platform_system_core/libcutils/sched_policy.cpp

529 lines
14 KiB
C++

/*
** Copyright 2007, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#define LOG_TAG "SchedPolicy"
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <log/log.h>
#include <cutils/sched_policy.h>
#define UNUSED __attribute__((__unused__))
/* Re-map SP_DEFAULT to the system default policy, and leave other values unchanged.
* Call this any place a SchedPolicy is used as an input parameter.
* Returns the possibly re-mapped policy.
*/
static inline SchedPolicy _policy(SchedPolicy p)
{
return p == SP_DEFAULT ? SP_SYSTEM_DEFAULT : p;
}
#if defined(__ANDROID__)
#include <pthread.h>
#include <sched.h>
#include <sys/prctl.h>
#define POLICY_DEBUG 0
// timer slack value in nS enforced when the thread moves to background
#define TIMER_SLACK_BG 40000000
#define TIMER_SLACK_FG 50000
static pthread_once_t the_once = PTHREAD_ONCE_INIT;
static int __sys_supports_schedgroups = -1;
static int __sys_supports_timerslack = -1;
// File descriptors open to /dev/cpuctl/../tasks, setup by initialize, or -1 on error.
static int bg_cgroup_fd = -1;
static int fg_cgroup_fd = -1;
// File descriptors open to /dev/cpuset/../tasks, setup by initialize, or -1 on error
static int system_bg_cpuset_fd = -1;
static int bg_cpuset_fd = -1;
static int fg_cpuset_fd = -1;
static int ta_cpuset_fd = -1; // special cpuset for top app
// File descriptors open to /dev/stune/../tasks, setup by initialize, or -1 on error
static int bg_schedboost_fd = -1;
static int fg_schedboost_fd = -1;
static int ta_schedboost_fd = -1;
/* Add tid to the scheduling group defined by the policy */
static int add_tid_to_cgroup(int tid, int fd)
{
if (fd < 0) {
SLOGE("add_tid_to_cgroup failed; fd=%d\n", fd);
errno = EINVAL;
return -1;
}
// specialized itoa -- works for tid > 0
char text[22];
char *end = text + sizeof(text) - 1;
char *ptr = end;
*ptr = '\0';
while (tid > 0) {
*--ptr = '0' + (tid % 10);
tid = tid / 10;
}
if (write(fd, ptr, end - ptr) < 0) {
/*
* If the thread is in the process of exiting,
* don't flag an error
*/
if (errno == ESRCH)
return 0;
SLOGW("add_tid_to_cgroup failed to write '%s' (%s); fd=%d\n",
ptr, strerror(errno), fd);
errno = EINVAL;
return -1;
}
return 0;
}
/*
If CONFIG_CPUSETS for Linux kernel is set, "tasks" can be found under
/dev/cpuset mounted in init.rc; otherwise, that file does not exist
even though the directory, /dev/cpuset, is still created (by init.rc).
A couple of other candidates (under cpuset mount directory):
notify_on_release
release_agent
Yet another way to decide if cpuset is enabled is to parse
/proc/self/status and search for lines begin with "Mems_allowed".
If CONFIG_PROC_PID_CPUSET is set, the existence "/proc/self/cpuset" can
be used to decide if CONFIG_CPUSETS is set, so we don't have a dependency
on where init.rc mounts cpuset. That's why we'd better require this
configuration be set if CONFIG_CPUSETS is set.
With runtime check using the following function, build time
variables like ENABLE_CPUSETS (used in Android.mk) or cpusets (used
in Android.bp) are not needed.
*/
bool cpusets_enabled() {
static bool enabled = (access("/dev/cpuset/tasks", F_OK) == 0);
return enabled;
}
/*
Similar to CONFIG_CPUSETS above, but with a different configuration
CONFIG_SCHEDTUNE that's in Android common Linux kernel and Linaro
Stable Kernel (LSK), but not in mainline Linux as of v4.9.
With runtime check using the following function, build time
variables like ENABLE_SCHEDBOOST (used in Android.mk) or schedboost
(used in Android.bp) are not needed.
*/
bool schedboost_enabled() {
static bool enabled = (access("/dev/stune/tasks", F_OK) == 0);
return enabled;
}
static void __initialize() {
const char* filename;
if (!access("/dev/cpuctl/tasks", W_OK)) {
__sys_supports_schedgroups = 1;
filename = "/dev/cpuctl/tasks";
fg_cgroup_fd = open(filename, O_WRONLY | O_CLOEXEC);
if (fg_cgroup_fd < 0) {
SLOGE("open of %s failed: %s\n", filename, strerror(errno));
}
filename = "/dev/cpuctl/bg_non_interactive/tasks";
bg_cgroup_fd = open(filename, O_WRONLY | O_CLOEXEC);
if (bg_cgroup_fd < 0) {
SLOGE("open of %s failed: %s\n", filename, strerror(errno));
}
} else {
__sys_supports_schedgroups = 0;
}
if (cpusets_enabled()) {
if (!access("/dev/cpuset/tasks", W_OK)) {
filename = "/dev/cpuset/foreground/tasks";
fg_cpuset_fd = open(filename, O_WRONLY | O_CLOEXEC);
filename = "/dev/cpuset/background/tasks";
bg_cpuset_fd = open(filename, O_WRONLY | O_CLOEXEC);
filename = "/dev/cpuset/system-background/tasks";
system_bg_cpuset_fd = open(filename, O_WRONLY | O_CLOEXEC);
filename = "/dev/cpuset/top-app/tasks";
ta_cpuset_fd = open(filename, O_WRONLY | O_CLOEXEC);
if (schedboost_enabled()) {
filename = "/dev/stune/top-app/tasks";
ta_schedboost_fd = open(filename, O_WRONLY | O_CLOEXEC);
filename = "/dev/stune/foreground/tasks";
fg_schedboost_fd = open(filename, O_WRONLY | O_CLOEXEC);
filename = "/dev/stune/background/tasks";
bg_schedboost_fd = open(filename, O_WRONLY | O_CLOEXEC);
}
}
}
char buf[64];
snprintf(buf, sizeof(buf), "/proc/%d/timerslack_ns", getpid());
__sys_supports_timerslack = !access(buf, W_OK);
}
/*
* Returns the path under the requested cgroup subsystem (if it exists)
*
* The data from /proc/<pid>/cgroup looks (something) like:
* 2:cpu:/bg_non_interactive
* 1:cpuacct:/
*
* We return the part after the "/", which will be an empty string for
* the default cgroup. If the string is longer than "bufLen", the string
* will be truncated.
*/
static int getCGroupSubsys(int tid, const char* subsys, char* buf, size_t bufLen)
{
#if defined(__ANDROID__)
char pathBuf[32];
char lineBuf[256];
FILE *fp;
snprintf(pathBuf, sizeof(pathBuf), "/proc/%d/cgroup", tid);
if (!(fp = fopen(pathBuf, "re"))) {
return -1;
}
while(fgets(lineBuf, sizeof(lineBuf) -1, fp)) {
char *next = lineBuf;
char *found_subsys;
char *grp;
size_t len;
/* Junk the first field */
if (!strsep(&next, ":")) {
goto out_bad_data;
}
if (!(found_subsys = strsep(&next, ":"))) {
goto out_bad_data;
}
if (strcmp(found_subsys, subsys)) {
/* Not the subsys we're looking for */
continue;
}
if (!(grp = strsep(&next, ":"))) {
goto out_bad_data;
}
grp++; /* Drop the leading '/' */
len = strlen(grp);
grp[len-1] = '\0'; /* Drop the trailing '\n' */
if (bufLen <= len) {
len = bufLen - 1;
}
strncpy(buf, grp, len);
buf[len] = '\0';
fclose(fp);
return 0;
}
SLOGE("Failed to find subsys %s", subsys);
fclose(fp);
return -1;
out_bad_data:
SLOGE("Bad cgroup data {%s}", lineBuf);
fclose(fp);
return -1;
#else
errno = ENOSYS;
return -1;
#endif
}
int get_sched_policy(int tid, SchedPolicy *policy)
{
if (tid == 0) {
tid = gettid();
}
pthread_once(&the_once, __initialize);
if (__sys_supports_schedgroups) {
char grpBuf[32];
if (cpusets_enabled()) {
if (getCGroupSubsys(tid, "cpuset", grpBuf, sizeof(grpBuf)) < 0)
return -1;
if (grpBuf[0] == '\0') {
*policy = SP_FOREGROUND;
} else if (!strcmp(grpBuf, "foreground")) {
*policy = SP_FOREGROUND;
} else if (!strcmp(grpBuf, "background")) {
*policy = SP_BACKGROUND;
} else if (!strcmp(grpBuf, "top-app")) {
*policy = SP_TOP_APP;
} else {
errno = ERANGE;
return -1;
}
} else {
if (getCGroupSubsys(tid, "cpu", grpBuf, sizeof(grpBuf)) < 0)
return -1;
if (grpBuf[0] == '\0') {
*policy = SP_FOREGROUND;
} else if (!strcmp(grpBuf, "bg_non_interactive")) {
*policy = SP_BACKGROUND;
} else {
errno = ERANGE;
return -1;
}
}
} else {
int rc = sched_getscheduler(tid);
if (rc < 0)
return -1;
else if (rc == SCHED_NORMAL)
*policy = SP_FOREGROUND;
else if (rc == SCHED_BATCH)
*policy = SP_BACKGROUND;
else {
errno = ERANGE;
return -1;
}
}
return 0;
}
int set_cpuset_policy(int tid, SchedPolicy policy)
{
// in the absence of cpusets, use the old sched policy
if (!cpusets_enabled()) {
return set_sched_policy(tid, policy);
}
if (tid == 0) {
tid = gettid();
}
policy = _policy(policy);
pthread_once(&the_once, __initialize);
int fd = -1;
int boost_fd = -1;
switch (policy) {
case SP_BACKGROUND:
fd = bg_cpuset_fd;
boost_fd = bg_schedboost_fd;
break;
case SP_FOREGROUND:
case SP_AUDIO_APP:
case SP_AUDIO_SYS:
fd = fg_cpuset_fd;
boost_fd = fg_schedboost_fd;
break;
case SP_TOP_APP :
fd = ta_cpuset_fd;
boost_fd = ta_schedboost_fd;
break;
case SP_SYSTEM:
fd = system_bg_cpuset_fd;
break;
default:
boost_fd = fd = -1;
break;
}
if (add_tid_to_cgroup(tid, fd) != 0) {
if (errno != ESRCH && errno != ENOENT)
return -errno;
}
if (schedboost_enabled()) {
if (boost_fd > 0 && add_tid_to_cgroup(tid, boost_fd) != 0) {
if (errno != ESRCH && errno != ENOENT)
return -errno;
}
}
return 0;
}
static void set_timerslack_ns(int tid, unsigned long long slack) {
// v4.6+ kernels support the /proc/<tid>/timerslack_ns interface.
// TODO: once we've backported this, log if the open(2) fails.
char buf[64];
snprintf(buf, sizeof(buf), "/proc/%d/timerslack_ns", tid);
int fd = open(buf, O_WRONLY | O_CLOEXEC);
if (fd != -1) {
int len = snprintf(buf, sizeof(buf), "%llu", slack);
if (write(fd, buf, len) != len) {
SLOGE("set_timerslack_ns write failed: %s\n", strerror(errno));
}
close(fd);
return;
}
}
int set_sched_policy(int tid, SchedPolicy policy)
{
if (tid == 0) {
tid = gettid();
}
policy = _policy(policy);
pthread_once(&the_once, __initialize);
#if POLICY_DEBUG
char statfile[64];
char statline[1024];
char thread_name[255];
snprintf(statfile, sizeof(statfile), "/proc/%d/stat", tid);
memset(thread_name, 0, sizeof(thread_name));
int fd = open(statfile, O_RDONLY | O_CLOEXEC);
if (fd >= 0) {
int rc = read(fd, statline, 1023);
close(fd);
statline[rc] = 0;
char *p = statline;
char *q;
for (p = statline; *p != '('; p++);
p++;
for (q = p; *q != ')'; q++);
strncpy(thread_name, p, (q-p));
}
switch (policy) {
case SP_BACKGROUND:
SLOGD("vvv tid %d (%s)", tid, thread_name);
break;
case SP_FOREGROUND:
case SP_AUDIO_APP:
case SP_AUDIO_SYS:
case SP_TOP_APP:
SLOGD("^^^ tid %d (%s)", tid, thread_name);
break;
case SP_SYSTEM:
SLOGD("/// tid %d (%s)", tid, thread_name);
break;
default:
SLOGD("??? tid %d (%s)", tid, thread_name);
break;
}
#endif
if (__sys_supports_schedgroups) {
int fd = -1;
int boost_fd = -1;
switch (policy) {
case SP_BACKGROUND:
fd = bg_cgroup_fd;
boost_fd = bg_schedboost_fd;
break;
case SP_FOREGROUND:
case SP_AUDIO_APP:
case SP_AUDIO_SYS:
fd = fg_cgroup_fd;
boost_fd = fg_schedboost_fd;
break;
case SP_TOP_APP:
fd = fg_cgroup_fd;
boost_fd = ta_schedboost_fd;
break;
default:
fd = -1;
boost_fd = -1;
break;
}
if (add_tid_to_cgroup(tid, fd) != 0) {
if (errno != ESRCH && errno != ENOENT)
return -errno;
}
if (schedboost_enabled()) {
if (boost_fd > 0 && add_tid_to_cgroup(tid, boost_fd) != 0) {
if (errno != ESRCH && errno != ENOENT)
return -errno;
}
}
} else {
struct sched_param param;
param.sched_priority = 0;
sched_setscheduler(tid,
(policy == SP_BACKGROUND) ?
SCHED_BATCH : SCHED_NORMAL,
&param);
}
if (__sys_supports_timerslack) {
set_timerslack_ns(tid, policy == SP_BACKGROUND ?
TIMER_SLACK_BG : TIMER_SLACK_FG);
}
return 0;
}
#else
/* Stubs for non-Android targets. */
int set_sched_policy(int tid UNUSED, SchedPolicy policy UNUSED)
{
return 0;
}
int get_sched_policy(int tid UNUSED, SchedPolicy *policy)
{
*policy = SP_SYSTEM_DEFAULT;
return 0;
}
#endif
const char *get_sched_policy_name(SchedPolicy policy)
{
policy = _policy(policy);
static const char * const strings[SP_CNT] = {
[SP_BACKGROUND] = "bg",
[SP_FOREGROUND] = "fg",
[SP_SYSTEM] = " ",
[SP_AUDIO_APP] = "aa",
[SP_AUDIO_SYS] = "as",
[SP_TOP_APP] = "ta",
};
if ((policy < SP_CNT) && (strings[policy] != NULL))
return strings[policy];
else
return "error";
}