TLS是由于多线程编程带来的产物，主要是为了解决线程资源局部化，具体内容网上有很多介绍。有很多地方已经支持了该功能，但有些地方没有，下面是GCC的一些介绍，反正具体看实际使用情况：
5.51 Thread-Local Storage
=========================
Thread-local storage (TLS) is a mechanism by which variables are
allocated such that there is one instance of the variable per extant
thread. The run-time model GCC uses to implement this originates in
the IA-64 processor-specific ABI, but has since been migrated to other
processors as well. It requires significant support from the linker
(`ld'), dynamic linker (`ld.so'), and system libraries (`libc.so' and
`libpthread.so'), so it is not available everywhere.
At the user level, the extension is visible with a new storage class
keyword: `__thread'. For example:
__thread int i;
extern __thread struct state s;
static __thread char *p;
The `__thread' specifier may be used alone, with the `extern' or
`static' specifiers, but with no other storage class specifier. When
used with `extern' or `static', `__thread' must appear immediately
after the other storage class specifier.
The `__thread' specifier may be applied to any global, file-scoped
static, function-scoped static, or static data member of a class. It
may not be applied to block-scoped automatic or non-static data member.
When the address-of operator is applied to a thread-local variable, it
is evaluated at run-time and returns the address of the current thread's
instance of that variable. An address so obtained may be used by any
thread. When a thread terminates, any pointers to thread-local
variables in that thread become invalid.
No static initialization may refer to the address of a thread-local
variable.
In C++, if an initializer is present for a thread-local variable, it
must be a CONSTANT-EXPRESSION, as defined in 5.19.2 of the ANSI/ISO C++
standard.
See ELF Handling For Thread-Local Storage
(http://people.redhat.com/drepper/tls.pdf) for a detailed explanation of
the four thread-local storage addressing models, and how the run-time
is expected to function.
为了防止现有资源不支持TLS的情况，下面提供一种绕开__thread的一种实现。
//类文件Tls.h：
#ifndef __SAP_UTIL_TLS_H_
#define __SAP_UTIL_TLS_H_
#include <pthread.h>
#include "Tlsconf.h"
typedef struct pthread_atexit
{
int key; //线程局部变量标记
void (*free_fn)(void*); //线程结束时资源释放回调函数
void *arg; //线程局部变量地址
}pthread_atexit_t;
typedef std::list<pthread_atexit_t *> TlsList; //一个线程的线程局部变量构成一条链
class Tls
{
public:
Tls();
~Tls();
static char *pthread_atexit_get_buf(variable_key_t key, int len); //主要外部接口，获得线程局部变量
static int pthread_atexit_add(void *arg, variable_key_t key, void (*free_fn)(void*)); //新的线程局部变量使用该接口进行存储
static int pthread_atexit_remove(void *arg, variable_key_t key, void (*free_fn)(void*)); //删除特定的某个线程局部变量，暂时没啥用
protected:
static void pthread_atexit_done(void *arg); //线程结束回调函数，用于回收该线程所有线程局部变量资源
static void pthread_atexit_init(void); //设置线程结束时的回调函数
static void pthread_atexit_release(void *pbuf); //资源释放函数
static char *pthread_atexit_get_buf_from_list(TlsList *ptlslist, variable_key_t key); //从线程局部变量链表当中找到key值的变量
protected:
static pthread_key_t _pthread_atexit_key; //线程存储键值，每个线程通过它来读取数据链
static pthread_once_t _pthread_atexit_control_once; //初始化状态标志，pthread_once使用
};
#endif
#endif
//外部接口声明tls_api.h:
#ifndef _TLS_API_H_
#define _TLS_API_H_
#include "Tlsconf.h"
#ifdef __cplusplus
extern "C"
{
#endif
char* get_buf(variable_key_t key, int len);
#ifdef __cplusplus
}
#endif
#endif
//参数键值表，用于标记哪个函数中的哪个参数Tlsconf.h:
#ifndef _TLS_CONF_H
#define _TLS_CONF_H
typedef enum
{
ENUM_0,
ENUM_1,
ENUM_2,
ENUM_3,
ENUM_4,
ENUM_5,
ENUM_6,
ENUM_7,
}variable_key_t;
#endif
//类实现及外部接口Tls.cpp:
#include <sys/syscall.h>
#include <list>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "Tls.h"
#include "Tlsconf.h"
#include "tls_api.h"
using namespace std;
//#define gettid() syscall(__NR_gettid)
#define TLS_OUT_OF_INDEXES 0xffffffff
pthread_key_t Tls::_pthread_atexit_key = TLS_OUT_OF_INDEXES;
pthread_once_t Tls::_pthread_atexit_control_once = PTHREAD_ONCE_INIT;
Tls::Tls()
{
//_pthread_atexit_key = TLS_OUT_OF_INDEXES;
//_pthread_atexit_control_once = PTHREAD_ONCE_INIT;
}
Tls::~Tls()
{
}
char* Tls::pthread_atexit_get_buf_from_list(TlsList *ptlslist, variable_key_t key)
{
pthread_atexit_t *id_ptr=NULL;
if (NULL == ptlslist)
{
return NULL;
}
for (TlsList::iterator iter=ptlslist->begin(); iter !=ptlslist->end(); ++iter)
{
id_ptr = *iter;
if (id_ptr == NULL)
continue;
if (key == id_ptr->key) //通过key值区分变量
{
return (char*)(id_ptr->arg);
}
}
return NULL;
}
char* Tls::pthread_atexit_get_buf(variable_key_t key, int len)
{
TlsList* id_list;
char* ptrtmp = NULL;
int iret = -1;
if (len <= 0)
{
return NULL;
}
pthread_once(&_pthread_atexit_control_once, pthread_atexit_init); //该函数只调用一次pthread_atexit_init
if (_pthread_atexit_key == (pthread_key_t) TLS_OUT_OF_INDEXES)
{
printf("%s(%d): _pthread_atexit_key(%d) invalid
", __func__, __LINE__, _pthread_atexit_key);
return NULL;
}
id_list = (TlsList*) pthread_getspecific(_pthread_atexit_key);
ptrtmp = pthread_atexit_get_buf_from_list(id_list, key);
if (NULL != ptrtmp)
{
return ptrtmp;
}
ptrtmp = (char*)malloc(len);
if (NULL == ptrtmp)
{
return NULL;
}
memset(ptrtmp, 0, len);
iret = pthread_atexit_add(ptrtmp, key, pthread_atexit_release);
if (-1 == iret)
{
free(ptrtmp);
return NULL;
}
return ptrtmp;
}
void Tls::pthread_atexit_done(void *arg)
{
TlsList *id_list = (TlsList*) arg;
pthread_atexit_t *id_ptr=NULL;
//printf("invoke Tls::pthread_atexit_done(): tid=%ld
",gettid());
for(TlsList::iterator iter=id_list->begin(); iter !=id_list->end(); ++iter)
{
id_ptr = *iter;
if (id_ptr == NULL)
continue;
printf("pthread(%u) realease resouce %d!
", pthread_self(), id_ptr->key);
if (id_ptr->free_fn)
id_ptr->free_fn(id_ptr->arg);
delete id_ptr;
}
delete id_list;
}
void Tls::pthread_atexit_init(void)
{
pthread_key_create(&_pthread_atexit_key, pthread_atexit_done);
}
void Tls::pthread_atexit_release(void *pbuf)
{
if (NULL == pbuf)
{
return;
}
free((char*)pbuf);
return;
}
int Tls::pthread_atexit_add(void *arg, variable_key_t key, void (*free_fn)(void*))
{
const char *myname = "pthread_atexit_add";
pthread_atexit_t *id;
TlsList *id_list;
if (arg == NULL)
{
return 0;
}
id = new pthread_atexit_t;
if (id == NULL)
{
printf("%s(%d): new pthread_atexit_t error
", myname, __LINE__);
return -1;
}
id->key = key;
id->free_fn = free_fn;
id->arg = arg;
id_list = (TlsList*) pthread_getspecific(_pthread_atexit_key);
if (id_list == NULL)
{
id_list = new TlsList();
if (pthread_setspecific(_pthread_atexit_key, id_list) != 0)
{
printf("%s(%d): pthread_setspecific error, key(%d)
", myname, __LINE__, _pthread_atexit_key);
goto errExit;
}
}
id_list->push_back(id);
printf("pthread(%u) get resouce %d!
", pthread_self(), id->key);
return 0;
errExit:
if (id)
{
delete id;
}
if (id_list)
{
delete id_list;
}
return -1;
}
int Tls::pthread_atexit_remove(void *arg, variable_key_t key, void (*free_fn)(void*))
{
const char *myname = "pthread_atexit_remove";
TlsList *id_list;
if (arg == NULL)
{
return (-1);
}
if (_pthread_atexit_key == (pthread_key_t) TLS_OUT_OF_INDEXES)
{
printf("%s(%d): _pthread_atexit_key(%d) invalid
", myname, __LINE__, _pthread_atexit_key);
return (-1);
}
id_list = (TlsList*) pthread_getspecific(_pthread_atexit_key);
if (id_list == NULL)
{
printf("%s(%d): _pthread_atexit_key(%d) no exist in tid(%lu)
", myname, __LINE__, _pthread_atexit_key,(unsigned long) pthread_self());
return (-1);
}
pthread_atexit_t *id_ptr =NULL;
TlsList::iterator iter=id_list->begin();
for(; iter !=id_list->end(); ++iter)
{
id_ptr = *iter;
if (id_ptr == NULL)
continue;
if (id_ptr->free_fn == free_fn && id_ptr->arg == arg)
{
break;
}
}
if(id_ptr != NULL)
{
id_list->erase(iter);
delete id_ptr;
}
return (0);
}
char* get_buf(variable_key_t key, int len)
{
static Tls tls;
return tls.pthread_atexit_get_buf(key, len);
}
这个类里用到了几个令人蛋疼的库函数，初看之下还以为是“山寨”的。不过幸好man都能找到，每个函数的功能就不具体介绍了，很多地方都有介绍的。
这个类主要采用了这样的一种结构：每个线程拥有一个TlsList链，它存储了该线程使用的所有的线程局部变量。每个线程可以通过pthread_setspecific存储TlsList，也可以通过pthread_getspecific读出TlsList，进而通过variable_key_t key来访问TlsList中的数据。Key与线程无关，每个函数中的每个线程局部变量需要拥有一个独立的key值。因此，在实际使用时一个增加一个线程局部变量就需要在Tlsconf.h中加一个字段，同样每删除一个变量都需要将Tlsconf.h中对应的key值删除。线程在结束时，会自动调用回调函数，遍历该线程的TlsList，调用每个资源的清理回调函数。
Tls类中基本所有的成员都是静态变量，一来是为了方便，因为成员函数里用到了其它成员函数的地址，不用静态成员不太好处理；二来这样也已经满足使用需求了，具体为什么，大家可以自己思考一下，欢迎一起讨论。
下面是测试函数，主要是看内存的申请、使用及内存释放：
//Test4：
#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 ", pthread_self(), psz);
memset(psz, 0, 32);
sprintf(psz, "%u_%d", pthread_self(), a);
return;
}
void* test_fn_main2(void* arg)
{
int i = 0;
for (i = 0; i < 3; i++)
{
test_fn2(i);
sleep(1);
}
return;
}
void test_fn1(void)
{
int *reti = (int*)get_buf(ENUM_0, sizeof(int));
printf("thread(%u) reti: %d ", pthread_self(), *reti);
(*reti)++;
return;
}
void* test_fn_main1(void* arg)
{
int i = 0;
for (i = 0; i < 3; i++)
{
test_fn1();
sleep(1);
}
return;
}
int main()
{
int iRet = ERROR;
int i = 0;
pthread_t tid;
printf("test start! ");
for (i = 0; i < 5; i++)
{
iRet = pthread_create(&tid, NULL, test_fn_main1, NULL);
if (OK != iRet)
{
printf("pthread_create error! ");
return ERROR;
}
iRet = pthread_create(&tid, NULL, test_fn_main2, NULL);
if (OK != iRet)
{
printf("pthread_create error! ");
return ERROR;
}
}
sleep(20);
return OK;
}
//result
[root@localhost 20130713]# ./tls4
test start!
pthread(3086523280) get resouce 0!
thread(3086523280) reti: 0
pthread(3076033424) get resouce 1!
thread(3076033424) psz:
pthread(3065543568) get resouce 0!
thread(3065543568) reti: 0
pthread(3055053712) get resouce 1!
thread(3055053712) psz:
pthread(3044563856) get resouce 0!
thread(3044563856) reti: 0
pthread(3034074000) get resouce 1!
thread(3034074000) psz:
pthread(3023584144) get resouce 0!
thread(3023584144) reti: 0
pthread(3013094288) get resouce 1!
thread(3013094288) psz:
pthread(3002604432) get resouce 0!
thread(3002604432) reti: 0
pthread(2992114576) get resouce 1!
thread(2992114576) psz:
thread(3086523280) reti: 1
thread(3076033424) psz: 3076033424_0
thread(3065543568) reti: 1
thread(3055053712) psz: 3055053712_0
thread(3044563856) reti: 1
thread(3034074000) psz: 3034074000_0
thread(3023584144) reti: 1
thread(3013094288) psz: 3013094288_0
thread(3002604432) reti: 1
thread(2992114576) psz: 2992114576_0
thread(3086523280) reti: 2
thread(3076033424) psz: 3076033424_1
thread(3065543568) reti: 2
thread(3055053712) psz: 3055053712_1
thread(3044563856) reti: 2
thread(3034074000) psz: 3034074000_1
thread(3023584144) reti: 2
thread(3013094288) psz: 3013094288_1
thread(3002604432) reti: 2
thread(2992114576) psz: 2992114576_1
pthread(3086523280) realease resouce 0!
pthread(3076033424) realease resouce 1!
pthread(3065543568) realease resouce 0!
pthread(3055053712) realease resouce 1!
pthread(3044563856) realease resouce 0!
pthread(3034074000) realease resouce 1!
pthread(3023584144) realease resouce 0!
pthread(3013094288) realease resouce 1!
pthread(3002604432) realease resouce 0!
pthread(2992114576) realease resouce 1!
//Test5:
#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 ", 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 ", 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! ");
for (i = 0; i < 5; i++)
{
iRet = pthread_create(&tid, NULL, test_fn_main1, NULL);
if (OK != iRet)
{
printf("pthread_create error! ");
return ERROR;
}
}
sleep(20);
return OK;
}
//result
[root@localhost 20130713]# ./tls5
test start!
pthread(3086138256) get resouce 0!
thread(3086138256) reti: 0
pthread(3086138256) get resouce 1!
thread(3086138256) psz:
pthread(3075648400) get resouce 0!
thread(3075648400) reti: 0
pthread(3075648400) get resouce 1!
thread(3075648400) psz:
pthread(3065158544) get resouce 0!
thread(3065158544) reti: 0
pthread(3065158544) get resouce 1!
thread(3065158544) psz:
pthread(3054668688) get resouce 0!
thread(3054668688) reti: 0
pthread(3054668688) get resouce 1!
thread(3054668688) psz:
pthread(3044178832) get resouce 0!
thread(3044178832) reti: 0
pthread(3044178832) get resouce 1!
thread(3044178832) psz:
thread(3086138256) reti: 1
thread(3086138256) psz: 3086138256_0
thread(3075648400) reti: 1
thread(3075648400) psz: 3075648400_0
thread(3065158544) reti: 1
thread(3065158544) psz: 3065158544_0
thread(3054668688) reti: 1
thread(3054668688) psz: 3054668688_0
thread(3044178832) reti: 1
thread(3044178832) psz: 3044178832_0
thread(3086138256) reti: 2
thread(3086138256) psz: 3086138256_1
thread(3075648400) reti: 2
thread(3075648400) psz: 3075648400_1
thread(3065158544) reti: 2
thread(3065158544) psz: 3065158544_1
thread(3054668688) reti: 2
thread(3054668688) psz: 3054668688_1
thread(3044178832) reti: 2
thread(3044178832) psz: 3044178832_1
pthread(3086138256) realease resouce 0!
pthread(3086138256) realease resouce 1!
pthread(3075648400) realease resouce 0!
pthread(3075648400) realease resouce 1!
pthread(3065158544) realease resouce 0!
pthread(3065158544) realease resouce 1!
pthread(3054668688) realease resouce 0!
pthread(3054668688) realease resouce 1!
pthread(3044178832) realease resouce 0!
pthread(3044178832) realease resouce 1
其中有个问题，主线程的资源最后没有被释放，这是不是问题呢？大家可以思考一下？