//tls5源代码：
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include “pthread.h”
#include “TLS/Tlsconf.h”
#include “TLS/tls_api.h”
#define ERROR -1
#define OK 0
void test_fn2(int a)
{
char *psz = (char*)get_buf(ENUM_1, 32);
printf(“thread(%u) psz: %s\n”, pthread_self(), psz);
memset(psz, 0, 32);
sprintf(psz, “%u_%d”, pthread_self(), a);
return;
}
void test_fn1(int a)
{
int *reti = (int*)get_buf(ENUM_0, sizeof(int));
printf(“thread(%u) reti: %d\n”, pthread_self(), *reti);
(*reti)++;
test_fn2(a);
return;
}
void* test_fn_main1(void* arg)
{
int i = 0;
for (i = 0; i < 3; i++)
{
test_fn1(i);
sleep(1);
}
return;
}
int main()
{
int iRet = ERROR;
int i = 0;
pthread_t tid;
printf(“test start!\n”);
for (i = 0; i < 5; i++)
{
iRet = pthread_create(&tid, NULL, test_fn_main1, NULL);
if (OK != iRet)
{
printf(“pthread_create error!\n”);
return ERROR;
}
//pthread_detach(tid);
}
sleep(10);
return OK;
}
//生成可执行程序：tls库见上一篇
gcc -o tls5 tlstest5.c TLS/libtls.a -lpthread -lstdc++
//用valgrind进行内存泄露监测：
valgrind -v –leak-check=full –tool=memcheck ./tls5
提示信息：
[root@localhost 20130713]# valgrind -v –leak-check=full –tool=memcheck ./tls5
…………
==31803== malloc/free: in use at exit: 720 bytes in 5 blocks.
==31803== malloc/free: 40 allocs, 35 frees, 1,220 bytes allocated.
==31803==
==31803== searching for pointers to 5 not-freed blocks.
==31803== checked 52,519,768 bytes.
==31803==
==31803==
==31803== 720 bytes in 5 blocks are possibly lost in loss record 1 of 1
==31803== at 0x4004824: calloc (vg_replace_malloc.c:279)
==31803== by 0x840B39: _dl_allocate_tls (in /lib/ld-2.6.so)
==31803== by 0x9F1B35: pthread_create@@GLIBC_2.1 (in /lib/libpthread-2.6.so)
==31803== by 0x8048ACA: main (in /mnt/hgfs/D/work/test/20130713/tls5)
==31803==
==31803== LEAK SUMMARY:
==31803== definitely lost: 0 bytes in 0 blocks.
==31803== possibly lost: 720 bytes in 5 blocks.
==31803== still reachable: 0 bytes in 0 blocks.
==31803== suppressed: 0 bytes in 0 blocks.
…………
IN SUMMARY: 16 errors from 8 contexts (suppressed: 0 from 0)主要是库里面使用了未初始化的变量（Conditional jump or move depends on uninitialised value(s)），不管它。主要问题是在malloc/free: 40 allocs, 35 frees, 1,220 bytes allocated。
查看代码及程序运行打印，malloc与free都是一一对应的，怎会出现未释放的内存块？另外，程序共创建了5个线程，每个线程中有两次malloc操作，应该10 allocs，怎会是40allocs？排查步骤如下：
步骤一：
void* test_fn_main1(void* arg)
{
int i = 0;
for (i = 0; i < 3; i++)
{
//test_fn1(i); //注释掉该行不再调用malloc
sleep(1);
}
return;
}
步骤二：
[root@localhost 20130713]# gcc -o tls5 tlstest5.c TLS/libtls.a -lpthread -lstdc++
[root@localhost 20130713]# valgrind -v –leak-check=full –tool=memcheck ./tls5
………………
==2237== malloc/free: in use at exit: 720 bytes in 5 blocks.
==2237== malloc/free: 5 allocs, 0 frees, 720 bytes allocated.
==2237==
==2237== searching for pointers to 5 not-freed blocks.
==2237== checked 52,526,684 bytes.
==2237==
==2237==
==2237== 720 bytes in 5 blocks are possibly lost in loss record 1 of 1
==2237== at 0x4004824: calloc (vg_replace_malloc.c:279)
==2237== by 0x840B39: _dl_allocate_tls (in /lib/ld-2.6.so)
==2237== by 0x9F1B35: pthread_create@@GLIBC_2.1 (in /lib/libpthread-2.6.so)
==2237== by 0x8048ABF: main (in /mnt/hgfs/D/work/test/20130713/tls5)
==2237==
==2237== LEAK SUMMARY:
==2237== definitely lost: 0 bytes in 0 blocks.
==2237== possibly lost: 720 bytes in 5 blocks.
==2237== still reachable: 0 bytes in 0 blocks.
==2237== suppressed: 0 bytes in 0 blocks.
–2237– memcheck: sanity checks: 2 cheap, 1 expensive
…………
我们发现，程序中即使没有调用malloc，也存在5 allocs，纵观测试代码，唯一有可能alloc的地方就只有pthread_create了。
上网查了一下，在http://blog.csdn.net/jiqiren007/article/details/5959810找到答案。
线程在创建时都会分配一些内存（valgrind里面alloc次数是怎么算的，不是很清楚！），这些内存默认情况下需要在主线程中调用pthread_join进行释放；或者可以将子线程从父线程剥离，子线程终止时会回收这些内存资源。如果不使用pthread_join或者pthread_detach，线程结束时，pthread_create产生的内存将一直存在，直到整个进程结束为止，造成内存泄露。
在pthread_create后面加上pthread_detach(tid)，结果如下：
…………………………
==3206== malloc/free: in use at exit: 0 bytes in 0 blocks.
==3206== malloc/free: 40 allocs, 40 frees, 1,220 bytes allocated.
==3206==
==3206== All heap blocks were freed — no leaks are possible.
–3206– memcheck: sanity checks: 2 cheap, 1 expensive
–3206– memcheck: auxmaps: 0 auxmap entries (0k, 0M) in use
–3206– memcheck: auxmaps: 0 searches, 0 comparisons
……………………
问题解决！
从这个问题可以看出，如果大量地启动子线程，却不对子线程进行任何处理（调用pthread_join或者pthread_detach），最终有可能导致内存大量地隐秘地被蚕食。这个过程缓慢而又难以发掘，不过带来的问题却是极严重的。
