Skip to contentSkip to conversation

perf tools: Handle sparse CPU numbers correctly

PATCH In review Paul Mackerras 1 participant 2 messages Mar 10, 2010 – Mar 11, 2010 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf tools: Handle sparse CPU numbers correctly

At present, the perf subcommands that do system-wide monitoring (perf stat, perf record and perf top) don't work properly unless the online cpus are numbered 0, 1, ..., N-1. These tools ask for the number of online cpus with sysconf(_SC_NPROCESSORS_ONLN) and then try to create events for cpus 0, 1, ..., N-1.

This creates problems for systems where the online cpus are numbered sparsely. For example, a POWER6 system in single-threaded mode (i.e. only running 1 hardware thread per core) will have only even-numbered cpus online.

This fixes the problem by reading the /sys/devices/system/cpu/online file to find out which cpus are online. The code that does that is in tools/perf/util/cpumap.[ch], and consists of a read_cpu_map() function that sets up a cpumap[] array and returns the number of online cpus. If /sys/devices/system/cpu/online can't be read or can't be parsed successfully, it falls back to using sysconf to ask how many cpus are online and sets up an identity map in cpumap[].

The perf record, perf stat and perf top code then calls read_cpu_map() in the system-wide monitoring case (instead of sysconf) and uses cpumap[] to get the cpu numbers to pass to perf_event_open.

---
I think this approach is a bit better than my previous approach of just keeping on incrementing the cpu number.

 tools/perf/Makefile         |    2 +
 tools/perf/builtin-record.c |    7 ++---
 tools/perf/builtin-stat.c   |   10 ++++---
 tools/perf/builtin-top.c    |    9 +++---
 tools/perf/util/cpumap.c    |   59 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/cpumap.h    |    7 +++++
 6 files changed, 81 insertions(+), 13 deletions(-)
 create mode 100644 tools/perf/util/cpumap.c
 create mode 100644 tools/perf/util/cpumap.h
tools/perf/Makefile +2 -0
index 54a5b50..aeb9aac 100644
@@ -387,6 +387,7 @@ LIB_H += util/thread.h
 LIB_H += util/trace-event.h
 LIB_H += util/probe-finder.h
 LIB_H += util/probe-event.h
+LIB_H += util/cpumap.h
 
 LIB_OBJS += util/abspath.o
 LIB_OBJS += util/alias.o
@@ -433,6 +434,7 @@ LIB_OBJS += util/sort.o
 LIB_OBJS += util/hist.o
 LIB_OBJS += util/probe-event.o
 LIB_OBJS += util/util.o
+LIB_OBJS += util/cpumap.o
 
 BUILTIN_OBJS += builtin-annotate.o
 
tools/perf/builtin-record.c +3 -4
index 771533c..cb28224 100644
@@ -22,6 +22,7 @@
 #include "util/debug.h"
 #include "util/session.h"
 #include "util/symbol.h"
+#include "util/cpumap.h"
 
 #include <unistd.h>
 #include <sched.h>
@@ -418,9 +419,6 @@ static int __cmd_record(int argc, const char **argv)
 	char buf;
 
 	page_size = sysconf(_SC_PAGE_SIZE);
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert(nr_cpus >= 0);
 
 	atexit(sig_atexit);
 	signal(SIGCHLD, sig_handler);
@@ -544,8 +542,9 @@ static int __cmd_record(int argc, const char **argv)
 	if ((!system_wide && !inherit) || profile_cpu != -1) {
 		open_counters(profile_cpu, target_pid);
 	} else {
+		nr_cpus = read_cpu_map();
 		for (i = 0; i < nr_cpus; i++)
-			open_counters(i, target_pid);
+			open_counters(cpumap[i], target_pid);
 	}
 
 	if (file_new) {
tools/perf/builtin-stat.c +6 -4
index e8c85d5..95db31c 100644
@@ -45,6 +45,7 @@
 #include "util/event.h"
 #include "util/debug.h"
 #include "util/header.h"
+#include "util/cpumap.h"
 
 #include <sys/prctl.h>
 #include <math.h>
@@ -151,7 +152,7 @@ static void create_perf_stat_counter(int counter, int pid)
 		unsigned int cpu;
 
 		for (cpu = 0; cpu < nr_cpus; cpu++) {
-			fd[cpu][counter] = sys_perf_event_open(attr, -1, cpu, -1, 0);
+			fd[cpu][counter] = sys_perf_event_open(attr, -1, cpumap[cpu], -1, 0);
 			if (fd[cpu][counter] < 0 && verbose)
 				fprintf(stderr, ERR_PERF_OPEN, counter,
 					fd[cpu][counter], strerror(errno));
@@ -519,9 +520,10 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 		nr_counters = ARRAY_SIZE(default_attrs);
 	}
 
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert((int)nr_cpus >= 0);
+	if (system_wide)
+		nr_cpus = read_cpu_map();
+	else
+		nr_cpus = 1;
 
 	/*
 	 * We dont want to block the signals - that would cause
tools/perf/builtin-top.c +4 -5
index 31f2e59..0b719e3 100644
@@ -28,6 +28,7 @@
 #include <linux/rbtree.h>
 #include "util/parse-options.h"
 #include "util/parse-events.h"
+#include "util/cpumap.h"
 
 #include "util/debug.h"
 
@@ -1123,7 +1124,7 @@ static void start_counter(int i, int counter)
 
 	cpu = profile_cpu;
 	if (target_pid == -1 && profile_cpu == -1)
-		cpu = i;
+		cpu = cpumap[i];
 
 	attr = attrs + counter;
 
@@ -1347,12 +1348,10 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 		attrs[counter].sample_period = default_interval;
 	}
 
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert(nr_cpus >= 0);
-
 	if (target_pid != -1 || profile_cpu != -1)
 		nr_cpus = 1;
+	else
+		nr_cpus = read_cpu_map();
 
 	get_term_dimensions(&winsize);
 	if (print_entries == 0) {
tools/perf/util/cpumap.c new file +59 -0

59 lines not on this page · raw message

tools/perf/util/cpumap.h new file +7 -0
new file mode 100644
index 0000000..86c78bb
@@ -0,0 +1,7 @@
+#ifndef __PERF_CPUMAP_H
+#define __PERF_CPUMAP_H
+
+extern int read_cpu_map(void);
+extern int cpumap[];
+
+#endif /* __PERF_CPUMAP_H */
Signature · 1 line
1.6.5.3.171.ge36e
#1 lore raw

[tip:perf/urgent] perf tools: Fix sparse CPU numbering related bugs

Commit-ID:  a12b51c478899fe0b7e874a559b05ba35f1128ee
Gitweb:     http://git.kernel.org/tip/a12b51c478899fe0b7e874a559b05ba35f1128ee
Author:     Paul Mackerras <[email protected]>
AuthorDate: Wed, 10 Mar 2010 20:36:09 +1100
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 11 Mar 2010 13:36:53 +0100

perf tools: Fix sparse CPU numbering related bugs

At present, the perf subcommands that do system-wide monitoring (perf stat, perf record and perf top) don't work properly unless the online cpus are numbered 0, 1, ..., N-1. These tools ask for the number of online cpus with sysconf(_SC_NPROCESSORS_ONLN) and then try to create events for cpus 0, 1, ..., N-1.

This creates problems for systems where the online cpus are numbered sparsely. For example, a POWER6 system in single-threaded mode (i.e. only running 1 hardware thread per core) will have only even-numbered cpus online.

This fixes the problem by reading the /sys/devices/system/cpu/online file to find out which cpus are online. The code that does that is in tools/perf/util/cpumap.[ch], and consists of a read_cpu_map() function that sets up a cpumap[] array and returns the number of online cpus. If /sys/devices/system/cpu/online can't be read or can't be parsed successfully, it falls back to using sysconf to ask how many cpus are online and sets up an identity map in cpumap[].

The perf record, perf stat and perf top code then calls read_cpu_map() in the system-wide monitoring case (instead of sysconf) and uses cpumap[] to get the cpu numbers to pass to perf_event_open.

Cc Anton Blanchard <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>

LKML-Reference: <[email protected]>

---
 tools/perf/Makefile         |    2 +
 tools/perf/builtin-record.c |    7 ++---
 tools/perf/builtin-stat.c   |   10 ++++---
 tools/perf/builtin-top.c    |    9 +++---
 tools/perf/util/cpumap.c    |   59 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/cpumap.h    |    7 +++++
 6 files changed, 81 insertions(+), 13 deletions(-)
tools/perf/Makefile +2 -0
index 2d53738..5840499 100644
@@ -387,6 +387,7 @@ LIB_H += util/thread.h
 LIB_H += util/trace-event.h
 LIB_H += util/probe-finder.h
 LIB_H += util/probe-event.h
+LIB_H += util/cpumap.h
 
 LIB_OBJS += util/abspath.o
 LIB_OBJS += util/alias.o
@@ -433,6 +434,7 @@ LIB_OBJS += util/sort.o
 LIB_OBJS += util/hist.o
 LIB_OBJS += util/probe-event.o
 LIB_OBJS += util/util.o
+LIB_OBJS += util/cpumap.o
 
 BUILTIN_OBJS += builtin-annotate.o
 
tools/perf/builtin-record.c +3 -4
index f573bbb..b09d3b2 100644
@@ -22,6 +22,7 @@
 #include "util/debug.h"
 #include "util/session.h"
 #include "util/symbol.h"
+#include "util/cpumap.h"
 
 #include <unistd.h>
 #include <sched.h>
@@ -421,9 +422,6 @@ static int __cmd_record(int argc, const char **argv)
 	char buf;
 
 	page_size = sysconf(_SC_PAGE_SIZE);
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert(nr_cpus >= 0);
 
 	atexit(sig_atexit);
 	signal(SIGCHLD, sig_handler);
@@ -547,8 +545,9 @@ static int __cmd_record(int argc, const char **argv)
 	if ((!system_wide && !inherit) || profile_cpu != -1) {
 		open_counters(profile_cpu, target_pid);
 	} else {
+		nr_cpus = read_cpu_map();
 		for (i = 0; i < nr_cpus; i++)
-			open_counters(i, target_pid);
+			open_counters(cpumap[i], target_pid);
 	}
 
 	if (file_new) {
tools/perf/builtin-stat.c +6 -4
index e8c85d5..95db31c 100644
@@ -45,6 +45,7 @@
 #include "util/event.h"
 #include "util/debug.h"
 #include "util/header.h"
+#include "util/cpumap.h"
 
 #include <sys/prctl.h>
 #include <math.h>
@@ -151,7 +152,7 @@ static void create_perf_stat_counter(int counter, int pid)
 		unsigned int cpu;
 
 		for (cpu = 0; cpu < nr_cpus; cpu++) {
-			fd[cpu][counter] = sys_perf_event_open(attr, -1, cpu, -1, 0);
+			fd[cpu][counter] = sys_perf_event_open(attr, -1, cpumap[cpu], -1, 0);
 			if (fd[cpu][counter] < 0 && verbose)
 				fprintf(stderr, ERR_PERF_OPEN, counter,
 					fd[cpu][counter], strerror(errno));
@@ -519,9 +520,10 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 		nr_counters = ARRAY_SIZE(default_attrs);
 	}
 
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert((int)nr_cpus >= 0);
+	if (system_wide)
+		nr_cpus = read_cpu_map();
+	else
+		nr_cpus = 1;
 
 	/*
 	 * We dont want to block the signals - that would cause
tools/perf/builtin-top.c +4 -5
index 31f2e59..0b719e3 100644
@@ -28,6 +28,7 @@
 #include <linux/rbtree.h>
 #include "util/parse-options.h"
 #include "util/parse-events.h"
+#include "util/cpumap.h"
 
 #include "util/debug.h"
 
@@ -1123,7 +1124,7 @@ static void start_counter(int i, int counter)
 
 	cpu = profile_cpu;
 	if (target_pid == -1 && profile_cpu == -1)
-		cpu = i;
+		cpu = cpumap[i];
 
 	attr = attrs + counter;
 
@@ -1347,12 +1348,10 @@ int cmd_top(int argc, const char **argv, const char *prefix __used)
 		attrs[counter].sample_period = default_interval;
 	}
 
-	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	assert(nr_cpus <= MAX_NR_CPUS);
-	assert(nr_cpus >= 0);
-
 	if (target_pid != -1 || profile_cpu != -1)
 		nr_cpus = 1;
+	else
+		nr_cpus = read_cpu_map();
 
 	get_term_dimensions(&winsize);
 	if (print_entries == 0) {
tools/perf/util/cpumap.c new file +59 -0

59 lines not on this page · raw message

tools/perf/util/cpumap.h new file +7 -0
new file mode 100644
index 0000000..86c78bb
@@ -0,0 +1,7 @@
+#ifndef __PERF_CPUMAP_H
+#define __PERF_CPUMAP_H
+
+extern int read_cpu_map(void);
+extern int cpumap[];
+
+#endif /* __PERF_CPUMAP_H */