使用线程的分段错误

Posted 2023-02-22

【中文标题】使用线程的分段错误

【英文标题】：Segmentation Fault Using threads

【发布时间】：2013-04-11 08:16:16

【问题描述】：

我正在尝试构建自己的库来使用和调度线程。

系统以您可以make maximum 100 threads 的方式工作，使它们“准备就绪”。然后有一个调度器，每个时间段，把活跃的线程，放到就绪链表中，并在就绪链表中取第一个等待的线程。

我的测试人员正在创建500000 线程，每个线程都会创建一个新线程并尝试自行终止。第二个线程（第一个线程创建）做同样的事情。第三个只是终止自己。

这是测试员：

void f3()

    printf("f3 before terminate\n");
    uthread_terminate(uthread_get_tid());
    printf("f3 after terminate\n");


void f2()

    printf("f2 before spawn\n");
    uthread_spawn(f3);
    printf("f2 after spawn\n");
    uthread_terminate(uthread_get_tid());
    printf("f2 after termination\n");


void f1()

    printf("f1 before spawn\n");
    uthread_spawn(f2);
    printf("f1 after spawn\n");
    uthread_terminate(uthread_get_tid());
    printf("f1 after termination\n");




int main(int argc, char **argv)

   printf("test8:\n--------------\n");
   cout << "* Output should be:\n";    
   cout << "--------------\n";
   cout << "***1***\n";
   cout << "***2***\n";
   cout << "Output is:\n";
   uthread_init(100);
   printf("***1***\n");
   for (volatile int i=0; i< 50000;++i)
        uthread_spawn(f1);
   
   printf("***2***\n");

   uthread_terminate(0);
   return 0;

当我的程序进入我的调度程序时出现“分段错误”：

static void scheduler()
DBG(("Schedular "))
nQuantum++;
if (ready.size()!=0)
    if (active != NULL)
        if (active->getState() == Thread::RUNNING)
            active->setState(Thread::READY);
            ready.push_back(active);
        

        int val = sigsetjmp(*active->getEnv(),1);
        if (val !=0)
            blockTimerSignal(UNBLOCK);
            return;
        
    

    // Set new Active
    active = ready.front();
    DBG(("Active thread Id: %d",active->getTid()))
    ready.pop_front();
    DBG(("Doing pop front on ready list"))
    DBG(("Number of threads in ready list is  - %d",(ready.size())))
    active->setQuantums(active->getQuantums()+1);
    active->setState(Thread::RUNNING);
    setTimer();
    blockTimerSignal(UNBLOCK);
    DBG(("UNBLOCKED"))
    siglongjmp(*active->getEnv(),1);
    DBG(("After siglong jmp"))
else
    active->setQuantums(active->getQuantums()+1);

DBG(("Number of threads in ready list is  - %d",(ready.size())))

blockTimerSignal(UNBLOCK);
 

在做siglongjmp(*active->getEnv(),1);时

它只发生在主线程（程序的第一个 id 为 0 的线程）。 它也会在程序运行一段时间后发生，这意味着在程序的中间，它可以为活动线程执行 siglongjmp，但是当它在一段时间后再次尝试时，它会给出错误。

如果有帮助，添加我的终止功能：

int uthread_terminate(int tid) 
    DBG(("Terminate - %d", tid))

    if (tid == 0)
        // delete all (threads) - don't think it's needed because using stl!
        // TODO : check if needed - and then delete all lists !
        //Added by Roni - Deleting all lists!
        while(!sleeping.empty())
            delete (sleeping.front());
            sleeping.pop_front();
        

        while(!suspended.empty())
            delete (suspended.front());
            suspended.pop_front();
        

        while(!ready.empty())
            delete (ready.front());
            ready.pop_front();
        


        exit(0);
    

    pThread t = getThread(tid);
    if (t == NULL)
        errmsgLibrary(THREAD_NOT_FOUND);
        return FAIL;
    
    Thread::threadState state = t->getState();
    DBG(("Terminate - %d in State %d", tid, state))
    DBG(("Number of threads in ready list is  - %d",(ready.size())))
    blockTimerSignal(BLOCK);
    switch (state)
    case (Thread::RUNNING):
        //DBG(("Running Case"))
        //DBG(("Active thread id is: %d ",active->getTid()))
            delete active;
            active = NULL;
    //DBG(("Finsihed running Case"))
            scheduler();

            break;
    case (Thread::READY):
        //DBG(("ready Case"))
            ready.remove(t);
            delete t;
            //DBG(("Finsihed ready Case"))
        break;
    case (Thread::SLEEP):
            sleeping.remove(t);
            delete t;
        break;
    case (Thread::SUSPENDED):
            suspended.remove(t);
            delete t;
            break;
    default:
        break;
    
    //DBG(("Number of threads in ready list is  - %d",(ready.size())))
    blockTimerSignal(UNBLOCK);
    return SUCCESS;

【问题讨论】：

active 必须是 NULL 所以导致 SIGSEGV 在那一刻。您应该在每次调用时检查uthread_spawn(f3); 是否成功，因为您可能遇到线程限制或内存不足而导致失败。

@ReckHou Hou Hou 实际上，当我检查时，我只看到 f1 在起作用。这是什么意思？你有什么建议？

【参考方案1】：

猜测一下，您在这里使用的是siglongjmp，并且可能不会在调用之间更改堆栈。

考虑改用getcontext()、swapcontext() 和makecontext()。