常用的查看app运行内存消耗的有adb shell procrank

lgy@lgy-PC:~$ adb shell procrank|grep dialer
19164  1646272K   68160K    8043K    6080K    9304K     284K  com.android.dialer

和adb shell dumpsys meminfo

lgy@lgy-PC:~$ adb shell dumpsys meminfo com.android.dialer
Applications Memory Usage (kB):
Uptime: 15528616 Realtime: 19602028

** MEMINFO in pid 19164 [com.android.dialer] **
                   Pss  Private  Private  Swapped     Heap     Heap     Heap
                 Total    Dirty    Clean    Dirty     Size    Alloc     Free
                ------   ------   ------   ------   ------   ------   ------
  Native Heap     2062     1384        0     3796    28928    27248     1679
  Dalvik Heap      885      848        0       20    17848    15944     1904
 Dalvik Other     1012      716        0     4572                           
        Stack      140      140        0        0                           
    Other dev        4        0        4        0                           
     .so mmap      286       68        0      796                           
    .apk mmap      645        0      620        0                           
    .dex mmap     2012        4     2008        4                           
    .oat mmap      316        0       24        0                           
    .art mmap     1032      440       24       76                           
   Other mmap       11        8        0        0                           
      Unknown       94       92        0       40                           
        TOTAL     8499     3700     2680     9304    46776    43192     3583
 
 App Summary
                       Pss(KB)
                        ------
           Java Heap:     1312
         Native Heap:     1384
                Code:     2724
               Stack:      140
            Graphics:        0
       Private Other:      820
              System:     2119
 
               TOTAL:     8499      TOTAL SWAP (KB):     9304
 
 Objects
               Views:        0         ViewRootImpl:        0
         AppContexts:        1           Activities:        0
              Assets:        4        AssetManagers:        2
       Local Binders:        3        Proxy Binders:        7
       Parcel memory:        3         Parcel count:       14
    Death Recipients:        0      OpenSSL Sockets:        0
 
 SQL
         MEMORY_USED:      101
  PAGECACHE_OVERFLOW:       19          MALLOC_SIZE:       62
 
 DATABASES
      pgsz     dbsz   Lookaside(b)          cache  Dbname
         4       28             22         2/22/3  /data/user/0/com.android.dialer/databases/ararat

都可以显示pss大小，但是会发现这两个显示的大小不一样，meminfo是8499，procrank显示的是8043。这两个为啥会不一样呢，哪个比较准确？

meminfo流程

dumsys程序不做分析，可以理解为它会调用xx服务的dump方法打印相关信息。meminfo服务在ActivityManagerService中添加

frameworks/base/services/core/java/com/android/server/am/ActivityManagerService.java

    public void setSystemProcess() {         
            ...
            ServiceManager.addService("meminfo", new MemBinder(this));
            ...
    }

服务代码实现是MemBinder，一个内部类

    static class MemBinder extends Binder {
        ActivityManagerService mActivityManagerService;
        MemBinder(ActivityManagerService activityManagerService) {
            mActivityManagerService = activityManagerService;
        }

        @Override
        protected void dump(FileDescriptor fd, PrintWriter pw, String[] args) {
            if (mActivityManagerService.checkCallingPermission(android.Manifest.permission.DUMP)
                    != PackageManager.PERMISSION_GRANTED) {
                pw.println("Permission Denial: can't dump meminfo from from pid="
                        + Binder.getCallingPid() + ", uid=" + Binder.getCallingUid()
                        + " without permission " + android.Manifest.permission.DUMP);
                return;
            }

            mActivityManagerService.dumpApplicationMemoryUsage(fd, pw, "  ", args, false, null);
        }
    }

dump就是实现打印信息的方法，先检测权限，然后打印相关信息

 final void dumpApplicationMemoryUsage(FileDescriptor fd,
            PrintWriter pw, String prefix, String[] args, boolean brief, PrintWriter categoryPw) {     
        ...
        ArrayList<ProcessRecord> procs = collectProcesses(pw, opti, packages, args);
        ...
    load_maps(pid, stats);
        for (int i = procs.size() - 1 ; i >= 0 ; i--) {
            ...
                if (mi == null) {
                    mi = new Debug.MemoryInfo();
                }
                if (dumpDetails || (!brief && !oomOnly)) {
                    Debug.getMemoryInfo(pid, mi);
                } else {
                    mi.dalvikPss = (int)Debug.getPss(pid, tmpLong, null);
                    mi.dalvikPrivateDirty = (int)tmpLong[0];
                }
            ...
        }
        ...
    }

主体是个循环，打印每个进程的信息，信息类是MemoryInfo，这个类定义在Debug中，而且获取信息内容的方法同样是在Debug中

frameworks/base/core/java/android/os/Debug.java

public static native void getMemoryInfo(int pid, MemoryInfo memoryInfo);

从native中看到这个类只是个跳转，真正的实现在C++代码中

frameworks/base/core/jni/android_os_Debug.cpp

    { "getMemoryInfo",          "(ILandroid/os/Debug$MemoryInfo;)V",
            (void*) android_os_Debug_getDirtyPagesPid },

从jni相关代码中看出对应方法是android_os_Debug_getDirtyPagesPid

static void android_os_Debug_getDirtyPagesPid(JNIEnv *env, jobject clazz,
        jint pid, jobject object)
{
    stats_t stats[_NUM_HEAP];
    memset(&stats, 0, sizeof(stats));

    load_maps(pid, stats);
    ...
}

stats是用于保存返回结果的，获取的代码就在load_maps中

static void load_maps(int pid, stats_t* stats)
{
    char tmp[128];
    FILE *fp;

    sprintf(tmp, "/proc/%d/smaps", pid);
    fp = fopen(tmp, "r");
    if (fp == 0) return;

    read_mapinfo(fp, stats);
    fclose(fp);
}

和名字一样，就是读取proc/（进程pid）/smaps文件，该文件片段如下

12c00000-12e0a000 rw-p 00000000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

Size:               2088 kB

Rss:                1864 kB

Pss:                1864 kB

Shared_Clean:          0 kB

Shared_Dirty:          0 kB

Private_Clean:         0 kB

Private_Dirty:      1864 kB

Referenced:         1864 kB

Anonymous:          1864 kB

AnonHugePages:         0 kB

Swap:                  0 kB

PSwap:                 0 kB

KernelPageSize:        4 kB

MMUPageSize:           4 kB

Locked:                0 kB

VmFlags: rd wr mr mw me ac 

12e0a000-133fe000 rw-p 0020a000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

Size:               6096 kB

Rss:                5900 kB

Pss:                5900 kB

Shared_Clean:          0 kB

Shared_Dirty:          0 kB

Private_Clean:         0 kB

Private_Dirty:      5900 kB

Referenced:         5900 kB

Anonymous:          5900 kB

AnonHugePages:         0 kB

Swap:                  0 kB

PSwap:                 0 kB

KernelPageSize:        4 kB

MMUPageSize:           4 kB

Locked:                0 kB

VmFlags: rd wr mr mw me ac 

133fe000-22c00000 ---p 007fe000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

Size:             253960 kB

Rss:                   0 kB

Pss:                   0 kB

Shared_Clean:          0 kB

Shared_Dirty:          0 kB

Private_Clean:         0 kB

Private_Dirty:         0 kB

Referenced:            0 kB

Anonymous:             0 kB

AnonHugePages:         0 kB

Swap:                  0 kB

PSwap:                 0 kB

KernelPageSize:        4 kB

MMUPageSize:           4 kB

Locked:                0 kB

VmFlags: mr mw me ac 

它详细的描述了内存中各个部分的信息。read_mapinfo方法就是逐行扫描这个文件，然后统计出信息。

procrank流程

/system/extras/procrank/procrank.c

procrank程序就一个源码文件

int main(int argc, char *argv[]) {
   ...
   for (i = 0; i < num_procs; i++) {
     ...
     error = pm_process_create(ker, pids[i], &proc);
     ...
   }
   ...
}

代码的主体还是个循环，通过pm_process_create方法获取相关信息，该方法实现在pm_process.c中

system/extras/libpagemap/pm_process.c

int pm_process_create(pm_kernel_t *ker, pid_t pid, pm_process_t **proc_out) {
   ...
   error = read_maps(proc);
   ...
}

static int read_maps(pm_process_t *proc) {
   ...
   error = snprintf(filename, MAX_FILENAME, "/proc/%d/maps", proc->pid);
   ...
}

可以看出最后和meminfo类似，从maps中获取信息，maps的片段如下

12c00000-12e0a000 rw-p 00000000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

12e0a000-135fe000 rw-p 0020a000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

135fe000-22c00000 ---p 009fe000 00:04 470401                             /dev/ashmem/dalvik-main space (deleted)

23807000-23827000 rw-p 00000000 00:00 0 

23827000-23a07000 ---p 00000000 00:00 0 

32c00000-32c01000 rw-p 00000000 00:04 470402                             /dev/ashmem/dalvik-main space 1 (deleted)

32c01000-42c00000 ---p 00001000 00:04 470402                             /dev/ashmem/dalvik-main space 1 (deleted)

43fd4000-44000000 ---p 00000000 00:00 0 

44000000-44020000 rw-p 00000000 00:00 0 

44020000-44034000 ---p 00000000 00:00 0 

45500000-45600000 rw-p 00000000 00:00 0 

4d900000-4da00000 rw-p 00000000 00:00 0 

52000000-52200000 rw-p 00000000 00:00 0 

52200000-53000000 ---p 00000000 00:00 0 

53000000-53200000 rw-p 00000000 00:00 0 

53200000-54000000 ---p 00000000 00:00 0 

56e00000-56f00000 rw-p 00000000 00:00 0 

5ba00000-5baca000 rw-p 00000000 00:00 0 

708e6000-717ba000 rw-p 00000000 fd:00 1368072                            /data/dalvik-cache/arm64/system@[email protected]

717ba000-73d4f000 r--p 00000000 fd:00 1368070                            /data/dalvik-cache/arm64/system@[email protected]

73d4f000-765f3000 r-xp 02595000 fd:00 1368070                            /data/dalvik-cache/arm64/system@[email protected]

765f3000-765f4000 rw-p 04e39000 fd:00 1368070                            /data/dalvik-cache/arm64/system@[email protected]

765f4000-7675d000 rw-p 00000000 00:04 470400                             /dev/ashmem/dalvik-zygote space (deleted)

7675d000-7675e000 rw-p 00000000 00:04 468941                             /dev/ashmem/dalvik-non moving space (deleted)

可以看出信息内容比smaps简略很多

总结

meminfo从proc/（进程pid）/smaps文件获取信息，而procrank从proc/（进程pid）/maps文件获取信息，由于smap比maps信息详细，所以meminfo得到的pss比procrank的pss精确。