Skip to contentSkip to conversation

perf: Add branch stack counters

PATCH In review [email protected] 3 participants 20 messages Oct 25, 2023 – Nov 9, 2023

Subsystems (3)
lore
Share & tools
lore
#1 lore raw

PATCH v5 2/8 perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

In reply to [email protected]

From: Kan Liang <[email protected]>

Currently, branch_sample_type !=0 is used to check whether a branch stack setup is required. But it doesn't check the sample type, unnecessary branch stack setup may be done for a counting event. E.g., perf record -e "{branch-instructions,branch-misses}:S" -j any Also, the event only with the new PERF_SAMPLE_BRANCH_COUNTERS branch sample type may not require a branch stack setup either.

Add a new flag NEEDS_BRANCH_STACK to indicate whether the event requires a branch stack setup. Replace the needs_branch_stack() by checking the new flag.

The counting event check is implemented here. The later patch will take the new PERF_SAMPLE_BRANCH_COUNTERS into account.

---

No changes since V4

 arch/x86/events/intel/core.c       | 14 +++++++++++---
 arch/x86/events/perf_event_flags.h |  1 +
 2 files changed, 12 insertions(+), 3 deletions(-)
arch/x86/events/intel/core.c +11 -3
index 41a164764a84..a99449c0d77c 100644
@@ -2527,9 +2527,14 @@ static void intel_pmu_assign_event(struct perf_event *event, int idx)
 		perf_report_aux_output_id(event, idx);
 }
 
+static __always_inline bool intel_pmu_needs_branch_stack(struct perf_event *event)
+{
+	return event->hw.flags & PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+}
+
 static void intel_pmu_del_event(struct perf_event *event)
 {
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_del(event);
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_del(event);
@@ -2820,7 +2825,7 @@ static void intel_pmu_add_event(struct perf_event *event)
 {
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_add(event);
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_add(event);
 }
 
@@ -3897,7 +3902,10 @@ static int intel_pmu_hw_config(struct perf_event *event)
 			x86_pmu.pebs_aliases(event);
 	}
 
-	if (needs_branch_stack(event)) {
+	if (needs_branch_stack(event) && is_sampling_event(event))
+		event->hw.flags  |= PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+
+	if (intel_pmu_needs_branch_stack(event)) {
 		ret = intel_pmu_setup_lbr_filter(event);
 		if (ret)
 			return ret;
arch/x86/events/perf_event_flags.h +1 -0
index 1dc19b9b4426..a1685981c520 100644
@@ -20,3 +20,4 @@ PERF_ARCH(TOPDOWN,		0x04000) /* Count Topdown slots/metrics events */
 PERF_ARCH(PEBS_STLAT,		0x08000) /* st+stlat data address sampling */
 PERF_ARCH(AMD_BRS,		0x10000) /* AMD Branch Sampling */
 PERF_ARCH(PEBS_LAT_HYBRID,	0x20000) /* ld and st lat for hybrid */
+PERF_ARCH(NEEDS_BRANCH_STACK,	0x40000) /* require branch stack setup */
Signature · 1 line
2.35.1
#11 lore raw

PATCH v5 3/8 perf: Add branch_sample_call_stack

In reply to [email protected]

From: Kan Liang <[email protected]>

Add a helper function to check call stack sample type.

The later patch will invoke the function in several places.

---

No changes since V4

 arch/x86/events/core.c     | 2 +-
 include/linux/perf_event.h | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)
arch/x86/events/core.c +1 -1
index 40c9af124128..09050641ce5d 100644
@@ -601,7 +601,7 @@ int x86_pmu_hw_config(struct perf_event *event)
 		}
 	}
 
-	if (event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_CALL_STACK)
+	if (branch_sample_call_stack(event))
 		event->attach_state |= PERF_ATTACH_TASK_DATA;
 
 	/*
include/linux/perf_event.h +5 -0
index 7897ef066027..ac1a59c1f252 100644
@@ -1144,6 +1144,11 @@ static inline bool branch_sample_counters(const struct perf_event *event)
 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS;
 }
 
+static inline bool branch_sample_call_stack(const struct perf_event *event)
+{
+	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_CALL_STACK;
+}
+
 struct perf_sample_data {
 	/*
 	 * Fields set by perf_sample_data_init() unconditionally,
Signature · 1 line
2.35.1
#0 lore raw

PATCH v5 1/8 perf: Add branch stack counters

From: Kan Liang <[email protected]>

Currently, the additional information of a branch entry is stored in a u64 space. With more and more information added, the space is running out. For example, the information of occurrences of events will be added for each branch.

Two places were suggested to append the counters.
https://lore.kernel.org/lkml/[email protected]/
One place is right after the flags of each branch entry. It changes the existing struct perf_branch_entry. The later ARCH specific implementation has to be really careful to consistently pick the right struct.
The other place is right after the entire struct perf_branch_stack. The disadvantage is that the pointer of the extra space has to be recorded. The common interface perf_sample_save_brstack() has to be updated.

The latter is much straightforward, and should be easily understood and maintained. It is implemented in the patch.

Add a new branch sample type, PERF_SAMPLE_BRANCH_COUNTERS, to indicate the event which is recorded in the branch info.

The "u64 counters" may store the occurrences of several events. The information regarding the number of events/counters and the width of each counter should be exposed via sysfs as a reference for the perf tool. Define the branch_counter_nr and branch_counter_width ABI here. The support will be implemented later in the Intel-specific patch.

---

No changes since V4

 .../testing/sysfs-bus-event_source-devices-caps |  6 ++++++
 arch/powerpc/perf/core-book3s.c                 |  2 +-
 arch/x86/events/amd/core.c                      |  2 +-
 arch/x86/events/core.c                          |  2 +-
 arch/x86/events/intel/core.c                    |  2 +-
 arch/x86/events/intel/ds.c                      |  4 ++--
 include/linux/perf_event.h                      | 17 ++++++++++++++++-
 include/uapi/linux/perf_event.h                 | 10 ++++++++++
 kernel/events/core.c                            |  8 ++++++++
 9 files changed, 46 insertions(+), 7 deletions(-)
Documentation/ABI/testing/sysfs-bus-event_source-devices-caps +6 -0
index 8757dcf41c08..a5f506f7d481 100644
@@ -16,3 +16,9 @@ Description:
 		Example output in powerpc:
 		grep . /sys/bus/event_source/devices/cpu/caps/*
 		/sys/bus/event_source/devices/cpu/caps/pmu_name:POWER9
+
+		The "branch_counter_nr" in the supported platform exposes the
+		maximum number of counters which can be shown in the u64 counters
+		of PERF_SAMPLE_BRANCH_COUNTERS, while the "branch_counter_width"
+		exposes the width of each counter. Both of them can be used by
+		the perf tool to parse the logged counters in each branch.
arch/powerpc/perf/core-book3s.c +1 -1
index 8c1f7def596e..3c14596bbfaf 100644
@@ -2313,7 +2313,7 @@ static void record_and_restart(struct perf_event *event, unsigned long val,
 			struct cpu_hw_events *cpuhw;
 			cpuhw = this_cpu_ptr(&cpu_hw_events);
 			power_pmu_bhrb_read(event, cpuhw);
-			perf_sample_save_brstack(&data, event, &cpuhw->bhrb_stack);
+			perf_sample_save_brstack(&data, event, &cpuhw->bhrb_stack, NULL);
 		}
 
 		if (event->attr.sample_type & PERF_SAMPLE_DATA_SRC &&
arch/x86/events/amd/core.c +1 -1
index e24976593a29..4ee6390b45c9 100644
@@ -940,7 +940,7 @@ static int amd_pmu_v2_handle_irq(struct pt_regs *regs)
 			continue;
 
 		if (has_branch_stack(event))
-			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack);
+			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack, NULL);
 
 		if (perf_event_overflow(event, &data, regs))
 			x86_pmu_stop(event, 0);
arch/x86/events/core.c +1 -1
index 40ad1425ffa2..40c9af124128 100644
@@ -1702,7 +1702,7 @@ int x86_pmu_handle_irq(struct pt_regs *regs)
 		perf_sample_data_init(&data, 0, event->hw.last_period);
 
 		if (has_branch_stack(event))
-			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack);
+			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack, NULL);
 
 		if (perf_event_overflow(event, &data, regs))
 			x86_pmu_stop(event, 0);
arch/x86/events/intel/core.c +1 -1
index a08f794a0e79..41a164764a84 100644
@@ -3047,7 +3047,7 @@ static int handle_pmi_common(struct pt_regs *regs, u64 status)
 		perf_sample_data_init(&data, 0, event->hw.last_period);
 
 		if (has_branch_stack(event))
-			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack);
+			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack, NULL);
 
 		if (perf_event_overflow(event, &data, regs))
 			x86_pmu_stop(event, 0);
arch/x86/events/intel/ds.c +2 -2
index dc8204afec76..1d309e97a629 100644
@@ -1755,7 +1755,7 @@ static void setup_pebs_fixed_sample_data(struct perf_event *event,
 		setup_pebs_time(event, data, pebs->tsc);
 
 	if (has_branch_stack(event))
-		perf_sample_save_brstack(data, event, &cpuc->lbr_stack);
+		perf_sample_save_brstack(data, event, &cpuc->lbr_stack, NULL);
 }
 
 static void adaptive_pebs_save_regs(struct pt_regs *regs,
@@ -1912,7 +1912,7 @@ static void setup_pebs_adaptive_sample_data(struct perf_event *event,
 
 		if (has_branch_stack(event)) {
 			intel_pmu_store_pebs_lbrs(lbr);
-			perf_sample_save_brstack(data, event, &cpuc->lbr_stack);
+			perf_sample_save_brstack(data, event, &cpuc->lbr_stack, NULL);
 		}
 	}
 
include/linux/perf_event.h +16 -1
index 0367d748fae0..7897ef066027 100644
@@ -1139,6 +1139,10 @@ static inline bool branch_sample_priv(const struct perf_event *event)
 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_PRIV_SAVE;
 }
 
+static inline bool branch_sample_counters(const struct perf_event *event)
+{
+	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS;
+}
 
 struct perf_sample_data {
 	/*
@@ -1173,6 +1177,7 @@ struct perf_sample_data {
 	struct perf_callchain_entry	*callchain;
 	struct perf_raw_record		*raw;
 	struct perf_branch_stack	*br_stack;
+	u64				*br_stack_cntr;
 	union perf_sample_weight	weight;
 	union  perf_mem_data_src	data_src;
 	u64				txn;
@@ -1250,7 +1255,8 @@ static inline void perf_sample_save_raw_data(struct perf_sample_data *data,
 
 static inline void perf_sample_save_brstack(struct perf_sample_data *data,
 					    struct perf_event *event,
-					    struct perf_branch_stack *brs)
+					    struct perf_branch_stack *brs,
+					    u64 *brs_cntr)
 {
 	int size = sizeof(u64); /* nr */
 
@@ -1258,7 +1264,16 @@ static inline void perf_sample_save_brstack(struct perf_sample_data *data,
 		size += sizeof(u64);
 	size += brs->nr * sizeof(struct perf_branch_entry);
 
+	/*
+	 * The extension space for counters is appended after the
+	 * struct perf_branch_stack. It is used to store the occurrences
+	 * of events of each branch.
+	 */
+	if (brs_cntr)
+		size += brs->nr * sizeof(u64);
+
 	data->br_stack = brs;
+	data->br_stack_cntr = brs_cntr;
 	data->dyn_size += size;
 	data->sample_flags |= PERF_SAMPLE_BRANCH_STACK;
 }
include/uapi/linux/perf_event.h +10 -0
index 39c6a250dd1b..4461f380425b 100644
@@ -204,6 +204,8 @@ enum perf_branch_sample_type_shift {
 
 	PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT	= 18, /* save privilege mode */
 
+	PERF_SAMPLE_BRANCH_COUNTERS_SHIFT	= 19, /* save occurrences of events on a branch */
+
 	PERF_SAMPLE_BRANCH_MAX_SHIFT		/* non-ABI */
 };
 
@@ -235,6 +237,8 @@ enum perf_branch_sample_type {
 
 	PERF_SAMPLE_BRANCH_PRIV_SAVE	= 1U << PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT,
 
+	PERF_SAMPLE_BRANCH_COUNTERS	= 1U << PERF_SAMPLE_BRANCH_COUNTERS_SHIFT,
+
 	PERF_SAMPLE_BRANCH_MAX		= 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT,
 };
 
@@ -982,6 +986,12 @@ enum perf_event_type {
 	 *	{ u64                   nr;
 	 *	  { u64	hw_idx; } && PERF_SAMPLE_BRANCH_HW_INDEX
 	 *        { u64 from, to, flags } lbr[nr];
+	 *        #
+	 *        # The format of the counters is decided by the
+	 *        # "branch_counter_nr" and "branch_counter_width",
+	 *        # which are defined in the ABI.
+	 *        #
+	 *        { u64 counters; } cntr[nr] && PERF_SAMPLE_BRANCH_COUNTERS
 	 *      } && PERF_SAMPLE_BRANCH_STACK
 	 *
 	 * 	{ u64			abi; # enum perf_sample_regs_abi
kernel/events/core.c +8 -0
index 3eb26c2c6e65..d27ffd80ed67 100644
@@ -7341,6 +7341,14 @@ void perf_output_sample(struct perf_output_handle *handle,
 			if (branch_sample_hw_index(event))
 				perf_output_put(handle, data->br_stack->hw_idx);
 			perf_output_copy(handle, data->br_stack->entries, size);
+			/*
+			 * Add the extension space which is appended
+			 * right after the struct perf_branch_stack.
+			 */
+			if (data->br_stack_cntr) {
+				size = data->br_stack->nr * sizeof(u64);
+				perf_output_copy(handle, data->br_stack_cntr, size);
+			}
 		} else {
 			/*
 			 * we always store at least the value of nr
Signature · 1 line
2.35.1
#14 lore raw

PATCH v5 6/8 tools headers UAPI: Sync include/uapi/linux/perf_event.h header with the kernel

From: Kan Liang <[email protected]>

Sync the new sample type for the branch counters feature.

---

Changes since V4
- Add PERF_BRANCH_ENTRY_INFO_BITS_MAX

 tools/include/uapi/linux/perf_event.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
tools/include/uapi/linux/perf_event.h +13 -0
index 39c6a250dd1b..3a64499b0f5d 100644
@@ -204,6 +204,8 @@ enum perf_branch_sample_type_shift {
 
 	PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT	= 18, /* save privilege mode */
 
+	PERF_SAMPLE_BRANCH_COUNTERS_SHIFT	= 19, /* save occurrences of events on a branch */
+
 	PERF_SAMPLE_BRANCH_MAX_SHIFT		/* non-ABI */
 };
 
@@ -235,6 +237,8 @@ enum perf_branch_sample_type {
 
 	PERF_SAMPLE_BRANCH_PRIV_SAVE	= 1U << PERF_SAMPLE_BRANCH_PRIV_SAVE_SHIFT,
 
+	PERF_SAMPLE_BRANCH_COUNTERS	= 1U << PERF_SAMPLE_BRANCH_COUNTERS_SHIFT,
+
 	PERF_SAMPLE_BRANCH_MAX		= 1U << PERF_SAMPLE_BRANCH_MAX_SHIFT,
 };
 
@@ -982,6 +986,12 @@ enum perf_event_type {
 	 *	{ u64                   nr;
 	 *	  { u64	hw_idx; } && PERF_SAMPLE_BRANCH_HW_INDEX
 	 *        { u64 from, to, flags } lbr[nr];
+	 *        #
+	 *        # The format of the counters is decided by the
+	 *        # "branch_counter_nr" and "branch_counter_width",
+	 *        # which are defined in the ABI.
+	 *        #
+	 *        { u64 counters; } cntr[nr] && PERF_SAMPLE_BRANCH_COUNTERS
 	 *      } && PERF_SAMPLE_BRANCH_STACK
 	 *
 	 * 	{ u64			abi; # enum perf_sample_regs_abi
@@ -1427,6 +1437,9 @@ struct perf_branch_entry {
 		reserved:31;
 };
 
+/* Size of used info bits in struct perf_branch_entry */
+#define PERF_BRANCH_ENTRY_INFO_BITS_MAX		33
+
 union perf_sample_weight {
 	__u64		full;
 #if defined(__LITTLE_ENDIAN_BITFIELD)
Signature · 1 line
2.35.1
#12 lore raw

PATCH v5 4/8 perf/x86/intel: Reorganize attrs and is_visible

In reply to [email protected]

From: Kan Liang <[email protected]>

Some attrs and is_visible implementations are rather far away from one another which makes the whole thing hard to interpret.

There are only two attribute groups which have both .attrs and .is_visible, group_default and group_caps_lbr. Move them together.

No functional changes.

---

New patch

 arch/x86/events/intel/core.c | 30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)
arch/x86/events/intel/core.c +15 -15
index a99449c0d77c..584b58df7bf6 100644
@@ -5540,6 +5540,12 @@ static struct attribute *lbr_attrs[] = {
 	NULL
 };
 
+static umode_t
+lbr_is_visible(struct kobject *kobj, struct attribute *attr, int i)
+{
+	return x86_pmu.lbr_nr ? attr->mode : 0;
+}
+
 static char pmu_name_str[30];
 
 static ssize_t pmu_name_show(struct device *cdev,
@@ -5566,6 +5572,15 @@ static struct attribute *intel_pmu_attrs[] = {
 	NULL,
 };
 
+static umode_t
+default_is_visible(struct kobject *kobj, struct attribute *attr, int i)
+{
+	if (attr == &dev_attr_allow_tsx_force_abort.attr)
+		return x86_pmu.flags & PMU_FL_TFA ? attr->mode : 0;
+
+	return attr->mode;
+}
+
 static umode_t
 tsx_is_visible(struct kobject *kobj, struct attribute *attr, int i)
 {
@@ -5587,27 +5602,12 @@ mem_is_visible(struct kobject *kobj, struct attribute *attr, int i)
 	return pebs_is_visible(kobj, attr, i);
 }
 
-static umode_t
-lbr_is_visible(struct kobject *kobj, struct attribute *attr, int i)
-{
-	return x86_pmu.lbr_nr ? attr->mode : 0;
-}
-
 static umode_t
 exra_is_visible(struct kobject *kobj, struct attribute *attr, int i)
 {
 	return x86_pmu.version >= 2 ? attr->mode : 0;
 }
 
-static umode_t
-default_is_visible(struct kobject *kobj, struct attribute *attr, int i)
-{
-	if (attr == &dev_attr_allow_tsx_force_abort.attr)
-		return x86_pmu.flags & PMU_FL_TFA ? attr->mode : 0;
-
-	return attr->mode;
-}
-
 static struct attribute_group group_events_td  = {
 	.name = "events",
 };
Signature · 1 line
2.35.1
#15 lore raw

PATCH v5 7/8 perf header: Support num and width of branch counters

In reply to [email protected]

From: Kan Liang <[email protected]>

To support the branch counters feature, the information of the maximum number of supported counters and the width of the counters is exposed in the sysfs caps folder. The perf tool can use the information to parse the logged counters in each branch.

Store the information in the perf_env for later usage.

---

No changes since V4

 tools/perf/util/env.h    |  5 +++++
 tools/perf/util/header.c | 18 +++++++++++++++---
 2 files changed, 20 insertions(+), 3 deletions(-)
tools/perf/util/env.h +5 -0
index 4566c51f2fd9..48d7f8759a2a 100644
@@ -46,6 +46,9 @@ struct hybrid_node {
 struct pmu_caps {
 	int		nr_caps;
 	unsigned int    max_branches;
+	unsigned int	br_cntr_nr;
+	unsigned int	br_cntr_width;
+
 	char            **caps;
 	char            *pmu_name;
 };
@@ -62,6 +65,8 @@ struct perf_env {
 	unsigned long long	total_mem;
 	unsigned int		msr_pmu_type;
 	unsigned int		max_branches;
+	unsigned int		br_cntr_nr;
+	unsigned int		br_cntr_width;
 	int			kernel_is_64_bit;
 
 	int			nr_cmdline;
tools/perf/util/header.c +15 -3
index d812e1e371a7..9664062ba835 100644
@@ -3256,7 +3256,9 @@ static int process_compressed(struct feat_fd *ff,
 }
 
 static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
-			      char ***caps, unsigned int *max_branches)
+			      char ***caps, unsigned int *max_branches,
+			      unsigned int *br_cntr_nr,
+			      unsigned int *br_cntr_width)
 {
 	char *name, *value, *ptr;
 	u32 nr_pmu_caps, i;
@@ -3291,6 +3293,12 @@ static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
 		if (!strcmp(name, "branches"))
 			*max_branches = atoi(value);
 
+		if (!strcmp(name, "branch_counter_nr"))
+			*br_cntr_nr = atoi(value);
+
+		if (!strcmp(name, "branch_counter_width"))
+			*br_cntr_width = atoi(value);
+
 		free(value);
 		free(name);
 	}
@@ -3315,7 +3323,9 @@ static int process_cpu_pmu_caps(struct feat_fd *ff,
 {
 	int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
 				     &ff->ph->env.cpu_pmu_caps,
-				     &ff->ph->env.max_branches);
+				     &ff->ph->env.max_branches,
+				     &ff->ph->env.br_cntr_nr,
+				     &ff->ph->env.br_cntr_width);
 
 	if (!ret && !ff->ph->env.cpu_pmu_caps)
 		pr_debug("cpu pmu capabilities not available\n");
@@ -3344,7 +3354,9 @@ static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
 	for (i = 0; i < nr_pmu; i++) {
 		ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
 					 &pmu_caps[i].caps,
-					 &pmu_caps[i].max_branches);
+					 &pmu_caps[i].max_branches,
+					 &pmu_caps[i].br_cntr_nr,
+					 &pmu_caps[i].br_cntr_width);
 		if (ret)
 			goto err;
 
Signature · 1 line
2.35.1
#17 lore raw

PATCH v5 8/8 perf tools: Add branch counter knob

In reply to [email protected]

From: Kan Liang <[email protected]>

Add a new branch filter, "counter", for the branch counter option. It is used to mark the events which should be logged in the branch. If it is applied with the -j option, the counters of all the events should be logged in the branch. If the legacy kernel doesn't support the new branch sample type, switching off the branch counter filter.

The stored counter values in each branch are displayed right after the regular branch stack information via perf report -D.

Usage examples:

perf record -e "{branch-instructions,branch-misses}:S" -j any,counter

Only the first event, branch-instructions, collect the LBR. Both branch-instructions and branch-misses are marked as logged events. The occurrences information of them can be found in the branch stack extension space of each branch.

perf record -e "{cpu/branch-instructions,branch_type=any/, cpu/branch-misses,branch_type=counter/}"

Only the first event, branch-instructions, collect the LBR. Only the branch-misses event is marked as a logged event.

---

No changes since V4

 tools/perf/Documentation/perf-record.txt  |  4 +++
 tools/perf/util/evsel.c                   | 31 ++++++++++++++++++++++-
 tools/perf/util/evsel.h                   |  1 +
 tools/perf/util/parse-branch-options.c    |  1 +
 tools/perf/util/perf_event_attr_fprintf.c |  1 +
 tools/perf/util/sample.h                  |  1 +
 tools/perf/util/session.c                 | 15 +++++++++--
 7 files changed, 51 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +4 -0
index d5217be012d7..b6afe7cc948d 100644
@@ -442,6 +442,10 @@ following filters are defined:
 		     4th-Gen Xeon+ server), the save branch type is unconditionally enabled
 		     when the taken branch stack sampling is enabled.
 	- priv: save privilege state during sampling in case binary is not available later
+	- counter: save occurrences of the event since the last branch entry. Currently, the
+		   feature is only supported by a newer CPU, e.g., Intel Sierra Forest and
+		   later platforms. An error out is expected if it's used on the unsupported
+		   kernel or CPUs.
 
 +
 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
tools/perf/util/evsel.c +30 -1
index a8a5ff87cc1f..58a9b8c82790 100644
@@ -1831,6 +1831,8 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
 
 static void evsel__disable_missing_features(struct evsel *evsel)
 {
+	if (perf_missing_features.branch_counters)
+		evsel->core.attr.branch_sample_type &= ~PERF_SAMPLE_BRANCH_COUNTERS;
 	if (perf_missing_features.read_lost)
 		evsel->core.attr.read_format &= ~PERF_FORMAT_LOST;
 	if (perf_missing_features.weight_struct) {
@@ -1884,7 +1886,12 @@ bool evsel__detect_missing_features(struct evsel *evsel)
 	 * Must probe features in the order they were added to the
 	 * perf_event_attr interface.
 	 */
-	if (!perf_missing_features.read_lost &&
+	if (!perf_missing_features.branch_counters &&
+	    (evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS)) {
+		perf_missing_features.branch_counters = true;
+		pr_debug2("switching off branch counters support\n");
+		return true;
+	} else if (!perf_missing_features.read_lost &&
 	    (evsel->core.attr.read_format & PERF_FORMAT_LOST)) {
 		perf_missing_features.read_lost = true;
 		pr_debug2("switching off PERF_FORMAT_LOST support\n");
@@ -2344,6 +2351,18 @@ u64 evsel__bitfield_swap_branch_flags(u64 value)
 	return new_val;
 }
 
+static inline bool evsel__has_branch_counters(const struct evsel *evsel)
+{
+	struct evsel *cur, *leader = evsel__leader(evsel);
+
+	evlist__for_each_entry(evsel->evlist, cur) {
+		if ((leader == evsel__leader(cur)) &&
+		    (cur->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
+			return true;
+	}
+	return false;
+}
+
 int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
 			struct perf_sample *data)
 {
@@ -2577,6 +2596,16 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
 
 		OVERFLOW_CHECK(array, sz, max_size);
 		array = (void *)array + sz;
+
+		if (evsel__has_branch_counters(evsel)) {
+			OVERFLOW_CHECK_u64(array);
+
+			data->branch_stack_cntr = (u64 *)array;
+			sz = data->branch_stack->nr * sizeof(u64);
+
+			OVERFLOW_CHECK(array, sz, max_size);
+			array = (void *)array + sz;
+		}
 	}
 
 	if (type & PERF_SAMPLE_REGS_USER) {
tools/perf/util/evsel.h +1 -0
index 848534ec74fa..85f24c986392 100644
@@ -191,6 +191,7 @@ struct perf_missing_features {
 	bool code_page_size;
 	bool weight_struct;
 	bool read_lost;
+	bool branch_counters;
 };
 
 extern struct perf_missing_features perf_missing_features;
tools/perf/util/parse-branch-options.c +1 -0
index fd67d204d720..f7f7aff3d85a 100644
@@ -36,6 +36,7 @@ static const struct branch_mode branch_modes[] = {
 	BRANCH_OPT("stack", PERF_SAMPLE_BRANCH_CALL_STACK),
 	BRANCH_OPT("hw_index", PERF_SAMPLE_BRANCH_HW_INDEX),
 	BRANCH_OPT("priv", PERF_SAMPLE_BRANCH_PRIV_SAVE),
+	BRANCH_OPT("counter", PERF_SAMPLE_BRANCH_COUNTERS),
 	BRANCH_END
 };
 
tools/perf/util/perf_event_attr_fprintf.c +1 -0
index 2247991451f3..8f04d3b7f3ec 100644
@@ -55,6 +55,7 @@ static void __p_branch_sample_type(char *buf, size_t size, u64 value)
 		bit_name(COND), bit_name(CALL_STACK), bit_name(IND_JUMP),
 		bit_name(CALL), bit_name(NO_FLAGS), bit_name(NO_CYCLES),
 		bit_name(TYPE_SAVE), bit_name(HW_INDEX), bit_name(PRIV_SAVE),
+		bit_name(COUNTERS),
 		{ .name = NULL, }
 	};
 #undef bit_name
tools/perf/util/sample.h +1 -0
index c92ad0f51ecd..70b2c3135555 100644
@@ -113,6 +113,7 @@ struct perf_sample {
 	void *raw_data;
 	struct ip_callchain *callchain;
 	struct branch_stack *branch_stack;
+	u64 *branch_stack_cntr;
 	struct regs_dump  user_regs;
 	struct regs_dump  intr_regs;
 	struct stack_dump user_stack;
tools/perf/util/session.c +13 -2
index 1e9aa8ed15b6..4a094ab0362b 100644
@@ -1150,9 +1150,13 @@ static void callchain__printf(struct evsel *evsel,
 		       i, callchain->ips[i]);
 }
 
-static void branch_stack__printf(struct perf_sample *sample, bool callstack)
+static void branch_stack__printf(struct perf_sample *sample,
+				 struct evsel *evsel)
 {
 	struct branch_entry *entries = perf_sample__branch_entries(sample);
+	bool callstack = evsel__has_branch_callstack(evsel);
+	u64 *branch_stack_cntr = sample->branch_stack_cntr;
+	struct perf_env *env = evsel__env(evsel);
 	uint64_t i;
 
 	if (!callstack) {
@@ -1194,6 +1198,13 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack)
 			}
 		}
 	}
+
+	if (branch_stack_cntr) {
+		printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n",
+			sample->branch_stack->nr, env->br_cntr_width, env->br_cntr_nr);
+		for (i = 0; i < sample->branch_stack->nr; i++)
+			printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]);
+	}
 }
 
 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
@@ -1355,7 +1366,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,
 		callchain__printf(evsel, sample);
 
 	if (evsel__has_br_stack(evsel))
-		branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
+		branch_stack__printf(sample, evsel);
 
 	if (sample_type & PERF_SAMPLE_REGS_USER)
 		regs_user__printf(sample, arch);
Signature · 1 line
2.35.1
#13 lore raw

PATCH v5 5/8 perf/x86/intel: Support branch counters logging

In reply to [email protected]

From: Kan Liang <[email protected]>

The branch counters logging (A.K.A LBR event logging) introduces a per-counter indication of precise event occurrences in LBRs. It can provide a means to attribute exposed retirement latency to combinations of events across a block of instructions. It also provides a means of attributing Timed LBR latencies to events.

The feature is first introduced on SRF/GRR. It is an enhancement of the ARCH LBR. It adds new fields in the LBR_INFO MSRs to log the occurrences of events on the GP counters. The information is displayed by the order of counters.

The design proposed in this patch requires that the events which are logged must be in a group with the event that has LBR. If there are more than one LBR group, the counters logging information only from the current group (overflowed) are stored for the perf tool, otherwise the perf tool cannot know which and when other groups are scheduled especially when multiplexing is triggered. The user can ensure it uses the maximum number of counters that support LBR info (4 by now) by making the group large enough.

The HW only logs events by the order of counters. The order may be different from the order of enabling which the perf tool can understand. When parsing the information of each branch entry, convert the counter order to the enabled order, and store the enabled order in the extension space.

Unconditionally reset LBRs for an LBR event group when it's deleted. The logged counter information is only valid for the current LBR group. If another LBR group is scheduled later, the information from the stale LBRs would be otherwise wrongly interpreted.

Add a sanity check in intel_pmu_hw_config(). Disable the feature if other counter filters (inv, cmask, edge, in_tx) are set or LBR call stack mode is enabled. (For the LBR call stack mode, we cannot simply flush the LBR, since it will break the call stack. Also, there is no obvious usage with the call stack mode for now.)

Only applying the PERF_SAMPLE_BRANCH_COUNTERS doesn't require any branch stack setup.

Expose the maximum number of supported counters and the width of the counters into the sysfs. The perf tool can use the information to parse the logged counters in each branch.

---

Changes since V4:
- Only reset LBR once for a branch counter group when the group is deleted. Force the leader event of a branch counter group a LBR event.
- Use the name "branch counters" to replace "event logging"
- Use the suggested code from PeterZ for branch counter reordering
- Check the max number for a branch counter group in hw_config
- Use the reserved field to store the temporary branch counters information

 arch/x86/events/intel/core.c       | 103 +++++++++++++++++++++++++++--
 arch/x86/events/intel/ds.c         |   2 +-
 arch/x86/events/intel/lbr.c        |  85 +++++++++++++++++++++++-
 arch/x86/events/perf_event.h       |  12 ++++
 arch/x86/events/perf_event_flags.h |   1 +
 arch/x86/include/asm/msr-index.h   |   5 ++
 arch/x86/include/asm/perf_event.h  |   4 ++
 include/uapi/linux/perf_event.h    |   3 +
 8 files changed, 207 insertions(+), 8 deletions(-)
arch/x86/events/intel/core.c +97 -6
index 584b58df7bf6..3a249b026d5c 100644
@@ -2792,6 +2792,7 @@ static void intel_pmu_enable_fixed(struct perf_event *event)
 
 static void intel_pmu_enable_event(struct perf_event *event)
 {
+	u64 enable_mask = ARCH_PERFMON_EVENTSEL_ENABLE;
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 
@@ -2800,8 +2801,10 @@ static void intel_pmu_enable_event(struct perf_event *event)
 
 	switch (idx) {
 	case 0 ... INTEL_PMC_IDX_FIXED - 1:
+		if (branch_sample_counters(event))
+			enable_mask |= ARCH_PERFMON_EVENTSEL_BR_CNTR;
 		intel_set_masks(event, idx);
-		__x86_pmu_enable_event(hwc, ARCH_PERFMON_EVENTSEL_ENABLE);
+		__x86_pmu_enable_event(hwc, enable_mask);
 		break;
 	case INTEL_PMC_IDX_FIXED ... INTEL_PMC_IDX_FIXED_BTS - 1:
 	case INTEL_PMC_IDX_METRIC_BASE ... INTEL_PMC_IDX_METRIC_END:
@@ -3052,7 +3055,7 @@ static int handle_pmi_common(struct pt_regs *regs, u64 status)
 		perf_sample_data_init(&data, 0, event->hw.last_period);
 
 		if (has_branch_stack(event))
-			perf_sample_save_brstack(&data, event, &cpuc->lbr_stack, NULL);
+			intel_pmu_lbr_save_brstack(&data, cpuc, event);
 
 		if (perf_event_overflow(event, &data, regs))
 			x86_pmu_stop(event, 0);
@@ -3617,6 +3620,13 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
 	if (cpuc->excl_cntrs)
 		return intel_get_excl_constraints(cpuc, event, idx, c2);
 
+	/* Not all counters support the branch counter feature. */
+	if (branch_sample_counters(event)) {
+		c2 = dyn_constraint(cpuc, c2, idx);
+		c2->idxmsk64 &= x86_pmu.lbr_counters;
+		c2->weight = hweight64(c2->idxmsk64);
+	}
+
 	return c2;
 }
 
@@ -3905,6 +3915,58 @@ static int intel_pmu_hw_config(struct perf_event *event)
 	if (needs_branch_stack(event) && is_sampling_event(event))
 		event->hw.flags  |= PERF_X86_EVENT_NEEDS_BRANCH_STACK;
 
+	if (branch_sample_counters(event)) {
+		struct perf_event *leader, *sibling;
+		int num = 0;
+
+		if (!(x86_pmu.flags & PMU_FL_BR_CNTR) ||
+		    (event->attr.config & ~INTEL_ARCH_EVENT_MASK))
+			return -EINVAL;
+
+		/*
+		 * The branch counter logging is not supported in the call stack
+		 * mode yet, since we cannot simply flush the LBR during e.g.,
+		 * multiplexing. Also, there is no obvious usage with the call
+		 * stack mode. Simply forbids it for now.
+		 *
+		 * If any events in the group enable the branch counter logging
+		 * feature, the group is treated as a branch counter logging
+		 * group, which requires the extra space to store the counters.
+		 */
+		leader = event->group_leader;
+		if (branch_sample_call_stack(leader))
+			return -EINVAL;
+		if (branch_sample_counters(leader))
+			num++;
+		leader->hw.flags |= PERF_X86_EVENT_BRANCH_COUNTERS;
+
+		for_each_sibling_event(sibling, leader) {
+			if (branch_sample_call_stack(sibling))
+				return -EINVAL;
+			if (branch_sample_counters(sibling))
+				num++;
+		}
+
+		if (num > fls(x86_pmu.lbr_counters))
+			return -EINVAL;
+		/*
+		 * Only applying the PERF_SAMPLE_BRANCH_COUNTERS doesn't
+		 * require any branch stack setup.
+		 * Clear the bit to avoid unnecessary branch stack setup.
+		 */
+		if (0 == (event->attr.branch_sample_type &
+			  ~(PERF_SAMPLE_BRANCH_PLM_ALL |
+			    PERF_SAMPLE_BRANCH_COUNTERS)))
+			event->hw.flags  &= ~PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+
+		/*
+		 * Force the leader to be a LBR event. So LBRs can be reset
+		 * with the leader event. See intel_pmu_lbr_del() for details.
+		 */
+		if (!intel_pmu_needs_branch_stack(leader))
+			return -EINVAL;
+	}
+
 	if (intel_pmu_needs_branch_stack(event)) {
 		ret = intel_pmu_setup_lbr_filter(event);
 		if (ret)
@@ -4383,8 +4445,13 @@ cmt_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
 	 */
 	if (event->attr.precise_ip == 3) {
 		/* Force instruction:ppp on PMC0, 1 and Fixed counter 0 */
-		if (constraint_match(&fixed0_constraint, event->hw.config))
-			return &fixed0_counter0_1_constraint;
+		if (constraint_match(&fixed0_constraint, event->hw.config)) {
+			/* The fixed counter 0 doesn't support LBR event logging. */
+			if (branch_sample_counters(event))
+				return &counter0_1_constraint;
+			else
+				return &fixed0_counter0_1_constraint;
+		}
 
 		switch (c->idxmsk64 & 0x3ull) {
 		case 0x1:
@@ -4563,7 +4630,7 @@ int intel_cpuc_prepare(struct cpu_hw_events *cpuc, int cpu)
 			goto err;
 	}
 
-	if (x86_pmu.flags & (PMU_FL_EXCL_CNTRS | PMU_FL_TFA)) {
+	if (x86_pmu.flags & (PMU_FL_EXCL_CNTRS | PMU_FL_TFA | PMU_FL_BR_CNTR)) {
 		size_t sz = X86_PMC_IDX_MAX * sizeof(struct event_constraint);
 
 		cpuc->constraint_list = kzalloc_node(sz, GFP_KERNEL, cpu_to_node(cpu));
@@ -5535,15 +5602,39 @@ static ssize_t branches_show(struct device *cdev,
 
 static DEVICE_ATTR_RO(branches);
 
+static ssize_t branch_counter_nr_show(struct device *cdev,
+				      struct device_attribute *attr,
+				      char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%d\n", fls(x86_pmu.lbr_counters));
+}
+
+static DEVICE_ATTR_RO(branch_counter_nr);
+
+static ssize_t branch_counter_width_show(struct device *cdev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "2\n");
+}
+
+static DEVICE_ATTR_RO(branch_counter_width);
+
 static struct attribute *lbr_attrs[] = {
 	&dev_attr_branches.attr,
+	&dev_attr_branch_counter_nr.attr,
+	&dev_attr_branch_counter_width.attr,
 	NULL
 };
 
 static umode_t
 lbr_is_visible(struct kobject *kobj, struct attribute *attr, int i)
 {
-	return x86_pmu.lbr_nr ? attr->mode : 0;
+	/* branches */
+	if (i == 0)
+		return x86_pmu.lbr_nr ? attr->mode : 0;
+
+	return (x86_pmu.flags & PMU_FL_BR_CNTR) ? attr->mode : 0;
 }
 
 static char pmu_name_str[30];
arch/x86/events/intel/ds.c +1 -1
index 1d309e97a629..98bda9796a28 100644
@@ -1912,7 +1912,7 @@ static void setup_pebs_adaptive_sample_data(struct perf_event *event,
 
 		if (has_branch_stack(event)) {
 			intel_pmu_store_pebs_lbrs(lbr);
-			perf_sample_save_brstack(data, event, &cpuc->lbr_stack, NULL);
+			intel_pmu_lbr_save_brstack(data, cpuc, event);
 		}
 	}
 
arch/x86/events/intel/lbr.c +84 -1

123 lines not on this page · raw message

arch/x86/events/perf_event.h +12 -0
index 53dd5d495ba6..fb56518356ec 100644
@@ -110,6 +110,11 @@ static inline bool is_topdown_event(struct perf_event *event)
 	return is_metric_event(event) || is_slots_event(event);
 }
 
+static inline bool is_branch_counters_group(struct perf_event *event)
+{
+	return event->group_leader->hw.flags & PERF_X86_EVENT_BRANCH_COUNTERS;
+}
+
 struct amd_nb {
 	int nb_id;  /* NorthBridge id */
 	int refcnt; /* reference count */
@@ -283,6 +288,7 @@ struct cpu_hw_events {
 	int				lbr_pebs_users;
 	struct perf_branch_stack	lbr_stack;
 	struct perf_branch_entry	lbr_entries[MAX_LBR_ENTRIES];
+	u64				lbr_counters[MAX_LBR_ENTRIES]; /* branch stack extra */
 	union {
 		struct er_account		*lbr_sel;
 		struct er_account		*lbr_ctl;
@@ -888,6 +894,7 @@ struct x86_pmu {
 	unsigned int	lbr_mispred:1;
 	unsigned int	lbr_timed_lbr:1;
 	unsigned int	lbr_br_type:1;
+	unsigned int	lbr_counters:4;
 
 	void		(*lbr_reset)(void);
 	void		(*lbr_read)(struct cpu_hw_events *cpuc);
@@ -1012,6 +1019,7 @@ do {									\
 #define PMU_FL_INSTR_LATENCY	0x80 /* Support Instruction Latency in PEBS Memory Info Record */
 #define PMU_FL_MEM_LOADS_AUX	0x100 /* Require an auxiliary event for the complete memory info */
 #define PMU_FL_RETIRE_LATENCY	0x200 /* Support Retire Latency in PEBS */
+#define PMU_FL_BR_CNTR		0x400 /* Support branch counter logging */
 
 #define EVENT_VAR(_id)  event_attr_##_id
 #define EVENT_PTR(_id) &event_attr_##_id.attr.attr
@@ -1552,6 +1560,10 @@ void intel_pmu_store_pebs_lbrs(struct lbr_entry *lbr);
 
 void intel_ds_init(void);
 
+void intel_pmu_lbr_save_brstack(struct perf_sample_data *data,
+				struct cpu_hw_events *cpuc,
+				struct perf_event *event);
+
 void intel_pmu_lbr_swap_task_ctx(struct perf_event_pmu_context *prev_epc,
 				 struct perf_event_pmu_context *next_epc);
 
arch/x86/events/perf_event_flags.h +1 -0
index a1685981c520..6c977c19f2cd 100644
@@ -21,3 +21,4 @@ PERF_ARCH(PEBS_STLAT,		0x08000) /* st+stlat data address sampling */
 PERF_ARCH(AMD_BRS,		0x10000) /* AMD Branch Sampling */
 PERF_ARCH(PEBS_LAT_HYBRID,	0x20000) /* ld and st lat for hybrid */
 PERF_ARCH(NEEDS_BRANCH_STACK,	0x40000) /* require branch stack setup */
+PERF_ARCH(BRANCH_COUNTERS,	0x80000) /* logs the counters in the extra space of each branch */
arch/x86/include/asm/msr-index.h +5 -0
index dc159acb350a..27e6f1c89fbf 100644
@@ -236,6 +236,11 @@
 #define LBR_INFO_CYCLES			0xffff
 #define LBR_INFO_BR_TYPE_OFFSET		56
 #define LBR_INFO_BR_TYPE		(0xfull << LBR_INFO_BR_TYPE_OFFSET)
+#define LBR_INFO_BR_CNTR_OFFSET		32
+#define LBR_INFO_BR_CNTR_NUM		4
+#define LBR_INFO_BR_CNTR_BITS		2
+#define LBR_INFO_BR_CNTR_MASK		GENMASK_ULL(LBR_INFO_BR_CNTR_BITS - 1, 0)
+#define LBR_INFO_BR_CNTR_FULL_MASK	GENMASK_ULL(LBR_INFO_BR_CNTR_NUM * LBR_INFO_BR_CNTR_BITS - 1, 0)
 
 #define MSR_ARCH_LBR_CTL		0x000014ce
 #define ARCH_LBR_CTL_LBREN		BIT(0)
arch/x86/include/asm/perf_event.h +4 -0
index 2618ec7c3d1d..3736b8a46c04 100644
@@ -31,6 +31,7 @@
 #define ARCH_PERFMON_EVENTSEL_ENABLE			(1ULL << 22)
 #define ARCH_PERFMON_EVENTSEL_INV			(1ULL << 23)
 #define ARCH_PERFMON_EVENTSEL_CMASK			0xFF000000ULL
+#define ARCH_PERFMON_EVENTSEL_BR_CNTR			(1ULL << 35)
 
 #define INTEL_FIXED_BITS_MASK				0xFULL
 #define INTEL_FIXED_BITS_STRIDE			4
@@ -223,6 +224,9 @@ union cpuid28_ecx {
 		unsigned int    lbr_timed_lbr:1;
 		/* Branch Type Field Supported */
 		unsigned int    lbr_br_type:1;
+		unsigned int	reserved:13;
+		/* Branch counters (Event Logging) Supported */
+		unsigned int	lbr_counters:4;
 	} split;
 	unsigned int            full;
 };
include/uapi/linux/perf_event.h +3 -0
index 4461f380425b..3a64499b0f5d 100644
@@ -1437,6 +1437,9 @@ struct perf_branch_entry {
 		reserved:31;
 };
 
+/* Size of used info bits in struct perf_branch_entry */
+#define PERF_BRANCH_ENTRY_INFO_BITS_MAX		33
+
 union perf_sample_weight {
 	__u64		full;
 #if defined(__LITTLE_ENDIAN_BITFIELD)
Signature · 1 line
2.35.1
reviewer Reviewed-by on v5 7/8
#16 lore raw

Re: [PATCH V5 7/8] perf header: Support num and width of branch counters

In reply to [email protected]
Quoted 11 lines · On Wed, Oct 25, 2023 at 1:16 PM <[email protected]> wrote:

From: Kan Liang <[email protected]>

To support the branch counters feature, the information of the maximum number of supported counters and the width of the counters is exposed in the sysfs caps folder. The perf tool can use the information to parse the logged counters in each branch.

Store the information in the perf_env for later usage.

Thanks,
Ian

Quoted 84 lines

---

No changes since V4

 tools/perf/util/env.h    |  5 +++++
 tools/perf/util/header.c | 18 +++++++++++++++---
 2 files changed, 20 insertions(+), 3 deletions(-)
tools/perf/util/env.h +5 -0
index 4566c51f2fd9..48d7f8759a2a 100644
@@ -46,6 +46,9 @@ struct hybrid_node {
 struct pmu_caps {
        int             nr_caps;
        unsigned int    max_branches;
+       unsigned int    br_cntr_nr;
+       unsigned int    br_cntr_width;
+
        char            **caps;
        char            *pmu_name;
 };
@@ -62,6 +65,8 @@ struct perf_env {
        unsigned long long      total_mem;
        unsigned int            msr_pmu_type;
        unsigned int            max_branches;
+       unsigned int            br_cntr_nr;
+       unsigned int            br_cntr_width;
        int                     kernel_is_64_bit;
 
        int                     nr_cmdline;
tools/perf/util/header.c +15 -4
index d812e1e371a7..9664062ba835 100644
@@ -3256,7 +3256,9 @@ static int process_compressed(struct feat_fd *ff,
 }
 
 static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
-                             char ***caps, unsigned int *max_branches)
+                             char ***caps, unsigned int *max_branches,
+                             unsigned int *br_cntr_nr,
+                             unsigned int *br_cntr_width)
 {
        char *name, *value, *ptr;
        u32 nr_pmu_caps, i;
@@ -3291,6 +3293,12 @@ static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps,
                if (!strcmp(name, "branches"))
                        *max_branches = atoi(value);
 
+               if (!strcmp(name, "branch_counter_nr"))
+                       *br_cntr_nr = atoi(value);
+
+               if (!strcmp(name, "branch_counter_width"))
+                       *br_cntr_width = atoi(value);
+
                free(value);
                free(name);
        }
@@ -3315,7 +3323,9 @@ static int process_cpu_pmu_caps(struct feat_fd *ff,
 {
        int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
                                     &ff->ph->env.cpu_pmu_caps,
-                                    &ff->ph->env.max_branches);
+                                    &ff->ph->env.max_branches,
+                                    &ff->ph->env.br_cntr_nr,
+                                    &ff->ph->env.br_cntr_width);
 
        if (!ret && !ff->ph->env.cpu_pmu_caps)
                pr_debug("cpu pmu capabilities not available\n");
@@ -3344,7 +3354,9 @@ static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused)
        for (i = 0; i < nr_pmu; i++) {
                ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps,
                                         &pmu_caps[i].caps,
-                                        &pmu_caps[i].max_branches);
+                                        &pmu_caps[i].max_branches,
+                                        &pmu_caps[i].br_cntr_nr,
+                                        &pmu_caps[i].br_cntr_width);
                if (ret)
                        goto err;
 
--

2.35.1

reviewer Reviewed-by on v5 8/8
#18 lore raw

Re: [PATCH V5 8/8] perf tools: Add branch counter knob

In reply to [email protected]
Quoted 21 lines · On Wed, Oct 25, 2023 at 1:16 PM <[email protected]> wrote:

From: Kan Liang <[email protected]>

Add a new branch filter, "counter", for the branch counter option. It is used to mark the events which should be logged in the branch. If it is applied with the -j option, the counters of all the events should be logged in the branch. If the legacy kernel doesn't support the new branch sample type, switching off the branch counter filter.

The stored counter values in each branch are displayed right after the regular branch stack information via perf report -D.

Usage examples:

perf record -e "{branch-instructions,branch-misses}:S" -j any,counter

Only the first event, branch-instructions, collect the LBR. Both branch-instructions and branch-misses are marked as logged events. The occurrences information of them can be found in the branch stack extension space of each branch.

Quoted 7 lines

perf record -e "{cpu/branch-instructions,branch_type=any/, cpu/branch-misses,branch_type=counter/}"

Only the first event, branch-instructions, collect the LBR. Only the branch-misses event is marked as a logged event.

Perhaps add a test somewhere like:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/record.sh
for coverage when hardware supports the feature.

Thanks,
Ian

Quoted 184 lines

---

No changes since V4

 tools/perf/Documentation/perf-record.txt  |  4 +++
 tools/perf/util/evsel.c                   | 31 ++++++++++++++++++++++-
 tools/perf/util/evsel.h                   |  1 +
 tools/perf/util/parse-branch-options.c    |  1 +
 tools/perf/util/perf_event_attr_fprintf.c |  1 +
 tools/perf/util/sample.h                  |  1 +
 tools/perf/util/session.c                 | 15 +++++++++--
 7 files changed, 51 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +4 -0
index d5217be012d7..b6afe7cc948d 100644
@@ -442,6 +442,10 @@ following filters are defined:
                     4th-Gen Xeon+ server), the save branch type is unconditionally enabled
                     when the taken branch stack sampling is enabled.
        - priv: save privilege state during sampling in case binary is not available later
+       - counter: save occurrences of the event since the last branch entry. Currently, the
+                  feature is only supported by a newer CPU, e.g., Intel Sierra Forest and
+                  later platforms. An error out is expected if it's used on the unsupported
+                  kernel or CPUs.
 
 +
 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
tools/perf/util/evsel.c +30 -1
index a8a5ff87cc1f..58a9b8c82790 100644
@@ -1831,6 +1831,8 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
 
 static void evsel__disable_missing_features(struct evsel *evsel)
 {
+       if (perf_missing_features.branch_counters)
+               evsel->core.attr.branch_sample_type &= ~PERF_SAMPLE_BRANCH_COUNTERS;
        if (perf_missing_features.read_lost)
                evsel->core.attr.read_format &= ~PERF_FORMAT_LOST;
        if (perf_missing_features.weight_struct) {
@@ -1884,7 +1886,12 @@ bool evsel__detect_missing_features(struct evsel *evsel)
         * Must probe features in the order they were added to the
         * perf_event_attr interface.
         */
-       if (!perf_missing_features.read_lost &&
+       if (!perf_missing_features.branch_counters &&
+           (evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS)) {
+               perf_missing_features.branch_counters = true;
+               pr_debug2("switching off branch counters support\n");
+               return true;
+       } else if (!perf_missing_features.read_lost &&
            (evsel->core.attr.read_format & PERF_FORMAT_LOST)) {
                perf_missing_features.read_lost = true;
                pr_debug2("switching off PERF_FORMAT_LOST support\n");
@@ -2344,6 +2351,18 @@ u64 evsel__bitfield_swap_branch_flags(u64 value)
        return new_val;
 }
 
+static inline bool evsel__has_branch_counters(const struct evsel *evsel)
+{
+       struct evsel *cur, *leader = evsel__leader(evsel);
+
+       evlist__for_each_entry(evsel->evlist, cur) {
+               if ((leader == evsel__leader(cur)) &&
+                   (cur->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
+                       return true;
+       }
+       return false;
+}
+
 int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
                        struct perf_sample *data)
 {
@@ -2577,6 +2596,16 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
 
                OVERFLOW_CHECK(array, sz, max_size);
                array = (void *)array + sz;
+
+               if (evsel__has_branch_counters(evsel)) {
+                       OVERFLOW_CHECK_u64(array);
+
+                       data->branch_stack_cntr = (u64 *)array;
+                       sz = data->branch_stack->nr * sizeof(u64);
+
+                       OVERFLOW_CHECK(array, sz, max_size);
+                       array = (void *)array + sz;
+               }
        }
 
        if (type & PERF_SAMPLE_REGS_USER) {
tools/perf/util/evsel.h +1 -0
index 848534ec74fa..85f24c986392 100644
@@ -191,6 +191,7 @@ struct perf_missing_features {
        bool code_page_size;
        bool weight_struct;
        bool read_lost;
+       bool branch_counters;
 };
 
 extern struct perf_missing_features perf_missing_features;
tools/perf/util/parse-branch-options.c +1 -0
index fd67d204d720..f7f7aff3d85a 100644
@@ -36,6 +36,7 @@ static const struct branch_mode branch_modes[] = {
        BRANCH_OPT("stack", PERF_SAMPLE_BRANCH_CALL_STACK),
        BRANCH_OPT("hw_index", PERF_SAMPLE_BRANCH_HW_INDEX),
        BRANCH_OPT("priv", PERF_SAMPLE_BRANCH_PRIV_SAVE),
+       BRANCH_OPT("counter", PERF_SAMPLE_BRANCH_COUNTERS),
        BRANCH_END
 };
 
tools/perf/util/perf_event_attr_fprintf.c +1 -0
index 2247991451f3..8f04d3b7f3ec 100644
@@ -55,6 +55,7 @@ static void __p_branch_sample_type(char *buf, size_t size, u64 value)
                bit_name(COND), bit_name(CALL_STACK), bit_name(IND_JUMP),
                bit_name(CALL), bit_name(NO_FLAGS), bit_name(NO_CYCLES),
                bit_name(TYPE_SAVE), bit_name(HW_INDEX), bit_name(PRIV_SAVE),
+               bit_name(COUNTERS),
                { .name = NULL, }
        };
 #undef bit_name
tools/perf/util/sample.h +1 -0
index c92ad0f51ecd..70b2c3135555 100644
@@ -113,6 +113,7 @@ struct perf_sample {
        void *raw_data;
        struct ip_callchain *callchain;
        struct branch_stack *branch_stack;
+       u64 *branch_stack_cntr;
        struct regs_dump  user_regs;
        struct regs_dump  intr_regs;
        struct stack_dump user_stack;
tools/perf/util/session.c +13 -3
index 1e9aa8ed15b6..4a094ab0362b 100644
@@ -1150,9 +1150,13 @@ static void callchain__printf(struct evsel *evsel,
                       i, callchain->ips[i]);
 }
 
-static void branch_stack__printf(struct perf_sample *sample, bool callstack)
+static void branch_stack__printf(struct perf_sample *sample,
+                                struct evsel *evsel)
 {
        struct branch_entry *entries = perf_sample__branch_entries(sample);
+       bool callstack = evsel__has_branch_callstack(evsel);
+       u64 *branch_stack_cntr = sample->branch_stack_cntr;
+       struct perf_env *env = evsel__env(evsel);
        uint64_t i;
 
        if (!callstack) {
@@ -1194,6 +1198,13 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack)
                        }
                }
        }
+
+       if (branch_stack_cntr) {
+               printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n",
+                       sample->branch_stack->nr, env->br_cntr_width, env->br_cntr_nr);
+               for (i = 0; i < sample->branch_stack->nr; i++)
+                       printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]);
+       }
 }
 
 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
@@ -1355,7 +1366,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,
                callchain__printf(evsel, sample);
 
        if (evsel__has_br_stack(evsel))
-               branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
+               branch_stack__printf(sample, evsel);
 
        if (sample_type & PERF_SAMPLE_REGS_USER)
                regs_user__printf(sample, arch);
--

2.35.1

author
#19 lore raw

Re: [PATCH V5 8/8] perf tools: Add branch counter knob

Quoted 25 lines · Ian Rogers
Quoted 24 lines · On Wed, Oct 25, 2023 at 1:16 PM <[email protected]> wrote:

From: Kan Liang <[email protected]>

Add a new branch filter, "counter", for the branch counter option. It is used to mark the events which should be logged in the branch. If it is applied with the -j option, the counters of all the events should be logged in the branch. If the legacy kernel doesn't support the new branch sample type, switching off the branch counter filter.

The stored counter values in each branch are displayed right after the regular branch stack information via perf report -D.

Usage examples:

perf record -e "{branch-instructions,branch-misses}:S" -j any,counter

Only the first event, branch-instructions, collect the LBR. Both branch-instructions and branch-misses are marked as logged events. The occurrences information of them can be found in the branch stack extension space of each branch.

perf record -e "{cpu/branch-instructions,branch_type=any/, cpu/branch-misses,branch_type=counter/}"

Quoted 6 lines
Quoted 4 lines

Only the first event, branch-instructions, collect the LBR. Only the branch-misses event is marked as a logged event.

Thanks Ian for the review.

Quoted 4 lines

Perhaps add a test somewhere like:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/record.sh
for coverage when hardware supports the feature.

Sure, I will add a test when posting the perf tool patches.

Thanks,
Kan

Quoted 188 lines

Thanks,
Ian

Quoted 184 lines

---

No changes since V4

 tools/perf/Documentation/perf-record.txt  |  4 +++
 tools/perf/util/evsel.c                   | 31 ++++++++++++++++++++++-
 tools/perf/util/evsel.h                   |  1 +
 tools/perf/util/parse-branch-options.c    |  1 +
 tools/perf/util/perf_event_attr_fprintf.c |  1 +
 tools/perf/util/sample.h                  |  1 +
 tools/perf/util/session.c                 | 15 +++++++++--
 7 files changed, 51 insertions(+), 3 deletions(-)
tools/perf/Documentation/perf-record.txt +4 -0
index d5217be012d7..b6afe7cc948d 100644
@@ -442,6 +442,10 @@ following filters are defined:
                     4th-Gen Xeon+ server), the save branch type is unconditionally enabled
                     when the taken branch stack sampling is enabled.
        - priv: save privilege state during sampling in case binary is not available later
+       - counter: save occurrences of the event since the last branch entry. Currently, the
+                  feature is only supported by a newer CPU, e.g., Intel Sierra Forest and
+                  later platforms. An error out is expected if it's used on the unsupported
+                  kernel or CPUs.
 
 +
 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
tools/perf/util/evsel.c +30 -1
index a8a5ff87cc1f..58a9b8c82790 100644
@@ -1831,6 +1831,8 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
 
 static void evsel__disable_missing_features(struct evsel *evsel)
 {
+       if (perf_missing_features.branch_counters)
+               evsel->core.attr.branch_sample_type &= ~PERF_SAMPLE_BRANCH_COUNTERS;
        if (perf_missing_features.read_lost)
                evsel->core.attr.read_format &= ~PERF_FORMAT_LOST;
        if (perf_missing_features.weight_struct) {
@@ -1884,7 +1886,12 @@ bool evsel__detect_missing_features(struct evsel *evsel)
         * Must probe features in the order they were added to the
         * perf_event_attr interface.
         */
-       if (!perf_missing_features.read_lost &&
+       if (!perf_missing_features.branch_counters &&
+           (evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS)) {
+               perf_missing_features.branch_counters = true;
+               pr_debug2("switching off branch counters support\n");
+               return true;
+       } else if (!perf_missing_features.read_lost &&
            (evsel->core.attr.read_format & PERF_FORMAT_LOST)) {
                perf_missing_features.read_lost = true;
                pr_debug2("switching off PERF_FORMAT_LOST support\n");
@@ -2344,6 +2351,18 @@ u64 evsel__bitfield_swap_branch_flags(u64 value)
        return new_val;
 }
 
+static inline bool evsel__has_branch_counters(const struct evsel *evsel)
+{
+       struct evsel *cur, *leader = evsel__leader(evsel);
+
+       evlist__for_each_entry(evsel->evlist, cur) {
+               if ((leader == evsel__leader(cur)) &&
+                   (cur->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS))
+                       return true;
+       }
+       return false;
+}
+
 int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
                        struct perf_sample *data)
 {
@@ -2577,6 +2596,16 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
 
                OVERFLOW_CHECK(array, sz, max_size);
                array = (void *)array + sz;
+
+               if (evsel__has_branch_counters(evsel)) {
+                       OVERFLOW_CHECK_u64(array);
+
+                       data->branch_stack_cntr = (u64 *)array;
+                       sz = data->branch_stack->nr * sizeof(u64);
+
+                       OVERFLOW_CHECK(array, sz, max_size);
+                       array = (void *)array + sz;
+               }
        }
 
        if (type & PERF_SAMPLE_REGS_USER) {
tools/perf/util/evsel.h +1 -0
index 848534ec74fa..85f24c986392 100644
@@ -191,6 +191,7 @@ struct perf_missing_features {
        bool code_page_size;
        bool weight_struct;
        bool read_lost;
+       bool branch_counters;
 };
 
 extern struct perf_missing_features perf_missing_features;
tools/perf/util/parse-branch-options.c +1 -0
index fd67d204d720..f7f7aff3d85a 100644
@@ -36,6 +36,7 @@ static const struct branch_mode branch_modes[] = {
        BRANCH_OPT("stack", PERF_SAMPLE_BRANCH_CALL_STACK),
        BRANCH_OPT("hw_index", PERF_SAMPLE_BRANCH_HW_INDEX),
        BRANCH_OPT("priv", PERF_SAMPLE_BRANCH_PRIV_SAVE),
+       BRANCH_OPT("counter", PERF_SAMPLE_BRANCH_COUNTERS),
        BRANCH_END
 };
 
tools/perf/util/perf_event_attr_fprintf.c +1 -0
index 2247991451f3..8f04d3b7f3ec 100644
@@ -55,6 +55,7 @@ static void __p_branch_sample_type(char *buf, size_t size, u64 value)
                bit_name(COND), bit_name(CALL_STACK), bit_name(IND_JUMP),
                bit_name(CALL), bit_name(NO_FLAGS), bit_name(NO_CYCLES),
                bit_name(TYPE_SAVE), bit_name(HW_INDEX), bit_name(PRIV_SAVE),
+               bit_name(COUNTERS),
                { .name = NULL, }
        };
 #undef bit_name
tools/perf/util/sample.h +1 -0
index c92ad0f51ecd..70b2c3135555 100644
@@ -113,6 +113,7 @@ struct perf_sample {
        void *raw_data;
        struct ip_callchain *callchain;
        struct branch_stack *branch_stack;
+       u64 *branch_stack_cntr;
        struct regs_dump  user_regs;
        struct regs_dump  intr_regs;
        struct stack_dump user_stack;
tools/perf/util/session.c +13 -3
index 1e9aa8ed15b6..4a094ab0362b 100644
@@ -1150,9 +1150,13 @@ static void callchain__printf(struct evsel *evsel,
                       i, callchain->ips[i]);
 }
 
-static void branch_stack__printf(struct perf_sample *sample, bool callstack)
+static void branch_stack__printf(struct perf_sample *sample,
+                                struct evsel *evsel)
 {
        struct branch_entry *entries = perf_sample__branch_entries(sample);
+       bool callstack = evsel__has_branch_callstack(evsel);
+       u64 *branch_stack_cntr = sample->branch_stack_cntr;
+       struct perf_env *env = evsel__env(evsel);
        uint64_t i;
 
        if (!callstack) {
@@ -1194,6 +1198,13 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack)
                        }
                }
        }
+
+       if (branch_stack_cntr) {
+               printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n",
+                       sample->branch_stack->nr, env->br_cntr_width, env->br_cntr_nr);
+               for (i = 0; i < sample->branch_stack->nr; i++)
+                       printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]);
+       }
 }
 
 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
@@ -1355,7 +1366,7 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,
                callchain__printf(evsel, sample);
 
        if (evsel__has_br_stack(evsel))
-               branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
+               branch_stack__printf(sample, evsel);
 
        if (sample_type & PERF_SAMPLE_REGS_USER)
                regs_user__printf(sample, arch);
--

2.35.1

#2 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

In reply to [email protected]

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Quoted 13 lines

From: Kan Liang <[email protected]>

Currently, branch_sample_type !=0 is used to check whether a branch stack setup is required. But it doesn't check the sample type, unnecessary branch stack setup may be done for a counting event. E.g., perf record -e "{branch-instructions,branch-misses}:S" -j any Also, the event only with the new PERF_SAMPLE_BRANCH_COUNTERS branch sample type may not require a branch stack setup either.

Add a new flag NEEDS_BRANCH_STACK to indicate whether the event requires a branch stack setup. Replace the needs_branch_stack() by checking the new flag.

Quoted 7 lines

The counting event check is implemented here. The later patch will take the new PERF_SAMPLE_BRANCH_COUNTERS into account.

---

No changes since V4

So I saw this on tip/perf/urgent, I'm picking the tools bits then.

- Arnaldo

Quoted 57 lines
 arch/x86/events/intel/core.c       | 14 +++++++++++---
 arch/x86/events/perf_event_flags.h |  1 +
 2 files changed, 12 insertions(+), 3 deletions(-)
arch/x86/events/intel/core.c +11 -3
index 41a164764a84..a99449c0d77c 100644
@@ -2527,9 +2527,14 @@ static void intel_pmu_assign_event(struct perf_event *event, int idx)
 		perf_report_aux_output_id(event, idx);
 }
 
+static __always_inline bool intel_pmu_needs_branch_stack(struct perf_event *event)
+{
+	return event->hw.flags & PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+}
+
 static void intel_pmu_del_event(struct perf_event *event)
 {
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_del(event);
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_del(event);
@@ -2820,7 +2825,7 @@ static void intel_pmu_add_event(struct perf_event *event)
 {
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_add(event);
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_add(event);
 }
 
@@ -3897,7 +3902,10 @@ static int intel_pmu_hw_config(struct perf_event *event)
 			x86_pmu.pebs_aliases(event);
 	}
 
-	if (needs_branch_stack(event)) {
+	if (needs_branch_stack(event) && is_sampling_event(event))
+		event->hw.flags  |= PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+
+	if (intel_pmu_needs_branch_stack(event)) {
 		ret = intel_pmu_setup_lbr_filter(event);
 		if (ret)
 			return ret;
arch/x86/events/perf_event_flags.h +1 -0
index 1dc19b9b4426..a1685981c520 100644
@@ -20,3 +20,4 @@ PERF_ARCH(TOPDOWN,		0x04000) /* Count Topdown slots/metrics events */
 PERF_ARCH(PEBS_STLAT,		0x08000) /* st+stlat data address sampling */
 PERF_ARCH(AMD_BRS,		0x10000) /* AMD Branch Sampling */
 PERF_ARCH(PEBS_LAT_HYBRID,	0x20000) /* ld and st lat for hybrid */
+PERF_ARCH(NEEDS_BRANCH_STACK,	0x40000) /* require branch stack setup */
Signature · 1 line
2.35.1
Signature · 1 line
- Arnaldo
author
#3 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Quoted 17 lines · Arnaldo Carvalho de Melo

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Quoted 16 lines

From: Kan Liang <[email protected]>

Currently, branch_sample_type !=0 is used to check whether a branch stack setup is required. But it doesn't check the sample type, unnecessary branch stack setup may be done for a counting event. E.g., perf record -e "{branch-instructions,branch-misses}:S" -j any Also, the event only with the new PERF_SAMPLE_BRANCH_COUNTERS branch sample type may not require a branch stack setup either.

Add a new flag NEEDS_BRANCH_STACK to indicate whether the event requires a branch stack setup. Replace the needs_branch_stack() by checking the new flag.

The counting event check is implemented here. The later patch will take the new PERF_SAMPLE_BRANCH_COUNTERS into account.

Quoted 6 lines
Quoted 4 lines

---

No changes since V4

So I saw this on tip/perf/urgent, I'm picking the tools bits then.

Thanks Arnaldo.

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Kan

Quoted 61 lines

- Arnaldo

Quoted 57 lines
 arch/x86/events/intel/core.c       | 14 +++++++++++---
 arch/x86/events/perf_event_flags.h |  1 +
 2 files changed, 12 insertions(+), 3 deletions(-)
arch/x86/events/intel/core.c +11 -3
index 41a164764a84..a99449c0d77c 100644
@@ -2527,9 +2527,14 @@ static void intel_pmu_assign_event(struct perf_event *event, int idx)
 		perf_report_aux_output_id(event, idx);
 }
 
+static __always_inline bool intel_pmu_needs_branch_stack(struct perf_event *event)
+{
+	return event->hw.flags & PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+}
+
 static void intel_pmu_del_event(struct perf_event *event)
 {
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_del(event);
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_del(event);
@@ -2820,7 +2825,7 @@ static void intel_pmu_add_event(struct perf_event *event)
 {
 	if (event->attr.precise_ip)
 		intel_pmu_pebs_add(event);
-	if (needs_branch_stack(event))
+	if (intel_pmu_needs_branch_stack(event))
 		intel_pmu_lbr_add(event);
 }
 
@@ -3897,7 +3902,10 @@ static int intel_pmu_hw_config(struct perf_event *event)
 			x86_pmu.pebs_aliases(event);
 	}
 
-	if (needs_branch_stack(event)) {
+	if (needs_branch_stack(event) && is_sampling_event(event))
+		event->hw.flags  |= PERF_X86_EVENT_NEEDS_BRANCH_STACK;
+
+	if (intel_pmu_needs_branch_stack(event)) {
 		ret = intel_pmu_setup_lbr_filter(event);
 		if (ret)
 			return ret;
arch/x86/events/perf_event_flags.h +1 -0
index 1dc19b9b4426..a1685981c520 100644
@@ -20,3 +20,4 @@ PERF_ARCH(TOPDOWN,		0x04000) /* Count Topdown slots/metrics events */
 PERF_ARCH(PEBS_STLAT,		0x08000) /* st+stlat data address sampling */
 PERF_ARCH(AMD_BRS,		0x10000) /* AMD Branch Sampling */
 PERF_ARCH(PEBS_LAT_HYBRID,	0x20000) /* ld and st lat for hybrid */
+PERF_ARCH(NEEDS_BRANCH_STACK,	0x40000) /* require branch stack setup */
Signature · 1 line
2.35.1
#4 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Em Mon, Nov 06, 2023 at 04:19:13PM -0500, Liang, Kan escreveu:

Quoted 25 lines · Arnaldo Carvalho de Melo
Quoted 22 lines

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Quoted 20 lines

From: Kan Liang <[email protected]>

Currently, branch_sample_type !=0 is used to check whether a branch stack setup is required. But it doesn't check the sample type, unnecessary branch stack setup may be done for a counting event. E.g., perf record -e "{branch-instructions,branch-misses}:S" -j any Also, the event only with the new PERF_SAMPLE_BRANCH_COUNTERS branch sample type may not require a branch stack setup either.

Add a new flag NEEDS_BRANCH_STACK to indicate whether the event requires a branch stack setup. Replace the needs_branch_stack() by checking the new flag.

The counting event check is implemented here. The later patch will take the new PERF_SAMPLE_BRANCH_COUNTERS into account.

---

No changes since V4

Quoted 8 lines
Quoted 1 line

So I saw this on tip/perf/urgent, I'm picking the tools bits then.

Thanks Arnaldo.

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

I saw Ian's suggestion, and agree with it, we need to pair new features with regression tests in 'perf test', thanks for working on it!

- Arnaldo

#5 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Em Tue, Nov 07, 2023 at 12:11:50PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 10 lines

Em Mon, Nov 06, 2023 at 04:19:13PM -0500, Liang, Kan escreveu:

Quoted 6 lines · Arnaldo Carvalho de Melo
Quoted 1 line

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

I saw Ian's suggestion, and agree with it, we need to pair new features with regression tests in 'perf test', thanks for working on it!

Kan,

I still have to bisect, but can you check if this works for you?

(gdb) run test -F -v 27
Starting program: /root/bin/perf test -F -v 27

27: Sample parsing :
--- start ---

Program received signal SIGSEGV, Segmentation fault.
0x00000000004e4aa6 in evsel.parse_sample ()
Missing separate debuginfos, use: dnf debuginfo-install bzip2-libs-1.0.8-13.fc38.x86_64 cyrus-sasl-lib-2.1.28-9.fc38.x86_64 elfutils-debuginfod-client-0.189-3.fc38.x86_64 elfutils-libelf-0.189-3.fc38.x86_64 elfutils-libs-0.189-3.fc38.x86_64 glib2-2.76.5-2.fc38.x86_64 glibc-2.37-13.fc38.x86_64 keyutils-libs-1.6.1-6.fc38.x86_64 krb5-libs-1.21-3.fc38.x86_64 libbabeltrace-1.5.11-2.fc38.x86_64 libbrotli-1.0.9-11.fc38.x86_64 libcap-2.48-6.fc38.x86_64 libcom_err-1.46.5-4.fc38.x86_64 libcurl-8.0.1-5.fc38.x86_64 libevent-2.1.12-8.fc38.x86_64 libgcc-13.2.1-4.fc38.x86_64 libidn2-2.3.4-2.fc38.x86_64 libnghttp2-1.52.0-2.fc38.x86_64 libpfm-4.11.0-11.fc38.x86_64 libpsl-0.21.2-2.fc38.x86_64 libselinux-3.5-1.fc38.x86_64 libssh-0.10.5-1.fc38.x86_64 libstdc++-13.2.1-4.fc38.x86_64 libtraceevent-1.7.2-1.fc38.x86_64 libunistring1.0-1.0-1.fc38.x86_64 libunwind-1.6.2-7.fc38.x86_64 libuuid-2.38.1-4.fc38.x86_64 libxcrypt-4.4.36-1.fc38.x86_64 libzstd-1.5.5-1.fc38.x86_64 opencsd-1.3.3-1.fc38.x86_64 openldap-2.6.6-1.fc38.x86_64 openssl-libs-3.0.9-2.fc38.x86_64 perl-libs-5.36.1-497.fc38.x86_64 popt-1.19-2.fc38.x86_64 python3-libs-3.11.6-1.fc38.x86_64 slang-2.3.3-3.fc38.x86_64 zlib-1.2.13-3.fc38.x86_64 (gdb) bt

#0  0x00000000004e4aa6 in evsel.parse_sample ()
#1  0x00000000004b28dc in do_test ()
#2  0x00000000004b2acd in test.sample_parsing ()
#3  0x0000000000495348 in test_and_print.isra ()
#4  0x0000000000495f5d in cmd_test ()
#5  0x00000000004c2a29 in run_builtin ()
#6  0x000000000041053f in main ()

(gdb)

author
#6 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Quoted 9 lines · Arnaldo Carvalho de Melo

Em Tue, Nov 07, 2023 at 12:11:50PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 7 lines

Em Mon, Nov 06, 2023 at 04:19:13PM -0500, Liang, Kan escreveu:

Quoted 6 lines · Arnaldo Carvalho de Melo
Quoted 1 line

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

Quoted 6 lines
Quoted 2 lines

I saw Ian's suggestion, and agree with it, we need to pair new features with regression tests in 'perf test', thanks for working on it!

Kan,

I still have to bisect, but can you check if this works for you?

The branch counters feature requires all the events to belong to a group. There is no problem for the normal perf usage which usually initializes an evlist even for a single evsel.
But perf test is special, which may not initialize an evlist. The Sample parsing test case is one of the examples. It crashes with the !evsel->evlist.

The below change should fix it. I will post a complete patch shortly.

tools/perf/util/evsel.c +0 -0
index 58a9b8c82790..7a6a2d1f96db 100644
@@ -2355,6 +2355,10 @@ static inline bool

evsel__has_branch_counters(const struct evsel *evsel)

 {
        struct evsel *cur, *leader = evsel__leader(evsel);

+ /* The branch counters feature only supports group */
+ if (!leader || !evsel->evlist)
+ return false;
+

        evlist__for_each_entry(evsel->evlist, cur) {
                if ((leader == evsel__leader(cur)) &&
                    (cur->core.attr.branch_sample_type &

PERF_SAMPLE_BRANCH_COUNTERS))

Thanks,
Kan

Quoted 21 lines

(gdb) run test -F -v 27
Starting program: /root/bin/perf test -F -v 27

27: Sample parsing :
--- start ---

Program received signal SIGSEGV, Segmentation fault.
0x00000000004e4aa6 in evsel.parse_sample ()
Missing separate debuginfos, use: dnf debuginfo-install bzip2-libs-1.0.8-13.fc38.x86_64 cyrus-sasl-lib-2.1.28-9.fc38.x86_64 elfutils-debuginfod-client-0.189-3.fc38.x86_64 elfutils-libelf-0.189-3.fc38.x86_64 elfutils-libs-0.189-3.fc38.x86_64 glib2-2.76.5-2.fc38.x86_64 glibc-2.37-13.fc38.x86_64 keyutils-libs-1.6.1-6.fc38.x86_64 krb5-libs-1.21-3.fc38.x86_64 libbabeltrace-1.5.11-2.fc38.x86_64 libbrotli-1.0.9-11.fc38.x86_64 libcap-2.48-6.fc38.x86_64 libcom_err-1.46.5-4.fc38.x86_64 libcurl-8.0.1-5.fc38.x86_64 libevent-2.1.12-8.fc38.x86_64 libgcc-13.2.1-4.fc38.x86_64 libidn2-2.3.4-2.fc38.x86_64 libnghttp2-1.52.0-2.fc38.x86_64 libpfm-4.11.0-11.fc38.x86_64 libpsl-0.21.2-2.fc38.x86_64 libselinux-3.5-1.fc38.x86_64 libssh-0.10.5-1.fc38.x86_64 libstdc++-13.2.1-4.fc38.x86_64 libtraceevent-1.7.2-1.fc38.x86_64 libunistring1.0-1.0-1.fc38.x86_64 libunwind-1.6.2-7.fc38.x86_64 libuuid-2.38.1-4.fc38.x86_64 libxcrypt-4.4.36-1.fc38.x86_64 libzstd-1.5.5-1.fc38.x86_64 opencsd-1.3.3-1.fc38.x86_64 openldap-2.6.6-1.fc38.x86_64 openssl-libs-3.0.9-2.fc38.x86_64 perl-libs-5.36.1-497.fc38.x86_64 popt-1.19-2.fc38.x86_64 python3-libs-3.11.6-1.fc38.x86_64 slang-2.3.3-3.fc38.x86_64 zlib-1.2.13-3.fc38.x86_64 (gdb) bt

#0  0x00000000004e4aa6 in evsel.parse_sample ()
#1  0x00000000004b28dc in do_test ()
#2  0x00000000004b2acd in test.sample_parsing ()
#3  0x0000000000495348 in test_and_print.isra ()
#4  0x0000000000495f5d in cmd_test ()
#5  0x00000000004c2a29 in run_builtin ()
#6  0x000000000041053f in main ()

(gdb)

#7 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Em Thu, Nov 09, 2023 at 11:14:31AM -0500, Liang, Kan escreveu:

Quoted 19 lines · Arnaldo Carvalho de Melo
Quoted 9 lines

Em Tue, Nov 07, 2023 at 12:11:50PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 7 lines

Em Mon, Nov 06, 2023 at 04:19:13PM -0500, Liang, Kan escreveu:

Quoted 6 lines · Arnaldo Carvalho de Melo
Quoted 1 line

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

Quoted 6 lines
Quoted 2 lines

I saw Ian's suggestion, and agree with it, we need to pair new features with regression tests in 'perf test', thanks for working on it!

Kan,

I still have to bisect, but can you check if this works for you?

Quoted 8 lines

The branch counters feature requires all the events to belong to a group. There is no problem for the normal perf usage which usually initializes an evlist even for a single evsel.
But perf test is special, which may not initialize an evlist. The Sample parsing test case is one of the examples. It crashes with the !evsel->evlist.

The below change should fix it. I will post a complete patch shortly.

Thanks for the quick response, if all that is needed are the checks below, I'll fold it into your original patch:

2ae01908298426e4 perf tools: Add branch counter knob

So that we don't regress, ok?

I'll add a note and the Link tag points to this discussion in case people want to do historical digs in the future :-)

- Arnaldo

Quoted 42 lines
tools/perf/util/evsel.c +4 -0
index 58a9b8c82790..7a6a2d1f96db 100644
@@ -2355,6 +2355,10 @@ static inline bool
 evsel__has_branch_counters(const struct evsel *evsel)
 {
        struct evsel *cur, *leader = evsel__leader(evsel);
 
+       /* The branch counters feature only supports group */
+       if (!leader || !evsel->evlist)
+               return false;
+
        evlist__for_each_entry(evsel->evlist, cur) {
                if ((leader == evsel__leader(cur)) &&
                    (cur->core.attr.branch_sample_type &

PERF_SAMPLE_BRANCH_COUNTERS))

Thanks,
Kan

Quoted 21 lines

(gdb) run test -F -v 27
Starting program: /root/bin/perf test -F -v 27

27: Sample parsing :
--- start ---

Program received signal SIGSEGV, Segmentation fault.
0x00000000004e4aa6 in evsel.parse_sample ()
Missing separate debuginfos, use: dnf debuginfo-install bzip2-libs-1.0.8-13.fc38.x86_64 cyrus-sasl-lib-2.1.28-9.fc38.x86_64 elfutils-debuginfod-client-0.189-3.fc38.x86_64 elfutils-libelf-0.189-3.fc38.x86_64 elfutils-libs-0.189-3.fc38.x86_64 glib2-2.76.5-2.fc38.x86_64 glibc-2.37-13.fc38.x86_64 keyutils-libs-1.6.1-6.fc38.x86_64 krb5-libs-1.21-3.fc38.x86_64 libbabeltrace-1.5.11-2.fc38.x86_64 libbrotli-1.0.9-11.fc38.x86_64 libcap-2.48-6.fc38.x86_64 libcom_err-1.46.5-4.fc38.x86_64 libcurl-8.0.1-5.fc38.x86_64 libevent-2.1.12-8.fc38.x86_64 libgcc-13.2.1-4.fc38.x86_64 libidn2-2.3.4-2.fc38.x86_64 libnghttp2-1.52.0-2.fc38.x86_64 libpfm-4.11.0-11.fc38.x86_64 libpsl-0.21.2-2.fc38.x86_64 libselinux-3.5-1.fc38.x86_64 libssh-0.10.5-1.fc38.x86_64 libstdc++-13.2.1-4.fc38.x86_64 libtraceevent-1.7.2-1.fc38.x86_64 libunistring1.0-1.0-1.fc38.x86_64 libunwind-1.6.2-7.fc38.x86_64 libuuid-2.38.1-4.fc38.x86_64 libxcrypt-4.4.36-1.fc38.x86_64 libzstd-1.5.5-1.fc38.x86_64 opencsd-1.3.3-1.fc38.x86_64 openldap-2.6.6-1.fc38.x86_64 openssl-libs-3.0.9-2.fc38.x86_64 perl-libs-5.36.1-497.fc38.x86_64 popt-1.19-2.fc38.x86_64 python3-libs-3.11.6-1.fc38.x86_64 slang-2.3.3-3.fc38.x86_64 zlib-1.2.13-3.fc38.x86_64 (gdb) bt

#0  0x00000000004e4aa6 in evsel.parse_sample ()
#1  0x00000000004b28dc in do_test ()
#2  0x00000000004b2acd in test.sample_parsing ()
#3  0x0000000000495348 in test_and_print.isra ()
#4  0x0000000000495f5d in cmd_test ()
#5  0x00000000004c2a29 in run_builtin ()
#6  0x000000000041053f in main ()

(gdb)

Signature · 1 line
- Arnaldo
author
#8 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Quoted 27 lines · Arnaldo Carvalho de Melo

Em Thu, Nov 09, 2023 at 11:14:31AM -0500, Liang, Kan escreveu:

Quoted 26 lines · Arnaldo Carvalho de Melo
Quoted 9 lines

Em Tue, Nov 07, 2023 at 12:11:50PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 7 lines

Em Mon, Nov 06, 2023 at 04:19:13PM -0500, Liang, Kan escreveu:

Quoted 6 lines · Arnaldo Carvalho de Melo
Quoted 1 line

Em Wed, Oct 25, 2023 at 01:16:20PM -0700, [email protected] escreveu:

Ian has already reviewed the tool parts.

But I still owe a test case for the feature. I will post a patch later.
https://lore.kernel.org/lkml/[email protected]/

Quoted 6 lines
Quoted 2 lines

I saw Ian's suggestion, and agree with it, we need to pair new features with regression tests in 'perf test', thanks for working on it!

Kan,

I still have to bisect, but can you check if this works for you?

The branch counters feature requires all the events to belong to a group. There is no problem for the normal perf usage which usually initializes an evlist even for a single evsel.
But perf test is special, which may not initialize an evlist. The Sample parsing test case is one of the examples. It crashes with the !evsel->evlist.

Quoted 8 lines
Quoted 1 line

The below change should fix it. I will post a complete patch shortly.

Thanks for the quick response, if all that is needed are the checks below, I'll fold it into your original patch:

2ae01908298426e4 perf tools: Add branch counter knob

So that we don't regress, ok?

Sure.

I also post the patch to
https://lore.kernel.org/lkml/[email protected]/
Either folding it or using the new patch is fine for me.

BTW: the new perf test case for the feature is posted here. I think Ian is reviewing it.
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Kan

Quoted 49 lines

I'll add a note and the Link tag points to this discussion in case people want to do historical digs in the future :-)

- Arnaldo

Quoted 42 lines
tools/perf/util/evsel.c +4 -0
index 58a9b8c82790..7a6a2d1f96db 100644
@@ -2355,6 +2355,10 @@ static inline bool
 evsel__has_branch_counters(const struct evsel *evsel)
 {
        struct evsel *cur, *leader = evsel__leader(evsel);
 
+       /* The branch counters feature only supports group */
+       if (!leader || !evsel->evlist)
+               return false;
+
        evlist__for_each_entry(evsel->evlist, cur) {
                if ((leader == evsel__leader(cur)) &&
                    (cur->core.attr.branch_sample_type &

PERF_SAMPLE_BRANCH_COUNTERS))

Thanks,
Kan

Quoted 21 lines

(gdb) run test -F -v 27
Starting program: /root/bin/perf test -F -v 27

27: Sample parsing :
--- start ---

Program received signal SIGSEGV, Segmentation fault.
0x00000000004e4aa6 in evsel.parse_sample ()
Missing separate debuginfos, use: dnf debuginfo-install bzip2-libs-1.0.8-13.fc38.x86_64 cyrus-sasl-lib-2.1.28-9.fc38.x86_64 elfutils-debuginfod-client-0.189-3.fc38.x86_64 elfutils-libelf-0.189-3.fc38.x86_64 elfutils-libs-0.189-3.fc38.x86_64 glib2-2.76.5-2.fc38.x86_64 glibc-2.37-13.fc38.x86_64 keyutils-libs-1.6.1-6.fc38.x86_64 krb5-libs-1.21-3.fc38.x86_64 libbabeltrace-1.5.11-2.fc38.x86_64 libbrotli-1.0.9-11.fc38.x86_64 libcap-2.48-6.fc38.x86_64 libcom_err-1.46.5-4.fc38.x86_64 libcurl-8.0.1-5.fc38.x86_64 libevent-2.1.12-8.fc38.x86_64 libgcc-13.2.1-4.fc38.x86_64 libidn2-2.3.4-2.fc38.x86_64 libnghttp2-1.52.0-2.fc38.x86_64 libpfm-4.11.0-11.fc38.x86_64 libpsl-0.21.2-2.fc38.x86_64 libselinux-3.5-1.fc38.x86_64 libssh-0.10.5-1.fc38.x86_64 libstdc++-13.2.1-4.fc38.x86_64 libtraceevent-1.7.2-1.fc38.x86_64 libunistring1.0-1.0-1.fc38.x86_64 libunwind-1.6.2-7.fc38.x86_64 libuuid-2.38.1-4.fc38.x86_64 libxcrypt-4.4.36-1.fc38.x86_64 libzstd-1.5.5-1.fc38.x86_64 opencsd-1.3.3-1.fc38.x86_64 openldap-2.6.6-1.fc38.x86_64 openssl-libs-3.0.9-2.fc38.x86_64 perl-libs-5.36.1-497.fc38.x86_64 popt-1.19-2.fc38.x86_64 python3-libs-3.11.6-1.fc38.x86_64 slang-2.3.3-3.fc38.x86_64 zlib-1.2.13-3.fc38.x86_64 (gdb) bt

#0  0x00000000004e4aa6 in evsel.parse_sample ()
#1  0x00000000004b28dc in do_test ()
#2  0x00000000004b2acd in test.sample_parsing ()
#3  0x0000000000495348 in test_and_print.isra ()
#4  0x0000000000495f5d in cmd_test ()
#5  0x00000000004c2a29 in run_builtin ()
#6  0x000000000041053f in main ()

(gdb)

#9 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Em Thu, Nov 09, 2023 at 12:05:27PM -0500, Liang, Kan escreveu:

Quoted 9 lines · Arnaldo Carvalho de Melo
Quoted 7 lines

Em Thu, Nov 09, 2023 at 11:14:31AM -0500, Liang, Kan escreveu:

Quoted 1 line

The below change should fix it. I will post a complete patch shortly.

Thanks for the quick response, if all that is needed are the checks below, I'll fold it into your original patch:

2ae01908298426e4 perf tools: Add branch counter knob

Quoted 7 lines
Quoted 1 line

So that we don't regress, ok?

Sure.

I also post the patch to
https://lore.kernel.org/lkml/[email protected]/
Either folding it or using the new patch is fine for me.

I folded it, retested, pushed out perf-tools-next.

Quoted 3 lines

BTW: the new perf test case for the feature is posted here. I think Ian is reviewing it.
https://lore.kernel.org/lkml/[email protected]/

Ok, lets wait some more.

Hey, what is SFR/GRR? Sapphire Rapids/Granite Rapids? I thought about testing this somehow, if possible.

- Arnaldo

author
#10 lore raw

Re: [PATCH V5 2/8] perf/x86: Add PERF_X86_EVENT_NEEDS_BRANCH_STACK flag

Quoted 12 lines · Arnaldo Carvalho de Melo

Em Thu, Nov 09, 2023 at 12:05:27PM -0500, Liang, Kan escreveu:

Quoted 10 lines · Arnaldo Carvalho de Melo
Quoted 9 lines

Em Thu, Nov 09, 2023 at 11:14:31AM -0500, Liang, Kan escreveu:

Quoted 1 line

The below change should fix it. I will post a complete patch shortly.

Thanks for the quick response, if all that is needed are the checks below, I'll fold it into your original patch:

2ae01908298426e4 perf tools: Add branch counter knob

So that we don't regress, ok?

Quoted 7 lines
Quoted 5 lines

Sure.

I also post the patch to
https://lore.kernel.org/lkml/[email protected]/
Either folding it or using the new patch is fine for me.

I folded it, retested, pushed out perf-tools-next.

Thanks!

Quoted 10 lines
Quoted 4 lines

BTW: the new perf test case for the feature is posted here. I think Ian is reviewing it.
https://lore.kernel.org/lkml/[email protected]/

Ok, lets wait some more.

Hey, what is SFR/GRR? Sapphire Rapids/Granite Rapids? I thought about testing this somehow, if possible.

Sierra Forest/Grand Ridge. The feature is only available on the E-core based server for now.

Thanks,
Kan