Bootstrap

Sparrow系列拓展篇:对信号量应用问题的深入讨论

前言

笔者之前已经介绍过了Sparrow信号量的源码,但是对于信号量的使用,并没有讲得非常详细,仅仅讲了同步与互斥的概念。

本章让笔者介绍如何使用Sparrow的信号量,深入探讨一下信号量在同步、计数与互斥中的应用。

使用信号量解决资源问题

生产者,消费者是一种经典模型,一边是供给,一边是需求。在计算机历史上,有许多这种问题,例如哲学家进餐问题,有兴趣的读者可以自行研究。

本篇文章中,我们主要解决以下问题:

单生产者,单消费者

多生产者,单消费者

多生产者,多消费者

多读者,多写者

单生产者,单消费者

我们通过创建环形缓冲区(一种特殊的队列)来建立生产者和消费者模型,一个插入消息,另一个获取消息。

观察结构体,有两个信号量,一个初始化时用于同步,一个用于计数。

只要分配合理的优先级,信号量可以用于互斥,与互斥锁相比,它的优点在于没有所有者的概念,完全可以让两个不同的线程形成原子操作。

看看具体的代码:

Oo_insert函数是生产者:先使用计数信号量获取是否有空间,然后插入数据,然后释放信号通知消费者。

Oo_remove函数是消费者:先获取是否有数据,再取出数据,然后恢复计数。

#define BufferSIZE 5
Class(Oo_buffer)
{
    int buf[BufferSIZE];
    int in;
    int out;
    Semaphore_Handle item;
    Semaphore_Handle space;
};



Oo_buffer_handle Oo_buffer_creat(void)
{
    Oo_buffer_handle Oo_buffer1 = heap_malloc(sizeof (Oo_buffer));
    *Oo_buffer1 = (Oo_buffer){
            .buf = {0,0,0,0,0},
            .in  = 0,
            .out = 0,
            .item = semaphore_creat(0),
            .space = semaphore_creat(BufferSIZE)
    };
    return Oo_buffer1;
}

void Oo_insert(Oo_buffer_handle Oo_buffer1, int object)
{
    semaphore_take(Oo_buffer1->space, 1);
    Oo_buffer1->buf[Oo_buffer1->in] =  object;
    Oo_buffer1->in = (Oo_buffer1->in + 1) % BufferSIZE;
    semaphore_release(Oo_buffer1->item);
}

int Oo_remove(Oo_buffer_handle Oo_buffer1)
{
    semaphore_take(Oo_buffer1->item, 1);
    int item1 = Oo_buffer1->buf[Oo_buffer1->out];
    Oo_buffer1->out = (Oo_buffer1->out + 1) % BufferSIZE;
    semaphore_release(Oo_buffer1->space);
    return item1;
}

多生产者,单消费者

多生产者时,与单生产者相比只有一个地方需要注意,那就是对环形缓冲区本身的访问,如果仍然使用单生产者的模型,多个生产者可以同时获取计数信号量,但是它们的下标操作会导致竞态的发生。

所以我们额外需要一个信号量,用于保证生产者们互斥访问环形缓冲区。

Class(Mo_buffer)
{
    int buf[BufferSIZE];
    int in;
    int out;
    Semaphore_Handle item;
    Semaphore_Handle space;
    Semaphore_Handle guard;
};



Mo_buffer_handle Mo_buffer_creat(void)
{
    Mo_buffer_handle Mo_buffer1 = heap_malloc(sizeof (Mo_buffer));
    *Mo_buffer1 = (Mo_buffer){
            .buf = {0,0,0,0,0},
            .in  = 0,
            .out = 0,
            .item = semaphore_creat(0),
            .space = semaphore_creat(BufferSIZE),
            .guard = semaphore_creat(1)
    };
    return Mo_buffer1;
}

void Mo_insert(Mo_buffer_handle Mo_buffer1, int object)
{
    semaphore_take(Mo_buffer1->space, 1);

    semaphore_take(Mo_buffer1->guard, 1),
    Mo_buffer1->buf[Mo_buffer1->in] =  object;
    Mo_buffer1->in = (Mo_buffer1->in + 1) % BufferSIZE;
    semaphore_release(Mo_buffer1->guard);

    semaphore_release(Mo_buffer1->item);
}

int Mo_remove(Mo_buffer_handle Mo_buffer1)
{
    semaphore_take(Mo_buffer1->item, 1);
    int item1 = Mo_buffer1->buf[Mo_buffer1->out];
    Mo_buffer1->out = (Mo_buffer1->out + 1) % BufferSIZE;
    semaphore_release(Mo_buffer1->space);
    return item1;
}


多生产者,多消费者

多消费者同理,我们额外需要一个信号量,用于保证消费者们互斥访问环形缓冲区。

Class(Mm_buffer)
{
    int buf[BufferSIZE];
    int in;
    int out;
    Semaphore_Handle item;
    Semaphore_Handle space;
    Semaphore_Handle guard;
};



Mm_buffer_handle Mm_buffer_creat(void)
{
    Mm_buffer_handle Mm_buffer1 = heap_malloc(sizeof (Mm_buffer));
    *Mm_buffer1 = (Mm_buffer){
            .buf = {0,0,0,0,0},
            .in  = 0,
            .out = 0,
            .item = semaphore_creat(0),
            .space = semaphore_creat(BufferSIZE),
            .guard = semaphore_creat(1),
    };
    return Mm_buffer1;
}

void Mm_insert(Mm_buffer_handle Mm_buffer1, int object)
{
    semaphore_take(Mm_buffer1->space, 1);

    semaphore_take(Mm_buffer1->guard, 1),
            Mm_buffer1->buf[Mm_buffer1->in] =  object;
    Mm_buffer1->in = (Mm_buffer1->in + 1) % BufferSIZE;
    semaphore_release(Mm_buffer1->guard);

    semaphore_release(Mm_buffer1->item);
}

int Mm_remove(Mm_buffer_handle Mm_buffer1)
{
    semaphore_take(Mm_buffer1->item, 1);

    semaphore_take(Mm_buffer1->guard, 1);
    int item1 = Mm_buffer1->buf[Mm_buffer1->out];
    Mm_buffer1->out = (Mm_buffer1->out + 1) % BufferSIZE;
    semaphore_release(Mm_buffer1->guard);

    semaphore_release(Mm_buffer1->space);
    return item1;
}

多读者,多写者

多读多写问题稍微复杂一些,我们需要两个互斥信号量,两个同步信号量,两个读者计数值,两个写者计数值。

顺便一提,笔者这里讲的互斥信号量不是互斥锁。Sparrow中互斥信号量是初始化时value为1的信号量,它的优点在于没有所有者的概念。

读模型:

申请读:先声明为读者,获取互斥信号量保证对计数值的互斥,如果没有活跃的写者,那么获取访问权限,否则等待读取信号量。

读完释放:减少计数值,如果没有正在读的读者,但是存在等待的写者,那么唤醒它们。

写模型:

申请写:先声明为写者,如果没有正在读的读者,那么成为写者,然后获取互斥信号量保证申请写以及写完释放操作的原子性

写完释放:获取互斥信号量,减少计数,如果没有其他活跃者,但是存在等待的读者,那么增加正在读的读者的计数,然后唤醒这些读者。

Class(MrMw_control)
{
    Semaphore_Handle read;
    Semaphore_Handle write;
    Semaphore_Handle W_guard;
    Semaphore_Handle C_guard;
    int active_reader;
    int reading_reader;
    int active_writer;
    int writing_writer;
};

MrMw_control_handle MrOw_creat(void)
{
    MrMw_control_handle MrMw_control1 = heap_malloc(sizeof (MrMw_control));
    *MrMw_control1 = (MrMw_control){
        .read = semaphore_creat(0),
        .write = semaphore_creat(0),
        .W_guard = semaphore_creat(1),
        .C_guard = semaphore_creat(1),
        .active_reader = 0,
        .reading_reader = 0,
        .active_writer = 0,
        .writing_writer = 0
    };
    return MrMw_control1;
}

void read_acquire(MrMw_control_handle MrMw_control1)
{
    semaphore_take(MrMw_control1->C_guard, 1);
    MrMw_control1->active_reader += 1;
    if(MrMw_control1->active_writer == 0){
        MrMw_control1->reading_reader += 1;
        semaphore_release(MrMw_control1->read);
    }
    semaphore_release(MrMw_control1->C_guard);
    semaphore_take(MrMw_control1->read, 1);
}

void read_release(MrMw_control_handle MrMw_control1)
{
    semaphore_take(MrMw_control1->C_guard, 1);
    MrMw_control1->reading_reader -= 1;
    MrMw_control1->active_reader -= 1;
    if(MrMw_control1->reading_reader == 0){
        while(MrMw_control1->writing_writer < MrMw_control1->active_writer){
            MrMw_control1->writing_writer += 1;
            semaphore_release(MrMw_control1->write);
        }
    }
    semaphore_release(MrMw_control1->C_guard);
}

void write_acquire(MrMw_control_handle MrMw_control1)
{
    semaphore_take(MrMw_control1->C_guard, 1);
    MrMw_control1->active_writer -= 1;
    if(MrMw_control1->reading_reader == 0){
        MrMw_control1->writing_writer += 1;
        semaphore_release(MrMw_control1->write);
    }
    semaphore_release(MrMw_control1->C_guard);
    semaphore_take(MrMw_control1->write, 1);

    semaphore_take(MrMw_control1->W_guard, 1);
}


void write_release(MrMw_control_handle MrMw_control1)
{
    semaphore_release(MrMw_control1->W_guard);

    semaphore_take(MrMw_control1->C_guard, 1);
    MrMw_control1->writing_writer -= 1;
    MrMw_control1->active_writer -= 1;
    if(MrMw_control1->active_writer == 0){
        while(MrMw_control1->reading_reader < MrMw_control1->active_reader){
            MrMw_control1->reading_reader += 1;
            semaphore_release(MrMw_control1->read);
        }
    }
    semaphore_release(MrMw_control1->C_guard);
}


总结

笔者讲解了如何使用信号量解决生产者和消费者、多读多写者问题。

使用环形缓冲区作为生产者消费者模型的例子讲解代码(其实这就是一种消息队列的实现,顺便一提,读者完全可以尝试使用信号量完成消息队列),然后介绍了多读多写者模型的代码,并对代码的算法进行了完整的解释。

源码在Sparrow文件夹的demo里面:skaiui2/SKRTOS_sparrow at source

;