梳理Process進(jìn)程相關(guān)知識(shí)點(diǎn)，再繼續(xù)補(bǔ)充點(diǎn)內(nèi)容。
Linux系統(tǒng)中對(duì)進(jìn)程的管理無(wú)非是從調(diào)度策略、優(yōu)先級(jí)以及CPU限制三個(gè)角度進(jìn)行配置與管理，那么Android中主要是通過(guò)AMS來(lái)管理應(yīng)用程序進(jìn)程的，是不是也是從這三個(gè)方面進(jìn)行管理的呢？答案是肯定的，那么本篇文章先來(lái)看看cpuset負(fù)載均衡在AMS中是如何應(yīng)用的。
一、cpuset介紹
cpuset是Linux cgroup子系統(tǒng)，它為cgroup任務(wù)分配單獨(dú)的CPU和內(nèi)存。單獨(dú)分配CPU即表明進(jìn)程可調(diào)度cpu范圍。cpu按不同的芯片，大小核數(shù)目和頻率都有差別，大核頻率高處理速度相對(duì)比小核快，而Android系統(tǒng)實(shí)際上還是響應(yīng)優(yōu)先于吞吐的交互型系統(tǒng)，因此Android AMS對(duì)進(jìn)程管理于不同優(yōu)先級(jí)的進(jìn)程在調(diào)度cpu限制上會(huì)做有一些策略，以保證更好的交互響應(yīng)。
二、進(jìn)程的cpuset設(shè)置
還是回到AMS中與adj相關(guān)的有三個(gè)方法，這三個(gè)方法值得看一萬(wàn)遍，每一遍都會(huì)有新收獲：
- updateOomAdjLocked：在進(jìn)程組件生命周期變化時(shí)更新adj，然后分別執(zhí)行以下兩個(gè)方法
- computeOomAdjLocked：計(jì)算adj
- applyOomAdjLocked：應(yīng)用adj
聚焦到computeOomAdjLocked方法，該方法主要是根據(jù)進(jìn)程的四大組件狀態(tài)決定當(dāng)前進(jìn)程的adj優(yōu)先級(jí)。
frameworks/base/services/core/java/com/android/server/am/ActivityManagerService.java
private final int computeOomAdjLocked(ProcessRecord app, int cachedAdj, ProcessRecord TOP_APP, boolean doingAll, long now) {
...
if (app.maxAdj <= ProcessList.FOREGROUND_APP_ADJ) {
...
if (app == TOP_APP) {
… //以top app為例
app.curSchedGroup = ProcessList.SCHED_GROUP_TOP_APP;
...
}
...
}
以TOP_APP為例，這里ProcessRecord 的curSchedGroup屬性對(duì)應(yīng)的是cup調(diào)度組，而在后續(xù)applyOomAdjLocked中會(huì)執(zhí)行Process的setProcessGroup方法。
frameworks/base/services/core/java/com/android/server/am/ActivityManagerService.java
import static android.os.Process.setProcessGroup;
private final boolean applyOomAdjLocked(ProcessRecord app, boolean doingAll, long now,
long nowElapsed) {
...
setProcessGroup(app.pid, processGroup);
...
}
調(diào)用Process的setProcessGroup方法
frameworks/base/core/java/android/os/Process.java
public static final int THREAD_GROUP_DEFAULT = -1;
public static final int THREAD_GROUP_BG_NONINTERACTIVE = 0;
private static final int THREAD_GROUP_FOREGROUND = 1;
public static final int THREAD_GROUP_SYSTEM = 2;
public static final int THREAD_GROUP_AUDIO_APP = 3;
public static final int THREAD_GROUP_AUDIO_SYS = 4;
public static final int THREAD_GROUP_TOP_APP = 5;
...
public static final native void setProcessGroup(int pid, int group)
throws IllegalArgumentException, SecurityException;
setProcessGroup是個(gè)native方法，并且這里分了若干類(lèi)型的group，這里看top app優(yōu)先級(jí)是最高的。接著jni到native
frameworks/base/core/jni/android_util_Process.cpp
void android_os_Process_setProcessGroup(JNIEnv* env, jobject clazz, int pid, jint grp)
{
...
if ((grp == SP_FOREGROUND) || (grp > SP_MAX)) {
signalExceptionForGroupError(env, EINVAL, pid);
return;
}
bool isDefault = false;
if (grp < 0) {
grp = SP_FOREGROUND; //SP_FOREGROUND = 1
isDefault = true;
}
SchedPolicy sp = (SchedPolicy) grp;
...
if (cpusets_enabled()) {
// set both cpuset and cgroup for general threads
err = set_cpuset_policy(t_pid, sp);
...
}
...
}
這里直接調(diào)用sched_policy.cpp的set_cpuset_policy，并傳入對(duì)應(yīng)的pid和SchedPolicy
system/core/libcutils/sched_policy.cpp
int set_cpuset_policy(int tid, SchedPolicy policy)
{
// in the absence of cpusets, use the old sched policy
if (!cpusets_enabled()) {//這里要注意，如果cpusets_enabled為false的話，會(huì)走set_sched_policy
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;
}
這里主要就是通過(guò)policy對(duì)應(yīng)具體的fd句柄，然后通過(guò)add_tid_to_cgroup()寫(xiě)cpuset對(duì)應(yīng)節(jié)點(diǎn)。這里要注意，如果cpusets_enabled為false的話，會(huì)走set_sched_policy，這部分下篇會(huì)講到。
看看對(duì)應(yīng)的fd是什么：
static void __initialize() {
const char* filename;
if (cpusets_enabled()) {
if (!access("/dev/cpuset/tasks", W_OK)) {
filename = "/dev/cpuset/foreground/tasks”; //這里對(duì)應(yīng)的是cpuset路徑
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);
}
}
...
}
那我們來(lái)看看對(duì)應(yīng)節(jié)點(diǎn)是什么內(nèi)容：
/dev/cpuset # ls -al
drwxr-xr-x 2 system system 0 1970-05-22 03:35 background
drwxr-xr-x 2 system system 0 1970-05-22 03:35 foreground
drwxr-xr-x 2 system system 0 1970-05-22 03:35 top-app
然后看看對(duì)應(yīng)的cpuset配置：
/dev/cpuset # cat top-app/cpus
XXX 所有核
/dev/cpuset # cat background/cpus
XXX 小核
顯然，top app 滿核隨便跑，foreground跑在除了3這個(gè)核以外的所有核上, 而background只能跑在小核上。
三、cpuset配置點(diǎn)
不同芯片平臺(tái)配置會(huì)有差別。
四、cpuset開(kāi)關(guān)
kernel/msm-4.14/arch/arm64/configs/XXXX
CONFIG_CPUSETS=y //打開(kāi)cpuset
