[PATCH V9 3/8] perf, x86: handle multiple records in PEBS buffer

[Kan Liang]

From: Yan, Zheng <zheng.z.yan@xxxxxxxxx>

When the PEBS interrupt threshold is larger than one record and the
machine supports multiple PEBS events, the records of these events are
mixed up and we need to demultiplex them.

Demuxing the records is hard because the hardware is deficient. The
hardware has two issues that, when combined, create impossible scenarios
to demux. (The deficiency has been fixed in Skylake. The two issues here
are for pre-SKL platforms.)

The first issue is that the 'status' field of the PEBS record is a copy
of the GLOBAL_STATUS MSR at PEBS assist time. To see why this is a
problem let us first describe the regular PEBS cycle:

A) the CTRn value reaches 0:
  - the corresponding bit in GLOBAL_STATUS gets set
  - we start arming the hardware assist
  < some unspecified amount of time later -- this could cover multiple
events of interest >

B) the hardware assist is armed, any next event will trigger it

C) a matching event happens:
  - the hardware assist triggers and generates a PEBS record
    this includes a copy of GLOBAL_STATUS at this moment
  - if we auto-reload we (re)set CTRn
  - we clear the relevant bit in GLOBAL_STATUS

Now consider the following chain of events:

A0, B0, A1, C0

The event generated for counter 0 will include a status with counter 1
set, even though its not at all related to the record. A similar thing
can happen with a !PEBS event if it just happens to overflow at the
right moment.

The second issue is that the hardware will only emit one record for two
or more counters if the event that triggers the assist is 'close'. The
'close' can be several cycles. In some cases even the complete assist,
if the event is something that doesn't need retirement.

For instance, consider this chain of events:
A0, B0, A1, B1, C01

Where C01 is an event that triggers both hardware assists, we will
generate but a single record, but again with both counters listed in the
status field.

This time the record pertains to both events.

Note that these two cases are different but undistinguishable with the
data as generated. Therefore demuxing records with multiple PEBS bits
(we can safely ignore status bits for !PEBS counters) is impossible.

Furthermore we cannot emit the record to both events because that might
cause a data leak -- the events might not have the same privileges -- so
what this patch does is discard such events.

The assumption/hope is that such discards will be rare.

Here lists some possible ways you may get high discard rate.
  - when you count the same thing multiple times. But it is not a useful
    configuration.
  - you can be unfortunate if you measure with a userspace only PEBS
    event along with either a kernel or unrestricted PEBS event. Imagine
    the event triggering and setting the overflow flag right before
    entering the kernel. Then all kernel side events will end up with
    multiple bits set.

Signed-off-by: Yan, Zheng <zheng.z.yan@xxxxxxxxx>
Signed-off-by: Kan Liang <kan.liang@xxxxxxxxx>
---
 arch/x86/kernel/cpu/perf_event_intel_ds.c | 144 +++++++++++++++++++++---------
 include/linux/perf_event.h                |  13 +++
 kernel/events/core.c                      |   6 +-
 kernel/events/internal.h                  |   9 --
 4 files changed, 118 insertions(+), 54 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index f26c8b4..971c77e 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -872,6 +872,9 @@ static void setup_pebs_sample_data(struct perf_event *event,
 	int fll, fst, dsrc;
 	int fl = event->hw.flags;
 
+	if (pebs == NULL)
+		return;
+
 	sample_type = event->attr.sample_type;
 	dsrc = sample_type & PERF_SAMPLE_DATA_SRC;
 
@@ -966,19 +969,68 @@ static void setup_pebs_sample_data(struct perf_event *event,
 		data->br_stack = &cpuc->lbr_stack;
 }
 
+static inline void *
+get_next_pebs_record_by_bit(void *base, void *top, int bit)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	void *at;
+	u64 pebs_status;
+
+	if (base == NULL)
+		return NULL;
+
+	for (at = base; at < top; at += x86_pmu.pebs_record_size) {
+		struct pebs_record_nhm *p = at;
+
+		if (test_bit(bit, (unsigned long *)&p->status)) {
+
+			if (p->status == (1 << bit))
+				return at;
+
+			/* clear non-PEBS bit and re-check */
+			pebs_status = p->status & cpuc->pebs_enabled;
+			pebs_status &= (1ULL << MAX_PEBS_EVENTS) - 1;
+			if (pebs_status == (1 << bit))
+				return at;
+		}
+	}
+	return NULL;
+}
+
 static void __intel_pmu_pebs_event(struct perf_event *event,
-				   struct pt_regs *iregs, void *__pebs)
+				   struct pt_regs *iregs,
+				   void *base, void *top,
+				   int bit, int count)
 {
 	struct perf_sample_data data;
 	struct pt_regs regs;
+	int i;
+	void *at = get_next_pebs_record_by_bit(base, top, bit);
 
-	if (!intel_pmu_save_and_restart(event))
+	if (!intel_pmu_save_and_restart(event) &&
+	    !(event->hw.flags & PERF_X86_EVENT_AUTO_RELOAD))
 		return;
 
-	setup_pebs_sample_data(event, iregs, __pebs, &data, &regs);
+	if (count > 1) {
+		for (i = 0; i < count - 1; i++) {
+			setup_pebs_sample_data(event, iregs, at, &data, &regs);
+			perf_event_output(event, &data, &regs);
+			at += x86_pmu.pebs_record_size;
+			at = get_next_pebs_record_by_bit(at, top, bit);
+		}
+	}
+
+	setup_pebs_sample_data(event, iregs, at, &data, &regs);
 
-	if (perf_event_overflow(event, &data, &regs))
+	/*
+	 * All but the last records are processed.
+	 * The last one is left to be able to call the overflow handler.
+	 */
+	if (perf_event_overflow(event, &data, &regs)) {
 		x86_pmu_stop(event, 0);
+		return;
+	}
+
 }
 
 static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
@@ -1008,72 +1060,80 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
 	if (!event->attr.precise_ip)
 		return;
 
-	n = top - at;
+	n = (top - at) / x86_pmu.pebs_record_size;
 	if (n <= 0)
 		return;
 
-	/*
-	 * Should not happen, we program the threshold at 1 and do not
-	 * set a reset value.
-	 */
-	WARN_ONCE(n > 1, "bad leftover pebs %d\n", n);
-	at += n - 1;
-
-	__intel_pmu_pebs_event(event, iregs, at);
+	__intel_pmu_pebs_event(event, iregs, at,
+			       top, 0, n);
 }
 
 static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs)
 {
 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
 	struct debug_store *ds = cpuc->ds;
-	struct perf_event *event = NULL;
-	void *at, *top;
-	u64 status = 0;
+	struct perf_event *event;
+	void *base, *at, *top;
 	int bit;
+	short counts[MAX_PEBS_EVENTS] = {};
 
 	if (!x86_pmu.pebs_active)
 		return;
 
-	at  = (struct pebs_record_nhm *)(unsigned long)ds->pebs_buffer_base;
+	base = (struct pebs_record_nhm *)(unsigned long)ds->pebs_buffer_base;
 	top = (struct pebs_record_nhm *)(unsigned long)ds->pebs_index;
 
 	ds->pebs_index = ds->pebs_buffer_base;
 
-	if (unlikely(at > top))
+	if (unlikely(base >= top))
 		return;
 
-	/*
-	 * Should not happen, we program the threshold at 1 and do not
-	 * set a reset value.
-	 */
-	WARN_ONCE(top - at > x86_pmu.max_pebs_events * x86_pmu.pebs_record_size,
-		  "Unexpected number of pebs records %ld\n",
-		  (long)(top - at) / x86_pmu.pebs_record_size);
-
-	for (; at < top; at += x86_pmu.pebs_record_size) {
+	for (at = base; at < top; at += x86_pmu.pebs_record_size) {
 		struct pebs_record_nhm *p = at;
 
-		for_each_set_bit(bit, (unsigned long *)&p->status,
-				 x86_pmu.max_pebs_events) {
-			event = cpuc->events[bit];
-			if (!test_bit(bit, cpuc->active_mask))
-				continue;
-
-			WARN_ON_ONCE(!event);
-
-			if (!event->attr.precise_ip)
-				continue;
+		bit = find_first_bit((unsigned long *)&p->status,
+					x86_pmu.max_pebs_events);
+		if (bit >= x86_pmu.max_pebs_events)
+			continue;
+		if (!test_bit(bit, cpuc->active_mask))
+			continue;
+		/*
+		 * The PEBS hardware does not deal well with the situation
+		 * when events happen near to each other and multiple bits
+		 * are set. But it should happen rarely.
+		 *
+		 * If these events include one PEBS and multiple non-PEBS
+		 * events, it doesn't impact PEBS record. The record will
+		 * be handled normally. (slow path)
+		 *
+		 * If these events include two or more PEBS events, the
+		 * records for the events can be collapsed into a single
+		 * one, and it's not possible to reconstruct all events
+		 * that caused the PEBS record. It's called collision.
+		 * If collision happened, the record will be dropped.
+		 *
+		 */
+		if (p->status != (1 << bit)) {
+			u64 pebs_status;
 
-			if (__test_and_set_bit(bit, (unsigned long *)&status))
+			/* slow path */
+			pebs_status = p->status & cpuc->pebs_enabled;
+			pebs_status &= (1ULL << MAX_PEBS_EVENTS) - 1;
+			if (pebs_status != (1 << bit))
 				continue;
-
-			break;
 		}
+		counts[bit]++;
+	}
 
-		if (!event || bit >= x86_pmu.max_pebs_events)
+	for (bit = 0; bit < x86_pmu.max_pebs_events; bit++) {
+		if (counts[bit] == 0)
 			continue;
+		event = cpuc->events[bit];
+		WARN_ON_ONCE(!event);
+		WARN_ON_ONCE(!event->attr.precise_ip);
 
-		__intel_pmu_pebs_event(event, iregs, at);
+		__intel_pmu_pebs_event(event, iregs, base,
+				       top, bit, counts[bit]);
 	}
 }
 
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 61992cf..bed1b6f 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -734,6 +734,19 @@ extern int perf_event_overflow(struct perf_event *event,
 				 struct perf_sample_data *data,
 				 struct pt_regs *regs);
 
+extern void perf_event_output(struct perf_event *event,
+				struct perf_sample_data *data,
+				struct pt_regs *regs);
+
+extern void
+perf_event_header__init_id(struct perf_event_header *header,
+			   struct perf_sample_data *data,
+			   struct perf_event *event);
+extern void
+perf_event__output_id_sample(struct perf_event *event,
+			     struct perf_output_handle *handle,
+			     struct perf_sample_data *sample);
+
 static inline bool is_sampling_event(struct perf_event *event)
 {
 	return event->attr.sample_period != 0;
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f528829..4d221a4 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5333,9 +5333,9 @@ void perf_prepare_sample(struct perf_event_header *header,
 	}
 }
 
-static void perf_event_output(struct perf_event *event,
-				struct perf_sample_data *data,
-				struct pt_regs *regs)
+void perf_event_output(struct perf_event *event,
+			struct perf_sample_data *data,
+			struct pt_regs *regs)
 {
 	struct perf_output_handle handle;
 	struct perf_event_header header;
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index 9f6ce9b..2deb24c 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -72,15 +72,6 @@ static inline bool rb_has_aux(struct ring_buffer *rb)
 void perf_event_aux_event(struct perf_event *event, unsigned long head,
 			  unsigned long size, u64 flags);
 
-extern void
-perf_event_header__init_id(struct perf_event_header *header,
-			   struct perf_sample_data *data,
-			   struct perf_event *event);
-extern void
-perf_event__output_id_sample(struct perf_event *event,
-			     struct perf_output_handle *handle,
-			     struct perf_sample_data *sample);
-
 extern struct page *
 perf_mmap_to_page(struct ring_buffer *rb, unsigned long pgoff);
 
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/