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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
#include "util.h"
|
2013-12-09 23:14:24 +07:00
|
|
|
#include <api/fs/fs.h>
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
#include "../perf.h"
|
|
|
|
#include "cpumap.h"
|
|
|
|
#include <assert.h>
|
|
|
|
#include <stdio.h>
|
2013-02-14 19:57:27 +07:00
|
|
|
#include <stdlib.h>
|
2015-10-25 21:51:25 +07:00
|
|
|
#include <linux/bitmap.h>
|
2015-06-23 05:36:04 +07:00
|
|
|
#include "asm/bug.h"
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
|
2016-01-27 01:51:46 +07:00
|
|
|
static int max_cpu_num;
|
2017-02-17 18:10:24 +07:00
|
|
|
static int max_present_cpu_num;
|
2016-01-27 01:51:46 +07:00
|
|
|
static int max_node_num;
|
|
|
|
static int *cpunode_map;
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
static struct cpu_map *cpu_map__default_new(void)
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
{
|
2011-01-04 02:49:48 +07:00
|
|
|
struct cpu_map *cpus;
|
|
|
|
int nr_cpus;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
|
|
|
|
nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
|
2011-01-04 02:49:48 +07:00
|
|
|
if (nr_cpus < 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
cpus = malloc(sizeof(*cpus) + nr_cpus * sizeof(int));
|
|
|
|
if (cpus != NULL) {
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < nr_cpus; ++i)
|
|
|
|
cpus->map[i] = i;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
cpus->nr = nr_cpus;
|
2015-06-23 05:36:04 +07:00
|
|
|
atomic_set(&cpus->refcnt, 1);
|
2011-01-04 02:49:48 +07:00
|
|
|
}
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
return cpus;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
}
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
static struct cpu_map *cpu_map__trim_new(int nr_cpus, int *tmp_cpus)
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
{
|
2011-01-04 02:49:48 +07:00
|
|
|
size_t payload_size = nr_cpus * sizeof(int);
|
|
|
|
struct cpu_map *cpus = malloc(sizeof(*cpus) + payload_size);
|
|
|
|
|
|
|
|
if (cpus != NULL) {
|
|
|
|
cpus->nr = nr_cpus;
|
|
|
|
memcpy(cpus->map, tmp_cpus, payload_size);
|
2015-06-23 05:36:04 +07:00
|
|
|
atomic_set(&cpus->refcnt, 1);
|
2011-01-04 02:49:48 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return cpus;
|
|
|
|
}
|
|
|
|
|
2012-09-10 14:53:50 +07:00
|
|
|
struct cpu_map *cpu_map__read(FILE *file)
|
2011-01-04 02:49:48 +07:00
|
|
|
{
|
|
|
|
struct cpu_map *cpus = NULL;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
int nr_cpus = 0;
|
2011-01-04 02:49:48 +07:00
|
|
|
int *tmp_cpus = NULL, *tmp;
|
|
|
|
int max_entries = 0;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
int n, cpu, prev;
|
|
|
|
char sep;
|
|
|
|
|
|
|
|
sep = 0;
|
|
|
|
prev = -1;
|
|
|
|
for (;;) {
|
2012-09-10 14:53:50 +07:00
|
|
|
n = fscanf(file, "%u%c", &cpu, &sep);
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
if (n <= 0)
|
|
|
|
break;
|
|
|
|
if (prev >= 0) {
|
2011-01-04 02:49:48 +07:00
|
|
|
int new_max = nr_cpus + cpu - prev - 1;
|
|
|
|
|
|
|
|
if (new_max >= max_entries) {
|
|
|
|
max_entries = new_max + MAX_NR_CPUS / 2;
|
|
|
|
tmp = realloc(tmp_cpus, max_entries * sizeof(int));
|
|
|
|
if (tmp == NULL)
|
|
|
|
goto out_free_tmp;
|
|
|
|
tmp_cpus = tmp;
|
|
|
|
}
|
|
|
|
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
while (++prev < cpu)
|
2011-01-04 02:49:48 +07:00
|
|
|
tmp_cpus[nr_cpus++] = prev;
|
|
|
|
}
|
|
|
|
if (nr_cpus == max_entries) {
|
|
|
|
max_entries += MAX_NR_CPUS;
|
|
|
|
tmp = realloc(tmp_cpus, max_entries * sizeof(int));
|
|
|
|
if (tmp == NULL)
|
|
|
|
goto out_free_tmp;
|
|
|
|
tmp_cpus = tmp;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
}
|
2011-01-04 02:49:48 +07:00
|
|
|
|
|
|
|
tmp_cpus[nr_cpus++] = cpu;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
if (n == 2 && sep == '-')
|
|
|
|
prev = cpu;
|
|
|
|
else
|
|
|
|
prev = -1;
|
|
|
|
if (n == 1 || sep == '\n')
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
if (nr_cpus > 0)
|
|
|
|
cpus = cpu_map__trim_new(nr_cpus, tmp_cpus);
|
|
|
|
else
|
|
|
|
cpus = cpu_map__default_new();
|
|
|
|
out_free_tmp:
|
|
|
|
free(tmp_cpus);
|
2012-09-10 14:53:50 +07:00
|
|
|
return cpus;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct cpu_map *cpu_map__read_all_cpu_map(void)
|
|
|
|
{
|
|
|
|
struct cpu_map *cpus = NULL;
|
|
|
|
FILE *onlnf;
|
|
|
|
|
|
|
|
onlnf = fopen("/sys/devices/system/cpu/online", "r");
|
|
|
|
if (!onlnf)
|
|
|
|
return cpu_map__default_new();
|
|
|
|
|
|
|
|
cpus = cpu_map__read(onlnf);
|
2011-01-04 02:49:48 +07:00
|
|
|
fclose(onlnf);
|
|
|
|
return cpus;
|
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.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
LKML-Reference: <20100310093609.GA3959@brick.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-10 16:36:09 +07:00
|
|
|
}
|
2010-05-28 17:00:01 +07:00
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
struct cpu_map *cpu_map__new(const char *cpu_list)
|
2010-05-28 17:00:01 +07:00
|
|
|
{
|
2011-01-04 02:49:48 +07:00
|
|
|
struct cpu_map *cpus = NULL;
|
2010-05-28 17:00:01 +07:00
|
|
|
unsigned long start_cpu, end_cpu = 0;
|
|
|
|
char *p = NULL;
|
|
|
|
int i, nr_cpus = 0;
|
2011-01-04 02:49:48 +07:00
|
|
|
int *tmp_cpus = NULL, *tmp;
|
|
|
|
int max_entries = 0;
|
2010-05-28 17:00:01 +07:00
|
|
|
|
|
|
|
if (!cpu_list)
|
2011-01-04 02:49:48 +07:00
|
|
|
return cpu_map__read_all_cpu_map();
|
2010-05-28 17:00:01 +07:00
|
|
|
|
|
|
|
if (!isdigit(*cpu_list))
|
2011-01-04 02:49:48 +07:00
|
|
|
goto out;
|
2010-05-28 17:00:01 +07:00
|
|
|
|
|
|
|
while (isdigit(*cpu_list)) {
|
|
|
|
p = NULL;
|
|
|
|
start_cpu = strtoul(cpu_list, &p, 0);
|
|
|
|
if (start_cpu >= INT_MAX
|
|
|
|
|| (*p != '\0' && *p != ',' && *p != '-'))
|
|
|
|
goto invalid;
|
|
|
|
|
|
|
|
if (*p == '-') {
|
|
|
|
cpu_list = ++p;
|
|
|
|
p = NULL;
|
|
|
|
end_cpu = strtoul(cpu_list, &p, 0);
|
|
|
|
|
|
|
|
if (end_cpu >= INT_MAX || (*p != '\0' && *p != ','))
|
|
|
|
goto invalid;
|
|
|
|
|
|
|
|
if (end_cpu < start_cpu)
|
|
|
|
goto invalid;
|
|
|
|
} else {
|
|
|
|
end_cpu = start_cpu;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (; start_cpu <= end_cpu; start_cpu++) {
|
|
|
|
/* check for duplicates */
|
|
|
|
for (i = 0; i < nr_cpus; i++)
|
2011-01-04 02:49:48 +07:00
|
|
|
if (tmp_cpus[i] == (int)start_cpu)
|
2010-05-28 17:00:01 +07:00
|
|
|
goto invalid;
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
if (nr_cpus == max_entries) {
|
|
|
|
max_entries += MAX_NR_CPUS;
|
|
|
|
tmp = realloc(tmp_cpus, max_entries * sizeof(int));
|
|
|
|
if (tmp == NULL)
|
|
|
|
goto invalid;
|
|
|
|
tmp_cpus = tmp;
|
|
|
|
}
|
|
|
|
tmp_cpus[nr_cpus++] = (int)start_cpu;
|
2010-05-28 17:00:01 +07:00
|
|
|
}
|
|
|
|
if (*p)
|
|
|
|
++p;
|
|
|
|
|
|
|
|
cpu_list = p;
|
|
|
|
}
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
if (nr_cpus > 0)
|
|
|
|
cpus = cpu_map__trim_new(nr_cpus, tmp_cpus);
|
|
|
|
else
|
|
|
|
cpus = cpu_map__default_new();
|
2010-05-28 17:00:01 +07:00
|
|
|
invalid:
|
2011-01-04 02:49:48 +07:00
|
|
|
free(tmp_cpus);
|
|
|
|
out:
|
|
|
|
return cpus;
|
|
|
|
}
|
|
|
|
|
2015-10-25 21:51:25 +07:00
|
|
|
static struct cpu_map *cpu_map__from_entries(struct cpu_map_entries *cpus)
|
|
|
|
{
|
|
|
|
struct cpu_map *map;
|
|
|
|
|
|
|
|
map = cpu_map__empty_new(cpus->nr);
|
|
|
|
if (map) {
|
|
|
|
unsigned i;
|
|
|
|
|
2016-01-06 17:49:55 +07:00
|
|
|
for (i = 0; i < cpus->nr; i++) {
|
|
|
|
/*
|
|
|
|
* Special treatment for -1, which is not real cpu number,
|
|
|
|
* and we need to use (int) -1 to initialize map[i],
|
|
|
|
* otherwise it would become 65535.
|
|
|
|
*/
|
|
|
|
if (cpus->cpu[i] == (u16) -1)
|
|
|
|
map->map[i] = -1;
|
|
|
|
else
|
|
|
|
map->map[i] = (int) cpus->cpu[i];
|
|
|
|
}
|
2015-10-25 21:51:25 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return map;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct cpu_map *cpu_map__from_mask(struct cpu_map_mask *mask)
|
|
|
|
{
|
|
|
|
struct cpu_map *map;
|
|
|
|
int nr, nbits = mask->nr * mask->long_size * BITS_PER_BYTE;
|
|
|
|
|
|
|
|
nr = bitmap_weight(mask->mask, nbits);
|
|
|
|
|
|
|
|
map = cpu_map__empty_new(nr);
|
|
|
|
if (map) {
|
|
|
|
int cpu, i = 0;
|
|
|
|
|
|
|
|
for_each_set_bit(cpu, mask->mask, nbits)
|
|
|
|
map->map[i++] = cpu;
|
|
|
|
}
|
|
|
|
return map;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
struct cpu_map *cpu_map__new_data(struct cpu_map_data *data)
|
|
|
|
{
|
|
|
|
if (data->type == PERF_CPU_MAP__CPUS)
|
|
|
|
return cpu_map__from_entries((struct cpu_map_entries *)data->data);
|
|
|
|
else
|
|
|
|
return cpu_map__from_mask((struct cpu_map_mask *)data->data);
|
|
|
|
}
|
|
|
|
|
2012-01-19 23:07:23 +07:00
|
|
|
size_t cpu_map__fprintf(struct cpu_map *map, FILE *fp)
|
|
|
|
{
|
2016-06-28 18:29:04 +07:00
|
|
|
#define BUFSIZE 1024
|
|
|
|
char buf[BUFSIZE];
|
2012-01-19 23:07:23 +07:00
|
|
|
|
2016-06-28 18:29:04 +07:00
|
|
|
cpu_map__snprint(map, buf, sizeof(buf));
|
|
|
|
return fprintf(fp, "%s\n", buf);
|
|
|
|
#undef BUFSIZE
|
2012-01-19 23:07:23 +07:00
|
|
|
}
|
|
|
|
|
2011-01-04 02:49:48 +07:00
|
|
|
struct cpu_map *cpu_map__dummy_new(void)
|
|
|
|
{
|
|
|
|
struct cpu_map *cpus = malloc(sizeof(*cpus) + sizeof(int));
|
|
|
|
|
|
|
|
if (cpus != NULL) {
|
|
|
|
cpus->nr = 1;
|
|
|
|
cpus->map[0] = -1;
|
2015-06-23 05:36:04 +07:00
|
|
|
atomic_set(&cpus->refcnt, 1);
|
2011-01-04 02:49:48 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return cpus;
|
2010-05-28 17:00:01 +07:00
|
|
|
}
|
2011-01-15 01:19:12 +07:00
|
|
|
|
2015-10-25 21:51:17 +07:00
|
|
|
struct cpu_map *cpu_map__empty_new(int nr)
|
|
|
|
{
|
|
|
|
struct cpu_map *cpus = malloc(sizeof(*cpus) + sizeof(int) * nr);
|
|
|
|
|
|
|
|
if (cpus != NULL) {
|
|
|
|
int i;
|
|
|
|
|
|
|
|
cpus->nr = nr;
|
|
|
|
for (i = 0; i < nr; i++)
|
|
|
|
cpus->map[i] = -1;
|
|
|
|
|
|
|
|
atomic_set(&cpus->refcnt, 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
return cpus;
|
|
|
|
}
|
|
|
|
|
2015-06-23 05:36:04 +07:00
|
|
|
static void cpu_map__delete(struct cpu_map *map)
|
2011-01-15 01:19:12 +07:00
|
|
|
{
|
2015-06-23 05:36:04 +07:00
|
|
|
if (map) {
|
|
|
|
WARN_ONCE(atomic_read(&map->refcnt) != 0,
|
|
|
|
"cpu_map refcnt unbalanced\n");
|
|
|
|
free(map);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
struct cpu_map *cpu_map__get(struct cpu_map *map)
|
|
|
|
{
|
|
|
|
if (map)
|
|
|
|
atomic_inc(&map->refcnt);
|
|
|
|
return map;
|
|
|
|
}
|
|
|
|
|
|
|
|
void cpu_map__put(struct cpu_map *map)
|
|
|
|
{
|
|
|
|
if (map && atomic_dec_and_test(&map->refcnt))
|
|
|
|
cpu_map__delete(map);
|
2011-01-15 01:19:12 +07:00
|
|
|
}
|
2013-02-06 21:46:01 +07:00
|
|
|
|
2015-09-11 20:49:45 +07:00
|
|
|
static int cpu__get_topology_int(int cpu, const char *name, int *value)
|
2013-02-06 21:46:01 +07:00
|
|
|
{
|
|
|
|
char path[PATH_MAX];
|
|
|
|
|
2013-02-14 19:57:27 +07:00
|
|
|
snprintf(path, PATH_MAX,
|
2015-09-11 20:49:45 +07:00
|
|
|
"devices/system/cpu/cpu%d/topology/%s", cpu, name);
|
2013-02-06 21:46:01 +07:00
|
|
|
|
2015-09-11 20:49:45 +07:00
|
|
|
return sysfs__read_int(path, value);
|
|
|
|
}
|
2015-09-01 20:58:11 +07:00
|
|
|
|
2015-09-11 20:49:45 +07:00
|
|
|
int cpu_map__get_socket_id(int cpu)
|
|
|
|
{
|
|
|
|
int value, ret = cpu__get_topology_int(cpu, "physical_package_id", &value);
|
|
|
|
return ret ?: value;
|
2015-09-01 20:58:11 +07:00
|
|
|
}
|
|
|
|
|
2015-10-16 17:41:15 +07:00
|
|
|
int cpu_map__get_socket(struct cpu_map *map, int idx, void *data __maybe_unused)
|
2015-09-01 20:58:11 +07:00
|
|
|
{
|
|
|
|
int cpu;
|
|
|
|
|
|
|
|
if (idx > map->nr)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
cpu = map->map[idx];
|
|
|
|
|
|
|
|
return cpu_map__get_socket_id(cpu);
|
2013-02-06 21:46:01 +07:00
|
|
|
}
|
|
|
|
|
2013-02-14 19:57:27 +07:00
|
|
|
static int cmp_ids(const void *a, const void *b)
|
2013-02-06 21:46:01 +07:00
|
|
|
{
|
2013-02-14 19:57:27 +07:00
|
|
|
return *(int *)a - *(int *)b;
|
|
|
|
}
|
|
|
|
|
2015-10-16 17:41:14 +07:00
|
|
|
int cpu_map__build_map(struct cpu_map *cpus, struct cpu_map **res,
|
2015-10-16 17:41:15 +07:00
|
|
|
int (*f)(struct cpu_map *map, int cpu, void *data),
|
|
|
|
void *data)
|
2013-02-14 19:57:27 +07:00
|
|
|
{
|
|
|
|
struct cpu_map *c;
|
2013-02-06 21:46:01 +07:00
|
|
|
int nr = cpus->nr;
|
|
|
|
int cpu, s1, s2;
|
|
|
|
|
2013-02-14 19:57:27 +07:00
|
|
|
/* allocate as much as possible */
|
|
|
|
c = calloc(1, sizeof(*c) + nr * sizeof(int));
|
|
|
|
if (!c)
|
2013-02-06 21:46:01 +07:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
for (cpu = 0; cpu < nr; cpu++) {
|
2015-10-16 17:41:15 +07:00
|
|
|
s1 = f(cpus, cpu, data);
|
2013-02-14 19:57:27 +07:00
|
|
|
for (s2 = 0; s2 < c->nr; s2++) {
|
|
|
|
if (s1 == c->map[s2])
|
2013-02-06 21:46:01 +07:00
|
|
|
break;
|
|
|
|
}
|
2013-02-14 19:57:27 +07:00
|
|
|
if (s2 == c->nr) {
|
|
|
|
c->map[c->nr] = s1;
|
|
|
|
c->nr++;
|
2013-02-06 21:46:01 +07:00
|
|
|
}
|
|
|
|
}
|
2013-02-14 19:57:27 +07:00
|
|
|
/* ensure we process id in increasing order */
|
|
|
|
qsort(c->map, c->nr, sizeof(int), cmp_ids);
|
|
|
|
|
2015-10-09 17:59:23 +07:00
|
|
|
atomic_set(&c->refcnt, 1);
|
2013-02-14 19:57:27 +07:00
|
|
|
*res = c;
|
2013-02-06 21:46:01 +07:00
|
|
|
return 0;
|
|
|
|
}
|
2013-02-14 19:57:27 +07:00
|
|
|
|
2015-09-01 20:58:11 +07:00
|
|
|
int cpu_map__get_core_id(int cpu)
|
2013-02-14 19:57:29 +07:00
|
|
|
{
|
2015-09-11 20:49:45 +07:00
|
|
|
int value, ret = cpu__get_topology_int(cpu, "core_id", &value);
|
|
|
|
return ret ?: value;
|
2015-09-01 20:58:11 +07:00
|
|
|
}
|
|
|
|
|
2015-10-16 17:41:15 +07:00
|
|
|
int cpu_map__get_core(struct cpu_map *map, int idx, void *data)
|
2015-09-01 20:58:11 +07:00
|
|
|
{
|
|
|
|
int cpu, s;
|
|
|
|
|
|
|
|
if (idx > map->nr)
|
2013-02-14 19:57:29 +07:00
|
|
|
return -1;
|
|
|
|
|
2015-09-01 20:58:11 +07:00
|
|
|
cpu = map->map[idx];
|
|
|
|
|
|
|
|
cpu = cpu_map__get_core_id(cpu);
|
|
|
|
|
2015-10-16 17:41:15 +07:00
|
|
|
s = cpu_map__get_socket(map, idx, data);
|
2013-02-14 19:57:29 +07:00
|
|
|
if (s == -1)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* encode socket in upper 16 bits
|
|
|
|
* core_id is relative to socket, and
|
|
|
|
* we need a global id. So we combine
|
|
|
|
* socket+ core id
|
|
|
|
*/
|
|
|
|
return (s << 16) | (cpu & 0xffff);
|
|
|
|
}
|
|
|
|
|
2013-02-14 19:57:27 +07:00
|
|
|
int cpu_map__build_socket_map(struct cpu_map *cpus, struct cpu_map **sockp)
|
|
|
|
{
|
2015-10-16 17:41:15 +07:00
|
|
|
return cpu_map__build_map(cpus, sockp, cpu_map__get_socket, NULL);
|
2013-02-14 19:57:27 +07:00
|
|
|
}
|
2013-02-14 19:57:29 +07:00
|
|
|
|
|
|
|
int cpu_map__build_core_map(struct cpu_map *cpus, struct cpu_map **corep)
|
|
|
|
{
|
2015-10-16 17:41:15 +07:00
|
|
|
return cpu_map__build_map(cpus, corep, cpu_map__get_core, NULL);
|
2013-02-14 19:57:29 +07:00
|
|
|
}
|
2014-04-08 01:55:21 +07:00
|
|
|
|
|
|
|
/* setup simple routines to easily access node numbers given a cpu number */
|
|
|
|
static int get_max_num(char *path, int *max)
|
|
|
|
{
|
|
|
|
size_t num;
|
|
|
|
char *buf;
|
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
if (filename__read_str(path, &buf, &num))
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
buf[num] = '\0';
|
|
|
|
|
|
|
|
/* start on the right, to find highest node num */
|
|
|
|
while (--num) {
|
|
|
|
if ((buf[num] == ',') || (buf[num] == '-')) {
|
|
|
|
num++;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (sscanf(&buf[num], "%d", max) < 1) {
|
|
|
|
err = -1;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* convert from 0-based to 1-based */
|
|
|
|
(*max)++;
|
|
|
|
|
|
|
|
out:
|
|
|
|
free(buf);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Determine highest possible cpu in the system for sparse allocation */
|
|
|
|
static void set_max_cpu_num(void)
|
|
|
|
{
|
|
|
|
const char *mnt;
|
|
|
|
char path[PATH_MAX];
|
|
|
|
int ret = -1;
|
|
|
|
|
|
|
|
/* set up default */
|
|
|
|
max_cpu_num = 4096;
|
2017-02-17 18:10:24 +07:00
|
|
|
max_present_cpu_num = 4096;
|
2014-04-08 01:55:21 +07:00
|
|
|
|
|
|
|
mnt = sysfs__mountpoint();
|
|
|
|
if (!mnt)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* get the highest possible cpu number for a sparse allocation */
|
2014-04-08 01:55:22 +07:00
|
|
|
ret = snprintf(path, PATH_MAX, "%s/devices/system/cpu/possible", mnt);
|
2014-04-08 01:55:21 +07:00
|
|
|
if (ret == PATH_MAX) {
|
|
|
|
pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = get_max_num(path, &max_cpu_num);
|
2017-02-17 18:10:24 +07:00
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* get the highest present cpu number for a sparse allocation */
|
|
|
|
ret = snprintf(path, PATH_MAX, "%s/devices/system/cpu/present", mnt);
|
|
|
|
if (ret == PATH_MAX) {
|
|
|
|
pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = get_max_num(path, &max_present_cpu_num);
|
2014-04-08 01:55:21 +07:00
|
|
|
|
|
|
|
out:
|
|
|
|
if (ret)
|
|
|
|
pr_err("Failed to read max cpus, using default of %d\n", max_cpu_num);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Determine highest possible node in the system for sparse allocation */
|
|
|
|
static void set_max_node_num(void)
|
|
|
|
{
|
|
|
|
const char *mnt;
|
|
|
|
char path[PATH_MAX];
|
|
|
|
int ret = -1;
|
|
|
|
|
|
|
|
/* set up default */
|
|
|
|
max_node_num = 8;
|
|
|
|
|
|
|
|
mnt = sysfs__mountpoint();
|
|
|
|
if (!mnt)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* get the highest possible cpu number for a sparse allocation */
|
|
|
|
ret = snprintf(path, PATH_MAX, "%s/devices/system/node/possible", mnt);
|
|
|
|
if (ret == PATH_MAX) {
|
|
|
|
pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = get_max_num(path, &max_node_num);
|
|
|
|
|
|
|
|
out:
|
|
|
|
if (ret)
|
|
|
|
pr_err("Failed to read max nodes, using default of %d\n", max_node_num);
|
|
|
|
}
|
|
|
|
|
2016-01-27 01:51:46 +07:00
|
|
|
int cpu__max_node(void)
|
|
|
|
{
|
|
|
|
if (unlikely(!max_node_num))
|
|
|
|
set_max_node_num();
|
|
|
|
|
|
|
|
return max_node_num;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpu__max_cpu(void)
|
|
|
|
{
|
|
|
|
if (unlikely(!max_cpu_num))
|
|
|
|
set_max_cpu_num();
|
|
|
|
|
|
|
|
return max_cpu_num;
|
|
|
|
}
|
|
|
|
|
2017-02-17 18:10:24 +07:00
|
|
|
int cpu__max_present_cpu(void)
|
|
|
|
{
|
|
|
|
if (unlikely(!max_present_cpu_num))
|
|
|
|
set_max_cpu_num();
|
|
|
|
|
|
|
|
return max_present_cpu_num;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2016-01-27 01:51:46 +07:00
|
|
|
int cpu__get_node(int cpu)
|
|
|
|
{
|
|
|
|
if (unlikely(cpunode_map == NULL)) {
|
|
|
|
pr_debug("cpu_map not initialized\n");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return cpunode_map[cpu];
|
|
|
|
}
|
|
|
|
|
2014-04-08 01:55:21 +07:00
|
|
|
static int init_cpunode_map(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
set_max_cpu_num();
|
|
|
|
set_max_node_num();
|
|
|
|
|
|
|
|
cpunode_map = calloc(max_cpu_num, sizeof(int));
|
|
|
|
if (!cpunode_map) {
|
|
|
|
pr_err("%s: calloc failed\n", __func__);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < max_cpu_num; i++)
|
|
|
|
cpunode_map[i] = -1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpu__setup_cpunode_map(void)
|
|
|
|
{
|
|
|
|
struct dirent *dent1, *dent2;
|
|
|
|
DIR *dir1, *dir2;
|
|
|
|
unsigned int cpu, mem;
|
|
|
|
char buf[PATH_MAX];
|
|
|
|
char path[PATH_MAX];
|
|
|
|
const char *mnt;
|
|
|
|
int n;
|
|
|
|
|
|
|
|
/* initialize globals */
|
|
|
|
if (init_cpunode_map())
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
mnt = sysfs__mountpoint();
|
|
|
|
if (!mnt)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
n = snprintf(path, PATH_MAX, "%s/devices/system/node", mnt);
|
|
|
|
if (n == PATH_MAX) {
|
|
|
|
pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
dir1 = opendir(path);
|
|
|
|
if (!dir1)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* walk tree and setup map */
|
|
|
|
while ((dent1 = readdir(dir1)) != NULL) {
|
|
|
|
if (dent1->d_type != DT_DIR || sscanf(dent1->d_name, "node%u", &mem) < 1)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
n = snprintf(buf, PATH_MAX, "%s/%s", path, dent1->d_name);
|
|
|
|
if (n == PATH_MAX) {
|
|
|
|
pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
dir2 = opendir(buf);
|
|
|
|
if (!dir2)
|
|
|
|
continue;
|
|
|
|
while ((dent2 = readdir(dir2)) != NULL) {
|
|
|
|
if (dent2->d_type != DT_LNK || sscanf(dent2->d_name, "cpu%u", &cpu) < 1)
|
|
|
|
continue;
|
|
|
|
cpunode_map[cpu] = mem;
|
|
|
|
}
|
|
|
|
closedir(dir2);
|
|
|
|
}
|
|
|
|
closedir(dir1);
|
|
|
|
return 0;
|
|
|
|
}
|
2016-04-12 20:29:25 +07:00
|
|
|
|
|
|
|
bool cpu_map__has(struct cpu_map *cpus, int cpu)
|
2016-07-15 17:08:11 +07:00
|
|
|
{
|
|
|
|
return cpu_map__idx(cpus, cpu) != -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpu_map__idx(struct cpu_map *cpus, int cpu)
|
2016-04-12 20:29:25 +07:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < cpus->nr; ++i) {
|
|
|
|
if (cpus->map[i] == cpu)
|
2016-07-15 17:08:11 +07:00
|
|
|
return i;
|
2016-04-12 20:29:25 +07:00
|
|
|
}
|
|
|
|
|
2016-07-15 17:08:11 +07:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpu_map__cpu(struct cpu_map *cpus, int idx)
|
|
|
|
{
|
|
|
|
return cpus->map[idx];
|
2016-04-12 20:29:25 +07:00
|
|
|
}
|
2016-06-28 18:29:04 +07:00
|
|
|
|
|
|
|
size_t cpu_map__snprint(struct cpu_map *map, char *buf, size_t size)
|
|
|
|
{
|
|
|
|
int i, cpu, start = -1;
|
|
|
|
bool first = true;
|
|
|
|
size_t ret = 0;
|
|
|
|
|
|
|
|
#define COMMA first ? "" : ","
|
|
|
|
|
|
|
|
for (i = 0; i < map->nr + 1; i++) {
|
|
|
|
bool last = i == map->nr;
|
|
|
|
|
|
|
|
cpu = last ? INT_MAX : map->map[i];
|
|
|
|
|
|
|
|
if (start == -1) {
|
|
|
|
start = i;
|
|
|
|
if (last) {
|
|
|
|
ret += snprintf(buf + ret, size - ret,
|
|
|
|
"%s%d", COMMA,
|
|
|
|
map->map[i]);
|
|
|
|
}
|
|
|
|
} else if (((i - start) != (cpu - map->map[start])) || last) {
|
|
|
|
int end = i - 1;
|
|
|
|
|
|
|
|
if (start == end) {
|
|
|
|
ret += snprintf(buf + ret, size - ret,
|
|
|
|
"%s%d", COMMA,
|
|
|
|
map->map[start]);
|
|
|
|
} else {
|
|
|
|
ret += snprintf(buf + ret, size - ret,
|
|
|
|
"%s%d-%d", COMMA,
|
|
|
|
map->map[start], map->map[end]);
|
|
|
|
}
|
|
|
|
first = false;
|
|
|
|
start = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#undef COMMA
|
|
|
|
|
|
|
|
pr_debug("cpumask list: %s\n", buf);
|
|
|
|
return ret;
|
|
|
|
}
|