c++ - 在使用 PIN 的模拟器上对 Pthread 进行安全编程

Question

我正在使用硬件模拟器，它使用 PIN 工具来执行工作负载。作为工作负载，我使用以下代码。虽然它可以在带有 -lpthread 标志的 Ubuntu 上运行，但在加入线程时它会在模拟器上卡住。

我认为 native 操作系统可以容忍但模拟器不能容忍的代码中存在一些不安全的地方。最合适的编码方式是什么？

ma​​in.h:

#include <stdlib.h>
#include <iostream>
#include <fstream>
#include <string>
#include <pthread.h>
#include <stdint.h>
#include <getopt.h>
#include <set>
#include <vector>
#include <algorithm>
#include <iterator>

#define NUM_OF_VERTICES 4
#define NUM_OF_PTHREADS (NUM_OF_VERTICES*(NUM_OF_VERTICES-1)/2)

std::string payload_texts[NUM_OF_VERTICES];
void payload_text_initialize();
double indices [NUM_OF_VERTICES][NUM_OF_VERTICES];

class thread_args {   
    public: 
        uint index1, index2;
        unsigned  long value = 0;
        unsigned  long * valuePointer = &value;
};

ma​​in.cc:

#include "main.h"

extern "C" {
    extern void mcsim_skip_instrs_begin();
    extern void mcsim_skip_instrs_end();
    extern void mcsim_spinning_begin();
    extern void mcsim_spinning_end();
    int32_t log_2(uint64_t);
}

using namespace std;

set<char> find_uniques(string str){
    set<char> unique_chars;
    for (int i = 0 ; i < str.length() ; i++ ){
        char c = str.at(i);
        if (unique_chars.find(c) == unique_chars.end())
            unique_chars.insert(c);
    }
    return unique_chars;
}

void * jaccard_visit(void *arg){
    thread_args * args = (thread_args *) arg;
    set<char> setunion;
    set<char> intersect;

    set<char> set1 = find_uniques(payload_texts[args->index1]);
    set<char> set2 = find_uniques(payload_texts[args->index2]);


    std::set_intersection(set1.begin(),set1.end(),set2.begin(),set2.end(),std::inserter(intersect,intersect.begin()));
    std::set_union(set1.begin(),set1.end(),set2.begin(),set2.end(),std::inserter(setunion,setunion.begin()));

    double similarity = ((double) intersect.size()) / ((double) setunion.size());
    indices[args->index1][args->index2] = similarity;
    indices[args->index2][args->index1] = similarity;

    unsigned long a = 1;
    unsigned long b = 1;
    unsigned long c = a + b;
    for (int i = 3 ; i < 100000 * (similarity - 0.9) ; i++){
        a = b;
        b = c; 
        c = a + b;
    }

    *(args->valuePointer) = c;

    return NULL;
}

void execute_parallel(){      
    pthread_t threads[NUM_OF_PTHREADS]; //array to hold thread information 
    thread_args *th_args = (thread_args*) malloc(NUM_OF_PTHREADS * sizeof(thread_args)); 

    cout << "NUM_OF_PTHREADS is " << NUM_OF_PTHREADS << endl;
    uint k = 0 ;
    for (int i = 0 ; i < NUM_OF_VERTICES ; i++){
        for (int j = i+1 ; j < NUM_OF_VERTICES ; j++){
            th_args[k].index1 = i;
            th_args[k].index2 = j;
            th_args[k].value = i+j; 
            th_args[k].valuePointer = &(th_args[k].value);
            pthread_create(&threads[k], NULL, jaccard_visit, (void*) &th_args[k]);
            cout << "Thread " << k << " is started" << endl;
            k++;
        }
    }

    cout << "k is " << k << endl;

    for(int i = 0; i < NUM_OF_PTHREADS; i++){ 
        cout << "Thread " << i << " is joined" << endl;
        pthread_join(threads[i], NULL); 
    } 

    cout << "Free threads" << endl ;

    free(th_args); 

} 

void manual_schedule(){
    pthread_t th0, th1, th2, th3, th4, th5;
    thread_args arg0, arg1, arg2, arg3, arg4, arg5;

    arg0.index1 = 0;    arg0.index2 = 1;    arg0.value = 0; arg0.valuePointer = &arg0.value;
    arg1.index1 = 0;    arg1.index2 = 2;    arg1.value = 1; arg1.valuePointer = &arg1.value;
    arg2.index1 = 0;    arg2.index2 = 3;    arg2.value = 2; arg2.valuePointer = &arg2.value;
    arg3.index1 = 1;    arg3.index2 = 2;    arg3.value = 3; arg3.valuePointer = &arg3.value;
    arg4.index1 = 1;    arg4.index2 = 3;    arg4.value = 4; arg4.valuePointer = &arg4.value;
    arg5.index1 = 2;    arg5.index2 = 3;    arg5.value = 5; arg5.valuePointer = &arg5.value;

    cout << "Arguments are done ";

    pthread_create(&th0, NULL, jaccard_visit, (void*) &arg0);
    pthread_create(&th1, NULL, jaccard_visit, (void*) &arg1);
    pthread_create(&th2, NULL, jaccard_visit, (void*) &arg2);
    pthread_create(&th3, NULL, jaccard_visit, (void*) &arg3);
    pthread_create(&th4, NULL, jaccard_visit, (void*) &arg4);
    pthread_create(&th5, NULL, jaccard_visit, (void*) &arg5);

    cout << "Threads are created" << endl;

    cout << "Join starts here" << endl;
    pthread_join(th0, NULL); 
    pthread_join(th1, NULL); 
    pthread_join(th2, NULL); 
    pthread_join(th3, NULL); 
    pthread_join(th4, NULL); 
    pthread_join(th5, NULL); 

    cout << "Fibonaccis: " <<endl;
    cout << *(arg0.valuePointer) << endl;
    cout << *(arg1.valuePointer) << endl;
    cout << *(arg2.valuePointer) << endl;
    cout << *(arg3.valuePointer) << endl;
    cout << *(arg4.valuePointer) << endl;
    cout << *(arg5.valuePointer) << endl;
}

int main(int argc, const char * argv[]){
    cout << "Jaccard process is started"<<endl;
    mcsim_skip_instrs_begin();
    payload_text_initialize();
    mcsim_skip_instrs_end();

    cout << "Parallel part begins"<< endl;
    manual_schedule();

    cout << "Calculated results are being logged"<<endl;
    for (int i = 0 ; i < NUM_OF_VERTICES ; i++){
        for (int j = 0 ; j < NUM_OF_VERTICES ; j++){
            cout << indices[i][j] << " ";
        }
        cout << endl;
    }
}



void payload_text_initialize(){
    payload_texts[0] = "l5IC5uC9AzcROkE3YkDJ2lEzLts8XP8a9WqDgDLWjg1M7HysAUfDFwzLWjc7875PnZVUHLzi6nQaUMQDNUeG4Wn2UkiOB79tOlE1t6LaKYbYiCJwJ34CAOFZCIbFSmcLTAAoB1rvPfeA6oM3kV3C8BDvraGvXjUORLGFAcBRQCerb3WD0qhrrM0MVW0t93bBqlTsrkxg";
    payload_texts[1] = "tILKwAhbUkoqouKZ1G1VrZRmKwQnwzBgQirLkdedsYIAplKdEfk8oSmqdJmCJd5g0Q3VcJ8RYoxtIwA7jL1L01DcagIOuld0whcyM0yvSP0pMWO2yVTwOQPGkW2k7AHqzSEvb5BWkKsTexBsCUepjbG50T6vKsEHXGJ9aZwn2274Ekhnu1hlvuTqsS8jgwr0kQwhbwxN";
    payload_texts[2] = "LNyQgx3mox3szmRNn1tSB4ibVuLsTr7MfANlj41Y0hKStx3NJx1O52XxNiqTMDCu4eGwWYcBvFMEC5tl1E7Rsm0Q9NZsPAJIwuiPYQuXeUyhMmbFiwRk6PlziXne0QaFJ3TrncsHsL3LxIDyaDPScSRdEvX72IJmi2gQTHgASi0KkKH4Sr6VJV3FjdNjKwY2ncT5oSXZ";
    payload_texts[3] = "UxynTAvEWF4CcY9wUJRFnrX7sgrvvubcXUqH5DXK12UjSHDUME397S3BdB38FeMQJq8r7P7RILAY0qkw7OxUhGsZHRPmuY7VwKULqb6fx0Oy2McW2u07yqdAEMCN6AkQ1jTn2sXB4uWH21uLbjCf9i2V7W9tyw3cx6piE7XJb3vfbLI34OG5LKQXmVAGT0D6nbibaN8M";
}

execute_parallel() 包含 2 个用于创建和加入 pthread 的 for 循环。 manual_schedule() 具有相同代码的展开版本。在 PIN 上执行时，两者都运行良好，直到第一个线程的 join 函数。当 join 到来时，它会卡住并永远保持这样，没有任何信号或错误。在带有 -lpthread 标志的 Ubuntu 上执行时，它运行完美并生成结果。

在这种情况下，实现 pthread 最安全、最合适的方法是什么？

提前致谢

编辑

我注意到程序在读取 payload_texts[args->index1] 之前卡住。添加互斥量有助于在这一点上进行。它也正常工作了一次。它现在是不确定的，在同一二进制文件的多次执行中，它很少能正确完成。我认为 jaccard_visit 函数内的死锁应该是有原因的。我将其更改如下:

void * jaccard_visit(void *arg){
    thread_args * args = (thread_args *) arg;
    set<char> setunion;
    set<char> intersect;
    int id = args->index1 * 10 + args->index2;

    pthread_mutex_lock(&cout_mutex);    cout << "Thread "<< id << " started with indices: " << args->index1 << " " << args->index2 << endl;     pthread_mutex_unlock(&cout_mutex);

    pthread_mutex_lock(&payload_mutex);
    set<char> set1 = find_uniques(payload_texts[args->index1]);
    set<char> set2 = find_uniques(payload_texts[args->index2]);
    pthread_mutex_unlock(&payload_mutex);

    pthread_mutex_lock(&cout_mutex);    cout << id << " : payload_texts were read" << endl;         pthread_mutex_unlock(&cout_mutex);
    pthread_mutex_lock(&cout_mutex);    cout << id << " : intersect was created, scan begins" << endl;         pthread_mutex_unlock(&cout_mutex);
    for (set<char>::iterator i = set1.begin(); i != set1.end(); i++) {
        char c1 = *i;
        for (set<char>::iterator j = set2.begin(); j != set2.end(); j++) {
            char c2 = *j;
            if (c1 == c2){
                intersect.insert(c1);
                pthread_mutex_lock(&cout_mutex);    cout << id << " : char" << c1 << " was inserted to intersection" << endl;         pthread_mutex_unlock(&cout_mutex);
                break;
            }
        }
    }
    pthread_mutex_lock(&cout_mutex);    cout << id << " : intersection is calculated" << endl;     pthread_mutex_unlock(&cout_mutex);
    for (set<char>::iterator i = set1.begin(); i != set1.end(); i++) {
        setunion.insert(*i);
    }
    for (set<char>::iterator i = set2.begin(); i != set2.end(); i++) {
        char c = *i;
        bool exists = false;
        for (set<char>::iterator j = set1.begin(); j != set1.end(); j++) {
            if (c == *j)
                exists = true;
        }
        if (exists == false)
            setunion.insert(c);
    }
    pthread_mutex_lock(&cout_mutex);    cout << id << " : union is calculated" << endl;            pthread_mutex_unlock(&cout_mutex);


    double similarity = ((double) intersect.size()) / ((double) setunion.size());
    pthread_mutex_lock(&cout_mutex);
    cout << id << " : similarity is calculated as " << similarity << endl;
    pthread_mutex_unlock(&cout_mutex);
    indices[args->index1][args->index2] = similarity;
    indices[args->index2][args->index1] = similarity;

    unsigned long a = 1;
    unsigned long b = 1;
    unsigned long c = a + b;

    pthread_mutex_lock(&cout_mutex);
    cout << id << " : fibonacci starts" << endl;
    pthread_mutex_unlock(&cout_mutex);

    for (int i = 3 ; i < 100000 * (similarity - 0.9) ; i++){
        a = b;
        b = c; 
        c = a + b;
    }
    *(args->valuePointer) = c;

    //pthread_exit(args->valuePointer);
    return NULL;
}

Answer 1

最后，我通过对每个线程执行的函数 (jaccard_visit) 进行以下修改使其工作:

1. 用mutex_lock包裹全局变量的读操作
2. 删除函数调用并内联实现
3. 避免使用集合，而是使用字符串或字符数组
4. 将字符串函数的目标和源分开

以下代码运行良好:

void * jaccard_visit(void *arg){
    thread_args * args = (thread_args *) arg;
    int id = args->id;
    pthread_mutex_lock(&cout_mutex);    cout << id << " : thread started" << endl;    pthread_mutex_unlock(&cout_mutex);

    pthread_mutex_lock(&payload_mutex);
    string str1 = my_graph[args->index1].payload_text;
    string str2 = my_graph[args->index2].payload_text;
    pthread_mutex_unlock(&payload_mutex);

    pthread_mutex_lock(&cout_mutex);    cout << id << " : payload texts are read" << endl;    pthread_mutex_unlock(&cout_mutex);

    int stringLength = str1.length() - 1;
    for (int i = 0; i < stringLength; i++) {
        for (int j = i + 1; j < stringLength;) {
            if (str1[i] == str1[j]) 
                str1[j] = str1[--stringLength];
            else   
                j++;
        }
    }
    string set1 = str1.substr(0, stringLength);

    pthread_mutex_lock(&cout_mutex);    cout << id << " : unique chars of first node were extracted" << endl;    pthread_mutex_unlock(&cout_mutex);

    stringLength = str2.length() - 1;
    for (int i = 0; i < stringLength; i++) {
        for (int j = i + 1; j < stringLength;) {
            if (str2[i] == str2[j]) 
                str2[j] = str2[--stringLength];
            else   
                j++;
        }
    }
    string set2 = str2.substr(0, stringLength);

    pthread_mutex_lock(&cout_mutex);    cout << id << " : unique chars of second node were extracted" << endl;    pthread_mutex_unlock(&cout_mutex);

    int intersection_index = 0;
    int union_index = 0;
    for (int i = 0 ; i < set1.length() ; i++){
        bool exists_in_set2 = false;
        for (int j = 0 ; j < set2.length() && exists_in_set2 == false; j++){
            if (set1[i] == set2[j]) {
                intersection_index ++;
                exists_in_set2 = true;
            }
        }
        if (!exists_in_set2) {
            union_index ++;
        }       
    }
    union_index += set2.length();

    pthread_mutex_lock(&cout_mutex);  cout << id << " : set1={" << set1 << "}, set2={" << set2 << "}" << endl;    pthread_mutex_unlock(&cout_mutex);
    pthread_mutex_lock(&cout_mutex);  cout << id << " : |n|=" << intersection_index << ", |u|=" << union_index <<  endl;    pthread_mutex_unlock(&cout_mutex);

    double similarity = ((double) intersection_index / union_index); 
    pthread_mutex_lock(&cout_mutex); cout<<id<<" : similarity is: " << similarity << endl;    pthread_mutex_unlock(&cout_mutex);

    pthread_mutex_lock(&indices_mutex); 
    my_graph[args->index1].jaccardList[args->index2] = similarity;
    my_graph[args->index2].jaccardList[args->index1] = similarity;
    pthread_mutex_unlock(&indices_mutex); 
    return NULL;

}