Assignment: Preemption and Synchronization

From Embedded Xinu
Revision as of 07:29, 12 December 2008 by Akoehler (talk | contribs) (New page: Category:Student Built Xinu = Preemption and Synchronization = This assignment is a Xinu assignment allowing the student to grow a more firm understanding of how an operating syste...)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Preemption and Synchronization

This assignment is a Xinu assignment allowing the student to grow a more firm understanding of how an operating system works. This assignment is part of the Student Built Xinu track for professors that are Teaching With Xinu. Written answers to the five analysis questions below should be put into a file called system/README and submitted with your code. The entire directory containing the operating system should be turned in during the submission process.

After completion of this assignment we will have a basic operating system with preemptive, priority scheduling of processes, and counting semaphores with wait queues as synchronization primitives. Explore how the system behaves with various numbers of producers and consumers at different levels of priority and with both blocking and non-blocking signals. You will be responsible for making predictions about how such a system behaves in subsequent projects. As an added challenge, you might like to try implementing some of the other classic synchronization problems outlined in the textbook.


First, make a fresh copy of your work thus far.

  cp -R <old Xinu directory> <new Xinu directory>

Untar the new project files on top of this new directory:

  tar xvzf <tar-ball location>

You should now see the new project files in with your old files. Be certain to make clean before compiling for the first time.


The new files system/clock* (as well as some adjustments to various loader configuration files) will provide you with basic preemption, as discussed in class. Take time to familiarize yourself with the contents of these files, as you will be responsible for understanding how these components of the operating system work.

How can you test that preemption is working in your system? Create a main program that demonstrates preemptive scheduling. Call it main-preempt.c, and make sure that it works when copied over main.c.


In a system with preemption, we must now guard against being interrupted while in the middle of an atomic operation. There are many points in operating system code where an unfortunately timed interrupt could leave the system in an inconsistent state.

For example, in the resched() function, there are several lines of C code (which translate into many more lines of machine code,) between the point where the outgoing running process is put back into the queue of ready processes and the point where an incoming process is dequeued and set to running. If an interrupt were to take place in the midst of this transition, the interrupt handling code might see the system in an inconsistent state, in which there doesn't seem to be a currently running process. For this reason, we consider process rescheduling to be atomic with respect to the rest of the operating system. The easiest way to enforce this is to temporarily disable interrupts during a critical, atomic section of code, and then reenable them when done.

The new file system/intr.S contains functions enable(), disable(), and restore() for manipulating the master bit that enables processor interrupts. From this point forward, these functions must be used judiciously to guard atomic sections in the operating system functions you write.


Implement classic semaphores with waiting queues. A definition of the basic semaphore structure is given in include/semaphore.h, and is initialized in the new system/initialize.c.

Fill in the functions in system/newsem.c, system/signal.c and system/wait.c. A simple main.c is given that implements a trivial case of the Producers/Consumers Problem.


  • Create a main program that spawns multiple producer processes (each producing a distinctive character output) and multiple consumer processes. What happens, and why? (Put your answers in system/README.) Save this test case as system/main-q1.c.
  • Using the same test case as in the previous question, briefly adjust the value of the preemption counter ("QUANTUM" in include/clock.h) to be 100 mS. How does this change the output? (Put your answers in system/README.)
  • Create a test case system/main-q3.c in which the consumer processes are all of a higher priority than the producer processes. Repeat the first two questions.
  • In the comment blocks for system/signal.c, you were directed to reschedule after taking a waiting process off of the semaphore's wait queue. What happens if you call only ready() when taking a process off of the wait queue, and do not voluntarily yield at that time? (This is called non-blocking or non-yielding signal().) Repeat the first three questions with non-blocking signal().
  • One of the interesting things about the classic Producer/Consumer problem is how finely balanced the solution turns out to be. A variety of simple one-line swaps in either the producer code or the consumer code can produce a system of processes that readily deadlocks. Find one of these combinations, and save it as system/main-q5.c