pmap學習：系統測試中怎麼肯定內存泄露（memory leak)

pmap is a good tool to report the memory map of a process. But how to make sure a process has memory leak, here is the tips:

1, if the virtual memory (VIRT) or writeable/private (‘pmap –d’ output) keeps increasing stably when you repeat the same operation, it should be have memory leak.

2, The RSS is just for your reference, it can’t used as the factor to detect memory leak.

3, although you call ‘free’ or ‘delete’, the virtual memory may not shrink immediately.

4, the same new operation may mapped to different [anon] virtual memory space. Here 1M was alloated to000000001b56a000    and00002ac25a77c000.   

000000001b56a000   1156      12      12 rw---   [ anon ]

00002ac25a77c000   1040      16      16 rw---   [ anon ]

Test Code:

#include "stdio.h"
int global_i_init=0;
static int static_global_j=11;
int global_k;

int main () {
        int tmp;
        printf ("global_i_init: 0x%lx\n",&global_i_init );
        printf ("global_k non_init : 0x%lx\n",&global_k );
        printf ("static_global_j  : 0x%lx\n",&static_global_j );
        printf ("stack i  : 0x%lx\n",&tmp );

        getchar (); // step2
        char *x= new char[100*1024];
        printf ("data allocate 100k @0x%lx\n", x);
        getchar();  // step3
        char *x2= new char[1024*1024];
        printf ("data allocate 1M i@0x%lx\n", x2);

        getchar(); // step4
        printf ("data write at x2[1024*1024-1] ");
        x2[1024*1024-1]=0;

        getchar(); // step5
        printf ("data read at x2 ");
        int j;
        for (int i=0;i<1024*1024; i++)
                j=x2[i];
        getchar (); // step6
        printf ("delete x2");
        delete x2;

        getchar ();  //step 7
        printf ("delete x");
        delete x;

        getchar();  //step8

       x2= new char[1024*1024];
        printf ("data allocate 1M i@0x%lx\n", x2);

        getchar ();
}

Program output:

ATCA48-0-0-1:/root-# ./a.out
global_i_init: 0x500cc0
global_k non_init : 0x500cc4
static_global_j  : 0x500cb8
stack i  : 0x7fff0120ab5c
data allocate 100k @0x1b56a010
data allocate 1M i@0x2ac25a77f010
data write at x2[1024*1024-1]
data read at x2
delete x2
delete x
data allocate 1M i@0x1b56a010

最後一行的值
mapped 表示該進程映射的虛擬地址空間大小，也就是該進程預先分配的虛擬內存大小，即ps出的vsz
writeable/private  表示進程所佔用的私有地址空間大小，也就是該進程實際使用的內存大小      
shared 表示進程和其餘進程共享的內存大小

pmap -d output:

		mapped	writeable/private:	shared
STEP 1	init	19616K	236K	0
STEP 2	new 100k	19848K	468K	0
STEP 3	new 1M	20876K	1496K	0
STEP 4	write 1 byte	20876K	1496K	0
STEP 5	read 1M	20876K	1496K	0
STEP 6	delete 1M	19848K	468K(沒有變化)	0
STEP 7	delete 100k	19848K	468K（沒有變化）	0
STEP 8	new 1M	20772K	1392K	0

pmap -d conclusion:

1, mapped and writeable/private can reflect the memory variation.

     mapped 和writeable/private 可以反映內存的變化.

2, delete or free may not decrease mapped or writeable/private immediately.

 delete 和free 不能在 mapped 和writeable/private 當即反映出來，比方刪除100k就沒有變化

pmap -x output: 

		Total kbytes	RSS	Dirty
STEP 1	init	19616	912	124
STEP 2	new 100k	19848	976	140
STEP 3	new 1M	20876	980	144
STEP 4	write 1 byte	20876	984	148
STEP 5	read 1M	20876	2004	148
STEP 6	delete 1M	19848	976	140
STEP 7	delete 100k	19848	980	140
STEP 8	new 1M	20772	984	144

		Address	kbytes	RSS	Dirty
STEP 1	init
STEP 2	new 100k	000000001b56a000	232	8	8
STEP 3	new 1M	000000001b56a000	232	8	8
STEP 4	write 1 byte	000000001b56a000	232	8	8
STEP 5	read 1M	000000001b56a000	232	8	8
STEP 6	delete 1M	000000001b56a000	232	8	8
STEP 7	delete 100k	000000001b56a000	232	8	8
STEP 8	new 1M	000000001b56a000	1156	12	12

 

		Address	kbytes	RSS	Dirty
STEP 1	init	00002ac25a77c000	12	12	12
STEP 2	new 100k	00002ac25a77c000	12	12	12
STEP 3	new 1M	00002ac25a77c000	1040	16	16
STEP 4	write 1 byte	00002ac25a77c000	1040	20	20
STEP 5	read 1M	00002ac25a77c000	1040	1040	20
STEP 6	delete 1M	00002ac25a77c000	12	12	12
STEP 7	delete 100k	00002ac25a77c000	12	12	12
STEP 8	new 1M	00002ac25a77c000	12	12	12

 

pmap -x conclusion:

pmap -x 結論

1, although you used new(step 3), the OS didn't commit the memory page until you read (STEP5)or write(STEP4) that memory space.

儘管你已經調用了 new分配內存(step 3), 可是 OS不必定一下把全部的page都分配。本例中只有對那一塊已經分配內存進行讀 (STEP5)或者寫(STEP4) 的時候才提交實際內存

pmap -x:

2, the new memory operation maybe happened on different Address range. like the first 1M allcation at00002ac25a77c000, but the 2nd 1M allcation happened at 000000001b56a000

 

Final pmap -d:

Address           Kbytes Mode  Offset           Device    Mapping
0000000000400000       4 r-x-- 0000000000000000 008:00001 a.out
0000000000500000       4 rw--- 0000000000000000 008:00001 a.out
000000001b56a000    1156 rw--- 000000001b56a000 000:00000   [ anon ]
0000003677000000     112 r-x-- 0000000000000000 008:00001 ld-2.5.so
000000367721c000       4 r---- 000000000001c000 008:00001 ld-2.5.so
000000367721d000       4 rw--- 000000000001d000 008:00001 ld-2.5.so
0000003677400000    1336 r-x-- 0000000000000000 008:00001 libc-2.5.so
000000367754e000    2048 ----- 000000000014e000 008:00001 libc-2.5.so
000000367774e000      16 r---- 000000000014e000 008:00001 libc-2.5.so
0000003677752000       4 rw--- 0000000000152000 008:00001 libc-2.5.so
0000003677753000      20 rw--- 0000003677753000 000:00000   [ anon ]
0000003677c00000     520 r-x-- 0000000000000000 008:00001 libm-2.5.so
0000003677c82000    2044 ----- 0000000000082000 008:00001 libm-2.5.so
0000003677e81000       4 r---- 0000000000081000 008:00001 libm-2.5.so
0000003677e82000       4 rw--- 0000000000082000 008:00001 libm-2.5.so
0000003678c00000      52 r-x-- 0000000000000000 008:00001 libgcc_s-4.1.2-20080825.so.1
0000003678c0d000    2048 ----- 000000000000d000 008:00001 libgcc_s-4.1.2-20080825.so.1
0000003678e0d000       4 rw--- 000000000000d000 008:00001 libgcc_s-4.1.2-20080825.so.1
0000003679400000     920 r-x-- 0000000000000000 008:00001 libstdc++.so.6.0.8
00000036794e6000    2044 ----- 00000000000e6000 008:00001 libstdc++.so.6.0.8
00000036796e5000      24 r---- 00000000000e5000 008:00001 libstdc++.so.6.0.8
00000036796eb000      12 rw--- 00000000000eb000 008:00001 libstdc++.so.6.0.8
00000036796ee000      72 rw--- 00000036796ee000 000:00000   [ anon ]
00002ac25a76f000      16 rw--- 00002ac25a76f000 000:00000   [ anon ]
00002ac25a77c000      12 rw--- 00002ac25a77c000 000:00000   [ anon ]
00007fff011f8000      84 rw--- 00007ffffffe9000 000:00000   [ stack ]
00007fff013c2000      12 r-x-- 00007fff013c2000 000:00000   [ anon ]
ffffffffff600000    8192 ----- 0000000000000000 000:00000   [ anon ]
mapped: 20772K    writeable/private: 1392K    shared: 0K

Final pmap -x

Address           Kbytes     RSS   Dirty Mode   Mapping
0000000000400000       4       4       0 r-x--  a.out
0000000000500000       4       4       4 rw---  a.out
000000001b56a000    1156      12      12 rw---    [ anon ]
0000003677000000     112      96       0 r-x--  ld-2.5.so
000000367721c000       4       4       4 r----  ld-2.5.so
000000367721d000       4       4       4 rw---  ld-2.5.so
0000003677400000    1336     284       0 r-x--  libc-2.5.so
000000367754e000    2048       0       0 -----  libc-2.5.so
000000367774e000      16      16       8 r----  libc-2.5.so
0000003677752000       4       4       4 rw---  libc-2.5.so
0000003677753000      20      16      16 rw---    [ anon ]
0000003677c00000     520      20       0 r-x--  libm-2.5.so
0000003677c82000    2044       0       0 -----  libm-2.5.so
0000003677e81000       4       4       4 r----  libm-2.5.so
0000003677e82000       4       4       4 rw---  libm-2.5.so
0000003678c00000      52      16       0 r-x--  libgcc_s-4.1.2-20080825.so.1
0000003678c0d000    2048       0       0 -----  libgcc_s-4.1.2-20080825.so.1
0000003678e0d000       4       4       4 rw---  libgcc_s-4.1.2-20080825.so.1
0000003679400000     920     404       0 r-x--  libstdc++.so.6.0.8
00000036794e6000    2044       0       0 -----  libstdc++.so.6.0.8
00000036796e5000      24      24      20 r----  libstdc++.so.6.0.8
00000036796eb000      12      12      12 rw---  libstdc++.so.6.0.8
00000036796ee000      72       8       8 rw---    [ anon ]
00002ac25a76f000      16      16      16 rw---    [ anon ]
00002ac25a77c000      12      12      12 rw---    [ anon ]
00007fff011f8000      84      12      12 rw---    [ stack ]
00007fff013c2000      12       4       0 r-x--    [ anon ]
ffffffffff600000    8192       0       0 -----    [ anon ]
----------------  ------  ------  ------
total kB           20772     984     144

  

3 ps和pamp比較

man ps 看看它們的含義：

rss       RSS    resident set size, the non-swapped physical memory that a task has used (in kiloBytes). (alias rssize, rsz).

vsz       VSZ    virtual memory size of the process in KiB (1024-byte units). Device mappings are currently excluded; this is subject to change. (alias vsize).

簡單一點說，RSS 就是這個process 實際佔用的物理內存，VSZ 就是process 的虛擬內存，就是process 如今沒有使用但將來可能會分配的內存大小。

其實這裏的ps 出來的結果，是有點不正確的，若是把全部程序的 RSS 加起來，恐怕比你的實際內存還要大呢。爲何呢？？由於 ps 的結果，RSS 那部分，是包括共享內存的。這裏我用 pmap 來看看。