ltp测试套件-pthread_rwlock_rdlock_2-1分析记录

测试pthread_rwlock_rdlock_2-1.c用例失败。log如下：

[83#yuchen@ubuntu ltp]# gcc 2-1.c -o 21 -pthread [84#yuchen@ubuntu ltp]# ./21 main: has priority: 3 main: attempt read lock main: acquired read lock main: create wr_thread, with priority: 2 wr_thread: attempt write lock main: create rd_thread, with priority: 1 rd_thread: attempt read lock rd_thread: acquired read lock rd_thread: unlock read lock Test FAILED: rd_thread did not block on read lock, when a reader owns the lock, and a higher priority writer is waiting for the lock [85#yuchen@ubuntu ltp]#

 这个测试用例的流程：

Steps:
 * We have three threads, main(also a reader), writer, reader
 *
 * 1.  Main thread set its shcedule policy as "SCHED_FIFO", with highest priority
 *     the three: sched_get_priority_min()+2.
 * 2.  Main thread read lock 'rwlock'
 * 3.  Create a writer thread, with schedule policy as "SCHED_FIFO", and priority
 *     using sched_get_priority_min()+1.
 * 4.  The thread write lock 'rwlock', should block.
 * 5.  Main thread create a reader thread, with schedule policy as "SCHED_FIFO", and
 *     priority sched_get_priority_min()
 * 6.  Reader thread read lock 'rwlock', should block, since there is a higher priority
 *     writer blocked on 'rwlock'
 * 7.  Main thread release the 'rwlock', the writer should get the lock first
 */

用例失败的原因：

rwlock是读写锁，读者是可以递归调用的，因为多个读者不存在竞争问题，如果写者先拿到了锁，则后面来的读者必须阻塞。读和写之间必然会造成竞争，但是读和读之间没有竞争。读写锁的实现比较复杂，基于futex实现（用户态和内核态混合同步机制）。在glibc库有很多分支判断，如果对futex机制不了解的话，是比较难理解的。针对这个用例，简单说下流程：

默认情况下，如果没有写者持有锁，并且锁的属性设置了读优先（默认情况下rwlock就是读优先的），那么读者可以重复占用锁（读和读没有竞争）。代码见nptl/pthread_rwlock_rdlock.c，注意140行。那么在该测试用例中有2个读者，读者1先拿到了锁，读者2在去拿锁，是不会被阻塞的，因为写者并没有持有锁。持有锁的是一个读者（main主程序）。

但是测试用例期望，在有高优先级写者竞争锁的时候，读者要阻塞。从c库代码上看，要想让读者阻塞，必须设置rwlock写者优先属性，也就是让pthread_rwlock_rdlock.c的140行的条件判断为假。从而让代码走到slowfast，在pthread_rwlock_rdlock_slow函数中会挂起当前线程，等待其他任务unlock时才有可能被唤醒。

119 __pthread_rwlock_rdlock (pthread_rwlock_t *rwlock)
120 {
121   int result = 0;
122   bool wake = false;
123   int futex_shared =
124       rwlock->__data.__shared == LLL_PRI