This program demonstrates the technique of using priority inheritance (PI) mutexes with processes instead of threads. The way to do this is to obtain some shared memory - in this case with mmap that backs a pthread_mutex_t since this will support PI. Pay particular attention to how this is intialized to support processes. Function init_shared_pthread_mutex() does this by setting the pthread_mutexattr to PTHREAD_PROCESS_SHARED and the mutex protocol to PTHREAD_PRIO_INHERIT. In this program we purposely try to invoke a classic priority inversion. A low priority process grabs the mutex and does some work. A high priority process comes a long and is blocked since the mutex is taken. A medium priority process that doesn't require the mutex then takes the processor. Because the processes are restricted to one cpu, the low priority processes never makes any progress because the medium priority process runs in an infinite loop. This is a priority inversion because the medium priority process is running at the expensive of the high priority process. However, since we have used PRIO_INHERIT and are running on a machine that supports preemption, the high priority process will lend it's priority to the low priority process which will preempt the medium priority process. The low priority process will then release the mutex which the high priority process can obtain. When the high priority process gets to run it kills the medium priority process. The state structure keeps track of the progress. Although this program is set up to likely trigger an inversion, there is no guarantee that scheduling will make that happen. After the program completes it reports whether a priority inversion occurred or not. In either case this program demonstrates how to use priority inheritance mutexes with processes. In fact, you would be better off to avoid scenarios in which a priority inversion occurs if possible - this program tries to trigger them just to show that it works. If you are having difficulty triggering an inversion, merely increase the time that the low priority process sleeps while holding the lock. (usleep); Also note that you have to run as a user with permission to change scheduling priorities.
pip_stress was written by John Kacur <jkacur at redhat.com>
This manual page was also written by John Kacur