A brief guide to creating a new scheduling plugin in LITMUS^RT (based on version 2012.2.)

As an example, we are going to implement a simple semi-partitioned EDF scheduler that migrates jobs among CPUs in a round-robin fashion every 500us of execution. While this is not an overly practical scheduler, it's a good example because it allows to demonstrate how to implement migrations, per-processor state, and custom scheduling timers.

Prerequisites

General Guidelines

Work in a checked-out git repository. Do not use the tarball from the web page. This will make keeping track of your edits and incorporating upstream changes much easier.

Commit early, commit often. Make many small commits a you go along. You can clean them up later using git rebase.

Do not commit to the master branch. This will make incorporating upstream changes much easier.

Follow the kernel coding standard defined in Documentation/CodingStyle, even when writing “throw away” code.

Step 0: Checkout and Compile the Kernel

First, obtain a copy of the LITMUSRT kernel.

$ git clone http://www.litmus-rt.org/src/litmus-rt.git

Next, obtain a copy of liblitmus, the corresponding userspace library

$ git clone http://www.litmus-rt.org/src/liblitmus.git
Cloning into liblitmus...
$ cd liblitmus/
$ ls
INSTALL  Makefile  README  arch  bin  inc  include  setsched  showsched  src  tests

Step 1: Dummy File

Create a new file and add it to the build system.

  1. Create the file litmus/sched_demo.c (all file names are relative to the kernel repository).

  2. Edit the file litmus/Makefile to add sched_demo.o to the list named obj-y.

  3. Compile the kernel to see if everything works.

Step 2: Dummy Plugin

Edit the file litmus/sched_demo.c to declare a dummy plugin (that implements no particular policy and that rejects all tasks).

Step 3: Define the Per-Processor State

In this step, we define the scheduler state on each processor.

Step 4: Augment the Per-Task State

Our scheduler needs to keep track of...

Step 5: Initialize the Scheduler Plugin

Step 6: Initialize new Tasks

Step 7: Enable Admission of Tasks

Step 8: Boot the Kernel in KVM

kvm -gdb tcp::3008 -smp 4 -hda /RTS/litmus-rt/work/kvm-images/bbb-litmus-rt.img -m 2000 -net nic,model=e1000 -net user -k en-us -kernel /home/bbb/ldev/litmus-rt/arch/x86/boot/bzImage -append console=ttyS0 ro root=/dev/hda1 no_timer_check  no_timer_check -nographic -redir tcp:2104::22

Step 9: Observe the Debugging TRACE() Log

Step 10: Trace Task Execution

Step 11: Visualize the Trace

Step 12: Record Scheduling Overheads

Step 13: Analyze Scheduling Overheads