[GIT PULL] Perf updates

[GIT PULL] Perf updates

[Frederic Weisbecker]

Ingo,

Please pull the perf/core branch that can be found at:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
	perf/core

Thanks,
	Frederic
---

Frederic Weisbecker (3):
      perf lock: Drop the buffers multiplexing dependency
      perf: Remove pointless breakpoint union
      x86/hw-breakpoints: Remove the name field

Hitoshi Mitake (1):
      perf lock: Fix and add misc documentally things

Tejun Heo (1):
      percpu: Add __percpu sparse annotations to hw_breakpoint


 arch/x86/include/asm/hw_breakpoint.h    |    1 -
 arch/x86/kernel/hw_breakpoint.c         |    7 --
 include/linux/hw_breakpoint.h           |    8 +-
 include/linux/perf_event.h              |    5 +-
 kernel/hw_breakpoint.c                  |   10 +-
 lib/Kconfig.debug                       |    8 ++
 samples/hw_breakpoint/data_breakpoint.c |    6 +-
 tools/perf/Documentation/perf-lock.txt  |   29 ++++++
 tools/perf/builtin-lock.c               |  148 ++++++++++++++++++++++++++++++-
 tools/perf/command-list.txt             |    1 +
 10 files changed, 198 insertions(+), 25 deletions(-)

[PATCH 1/5] percpu: Add __percpu sparse annotations to hw_breakpoint

[Frederic Weisbecker]

From: Tejun Heo <tj@kernel.org>

Add __percpu sparse annotations to hw_breakpoint.

These annotations are to make sparse consider percpu variables to be
in a different address space and warn if accessed without going
through percpu accessors.  This patch doesn't affect normal builds.

In kernel/hw_breakpoint.c, per_cpu(nr_task_bp_pinned, cpu)'s will
trigger spurious noderef related warnings from sparse.  Changing it to
&per_cpu(nr_task_bp_pinned[0], cpu) will work around the problem but
deemed to ugly by the maintainer.  Leave it alone until better
solution can be found.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
LKML-Reference: <4B7B4B7A.9050902@kernel.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 include/linux/hw_breakpoint.h           |    8 ++++----
 kernel/hw_breakpoint.c                  |   10 +++++-----
 samples/hw_breakpoint/data_breakpoint.c |    6 +++---
 3 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h
index 5977b72..c70d27a 100644
--- a/include/linux/hw_breakpoint.h
+++ b/include/linux/hw_breakpoint.h
@@ -66,14 +66,14 @@ register_wide_hw_breakpoint_cpu(struct perf_event_attr *attr,
 				perf_overflow_handler_t	triggered,
 				int cpu);
 
-extern struct perf_event **
+extern struct perf_event * __percpu *
 register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			    perf_overflow_handler_t triggered);
 
 extern int register_perf_hw_breakpoint(struct perf_event *bp);
 extern int __register_perf_hw_breakpoint(struct perf_event *bp);
 extern void unregister_hw_breakpoint(struct perf_event *bp);
-extern void unregister_wide_hw_breakpoint(struct perf_event **cpu_events);
+extern void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events);
 
 extern int dbg_reserve_bp_slot(struct perf_event *bp);
 extern int dbg_release_bp_slot(struct perf_event *bp);
@@ -100,7 +100,7 @@ static inline struct perf_event *
 register_wide_hw_breakpoint_cpu(struct perf_event_attr *attr,
 				perf_overflow_handler_t	 triggered,
 				int cpu)		{ return NULL; }
-static inline struct perf_event **
+static inline struct perf_event * __percpu *
 register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			    perf_overflow_handler_t triggered)	{ return NULL; }
 static inline int
@@ -109,7 +109,7 @@ static inline int
 __register_perf_hw_breakpoint(struct perf_event *bp) 	{ return -ENOSYS; }
 static inline void unregister_hw_breakpoint(struct perf_event *bp)	{ }
 static inline void
-unregister_wide_hw_breakpoint(struct perf_event **cpu_events)		{ }
+unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)	{ }
 static inline int
 reserve_bp_slot(struct perf_event *bp)			{return -ENOSYS; }
 static inline void release_bp_slot(struct perf_event *bp) 		{ }
diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c
index 967e661..6542eac 100644
--- a/kernel/hw_breakpoint.c
+++ b/kernel/hw_breakpoint.c
@@ -413,17 +413,17 @@ EXPORT_SYMBOL_GPL(unregister_hw_breakpoint);
  *
  * @return a set of per_cpu pointers to perf events
  */
-struct perf_event **
+struct perf_event * __percpu *
 register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			    perf_overflow_handler_t triggered)
 {
-	struct perf_event **cpu_events, **pevent, *bp;
+	struct perf_event * __percpu *cpu_events, **pevent, *bp;
 	long err;
 	int cpu;
 
 	cpu_events = alloc_percpu(typeof(*cpu_events));
 	if (!cpu_events)
-		return ERR_PTR(-ENOMEM);
+		return (void __percpu __force *)ERR_PTR(-ENOMEM);
 
 	get_online_cpus();
 	for_each_online_cpu(cpu) {
@@ -451,7 +451,7 @@ fail:
 	put_online_cpus();
 
 	free_percpu(cpu_events);
-	return ERR_PTR(err);
+	return (void __percpu __force *)ERR_PTR(err);
 }
 EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
 
@@ -459,7 +459,7 @@ EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
  * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel
  * @cpu_events: the per cpu set of events to unregister
  */
-void unregister_wide_hw_breakpoint(struct perf_event **cpu_events)
+void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)
 {
 	int cpu;
 	struct perf_event **pevent;
diff --git a/samples/hw_breakpoint/data_breakpoint.c b/samples/hw_breakpoint/data_breakpoint.c
index c69cbe9..bd0f337 100644
--- a/samples/hw_breakpoint/data_breakpoint.c
+++ b/samples/hw_breakpoint/data_breakpoint.c
@@ -34,7 +34,7 @@
 #include <linux/perf_event.h>
 #include <linux/hw_breakpoint.h>
 
-struct perf_event **sample_hbp;
+struct perf_event * __percpu *sample_hbp;
 
 static char ksym_name[KSYM_NAME_LEN] = "pid_max";
 module_param_string(ksym, ksym_name, KSYM_NAME_LEN, S_IRUGO);
@@ -61,8 +61,8 @@ static int __init hw_break_module_init(void)
 	attr.bp_type = HW_BREAKPOINT_W | HW_BREAKPOINT_R;
 
 	sample_hbp = register_wide_hw_breakpoint(&attr, sample_hbp_handler);
-	if (IS_ERR(sample_hbp)) {
-		ret = PTR_ERR(sample_hbp);
+	if (IS_ERR((void __force *)sample_hbp)) {
+		ret = PTR_ERR((void __force *)sample_hbp);
 		goto fail;
 	}
 
-- 
1.6.2.3

[PATCH 2/5] perf lock: Fix and add misc documentally things

[Frederic Weisbecker]

From: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>

I've forgot to add 'perf lock' line to command-list.txt,
so users of perf could not find perf lock when they type 'perf'.

Fixing command-list.txt requires document
(tools/perf/Documentation/perf-lock.txt).
But perf lock is too much "under construction" to write a
stable document, so this is something like pseudo document for now.

And I wrote description of perf lock at help section of
CONFIG_LOCK_STAT, this will navigate users of lock trace events.

Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
LKML-Reference: <1265267295-8388-1-git-send-email-mitake@dcl.info.waseda.ac.jp>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 lib/Kconfig.debug                      |    6 ++++++
 tools/perf/Documentation/perf-lock.txt |   29 +++++++++++++++++++++++++++++
 tools/perf/command-list.txt            |    1 +
 3 files changed, 36 insertions(+), 0 deletions(-)
 create mode 100644 tools/perf/Documentation/perf-lock.txt

diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 25c3ed5..65f964e 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -520,6 +520,12 @@ config LOCK_STAT
 
 	 For more details, see Documentation/lockstat.txt
 
+	 You can analyze lock events with "perf lock", subcommand of perf.
+	 If you want to use "perf lock", you need to turn on CONFIG_EVENT_TRACING.
+
+	 CONFIG_LOCK_STAT defines "contended" and "acquired" lock events.
+ 	 (CONFIG_LOCKDEP defines "acquire" and "release" events.)
+
 config DEBUG_LOCKDEP
 	bool "Lock dependency engine debugging"
 	depends on DEBUG_KERNEL && LOCKDEP
diff --git a/tools/perf/Documentation/perf-lock.txt b/tools/perf/Documentation/perf-lock.txt
new file mode 100644
index 0000000..b317102
--- /dev/null
+++ b/tools/perf/Documentation/perf-lock.txt
@@ -0,0 +1,29 @@
+perf-lock(1)
+============
+
+NAME
+----
+perf-lock - Analyze lock events
+
+SYNOPSIS
+--------
+[verse]
+'perf lock' {record|report|trace}
+
+DESCRIPTION
+-----------
+You can analyze various lock behaviours
+and statistics with this 'perf lock' command.
+
+  'perf lock record <command>' records lock events
+  between start and end <command>. And this command
+  produces the file "perf.data" which contains tracing
+  results of lock events.
+
+  'perf lock trace' shows raw lock events.
+
+  'perf lock report' reports statistical data.
+
+SEE ALSO
+--------
+linkperf:perf[1]
diff --git a/tools/perf/command-list.txt b/tools/perf/command-list.txt
index 9afcff2..db6ee94 100644
--- a/tools/perf/command-list.txt
+++ b/tools/perf/command-list.txt
@@ -18,3 +18,4 @@ perf-top			mainporcelain common
 perf-trace			mainporcelain common
 perf-probe			mainporcelain common
 perf-kmem			mainporcelain common
+perf-lock			mainporcelain common
-- 
1.6.2.3

[PATCH 3/5] perf lock: Drop the buffers multiplexing dependency

[Frederic Weisbecker]

We need to deal with time ordered events to build a correct
state machine of lock events. This is why we multiplex the lock
events buffers. But the ordering is done from the kernel, on
the tracing fast path, leading to high contention between cpus.

Without multiplexing, the events appears in a weak order.
If we have four events, each split per cpu, perf record will
read the events buffers in the following order:

[ CPU0 ev0, CPU0 ev1, CPU0 ev3, CPU0 ev4, CPU1 ev0, CPU1 ev0....]

To handle a post processing reordering, we could just read and sort
the whole in memory, but it just doesn't scale with high amounts
of events: lock events can fill huge amounts in few times.

Basically we need to sort in memory and find a "grace period"
point when we know that a given slice of previously sorted events
can be committed for post-processing, so that we can unload the
memory usage step by step and keep a scalable sorting list.

There is no strong rules about how to define such "grace period".
What does this patch is:

We define a FLUSH_PERIOD value that defines a grace period in
seconds.
We want to have a slice of events covering 2 * FLUSH_PERIOD in our
sorted list.
If FLUSH_PERIOD is big enough, it ensures every events that occured
in the first half of the timeslice have all been buffered and there
are none remaining and there won't be further to put inside this
first timeslice. Then once we reach the 2 * FLUSH_PERIOD
timeslice, we flush the first half to be gentle with the memory
(the second half can still get new events in the middle, so wait
another period to flush it)

FLUSH_PERIOD is defined to 5 seconds. Say the first event started on
time t0. We can safely assume that at the time we are processing
events of t0 + 10 seconds, ther won't be anymore events to read
from perf.data that occured between t0 and t0 + 5 seconds. Hence
we can safely flush the first half.

To point out funky bugs, we have a guardian that checks a new event
timestamp is not below the last event's timestamp flushed and that
displays a warning in this case.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
---
 tools/perf/builtin-lock.c |  148 ++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 146 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-lock.c b/tools/perf/builtin-lock.c
index fb9ab2a..e12c844 100644
--- a/tools/perf/builtin-lock.c
+++ b/tools/perf/builtin-lock.c
@@ -460,6 +460,150 @@ process_raw_event(void *data, int cpu,
 		process_lock_release_event(data, event, cpu, timestamp, thread);
 }
 
+struct raw_event_queue {
+	u64			timestamp;
+	int			cpu;
+	void			*data;
+	struct thread		*thread;
+	struct list_head	list;
+};
+
+static LIST_HEAD(raw_event_head);
+
+#define FLUSH_PERIOD	(5 * NSEC_PER_SEC)
+
+static u64 flush_limit = ULLONG_MAX;
+static u64 last_flush = 0;
+struct raw_event_queue *last_inserted;
+
+static void flush_raw_event_queue(u64 limit)
+{
+	struct raw_event_queue *tmp, *iter;
+
+	list_for_each_entry_safe(iter, tmp, &raw_event_head, list) {
+		if (iter->timestamp > limit)
+			return;
+
+		if (iter == last_inserted)
+			last_inserted = NULL;
+
+		process_raw_event(iter->data, iter->cpu, iter->timestamp,
+				  iter->thread);
+
+		last_flush = iter->timestamp;
+		list_del(&iter->list);
+		free(iter->data);
+		free(iter);
+	}
+}
+
+static void __queue_raw_event_end(struct raw_event_queue *new)
+{
+	struct raw_event_queue *iter;
+
+	list_for_each_entry_reverse(iter, &raw_event_head, list) {
+		if (iter->timestamp < new->timestamp) {
+			list_add(&new->list, &iter->list);
+			return;
+		}
+	}
+
+	list_add(&new->list, &raw_event_head);
+}
+
+static void __queue_raw_event_before(struct raw_event_queue *new,
+				     struct raw_event_queue *iter)
+{
+	list_for_each_entry_continue_reverse(iter, &raw_event_head, list) {
+		if (iter->timestamp < new->timestamp) {
+			list_add(&new->list, &iter->list);
+			return;
+		}
+	}
+
+	list_add(&new->list, &raw_event_head);
+}
+
+static void __queue_raw_event_after(struct raw_event_queue *new,
+				     struct raw_event_queue *iter)
+{
+	list_for_each_entry_continue(iter, &raw_event_head, list) {
+		if (iter->timestamp > new->timestamp) {
+			list_add_tail(&new->list, &iter->list);
+			return;
+		}
+	}
+	list_add_tail(&new->list, &raw_event_head);
+}
+
+/* The queue is ordered by time */
+static void __queue_raw_event(struct raw_event_queue *new)
+{
+	if (!last_inserted) {
+		__queue_raw_event_end(new);
+		return;
+	}
+
+	/*
+	 * Most of the time the current event has a timestamp
+	 * very close to the last event inserted, unless we just switched
+	 * to another event buffer. Having a sorting based on a list and
+	 * on the last inserted event that is close to the current one is
+	 * probably more efficient than an rbtree based sorting.
+	 */
+	if (last_inserted->timestamp >= new->timestamp)
+		__queue_raw_event_before(new, last_inserted);
+	else
+		__queue_raw_event_after(new, last_inserted);
+}
+
+static void queue_raw_event(void *data, int raw_size, int cpu,
+			    u64 timestamp, struct thread *thread)
+{
+	struct raw_event_queue *new;
+
+	if (flush_limit == ULLONG_MAX)
+		flush_limit = timestamp + FLUSH_PERIOD;
+
+	if (timestamp < last_flush) {
+		printf("Warning: Timestamp below last timeslice flush\n");
+		return;
+	}
+
+	new = malloc(sizeof(*new));
+	if (!new)
+		die("Not enough memory\n");
+
+	new->timestamp = timestamp;
+	new->cpu = cpu;
+	new->thread = thread;
+
+	new->data = malloc(raw_size);
+	if (!new->data)
+		die("Not enough memory\n");
+
+	memcpy(new->data, data, raw_size);
+
+	__queue_raw_event(new);
+	last_inserted = new;
+
+	/*
+	 * We want to have a slice of events covering 2 * FLUSH_PERIOD
+	 * If FLUSH_PERIOD is big enough, it ensures every events that occured
+	 * in the first half of the timeslice have all been buffered and there
+	 * are none remaining (we need that because of the weakly ordered
+	 * event recording we have). Then once we reach the 2 * FLUSH_PERIOD
+	 * timeslice, we flush the first half to be gentle with the memory
+	 * (the second half can still get new events in the middle, so wait
+	 * another period to flush it)
+	 */
+	if (new->timestamp > flush_limit &&
+		new->timestamp - flush_limit > FLUSH_PERIOD) {
+		flush_limit += FLUSH_PERIOD;
+		flush_raw_event_queue(flush_limit);
+	}
+}
+
 static int process_sample_event(event_t *event, struct perf_session *session)
 {
 	struct thread *thread;
@@ -480,7 +624,7 @@ static int process_sample_event(event_t *event, struct perf_session *session)
 	if (profile_cpu != -1 && profile_cpu != (int) data.cpu)
 		return 0;
 
-	process_raw_event(data.raw_data, data.cpu, data.time, thread);
+	queue_raw_event(data.raw_data, data.raw_size, data.cpu, data.time, thread);
 
 	return 0;
 }
@@ -576,6 +720,7 @@ static void __cmd_report(void)
 	setup_pager();
 	select_key();
 	read_events();
+	flush_raw_event_queue(ULLONG_MAX);
 	sort_result();
 	print_result();
 }
@@ -608,7 +753,6 @@ static const char *record_args[] = {
 	"record",
 	"-a",
 	"-R",
-	"-M",
 	"-f",
 	"-m", "1024",
 	"-c", "1",
-- 
1.6.2.3

[PATCH 4/5] perf: Remove pointless breakpoint union

[Frederic Weisbecker]

Remove pointless union in the breakpoint field of hw_perf_event.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
---
 include/linux/perf_event.h |    5 ++---
 lib/Kconfig.debug          |    8 +++++---
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 7b18b4f..04f06b4 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -487,9 +487,8 @@ struct hw_perf_event {
 			struct hrtimer	hrtimer;
 		};
 #ifdef CONFIG_HAVE_HW_BREAKPOINT
-		union { /* breakpoint */
-			struct arch_hw_breakpoint	info;
-		};
+		/* breakpoint */
+		struct arch_hw_breakpoint	info;
 #endif
 	};
 	atomic64_t			prev_count;
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 65f964e..4dc24cc 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -520,11 +520,13 @@ config LOCK_STAT
 
 	 For more details, see Documentation/lockstat.txt
 
-	 You can analyze lock events with "perf lock", subcommand of perf.
-	 If you want to use "perf lock", you need to turn on CONFIG_EVENT_TRACING.
+	 This also enables lock events required by "perf lock",
+	 subcommand of perf.
+	 If you want to use "perf lock", you also need to turn on
+	 CONFIG_EVENT_TRACING.
 
 	 CONFIG_LOCK_STAT defines "contended" and "acquired" lock events.
- 	 (CONFIG_LOCKDEP defines "acquire" and "release" events.)
+	 (CONFIG_LOCKDEP defines "acquire" and "release" events.)
 
 config DEBUG_LOCKDEP
 	bool "Lock dependency engine debugging"
-- 
1.6.2.3

[PATCH 5/5] x86/hw-breakpoints: Remove the name field

[Frederic Weisbecker]

Remove the name field from the arch_hw_breakpoint. We never deal
with target symbols in the arch level, neither do we need to ever
store it. It's a legacy for the previous version of the x86
breakpoint backend.

Let's remove it.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: K.Prasad <prasad@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
---
 arch/x86/include/asm/hw_breakpoint.h |    1 -
 arch/x86/kernel/hw_breakpoint.c      |    7 -------
 2 files changed, 0 insertions(+), 8 deletions(-)

diff --git a/arch/x86/include/asm/hw_breakpoint.h b/arch/x86/include/asm/hw_breakpoint.h
index 0675a7c..2a1bd8f 100644
--- a/arch/x86/include/asm/hw_breakpoint.h
+++ b/arch/x86/include/asm/hw_breakpoint.h
@@ -10,7 +10,6 @@
  * (display/resolving)
  */
 struct arch_hw_breakpoint {
-	char		*name; /* Contains name of the symbol to set bkpt */
 	unsigned long	address;
 	u8		len;
 	u8		type;
diff --git a/arch/x86/kernel/hw_breakpoint.c b/arch/x86/kernel/hw_breakpoint.c
index dca2802..41e08df 100644
--- a/arch/x86/kernel/hw_breakpoint.c
+++ b/arch/x86/kernel/hw_breakpoint.c
@@ -344,13 +344,6 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp,
 	}
 
 	/*
-	 * For kernel-addresses, either the address or symbol name can be
-	 * specified.
-	 */
-	if (info->name)
-		info->address = (unsigned long)
-				kallsyms_lookup_name(info->name);
-	/*
 	 * Check that the low-order bits of the address are appropriate
 	 * for the alignment implied by len.
 	 */
-- 
1.6.2.3