Lamport的Bakery算法的实现具有1个以上线程的段错误

Question

I'm implementing Lamport's Bakery Algorithm using pthreads and a Processor class to act as shared memory. With a single thread it works fine, with 2 threads I get the seg fault after thread 2 runs through all 30 attempts to access the 'bakery':

dl-tls.c: No such file or directory.

With 3 or more threads I get the seg fault after outputting "here" twice from the bakeryAlgo function:

0x0804ae52 in Processor::getNumber (this=0x5b18c483) at Processor.cpp:33

bakery.cpp

struct argStruct {
    vector<Processor>* processors;
    Processor* processor;
};

int findMax(vector<Processor>* processors) {
    int max = -99;
    for (int i = 0; i < processors->size(); i++) {
        if (processors->at(i).getNumber() > max) {
            max = processors->at(i).getNumber();
        }
    }
    return max;
}

void* bakeryAlgo(void* arg) {
    struct argStruct* args = static_cast<struct argStruct *>(arg);
    cout << "here" << endl;
    for (int i = 0; i < 30; i++) {
        args->processor->setChoosing(1);
        args->processor->setNumber(findMax(args->processors));
        args->processor->setChoosing(0);
        for (int j = 0; j < args->processors->size(); j++) {
            int jChoosing = args->processors->at(j).getChoosing();
            int jNumber = args->processors->at(j).getNumber();
            int jId = args->processors->at(j).getId();
            int pNumber = args->processor->getNumber();
            int pId = args->processor->getId();
            if (jId != pId) {
                while (jChoosing != 0) {}
                while (jNumber != 0 && ((jNumber < pNumber) || ((jNumber == pNumber) && (jId < pId)))) { }
            }
        }
        cout << "Processor: " << args->processor->getId() << " executing critical section!" << endl;
        args->processor->setNumber(0);
    }
}

int main(int argc, char *argv[]) {

    // Check that a command line argument was provided
    if (2 == argc) {
        int numProcessors = atoi(argv[1]);
        vector<Processor> processors;
        vector<argStruct> argVect;
        vector < pthread_t > threads;
        for (int i = 0; i < numProcessors; i++) {
            Processor p = Processor(i);
            processors.push_back(p);
        }
        for (int i = 0; i < numProcessors; i++) {
            pthread_t processorThread;
            struct argStruct args;
            args.processors = &processors;
            args.processor = &processors.at(i);
            argVect.push_back(args);
            threads.push_back(processorThread);
            pthread_create(&threads.at(i), NULL, &bakeryAlgo, &argVect.at(i));
        }
        for (int i = 0; i < numProcessors; i++) {
            pthread_join(threads.at(i), NULL);
        }
    } 
    else {
        cout << "Usage: bakery num, num is number of threads." << endl;
    }
    return 0;
}

The code in Processor.cpp / Processor.h is simple, it's just a few getters and setters on the values id, choosing, and number, with a default constructor and a constructor that takes an int id.

Processor::Processor() {

}

Processor::Processor(int idval) {
    id = idval;
    choosing = 0;
    number = 0;
}

Processor::~Processor() {

}

int Processor::getChoosing() {
    return choosing;
}

int Processor::getNumber() {
    return number;
}

int Processor::getId() {
    return id;
}

void Processor::setChoosing(int c) {
    choosing = c;
}

void Processor::setNumber(int n) {
    number = n;
}

Does anyone have any idea why these seg faults are occuring? The places gdb says they are occuring look like innocent lines of code to me.

Answer 1

Are you using a pointer to a vector defined in main as your data? Stack is not shared among threads, so the other threads accessing this memory would be undefined behavior at best. I expect this is the source of your troubles.

Answer 2

You are taking the address of an element in a vector which is changing:

    for (int i = 0; i < numProcessors; i++) {
        pthread_t processorThread;
        struct argStruct args;
        args.processors = &processors;
        args.processor = &processors.at(i);
        argVect.push_back(args);
        threads.push_back(processorThread);

        // danger! 
        pthread_create(&threads.at(i), NULL, &bakeryAlgo, &argVect.at(i));
    }

Each time a new thread is pushed onto the threads vector , the vector may be relocated in memory, so the pointer that you passed to pthread_create may be pointing to garbage when the next thread is added.

You want to do this instead:

    for (int i = 0; i < numProcessors; i++) {
        pthread_t processorThread;
        struct argStruct args;
        args.processors = &processors;
        args.processor = &processors.at(i);
        argVect.push_back(args);
        threads.push_back(processorThread);
    }
    for (int i = 0; i < numProcessors; i++) {
        pthread_create(&threads.at(i), NULL, &bakeryAlgo, &argVect.at(i));
    }

By waiting until all the elements are added to the vector before you create your threads, you are now passing pointers that stay good while the threads are running.