[PATCH V2 1/5] perf cpumap: Retrieve die id information

[PATCH V2 1/5] perf cpumap: Retrieve die id information

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

There is no function to retrieve die id information of a given CPU.

Add cpu_map__get_die_id() to retrieve die id information.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---

New patch for V2.

 tools/perf/util/cpumap.c | 7 +++++++
 tools/perf/util/cpumap.h | 1 +
 2 files changed, 8 insertions(+)

diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c
index 0b59922..7db1365 100644
--- a/tools/perf/util/cpumap.c
+++ b/tools/perf/util/cpumap.c
@@ -373,6 +373,13 @@ int cpu_map__build_map(struct cpu_map *cpus, struct cpu_map **res,
 	return 0;
 }
 
+int cpu_map__get_die_id(int cpu)
+{
+	int value, ret = cpu__get_topology_int(cpu, "die_id", &value);
+
+	return ret ?: value;
+}
+
 int cpu_map__get_core_id(int cpu)
 {
 	int value, ret = cpu__get_topology_int(cpu, "core_id", &value);
diff --git a/tools/perf/util/cpumap.h b/tools/perf/util/cpumap.h
index f00ce62..6762ff9 100644
--- a/tools/perf/util/cpumap.h
+++ b/tools/perf/util/cpumap.h
@@ -25,6 +25,7 @@ size_t cpu_map__snprint_mask(struct cpu_map *map, char *buf, size_t size);
 size_t cpu_map__fprintf(struct cpu_map *map, FILE *fp);
 int cpu_map__get_socket_id(int cpu);
 int cpu_map__get_socket(struct cpu_map *map, int idx, void *data);
+int cpu_map__get_die_id(int cpu);
 int cpu_map__get_core_id(int cpu);
 int cpu_map__get_core(struct cpu_map *map, int idx, void *data);
 int cpu_map__build_socket_map(struct cpu_map *cpus, struct cpu_map **sockp);
-- 
2.7.4

[PATCH V2 2/5] perf header: Add die information in CPU topology

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

With the new CPUID.1F, a new level type of CPU topology, 'die', is
introduced. The 'die' information in CPU topology should be added in
perf header.

To be compatible with old perf.data, the patch checks the section size
before reading the die information. The new info is added at the end of
the cpu_topology section, the old perf tool ignores the extra data.
It never reads data crossing the section boundary.

The new perf tool with the patch can be used on legacy kernel. Add a
new function has_die_topology() to check if die topology information is
supported by kernel. The function only check X86 and CPU 0. Assuming
other CPUs have same topology.

Use similar method for core and socket to support die id and sibling
dies string.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---

Changes since V1:
- Avoid has_die parameters for many functions
- Move cpu_map__get_die_id() to separate patch
- Rename check_x86_die_exists() to has_die_topology()

 tools/perf/Documentation/perf.data-file-format.txt |  9 ++-
 tools/perf/util/cputopo.c                          | 76 +++++++++++++++--
 tools/perf/util/cputopo.h                          |  2 +
 tools/perf/util/env.c                              |  1 +
 tools/perf/util/env.h                              |  3 +
 tools/perf/util/header.c                           | 94 ++++++++++++++++++++--
 6 files changed, 172 insertions(+), 13 deletions(-)

diff --git a/tools/perf/Documentation/perf.data-file-format.txt b/tools/perf/Documentation/perf.data-file-format.txt
index 6967e9b..c731416 100644
--- a/tools/perf/Documentation/perf.data-file-format.txt
+++ b/tools/perf/Documentation/perf.data-file-format.txt
@@ -153,7 +153,7 @@ struct {
 
 String lists defining the core and CPU threads topology.
 The string lists are followed by a variable length array
-which contains core_id and socket_id of each cpu.
+which contains core_id, die_id (for x86) and socket_id of each cpu.
 The number of entries can be determined by the size of the
 section minus the sizes of both string lists.
 
@@ -162,14 +162,19 @@ struct {
        struct perf_header_string_list threads; /* Variable length */
        struct {
 	      uint32_t core_id;
+	      uint32_t die_id;
 	      uint32_t socket_id;
        } cpus[nr]; /* Variable length records */
 };
 
 Example:
-	sibling cores   : 0-3
+	sibling cores   : 0-8
+	sibling dies	: 0-3
+	sibling dies	: 4-7
 	sibling threads : 0-1
 	sibling threads : 2-3
+	sibling threads : 4-5
+	sibling threads : 6-7
 
 	HEADER_NUMA_TOPOLOGY = 14,
 
diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c
index ece0710..85fa87f 100644
--- a/tools/perf/util/cputopo.c
+++ b/tools/perf/util/cputopo.c
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
 #include <sys/param.h>
+#include <sys/utsname.h>
 #include <inttypes.h>
 #include <api/fs/fs.h>
 
@@ -8,9 +9,10 @@
 #include "util.h"
 #include "env.h"
 
-
 #define CORE_SIB_FMT \
 	"%s/devices/system/cpu/cpu%d/topology/core_siblings_list"
+#define DIE_SIB_FMT \
+	"%s/devices/system/cpu/cpu%d/topology/die_cpus_list"
 #define THRD_SIB_FMT \
 	"%s/devices/system/cpu/cpu%d/topology/thread_siblings_list"
 #define NODE_ONLINE_FMT \
@@ -34,12 +36,12 @@ static int build_cpu_topology(struct cpu_topology *tp, int cpu)
 		  sysfs__mountpoint(), cpu);
 	fp = fopen(filename, "r");
 	if (!fp)
-		goto try_threads;
+		goto try_dies;
 
 	sret = getline(&buf, &len, fp);
 	fclose(fp);
 	if (sret <= 0)
-		goto try_threads;
+		goto try_dies;
 
 	p = strchr(buf, '\n');
 	if (p)
@@ -57,6 +59,37 @@ static int build_cpu_topology(struct cpu_topology *tp, int cpu)
 	}
 	ret = 0;
 
+try_dies:
+	if (!tp->die_siblings)
+		goto try_threads;
+
+	scnprintf(filename, MAXPATHLEN, DIE_SIB_FMT,
+		  sysfs__mountpoint(), cpu);
+	fp = fopen(filename, "r");
+	if (!fp)
+		goto try_threads;
+
+	sret = getline(&buf, &len, fp);
+	fclose(fp);
+	if (sret <= 0)
+		goto try_threads;
+
+	p = strchr(buf, '\n');
+	if (p)
+		*p = '\0';
+
+	for (i = 0; i < tp->die_sib; i++) {
+		if (!strcmp(buf, tp->die_siblings[i]))
+			break;
+	}
+	if (i == tp->die_sib) {
+		tp->die_siblings[i] = buf;
+		tp->die_sib++;
+		buf = NULL;
+		len = 0;
+	}
+	ret = 0;
+
 try_threads:
 	scnprintf(filename, MAXPATHLEN, THRD_SIB_FMT,
 		  sysfs__mountpoint(), cpu);
@@ -98,21 +131,46 @@ void cpu_topology__delete(struct cpu_topology *tp)
 	for (i = 0 ; i < tp->core_sib; i++)
 		zfree(&tp->core_siblings[i]);
 
+	if (tp->die_sib) {
+		for (i = 0 ; i < tp->die_sib; i++)
+			zfree(&tp->die_siblings[i]);
+	}
+
 	for (i = 0 ; i < tp->thread_sib; i++)
 		zfree(&tp->thread_siblings[i]);
 
 	free(tp);
 }
 
+static bool has_die_topology(void)
+{
+	char filename[MAXPATHLEN];
+	struct utsname uts;
+
+	if (uname(&uts) < 0)
+		return false;
+
+	if (strncmp(uts.machine, "x86_64", 6))
+		return false;
+
+	scnprintf(filename, MAXPATHLEN, DIE_SIB_FMT,
+		  sysfs__mountpoint(), 0);
+	if (access(filename, F_OK) == -1)
+		return false;
+
+	return true;
+}
+
 struct cpu_topology *cpu_topology__new(void)
 {
 	struct cpu_topology *tp = NULL;
 	void *addr;
-	u32 nr, i;
+	u32 nr, i, nr_addr;
 	size_t sz;
 	long ncpus;
 	int ret = -1;
 	struct cpu_map *map;
+	bool has_die = has_die_topology();
 
 	ncpus = cpu__max_present_cpu();
 
@@ -126,7 +184,11 @@ struct cpu_topology *cpu_topology__new(void)
 	nr = (u32)(ncpus & UINT_MAX);
 
 	sz = nr * sizeof(char *);
-	addr = calloc(1, sizeof(*tp) + 2 * sz);
+	if (has_die)
+		nr_addr = 3;
+	else
+		nr_addr = 2;
+	addr = calloc(1, sizeof(*tp) + nr_addr * sz);
 	if (!addr)
 		goto out_free;
 
@@ -134,6 +196,10 @@ struct cpu_topology *cpu_topology__new(void)
 	addr += sizeof(*tp);
 	tp->core_siblings = addr;
 	addr += sz;
+	if (has_die) {
+		tp->die_siblings = addr;
+		addr += sz;
+	}
 	tp->thread_siblings = addr;
 
 	for (i = 0; i < nr; i++) {
diff --git a/tools/perf/util/cputopo.h b/tools/perf/util/cputopo.h
index 47a97e7..bae2f1d 100644
--- a/tools/perf/util/cputopo.h
+++ b/tools/perf/util/cputopo.h
@@ -7,8 +7,10 @@
 
 struct cpu_topology {
 	u32	  core_sib;
+	u32	  die_sib;
 	u32	  thread_sib;
 	char	**core_siblings;
+	char	**die_siblings;
 	char	**thread_siblings;
 };
 
diff --git a/tools/perf/util/env.c b/tools/perf/util/env.c
index 6a3eaf7..1cc7a18 100644
--- a/tools/perf/util/env.c
+++ b/tools/perf/util/env.c
@@ -246,6 +246,7 @@ int perf_env__read_cpu_topology_map(struct perf_env *env)
 	for (cpu = 0; cpu < nr_cpus; ++cpu) {
 		env->cpu[cpu].core_id	= cpu_map__get_core_id(cpu);
 		env->cpu[cpu].socket_id	= cpu_map__get_socket_id(cpu);
+		env->cpu[cpu].die_id	= cpu_map__get_die_id(cpu);
 	}
 
 	env->nr_cpus_avail = nr_cpus;
diff --git a/tools/perf/util/env.h b/tools/perf/util/env.h
index 271a90b..d5d9865 100644
--- a/tools/perf/util/env.h
+++ b/tools/perf/util/env.h
@@ -9,6 +9,7 @@
 
 struct cpu_topology_map {
 	int	socket_id;
+	int	die_id;
 	int	core_id;
 };
 
@@ -49,6 +50,7 @@ struct perf_env {
 
 	int			nr_cmdline;
 	int			nr_sibling_cores;
+	int			nr_sibling_dies;
 	int			nr_sibling_threads;
 	int			nr_numa_nodes;
 	int			nr_memory_nodes;
@@ -57,6 +59,7 @@ struct perf_env {
 	char			*cmdline;
 	const char		**cmdline_argv;
 	char			*sibling_cores;
+	char			*sibling_dies;
 	char			*sibling_threads;
 	char			*pmu_mappings;
 	struct cpu_topology_map	*cpu;
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 847ae51..6497625 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -599,6 +599,27 @@ static int write_cpu_topology(struct feat_fd *ff,
 		if (ret < 0)
 			return ret;
 	}
+
+	if (!tp->die_sib)
+		goto done;
+
+	ret = do_write(ff, &tp->die_sib, sizeof(tp->die_sib));
+	if (ret < 0)
+		goto done;
+
+	for (i = 0; i < tp->die_sib; i++) {
+		ret = do_write_string(ff, tp->die_siblings[i]);
+		if (ret < 0)
+			goto done;
+	}
+
+	for (j = 0; j < perf_env.nr_cpus_avail; j++) {
+		ret = do_write(ff, &perf_env.cpu[j].die_id,
+			       sizeof(perf_env.cpu[j].die_id));
+		if (ret < 0)
+			return ret;
+	}
+
 done:
 	cpu_topology__delete(tp);
 	return ret;
@@ -1443,6 +1464,16 @@ static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
 		str += strlen(str) + 1;
 	}
 
+	if (ph->env.nr_sibling_dies) {
+		nr = ph->env.nr_sibling_dies;
+		str = ph->env.sibling_dies;
+
+		for (i = 0; i < nr; i++) {
+			fprintf(fp, "# sibling dies    : %s\n", str);
+			str += strlen(str) + 1;
+		}
+	}
+
 	nr = ph->env.nr_sibling_threads;
 	str = ph->env.sibling_threads;
 
@@ -1451,12 +1482,28 @@ static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
 		str += strlen(str) + 1;
 	}
 
-	if (ph->env.cpu != NULL) {
-		for (i = 0; i < cpu_nr; i++)
-			fprintf(fp, "# CPU %d: Core ID %d, Socket ID %d\n", i,
-				ph->env.cpu[i].core_id, ph->env.cpu[i].socket_id);
-	} else
-		fprintf(fp, "# Core ID and Socket ID information is not available\n");
+	if (ph->env.nr_sibling_dies) {
+		if (ph->env.cpu != NULL) {
+			for (i = 0; i < cpu_nr; i++)
+				fprintf(fp, "# CPU %d: Core ID %d, "
+					    "Die ID %d, Socket ID %d\n",
+					    i, ph->env.cpu[i].core_id,
+					    ph->env.cpu[i].die_id,
+					    ph->env.cpu[i].socket_id);
+		} else
+			fprintf(fp, "# Core ID, Die ID and Socket ID "
+				    "information is not available\n");
+	} else {
+		if (ph->env.cpu != NULL) {
+			for (i = 0; i < cpu_nr; i++)
+				fprintf(fp, "# CPU %d: Core ID %d, "
+					    "Socket ID %d\n",
+					    i, ph->env.cpu[i].core_id,
+					    ph->env.cpu[i].socket_id);
+		} else
+			fprintf(fp, "# Core ID and Socket ID "
+				    "information is not available\n");
+	}
 }
 
 static void print_clockid(struct feat_fd *ff, FILE *fp)
@@ -2214,6 +2261,7 @@ static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
 			goto free_cpu;
 
 		ph->env.cpu[i].core_id = nr;
+		size += sizeof(u32);
 
 		if (do_read_u32(ff, &nr))
 			goto free_cpu;
@@ -2225,6 +2273,40 @@ static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
 		}
 
 		ph->env.cpu[i].socket_id = nr;
+		size += sizeof(u32);
+	}
+
+	/*
+	 * The header may be from old perf,
+	 * which doesn't include die information.
+	 */
+	if (ff->size <= size)
+		return 0;
+
+	if (do_read_u32(ff, &nr))
+		return -1;
+
+	ph->env.nr_sibling_dies = nr;
+	size += sizeof(u32);
+
+	for (i = 0; i < nr; i++) {
+		str = do_read_string(ff);
+		if (!str)
+			goto error;
+
+		/* include a NULL character at the end */
+		if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
+			goto error;
+		size += string_size(str);
+		free(str);
+	}
+	ph->env.sibling_dies = strbuf_detach(&sb, NULL);
+
+	for (i = 0; i < (u32)cpu_nr; i++) {
+		if (do_read_u32(ff, &nr))
+			goto free_cpu;
+
+		ph->env.cpu[i].die_id = nr;
 	}
 
 	return 0;
-- 
2.7.4

[PATCH V2 3/5] perf stat: Support per-die aggregation

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

It is useful to aggregate counts per die. E.g. Uncore becomes die-scope
on Xeon Cascade Lake-AP.

Introduce a new option "--per-die" to support per-die aggregation.

The global id for each core has been changed to socket + die id + core
id. The global id for each die is socket + die id.

Add die information for per-core aggregation. The output of per-core
aggregation will be changed from "S0-C0" to "S0-D0-C0". Any scripts
which rely on the output format of per-core aggregation probably be
broken.

For perf stat record/report, there is no die information when processing
the old perf.data. The per-die result will be the same as per-socket.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---

Changes since V1
- Fix the title shift for the case of "--per-die -a -I 1000"
- Add warn to check the size of socket id and die id

 tools/perf/Documentation/perf-stat.txt | 10 ++++
 tools/perf/builtin-stat.c              | 85 +++++++++++++++++++++++++++++++---
 tools/perf/util/cpumap.c               | 60 ++++++++++++++++++++++--
 tools/perf/util/cpumap.h               |  9 +++-
 tools/perf/util/stat-display.c         | 27 +++++++++--
 tools/perf/util/stat-shadow.c          |  1 +
 tools/perf/util/stat.c                 |  1 +
 tools/perf/util/stat.h                 |  1 +
 8 files changed, 179 insertions(+), 15 deletions(-)

diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 1e312c2..930c51c0 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -200,6 +200,13 @@ use --per-socket in addition to -a. (system-wide).  The output includes the
 socket number and the number of online processors on that socket. This is
 useful to gauge the amount of aggregation.
 
+--per-die::
+Aggregate counts per processor die for system-wide mode measurements.  This
+is a useful mode to detect imbalance between dies.  To enable this mode,
+use --per-die in addition to -a. (system-wide).  The output includes the
+die number and the number of online processors on that die. This is
+useful to gauge the amount of aggregation.
+
 --per-core::
 Aggregate counts per physical processor for system-wide mode measurements.  This
 is a useful mode to detect imbalance between physical cores.  To enable this mode,
@@ -239,6 +246,9 @@ Input file name.
 --per-socket::
 Aggregate counts per processor socket for system-wide mode measurements.
 
+--per-die::
+Aggregate counts per processor die for system-wide mode measurements.
+
 --per-core::
 Aggregate counts per physical processor for system-wide mode measurements.
 
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 24b8e69..e3aa0b1 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -777,6 +777,8 @@ static struct option stat_options[] = {
 		    "stop workload and print counts after a timeout period in ms (>= 10ms)"),
 	OPT_SET_UINT(0, "per-socket", &stat_config.aggr_mode,
 		     "aggregate counts per processor socket", AGGR_SOCKET),
+	OPT_SET_UINT(0, "per-die", &stat_config.aggr_mode,
+		     "aggregate counts per processor die", AGGR_DIE),
 	OPT_SET_UINT(0, "per-core", &stat_config.aggr_mode,
 		     "aggregate counts per physical processor core", AGGR_CORE),
 	OPT_SET_UINT(0, "per-thread", &stat_config.aggr_mode,
@@ -801,6 +803,12 @@ static int perf_stat__get_socket(struct perf_stat_config *config __maybe_unused,
 	return cpu_map__get_socket(map, cpu, NULL);
 }
 
+static int perf_stat__get_die(struct perf_stat_config *config __maybe_unused,
+			      struct cpu_map *map, int cpu)
+{
+	return cpu_map__get_die(map, cpu, NULL);
+}
+
 static int perf_stat__get_core(struct perf_stat_config *config __maybe_unused,
 			       struct cpu_map *map, int cpu)
 {
@@ -841,6 +849,12 @@ static int perf_stat__get_socket_cached(struct perf_stat_config *config,
 	return perf_stat__get_aggr(config, perf_stat__get_socket, map, idx);
 }
 
+static int perf_stat__get_die_cached(struct perf_stat_config *config,
+					struct cpu_map *map, int idx)
+{
+	return perf_stat__get_aggr(config, perf_stat__get_die, map, idx);
+}
+
 static int perf_stat__get_core_cached(struct perf_stat_config *config,
 				      struct cpu_map *map, int idx)
 {
@@ -871,6 +885,13 @@ static int perf_stat_init_aggr_mode(void)
 		}
 		stat_config.aggr_get_id = perf_stat__get_socket_cached;
 		break;
+	case AGGR_DIE:
+		if (cpu_map__build_die_map(evsel_list->cpus, &stat_config.aggr_map)) {
+			perror("cannot build die map");
+			return -1;
+		}
+		stat_config.aggr_get_id = perf_stat__get_die_cached;
+		break;
 	case AGGR_CORE:
 		if (cpu_map__build_core_map(evsel_list->cpus, &stat_config.aggr_map)) {
 			perror("cannot build core map");
@@ -936,21 +957,53 @@ static int perf_env__get_socket(struct cpu_map *map, int idx, void *data)
 	return cpu == -1 ? -1 : env->cpu[cpu].socket_id;
 }
 
+static int perf_env__get_die(struct cpu_map *map, int idx, void *data)
+{
+	struct perf_env *env = data;
+	int die = -1, cpu = perf_env__get_cpu(env, map, idx);
+
+	if (cpu != -1) {
+		/*
+		 * Encode socket in upper 8 bits
+		 * die_id is relative to socket,
+		 * we need a global id. So we combine
+		 * socket + die id
+		 */
+		if (WARN_ONCE(env->cpu[cpu].die_id >> 8,
+		    "The die_id number is too big. Please upgrade the perf tool.\n"))
+			return -1;
+
+		die = (env->cpu[cpu].socket_id << 8) |
+		      (env->cpu[cpu].die_id & 0xff);
+	}
+
+	return die;
+}
+
 static int perf_env__get_core(struct cpu_map *map, int idx, void *data)
 {
 	struct perf_env *env = data;
 	int core = -1, cpu = perf_env__get_cpu(env, map, idx);
 
 	if (cpu != -1) {
-		int socket_id = env->cpu[cpu].socket_id;
-
 		/*
-		 * Encode socket in upper 16 bits
-		 * core_id is relative to socket, and
+		 * Encode socket in upper 24 bits
+		 * encode die id in upper 16 bits
+		 * core_id is relative to socket and die,
 		 * we need a global id. So we combine
-		 * socket + core id.
+		 * socket + die id + core id
 		 */
-		core = (socket_id << 16) | (env->cpu[cpu].core_id & 0xffff);
+		if (WARN_ONCE(env->cpu[cpu].socket_id >> 8,
+		    "The socket_id number is too big. Please upgrade the perf tool.\n"))
+			return -1;
+
+		if (WARN_ONCE(env->cpu[cpu].die_id >> 8,
+		    "The die_id number is too big. Please upgrade the perf tool.\n"))
+			return -1;
+
+		core = (env->cpu[cpu].socket_id << 24) |
+		       (env->cpu[cpu].die_id << 16) |
+		       (env->cpu[cpu].core_id & 0xffff);
 	}
 
 	return core;
@@ -962,6 +1015,12 @@ static int perf_env__build_socket_map(struct perf_env *env, struct cpu_map *cpus
 	return cpu_map__build_map(cpus, sockp, perf_env__get_socket, env);
 }
 
+static int perf_env__build_die_map(struct perf_env *env, struct cpu_map *cpus,
+				   struct cpu_map **diep)
+{
+	return cpu_map__build_map(cpus, diep, perf_env__get_die, env);
+}
+
 static int perf_env__build_core_map(struct perf_env *env, struct cpu_map *cpus,
 				    struct cpu_map **corep)
 {
@@ -973,6 +1032,11 @@ static int perf_stat__get_socket_file(struct perf_stat_config *config __maybe_un
 {
 	return perf_env__get_socket(map, idx, &perf_stat.session->header.env);
 }
+static int perf_stat__get_die_file(struct perf_stat_config *config __maybe_unused,
+				   struct cpu_map *map, int idx)
+{
+	return perf_env__get_die(map, idx, &perf_stat.session->header.env);
+}
 
 static int perf_stat__get_core_file(struct perf_stat_config *config __maybe_unused,
 				    struct cpu_map *map, int idx)
@@ -992,6 +1056,13 @@ static int perf_stat_init_aggr_mode_file(struct perf_stat *st)
 		}
 		stat_config.aggr_get_id = perf_stat__get_socket_file;
 		break;
+	case AGGR_DIE:
+		if (perf_env__build_die_map(env, evsel_list->cpus, &stat_config.aggr_map)) {
+			perror("cannot build die map");
+			return -1;
+		}
+		stat_config.aggr_get_id = perf_stat__get_die_file;
+		break;
 	case AGGR_CORE:
 		if (perf_env__build_core_map(env, evsel_list->cpus, &stat_config.aggr_map)) {
 			perror("cannot build core map");
@@ -1542,6 +1613,8 @@ static int __cmd_report(int argc, const char **argv)
 	OPT_STRING('i', "input", &input_name, "file", "input file name"),
 	OPT_SET_UINT(0, "per-socket", &perf_stat.aggr_mode,
 		     "aggregate counts per processor socket", AGGR_SOCKET),
+	OPT_SET_UINT(0, "per-die", &perf_stat.aggr_mode,
+		     "aggregate counts per processor die", AGGR_DIE),
 	OPT_SET_UINT(0, "per-core", &perf_stat.aggr_mode,
 		     "aggregate counts per physical processor core", AGGR_CORE),
 	OPT_SET_UINT('A', "no-aggr", &perf_stat.aggr_mode,
diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c
index 7db1365..f70802f 100644
--- a/tools/perf/util/cpumap.c
+++ b/tools/perf/util/cpumap.c
@@ -380,6 +380,37 @@ int cpu_map__get_die_id(int cpu)
 	return ret ?: value;
 }
 
+int cpu_map__get_die(struct cpu_map *map, int idx, void *data)
+{
+	int cpu, die, s;
+
+	if (idx > map->nr)
+		return -1;
+
+	cpu = map->map[idx];
+
+	die = cpu_map__get_die_id(cpu);
+	/* There is no die_id on legacy system. */
+	if (die == -1)
+		die = 0;
+
+	s = cpu_map__get_socket(map, idx, data);
+	if (s == -1)
+		return -1;
+
+	/*
+	 * encode socket in upper 8 bits
+	 * die_id is relative to socket, and
+	 * we need a global id. So we combine
+	 * socket + die id
+	 */
+	if (WARN_ONCE(die >> 8,
+	    "The die_id number is too big. Please upgrade the perf tool.\n"))
+		return -1;
+
+	return (s << 8) | (die & 0xff);
+}
+
 int cpu_map__get_core_id(int cpu)
 {
 	int value, ret = cpu__get_topology_int(cpu, "core_id", &value);
@@ -388,7 +419,7 @@ int cpu_map__get_core_id(int cpu)
 
 int cpu_map__get_core(struct cpu_map *map, int idx, void *data)
 {
-	int cpu, s;
+	int cpu, s, die;
 
 	if (idx > map->nr)
 		return -1;
@@ -397,17 +428,31 @@ int cpu_map__get_core(struct cpu_map *map, int idx, void *data)
 
 	cpu = cpu_map__get_core_id(cpu);
 
+	die = cpu_map__get_die(map, idx, data);
+	/* There is no die_id on legacy system. */
+	if (die == -1)
+		die = 0;
+
 	s = cpu_map__get_socket(map, idx, data);
 	if (s == -1)
 		return -1;
 
 	/*
-	 * encode socket in upper 16 bits
-	 * core_id is relative to socket, and
+	 * encode socket in upper 24 bits
+	 * encode die id in upper 16 bits
+	 * core_id is relative to socket and die,
 	 * we need a global id. So we combine
-	 * socket+ core id
+	 * socket + die id + core id
 	 */
-	return (s << 16) | (cpu & 0xffff);
+	if (WARN_ONCE(s >> 8,
+	    "The socket_id number is too big. Please upgrade the perf tool.\n"))
+		return -1;
+
+	if (WARN_ONCE(die >> 8,
+	    "The die_id number is too big. Please upgrade the perf tool.\n"))
+		return -1;
+
+	return (s << 24) | (die << 16) | (cpu & 0xffff);
 }
 
 int cpu_map__build_socket_map(struct cpu_map *cpus, struct cpu_map **sockp)
@@ -415,6 +460,11 @@ int cpu_map__build_socket_map(struct cpu_map *cpus, struct cpu_map **sockp)
 	return cpu_map__build_map(cpus, sockp, cpu_map__get_socket, NULL);
 }
 
+int cpu_map__build_die_map(struct cpu_map *cpus, struct cpu_map **diep)
+{
+	return cpu_map__build_map(cpus, diep, cpu_map__get_die, NULL);
+}
+
 int cpu_map__build_core_map(struct cpu_map *cpus, struct cpu_map **corep)
 {
 	return cpu_map__build_map(cpus, corep, cpu_map__get_core, NULL);
diff --git a/tools/perf/util/cpumap.h b/tools/perf/util/cpumap.h
index 6762ff9..1265f0e 100644
--- a/tools/perf/util/cpumap.h
+++ b/tools/perf/util/cpumap.h
@@ -26,9 +26,11 @@ size_t cpu_map__fprintf(struct cpu_map *map, FILE *fp);
 int cpu_map__get_socket_id(int cpu);
 int cpu_map__get_socket(struct cpu_map *map, int idx, void *data);
 int cpu_map__get_die_id(int cpu);
+int cpu_map__get_die(struct cpu_map *map, int idx, void *data);
 int cpu_map__get_core_id(int cpu);
 int cpu_map__get_core(struct cpu_map *map, int idx, void *data);
 int cpu_map__build_socket_map(struct cpu_map *cpus, struct cpu_map **sockp);
+int cpu_map__build_die_map(struct cpu_map *cpus, struct cpu_map **diep);
 int cpu_map__build_core_map(struct cpu_map *cpus, struct cpu_map **corep);
 const struct cpu_map *cpu_map__online(void); /* thread unsafe */
 
@@ -44,7 +46,12 @@ static inline int cpu_map__socket(struct cpu_map *sock, int s)
 
 static inline int cpu_map__id_to_socket(int id)
 {
-	return id >> 16;
+	return id >> 24;
+}
+
+static inline int cpu_map__id_to_die(int id)
+{
+	return (id >> 16) & 0xff;
 }
 
 static inline int cpu_map__id_to_cpu(int id)
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index 4c53bae..f67ac2c 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -69,8 +69,9 @@ static void aggr_printout(struct perf_stat_config *config,
 {
 	switch (config->aggr_mode) {
 	case AGGR_CORE:
-		fprintf(config->output, "S%d-C%*d%s%*d%s",
+		fprintf(config->output, "S%d-D%d-C%*d%s%*d%s",
 			cpu_map__id_to_socket(id),
+			cpu_map__id_to_die(id),
 			config->csv_output ? 0 : -8,
 			cpu_map__id_to_cpu(id),
 			config->csv_sep,
@@ -78,6 +79,16 @@ static void aggr_printout(struct perf_stat_config *config,
 			nr,
 			config->csv_sep);
 		break;
+	case AGGR_DIE:
+		fprintf(config->output, "S%d-D%*d%s%*d%s",
+			cpu_map__id_to_socket(id << 16),
+			config->csv_output ? 0 : -8,
+			cpu_map__id_to_die(id << 16),
+			config->csv_sep,
+			config->csv_output ? 0 : 4,
+			nr,
+			config->csv_sep);
+		break;
 	case AGGR_SOCKET:
 		fprintf(config->output, "S%*d%s%*d%s",
 			config->csv_output ? 0 : -5,
@@ -89,8 +100,9 @@ static void aggr_printout(struct perf_stat_config *config,
 			break;
 	case AGGR_NONE:
 		if (evsel->percore) {
-			fprintf(config->output, "S%d-C%*d%s",
+			fprintf(config->output, "S%d-D%d-C%*d%s",
 				cpu_map__id_to_socket(id),
+				cpu_map__id_to_die(id),
 				config->csv_output ? 0 : -5,
 				cpu_map__id_to_cpu(id), config->csv_sep);
 		} else {
@@ -407,6 +419,7 @@ static void printout(struct perf_stat_config *config, int id, int nr,
 			[AGGR_THREAD] = 1,
 			[AGGR_NONE] = 1,
 			[AGGR_SOCKET] = 2,
+			[AGGR_DIE] = 2,
 			[AGGR_CORE] = 2,
 		};
 
@@ -879,7 +892,8 @@ static void print_no_aggr_metric(struct perf_stat_config *config,
 }
 
 static int aggr_header_lens[] = {
-	[AGGR_CORE] = 18,
+	[AGGR_CORE] = 24,
+	[AGGR_DIE] = 18,
 	[AGGR_SOCKET] = 12,
 	[AGGR_NONE] = 6,
 	[AGGR_THREAD] = 24,
@@ -888,6 +902,7 @@ static int aggr_header_lens[] = {
 
 static const char *aggr_header_csv[] = {
 	[AGGR_CORE] 	= 	"core,cpus,",
+	[AGGR_DIE] 	= 	"die,cpus",
 	[AGGR_SOCKET] 	= 	"socket,cpus",
 	[AGGR_NONE] 	= 	"cpu,",
 	[AGGR_THREAD] 	= 	"comm-pid,",
@@ -954,6 +969,11 @@ static void print_interval(struct perf_stat_config *config,
 			if (!metric_only)
 				fprintf(output, "             counts %*s events\n", unit_width, "unit");
 			break;
+		case AGGR_DIE:
+			fprintf(output, "#           time die          cpus");
+			if (!metric_only)
+				fprintf(output, "             counts %*s events\n", unit_width, "unit");
+			break;
 		case AGGR_CORE:
 			fprintf(output, "#           time core         cpus");
 			if (!metric_only)
@@ -1165,6 +1185,7 @@ perf_evlist__print_counters(struct perf_evlist *evlist,
 
 	switch (config->aggr_mode) {
 	case AGGR_CORE:
+	case AGGR_DIE:
 	case AGGR_SOCKET:
 		print_aggr(config, evlist, prefix);
 		break;
diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
index 83d8094..027b09a 100644
--- a/tools/perf/util/stat-shadow.c
+++ b/tools/perf/util/stat-shadow.c
@@ -12,6 +12,7 @@
 /*
  * AGGR_GLOBAL: Use CPU 0
  * AGGR_SOCKET: Use first CPU of socket
+ * AGGR_DIE: Use first CPU of die
  * AGGR_CORE: Use first CPU of core
  * AGGR_NONE: Use matching CPU
  * AGGR_THREAD: Not supported?
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index c3115d9..d91fe75 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -272,6 +272,7 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
 	switch (config->aggr_mode) {
 	case AGGR_THREAD:
 	case AGGR_CORE:
+	case AGGR_DIE:
 	case AGGR_SOCKET:
 	case AGGR_NONE:
 		if (!evsel->snapshot)
diff --git a/tools/perf/util/stat.h b/tools/perf/util/stat.h
index 2f9c915..7032dd1 100644
--- a/tools/perf/util/stat.h
+++ b/tools/perf/util/stat.h
@@ -44,6 +44,7 @@ enum aggr_mode {
 	AGGR_NONE,
 	AGGR_GLOBAL,
 	AGGR_SOCKET,
+	AGGR_DIE,
 	AGGR_CORE,
 	AGGR_THREAD,
 	AGGR_UNSET,
-- 
2.7.4

[PATCH V2 4/5] perf header: Rename "sibling cores" to "sibling sockets"

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

The "sibling cores" actually shows the sibling CPUs of a socket.
The name "sibling cores" is very misleading.

Rename "sibling cores" to "sibling sockets"

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---

No changes since V1

 tools/perf/Documentation/perf.data-file-format.txt | 2 +-
 tools/perf/util/header.c                           | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/Documentation/perf.data-file-format.txt b/tools/perf/Documentation/perf.data-file-format.txt
index c731416..dd85163 100644
--- a/tools/perf/Documentation/perf.data-file-format.txt
+++ b/tools/perf/Documentation/perf.data-file-format.txt
@@ -168,7 +168,7 @@ struct {
 };
 
 Example:
-	sibling cores   : 0-8
+	sibling sockets : 0-8
 	sibling dies	: 0-3
 	sibling dies	: 4-7
 	sibling threads : 0-1
diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 6497625..06ddb66 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -1460,7 +1460,7 @@ static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
 	str = ph->env.sibling_cores;
 
 	for (i = 0; i < nr; i++) {
-		fprintf(fp, "# sibling cores   : %s\n", str);
+		fprintf(fp, "# sibling sockets : %s\n", str);
 		str += strlen(str) + 1;
 	}
 
-- 
2.7.4

[PATCH V2 5/5] perf tools: Apply new CPU topology sysfs attributes

[kan.liang]

From: Kan Liang <kan.liang@linux.intel.com>

The existing "thread_siblings" and "thread_siblings_list" attribute will
be deprecated.
Use the new CPU topology sysfs attributes, "core_cpus" and
"core_cpus_list", which are synonymous with the deprecated attributes.

Check the new name first. If not available, use the deprecated name to
be compatible with old kernel.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
---

New for V2

 tools/perf/util/cputopo.c | 8 +++++++-
 tools/perf/util/smt.c     | 8 ++++++--
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/tools/perf/util/cputopo.c b/tools/perf/util/cputopo.c
index 85fa87f..26e73a4b 100644
--- a/tools/perf/util/cputopo.c
+++ b/tools/perf/util/cputopo.c
@@ -15,6 +15,8 @@
 	"%s/devices/system/cpu/cpu%d/topology/die_cpus_list"
 #define THRD_SIB_FMT \
 	"%s/devices/system/cpu/cpu%d/topology/thread_siblings_list"
+#define THRD_SIB_FMT_NEW \
+	"%s/devices/system/cpu/cpu%d/topology/core_cpus_list"
 #define NODE_ONLINE_FMT \
 	"%s/devices/system/node/online"
 #define NODE_MEMINFO_FMT \
@@ -91,8 +93,12 @@ static int build_cpu_topology(struct cpu_topology *tp, int cpu)
 	ret = 0;
 
 try_threads:
-	scnprintf(filename, MAXPATHLEN, THRD_SIB_FMT,
+	scnprintf(filename, MAXPATHLEN, THRD_SIB_FMT_NEW,
 		  sysfs__mountpoint(), cpu);
+	if (access(filename, F_OK) == -1) {
+		scnprintf(filename, MAXPATHLEN, THRD_SIB_FMT,
+			  sysfs__mountpoint(), cpu);
+	}
 	fp = fopen(filename, "r");
 	if (!fp)
 		goto done;
diff --git a/tools/perf/util/smt.c b/tools/perf/util/smt.c
index 453f6f6..3b791ef 100644
--- a/tools/perf/util/smt.c
+++ b/tools/perf/util/smt.c
@@ -23,8 +23,12 @@ int smt_on(void)
 		char fn[256];
 
 		snprintf(fn, sizeof fn,
-			"devices/system/cpu/cpu%d/topology/thread_siblings",
-			cpu);
+			"devices/system/cpu/cpu%d/topology/core_cpus", cpu);
+		if (access(fn, F_OK) == -1) {
+			snprintf(fn, sizeof fn,
+				"devices/system/cpu/cpu%d/topology/thread_siblings",
+				cpu);
+		}
 		if (sysfs__read_str(fn, &str, &strlen) < 0)
 			continue;
 		/* Entry is hex, but does not have 0x, so need custom parser */
-- 
2.7.4

Re: [PATCH V2 3/5] perf stat: Support per-die aggregation

[Jiri Olsa]

On Thu, May 30, 2019 at 07:53:47AM -0700, kan.liang@linux.intel.com wrote:

SNIP

> +
>  static int perf_env__get_core(struct cpu_map *map, int idx, void *data)
>  {
>  	struct perf_env *env = data;
>  	int core = -1, cpu = perf_env__get_cpu(env, map, idx);
>  
>  	if (cpu != -1) {
> -		int socket_id = env->cpu[cpu].socket_id;
> -
>  		/*
> -		 * Encode socket in upper 16 bits
> -		 * core_id is relative to socket, and
> +		 * Encode socket in upper 24 bits

please note we use upper 8 bits for socket number,
the comments suggests it's 24 bits

> +		 * encode die id in upper 16 bits
> +		 * core_id is relative to socket and die,
>  		 * we need a global id. So we combine
> -		 * socket + core id.
> +		 * socket + die id + core id
>  		 */
> -		core = (socket_id << 16) | (env->cpu[cpu].core_id & 0xffff);
> +		if (WARN_ONCE(env->cpu[cpu].socket_id >> 8,
> +		    "The socket_id number is too big. Please upgrade the perf tool.\n"))

hum, how's perf tool upgrade going to help in here?

> +			return -1;
> +
> +		if (WARN_ONCE(env->cpu[cpu].die_id >> 8,
> +		    "The die_id number is too big. Please upgrade the perf tool.\n"))
> +			return -1;
> +
> +		core = (env->cpu[cpu].socket_id << 24) |
> +		       (env->cpu[cpu].die_id << 16) |
> +		       (env->cpu[cpu].core_id & 0xffff);
>  	}

other than comments above, the patchset looks good to me

thanks,
jirka

Re: [PATCH V2 3/5] perf stat: Support per-die aggregation

[Liang, Kan]

On 6/3/2019 12:36 PM, Jiri Olsa wrote:
> On Thu, May 30, 2019 at 07:53:47AM -0700, kan.liang@linux.intel.com wrote:
> 
> SNIP
> 
>> +
>>   static int perf_env__get_core(struct cpu_map *map, int idx, void *data)
>>   {
>>   	struct perf_env *env = data;
>>   	int core = -1, cpu = perf_env__get_cpu(env, map, idx);
>>   
>>   	if (cpu != -1) {
>> -		int socket_id = env->cpu[cpu].socket_id;
>> -
>>   		/*
>> -		 * Encode socket in upper 16 bits
>> -		 * core_id is relative to socket, and
>> +		 * Encode socket in upper 24 bits
> 
> please note we use upper 8 bits for socket number,
> the comments suggests it's 24 bits

I will fix it in v3.

> 
>> +		 * encode die id in upper 16 bits
>> +		 * core_id is relative to socket and die,
>>   		 * we need a global id. So we combine
>> -		 * socket + core id.
>> +		 * socket + die id + core id
>>   		 */
>> -		core = (socket_id << 16) | (env->cpu[cpu].core_id & 0xffff);
>> +		if (WARN_ONCE(env->cpu[cpu].socket_id >> 8,
>> +		    "The socket_id number is too big. Please upgrade the perf tool.\n"))
> 
> hum, how's perf tool upgrade going to help in here?


I think there is nothing we can do for current encoding, if the socket 
number or die number is bigger than 8 bits. We have to design a new 
encoding method, which needs to update the perf tool.
So I use a similar expression from process_cpu_topology().
		if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
			pr_debug("socket_id number is too big."
				 "You may need to upgrade the perf tool.\n");
			goto free_cpu;
		}

Any suggestions for the warning message?



> 
>> +			return -1;
>> +
>> +		if (WARN_ONCE(env->cpu[cpu].die_id >> 8,
>> +		    "The die_id number is too big. Please upgrade the perf tool.\n"))
>> +			return -1;
>> +
>> +		core = (env->cpu[cpu].socket_id << 24) |
>> +		       (env->cpu[cpu].die_id << 16) |
>> +		       (env->cpu[cpu].core_id & 0xffff);
>>   	}
> 
> other than comments above, the patchset looks good to me
>

Thanks for the review. I will send out V3 after the comments are addressed.

Thanks,
Kan

Re: [PATCH V2 3/5] perf stat: Support per-die aggregation

[Jiri Olsa]

On Mon, Jun 03, 2019 at 01:42:58PM -0400, Liang, Kan wrote:
> 
> 
> On 6/3/2019 12:36 PM, Jiri Olsa wrote:
> > On Thu, May 30, 2019 at 07:53:47AM -0700, kan.liang@linux.intel.com wrote:
> > 
> > SNIP
> > 
> > > +
> > >   static int perf_env__get_core(struct cpu_map *map, int idx, void *data)
> > >   {
> > >   	struct perf_env *env = data;
> > >   	int core = -1, cpu = perf_env__get_cpu(env, map, idx);
> > >   	if (cpu != -1) {
> > > -		int socket_id = env->cpu[cpu].socket_id;
> > > -
> > >   		/*
> > > -		 * Encode socket in upper 16 bits
> > > -		 * core_id is relative to socket, and
> > > +		 * Encode socket in upper 24 bits
> > 
> > please note we use upper 8 bits for socket number,
> > the comments suggests it's 24 bits
> 
> I will fix it in v3.
> 
> > 
> > > +		 * encode die id in upper 16 bits
> > > +		 * core_id is relative to socket and die,
> > >   		 * we need a global id. So we combine
> > > -		 * socket + core id.
> > > +		 * socket + die id + core id
> > >   		 */
> > > -		core = (socket_id << 16) | (env->cpu[cpu].core_id & 0xffff);
> > > +		if (WARN_ONCE(env->cpu[cpu].socket_id >> 8,
> > > +		    "The socket_id number is too big. Please upgrade the perf tool.\n"))
> > 
> > hum, how's perf tool upgrade going to help in here?
> 
> 
> I think there is nothing we can do for current encoding, if the socket
> number or die number is bigger than 8 bits. We have to design a new encoding
> method, which needs to update the perf tool.
> So I use a similar expression from process_cpu_topology().
> 		if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
> 			pr_debug("socket_id number is too big."
> 				 "You may need to upgrade the perf tool.\n");
> 			goto free_cpu;
> 		}
> 
> Any suggestions for the warning message?

not really :-\ perhaps the original is more accurate
and does not "claim" the upgrade will help.. it's just
a nit.. feel free to keep it

thanks,
jirka