[PATCH v3 00/14] perf kvm: Support histograms and TUI mode

[PATCH v3 00/14] perf kvm: Support histograms and TUI mode

[Leo Yan]

This patch set is to enable histograms and (partially) TUI mode in perf
kvm tool.

Patch set v1 [1] gives brief introduction for the change and this series
has a minor update for the patch 13 'perf kvm: Add TUI mode for stat
report' to avoid building failure when the system doesn't support
HAVE_SLANG_SUPPORT.

We can use below commands for testing this series:

In a terminal, you could launch a virtual machine with qemu command; in
below case, I downloaded a Ubuntu (or Debian) iso file and used it as
the file system image:

  $ qemu-system-x86_64 -M pc -enable-kvm -cpu host -m 4096 -hda ubuntu-22.04-desktop-amd64.iso

Then in another terminal, I can use below command to capture KVM trace
data and report the result:

  # cd linux/tools/perf
  # ./perf kvm stat record
  # ./perf kvm stat report          => Output in TUI mode
  # ./perf kvm stat report --stdio  => Output in stdio mode

Changes from v2:
* Found building failure with command 'make VF=1 DEBUG=1 NO_SLANG=1',
  fixed it in the patch 13. (James Clark)

Changes from v1:
* Updated the patch 13 'perf kvm: Add TUI mode for stat report' to avoid
  building failure if no support HAVE_SLANG_SUPPORT.

[1] https://lore.kernel.org/lkml/20230226042053.1492409-1-leo.yan@linaro.org/


Leo Yan (14):
  perf kvm: Refactor overall statistics
  perf kvm: Add pointer to 'perf_kvm_stat' in kvm event
  perf kvm: Move up metrics helpers
  perf kvm: Use subtraction for comparison metrics
  perf kvm: Introduce histograms data structures
  perf kvm: Pass argument 'sample' to kvm_alloc_init_event()
  perf kvm: Parse address location for samples
  perf kvm: Add dimensions for KVM event statistics
  perf kvm: Use histograms list to replace cached list
  perf kvm: Polish sorting key
  perf kvm: Support printing attributions for dimensions
  perf kvm: Add dimensions for percentages
  perf kvm: Add TUI mode for stat report
  perf kvm: Update documentation to reflect new changes

 tools/perf/Documentation/perf-kvm.txt |   9 +-
 tools/perf/builtin-kvm.c              | 856 +++++++++++++++++++++-----
 tools/perf/util/kvm-stat.h            |  26 +-
 3 files changed, 716 insertions(+), 175 deletions(-)

-- 
2.34.1

[PATCH v3 01/14] perf kvm: Refactor overall statistics

[Leo Yan]

Currently the tool computes overall statistics when sort the results.
This patch refactors overall statistics during events processing,
therefore, the function update_total_coun() is not needed anymore, an
extra benefit is we can de-couple code between the statistics and the
sorting.

This patch is not expected any functionality changes.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 641e739c717c..0172e5b0d26e 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -278,9 +278,14 @@ static double kvm_event_rel_stddev(int vcpu_id, struct kvm_event *event)
 				avg_stats(&kvm_stats->stats));
 }
 
-static bool update_kvm_event(struct kvm_event *event, int vcpu_id,
+static bool update_kvm_event(struct perf_kvm_stat *kvm,
+			     struct kvm_event *event, int vcpu_id,
 			     u64 time_diff)
 {
+	/* Update overall statistics */
+	kvm->total_count++;
+	kvm->total_time += time_diff;
+
 	if (vcpu_id == -1) {
 		kvm_update_event_stats(&event->total, time_diff);
 		return true;
@@ -399,7 +404,7 @@ static bool handle_end_event(struct perf_kvm_stat *kvm,
 		}
 	}
 
-	return update_kvm_event(event, vcpu, time_diff);
+	return update_kvm_event(kvm, event, vcpu, time_diff);
 }
 
 static
@@ -526,15 +531,6 @@ static void insert_to_result(struct rb_root *result, struct kvm_event *event,
 	rb_insert_color(&event->rb, result);
 }
 
-static void
-update_total_count(struct perf_kvm_stat *kvm, struct kvm_event *event)
-{
-	int vcpu = kvm->trace_vcpu;
-
-	kvm->total_count += get_event_count(event, vcpu);
-	kvm->total_time += get_event_time(event, vcpu);
-}
-
 static bool event_is_valid(struct kvm_event *event, int vcpu)
 {
 	return !!get_event_count(event, vcpu);
@@ -549,7 +545,6 @@ static void sort_result(struct perf_kvm_stat *kvm)
 	for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
 		list_for_each_entry(event, &kvm->kvm_events_cache[i], hash_entry) {
 			if (event_is_valid(event, vcpu)) {
-				update_total_count(kvm, event);
 				insert_to_result(&kvm->result, event,
 						 kvm->compare, vcpu);
 			}
-- 
2.34.1

[PATCH v3 02/14] perf kvm: Add pointer to 'perf_kvm_stat' in kvm event

[Leo Yan]

Sometimes, handling kvm events needs to base on global variables, e.g.
when read event counts we need to know the target vcpu ID; the global
variables are stored in structure perf_kvm_stat.

This patch adds add a 'perf_kvm_stat' pointer in kvm event structure,
it is to be used by later refactoring.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 6 ++++--
 tools/perf/util/kvm-stat.h | 5 +++--
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 0172e5b0d26e..3d2560ec6b37 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -209,7 +209,8 @@ static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
 	return true;
 }
 
-static struct kvm_event *kvm_alloc_init_event(struct event_key *key)
+static struct kvm_event *kvm_alloc_init_event(struct perf_kvm_stat *kvm,
+					      struct event_key *key)
 {
 	struct kvm_event *event;
 
@@ -219,6 +220,7 @@ static struct kvm_event *kvm_alloc_init_event(struct event_key *key)
 		return NULL;
 	}
 
+	event->perf_kvm = kvm;
 	event->key = *key;
 	init_stats(&event->total.stats);
 	return event;
@@ -238,7 +240,7 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
 			return event;
 	}
 
-	event = kvm_alloc_init_event(key);
+	event = kvm_alloc_init_event(kvm, key);
 	if (!event)
 		return NULL;
 
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index 6f0fa05b62b6..40a4b66cfee6 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -24,10 +24,13 @@ struct kvm_event_stats {
 	struct stats stats;
 };
 
+struct perf_kvm_stat;
+
 struct kvm_event {
 	struct list_head hash_entry;
 	struct rb_node rb;
 
+	struct perf_kvm_stat *perf_kvm;
 	struct event_key key;
 
 	struct kvm_event_stats total;
@@ -44,8 +47,6 @@ struct kvm_event_key {
 	key_cmp_fun key;
 };
 
-struct perf_kvm_stat;
-
 struct child_event_ops {
 	void (*get_key)(struct evsel *evsel,
 			struct perf_sample *sample,
-- 
2.34.1

[PATCH v3 03/14] perf kvm: Move up metrics helpers

[Leo Yan]

This patch moves up the helper functions of event's metrics for later
adding code to call them.

No any functionality changes, but has a function renaming from
compare_kvm_event_{metric}() to cmp_event_{metric}().

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 72 ++++++++++++++++++++--------------------
 1 file changed, 36 insertions(+), 36 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 3d2560ec6b37..62c097a37da9 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -49,6 +49,42 @@
 #include <math.h>
 #include <perf/mmap.h>
 
+#define GET_EVENT_KEY(func, field)					\
+static u64 get_event_ ##func(struct kvm_event *event, int vcpu)		\
+{									\
+	if (vcpu == -1)							\
+		return event->total.field;				\
+									\
+	if (vcpu >= event->max_vcpu)					\
+		return 0;						\
+									\
+	return event->vcpu[vcpu].field;					\
+}
+
+#define COMPARE_EVENT_KEY(func, field)					\
+GET_EVENT_KEY(func, field)						\
+static int cmp_event_ ## func(struct kvm_event *one,			\
+			      struct kvm_event *two, int vcpu)		\
+{									\
+	return get_event_ ##func(one, vcpu) >				\
+	       get_event_ ##func(two, vcpu);				\
+}
+
+GET_EVENT_KEY(time, time);
+GET_EVENT_KEY(max, stats.max);
+GET_EVENT_KEY(min, stats.min);
+COMPARE_EVENT_KEY(count, stats.n);
+COMPARE_EVENT_KEY(mean, stats.mean);
+
+#define DEF_SORT_NAME_KEY(name, compare_key)				\
+	{ #name, cmp_event_ ## compare_key }
+
+static struct kvm_event_key keys[] = {
+	DEF_SORT_NAME_KEY(sample, count),
+	DEF_SORT_NAME_KEY(time, mean),
+	{ NULL, NULL }
+};
+
 static const char *get_filename_for_perf_kvm(void)
 {
 	const char *filename;
@@ -461,42 +497,6 @@ static bool handle_kvm_event(struct perf_kvm_stat *kvm,
 	return true;
 }
 
-#define GET_EVENT_KEY(func, field)					\
-static u64 get_event_ ##func(struct kvm_event *event, int vcpu)		\
-{									\
-	if (vcpu == -1)							\
-		return event->total.field;				\
-									\
-	if (vcpu >= event->max_vcpu)					\
-		return 0;						\
-									\
-	return event->vcpu[vcpu].field;					\
-}
-
-#define COMPARE_EVENT_KEY(func, field)					\
-GET_EVENT_KEY(func, field)						\
-static int compare_kvm_event_ ## func(struct kvm_event *one,		\
-					struct kvm_event *two, int vcpu)\
-{									\
-	return get_event_ ##func(one, vcpu) >				\
-				get_event_ ##func(two, vcpu);		\
-}
-
-GET_EVENT_KEY(time, time);
-COMPARE_EVENT_KEY(count, stats.n);
-COMPARE_EVENT_KEY(mean, stats.mean);
-GET_EVENT_KEY(max, stats.max);
-GET_EVENT_KEY(min, stats.min);
-
-#define DEF_SORT_NAME_KEY(name, compare_key)				\
-	{ #name, compare_kvm_event_ ## compare_key }
-
-static struct kvm_event_key keys[] = {
-	DEF_SORT_NAME_KEY(sample, count),
-	DEF_SORT_NAME_KEY(time, mean),
-	{ NULL, NULL }
-};
-
 static bool select_key(struct perf_kvm_stat *kvm)
 {
 	int i;
-- 
2.34.1

[PATCH v3 04/14] perf kvm: Use subtraction for comparison metrics

[Leo Yan]

Currently the metrics comparison uses greater operator (>), it returns
the boolean value (0 or 1).

This patch changes to use subtraction as comparison result, which can
be used by histograms sorting.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 62c097a37da9..6eec0db2dda6 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -66,7 +66,7 @@ GET_EVENT_KEY(func, field)						\
 static int cmp_event_ ## func(struct kvm_event *one,			\
 			      struct kvm_event *two, int vcpu)		\
 {									\
-	return get_event_ ##func(one, vcpu) >				\
+	return get_event_ ##func(one, vcpu) -				\
 	       get_event_ ##func(two, vcpu);				\
 }
 
@@ -523,7 +523,7 @@ static void insert_to_result(struct rb_root *result, struct kvm_event *event,
 		p = container_of(*rb, struct kvm_event, rb);
 		parent = *rb;
 
-		if (bigger(event, p, vcpu))
+		if (bigger(event, p, vcpu) > 0)
 			rb = &(*rb)->rb_left;
 		else
 			rb = &(*rb)->rb_right;
-- 
2.34.1

[PATCH v3 05/14] perf kvm: Introduce histograms data structures

[Leo Yan]

This is a preparation to support histograms in perf kvm tool.  As first
step, this patch defines histograms data structures and initialize them.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 18 ++++++++++++++++++
 tools/perf/util/kvm-stat.h |  1 +
 2 files changed, 19 insertions(+)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 6eec0db2dda6..4ae54ba4fdaf 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -85,6 +85,20 @@ static struct kvm_event_key keys[] = {
 	{ NULL, NULL }
 };
 
+struct kvm_hists {
+	struct hists		hists;
+	struct perf_hpp_list	list;
+};
+
+static struct kvm_hists kvm_hists;
+
+static int kvm_hists__init(void)
+{
+	__hists__init(&kvm_hists.hists, &kvm_hists.list);
+	perf_hpp_list__init(&kvm_hists.list);
+	return 0;
+}
+
 static const char *get_filename_for_perf_kvm(void)
 {
 	const char *filename;
@@ -957,6 +971,8 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
 	set_term_quiet_input(&save);
 	init_kvm_event_record(kvm);
 
+	kvm_hists__init();
+
 	signal(SIGINT, sig_handler);
 	signal(SIGTERM, sig_handler);
 
@@ -1152,6 +1168,8 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	init_kvm_event_record(kvm);
 	setup_pager();
 
+	kvm_hists__init();
+
 	ret = read_events(kvm);
 	if (ret)
 		goto exit;
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index 40a4b66cfee6..a3219829c448 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -5,6 +5,7 @@
 #ifdef HAVE_KVM_STAT_SUPPORT
 
 #include "tool.h"
+#include "sort.h"
 #include "stat.h"
 #include "record.h"
 
-- 
2.34.1

[PATCH v3 06/14] perf kvm: Pass argument 'sample' to kvm_alloc_init_event()

[Leo Yan]

This patch adds an argument 'sample' for kvm_alloc_init_event(), and its
caller functions are updated as well for passing down the 'sample'
pointer.

This is a preparation change to allow later patch to create histograms
entries for kvm event, no any functionality changes.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 4ae54ba4fdaf..779881901a05 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -260,7 +260,8 @@ static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
 }
 
 static struct kvm_event *kvm_alloc_init_event(struct perf_kvm_stat *kvm,
-					      struct event_key *key)
+					      struct event_key *key,
+					      struct perf_sample *sample __maybe_unused)
 {
 	struct kvm_event *event;
 
@@ -277,7 +278,8 @@ static struct kvm_event *kvm_alloc_init_event(struct perf_kvm_stat *kvm,
 }
 
 static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
-					       struct event_key *key)
+					       struct event_key *key,
+					       struct perf_sample *sample)
 {
 	struct kvm_event *event;
 	struct list_head *head;
@@ -290,7 +292,7 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
 			return event;
 	}
 
-	event = kvm_alloc_init_event(kvm, key);
+	event = kvm_alloc_init_event(kvm, key, sample);
 	if (!event)
 		return NULL;
 
@@ -300,15 +302,16 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
 
 static bool handle_begin_event(struct perf_kvm_stat *kvm,
 			       struct vcpu_event_record *vcpu_record,
-			       struct event_key *key, u64 timestamp)
+			       struct event_key *key,
+			       struct perf_sample *sample)
 {
 	struct kvm_event *event = NULL;
 
 	if (key->key != INVALID_KEY)
-		event = find_create_kvm_event(kvm, key);
+		event = find_create_kvm_event(kvm, key, sample);
 
 	vcpu_record->last_event = event;
-	vcpu_record->start_time = timestamp;
+	vcpu_record->start_time = sample->time;
 	return true;
 }
 
@@ -375,12 +378,12 @@ static bool is_child_event(struct perf_kvm_stat *kvm,
 static bool handle_child_event(struct perf_kvm_stat *kvm,
 			       struct vcpu_event_record *vcpu_record,
 			       struct event_key *key,
-			       struct perf_sample *sample __maybe_unused)
+			       struct perf_sample *sample)
 {
 	struct kvm_event *event = NULL;
 
 	if (key->key != INVALID_KEY)
-		event = find_create_kvm_event(kvm, key);
+		event = find_create_kvm_event(kvm, key, sample);
 
 	vcpu_record->last_event = event;
 
@@ -429,7 +432,7 @@ static bool handle_end_event(struct perf_kvm_stat *kvm,
 		return true;
 
 	if (!event)
-		event = find_create_kvm_event(kvm, key);
+		event = find_create_kvm_event(kvm, key, sample);
 
 	if (!event)
 		return false;
@@ -500,7 +503,7 @@ static bool handle_kvm_event(struct perf_kvm_stat *kvm,
 		return true;
 
 	if (kvm->events_ops->is_begin_event(evsel, sample, &key))
-		return handle_begin_event(kvm, vcpu_record, &key, sample->time);
+		return handle_begin_event(kvm, vcpu_record, &key, sample);
 
 	if (is_child_event(kvm, evsel, sample, &key))
 		return handle_child_event(kvm, vcpu_record, &key, sample);
-- 
2.34.1

[PATCH v3 07/14] perf kvm: Parse address location for samples

[Leo Yan]

Parse address location for samples and save it into the structure
'perf_kvm_stat', it is to be used by histograms entry.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 5 +++++
 tools/perf/util/kvm-stat.h | 4 ++++
 2 files changed, 9 insertions(+)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 779881901a05..a9f467926bdd 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -704,6 +704,11 @@ static int process_sample_event(struct perf_tool *tool,
 	if (skip_sample(kvm, sample))
 		return 0;
 
+	if (machine__resolve(machine, &kvm->al, sample) < 0) {
+		pr_warning("Fail to resolve address location, skip sample.\n");
+		return 0;
+	}
+
 	thread = machine__findnew_thread(machine, sample->pid, sample->tid);
 	if (thread == NULL) {
 		pr_debug("problem processing %d event, skipping it.\n",
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index a3219829c448..80d5c5a9ae31 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -7,6 +7,7 @@
 #include "tool.h"
 #include "sort.h"
 #include "stat.h"
+#include "symbol.h"
 #include "record.h"
 
 struct evsel;
@@ -86,6 +87,9 @@ struct perf_kvm_stat {
 	const char *sort_key;
 	int trace_vcpu;
 
+	/* Used when process events */
+	struct addr_location al;
+
 	struct exit_reasons_table *exit_reasons;
 	const char *exit_reasons_isa;
 
-- 
2.34.1

[PATCH v3 08/14] perf kvm: Add dimensions for KVM event statistics

[Leo Yan]

To support KVM event statistics, this patch firstly registers histograms
columns and sorting fields; every column or field has its own format
structure, the format structure is dereferenced to access the dimension,
finally the dimension provides the comparison callback for sorting
result.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 239 +++++++++++++++++++++++++++++++++++--
 tools/perf/util/kvm-stat.h |   2 +
 2 files changed, 234 insertions(+), 7 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index a9f467926bdd..da84f5063d4d 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -70,9 +70,9 @@ static int cmp_event_ ## func(struct kvm_event *one,			\
 	       get_event_ ##func(two, vcpu);				\
 }
 
-GET_EVENT_KEY(time, time);
-GET_EVENT_KEY(max, stats.max);
-GET_EVENT_KEY(min, stats.min);
+COMPARE_EVENT_KEY(time, time);
+COMPARE_EVENT_KEY(max, stats.max);
+COMPARE_EVENT_KEY(min, stats.min);
 COMPARE_EVENT_KEY(count, stats.n);
 COMPARE_EVENT_KEY(mean, stats.mean);
 
@@ -90,13 +90,238 @@ struct kvm_hists {
 	struct perf_hpp_list	list;
 };
 
+struct kvm_dimension {
+	const char *name;
+	int64_t (*cmp)(struct perf_hpp_fmt *fmt, struct hist_entry *left,
+		       struct hist_entry *right);
+};
+
+struct kvm_fmt {
+	struct perf_hpp_fmt	fmt;
+	struct kvm_dimension	*dim;
+};
+
 static struct kvm_hists kvm_hists;
 
-static int kvm_hists__init(void)
+static int64_t
+empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
+	  struct hist_entry *left __maybe_unused,
+	  struct hist_entry *right __maybe_unused)
+{
+	return 0;
+}
+
+static struct kvm_dimension dim_event = {
+	.name		= "name",
+	.cmp		= empty_cmp,
+};
+
+#define EV_METRIC_CMP(metric)						\
+static int64_t ev_cmp_##metric(struct perf_hpp_fmt *fmt __maybe_unused,	\
+			       struct hist_entry *left,			\
+			       struct hist_entry *right)		\
+{									\
+	struct kvm_event *event_left;					\
+	struct kvm_event *event_right;					\
+	struct perf_kvm_stat *perf_kvm;					\
+									\
+	event_left  = container_of(left, struct kvm_event, he);		\
+	event_right = container_of(right, struct kvm_event, he);	\
+									\
+	perf_kvm = event_left->perf_kvm;				\
+	return cmp_event_##metric(event_left, event_right,		\
+				  perf_kvm->trace_vcpu);		\
+}
+
+EV_METRIC_CMP(time)
+EV_METRIC_CMP(count)
+EV_METRIC_CMP(max)
+EV_METRIC_CMP(min)
+EV_METRIC_CMP(mean)
+
+static struct kvm_dimension dim_time = {
+	.name		= "time",
+	.cmp		= ev_cmp_time,
+};
+
+static struct kvm_dimension dim_count = {
+	.name		= "sample",
+	.cmp		= ev_cmp_count,
+};
+
+static struct kvm_dimension dim_max_time = {
+	.name		= "max_t",
+	.cmp		= ev_cmp_max,
+};
+
+static struct kvm_dimension dim_min_time = {
+	.name		= "min_t",
+	.cmp		= ev_cmp_min,
+};
+
+static struct kvm_dimension dim_mean_time = {
+	.name		= "mean_t",
+	.cmp		= ev_cmp_mean,
+};
+
+static struct kvm_dimension *dimensions[] = {
+	&dim_event,
+	&dim_time,
+	&dim_count,
+	&dim_max_time,
+	&dim_min_time,
+	&dim_mean_time,
+	NULL,
+};
+
+static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
+{
+	struct kvm_fmt *kvm_fmt_a = container_of(a, struct kvm_fmt, fmt);
+	struct kvm_fmt *kvm_fmt_b = container_of(b, struct kvm_fmt, fmt);
+
+	return kvm_fmt_a->dim == kvm_fmt_b->dim;
+}
+
+static void fmt_free(struct perf_hpp_fmt *fmt)
+{
+	struct kvm_fmt *kvm_fmt;
+
+	kvm_fmt = container_of(fmt, struct kvm_fmt, fmt);
+	free(kvm_fmt);
+}
+
+static struct kvm_dimension *get_dimension(const char *name)
 {
+	unsigned int i;
+
+	for (i = 0; dimensions[i] != NULL; i++) {
+		if (!strcmp(dimensions[i]->name, name))
+			return dimensions[i];
+	}
+
+	return NULL;
+}
+
+static struct kvm_fmt *get_format(const char *name)
+{
+	struct kvm_dimension *dim = get_dimension(name);
+	struct kvm_fmt *kvm_fmt;
+	struct perf_hpp_fmt *fmt;
+
+	if (!dim)
+		return NULL;
+
+	kvm_fmt = zalloc(sizeof(*kvm_fmt));
+	if (!kvm_fmt)
+		return NULL;
+
+	kvm_fmt->dim = dim;
+
+	fmt = &kvm_fmt->fmt;
+	INIT_LIST_HEAD(&fmt->list);
+	INIT_LIST_HEAD(&fmt->sort_list);
+	fmt->cmp	= dim->cmp;
+	fmt->sort	= dim->cmp;
+	fmt->color	= NULL;
+	fmt->entry	= NULL;
+	fmt->header	= NULL;
+	fmt->width	= NULL;
+	fmt->collapse	= dim->cmp;
+	fmt->equal	= fmt_equal;
+	fmt->free	= fmt_free;
+
+	return kvm_fmt;
+}
+
+static int kvm_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
+{
+	struct kvm_fmt *kvm_fmt = get_format(name);
+
+	if (!kvm_fmt) {
+		reset_dimensions();
+		return output_field_add(hpp_list, name);
+	}
+
+	perf_hpp_list__column_register(hpp_list, &kvm_fmt->fmt);
+	return 0;
+}
+
+static int kvm_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
+{
+	struct kvm_fmt *kvm_fmt = get_format(name);
+
+	if (!kvm_fmt) {
+		reset_dimensions();
+		return sort_dimension__add(hpp_list, name, NULL, 0);
+	}
+
+	perf_hpp_list__register_sort_field(hpp_list, &kvm_fmt->fmt);
+	return 0;
+}
+
+static int kvm_hpp_list__init(char *list,
+			      struct perf_hpp_list *hpp_list,
+			      int (*fn)(struct perf_hpp_list *hpp_list,
+					char *name))
+{
+	char *tmp, *tok;
+	int ret;
+
+	if (!list || !fn)
+		return 0;
+
+	for (tok = strtok_r(list, ", ", &tmp); tok;
+	     tok = strtok_r(NULL, ", ", &tmp)) {
+		ret = fn(hpp_list, tok);
+		if (!ret)
+			continue;
+
+		/* Handle errors */
+		if (ret == -EINVAL)
+			pr_err("Invalid field key: '%s'", tok);
+		else if (ret == -ESRCH)
+			pr_err("Unknown field key: '%s'", tok);
+		else
+			pr_err("Fail to initialize for field key: '%s'", tok);
+
+		break;
+	}
+
+	return ret;
+}
+
+static int kvm_hpp_list__parse(struct perf_hpp_list *hpp_list,
+			       const char *output_, const char *sort_)
+{
+	char *output = output_ ? strdup(output_) : NULL;
+	char *sort = sort_ ? strdup(sort_) : NULL;
+	int ret;
+
+	ret = kvm_hpp_list__init(output, hpp_list, kvm_hists__init_output);
+	if (ret)
+		goto out;
+
+	ret = kvm_hpp_list__init(sort, hpp_list, kvm_hists__init_sort);
+	if (ret)
+		goto out;
+
+	/* Copy sort keys to output fields */
+	perf_hpp__setup_output_field(hpp_list);
+
+out:
+	free(output);
+	free(sort);
+	return ret;
+}
+
+static int kvm_hists__init(struct perf_kvm_stat *kvm)
+{
+	const char *output_columns = "name,sample,time,max_t,min_t,mean_t";
+
 	__hists__init(&kvm_hists.hists, &kvm_hists.list);
 	perf_hpp_list__init(&kvm_hists.list);
-	return 0;
+	return kvm_hpp_list__parse(&kvm_hists.list, output_columns,
+				   kvm->sort_key);
 }
 
 static const char *get_filename_for_perf_kvm(void)
@@ -979,7 +1204,7 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
 	set_term_quiet_input(&save);
 	init_kvm_event_record(kvm);
 
-	kvm_hists__init();
+	kvm_hists__init(kvm);
 
 	signal(SIGINT, sig_handler);
 	signal(SIGTERM, sig_handler);
@@ -1176,7 +1401,7 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	init_kvm_event_record(kvm);
 	setup_pager();
 
-	kvm_hists__init();
+	kvm_hists__init(kvm);
 
 	ret = read_events(kvm);
 	if (ret)
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index 80d5c5a9ae31..ca5796959f66 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -40,6 +40,8 @@ struct kvm_event {
 	#define DEFAULT_VCPU_NUM 8
 	int max_vcpu;
 	struct kvm_event_stats *vcpu;
+
+	struct hist_entry he;
 };
 
 typedef int (*key_cmp_fun)(struct kvm_event*, struct kvm_event*, int);
-- 
2.34.1

Re: [PATCH v3 08/14] perf kvm: Add dimensions for KVM event statistics

[Namhyung Kim]

Hi Leo,

On Tue, Feb 28, 2023 at 3:53 AM Leo Yan <leo.yan@linaro.org> wrote:
>
> To support KVM event statistics, this patch firstly registers histograms
> columns and sorting fields; every column or field has its own format
> structure, the format structure is dereferenced to access the dimension,
> finally the dimension provides the comparison callback for sorting
> result.
>
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> Reviewed-by: James Clark <james.clark@arm.com>
> ---
>  tools/perf/builtin-kvm.c   | 239 +++++++++++++++++++++++++++++++++++--
>  tools/perf/util/kvm-stat.h |   2 +
>  2 files changed, 234 insertions(+), 7 deletions(-)
>
> diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
> index a9f467926bdd..da84f5063d4d 100644
> --- a/tools/perf/builtin-kvm.c
> +++ b/tools/perf/builtin-kvm.c
> @@ -70,9 +70,9 @@ static int cmp_event_ ## func(struct kvm_event *one,                  \
>                get_event_ ##func(two, vcpu);                            \
>  }
>
> -GET_EVENT_KEY(time, time);
> -GET_EVENT_KEY(max, stats.max);
> -GET_EVENT_KEY(min, stats.min);
> +COMPARE_EVENT_KEY(time, time);
> +COMPARE_EVENT_KEY(max, stats.max);
> +COMPARE_EVENT_KEY(min, stats.min);
>  COMPARE_EVENT_KEY(count, stats.n);
>  COMPARE_EVENT_KEY(mean, stats.mean);
>
> @@ -90,13 +90,238 @@ struct kvm_hists {
>         struct perf_hpp_list    list;
>  };
>
> +struct kvm_dimension {
> +       const char *name;
> +       int64_t (*cmp)(struct perf_hpp_fmt *fmt, struct hist_entry *left,
> +                      struct hist_entry *right);
> +};
> +
> +struct kvm_fmt {
> +       struct perf_hpp_fmt     fmt;
> +       struct kvm_dimension    *dim;
> +};
> +
>  static struct kvm_hists kvm_hists;
>
> -static int kvm_hists__init(void)
> +static int64_t
> +empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
> +         struct hist_entry *left __maybe_unused,
> +         struct hist_entry *right __maybe_unused)
> +{
> +       return 0;
> +}
> +
> +static struct kvm_dimension dim_event = {
> +       .name           = "name",
> +       .cmp            = empty_cmp,
> +};

I guess you can name it as 'dim_name' to reduce the confusion.
Also it can compare event names using strcmp() in case users
want to see event names in alphabetical order.

[SNIP]
> +
> +static int kvm_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
> +{
> +       struct kvm_fmt *kvm_fmt = get_format(name);
> +
> +       if (!kvm_fmt) {
> +               reset_dimensions();
> +               return output_field_add(hpp_list, name);

Hmm.. do you plan to support these generic output fields too?
I'm not sure you need reset_dimensions() here.

> +       }
> +
> +       perf_hpp_list__column_register(hpp_list, &kvm_fmt->fmt);
> +       return 0;
> +}
> +
> +static int kvm_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
> +{
> +       struct kvm_fmt *kvm_fmt = get_format(name);
> +
> +       if (!kvm_fmt) {
> +               reset_dimensions();
> +               return sort_dimension__add(hpp_list, name, NULL, 0);

Ditto.

> +       }
> +
> +       perf_hpp_list__register_sort_field(hpp_list, &kvm_fmt->fmt);
> +       return 0;
> +}
> +
> +static int kvm_hpp_list__init(char *list,
> +                             struct perf_hpp_list *hpp_list,
> +                             int (*fn)(struct perf_hpp_list *hpp_list,
> +                                       char *name))
> +{
> +       char *tmp, *tok;
> +       int ret;
> +
> +       if (!list || !fn)
> +               return 0;
> +
> +       for (tok = strtok_r(list, ", ", &tmp); tok;
> +            tok = strtok_r(NULL, ", ", &tmp)) {
> +               ret = fn(hpp_list, tok);
> +               if (!ret)
> +                       continue;
> +
> +               /* Handle errors */
> +               if (ret == -EINVAL)
> +                       pr_err("Invalid field key: '%s'", tok);
> +               else if (ret == -ESRCH)
> +                       pr_err("Unknown field key: '%s'", tok);
> +               else
> +                       pr_err("Fail to initialize for field key: '%s'", tok);
> +
> +               break;
> +       }
> +
> +       return ret;
> +}
> +
> +static int kvm_hpp_list__parse(struct perf_hpp_list *hpp_list,
> +                              const char *output_, const char *sort_)
> +{
> +       char *output = output_ ? strdup(output_) : NULL;
> +       char *sort = sort_ ? strdup(sort_) : NULL;
> +       int ret;
> +
> +       ret = kvm_hpp_list__init(output, hpp_list, kvm_hists__init_output);
> +       if (ret)
> +               goto out;
> +
> +       ret = kvm_hpp_list__init(sort, hpp_list, kvm_hists__init_sort);
> +       if (ret)
> +               goto out;
> +
> +       /* Copy sort keys to output fields */
> +       perf_hpp__setup_output_field(hpp_list);

I think you also need perf_hpp__append_sort_keys() as in
setup_sorting() to have secondary sort keys in case the
given sort key cannot determine the ordering.

Thanks,
Namhyung


> +
> +out:
> +       free(output);
> +       free(sort);
> +       return ret;
> +}
> +
> +static int kvm_hists__init(struct perf_kvm_stat *kvm)
> +{
> +       const char *output_columns = "name,sample,time,max_t,min_t,mean_t";
> +
>         __hists__init(&kvm_hists.hists, &kvm_hists.list);
>         perf_hpp_list__init(&kvm_hists.list);
> -       return 0;
> +       return kvm_hpp_list__parse(&kvm_hists.list, output_columns,
> +                                  kvm->sort_key);
>  }
>
>  static const char *get_filename_for_perf_kvm(void)
> @@ -979,7 +1204,7 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
>         set_term_quiet_input(&save);
>         init_kvm_event_record(kvm);
>
> -       kvm_hists__init();
> +       kvm_hists__init(kvm);
>
>         signal(SIGINT, sig_handler);
>         signal(SIGTERM, sig_handler);
> @@ -1176,7 +1401,7 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
>         init_kvm_event_record(kvm);
>         setup_pager();
>
> -       kvm_hists__init();
> +       kvm_hists__init(kvm);
>
>         ret = read_events(kvm);
>         if (ret)
> diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
> index 80d5c5a9ae31..ca5796959f66 100644
> --- a/tools/perf/util/kvm-stat.h
> +++ b/tools/perf/util/kvm-stat.h
> @@ -40,6 +40,8 @@ struct kvm_event {
>         #define DEFAULT_VCPU_NUM 8
>         int max_vcpu;
>         struct kvm_event_stats *vcpu;
> +
> +       struct hist_entry he;
>  };
>
>  typedef int (*key_cmp_fun)(struct kvm_event*, struct kvm_event*, int);
> --
> 2.34.1
>

Re: [PATCH v3 08/14] perf kvm: Add dimensions for KVM event statistics

[Leo Yan]

Hi Namhyung,

On Tue, Feb 28, 2023 at 03:34:28PM -0800, Namhyung Kim wrote:

[...]

> > +static struct kvm_dimension dim_event = {
> > +       .name           = "name",
> > +       .cmp            = empty_cmp,
> > +};
> 
> I guess you can name it as 'dim_name' to reduce the confusion.
> Also it can compare event names using strcmp() in case users
> want to see event names in alphabetical order.

I will rename as 'ev_name' or 'event_name'; and it's good point to
compare the name with string order.  Will do it.

[...]

> [SNIP]
> > +
> > +static int kvm_hists__init_output(struct perf_hpp_list *hpp_list, char *name)
> > +{
> > +       struct kvm_fmt *kvm_fmt = get_format(name);
> > +
> > +       if (!kvm_fmt) {
> > +               reset_dimensions();
> > +               return output_field_add(hpp_list, name);
> 
> Hmm.. do you plan to support these generic output fields too?
> I'm not sure you need reset_dimensions() here.

So far, we don't need to add any generic output field, I will remove
reset_dimensions() / output_field_add(), alternatively, in next spin I
will simply print an error info and return -EINVAL.

> > +       }
> > +
> > +       perf_hpp_list__column_register(hpp_list, &kvm_fmt->fmt);
> > +       return 0;
> > +}
> > +
> > +static int kvm_hists__init_sort(struct perf_hpp_list *hpp_list, char *name)
> > +{
> > +       struct kvm_fmt *kvm_fmt = get_format(name);
> > +
> > +       if (!kvm_fmt) {
> > +               reset_dimensions();
> > +               return sort_dimension__add(hpp_list, name, NULL, 0);
> 
> Ditto.

Will do the same change with the above statement.

[...]

> > +static int kvm_hpp_list__parse(struct perf_hpp_list *hpp_list,
> > +                              const char *output_, const char *sort_)
> > +{
> > +       char *output = output_ ? strdup(output_) : NULL;
> > +       char *sort = sort_ ? strdup(sort_) : NULL;
> > +       int ret;
> > +
> > +       ret = kvm_hpp_list__init(output, hpp_list, kvm_hists__init_output);
> > +       if (ret)
> > +               goto out;
> > +
> > +       ret = kvm_hpp_list__init(sort, hpp_list, kvm_hists__init_sort);
> > +       if (ret)
> > +               goto out;
> > +
> > +       /* Copy sort keys to output fields */
> > +       perf_hpp__setup_output_field(hpp_list);
> 
> I think you also need perf_hpp__append_sort_keys() as in
> setup_sorting() to have secondary sort keys in case the
> given sort key cannot determine the ordering.

Will do.

Thanks a lot for reviewing!

Leo

[PATCH v3 09/14] perf kvm: Use histograms list to replace cached list

[Leo Yan]

perf kvm tool defines its own cached list which is managed with RB tree,
histograms also provide RB tree to manage data entries.  Since now we
have introduced histograms in the tool, it's not necessary to use the
self defined list and we can directly use histograms list to manage
KVM events.

This patch changes to use histograms list to track KVM events, and it
invokes the common function hists__output_resort_cb() to sort result,
this also give us flexibility to extend more sorting key words easily.

After histograms list supported, the cached list is redundant so remove
the relevant code for it.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 186 +++++++++++++++++++------------------
 tools/perf/util/kvm-stat.h |   7 --
 2 files changed, 94 insertions(+), 99 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index da84f5063d4d..32dc697ff707 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -421,44 +421,37 @@ struct vcpu_event_record {
 	struct kvm_event *last_event;
 };
 
-
-static void init_kvm_event_record(struct perf_kvm_stat *kvm)
-{
-	unsigned int i;
-
-	for (i = 0; i < EVENTS_CACHE_SIZE; i++)
-		INIT_LIST_HEAD(&kvm->kvm_events_cache[i]);
-}
-
 #ifdef HAVE_TIMERFD_SUPPORT
-static void clear_events_cache_stats(struct list_head *kvm_events_cache)
+static void clear_events_cache_stats(void)
 {
-	struct list_head *head;
+	struct rb_root_cached *root;
+	struct rb_node *nd;
 	struct kvm_event *event;
-	unsigned int i;
-	int j;
-
-	for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
-		head = &kvm_events_cache[i];
-		list_for_each_entry(event, head, hash_entry) {
-			/* reset stats for event */
-			event->total.time = 0;
-			init_stats(&event->total.stats);
-
-			for (j = 0; j < event->max_vcpu; ++j) {
-				event->vcpu[j].time = 0;
-				init_stats(&event->vcpu[j].stats);
-			}
+	int i;
+
+	if (hists__has(&kvm_hists.hists, need_collapse))
+		root = &kvm_hists.hists.entries_collapsed;
+	else
+		root = kvm_hists.hists.entries_in;
+
+	for (nd = rb_first_cached(root); nd; nd = rb_next(nd)) {
+		struct hist_entry *he;
+
+		he = rb_entry(nd, struct hist_entry, rb_node_in);
+		event = container_of(he, struct kvm_event, he);
+
+		/* reset stats for event */
+		event->total.time = 0;
+		init_stats(&event->total.stats);
+
+		for (i = 0; i < event->max_vcpu; ++i) {
+			event->vcpu[i].time = 0;
+			init_stats(&event->vcpu[i].stats);
 		}
 	}
 }
 #endif
 
-static int kvm_events_hash_fn(u64 key)
-{
-	return key & (EVENTS_CACHE_SIZE - 1);
-}
-
 static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
 {
 	int old_max_vcpu = event->max_vcpu;
@@ -484,21 +477,51 @@ static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
 	return true;
 }
 
+static void *kvm_he_zalloc(size_t size)
+{
+	struct kvm_event *kvm_ev;
+
+	kvm_ev = zalloc(size + sizeof(*kvm_ev));
+	if (!kvm_ev)
+		return NULL;
+
+	return &kvm_ev->he;
+}
+
+static void kvm_he_free(void *he)
+{
+	struct kvm_event *kvm_ev;
+
+	kvm_ev = container_of(he, struct kvm_event, he);
+	free(kvm_ev);
+}
+
+static struct hist_entry_ops kvm_ev_entry_ops = {
+	.new	= kvm_he_zalloc,
+	.free	= kvm_he_free,
+};
+
 static struct kvm_event *kvm_alloc_init_event(struct perf_kvm_stat *kvm,
 					      struct event_key *key,
-					      struct perf_sample *sample __maybe_unused)
+					      struct perf_sample *sample)
 {
 	struct kvm_event *event;
+	struct hist_entry *he;
 
-	event = zalloc(sizeof(*event));
-	if (!event) {
-		pr_err("Not enough memory\n");
+	he = hists__add_entry_ops(&kvm_hists.hists, &kvm_ev_entry_ops,
+				  &kvm->al, NULL, NULL, NULL, sample, true);
+	if (he == NULL) {
+		pr_err("Failed to allocate hist entry\n");
 		return NULL;
 	}
 
+	hists__inc_nr_samples(&kvm_hists.hists, 0);
+
+	event = container_of(he, struct kvm_event, he);
 	event->perf_kvm = kvm;
 	event->key = *key;
 	init_stats(&event->total.stats);
+
 	return event;
 }
 
@@ -507,22 +530,26 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
 					       struct perf_sample *sample)
 {
 	struct kvm_event *event;
-	struct list_head *head;
+	struct rb_root_cached *root;
+	struct rb_node *nd;
 
 	BUG_ON(key->key == INVALID_KEY);
 
-	head = &kvm->kvm_events_cache[kvm_events_hash_fn(key->key)];
-	list_for_each_entry(event, head, hash_entry) {
+	if (hists__has(&kvm_hists.hists, need_collapse))
+		root = &kvm_hists.hists.entries_collapsed;
+	else
+		root = kvm_hists.hists.entries_in;
+
+	for (nd = rb_first_cached(root); nd; nd = rb_next(nd)) {
+		struct hist_entry *he = rb_entry(nd, struct hist_entry,
+						 rb_node_in);
+
+		event = container_of(he, struct kvm_event, he);
 		if (event->key.key == key->key && event->key.info == key->info)
 			return event;
 	}
 
-	event = kvm_alloc_init_event(kvm, key, sample);
-	if (!event)
-		return NULL;
-
-	list_add(&event->hash_entry, head);
-	return event;
+	return kvm_alloc_init_event(kvm, key, sample);
 }
 
 static bool handle_begin_event(struct perf_kvm_stat *kvm,
@@ -754,58 +781,29 @@ static bool select_key(struct perf_kvm_stat *kvm)
 	return false;
 }
 
-static void insert_to_result(struct rb_root *result, struct kvm_event *event,
-			     key_cmp_fun bigger, int vcpu)
-{
-	struct rb_node **rb = &result->rb_node;
-	struct rb_node *parent = NULL;
-	struct kvm_event *p;
-
-	while (*rb) {
-		p = container_of(*rb, struct kvm_event, rb);
-		parent = *rb;
-
-		if (bigger(event, p, vcpu) > 0)
-			rb = &(*rb)->rb_left;
-		else
-			rb = &(*rb)->rb_right;
-	}
-
-	rb_link_node(&event->rb, parent, rb);
-	rb_insert_color(&event->rb, result);
-}
-
 static bool event_is_valid(struct kvm_event *event, int vcpu)
 {
 	return !!get_event_count(event, vcpu);
 }
 
-static void sort_result(struct perf_kvm_stat *kvm)
+static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
 {
-	unsigned int i;
-	int vcpu = kvm->trace_vcpu;
 	struct kvm_event *event;
+	struct perf_kvm_stat *perf_kvm;
 
-	for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
-		list_for_each_entry(event, &kvm->kvm_events_cache[i], hash_entry) {
-			if (event_is_valid(event, vcpu)) {
-				insert_to_result(&kvm->result, event,
-						 kvm->compare, vcpu);
-			}
-		}
-	}
+	event = container_of(he, struct kvm_event, he);
+	perf_kvm = event->perf_kvm;
+	if (!event_is_valid(event, perf_kvm->trace_vcpu))
+		he->filtered = 1;
+	else
+		he->filtered = 0;
+	return 0;
 }
 
-/* returns left most element of result, and erase it */
-static struct kvm_event *pop_from_result(struct rb_root *result)
+static void sort_result(void)
 {
-	struct rb_node *node = rb_first(result);
-
-	if (!node)
-		return NULL;
-
-	rb_erase(node, result);
-	return container_of(node, struct kvm_event, rb);
+	hists__collapse_resort(&kvm_hists.hists, NULL);
+	hists__output_resort_cb(&kvm_hists.hists, NULL, filter_cb);
 }
 
 static void print_vcpu_info(struct perf_kvm_stat *kvm)
@@ -848,6 +846,7 @@ static void print_result(struct perf_kvm_stat *kvm)
 	char decode[decode_str_len];
 	struct kvm_event *event;
 	int vcpu = kvm->trace_vcpu;
+	struct rb_node *nd;
 
 	if (kvm->live) {
 		puts(CONSOLE_CLEAR);
@@ -866,9 +865,15 @@ static void print_result(struct perf_kvm_stat *kvm)
 	pr_info("%16s ", "Avg time");
 	pr_info("\n\n");
 
-	while ((event = pop_from_result(&kvm->result))) {
+	for (nd = rb_first_cached(&kvm_hists.hists.entries); nd; nd = rb_next(nd)) {
+		struct hist_entry *he;
 		u64 ecount, etime, max, min;
 
+		he = rb_entry(nd, struct hist_entry, rb_node);
+		if (he->filtered)
+			continue;
+
+		event = container_of(he, struct kvm_event, he);
 		ecount = get_event_count(event, vcpu);
 		etime = get_event_time(event, vcpu);
 		max = get_event_max(event, vcpu);
@@ -1142,11 +1147,11 @@ static int perf_kvm__handle_timerfd(struct perf_kvm_stat *kvm)
 		pr_debug("Missed timer beats: %" PRIu64 "\n", c-1);
 
 	/* update display */
-	sort_result(kvm);
+	sort_result();
 	print_result(kvm);
 
 	/* reset counts */
-	clear_events_cache_stats(kvm->kvm_events_cache);
+	clear_events_cache_stats();
 	kvm->total_count = 0;
 	kvm->total_time = 0;
 	kvm->lost_events = 0;
@@ -1202,8 +1207,6 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
 	}
 
 	set_term_quiet_input(&save);
-	init_kvm_event_record(kvm);
-
 	kvm_hists__init(kvm);
 
 	signal(SIGINT, sig_handler);
@@ -1250,7 +1253,7 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
 	evlist__disable(kvm->evlist);
 
 	if (err == 0) {
-		sort_result(kvm);
+		sort_result();
 		print_result(kvm);
 	}
 
@@ -1398,7 +1401,6 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	if (!register_kvm_events_ops(kvm))
 		goto exit;
 
-	init_kvm_event_record(kvm);
 	setup_pager();
 
 	kvm_hists__init(kvm);
@@ -1407,7 +1409,7 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	if (ret)
 		goto exit;
 
-	sort_result(kvm);
+	sort_result();
 	print_result(kvm);
 
 exit:
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index ca5796959f66..c38d320c7cbe 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -30,7 +30,6 @@ struct perf_kvm_stat;
 
 struct kvm_event {
 	struct list_head hash_entry;
-	struct rb_node rb;
 
 	struct perf_kvm_stat *perf_kvm;
 	struct event_key key;
@@ -75,9 +74,6 @@ struct exit_reasons_table {
 	const char *reason;
 };
 
-#define EVENTS_BITS		12
-#define EVENTS_CACHE_SIZE	(1UL << EVENTS_BITS)
-
 struct perf_kvm_stat {
 	struct perf_tool    tool;
 	struct record_opts  opts;
@@ -97,7 +93,6 @@ struct perf_kvm_stat {
 
 	struct kvm_events_ops *events_ops;
 	key_cmp_fun compare;
-	struct list_head kvm_events_cache[EVENTS_CACHE_SIZE];
 
 	u64 total_time;
 	u64 total_count;
@@ -106,8 +101,6 @@ struct perf_kvm_stat {
 
 	struct intlist *pid_list;
 
-	struct rb_root result;
-
 	int timerfd;
 	unsigned int display_time;
 	bool live;
-- 
2.34.1

Re: [PATCH v3 09/14] perf kvm: Use histograms list to replace cached list

[Namhyung Kim]

On Tue, Feb 28, 2023 at 3:53 AM Leo Yan <leo.yan@linaro.org> wrote:
>
> perf kvm tool defines its own cached list which is managed with RB tree,
> histograms also provide RB tree to manage data entries.  Since now we
> have introduced histograms in the tool, it's not necessary to use the
> self defined list and we can directly use histograms list to manage
> KVM events.
>
> This patch changes to use histograms list to track KVM events, and it
> invokes the common function hists__output_resort_cb() to sort result,
> this also give us flexibility to extend more sorting key words easily.
>
> After histograms list supported, the cached list is redundant so remove
> the relevant code for it.
>
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> Reviewed-by: James Clark <james.clark@arm.com>
> ---
>  tools/perf/builtin-kvm.c   | 186 +++++++++++++++++++------------------
>  tools/perf/util/kvm-stat.h |   7 --
>  2 files changed, 94 insertions(+), 99 deletions(-)
>
> diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
> index da84f5063d4d..32dc697ff707 100644
> --- a/tools/perf/builtin-kvm.c
> +++ b/tools/perf/builtin-kvm.c
> @@ -421,44 +421,37 @@ struct vcpu_event_record {
>         struct kvm_event *last_event;
>  };
>
> -
> -static void init_kvm_event_record(struct perf_kvm_stat *kvm)
> -{
> -       unsigned int i;
> -
> -       for (i = 0; i < EVENTS_CACHE_SIZE; i++)
> -               INIT_LIST_HEAD(&kvm->kvm_events_cache[i]);
> -}
> -
>  #ifdef HAVE_TIMERFD_SUPPORT
> -static void clear_events_cache_stats(struct list_head *kvm_events_cache)
> +static void clear_events_cache_stats(void)
>  {
> -       struct list_head *head;
> +       struct rb_root_cached *root;
> +       struct rb_node *nd;
>         struct kvm_event *event;
> -       unsigned int i;
> -       int j;
> -
> -       for (i = 0; i < EVENTS_CACHE_SIZE; i++) {
> -               head = &kvm_events_cache[i];
> -               list_for_each_entry(event, head, hash_entry) {
> -                       /* reset stats for event */
> -                       event->total.time = 0;
> -                       init_stats(&event->total.stats);
> -
> -                       for (j = 0; j < event->max_vcpu; ++j) {
> -                               event->vcpu[j].time = 0;
> -                               init_stats(&event->vcpu[j].stats);
> -                       }
> +       int i;
> +
> +       if (hists__has(&kvm_hists.hists, need_collapse))
> +               root = &kvm_hists.hists.entries_collapsed;
> +       else
> +               root = kvm_hists.hists.entries_in;
> +
> +       for (nd = rb_first_cached(root); nd; nd = rb_next(nd)) {
> +               struct hist_entry *he;
> +
> +               he = rb_entry(nd, struct hist_entry, rb_node_in);
> +               event = container_of(he, struct kvm_event, he);
> +
> +               /* reset stats for event */
> +               event->total.time = 0;
> +               init_stats(&event->total.stats);
> +
> +               for (i = 0; i < event->max_vcpu; ++i) {
> +                       event->vcpu[i].time = 0;
> +                       init_stats(&event->vcpu[i].stats);
>                 }
>         }
>  }
>  #endif
>
> -static int kvm_events_hash_fn(u64 key)
> -{
> -       return key & (EVENTS_CACHE_SIZE - 1);
> -}
> -
>  static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
>  {
>         int old_max_vcpu = event->max_vcpu;
> @@ -484,21 +477,51 @@ static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
>         return true;
>  }
>
> +static void *kvm_he_zalloc(size_t size)
> +{
> +       struct kvm_event *kvm_ev;
> +
> +       kvm_ev = zalloc(size + sizeof(*kvm_ev));
> +       if (!kvm_ev)
> +               return NULL;
> +
> +       return &kvm_ev->he;
> +}
> +
> +static void kvm_he_free(void *he)
> +{
> +       struct kvm_event *kvm_ev;
> +
> +       kvm_ev = container_of(he, struct kvm_event, he);
> +       free(kvm_ev);
> +}
> +
> +static struct hist_entry_ops kvm_ev_entry_ops = {
> +       .new    = kvm_he_zalloc,
> +       .free   = kvm_he_free,
> +};
> +
>  static struct kvm_event *kvm_alloc_init_event(struct perf_kvm_stat *kvm,
>                                               struct event_key *key,
> -                                             struct perf_sample *sample __maybe_unused)
> +                                             struct perf_sample *sample)
>  {
>         struct kvm_event *event;
> +       struct hist_entry *he;
>
> -       event = zalloc(sizeof(*event));
> -       if (!event) {
> -               pr_err("Not enough memory\n");
> +       he = hists__add_entry_ops(&kvm_hists.hists, &kvm_ev_entry_ops,
> +                                 &kvm->al, NULL, NULL, NULL, sample, true);
> +       if (he == NULL) {
> +               pr_err("Failed to allocate hist entry\n");
>                 return NULL;
>         }
>
> +       hists__inc_nr_samples(&kvm_hists.hists, 0);
> +
> +       event = container_of(he, struct kvm_event, he);
>         event->perf_kvm = kvm;
>         event->key = *key;
>         init_stats(&event->total.stats);
> +
>         return event;
>  }
>
> @@ -507,22 +530,26 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
>                                                struct perf_sample *sample)
>  {
>         struct kvm_event *event;
> -       struct list_head *head;
> +       struct rb_root_cached *root;
> +       struct rb_node *nd;
>
>         BUG_ON(key->key == INVALID_KEY);
>
> -       head = &kvm->kvm_events_cache[kvm_events_hash_fn(key->key)];
> -       list_for_each_entry(event, head, hash_entry) {
> +       if (hists__has(&kvm_hists.hists, need_collapse))
> +               root = &kvm_hists.hists.entries_collapsed;
> +       else
> +               root = kvm_hists.hists.entries_in;
> +
> +       for (nd = rb_first_cached(root); nd; nd = rb_next(nd)) {
> +               struct hist_entry *he = rb_entry(nd, struct hist_entry,
> +                                                rb_node_in);
> +
> +               event = container_of(he, struct kvm_event, he);
>                 if (event->key.key == key->key && event->key.info == key->info)
>                         return event;

This seems inefficient and even unnecessary.  You should find
the event based on the return value of hist_entry__cmp() from
the root and go down.

But I think that's what hists__add_entry_ops() does.  Maybe
you may need to move the init logic (like init_stats) to the
kvm_he_zalloc().

Thanks,
Namhyung


>         }
>
> -       event = kvm_alloc_init_event(kvm, key, sample);
> -       if (!event)
> -               return NULL;
> -
> -       list_add(&event->hash_entry, head);
> -       return event;
> +       return kvm_alloc_init_event(kvm, key, sample);
>  }

Re: [PATCH v3 09/14] perf kvm: Use histograms list to replace cached list

[Leo Yan]

On Tue, Feb 28, 2023 at 04:03:27PM -0800, Namhyung Kim wrote:

[...]

> > @@ -507,22 +530,26 @@ static struct kvm_event *find_create_kvm_event(struct perf_kvm_stat *kvm,
> >                                                struct perf_sample *sample)
> >  {
> >         struct kvm_event *event;
> > -       struct list_head *head;
> > +       struct rb_root_cached *root;
> > +       struct rb_node *nd;
> >
> >         BUG_ON(key->key == INVALID_KEY);
> >
> > -       head = &kvm->kvm_events_cache[kvm_events_hash_fn(key->key)];
> > -       list_for_each_entry(event, head, hash_entry) {
> > +       if (hists__has(&kvm_hists.hists, need_collapse))
> > +               root = &kvm_hists.hists.entries_collapsed;
> > +       else
> > +               root = kvm_hists.hists.entries_in;
> > +
> > +       for (nd = rb_first_cached(root); nd; nd = rb_next(nd)) {
> > +               struct hist_entry *he = rb_entry(nd, struct hist_entry,
> > +                                                rb_node_in);
> > +
> > +               event = container_of(he, struct kvm_event, he);
> >                 if (event->key.key == key->key && event->key.info == key->info)
> >                         return event;
> 
> This seems inefficient and even unnecessary.  You should find
> the event based on the return value of hist_entry__cmp() from
> the root and go down.
> 
> But I think that's what hists__add_entry_ops() does.  Maybe
> you may need to move the init logic (like init_stats) to the
> kvm_he_zalloc().

Indeed.  I will drop above code and rely on hists__add_entry_ops()
to check if have existed event entry.

Thanks a lot for suggestion!

Leo

[PATCH v3 10/14] perf kvm: Polish sorting key

[Leo Yan]

Since histograms supports sorting, the tool doesn't need to maintain the
mapping between the sorting keys and the corresponding comparison
callbacks, therefore, this patch removes structure kvm_event_key.

But we still need to validate the sorting key, this patch uses an array
for sorting keys and renames function select_key() to is_valid_key()
to validate the sorting key passed by user.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c   | 27 +++++++++------------------
 tools/perf/util/kvm-stat.h |  8 --------
 2 files changed, 9 insertions(+), 26 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 32dc697ff707..741ba65bf092 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -76,15 +76,6 @@ COMPARE_EVENT_KEY(min, stats.min);
 COMPARE_EVENT_KEY(count, stats.n);
 COMPARE_EVENT_KEY(mean, stats.mean);
 
-#define DEF_SORT_NAME_KEY(name, compare_key)				\
-	{ #name, cmp_event_ ## compare_key }
-
-static struct kvm_event_key keys[] = {
-	DEF_SORT_NAME_KEY(sample, count),
-	DEF_SORT_NAME_KEY(time, mean),
-	{ NULL, NULL }
-};
-
 struct kvm_hists {
 	struct hists		hists;
 	struct perf_hpp_list	list;
@@ -766,18 +757,18 @@ static bool handle_kvm_event(struct perf_kvm_stat *kvm,
 	return true;
 }
 
-static bool select_key(struct perf_kvm_stat *kvm)
+static bool is_valid_key(struct perf_kvm_stat *kvm)
 {
+	const char *key_array[] = {
+		"sample", "time", "max_t", "min_t", "mean_t",
+	};
 	int i;
 
-	for (i = 0; keys[i].name; i++) {
-		if (!strcmp(keys[i].name, kvm->sort_key)) {
-			kvm->compare = keys[i].key;
+	for (i = 0; ARRAY_SIZE(key_array); i++)
+		if (!strcmp(key_array[i], kvm->sort_key))
 			return true;
-		}
-	}
 
-	pr_err("Unknown compare key:%s\n", kvm->sort_key);
+	pr_err("Unsupported key: %s\n", kvm->sort_key);
 	return false;
 }
 
@@ -1201,7 +1192,7 @@ static int kvm_events_live_report(struct perf_kvm_stat *kvm)
 		return ret;
 
 	if (!verify_vcpu(kvm->trace_vcpu) ||
-	    !select_key(kvm) ||
+	    !is_valid_key(kvm) ||
 	    !register_kvm_events_ops(kvm)) {
 		goto out;
 	}
@@ -1395,7 +1386,7 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	if (!verify_vcpu(vcpu))
 		goto exit;
 
-	if (!select_key(kvm))
+	if (!is_valid_key(kvm))
 		goto exit;
 
 	if (!register_kvm_events_ops(kvm))
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index c38d320c7cbe..a8e919ca59f4 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -43,13 +43,6 @@ struct kvm_event {
 	struct hist_entry he;
 };
 
-typedef int (*key_cmp_fun)(struct kvm_event*, struct kvm_event*, int);
-
-struct kvm_event_key {
-	const char *name;
-	key_cmp_fun key;
-};
-
 struct child_event_ops {
 	void (*get_key)(struct evsel *evsel,
 			struct perf_sample *sample,
@@ -92,7 +85,6 @@ struct perf_kvm_stat {
 	const char *exit_reasons_isa;
 
 	struct kvm_events_ops *events_ops;
-	key_cmp_fun compare;
 
 	u64 total_time;
 	u64 total_count;
-- 
2.34.1

[PATCH v3 11/14] perf kvm: Support printing attributions for dimensions

[Leo Yan]

This patch adds header, entry callback and width for every dimension,
thus in TUI mode the tool can print items with the defined attributions.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 105 +++++++++++++++++++++++++++++++++++++--
 1 file changed, 102 insertions(+), 3 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 741ba65bf092..a56d0983c585 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -83,8 +83,12 @@ struct kvm_hists {
 
 struct kvm_dimension {
 	const char *name;
+	const char *header;
+	int width;
 	int64_t (*cmp)(struct perf_hpp_fmt *fmt, struct hist_entry *left,
 		       struct hist_entry *right);
+	int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
+		     struct hist_entry *he);
 };
 
 struct kvm_fmt {
@@ -102,9 +106,32 @@ empty_cmp(struct perf_hpp_fmt *fmt __maybe_unused,
 	return 0;
 }
 
+static int fmt_width(struct perf_hpp_fmt *fmt,
+		     struct perf_hpp *hpp __maybe_unused,
+		     struct hists *hists __maybe_unused);
+
+static int ev_name_entry(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
+			 struct hist_entry *he)
+{
+	struct kvm_event *event;
+	int width = fmt_width(fmt, hpp, he->hists);
+	char decode[decode_str_len];
+	struct perf_kvm_stat *perf_kvm;
+
+	event = container_of(he, struct kvm_event, he);
+
+	perf_kvm = event->perf_kvm;
+	perf_kvm->events_ops->decode_key(perf_kvm, &event->key, decode);
+
+	return scnprintf(hpp->buf, hpp->size, "%*s", width, decode);
+}
+
 static struct kvm_dimension dim_event = {
+	.header		= "Event name",
 	.name		= "name",
 	.cmp		= empty_cmp,
+	.entry		= ev_name_entry,
+	.width		= 40,
 };
 
 #define EV_METRIC_CMP(metric)						\
@@ -130,29 +157,77 @@ EV_METRIC_CMP(max)
 EV_METRIC_CMP(min)
 EV_METRIC_CMP(mean)
 
+#define EV_METRIC_ENTRY(metric)						\
+static int ev_entry_##metric(struct perf_hpp_fmt *fmt,			\
+			     struct perf_hpp *hpp,			\
+			     struct hist_entry *he)			\
+{									\
+	struct kvm_event *event;					\
+	int width = fmt_width(fmt, hpp, he->hists);			\
+	struct perf_kvm_stat *perf_kvm;					\
+									\
+	event = container_of(he, struct kvm_event, he);			\
+	perf_kvm = event->perf_kvm;					\
+	return scnprintf(hpp->buf, hpp->size, "%*lu", width,		\
+		get_event_##metric(event, perf_kvm->trace_vcpu));	\
+}
+
+EV_METRIC_ENTRY(time)
+EV_METRIC_ENTRY(count)
+EV_METRIC_ENTRY(max)
+EV_METRIC_ENTRY(min)
+
 static struct kvm_dimension dim_time = {
+	.header		= "Time (ns)",
 	.name		= "time",
 	.cmp		= ev_cmp_time,
+	.entry		= ev_entry_time,
+	.width		= 12,
 };
 
 static struct kvm_dimension dim_count = {
+	.header		= "Samples",
 	.name		= "sample",
 	.cmp		= ev_cmp_count,
+	.entry		= ev_entry_count,
+	.width		= 12,
 };
 
 static struct kvm_dimension dim_max_time = {
+	.header		= "Max Time (ns)",
 	.name		= "max_t",
 	.cmp		= ev_cmp_max,
+	.entry		= ev_entry_max,
+	.width		= 14,
 };
 
 static struct kvm_dimension dim_min_time = {
+	.header		= "Min Time (ns)",
 	.name		= "min_t",
 	.cmp		= ev_cmp_min,
+	.entry		= ev_entry_min,
+	.width		= 14,
 };
 
+static int ev_entry_mean(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
+			 struct hist_entry *he)
+{
+	struct kvm_event *event;
+	int width = fmt_width(fmt, hpp, he->hists);
+	struct perf_kvm_stat *perf_kvm;
+
+	event = container_of(he, struct kvm_event, he);
+	perf_kvm = event->perf_kvm;
+	return scnprintf(hpp->buf, hpp->size, "%*lu", width,
+			 get_event_mean(event, perf_kvm->trace_vcpu));
+}
+
 static struct kvm_dimension dim_mean_time = {
+	.header		= "Mean Time (ns)",
 	.name		= "mean_t",
 	.cmp		= ev_cmp_mean,
+	.entry		= ev_entry_mean,
+	.width		= 14,
 };
 
 static struct kvm_dimension *dimensions[] = {
@@ -165,6 +240,30 @@ static struct kvm_dimension *dimensions[] = {
 	NULL,
 };
 
+static int fmt_width(struct perf_hpp_fmt *fmt,
+		     struct perf_hpp *hpp __maybe_unused,
+		     struct hists *hists __maybe_unused)
+{
+	struct kvm_fmt *kvm_fmt;
+
+	kvm_fmt = container_of(fmt, struct kvm_fmt, fmt);
+	return kvm_fmt->dim->width;
+}
+
+static int fmt_header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
+		      struct hists *hists, int line __maybe_unused,
+		      int *span __maybe_unused)
+{
+	struct kvm_fmt *kvm_fmt;
+	struct kvm_dimension *dim;
+	int width = fmt_width(fmt, hpp, hists);
+
+	kvm_fmt = container_of(fmt, struct kvm_fmt, fmt);
+	dim = kvm_fmt->dim;
+
+	return scnprintf(hpp->buf, hpp->size, "%*s", width, dim->header);
+}
+
 static bool fmt_equal(struct perf_hpp_fmt *a, struct perf_hpp_fmt *b)
 {
 	struct kvm_fmt *kvm_fmt_a = container_of(a, struct kvm_fmt, fmt);
@@ -214,9 +313,9 @@ static struct kvm_fmt *get_format(const char *name)
 	fmt->cmp	= dim->cmp;
 	fmt->sort	= dim->cmp;
 	fmt->color	= NULL;
-	fmt->entry	= NULL;
-	fmt->header	= NULL;
-	fmt->width	= NULL;
+	fmt->entry	= dim->entry;
+	fmt->header	= fmt_header;
+	fmt->width	= fmt_width;
 	fmt->collapse	= dim->cmp;
 	fmt->equal	= fmt_equal;
 	fmt->free	= fmt_free;
-- 
2.34.1

[PATCH v3 12/14] perf kvm: Add dimensions for percentages

[Leo Yan]

Add dimensions for count and time percentages, it would useful for user
to review percentage statistics.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/builtin-kvm.c | 98 +++++++++++++++++++++++++++++++++++++++-
 1 file changed, 97 insertions(+), 1 deletion(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index a56d0983c585..5b1b2042dfed 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -230,10 +230,105 @@ static struct kvm_dimension dim_mean_time = {
 	.width		= 14,
 };
 
+#define PERC_STR(__s, __v)				\
+({							\
+	scnprintf(__s, sizeof(__s), "%.2F%%", __v);	\
+	__s;						\
+})
+
+static double percent(u64 st, u64 tot)
+{
+	return tot ? 100. * (double) st / (double) tot : 0;
+}
+
+#define EV_METRIC_PERCENT(metric)					\
+static int ev_percent_##metric(struct hist_entry *he)			\
+{									\
+	struct kvm_event *event;					\
+	struct perf_kvm_stat *perf_kvm;					\
+									\
+	event = container_of(he, struct kvm_event, he);			\
+	perf_kvm = event->perf_kvm;					\
+									\
+	return percent(get_event_##metric(event, perf_kvm->trace_vcpu),	\
+		       perf_kvm->total_##metric);			\
+}
+
+EV_METRIC_PERCENT(time)
+EV_METRIC_PERCENT(count)
+
+static int ev_entry_time_precent(struct perf_hpp_fmt *fmt,
+				 struct perf_hpp *hpp,
+				 struct hist_entry *he)
+{
+	int width = fmt_width(fmt, hpp, he->hists);
+	double per;
+	char buf[10];
+
+	per = ev_percent_time(he);
+	return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
+}
+
+static int64_t
+ev_cmp_time_precent(struct perf_hpp_fmt *fmt __maybe_unused,
+		    struct hist_entry *left, struct hist_entry *right)
+{
+	double per_left;
+	double per_right;
+
+	per_left  = ev_percent_time(left);
+	per_right = ev_percent_time(right);
+
+	return per_left - per_right;
+}
+
+static struct kvm_dimension dim_time_percent = {
+	.header		= "Time%",
+	.name		= "percent_time",
+	.cmp		= ev_cmp_time_precent,
+	.entry		= ev_entry_time_precent,
+	.width		= 12,
+};
+
+static int ev_entry_count_precent(struct perf_hpp_fmt *fmt,
+				  struct perf_hpp *hpp,
+				  struct hist_entry *he)
+{
+	int width = fmt_width(fmt, hpp, he->hists);
+	double per;
+	char buf[10];
+
+	per = ev_percent_count(he);
+	return scnprintf(hpp->buf, hpp->size, "%*s", width, PERC_STR(buf, per));
+}
+
+static int64_t
+ev_cmp_count_precent(struct perf_hpp_fmt *fmt __maybe_unused,
+		     struct hist_entry *left, struct hist_entry *right)
+{
+	double per_left;
+	double per_right;
+
+	per_left  = ev_percent_count(left);
+	per_right = ev_percent_count(right);
+
+	return per_left - per_right;
+}
+
+static struct kvm_dimension dim_count_percent = {
+	.header		= "Sample%",
+	.name		= "percent_sample",
+	.cmp		= ev_cmp_count_precent,
+	.entry		= ev_entry_count_precent,
+	.width		= 12,
+};
+
 static struct kvm_dimension *dimensions[] = {
 	&dim_event,
 	&dim_time,
+	&dim_time_percent,
 	&dim_count,
+	&dim_count_percent,
 	&dim_max_time,
 	&dim_min_time,
 	&dim_mean_time,
@@ -406,7 +501,8 @@ static int kvm_hpp_list__parse(struct perf_hpp_list *hpp_list,
 
 static int kvm_hists__init(struct perf_kvm_stat *kvm)
 {
-	const char *output_columns = "name,sample,time,max_t,min_t,mean_t";
+	const char *output_columns = "name,sample,percent_sample,"
+				     "time,percent_time,max_t,min_t,mean_t";
 
 	__hists__init(&kvm_hists.hists, &kvm_hists.list);
 	perf_hpp_list__init(&kvm_hists.list);
-- 
2.34.1

[PATCH v3 13/14] perf kvm: Add TUI mode for stat report

[Leo Yan]

Since we have supported histograms list and prepared the dimensions in
the tool, this patch adds TUI mode for stat report.  It also adds UI
progress for sorting for better user experience.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 tools/perf/builtin-kvm.c   | 110 ++++++++++++++++++++++++++++++++++++-
 tools/perf/util/kvm-stat.h |   1 +
 2 files changed, 109 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index 5b1b2042dfed..56ad3346ba08 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -23,6 +23,8 @@
 #include "util/data.h"
 #include "util/ordered-events.h"
 #include "util/kvm-stat.h"
+#include "ui/browsers/hists.h"
+#include "ui/progress.h"
 #include "ui/ui.h"
 #include "util/string2.h"
 
@@ -506,10 +508,98 @@ static int kvm_hists__init(struct perf_kvm_stat *kvm)
 
 	__hists__init(&kvm_hists.hists, &kvm_hists.list);
 	perf_hpp_list__init(&kvm_hists.list);
+	kvm_hists.list.nr_header_lines = 1;
 	return kvm_hpp_list__parse(&kvm_hists.list, output_columns,
 				   kvm->sort_key);
 }
 
+static void print_result(struct perf_kvm_stat *kvm);
+
+#ifdef HAVE_SLANG_SUPPORT
+static void kvm_browser__update_nr_entries(struct hist_browser *hb)
+{
+	struct rb_node *nd = rb_first_cached(&hb->hists->entries);
+	u64 nr_entries = 0;
+
+	for (; nd; nd = rb_next(nd)) {
+		struct hist_entry *he = rb_entry(nd, struct hist_entry,
+						 rb_node);
+
+		if (!he->filtered)
+			nr_entries++;
+	}
+
+	hb->nr_non_filtered_entries = nr_entries;
+}
+
+static int kvm_browser__title(struct hist_browser *browser,
+			      char *buf, size_t size)
+{
+	scnprintf(buf, size, "KVM event statistics (%lu entries)",
+		  browser->nr_non_filtered_entries);
+	return 0;
+}
+
+static struct hist_browser*
+perf_kvm_browser__new(struct hists *hists)
+{
+	struct hist_browser *browser = hist_browser__new(hists);
+
+	if (browser)
+		browser->title = kvm_browser__title;
+
+	return browser;
+}
+
+static int kvm__hists_browse(struct hists *hists)
+{
+	struct hist_browser *browser;
+	int key = -1;
+
+	browser = perf_kvm_browser__new(hists);
+	if (browser == NULL)
+		return -1;
+
+	/* reset abort key so that it can get Ctrl-C as a key */
+	SLang_reset_tty();
+	SLang_init_tty(0, 0, 0);
+
+	kvm_browser__update_nr_entries(browser);
+
+	while (1) {
+		key = hist_browser__run(browser, "? - help", true, 0);
+
+		switch (key) {
+		case 'q':
+			goto out;
+		default:
+			break;
+		}
+	}
+
+out:
+	hist_browser__delete(browser);
+	return 0;
+}
+
+static void kvm_display(struct perf_kvm_stat *kvm)
+{
+	if (!use_browser)
+		print_result(kvm);
+	else
+		kvm__hists_browse(&kvm_hists.hists);
+}
+
+#else
+
+static void kvm_display(struct perf_kvm_stat *kvm)
+{
+	use_browser = 0;
+	print_result(kvm);
+}
+
+#endif /* HAVE_SLANG_SUPPORT */
+
 static const char *get_filename_for_perf_kvm(void)
 {
 	const char *filename;
@@ -988,8 +1078,12 @@ static int filter_cb(struct hist_entry *he, void *arg __maybe_unused)
 
 static void sort_result(void)
 {
+	struct ui_progress prog;
+
+	ui_progress__init(&prog, kvm_hists.hists.nr_entries, "Sorting...");
 	hists__collapse_resort(&kvm_hists.hists, NULL);
 	hists__output_resort_cb(&kvm_hists.hists, NULL, filter_cb);
+	ui_progress__finish();
 }
 
 static void print_vcpu_info(struct perf_kvm_stat *kvm)
@@ -1587,7 +1681,14 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 	if (!register_kvm_events_ops(kvm))
 		goto exit;
 
-	setup_pager();
+	if (kvm->use_stdio) {
+		use_browser = 0;
+		setup_pager();
+	} else {
+		use_browser = 1;
+	}
+
+	setup_browser(false);
 
 	kvm_hists__init(kvm);
 
@@ -1596,7 +1697,7 @@ static int kvm_events_report_vcpu(struct perf_kvm_stat *kvm)
 		goto exit;
 
 	sort_result();
-	print_result(kvm);
+	kvm_display(kvm);
 
 exit:
 	return ret;
@@ -1703,6 +1804,7 @@ kvm_events_report(struct perf_kvm_stat *kvm, int argc, const char **argv)
 		OPT_STRING('p', "pid", &kvm->opts.target.pid, "pid",
 			   "analyze events only for given process id(s)"),
 		OPT_BOOLEAN('f', "force", &kvm->force, "don't complain, do it"),
+		OPT_BOOLEAN(0, "stdio", &kvm->use_stdio, "use the stdio interface"),
 		OPT_END()
 	};
 
@@ -1720,6 +1822,10 @@ kvm_events_report(struct perf_kvm_stat *kvm, int argc, const char **argv)
 					   kvm_events_report_options);
 	}
 
+#ifndef HAVE_SLANG_SUPPORT
+	kvm->use_stdio = true;
+#endif
+
 	if (!kvm->opts.target.pid)
 		kvm->opts.target.system_wide = true;
 
diff --git a/tools/perf/util/kvm-stat.h b/tools/perf/util/kvm-stat.h
index a8e919ca59f4..b66eb4e8c547 100644
--- a/tools/perf/util/kvm-stat.h
+++ b/tools/perf/util/kvm-stat.h
@@ -97,6 +97,7 @@ struct perf_kvm_stat {
 	unsigned int display_time;
 	bool live;
 	bool force;
+	bool use_stdio;
 };
 
 struct kvm_reg_events_ops {
-- 
2.34.1

[PATCH v3 14/14] perf kvm: Update documentation to reflect new changes

[Leo Yan]

Update documentation for new sorting and option '--stdio'.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
Reviewed-by: James Clark <james.clark@arm.com>
---
 tools/perf/Documentation/perf-kvm.txt | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/tools/perf/Documentation/perf-kvm.txt b/tools/perf/Documentation/perf-kvm.txt
index 2ad3f5d9f72b..b66be66fe836 100644
--- a/tools/perf/Documentation/perf-kvm.txt
+++ b/tools/perf/Documentation/perf-kvm.txt
@@ -58,7 +58,7 @@ There are a couple of variants of perf kvm:
   events.
 
   'perf kvm stat report' reports statistical data which includes events
-  handled time, samples, and so on.
+  handled sample, percent_sample, time, percent_time, max_t, min_t, mean_t.
 
   'perf kvm stat live' reports statistical data in a live mode (similar to
   record + report but with statistical data updated live at a given display
@@ -82,6 +82,8 @@ OPTIONS
 :GMEXAMPLESUBCMD: top
 include::guest-files.txt[]
 
+--stdio:: Use the stdio interface.
+
 -v::
 --verbose::
 	Be more verbose (show counter open errors, etc).
@@ -97,7 +99,10 @@ STAT REPORT OPTIONS
 -k::
 --key=<value>::
        Sorting key. Possible values: sample (default, sort by samples
-       number), time (sort by average time).
+       number), percent_sample (sort by sample percentage), time
+       (sort by average time), precent_time (sort by time percentage),
+       max_t (sort by maximum time), min_t (sort by minimum time), mean_t
+       (sort by mean time).
 -p::
 --pid=::
     Analyze events only for given process ID(s) (comma separated list).
-- 
2.34.1