眉目传情之并发无锁环形队列的实现
-
Author：Echo Chen（陈斌）
-
Email：chenb19870707@gmail.com
-
Date：October 10th, 2014
前面在《眉目传情之匠心独运的kfifo》一文中详细解析了 linux 内核并发无锁环形队列kfifo的原理和实现，kfifo鬼斧神工，博大精深，让人叹为观止，但遗憾的是kfifo为内核提供服务，并未开放出来。剑不试则利钝暗，弓不试则劲挠诬，鹰不试则巧拙惑，马不试则良驽疑，光说不练是不能学到精髓的，下面就动手实现自己的并发无锁队列UnlockQueue（单生产者单消费者）。
一、UnlockQueue声明
1: #ifndef _UNLOCK_QUEUE_H2: #define _UNLOCK_QUEUE_H3:4: class UnlockQueue
5: {6: public:
7: UnlockQueue(int nSize);
8: virtual ~UnlockQueue();
9:10: bool Initialize();
11:12: unsigned int Put(const unsigned char *pBuffer, unsigned int nLen);
13: unsigned int Get(unsigned char *pBuffer, unsigned int nLen);
14:15: inline void Clean() { m_nIn = m_nOut = 0; }
16: inline unsigned int GetDataLen() const { return m_nIn - m_nOut; }
17:18: private:
19: inline bool is_power_of_2(unsigned long n) { return (n != 0 && ((n & (n - 1)) == 0)); };
20: inline unsigned long roundup_power_of_two(unsigned long val);
21:22: private:
23: unsigned char *m_pBuffer; /* the buffer holding the data */
24: unsigned int m_nSize; /* the size of the allocated buffer */
25: unsigned int m_nIn; /* data is added at offset (in % size) */
26: unsigned int m_nOut; /* data is extracted from off. (out % size) */
27: };28:29: #endif
UnlockQueue与kfifo 结构相同相同，也是由一下变量组成：
UnlockQueue kfifo 作用 m_pBuffer buffer 用于存放数据的缓存 m_nSize size 缓冲区空间的大小，圆整为2的次幂 m_nIn in 指向buffer中队头 m_nOut out 指向buffer中的队尾 UnlockQueue的设计是用在单生产者单消费者情况下，所以不需要锁 lock 如果使用不能保证任何时间最多只有一个读线程和写线程，必须使用该lock实施同步。 二、UnlockQueue构造函数和初始化
1: UnlockQueue::UnlockQueue(int nSize)
2: :m_pBuffer(NULL)3: ,m_nSize(nSize)4: ,m_nIn(0)5: ,m_nOut(0)6: {7: //round up to the next power of 2
8: if (!is_power_of_2(nSize))
9: {10: m_nSize = roundup_power_of_two(nSize);11: }12: }13:14: UnlockQueue::~UnlockQueue()15: {16: if(NULL != m_pBuffer)
17: {18: delete[] m_pBuffer;
19: m_pBuffer = NULL;20: }21: }22:23: bool UnlockQueue::Initialize()
24: {25: m_pBuffer = new unsigned char[m_nSize];
26: if (!m_pBuffer)
27: {28: return false;
29: }30:31: m_nIn = m_nOut = 0;32:33: return true;
34: }35:36: unsigned long UnlockQueue::roundup_power_of_two(unsigned long val)
37: {38: if((val & (val-1)) == 0)
39: return val;
40:41: unsigned long maxulong = (unsigned long)((unsigned long)~0);
42: unsigned long andv = ~(maxulong&(maxulong>>1));
43: while((andv & val) == 0)
44: andv = andv>>1;45:46: return andv<<1;
47: }1.在构造函数中，对传入的size进行2的次幂圆整，圆整的好处是可以将m_nIn % m_nSize 可以转化为 m_nIn & (m_nSize – 1)，取模运算”的效率并没有 “位运算” 的效率高。
2.在构造函数中，未给buffer分配内存，而在Initialize中分配，这样做的原因是：我们知道在new UnlockQueue的时候有两步操作，第一步分配内存，第二步调用构造函数，如果将buffer的分配放在构造函数中，那么就可能 buffer 就可能分配失败，而后面用到buffer，还需要判空。
三、UnlockQueue入队和出队操作
1: unsigned int UnlockQueue::Put(const unsigned char *buffer, unsigned int len)
2: {3: unsigned int l;
4:5: len = std::min(len, m_nSize - m_nIn + m_nOut);6:7: /*
8: * Ensure that we sample the m_nOut index -before- we
9: * start putting bytes into the UnlockQueue.
10: */
11: __sync_synchronize();12:13: /* first put the data starting from fifo->in to buffer end */
14: l = std::min(len, m_nSize - (m_nIn & (m_nSize - 1)));15: memcpy(m_pBuffer + (m_nIn & (m_nSize - 1)), buffer, l);16:17: /* then put the rest (if any) at the beginning of the buffer */
18: memcpy(m_pBuffer, buffer + l, len - l);19:20: /*
21: * Ensure that we add the bytes to the kfifo -before-
22: * we update the fifo->in index.
23: */
24: __sync_synchronize();25:26: m_nIn += len;27:28: return len;
29: }30:31: unsigned int UnlockQueue::Get(unsigned char *buffer, unsigned int len)
32: {33: unsigned int l;
34:35: len = std::min(len, m_nIn - m_nOut);36:37: /*
38: * Ensure that we sample the fifo->in index -before- we
39: * start removing bytes from the kfifo.
40: */
41: __sync_synchronize();42:43: /* first get the data from fifo->out until the end of the buffer */
44: l = std::min(len, m_nSize - (m_nOut & (m_nSize - 1)));45: memcpy(buffer, m_pBuffer + (m_nOut & (m_nSize - 1)), l);46:47: /* then get the rest (if any) from the beginning of the buffer */
48: memcpy(buffer + l, m_pBuffer, len - l);49:50: /*
51: * Ensure that we remove the bytes from the kfifo -before-
52: * we update the fifo->out index.
53: */
54: __sync_synchronize();55:56: m_nOut += len;57:58: return len;
59: }入队和出队操作与kfifo相同，用到的技巧也完全相同，有不理解的童鞋可以参考前面一篇文章《眉目传情之匠心独运的kfifo》。这里需要指出的是__sync_synchronize()函数，由于linux并未开房出内存屏障函数，而在gcc4.2以上版本提供This builtin issues a full memory barrier，有兴趣同学可以参考Built-in functions for atomic memory access。
四、测试程序
如图所示，我们设计了两个线程，一个生产者随机生成学生信息放入队列，一个消费者从队列中取出学生信息并打印，可以看到整个代码是无锁的。
1: #include "UnlockQueue.h"
2: #include <iostream>3: #include <algorithm>4: #include <pthread.h>5: #include <time.h>6: #include <stdio.h>7: #include <errno.h>8: #include <string.h>9:10: struct student_info
11: {12: long stu_id;
13: unsigned int age;
14: unsigned int score;
15: };16:17: void print_student_info(const student_info *stu_info)
18: {19: if(NULL == stu_info)
20: return;
21:22: printf("id:%ld\t",stu_info->stu_id);
23: printf("age:%u\t",stu_info->age);
24: printf("score:%u\n",stu_info->score);
25: }26:27: student_info * get_student_info(time_t timer)28: {29: student_info *stu_info = (student_info *)malloc(sizeof(student_info));
30: if (!stu_info)
31: {32: fprintf(stderr, "Failed to malloc memory.\n");
33: return NULL;
34: }35: srand(timer);36: stu_info->stu_id = 10000 + rand() % 9999;37: stu_info->age = rand() % 30;38: stu_info->score = rand() % 101;39: //print_student_info(stu_info);
40: return stu_info;
41: }42:43: void * consumer_proc(void *arg)
44: {45: UnlockQueue* queue = (UnlockQueue *)arg;46: student_info stu_info;47: while(1)
48: {49: sleep(1);50: unsigned int len = queue->Get((unsigned char *)&stu_info, sizeof(student_info));
51: if(len > 0)
52: {53: printf("------------------------------------------\n");
54: printf("UnlockQueue length: %u\n", queue->GetDataLen());
55: printf("Get a student\n");
56: print_student_info(&stu_info);57: printf("------------------------------------------\n");
58: }59: }60: return (void *)queue;
61: }62:63: void * producer_proc(void *arg)
64: {65: time_t cur_time;66: UnlockQueue *queue = (UnlockQueue*)arg;67: while(1)
68: {69: time(&cur_time);70: srand(cur_time);71: int seed = rand() % 11111;
72: printf("******************************************\n");
73: student_info *stu_info = get_student_info(cur_time + seed);74: printf("put a student info to queue.\n");
75: queue->Put( (unsigned char *)stu_info, sizeof(student_info));
76: free(stu_info);77: printf("UnlockQueue length: %u\n", queue->GetDataLen());
78: printf("******************************************\n");
79: sleep(1);80: }81: return (void *)queue;
82: }83:84:85: int main()
86: {87: UnlockQueue unlockQueue(1024);88: if(!unlockQueue.Initialize())
89: {90: return -1;
91: }92:93: pthread_t consumer_tid, producer_tid;94:95: printf("multi thread test.......\n");
96:97: if(0 != pthread_create(&producer_tid, NULL, producer_proc, (void*)&unlockQueue))
98: {99: fprintf(stderr, "Failed to create consumer thread.errno:%u, reason:%s\n",
100: errno, strerror(errno));101: return -1;
102: }103:104: if(0 != pthread_create(&consumer_tid, NULL, consumer_proc, (void*)&unlockQueue))
105: {106: fprintf(stderr, "Failed to create consumer thread.errno:%u, reason:%s\n",
107: errno, strerror(errno));108: return -1;
109: }110:111: pthread_join(producer_tid, NULL);112: pthread_join(consumer_tid, NULL);113:114: return 0;
115: }运行结果：
-
Echo Chen：Blog.youkuaiyun.com/chen19870707
扫一扫
8159
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
