
 tools/perf/bench/bench.h                     |    1 +
 tools/perf/bench/mem-memset-arch.h           |   12 +
 tools/perf/bench/mem-memset-x86-64-asm-def.h |   12 +
 tools/perf/bench/mem-memset-x86-64-asm.S     |    6 +
 tools/perf/bench/mem-memset.c                |  291 ++++++++++++++++++++++++++
 tools/perf/builtin-bench.c                   |    3 +
 tools/perf/util/include/asm/dwarf2.h         |    4 +-
 8 files changed, 331 insertions(+), 2 deletions(-)
 create mode 100644 tools/perf/bench/mem-memset-arch.h
 create mode 100644 tools/perf/bench/mem-memset-x86-64-asm-def.h
 create mode 100644 tools/perf/bench/mem-memset-x86-64-asm.S
 create mode 100644 tools/perf/bench/mem-memset.c

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index ac86d67..599031a 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -61,7 +61,7 @@ ifeq ($(ARCH),x86_64)
 	ifeq (${IS_X86_64}, 1)
 		RAW_ARCH := x86_64
 		ARCH_CFLAGS := -DARCH_X86_64
-		ARCH_INCLUDE = ../../arch/x86/lib/memcpy_64.S
+		ARCH_INCLUDE = ../../arch/x86/lib/memcpy_64.S ../../arch/x86/lib/memset_64.S
 	endif
 endif
 
@@ -362,8 +362,10 @@ BUILTIN_OBJS += $(OUTPUT)bench/sched-messaging.o
 BUILTIN_OBJS += $(OUTPUT)bench/sched-pipe.o
 ifeq ($(RAW_ARCH),x86_64)
 BUILTIN_OBJS += $(OUTPUT)bench/mem-memcpy-x86-64-asm.o
+BUILTIN_OBJS += $(OUTPUT)bench/mem-memset-x86-64-asm.o
 endif
 BUILTIN_OBJS += $(OUTPUT)bench/mem-memcpy.o
+BUILTIN_OBJS += $(OUTPUT)bench/mem-memset.o
 
 BUILTIN_OBJS += $(OUTPUT)builtin-diff.o
 BUILTIN_OBJS += $(OUTPUT)builtin-evlist.o
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index f7781c6..a09bece 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -4,6 +4,7 @@
 extern int bench_sched_messaging(int argc, const char **argv, const char *prefix);
 extern int bench_sched_pipe(int argc, const char **argv, const char *prefix);
 extern int bench_mem_memcpy(int argc, const char **argv, const char *prefix __used);
+extern int bench_mem_memset(int argc, const char **argv, const char *prefix);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
 #define BENCH_FORMAT_DEFAULT		0
diff --git a/tools/perf/bench/mem-memset-arch.h b/tools/perf/bench/mem-memset-arch.h
new file mode 100644
index 0000000..a040fa7
--- /dev/null
+++ b/tools/perf/bench/mem-memset-arch.h
@@ -0,0 +1,12 @@
+
+#ifdef ARCH_X86_64
+
+#define MEMSET_FN(fn, name, desc)		\
+	extern void *fn(void *, int, size_t);
+
+#include "mem-memset-x86-64-asm-def.h"
+
+#undef MEMSET_FN
+
+#endif
+
diff --git a/tools/perf/bench/mem-memset-x86-64-asm-def.h b/tools/perf/bench/mem-memset-x86-64-asm-def.h
new file mode 100644
index 0000000..a71dff9
--- /dev/null
+++ b/tools/perf/bench/mem-memset-x86-64-asm-def.h
@@ -0,0 +1,12 @@
+
+MEMSET_FN(__memset,
+	"x86-64-unrolled",
+	"unrolled memset() in arch/x86/lib/memset_64.S")
+
+MEMSET_FN(memset_c,
+	"x86-64-stosq",
+	"movsq-based memset() in arch/x86/lib/memset_64.S")
+
+MEMSET_FN(memset_c_e,
+	"x86-64-stosb",
+	"movsb-based memset() in arch/x86/lib/memset_64.S")
diff --git a/tools/perf/bench/mem-memset-x86-64-asm.S b/tools/perf/bench/mem-memset-x86-64-asm.S
new file mode 100644
index 0000000..cb92170
--- /dev/null
+++ b/tools/perf/bench/mem-memset-x86-64-asm.S
@@ -0,0 +1,6 @@
+#define memset MEMSET /* don't hide glibc's memset() */
+#define altinstr_replacement text
+#define globl p2align 4; .globl
+#define Lmemset_c globl memset_c; memset_c
+#define Lmemset_c_e globl memset_c_e; memset_c_e
+#include "../../../arch/x86/lib/memset_64.S"
diff --git a/tools/perf/bench/mem-memset.c b/tools/perf/bench/mem-memset.c
new file mode 100644
index 0000000..9c0c6f0
--- /dev/null
+++ b/tools/perf/bench/mem-memset.c
@@ -0,0 +1,291 @@
+/*
+ * mem-memset.c
+ *
+ * memset: Simple memory set in various ways
+ *
+ * Trivial clone of mem-memcpy.c.
+ */
+#include <ctype.h>
+
+#include "../perf.h"
+#include "../util/util.h"
+#include "../util/parse-options.h"
+#include "../util/header.h"
+#include "bench.h"
+#include "mem-memset-arch.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/time.h>
+#include <errno.h>
+
+#define K 1024
+
+static const char	*length_str	= "1MB";
+static const char	*routine	= "default";
+static bool		use_clock;
+static int		clock_fd;
+static bool		only_prefault;
+static bool		no_prefault;
+
+static const struct option options[] = {
+	OPT_STRING('l', "length", &length_str, "1MB",
+		    "Specify length of memory to copy. "
+		    "available unit: B, MB, GB (upper and lower)"),
+	OPT_STRING('r', "routine", &routine, "default",
+		    "Specify routine to copy"),
+	OPT_BOOLEAN('c', "clock", &use_clock,
+		    "Use CPU clock for measuring"),
+	OPT_BOOLEAN('o', "only-prefault", &only_prefault,
+		    "Show only the result with page faults before memset()"),
+	OPT_BOOLEAN('n', "no-prefault", &no_prefault,
+		    "Show only the result without page faults before memset()"),
+	OPT_END()
+};
+
+typedef void *(*memset_t)(void *, int, size_t);
+
+struct routine {
+	const char *name;
+	const char *desc;
+	memset_t fn;
+};
+
+static const struct routine routines[] = {
+	{ "default",
+	  "Default memset() provided by glibc",
+	  memset },
+#ifdef ARCH_X86_64
+
+#define MEMSET_FN(fn, name, desc) { name, desc, fn },
+#include "mem-memset-x86-64-asm-def.h"
+#undef MEMSET_FN
+
+#endif
+
+	{ NULL,
+	  NULL,
+	  NULL   }
+};
+
+static const char * const bench_mem_memset_usage[] = {
+	"perf bench mem memset <options>",
+	NULL
+};
+
+static struct perf_event_attr clock_attr = {
+	.type		= PERF_TYPE_HARDWARE,
+	.config		= PERF_COUNT_HW_CPU_CYCLES
+};
+
+static void init_clock(void)
+{
+	clock_fd = sys_perf_event_open(&clock_attr, getpid(), -1, -1, 0);
+
+	if (clock_fd < 0 && errno == ENOSYS)
+		die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
+	else
+		BUG_ON(clock_fd < 0);
+}
+
+static u64 get_clock(void)
+{
+	int ret;
+	u64 clk;
+
+	ret = read(clock_fd, &clk, sizeof(u64));
+	BUG_ON(ret != sizeof(u64));
+
+	return clk;
+}
+
+static double timeval2double(struct timeval *ts)
+{
+	return (double)ts->tv_sec +
+		(double)ts->tv_usec / (double)1000000;
+}
+
+static void alloc_mem(void **dst, size_t length)
+{
+	*dst = zalloc(length);
+	if (!dst)
+		die("memory allocation failed - maybe length is too large?\n");
+}
+
+static u64 do_memset_clock(memset_t fn, size_t len, bool prefault)
+{
+	u64 clock_start = 0ULL, clock_end = 0ULL;
+	void *dst = NULL;
+
+	alloc_mem(&dst, len);
+
+	if (prefault)
+		fn(dst, -1, len);
+
+	clock_start = get_clock();
+	fn(dst, 0, len);
+	clock_end = get_clock();
+
+	free(dst);
+	return clock_end - clock_start;
+}
+
+static double do_memset_gettimeofday(memset_t fn, size_t len, bool prefault)
+{
+	struct timeval tv_start, tv_end, tv_diff;
+	void *dst = NULL;
+
+	alloc_mem(&dst, len);
+
+	if (prefault)
+		fn(dst, -1, len);
+
+	BUG_ON(gettimeofday(&tv_start, NULL));
+	fn(dst, 0, len);
+	BUG_ON(gettimeofday(&tv_end, NULL));
+
+	timersub(&tv_end, &tv_start, &tv_diff);
+
+	free(dst);
+	return (double)((double)len / timeval2double(&tv_diff));
+}
+
+#define pf (no_prefault ? 0 : 1)
+
+#define print_bps(x) do {					\
+		if (x < K)					\
+			printf(" %14lf B/Sec", x);		\
+		else if (x < K * K)				\
+			printf(" %14lfd KB/Sec", x / K);	\
+		else if (x < K * K * K)				\
+			printf(" %14lf MB/Sec", x / K / K);	\
+		else						\
+			printf(" %14lf GB/Sec", x / K / K / K); \
+	} while (0)
+
+int bench_mem_memset(int argc, const char **argv,
+		     const char *prefix __used)
+{
+	int i;
+	size_t len;
+	double result_bps[2];
+	u64 result_clock[2];
+
+	argc = parse_options(argc, argv, options,
+			     bench_mem_memset_usage, 0);
+
+	if (use_clock)
+		init_clock();
+
+	len = (size_t)perf_atoll((char *)length_str);
+
+	result_clock[0] = result_clock[1] = 0ULL;
+	result_bps[0] = result_bps[1] = 0.0;
+
+	if ((s64)len <= 0) {
+		fprintf(stderr, "Invalid length:%s\n", length_str);
+		return 1;
+	}
+
+	/* same to without specifying either of prefault and no-prefault */
+	if (only_prefault && no_prefault)
+		only_prefault = no_prefault = false;
+
+	for (i = 0; routines[i].name; i++) {
+		if (!strcmp(routines[i].name, routine))
+			break;
+	}
+	if (!routines[i].name) {
+		printf("Unknown routine:%s\n", routine);
+		printf("Available routines...\n");
+		for (i = 0; routines[i].name; i++) {
+			printf("\t%s ... %s\n",
+			       routines[i].name, routines[i].desc);
+		}
+		return 1;
+	}
+
+	if (bench_format == BENCH_FORMAT_DEFAULT)
+		printf("# Copying %s Bytes ...\n\n", length_str);
+
+	if (!only_prefault && !no_prefault) {
+		/* show both of results */
+		if (use_clock) {
+			result_clock[0] =
+				do_memset_clock(routines[i].fn, len, false);
+			result_clock[1] =
+				do_memset_clock(routines[i].fn, len, true);
+		} else {
+			result_bps[0] =
+				do_memset_gettimeofday(routines[i].fn,
+						len, false);
+			result_bps[1] =
+				do_memset_gettimeofday(routines[i].fn,
+						len, true);
+		}
+	} else {
+		if (use_clock) {
+			result_clock[pf] =
+				do_memset_clock(routines[i].fn,
+						len, only_prefault);
+		} else {
+			result_bps[pf] =
+				do_memset_gettimeofday(routines[i].fn,
+						len, only_prefault);
+		}
+	}
+
+	switch (bench_format) {
+	case BENCH_FORMAT_DEFAULT:
+		if (!only_prefault && !no_prefault) {
+			if (use_clock) {
+				printf(" %14lf Clock/Byte\n",
+					(double)result_clock[0]
+					/ (double)len);
+				printf(" %14lf Clock/Byte (with prefault)\n ",
+					(double)result_clock[1]
+					/ (double)len);
+			} else {
+				print_bps(result_bps[0]);
+				printf("\n");
+				print_bps(result_bps[1]);
+				printf(" (with prefault)\n");
+			}
+		} else {
+			if (use_clock) {
+				printf(" %14lf Clock/Byte",
+					(double)result_clock[pf]
+					/ (double)len);
+			} else
+				print_bps(result_bps[pf]);
+
+			printf("%s\n", only_prefault ? " (with prefault)" : "");
+		}
+		break;
+	case BENCH_FORMAT_SIMPLE:
+		if (!only_prefault && !no_prefault) {
+			if (use_clock) {
+				printf("%lf %lf\n",
+					(double)result_clock[0] / (double)len,
+					(double)result_clock[1] / (double)len);
+			} else {
+				printf("%lf %lf\n",
+					result_bps[0], result_bps[1]);
+			}
+		} else {
+			if (use_clock) {
+				printf("%lf\n", (double)result_clock[pf]
+					/ (double)len);
+			} else
+				printf("%lf\n", result_bps[pf]);
+		}
+		break;
+	default:
+		/* reaching this means there's some disaster: */
+		die("unknown format: %d\n", bench_format);
+		break;
+	}
+
+	return 0;
+}
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index fcb9626..b0e74ab 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -52,6 +52,9 @@ static struct bench_suite mem_suites[] = {
 	{ "memcpy",
 	  "Simple memory copy in various ways",
 	  bench_mem_memcpy },
+	{ "memset",
+	  "Simple memory set in various ways",
+	  bench_mem_memset },
 	suite_all,
 	{ NULL,
 	  NULL,
diff --git a/tools/perf/util/include/asm/dwarf2.h b/tools/perf/util/include/asm/dwarf2.h
index bb4198e..afe3819 100644
--- a/tools/perf/util/include/asm/dwarf2.h
+++ b/tools/perf/util/include/asm/dwarf2.h
@@ -2,10 +2,12 @@
 #ifndef PERF_DWARF2_H
 #define PERF_DWARF2_H
 
-/* dwarf2.h ... dummy header file for including arch/x86/lib/memcpy_64.S */
+/* dwarf2.h ... dummy header file for including arch/x86/lib/mem{cpy,set}_64.S */
 
 #define CFI_STARTPROC
 #define CFI_ENDPROC
+#define CFI_REMEMBER_STATE
+#define CFI_RESTORE_STATE
 
 #endif	/* PERF_DWARF2_H */
 
-- 
1.7.9.rc2.1.g69204

[PATCH 6/9] perf bench: Allow passing an iteration count to "bench mem mem{cpy,set}"

[Arnaldo Carvalho de Melo]

From: Jan Beulich <JBeulich@suse.com>

"perf stat ... perf bench mem mem..." is pretty meaningless when using
small block sizes (as the overhead of the invocation of each test run
basically hides the actual test result in the noise). Repeating the
actually interesting function's invocation a number of times allows the
results to become meaningful.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4F16D767020000780006D738@nat28.tlf.novell.com
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/bench/mem-memcpy.c |   11 +++++++++--
 tools/perf/bench/mem-memset.c |   11 +++++++++--
 2 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/tools/perf/bench/mem-memcpy.c b/tools/perf/bench/mem-memcpy.c
index db82021..6ad2b1c 100644
--- a/tools/perf/bench/mem-memcpy.c
+++ b/tools/perf/bench/mem-memcpy.c
@@ -24,6 +24,7 @@
 
 static const char	*length_str	= "1MB";
 static const char	*routine	= "default";
+static int		iterations	= 1;
 static bool		use_clock;
 static int		clock_fd;
 static bool		only_prefault;
@@ -35,6 +36,8 @@ static const struct option options[] = {
 		    "available unit: B, MB, GB (upper and lower)"),
 	OPT_STRING('r', "routine", &routine, "default",
 		    "Specify routine to copy"),
+	OPT_INTEGER('i', "iterations", &iterations,
+		    "repeat memcpy() invocation this number of times"),
 	OPT_BOOLEAN('c', "clock", &use_clock,
 		    "Use CPU clock for measuring"),
 	OPT_BOOLEAN('o', "only-prefault", &only_prefault,
@@ -121,6 +124,7 @@ static u64 do_memcpy_clock(memcpy_t fn, size_t len, bool prefault)
 {
 	u64 clock_start = 0ULL, clock_end = 0ULL;
 	void *src = NULL, *dst = NULL;
+	int i;
 
 	alloc_mem(&src, &dst, len);
 
@@ -128,7 +132,8 @@ static u64 do_memcpy_clock(memcpy_t fn, size_t len, bool prefault)
 		fn(dst, src, len);
 
 	clock_start = get_clock();
-	fn(dst, src, len);
+	for (i = 0; i < iterations; ++i)
+		fn(dst, src, len);
 	clock_end = get_clock();
 
 	free(src);
@@ -140,6 +145,7 @@ static double do_memcpy_gettimeofday(memcpy_t fn, size_t len, bool prefault)
 {
 	struct timeval tv_start, tv_end, tv_diff;
 	void *src = NULL, *dst = NULL;
+	int i;
 
 	alloc_mem(&src, &dst, len);
 
@@ -147,7 +153,8 @@ static double do_memcpy_gettimeofday(memcpy_t fn, size_t len, bool prefault)
 		fn(dst, src, len);
 
 	BUG_ON(gettimeofday(&tv_start, NULL));
-	fn(dst, src, len);
+	for (i = 0; i < iterations; ++i)
+		fn(dst, src, len);
 	BUG_ON(gettimeofday(&tv_end, NULL));
 
 	timersub(&tv_end, &tv_start, &tv_diff);
diff --git a/tools/perf/bench/mem-memset.c b/tools/perf/bench/mem-memset.c
index 9c0c6f0..59d4933 100644
--- a/tools/perf/bench/mem-memset.c
+++ b/tools/perf/bench/mem-memset.c
@@ -24,6 +24,7 @@
 
 static const char	*length_str	= "1MB";
 static const char	*routine	= "default";
+static int		iterations	= 1;
 static bool		use_clock;
 static int		clock_fd;
 static bool		only_prefault;
@@ -35,6 +36,8 @@ static const struct option options[] = {
 		    "available unit: B, MB, GB (upper and lower)"),
 	OPT_STRING('r', "routine", &routine, "default",
 		    "Specify routine to copy"),
+	OPT_INTEGER('i', "iterations", &iterations,
+		    "repeat memset() invocation this number of times"),
 	OPT_BOOLEAN('c', "clock", &use_clock,
 		    "Use CPU clock for measuring"),
 	OPT_BOOLEAN('o', "only-prefault", &only_prefault,
@@ -117,6 +120,7 @@ static u64 do_memset_clock(memset_t fn, size_t len, bool prefault)
 {
 	u64 clock_start = 0ULL, clock_end = 0ULL;
 	void *dst = NULL;
+	int i;
 
 	alloc_mem(&dst, len);
 
@@ -124,7 +128,8 @@ static u64 do_memset_clock(memset_t fn, size_t len, bool prefault)
 		fn(dst, -1, len);
 
 	clock_start = get_clock();
-	fn(dst, 0, len);
+	for (i = 0; i < iterations; ++i)
+		fn(dst, i, len);
 	clock_end = get_clock();
 
 	free(dst);
@@ -135,6 +140,7 @@ static double do_memset_gettimeofday(memset_t fn, size_t len, bool prefault)
 {
 	struct timeval tv_start, tv_end, tv_diff;
 	void *dst = NULL;
+	int i;
 
 	alloc_mem(&dst, len);
 
@@ -142,7 +148,8 @@ static double do_memset_gettimeofday(memset_t fn, size_t len, bool prefault)
 		fn(dst, -1, len);
 
 	BUG_ON(gettimeofday(&tv_start, NULL));
-	fn(dst, 0, len);
+	for (i = 0; i < iterations; ++i)
+		fn(dst, i, len);
 	BUG_ON(gettimeofday(&tv_end, NULL));
 
 	timersub(&tv_end, &tv_start, &tv_diff);
-- 
1.7.9.rc2.1.g69204

[PATCH 7/9] perf tools: Fix broken build by defining _GNU_SOURCE in Makefile

[Arnaldo Carvalho de Melo]

From: David Daney <david.daney@cavium.com>

When building on my Debian/mips system, util/util.c fails to build
because commit 1aed2671738785e8f5aea663a6fda91aa7ef59b5 (perf kvm: Do
guest-only counting by default) indirectly includes stdio.h before the
feature selection in util.h is done.  This prevents _GNU_SOURCE in
util.h from enabling the declaration of getline(), from now second
inclusion of stdio.h, and the build is broken.

There is another breakage in util/evsel.c caused by include ordering,
but I didn't fully track down the commit that caused it.

The root cause of all this is an inconsistent definition of _GNU_SOURCE,
so I move the definition into the Makefile so that it is passed to all
invocations of the compiler and used uniformly for all system header
files.  All other #define and #undef of _GNU_SOURCE are removed as they
cause conflicts with the definition passed to the compiler.

All the features.h definitions (_LARGEFILE64_SOURCE _FILE_OFFSET_BITS=64
and _GNU_SOURCE) are needed by the python glue code too, so they are
moved to BASIC_CFLAGS, and the misleading comments about BASIC_CFLAGS
are removed.

This gives me a clean build on x86_64 (fc12) and mips (Debian).

Cc: David Daney <david.daney@cavium.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Joerg Roedel <joerg.roedel@amd.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1326836461-11952-1-git-send-email-ddaney.cavm@gmail.com
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Makefile                 |    7 ++-----
 tools/perf/builtin-probe.c          |    2 --
 tools/perf/util/probe-event.c       |    2 --
 tools/perf/util/symbol.c            |    1 -
 tools/perf/util/trace-event-parse.c |    3 +--
 tools/perf/util/ui/browsers/hists.c |    2 --
 tools/perf/util/ui/helpline.c       |    1 -
 tools/perf/util/util.h              |    1 -
 8 files changed, 3 insertions(+), 16 deletions(-)

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 599031a..d64f581 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -104,7 +104,7 @@ endif
 
 CFLAGS = -fno-omit-frame-pointer -ggdb3 -Wall -Wextra -std=gnu99 $(CFLAGS_WERROR) $(CFLAGS_OPTIMIZE) -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS) $(EXTRA_CFLAGS)
 EXTLIBS = -lpthread -lrt -lelf -lm
-ALL_CFLAGS = $(CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64
+ALL_CFLAGS = $(CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE
 ALL_LDFLAGS = $(LDFLAGS)
 STRIP ?= strip
 
@@ -168,10 +168,7 @@ endif
 
 ### --- END CONFIGURATION SECTION ---
 
-# Those must not be GNU-specific; they are shared with perl/ which may
-# be built by a different compiler. (Note that this is an artifact now
-# but it still might be nice to keep that distinction.)
-BASIC_CFLAGS = -Iutil/include -Iarch/$(ARCH)/include
+BASIC_CFLAGS = -Iutil/include -Iarch/$(ARCH)/include -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE
 BASIC_LDFLAGS =
 
 # Guard against environment variables
diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
index 59d43ab..fb85661 100644
--- a/tools/perf/builtin-probe.c
+++ b/tools/perf/builtin-probe.c
@@ -20,7 +20,6 @@
  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  *
  */
-#define _GNU_SOURCE
 #include <sys/utsname.h>
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -31,7 +30,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#undef _GNU_SOURCE
 #include "perf.h"
 #include "builtin.h"
 #include "util/util.h"
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index eb25900..29cb654 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -19,7 +19,6 @@
  *
  */
 
-#define _GNU_SOURCE
 #include <sys/utsname.h>
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -33,7 +32,6 @@
 #include <limits.h>
 #include <elf.h>
 
-#undef _GNU_SOURCE
 #include "util.h"
 #include "event.h"
 #include "string.h"
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 215d50f..0975438 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -1,4 +1,3 @@
-#define _GNU_SOURCE
 #include <ctype.h>
 #include <dirent.h>
 #include <errno.h>
diff --git a/tools/perf/util/trace-event-parse.c b/tools/perf/util/trace-event-parse.c
index 6c164dc..1a8d4dc 100644
--- a/tools/perf/util/trace-event-parse.c
+++ b/tools/perf/util/trace-event-parse.c
@@ -21,14 +21,13 @@
  *  The parts for function graph printing was taken and modified from the
  *  Linux Kernel that were written by Frederic Weisbecker.
  */
-#define _GNU_SOURCE
+
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <ctype.h>
 #include <errno.h>
 
-#undef _GNU_SOURCE
 #include "../perf.h"
 #include "util.h"
 #include "trace-event.h"
diff --git a/tools/perf/util/ui/browsers/hists.c b/tools/perf/util/ui/browsers/hists.c
index 7b6669d..bfba049 100644
--- a/tools/perf/util/ui/browsers/hists.c
+++ b/tools/perf/util/ui/browsers/hists.c
@@ -1,6 +1,4 @@
-#define _GNU_SOURCE
 #include <stdio.h>
-#undef _GNU_SOURCE
 #include "../libslang.h"
 #include <stdlib.h>
 #include <string.h>
diff --git a/tools/perf/util/ui/helpline.c b/tools/perf/util/ui/helpline.c
index 6ef3c56..4f48f59 100644
--- a/tools/perf/util/ui/helpline.c
+++ b/tools/perf/util/ui/helpline.c
@@ -1,4 +1,3 @@
-#define _GNU_SOURCE
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index 061dbf8..232d17e 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -40,7 +40,6 @@
 #define decimal_length(x)	((int)(sizeof(x) * 2.56 + 0.5) + 1)
 
 #define _ALL_SOURCE 1
-#define _GNU_SOURCE 1
 #define _BSD_SOURCE 1
 #define HAS_BOOL
 
-- 
1.7.9.rc2.1.g69204

[PATCH 8/9] perf probe: Usability fixes

[Arnaldo Carvalho de Melo]

From: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

Ingo pointed out few perf probe usability related errors during his
review of uprobes.

Since these issues are independent of uprobes, fixing them in a separate
patch.

Suggested-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Link: http://lkml.kernel.org/r/20120120121354.GL15447@linux.vnet.ibm.com
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/probe-event.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index 29cb654..b9bbdd2 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -1729,7 +1729,7 @@ static int __add_probe_trace_events(struct perf_probe_event *pev,
 	}
 
 	ret = 0;
-	printf("Add new event%s\n", (ntevs > 1) ? "s:" : ":");
+	printf("Added new event%s\n", (ntevs > 1) ? "s:" : ":");
 	for (i = 0; i < ntevs; i++) {
 		tev = &tevs[i];
 		if (pev->event)
@@ -1784,7 +1784,7 @@ static int __add_probe_trace_events(struct perf_probe_event *pev,
 
 	if (ret >= 0) {
 		/* Show how to use the event. */
-		printf("\nYou can now use it on all perf tools, such as:\n\n");
+		printf("\nYou can now use it in all perf tools, such as:\n\n");
 		printf("\tperf record -e %s:%s -aR sleep 1\n\n", tev->group,
 			 tev->event);
 	}
@@ -1959,7 +1959,7 @@ static int __del_trace_probe_event(int fd, struct str_node *ent)
 		goto error;
 	}
 
-	printf("Remove event: %s\n", ent->s);
+	printf("Removed event: %s\n", ent->s);
 	return 0;
 error:
 	pr_warning("Failed to delete event: %s\n", strerror(-ret));
-- 
1.7.9.rc2.1.g69204

[PATCH 9/9] perf tools: Fix strlen() bug in perf_event__synthesize_event_type()

[Arnaldo Carvalho de Melo]

From: Stephane Eranian <eranian@google.com>

The event_type record has a max length for the event name.

It's called MAX_EVENT_NAME.

The name may be truncated to fit the max length. But the header.size still
reflects the original name length. If that length is > MAX_EVENT_NAME, then the
header.size field is bogus. Fix this by using the length of the name after the
potential truncation.

Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20120120094912.GA4882@quad
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/header.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 3e7e0b0..ecd7f4d 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -2105,7 +2105,7 @@ int perf_event__synthesize_event_type(struct perf_tool *tool,
 	strncpy(ev.event_type.event_type.name, name, MAX_EVENT_NAME - 1);
 
 	ev.event_type.header.type = PERF_RECORD_HEADER_EVENT_TYPE;
-	size = strlen(name);
+	size = strlen(ev.event_type.event_type.name);
 	size = ALIGN(size, sizeof(u64));
 	ev.event_type.header.size = sizeof(ev.event_type) -
 		(sizeof(ev.event_type.event_type.name) - size);
-- 
1.7.9.rc2.1.g69204

Re: [GIT PULL 0/9] perf/core improvements and fixes

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@infradead.org> wrote:

> Hi Ingo,
> 
> 	This is a signed tag, please lemme know if everything went well.
> 
> 	The --uid feature works for root, we still need to 
> sort out that paranoia with some threads owned by a user that 
> prevents 'perf --uid non-root-user' to work for 
> 'non-root-user'.

Just wondering what detail causes that failure - the whole point 
of --uid mingo would be to enable nonprivileged users to do 
'session wide' profiling, *especially* if paranoia is high.

So what does --uid do which perf record --pid 1234 wouldnt 
already do? By all means --uid ought to be a fancy way of doing 
a whole bunch of perf record --pid 1234 profiling sessions, at 
once.

[ Btw, we should probably alias --user to --uid as well, as that 
  might be the intuitive thing people would typically use? ]

> 	http://git.kernel.org/?p=linux/kernel/git/acme/linux.git;a=tag;h=ce9600c4c664ce7f97e8aa5e756b0b4ea5b017c7
> looks ok to me, need just to improve on the commit log message, I'll get
> used to it :-)
> 
> - Arnaldo
> 
> The following changes since commit 172d1b0b73256551f100fc00c69e356d047103f5:
> 
>   perf tools: Fix compile error on x86_64 Ubuntu (2012-01-08 13:34:55 -0200)
> 
> are available in the git repository at:
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux perf-core-for-mingo
> 
> for you to fetch changes up to f8f4b2872295dca88339ec0c403b2217b1197353:
> 
>   perf tools: Fix strlen() bug in perf_event__synthesize_event_type() (2012-01-24 20:31:34 -0200)
> 
> ----------------------------------------------------------------
> perf/core improvements and fixes
> 
> ----------------------------------------------------------------

Anyway, pulled, thanks a lot Arnaldo!

One detail: don't we want some of these fixes cherry-picked into 
perf/urgent as well?

Thanks,

	Ingo

Fixing perf top --user shortcoming was: Re: [GIT PULL 0/9] perf/core improvements and fixes

[Arnaldo Carvalho de Melo]

Em Thu, Jan 26, 2012 at 12:16:48PM +0100, Ingo Molnar escreveu:
> * Arnaldo Carvalho de Melo <acme@infradead.org> wrote:
> > 	The --uid feature works for root, we still need to 
> > sort out that paranoia with some threads owned by a user that 
> > prevents 'perf --uid non-root-user' to work for 
> > 'non-root-user'.

> Just wondering what detail causes that failure - the whole point 
> of --uid mingo would be to enable nonprivileged users to do 
> 'session wide' profiling, *especially* if paranoia is high.

> So what does --uid do which perf record --pid 1234 wouldnt 
> already do? By all means --uid ought to be a fancy way of doing 
> a whole bunch of perf record --pid 1234 profiling sessions, at 
> once.

I stopped at the kernel, i.e. used what can be done with what is
available from the kernel right now, the diagnosis was sent in private,
but boils down to:

+++ b/kernel/events/core.c
@@ -2636,7 +2636,8 @@ find_lively_task_by_vpid(pid_t vpid)
 
 	/* Reuse ptrace permission checks for now. */
 	err = -EACCES;
-	if (!ptrace_may_access(task, PTRACE_MODE_READ))
+	if (perf_paranoid_tracepoint_raw() &&
+	    !ptrace_may_access(task, PTRACE_MODE_READ))
 		goto errout;
 
 	return task;

ptrace_may_access(task, PTRACE_MODE_READ) fails for some tasks owned by
the user because, IIRC, in __ptrace_may_access:

        const struct cred *cred = current_cred(), *tcred;

        /* May we inspect the given task?
         * This check is used both for attaching with ptrace
         * and for allowing access to sensitive information in /proc.
         *
         * ptrace_attach denies several cases that /proc allows
         * because setting up the necessary parent/child relationship
         * or halting the specified task is impossible.
         */
        int dumpable = 0;
<SNIP>
        if (!dumpable && !task_ns_capable(task, CAP_SYS_PTRACE))
                return -EPERM;

fails.

The patch above is not any kind of solution, just a way to make it work
when paranoia is set to -1 (thus perf_paranoid_tracepoint_raw in the
POC) and show where the problem lies, ideas? Peter?

> [ Btw, we should probably alias --user to --uid as well, as that 
>   might be the intuitive thing people would typically use? ]

I'll do that
 
> Anyway, pulled, thanks a lot Arnaldo!
> 
> One detail: don't we want some of these fixes cherry-picked into 
> perf/urgent as well?

Yeah, I'll prepare a perf-urgent-for-ingo signed tag.
 
> Thanks,
> 
> 	Ingo

Re: Fixing perf top --user shortcoming was: Re: [GIT PULL 0/9] perf/core improvements and fixes

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@infradead.org> wrote:

> > So what does --uid do which perf record --pid 1234 wouldnt 
> > already do? By all means --uid ought to be a fancy way of 
> > doing a whole bunch of perf record --pid 1234 profiling 
> > sessions, at once.
> 
> I stopped at the kernel, i.e. used what can be done with what 
> is available from the kernel right now, the diagnosis was sent 
> in private, but boils down to:
> 
> +++ b/kernel/events/core.c
> @@ -2636,7 +2636,8 @@ find_lively_task_by_vpid(pid_t vpid)
>  
>  	/* Reuse ptrace permission checks for now. */
>  	err = -EACCES;
> -	if (!ptrace_may_access(task, PTRACE_MODE_READ))
> +	if (perf_paranoid_tracepoint_raw() &&
> +	    !ptrace_may_access(task, PTRACE_MODE_READ))
>  		goto errout;
>  
>  	return task;
> 
> ptrace_may_access(task, PTRACE_MODE_READ) fails for some tasks 
> owned by the user because, IIRC, in __ptrace_may_access:

Which tasks are these, are they privileged in any sense?

If yes and if most of the 'real' tasks a user have can be 
profiled just fine then i think we should just skip the 
privileged tasks and not abort the profiling session?

Thanks,

	Ingo

Re: Fixing perf top --user shortcoming was: Re: [GIT PULL 0/9] perf/core improvements and fixes

[Arnaldo Carvalho de Melo]

Em Thu, Jan 26, 2012 at 02:09:19PM +0100, Ingo Molnar escreveu:
> * Arnaldo Carvalho de Melo <acme@infradead.org> wrote:
> > > So what does --uid do which perf record --pid 1234 wouldnt 
> > > already do? By all means --uid ought to be a fancy way of 
> > > doing a whole bunch of perf record --pid 1234 profiling 
> > > sessions, at once.

> > I stopped at the kernel, i.e. used what can be done with what 
> > is available from the kernel right now, the diagnosis was sent 
> > in private, but boils down to:

> > +++ b/kernel/events/core.c
> > @@ -2636,7 +2636,8 @@ find_lively_task_by_vpid(pid_t vpid)
> >  	/* Reuse ptrace permission checks for now. */
> >  	err = -EACCES;
> > -	if (!ptrace_may_access(task, PTRACE_MODE_READ))
> > +	if (perf_paranoid_tracepoint_raw() &&
> > +	    !ptrace_may_access(task, PTRACE_MODE_READ))
> >  		goto errout;
> >  	return task;

> > ptrace_may_access(task, PTRACE_MODE_READ) fails for some tasks 
> > owned by the user because, IIRC, in __ptrace_may_access:

> Which tasks are these, are they privileged in any sense?

IIRC one of them was a child of sshd, that runs as root and then changes
the child ownership to the user logging in.

I'll continue investigation but probably for now the first thing to do
is to just remove them from the thread_map when they return -EPERM.
 
> If yes and if most of the 'real' tasks a user have can be 
> profiled just fine then i think we should just skip the 
> privileged tasks and not abort the profiling session?

Yeah, that can be done, while debugging I'll emit a warning with the
resulting thread_map of "special tasks" to figure out what makes them
special.

- Arnaldo

Re: Fixing perf top --user shortcoming was: Re: [GIT PULL 0/9] perf/core improvements and fixes

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@infradead.org> wrote:

> > > ptrace_may_access(task, PTRACE_MODE_READ) fails for some 
> > > tasks owned by the user because, IIRC, in 
> > > __ptrace_may_access:
> 
> > Which tasks are these, are they privileged in any sense?
> 
> IIRC one of them was a child of sshd, that runs as root and 
> then changes the child ownership to the user logging in.

It's probably privileged then - or at least not sufficiently 
deprivileged.

Skipping them ought to be the right solution - it's not like 
such tasks tend to create a lot of overhead worth profiling. 
They are also not debuggable via gdb so they are not part of the 
user's development session and such.

> I'll continue investigation but probably for now the first 
> thing to do is to just remove them from the thread_map when 
> they return -EPERM.

Yeah. Maybe warn about them in verbose mode or such.

> > If yes and if most of the 'real' tasks a user have can be 
> > profiled just fine then i think we should just skip the 
> > privileged tasks and not abort the profiling session?
> 
> Yeah, that can be done, while debugging I'll emit a warning 
> with the resulting thread_map of "special tasks" to figure out 
> what makes them special.

Ok, sounds great!

Thanks,

	Ingo

Re: Fixing perf top --user shortcoming was: Re: [GIT PULL 0/9] perf/core improvements and fixes

[Arnaldo Carvalho de Melo]

Em Thu, Jan 26, 2012 at 07:32:23PM +0100, Ingo Molnar escreveu:
> 
> * Arnaldo Carvalho de Melo <acme@infradead.org> wrote:
> 
> > > > ptrace_may_access(task, PTRACE_MODE_READ) fails for some 
> > > > tasks owned by the user because, IIRC, in 
> > > > __ptrace_may_access:
> > 
> > > Which tasks are these, are they privileged in any sense?
> > 
> > IIRC one of them was a child of sshd, that runs as root and 
> > then changes the child ownership to the user logging in.
> 
> It's probably privileged then - or at least not sufficiently 
> deprivileged.
> 
> Skipping them ought to be the right solution - it's not like 
> such tasks tend to create a lot of overhead worth profiling. 
> They are also not debuggable via gdb so they are not part of the 
> user's development session and such.

Least convoluted way seems to be just to try to call sys_perf_event_open
on then when building the thread list in thread_new__by_uid(), will try
that after soccer :-)

/me vanishes

- Arnaldo

> > I'll continue investigation but probably for now the first 
> > thing to do is to just remove them from the thread_map when 
> > they return -EPERM.
> 
> Yeah. Maybe warn about them in verbose mode or such.