Valgrind的多线程调试工具
Helgrind是Valgrind的一个重点功能 本节主要针对与多线程基本安全问题进行检测：【所有的代码环境都是在POSIX_THREAD模式下】
写线程代码时 经常碰到如下问题
1) 资源不安全访问 【就是多个线程在没有同步的情况下写某个资源体】
2) 死锁问题
3) POSIX pthreads API的错误使用
4) 在前面几个基础上都能安全无误的情况下 多于多线程程序就是要能够能好将同步块尽量缩到最小 【这是一个很大的课题】
解决问题：
问题1: 调用Helgrind能够很好的解决掉 已基本例子为例：
-
#include <pthread.h>
-
int var = 0;
-
void* child_fn ( void* arg ) {
-
var++;
-
return NULL;
-
}
-
int main ( void ) {
-
pthread_t child;
-
pthread_t child2;
-
pthread_create(&child,NULL, child_fn, NULL);
-
pthread_create(&child2,NULL,child_fn,NULL);
-
pthread_join(child,NULL);
-
pthread_join(child2,NULL);
-
return 0;
-
}
明显var是共享的 不安全访问，调用Helgrind看看怎么能够检测出来
gcc -g thread_helgrind.c -o thread_helgrind -lpthread
valgrind --tool=helgrind ./thread_helgrind
可以看出valgrind弹出如下输出信息：
==25516== Helgrind, a thread error detector
==25516== Copyright (C) 2007-2013, and GNU GPL'd, by OpenWorks LLP et al.
==25516== Using Valgrind-3.9.0 and LibVEX; rerun with -h for copyright info
==25516== Command: ./thread_helgrind
==25516==
==25516== ---Thread-Announcement------------------------------------------
==25516==
==25516== Thread #3 was created
==25516== at 0x415B3C8: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516== ---Thread-Announcement------------------------------------------
==25516==
==25516== Thread #2 was created
==25516== at 0x415B3C8: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516== ----------------------------------------------------------------
==25516==
==25516== Possible data race during read of size 4 at 0x804A028 by thread #3
==25516== Locks held: none
==25516== at 0x804851F: child_fn (thread_helgrind.c:12)
==25516== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==25516== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==25516== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516== This conflicts with a previous write of size 4 by thread #2
==25516== Locks held: none
==25516== at 0x8048527: child_fn (thread_helgrind.c:12)
==25516== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==25516== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==25516== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516== ----------------------------------------------------------------
==25516==
==25516== Possible data race during write of size 4 at 0x804A028 by thread #3
==25516== Locks held: none
==25516== at 0x8048527: child_fn (thread_helgrind.c:12)
==25516== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==25516== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==25516== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516== This conflicts with a previous write of size 4 by thread #2
==25516== Locks held: none
==25516== at 0x8048527: child_fn (thread_helgrind.c:12)
==25516== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==25516== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==25516== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==25516==
==25516==
==25516== For counts of detected and suppressed errors, rerun with: -v
==25516== Use --history-level=approx or =none to gain increased speed, at
==25516== the cost of reduced accuracy of conflicting-access information
==25516== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)
可以看出绿色就显示的data_race错误 可以直接定位到var前后没有locks/
问题2:
死锁问题是尽量避免 对于helgrind可以检测出加锁解锁顺序出现问题导致的死锁问题 这个问题我们可以好好看下：
首先 看下一个正常的程序
#include <pthread.h>
pthread_mutex_t mut_thread;
int var = 0;
void* child_fn ( void* arg ) {
pthread_mutex_lock(&mut_thread);
var++;
pthread_mutex_unlock(&mut_thread);
return NULL;
}
int main ( void ) {
pthread_t child;
pthread_t child2;
pthread_mutex_init(&mut_thread,NULL);
pthread_create(&child,NULL, child_fn, NULL);
pthread_create(&child2,NULL,child_fn,NULL);
pthread_join(child,NULL);
pthread_join(child2,NULL);
return 0;
}
正常加锁解锁 没有问题
在看下连续加2次锁的情况：
#include <pthread.h>
pthread_mutex_t mut_thread;
int var = 0;
void* child_fn ( void* arg ) {
pthread_mutex_lock(&mut_thread);
var++;
pthread_mutex_lock(&mut_thread);
return NULL;
}
int main ( void ) {
pthread_t child;
pthread_t child2;
pthread_mutex_init(&mut_thread,NULL);
pthread_create(&child,NULL, child_fn, NULL);
pthread_create(&child2,NULL,child_fn,NULL);
pthread_join(child,NULL);
pthread_join(child2,NULL);
return 0;
}
看下这个helgrind打印出来的东西 【当然要杀死 不然会一直卡在那里动都不动一下】
==26534== Helgrind, a thread error detector
==26534== Copyright (C) 2007-2013, and GNU GPL'd, by OpenWorks LLP et al.
==26534== Using Valgrind-3.9.0 and LibVEX; rerun with -h for copyright info
==26534== Command: ./deadlock_helgrind
==26534==
==26534== ---Thread-Announcement------------------------------------------
==26534==
==26534== Thread #2 was created
==26534== at 0x415B3C8: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26534==
==26534== ----------------------------------------------------------------
==26534==
==26534== Thread #2: Attempt to re-lock a non-recursive lock I already hold
==26534== at 0x402E8E5: pthread_mutex_lock (hg_intercepts.c:507)
==26534== by 0x80485C6: child_fn (deadlock_helgrind.c:14)
==26534== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26534== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26534== Lock was previously acquired
==26534== at 0x402E95D: pthread_mutex_lock (hg_intercepts.c:518)
==26534== by 0x80485AD: child_fn (deadlock_helgrind.c:12)
==26534== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26534== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26534==
^C==26534== ----------------------------------------------------------------
==26534==
==26534== Thread #2: Exiting thread still holds 1 lock
==26534== at 0x4001182: ??? (in /lib/i386-linux-gnu/ld-2.17.so)
==26534== by 0x405B4D1: __lll_lock_wait (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x4056ED3: _L_lock_776 (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x4056D11: pthread_mutex_lock (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x402E914: pthread_mutex_lock (hg_intercepts.c:510)
==26534== by 0x80485C6: child_fn (deadlock_helgrind.c:14)
==26534== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26534== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26534== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26534==
==26534==
==26534== For counts of detected and suppressed errors, rerun with: -v
==26534== Use --history-level=approx or =none to gain increased speed, at
==26534== the cost of reduced accuracy of conflicting-access information
==26534== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0
Lock was previously acquired 对于自身线程来讲 这样的做法明显出问题 它自身还在尝试获取这个lock 这样就导致死锁的发生。可以定位到==26534== by 0x80485C6: child_fn (deadlock_helgrind.c:14)出问题了 所以这个helgrind解决这类问题时还是非常的厉害的~！
接下来看一个2 mutex导致的问题：
#include <pthread.h>
pthread_mutex_t mut_thread;
pthread_mutex_t mut_thread1;
int var = 0;
void* child_fn ( void* arg ) {
pthread_mutex_lock(&mut_thread);
pthread_mutex_lock(&mut_thread1);
var++;
pthread_mutex_unlock(&mut_thread);
pthread_mutex_unlock(&mut_thread1);
return NULL;
}
void* child_fn1(void *arg)
{
pthread_mutex_lock(&mut_thread1);
pthread_mutex_lock(&mut_thread);
var++;
pthread_mutex_unlock(&mut_thread1);
pthread_mutex_unlock(&mut_thread);
return NULL;
}
int main ( void ) {
pthread_t child;
pthread_t child2;
pthread_mutex_init(&mut_thread,NULL);
pthread_mutex_init(&mut_thread1,NULL);
pthread_create(&child,NULL, child_fn, NULL);
pthread_create(&child2,NULL,child_fn1,NULL);
pthread_join(child,NULL);
pthread_join(child2,NULL);
return 0;
}
加锁顺序导致死锁问题
==26785== Helgrind, a thread error detector
==26785== Copyright (C) 2007-2013, and GNU GPL'd, by OpenWorks LLP et al.
==26785== Using Valgrind-3.9.0 and LibVEX; rerun with -h for copyright info
==26785== Command: ./deadlock_helgrind
==26785==
==26785== ---Thread-Announcement------------------------------------------
==26785==
==26785== Thread #3 was created
==26785== at 0x415B3C8: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26785==
==26785== ----------------------------------------------------------------
==26785==
==26785== Thread #3: lock order "0x804A038 before 0x804A050" violated
==26785==
==26785== Observed (incorrect) order is: acquisition of lock at 0x804A050
==26785== at 0x402E95D: pthread_mutex_lock (hg_intercepts.c:518)
==26785== by 0x8048637: child_fn1 (deadlock_helgrind.c:22)
==26785== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26785== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26785== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26785==
==26785== followed by a later acquisition of lock at 0x804A038
==26785== at 0x402E95D: pthread_mutex_lock (hg_intercepts.c:518)
==26785== by 0x8048643: child_fn1 (deadlock_helgrind.c:23)
==26785== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26785== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26785== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26785==
==26785== Required order was established by acquisition of lock at 0x804A038
==26785== at 0x402E95D: pthread_mutex_lock (hg_intercepts.c:518)
==26785== by 0x80485ED: child_fn (deadlock_helgrind.c:13)
==26785== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26785== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26785== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26785==
==26785== followed by a later acquisition of lock at 0x804A050
==26785== at 0x402E95D: pthread_mutex_lock (hg_intercepts.c:518)
==26785== by 0x80485F9: child_fn (deadlock_helgrind.c:14)
==26785== by 0x402E5F6: mythread_wrapper (hg_intercepts.c:233)
==26785== by 0x4054D77: start_thread (in /lib/i386-linux-gnu/libpthread-2.17.so)
==26785== by 0x415B3DD: clone (in /lib/i386-linux-gnu/libc-2.17.so)
==26785==
==26785==
==26785== For counts of detected and suppressed errors, rerun with: -v
==26785== Use --history-level=approx or =none to gain increased speed, at
==26785== the cost of reduced accuracy of conflicting-access information
==26785== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 9 from 9
这是观察法得出的 加锁顺序出错导致了这种情况的发生。
posix_thread erro这里就不列举了 这个完全是看基础。
下篇待续