A simple OS kernel for research, teaching, and fun.
C Assembly Other
Latest commit db99a00 Jan 30, 2017 @dthain committed on GitHub Merge pull request #7 from kevinwern/os-x-compat
Generify Makefile and add build instructions for Mac


The Basekernel Operating System Kernel

Basekernel is a simple operating system kernel for research, teaching, and fun.

Basekernel is not a complete operating system, but it is a starting point for those who wish to study and develop new operating system code. If you want to build something different than Windows, Linux, or Multics, Basekernel may be a good place to try out your new ideas.

Basekernel can boot an Intel PC-compatible machine (or VM) in 32-bit protected mode, with support for VESA framebuffer graphics, ATA hard disks, ATAPI optical devices, basic system calls, and paged virtual memory. From there, it's your job to write filesystem code, a windowing system, or whatever you like.

Our goal is to keep Basekernel simple and easy to understand, so that others can build on it and take it in different directions. We encourage you to fork the code, give it a new name, and develop something completely new on top of it.

To be clear, this is raw low-level code, and there is no guarantee that it will work on your particular machine, or work at all. If you fix bugs or solve build problems, we would be happy to accept your contributions.

How to Run Basekernel

Compilation: Linux x86 machine

From a standard Linux X86 machine with the GCC compiler:

cd src

Compilation: OS X

Getting basekernel to work with OS X will require installing a few dependencies. The easiest way to do this is with MacPorts.

By default, running make attempts to use genisoimage, which no version of OS X has or supports. You will have to install cdrtools and use the near-identical mkisofs, then update your environment so make will handle it:

port install cdrtools
export ISOGEN=mkisofs

You can then attempt to build the image:

cd src

On recent versions of OS X, gcc links to LLVM GCC or clang. If you try running make with the defaults and compilation or linking doesn't work, it's easiest to just install the complete gcc and binutils for i386 ELF:

port install i386-elf-gcc
port install i386-elf-binutils

Then update your environment like before:

export CC=i386-elf-gcc
export LD=i386-elf-ld

Run make again and, if the tools were referenced correctly, it should finish successfully and create the image.

Running the image

You should have a file called basekernel.img, which is an image of a floppy disk that can be mounted in a virtual machine.

Next, set up a virtual machine system like VMWare, VirtualBox, or QEUM, and direct it to use that image.

If you are using QEMU, this command-line should do it:

qemu-system-i386 -fda basekernel.img 

You should see something like this:

(Hint: VirtualBox doesn't have floppy support enabled by default. With your VM stopped, go to Settings->Storage and add a new floppy controller, then you can mount the image.)

If you want to try it on a real machine, then write out a physical floppy disk like this:

dd if=basekernel.img of=/dev/floppy

Of course, nobody uses floppies any more. To cold boot a physical machine, then you want to write out an optical disk containing the boot image. That leads to your first little starter project:

Starter Project

Read up on the mkisofs tool, and figure out how to create a CD-ROM image that has basekernel.img embedded within it as a bootable floppy. Use that CD as a bootable image with your virtual machine, or write it out to a physical CD, and use it to boot a real machine.

More Tweaks and Projects

Now that you have the code running, here is a whole raft of ideas to try, starting some single-line code changes to more elaborate projects:

  • Modify main.c to display a custom startup message to welcome the user.

  • Tweak console.c to display things in a color of your choosing.

  • Examine the rtc.c module, and write a little code in main.c that will display a real-time clock.

  • Modify the infinite loop in main.c to display a kernel command prompt, and read some commands from the user to do things like show system statistics, display the current time, reboot, etc.

  • Add some functions for drawing lines and shapes to graphics.c, and then use in main.c to make an interesting status display for your operating system.

  • Modify the console code so that it scrolls the display up when you reach the bottom, rather than clearing and starting over.

  • Try writing to an address somewhere into the (unmapped) user address space above 0x80000000. Notice how you get a page fault, and the system halts? Modify the page fault handler to automatically allocate a fresh memory page and map it into the process each time this happens.

  • Implement proper memory resource management. Create some system calls that allow each process to request/release memory in an orderly manner, and map it into the process in a suitable location. Establish a policy for how many pages each process can get, and what happens when memory is exhausted.

  • Write some easily-recognizable data to a file, then mount it as a virtual disk image in your virtual machine of choice. Use the ATA driver to load and display the raw data on the virtual disk.

  • Write a read-only driver for the ISO-9660 (CDROM) filesystem, and use it with the ATAPI driver to browse and access an entire filesystem tree.

  • Write a read-write driver for a Unix-style filesystem with proper inodes and directory entries. Copy data from your CDROM filesystem into the Unix filesystem, which you can now save across boot sessions.

  • Implement loadable programs. Using either a filesystem or just a raw disk image file, load a user-level program compiled outside the kernel, and execute it. The simplest way is to compile straight to binary format with a load address of 0x80000000, and then fall through process_switch(). A more sophisticated way is to create an A.OUT or ELF format executable, and then write the code to load that into the kernel.

  • Create a display system for multiple processes. Building on the bitmapped graphics module, create a set of virtual displays that allow each process to display independently without interfering with each other.

  • Port the system to 64-bit mode. Start by modifying kernelcore.S to make a second jump from 32 to 64 bit mode, then update all of the types and kernel structures to accommodate the new mode.

  • Create a driver for the PCI bus, which enumerates all devices and their configurations.

  • Create an AHCI driver for SATA disks.

  • Create an OHCI or EHCI driver for USB devices.

Machine Organization

The following memory organization is used throughout the code. To the extent possible, these definitions have been collected in memorylayout.h, but changing any of them may require some some significant code surgery.

Physical Memory Layout

0000 0000 Reserved area for initial interrupt vector and BIOS data.
0000 7C00 (BOOTBLOCK_START) 512 bytes where bootblock.S is initially loaded.
0000 fff0 (INTERUPT_STACK_TOP) Initial location of kernel/interrupt stack
          until paging and user processes allocate their own stacks.
0001 0000 (KERNEL_START) Start of kernel code and data in kernelcore.S
0010 0000 (KMALLOC_START) Start of kernel memory managed by kmalloc().
0020 0000 (MAIN_MEMORY_START)  Start of memory pages managed by memory.c
???? ???? Location of the video buffer, determined by video BIOS at runtime.
          Care must be taken in memory allocation and pagetable setup
          to avoid stomping on this area.

Virtual Memory Layout

0000 0000 First 2GB of VM space for all processes is directly mapped to
          physical memory in kernel mode.  That way, kernel space is
          inaccessible in user mode, but kernel code can run correctly
          with paging activated.
8000 0000 (PROCESS_ENTRY_POINT) The upper 2GB of VM space for all processes
          is private to that process.  Each page of VM here is mapped to
          physical page allocated by memory.c.
ffff fff0 (PROCESS_STACK_INIT) The high end of the user space is designated
          for the user level stack, which grows down towards the middle
          of memory.


        Start   Length      Type    Privilege
Segment 1   0   4GB     CODE    SUPERVISOR
Segment 2   0   4GB     DATA    SUPERVISOR
Segment 3   0   4GB     CODE    USER
Segment 4   0   4GB     DATA    USER
Segment 5   tss sizeof(tss) TSS SUPERVISOR

For technical reasons, the X86 requires that segmentation be turned on whenever paging is used. A few key places in interrupt handling and system startup require the use of segments. So, we set up four segments that span all of memory, and use the one with the desired privilege wherever a segment is required. The fifth segement identifies the Task State Structure (TSS) which is where the initial setup information for entering protected mode is stored.


The [OS Development Wiki][1] is an excellent source of sample code and general orientation toward the topics that you need to learn. However, to get the details of each element right, you will need to dig into the source documents for each component of the system:

The [Intel CPU Manuals][2] are the definitive CPU reference. Especially focus on the first five chapters of the * System Programming Guide * to fully understand the virtual memory model and exception handling.

The doc directory contains a few hard to find PDF specifications. While they describe older hardware, standardized hardware has a high degree of backwards compatibility, so these techniques continue to work.

[ATA Attachment-3 Interface][3] describes the interface to traditional ATA disks, including both I/O registers and commands interpreted by the disk.

[PhoenixBIOS Programmer's Guide][4] describes the BIOS functions available in the early boot environment, while the system is still in 16-bit mode. These are used by bootblock.S and kernelcore.S to load data from disk and initialize the system.

[VESA BIOS Extension (VBE)][5] describes extensions to the BIOS interrupts specifically for setting and querying graphics modes. This is used by kernelcore.S just before jumping to protected mode.

[Super I/O Controller][6] describes a typical chipset that provides basic support devices such as keyboard, serial ports, timer, and real-time clock. These can be accessed in any processor mode. While your machine may not have this specific chipset, the basic functions are widely compatible.

[1] http://wiki.osdev.org [2] http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html [3] doc/ata-3-std.pdf [4] doc/phoenix-bios.pdf [5] doc/vesa-bios.pdf [6] doc/super-io.pdf