OS161 waitpid

The waitpid system call in OS161 has implications in more places than just the files involved with adding a system call. The design of waitpid will have implications in the way you create and destroy threads, and how you exit threads cleanly.

Let us first go over the high level concepts of what the entire waitpid system will need to do.

PID Management

I will run on the assumption that you have created a way of getting the next available PID in your system along with the getpid system call. I will refer to the getting of the next PID as a function called get_new_pid.

So what does this system of allowing processes to wait on each other need to do.

First of all we need a way of getting the thread structure of a process based on it's PID. We need to do this because the semantics of waitpid state that we need to return some information about the process so the simplest way (although not the only) would be to have public members of the thread structure for waitpid's use. Additionally, the scheduler operates on thread structures and we will most likely need similar information in the scheduler and waitpid mechanisms so the thread structure seems like the simplest option.

So how can we get a thread structure from a pid? Data Structures are your friends! Hash Tables, Linked Lists, Binary Trees, Resizing Arrays, etc. Be creative, use something you are comfortable with and you know will have a stable implementation. For the purposes of this post I will refer to however you created this mechanism as get_thread_by_pid(). Along with this you will have things such as add_thread_by_pid(), remove_thread_by_pid(), etc. These are just standard methods for adding to your chosen data structure and removing

You are going to want to add_thread_by_pid() after get_new_pid() and when a thread is being destroyed you are going to need to remove_thread_by_pid() and allow the pid to be reclaimed.

Once this is implemented you will be able to get a thread structure based on a PID. The semantics of waitpid state that the parent must receive the child's exit code. A way of doing this would be to, during the exit system call, save the exit code into the thread structure and then get_thread_by_pid() and read from the thread structure in the thread that is waiting.

waitpid

Now lets talk about actually implementing waitpid. This call should not be that complicated, most of the code is just book keeping. You could implement this by recreating another piece of code and just sleeping on the address of the child thread and then having the child thread call thread_wakeupone (assuming you implemented something like this) when it exits. You could also choose to use a condition variable instead of recreating the CV code. You would probably want to store the CV object in the thread structure. Also remember that if you want to add something to the thread structure be sure to initialize it in thread_create.

One last thing that you need to worry about is a race condition when the thread exits. You will be trying to get_thread_by_pid at the same time that exorcise is going to be trying to delete that thread structure. One way you can get around this would be to add a new thread state that would represent a thread that has exited but is not yet a zombie because someone has been waiting on it. To do this you would also need to know if someone is waiting on you. You may also want to have a list of who is waiting on you, notify them in a FIFO order etc. You must conform to the semantics detailed in the man pages for OS161 and anything else is just going to be bonus. You would also need to modify thread_exit so that if it has any waiters it won't add the thread to the zombies array, waking all waiters first. This is the part that involves the biggest design decision. I hope that this gives you guidance on what exactly needs to be done but there is no way to be more specific without giving a solution. Just remember that you need to return your exit code to any waiters, handle memory correctly, and eliminate a race condition with exorcise.

This is a preliminary analysis from my standpoint so if there are any specific implementation problems people are having I can try and answer them in the comments.

To those who celebrate a very Happy Thanksgiving and to those who don't enjoy your long weekend! (Although I hope those who celebrate enjoy their weekends as well!)

- FlounderingZ