美文网首页
LowMemoryKiller机制分析

LowMemoryKiller机制分析

作者: 泡面先生_Jack | 来源:发表于2019-11-02 22:22 被阅读0次

    概述

    Linux系统当可用内存较低的时候oom killer机制会根据一定的规则去杀掉一些进程来释放内存,而Android系统的LowMemoryKiller机制就是以此功能为基础做了一些调整。Android系统中的APP在使用完成之后并不会马上被杀掉,而是驻留在内存中,当下一次在此进入此应用的时候可以省去进程创建的过程,加快启动速度。LowMemoryKiller机制会在内存资源紧张的时候,杀掉一些进程来回收内存。

    整体架构

    LowMemoryKiller机制分为三个部分


    LowMemoryKiller整体结构图
    1. AMS部分的ProcessList
    2. Native进程lmkd
    3. 内核中的LowMemoryKiller部分

    Framework中的ProcessList和Native的lmkd进程通过Socket进行进程间通信,而lmkd和内核中的LowMemoryKiller通过writeFileString向文件节点写内容方法进行通信。

    Framework层通过一定的规则调整进程的adj的值和内存空间阀值,然后通过socket发送给lmkd进程,lmkd两种处理方式, 一种将阀值写入文件节点发送给内核的LowMemoryKiller,由内核进行杀进程处理,另一种是lmkd通过cgroup监控内存使用情况,自行计算杀掉进程。

    lmkd的启动和初始化

    lmkd是一个native进程,由init进程启动,定义在/system/core/lmkd/lmkd.rc中

    service lmkd /system/bin/lmkd
        class core
        group root readproc
        critical
        socket lmkd seqpacket 0660 system system
        writepid /dev/cpuset/system-background/tasks
    

    在lmkd.rc中,启动了lmkd进程,并创建了一个名为lmkd的socket的描述符,用于socket进程间通信。lmkd启动后首先执行main方法。

    int main(int argc __unused, char **argv __unused) {
        struct sched_param param = {
                .sched_priority = 1,
        };
        sched_setscheduler(0, SCHED_FIFO, &param);
        if (!init())
            mainloop();
    }
    

    main方法首先设置了当前进程的调度规则,然后执行了init方法和mainLoop方法。

    static int init(void) {
        struct epoll_event epev;
        int i;
        int ret;
        //获取当前系统的页大小,单位kb
        page_k = sysconf(_SC_PAGESIZE);
        if (page_k == -1)
            page_k = PAGE_SIZE;
        page_k /= 1024;
        // 创建一个epollfd描述符
        epollfd = epoll_create(MAX_EPOLL_EVENTS);
        if (epollfd == -1) {
            ALOGE("epoll_create failed (errno=%d)", errno);
            return -1;
        }
    
        // mark data connections as not connected
        for (int i = 0; i < MAX_DATA_CONN; i++) {
            data_sock[i].sock = -1;
        }
    
        // 获取init.rc中创建的lmkd socket描述符
        ctrl_sock.sock = android_get_control_socket("lmkd");
        if (ctrl_sock.sock < 0) {
            ALOGE("get lmkd control socket failed");
            return -1;
        }
        
        // 监听socket的连接,即ProcessList的Socket连接
        ret = listen(ctrl_sock.sock, MAX_DATA_CONN);
        if (ret < 0) {
            ALOGE("lmkd control socket listen failed (errno=%d)", errno);
            return -1;
        }
    
        // Epoll 设置监听socket中的可读事件,当有可读事件的时候回调ctrl_connect_hander方法
       //处理socket连接过程
        epev.events = EPOLLIN;
        ctrl_sock.handler_info.handler = ctrl_connect_handler;
        epev.data.ptr = (void *)&(ctrl_sock.handler_info);
        if (epoll_ctl(epollfd, EPOLL_CTL_ADD, ctrl_sock.sock, &epev) == -1) {
            ALOGE("epoll_ctl for lmkd control socket failed (errno=%d)", errno);
            return -1;
        }
        maxevents++;
       
        //检测内核是否支持lowMemoryKiller机制
        has_inkernel_module = !access(INKERNEL_MINFREE_PATH, W_OK);
        use_inkernel_interface = has_inkernel_module;
    
        //如果内核不支持LowMemoryKiller,则调用init_mp_common初始化,在lmkd中实现进程查杀过程
        if (use_inkernel_interface) {
            ALOGI("Using in-kernel low memory killer interface");
        } else {
            if (!init_mp_common(VMPRESS_LEVEL_LOW) ||
                !init_mp_common(VMPRESS_LEVEL_MEDIUM) ||
                !init_mp_common(VMPRESS_LEVEL_CRITICAL)) {
                ALOGE("Kernel does not support memory pressure events or in-kernel low memory killer");
                return -1;
            }
        }
    
        //初始化lmkd中的进程列表
        for (i = 0; i <= ADJTOSLOT(OOM_SCORE_ADJ_MAX); i++) {
            procadjslot_list[i].next = &procadjslot_list[i];
            procadjslot_list[i].prev = &procadjslot_list[i];
        }
    
        return 0;
    }
    

    lmkd的init方法中做的工作

    1. 获取lmkd的socket描述符
    2. 创建epoll来监听socket的连接,如果有连接则回调ctrl_connect_handler方法来处理。
    3. 检测是否有minfree接口,即内核是否支持lowmemorykiller,如果内核不支持则调用init_mp_common初始化,在lmkd中实现进程查杀。

    为了防止剩余内存过低,Android在内核空间有lowmemorykiller(简称LMK),LMK是通过注册shrinker来触发低内存回收的,这个机制并不太优雅,可能会拖慢Shrinkers内存扫描速度,已从内核4.12中移除,后续会采用用户空间的LMKD + memory cgroups机制

    我们先分析内核实现的LowMemoryKiller进程查杀机制, 然后再分析lmkd实现的机制。两者最终的结果都是在内存紧张的时候杀死一些进程来释放内存, 但是实现机制去不太一样。

    static void mainloop(void) {
        struct event_handler_info* handler_info;
        struct epoll_event *evt;
    
        //循环等待epoll事件的上报
        while (1) {
            struct epoll_event events[maxevents];
            int nevents;
            int i;
    
            nevents = epoll_wait(epollfd, events, maxevents, -1);
    
            if (nevents == -1) {
                if (errno == EINTR)
                    continue;
                ALOGE("epoll_wait failed (errno=%d)", errno);
                continue;
            }
    
            //获取到对应的epoll事件,分发给对应的handler处理
            for (i = 0, evt = &events[0]; i < nevents; ++i, evt++) {
                if (evt->events & EPOLLERR)
                    ALOGD("EPOLLERR on event #%d", i);
                if (evt->events & EPOLLHUP) {
                    /* This case was handled in the first pass */
                    continue;
                }
                if (evt->data.ptr) {
                    handler_info = (struct event_handler_info*)evt->data.ptr;
                    handler_info->handler(handler_info->data, evt->events);
                }
            }
        }
    }
    

    init执行初始化完成之后, 进入mainloop方法,循环等待epoll事件的上报,init的时候epoll监听的socket连接, 当有socket连接的时候就会调用ctrl_connect_handler方法。

    static void ctrl_connect_handler(int data __unused, uint32_t events __unused) {
        struct epoll_event epev;
        int free_dscock_idx = get_free_dsock();
    
        if (free_dscock_idx < 0) {
            /*
             * Number of data connections exceeded max supported. This should not
             * happen but if it does we drop all existing connections and accept
             * the new one. This prevents inactive connections from monopolizing
             * data socket and if we drop ActivityManager connection it will
             * immediately reconnect.
             */
            for (int i = 0; i < MAX_DATA_CONN; i++) {
                ctrl_data_close(i);
            }
            free_dscock_idx = 0;
        }
        //接受framework的socket连接
        data_sock[free_dscock_idx].sock = accept(ctrl_sock.sock, NULL, NULL);
        if (data_sock[free_dscock_idx].sock < 0) {
            ALOGE("lmkd control socket accept failed; errno=%d", errno);
            return;
        }
    
        ALOGI("lmkd data connection established");
        //监听连接的socket通信,当socket有消息的时候会掉ctrl_data_handler方法。
        data_sock[free_dscock_idx].handler_info.data = free_dscock_idx;
        data_sock[free_dscock_idx].handler_info.handler = ctrl_data_handler;
        epev.events = EPOLLIN;
        epev.data.ptr = (void *)&(data_sock[free_dscock_idx].handler_info);
        if (epoll_ctl(epollfd, EPOLL_CTL_ADD, data_sock[free_dscock_idx].sock, &epev) == -1) {
            ALOGE("epoll_ctl for data connection socket failed; errno=%d", errno);
            ctrl_data_close(free_dscock_idx);
            return;
        }
        maxevents++;
    }
    

    监听到socket连接, 我们知道此时连接lmkd的socket客户端就是framework,当有连接到来的时候accept方法返回连接的socketFD, 然后将连接的socketFD同样加入epoll中, 当socketFD中有可读消息,即framework给lmkd发送消息的时候,epoll唤醒然后会掉ctrl_data_handler方法来处理。

    Framework和lmkd通信

    Framework和lmkd进程通过socket来进行进程间通信,在lmkd初始化的时候,通过监听socket描述符lmkd来等待Framework发送的消息。
    Framework向lmkd发送命令相关的方法有三个。

    1. AMS.updateConfiguration
      更新配置,手机屏幕的尺寸和内存大小不一样,对应的最小内存阀值和adj值也不一样, 最终调用ProcessList的updateOomLevel方法向lmkd发送调整命令
    2. AMS.applyOomAdjLocked AMS根据一定的规则调整进程的adj值,最用通过ProcessList的setOomAdj方法发送给lmkd调整命令
    3. AMS.cleanUpApplicationRecordLocked & AMS.handleAppDiedLocked 进程死亡后,调用ProcessList的remove方法移除进程

    上面的三种情况Framework最终是通过socket向lmkd发送了三种消息。

        // LMK_TARGET <minfree> <minkillprio> ... (up to 6 pairs)
        // LMK_PROCPRIO <pid> <uid> <prio>
        // LMK_PROCREMOVE <pid>
        //调整minfree和adj的值
        static final byte LMK_TARGET = 0;
        //设置对应进程的adj值
        static final byte LMK_PROCPRIO = 1;
        //移除对应进程
        static final byte LMK_PROCREMOVE = 2;
    

    lmkd接收命令处理逻辑

    static void ctrl_command_handler(int dsock_idx) {
        LMKD_CTRL_PACKET packet;
        int len;
        enum lmk_cmd cmd;
        int nargs;
        int targets;
        //从socket中读取数据
        len = ctrl_data_read(dsock_idx, (char *)packet, CTRL_PACKET_MAX_SIZE);
        if (len <= 0)
            return;
    
        if (len < (int)sizeof(int)) {
            ALOGE("Wrong control socket read length len=%d", len);
            return;
        }
    
        //解析Socket的命令和参数
        cmd = lmkd_pack_get_cmd(packet);
        nargs = len / sizeof(int) - 1;
        if (nargs < 0)
            goto wronglen;
    
        switch(cmd) {
        case LMK_TARGET:
            targets = nargs / 2;
            if (nargs & 0x1 || targets > (int)ARRAY_SIZE(lowmem_adj))
                goto wronglen;
            //调整minfree和adj阀值
            cmd_target(targets, packet);
            break;
        case LMK_PROCPRIO:
            if (nargs != 3)
                goto wronglen;
            //设置对应进程的adj值
            cmd_procprio(packet);
            break;
        case LMK_PROCREMOVE:
            if (nargs != 1)
                goto wronglen;
            //移除对应的进程
            cmd_procremove(packet);
            break;
        default:
            ALOGE("Received unknown command code %d", cmd);
            return;
        }
    
        return;
    

    lmkd通过epoll监听socket中是否有数据, 当接受的framework发送的socket命令之后,调用ctrl_cmmand_handler方法处理,显示解析socket中的命令和参数,根据对于的命令来调用不同的方法处理。

    • cmd_target 调整最小内存阀值和adj值
    • cmd_procprio 调整进程的adj值
    • cmd_procremove 移除对应的进程

    对于进程查杀有两种实现方式,一种是内核的LMK,通过shrinker来触发低内存回收, 另一种是lmkd通过cgroup监控内存使用情况,自行计算杀掉进程。两种实现不太一样,需要逐个分析。

    内核LMK的实现

    cmd_target

    static void cmd_target(int ntargets, LMKD_CTRL_PACKET packet) {
        int i;
        struct lmk_target target;
        lowmem_targets_size = ntargets;
    
        //使用kernel中的LMK
        if (has_inkernel_module) {
            char minfreestr[128];
            char killpriostr[128];
    
            minfreestr[0] = '\0';
            killpriostr[0] = '\0';
    
            //将从framework收到的内存阀值和adj值封装成字符串,以,分隔
           //如 18432,23040,27648,32256,55296,80640 
           //0,100,200,300,900,906
            for (i = 0; i < lowmem_targets_size; i++) {
                char val[40];
               
                if (i) {
                    strlcat(minfreestr, ",", sizeof(minfreestr));
                    strlcat(killpriostr, ",", sizeof(killpriostr));
                }
    
                snprintf(val, sizeof(val), "%d", use_inkernel_interface ? lowmem_minfree[i] : 0);
                strlcat(minfreestr, val, sizeof(minfreestr));
                snprintf(val, sizeof(val), "%d", use_inkernel_interface ? lowmem_adj[i] : 0);
                strlcat(killpriostr, val, sizeof(killpriostr));
            }
            // 将字符串分别写入 /sys/module/lowmemorykiller/parameters/minfree
            // 和/sys/module/lowmemorykiller/parameters/adj
            writefilestring(INKERNEL_MINFREE_PATH, minfreestr);
            writefilestring(INKERNEL_ADJ_PATH, killpriostr);
        }
    }
    

    设置内存阀值和adj的值就是将从framework收到的数据封装成字符串,通过writefilestring写入到两个文件节点,以供内核LMK使用。
    /sys/module/lowmemorykiller/parameters/minfree : 内存级别限额
    /sys/module/lowmemorykiller/parameters/adj :内存级别限额对应的要杀掉的进程的adj值.

    cmd_procprio

    static void cmd_procprio(LMKD_CTRL_PACKET packet) {
        struct proc *procp;
        char path[80];
        char val[20];
        int soft_limit_mult;
        struct lmk_procprio params;
    
        //解析进程adj相关的参数
        lmkd_pack_get_procprio(packet, &params);
    
        if (params.oomadj < OOM_SCORE_ADJ_MIN ||
            params.oomadj > OOM_SCORE_ADJ_MAX) {
            ALOGE("Invalid PROCPRIO oomadj argument %d", params.oomadj);
            return;
        }
    
        //将进程优先级写入到/proc/进程id/oom_score_adj文件中
        snprintf(path, sizeof(path), "/proc/%d/oom_score_adj", params.pid);
        snprintf(val, sizeof(val), "%d", params.oomadj);
        writefilestring(path, val);
    
        if (use_inkernel_interface)
            return;
    }
    

    由于使用内核LMK, 所以调整进程优先级直接将优先级写入对应进程的oom_adj_score文件即可。

    cmd_procremove

    static void cmd_procremove(LMKD_CTRL_PACKET packet) {
        struct lmk_procremove params;
    
        //内核LMK,移除进程什么都不需要做,全部有内核处理
        if (use_inkernel_interface)
            return;
    }
    

    移除进程的时候不需要做任何操作

    内核LowMemoryKiller的实现原理

    在linux中,有一个名为kswapd的内核线程,当linux回收存放分页的时候,kswapd线程将会遍历一张shrinker链表,并执行回调,或者某个app启动,发现可用内存不足时,则内核会阻塞请求分配内存的进程分配内存的过程,并在该进程中去执行lowmemorykiller来释放内存。虽然之前没有接触过,大体的理解就是向系统注册了这个shrinker回调函数之后,当系统空闲内存页面不足时会调用这个回调函数。 struct shrinker的定义在linux/kernel/include/linux/shrinker.h中:
    内核LowMemoryKiller shrinker的注册过程如下:

    static struct shrinker lowmem_shrinker = {
        .scan_objects = lowmem_scan,
        .count_objects = lowmem_count,
        .seeks = DEFAULT_SEEKS * 16
    };
    
    static int __init lowmem_init(void)
    {
        register_shrinker(&lowmem_shrinker);
        return 0;
    }
    
    static void __exit lowmem_exit(void)
    {
        unregister_shrinker(&lowmem_shrinker);
    }
    

    注册完成之后, 在内存紧张的时候就会回调shrinker, 其中最主要的是lowmem_scan方法。具体实现如下:

    static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
    {
        struct task_struct *tsk;
            //选中要杀掉进程的task_struct
        struct task_struct *selected = NULL;
        unsigned long rem = 0;
        int tasksize;
        int I;
            //min_score_adj 初始值=1000
        short min_score_adj = OOM_SCORE_ADJ_MAX + 1;
        int minfree = 0;
            //要杀掉进程占用内存的大小
        int selected_tasksize = 0;
            //要杀掉进程的adj值
        short selected_oom_score_adj;
            //获取并计算剩余内存的大小
        int array_size = ARRAY_SIZE(lowmem_adj);
        int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
        int other_file = global_page_state(NR_FILE_PAGES) -
                            global_page_state(NR_SHMEM) -
                            total_swapcache_pages();
    
            //从minfree中和adj节点中获取的数据初始化内存限额和adj
        if (lowmem_adj_size < array_size)
            array_size = lowmem_adj_size;
        if (lowmem_minfree_size < array_size)
            array_size = lowmem_minfree_size;
            //  内存限额从小到大遍历, 找到剩余内存属于哪一个阀值,并获取它对应的adj值
        for (i = 0; i < array_size; i++) {
            minfree = lowmem_minfree[i];
            if (other_free < minfree && other_file < minfree) {
                min_score_adj = lowmem_adj[i];
                break;
            }
        }
    
        lowmem_print(3, "lowmem_scan %lu, %x, ofree %d %d, ma %hd\n",
                sc->nr_to_scan, sc->gfp_mask, other_free,
                other_file, min_score_adj);
    
            //如果adj值为初始值1000,则内存情况良好无需杀掉进程
        if (min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
            lowmem_print(5, "lowmem_scan %lu, %x, return 0\n",
                     sc->nr_to_scan, sc->gfp_mask);
            return 0;
        }
    
        selected_oom_score_adj = min_score_adj;
    
        rcu_read_lock();
            //遍历进程列表
        for_each_process(tsk) {
            struct task_struct *p;
            short oom_score_adj;
    
            if (tsk->flags & PF_KTHREAD)
                continue;
    
            p = find_lock_task_mm(tsk);
            if (!p)
                continue;
    
            if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
                time_before_eq(jiffies, lowmem_deathpending_timeout)) {
                task_unlock(p);
                rcu_read_unlock();
                return 0;
            }
                    //获取进程的adj值,如果进程的adj值小于目标adj值,则继续寻找
            oom_score_adj = p->signal->oom_score_adj;
            if (oom_score_adj < min_score_adj) {
                task_unlock(p);
                continue;
            }
                    //找到adj值大于目标adj的进程后,首先计算该进程占用的内存大小,
                   //如果两个进程的adj一般大,则找占用内存大的一个进程
            tasksize = get_mm_rss(p->mm);
            task_unlock(p);
            if (tasksize <= 0)
                continue;
            if (selected) {
                if (oom_score_adj < selected_oom_score_adj)
                    continue;
                if (oom_score_adj == selected_oom_score_adj &&
                    tasksize <= selected_tasksize)
                    continue;
            }
            selected = p;
            selected_tasksize = tasksize;
            selected_oom_score_adj = oom_score_adj;
            lowmem_print(2, "select '%s' (%d), adj %hd, size %d, to kill\n",
                     p->comm, p->pid, oom_score_adj, tasksize);
        }
           //将找到的进程发送SIGKILL杀掉该进程从而释放内存
        if (selected) {
            long cache_size = other_file * (long)(PAGE_SIZE / 1024);
            long cache_limit = minfree * (long)(PAGE_SIZE / 1024);
            long free = other_free * (long)(PAGE_SIZE / 1024);
            trace_lowmemory_kill(selected, cache_size, cache_limit, free);
            lowmem_print(1, "Killing '%s' (%d), adj %hd,\n" \
                    "   to free %ldkB on behalf of '%s' (%d) because\n" \
                    "   cache %ldkB is below limit %ldkB for oom_score_adj %hd\n" \
                    "   Free memory is %ldkB above reserved\n",
                     selected->comm, selected->pid,
                     selected_oom_score_adj,
                     selected_tasksize * (long)(PAGE_SIZE / 1024),
                     current->comm, current->pid,
                     cache_size, cache_limit,
                     min_score_adj,
                     free);
            lowmem_deathpending_timeout = jiffies + HZ;
            set_tsk_thread_flag(selected, TIF_MEMDIE);
            send_sig(SIGKILL, selected, 0);
            rem += selected_tasksize;
        }
    
        lowmem_print(4, "lowmem_scan %lu, %x, return %lu\n",
                 sc->nr_to_scan, sc->gfp_mask, rem);
        rcu_read_unlock();
        return rem;
    }
    
    

    内核LMK的原理很简单:首先注册了shrinker,在内存紧张的时候会触发lowmem_scan方法,这个方法要做的就是找打一个进程,然后杀掉他,释放一些内存。

    1. 获取剩余内存的大小,和Minfree内存阀值做比较,找到对应的内存阀值,找到对应的adj值。
    2. 遍历所有的进程,大于该adj的值的进程是要杀掉的目标进程, 但是并不是全部杀掉,而是找到adj最大的进程杀掉,如果最大adj有多个相同adj进程,则杀掉占用内存最大的一个

    内核LMK的实现逻辑已经分析完了

    lmkd内存查杀的实现

    lmkd实现内存查实的方式是基于cgroup memory来实现的。
    什么是cgroup memory?
    Cgroup的memory子系统,即memory cgroup(本文以下简称memcg),提供了对系统中一组进程的内存行为的管理,从而对整个系统中对内存有不用需求的进程或应用程序区分管理,实现更有效的资源利用和隔离。

    cgroup memory相关的文件

    cgroup.event_control       #用于eventfd的接口
     memory.usage_in_bytes      #显示当前已用的内存
     memory.limit_in_bytes      #设置/显示当前限制的内存额度
     memory.failcnt             #显示内存使用量达到限制值的次数
     memory.max_usage_in_bytes  #历史内存最大使用量
     memory.soft_limit_in_bytes #设置/显示当前限制的内存软额度
     memory.stat                #显示当前cgroup的内存使用情况
     memory.use_hierarchy       #设置/显示是否将子cgroup的内存使用情况统计到当前cgroup里面
     memory.force_empty         #触发系统立即尽可能的回收当前cgroup中可以回收的内存
     memory.pressure_level      #设置内存压力的通知事件,配合cgroup.event_control一起使用
     memory.swappiness          #设置和显示当前的swappiness
     memory.move_charge_at_immigrate #设置当进程移动到其他cgroup中时,它所占用的内存是否也随着移动过去
     memory.oom_control         #设置/显示oom controls相关的配置
     memory.numa_stat           #显示numa相关的内存
    

    简单的了解了下cgroup的原理,再来看lmkd的init方法

    static int init(void) {
        has_inkernel_module = !access(INKERNEL_MINFREE_PATH, W_OK);
        use_inkernel_interface = has_inkernel_module;
    
        if (use_inkernel_interface) {
            ALOGI("Using in-kernel low memory killer interface");
        } else {
        //如果没有使用内核LMK机制,则初始化memory pressure
            if (!init_mp_common(VMPRESS_LEVEL_LOW) ||
                !init_mp_common(VMPRESS_LEVEL_MEDIUM) ||
                !init_mp_common(VMPRESS_LEVEL_CRITICAL)) {
                ALOGE("Kernel does not support memory pressure events or in-kernel low memory killer");
                return -1;
            }
        }
    
        for (i = 0; i <= ADJTOSLOT(OOM_SCORE_ADJ_MAX); i++) {
            procadjslot_list[i].next = &procadjslot_list[i];
            procadjslot_list[i].prev = &procadjslot_list[i];
        }
    
        return 0;
    }
    
    static bool init_mp_common(enum vmpressure_level level) {
        int mpfd;
        int evfd;
        int evctlfd;
        char buf[256];
        struct epoll_event epev;
        int ret;
        int level_idx = (int)level;
        const char *levelstr = level_name[level_idx];
        
        //打开pressure_level的文件节点
        mpfd = open(MEMCG_SYSFS_PATH "memory.pressure_level", O_RDONLY | O_CLOEXEC);
        if (mpfd < 0) {
            ALOGI("No kernel memory.pressure_level support (errno=%d)", errno);
            goto err_open_mpfd;
        }
        //打开event_control的文件节点
        evctlfd = open(MEMCG_SYSFS_PATH "cgroup.event_control", O_WRONLY | O_CLOEXEC);
        if (evctlfd < 0) {
            ALOGI("No kernel memory cgroup event control (errno=%d)", errno);
            goto err_open_evctlfd;
        }
        //创建一个eventfd
        evfd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
        if (evfd < 0) {
            ALOGE("eventfd failed for level %s; errno=%d", levelstr, errno);
            goto err_eventfd;
        }
    
        //往cgroup.event_control中写入:<event_fd> <pressure_level_fd> <level>
        ret = snprintf(buf, sizeof(buf), "%d %d %s", evfd, mpfd, levelstr);
        if (ret >= (ssize_t)sizeof(buf)) {
            ALOGE("cgroup.event_control line overflow for level %s", levelstr);
            goto err;
        }
    
        ret = TEMP_FAILURE_RETRY(write(evctlfd, buf, strlen(buf) + 1));
        if (ret == -1) {
            ALOGE("cgroup.event_control write failed for level %s; errno=%d",
                  levelstr, errno);
            goto err;
        }
        //然后使用epoll监听evfd, 等待memory pressure_level的事件通知
        epev.events = EPOLLIN;
        /* use data to store event level */
        vmpressure_hinfo[level_idx].data = level_idx;
        vmpressure_hinfo[level_idx].handler = mp_event_common;
        epev.data.ptr = (void *)&vmpressure_hinfo[level_idx];
        ret = epoll_ctl(epollfd, EPOLL_CTL_ADD, evfd, &epev);
        if (ret == -1) {
            ALOGE("epoll_ctl for level %s failed; errno=%d", levelstr, errno);
            goto err;
        }
        maxevents++;
        mpevfd[level] = evfd;
        close(evctlfd);
        return true;
    
    err:
        close(evfd);
    err_eventfd:
        close(evctlfd);
    err_open_evctlfd:
        close(mpfd);
    err_open_mpfd:
        return false;
    }
    

    先了解下memory pressure_level的用法

    memory.pressure_level
    这个文件主要用来监控当前cgroup的内存压力,当内存压力大时(即已使用内存快达到设置的限额),在分配内存之前需要先回收部分内存,从而影响内存分配速度,影响性能,而通过监控当前cgroup的内存压力,可以在有压力的时候采取一定的行动来改善当前cgroup的性能,比如关闭当前cgroup中不重要的服务等。目前有三种压力水平:
    low
    意味着系统在开始为当前cgroup分配内存之前,需要先回收内存中的数据了,这时候回收的是在磁盘上有对应文件的内存数据。
    medium
    意味着系统已经开始频繁为当前cgroup使用交换空间了。
    critical
    快撑不住了,系统随时有可能kill掉cgroup中的进程。

    如何配置相关的监听事件呢?和memory.oom_control类似,大概步骤如下:

    利用函数eventfd(2)创建一个event_fd
    打开文件memory.pressure_level,得到pressure_level_fd
    往cgroup.event_control中写入这么一串:<event_fd> <pressure_level_fd> <level>
    然后通过读event_fd得到通知

    init_mp_common方法严格的按照pressure_level的用法,注册了pressure_level的事件回调, pressure_level分为三个等级

    static const char *level_name[] = {
    "low",
    "medium",
    "critical"
    };

    当内存达到相应的等级,就会回调mp_event_common方法, 由mp_event_common方法来处理。

    static void mp_event_common(int data, uint32_t events __unused) {
        int ret;
        unsigned long long evcount;
        int64_t mem_usage, memsw_usage;
        int64_t mem_pressure;
        enum vmpressure_level lvl;
        union meminfo mi;
        union zoneinfo zi;
        static struct timeval last_report_tm;
        static unsigned long skip_count = 0;
        enum vmpressure_level level = (enum vmpressure_level)data;
        long other_free = 0, other_file = 0;
        int min_score_adj;
        int pages_to_free = 0;
        int minfree = 0;
        static struct reread_data mem_usage_file_data = {
            .filename = MEMCG_MEMORY_USAGE,
            .fd = -1,
        };
        static struct reread_data memsw_usage_file_data = {
            .filename = MEMCG_MEMORYSW_USAGE,
            .fd = -1,
        };
    
        //检查触发该方法压力类型,如果大于一种类型出发,则选择压力大的一个
        for (lvl = VMPRESS_LEVEL_LOW; lvl < VMPRESS_LEVEL_COUNT; lvl++) {
            if (mpevfd[lvl] != -1 &&
                TEMP_FAILURE_RETRY(read(mpevfd[lvl],
                                   &evcount, sizeof(evcount))) > 0 &&
                evcount > 0 && lvl > level) {
                level = lvl;
            }
        }
    
        //获取剩余内存的大小
        if (meminfo_parse(&mi) < 0 || zoneinfo_parse(&zi) < 0) {
            ALOGE("Failed to get free memory!");
            return;
        }
     
        //使用设置的最小内存阀值
        if (use_minfree_levels) {
            int i;
    
            //获取剩余内存大小
            other_free = mi.field.nr_free_pages - zi.field.totalreserve_pages;
            if (mi.field.nr_file_pages > (mi.field.shmem + mi.field.unevictable + mi.field.swap_cached)) {
                other_file = (mi.field.nr_file_pages - mi.field.shmem -
                              mi.field.unevictable - mi.field.swap_cached);
            } else {
                other_file = 0;
            }
            //根据剩余内存大小,找到对应的内存阀值及adj的值
            min_score_adj = OOM_SCORE_ADJ_MAX + 1;
            for (i = 0; i < lowmem_targets_size; i++) {
                minfree = lowmem_minfree[i];
                if (other_free < minfree && other_file < minfree) {
                    min_score_adj = lowmem_adj[i];
                    break;
                }
            }
    
           //如果得到的adj值=1000, 则表示内存状况良好,无需查杀内存直接返回
            if (min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
                if (debug_process_killing) {
                    ALOGI("Ignore %s memory pressure event "
                          "(free memory=%ldkB, cache=%ldkB, limit=%ldkB)",
                          level_name[level], other_free * page_k, other_file * page_k,
                          (long)lowmem_minfree[lowmem_targets_size - 1] * page_k);
                }
                return;
            }
             //计算要达到最大内存阀值情况下需要释放的内存大小
            /* Free up enough pages to push over the highest minfree level */
            pages_to_free = lowmem_minfree[lowmem_targets_size - 1] -
                ((other_free < other_file) ? other_free : other_file);
            goto do_kill;
        }
    
        if (level == VMPRESS_LEVEL_LOW) {
            record_low_pressure_levels(&mi);
        }
       //内存状况良好, 无需查杀,直接退出
        if (level_oomadj[level] > OOM_SCORE_ADJ_MAX) {
            /* Do not monitor this pressure level */
            return;
        }
    
        if ((mem_usage = get_memory_usage(&mem_usage_file_data)) < 0) {
            goto do_kill;
        }
        if ((memsw_usage = get_memory_usage(&memsw_usage_file_data)) < 0) {
            goto do_kill;
        }
    
        // Calculate percent for swappinness.
        mem_pressure = (mem_usage * 100) / memsw_usage;
    
        //根据一定的规则,计算出压力level
        if (enable_pressure_upgrade && level != VMPRESS_LEVEL_CRITICAL) {
            // We are swapping too much.
            if (mem_pressure < upgrade_pressure) {
                level = upgrade_level(level);
                if (debug_process_killing) {
                    ALOGI("Event upgraded to %s", level_name[level]);
                }
            }
        }
    
        // If the pressure is larger than downgrade_pressure lmk will not
        // kill any process, since enough memory is available.
        if (mem_pressure > downgrade_pressure) {
            if (debug_process_killing) {
                ALOGI("Ignore %s memory pressure", level_name[level]);
            }
            return;
        } else if (level == VMPRESS_LEVEL_CRITICAL &&
                   mem_pressure > upgrade_pressure) {
            if (debug_process_killing) {
                ALOGI("Downgrade critical memory pressure");
            }
            // Downgrade event, since enough memory available.
            level = downgrade_level(level);
        }
    
    do_kill:
        if (low_ram_device) {
            //只杀掉一个进程
            if (find_and_kill_processes(level, level_oomadj[level], 0) == 0) {
                if (debug_process_killing) {
                    ALOGI("Nothing to kill");
                }
            }
        } else {
            int pages_freed;
            //如果不使用设置的minfree内存阀值,则根据计算level对应的adj来查杀内存
            if (!use_minfree_levels) {
                /* If pressure level is less than critical and enough free swap then ignore */
                if (level < VMPRESS_LEVEL_CRITICAL &&
                    mi.field.free_swap > low_pressure_mem.max_nr_free_pages) {
                    if (debug_process_killing) {
                        ALOGI("Ignoring pressure since %" PRId64
                              " swap pages are available ",
                              mi.field.free_swap);
                    }
                    return;
                }
                //计算出需要释放的内存大小
                if (mi.field.nr_free_pages < low_pressure_mem.max_nr_free_pages) {
                    pages_to_free = low_pressure_mem.max_nr_free_pages -
                        mi.field.nr_free_pages;
                } else {
                    if (debug_process_killing) {
                        ALOGI("Ignoring pressure since more memory is "
                            "available (%" PRId64 ") than watermark (%" PRId64 ")",
                            mi.field.nr_free_pages, low_pressure_mem.max_nr_free_pages);
                    }
                    return;
                }
                min_score_adj = level_oomadj[level];
            }
    
           //根据min_score_adj和page_to_free来杀掉进程
           //大于min_score_adj的进程都属于目标进程,循环查杀,并累计杀掉线程释放内存的大小
           //当释放的内存大于page_to_free的大小的时候,就可以停止了
            pages_freed = find_and_kill_processes(level, min_score_adj, pages_to_free);
    
            if (use_minfree_levels) {
                ALOGI("Killing because cache %ldkB is below "
                      "limit %ldkB for oom_adj %d\n"
                      "   Free memory is %ldkB %s reserved",
                      other_file * page_k, minfree * page_k, min_score_adj,
                      other_free * page_k, other_free >= 0 ? "above" : "below");
            }
    
            if (pages_freed < pages_to_free) {
                ALOGI("Unable to free enough memory (pages to free=%d, pages freed=%d)",
                      pages_to_free, pages_freed);
            } else {
                ALOGI("Reclaimed enough memory (pages to free=%d, pages freed=%d)",
                      pages_to_free, pages_freed);
                gettimeofday(&last_report_tm, NULL);
            }
        }
    }
    

    lmkd内存查杀原理:

    1. 当内存压力出发该方法的时候,读取当前的内存压力类型
    2. 获取当前剩余内存的大小
    3. 根据剩余内存计算要杀掉进程的adj值,以及需要释放内存的大小
      分为两种情况:
      a: 如果使用的是设置的minfree和adj值,则根据剩余内存大小找到对应的adj,和需要释放内存大小
      b:如果不使用设置的内存阀值,则根据当前压力类型计算adj,以及达到上一个等级需要释放内存大小
    4. 根据以上计算的adj和需要释放内存大小来查杀内存,从大于adj的进程中开始查杀进程并释放内存,当释放内存大小达到需求 就可以停止查杀内存了。

    总结

    进程查杀的两种实现方式原理类似,都是注册是的回调,当内存紧张的时候根据剩余内存的adj来查杀大于该adj的内存。内核shrinker方式是只有内存紧张的时候才会去释放,而cgroup方式控制更加精细, 根据不同等级来触发内存回收。

    相关文章

      网友评论

          本文标题:LowMemoryKiller机制分析

          本文链接:https://www.haomeiwen.com/subject/tzndvctx.html