Home ›

Linux KVM as a Learning Tool

Issue 186

From Issue #186
October 2009

Oct 01, 2009  By Duilio Javier Protti
 in
Low-level system programming is a difficult task, but with Linux KVM, it's a whole lot easier.

Listing 5. Linker Script kernel16.lds

OUTPUT_FORMAT(binary)

SECTIONS {
        . = 0;
        .text : { *(.init) *(.text) }
        . = ALIGN(4K);
        .data : { *(.data) }
        . = ALIGN(16);
        .bss : { *(.bss) }
        . = ALIGN(4K);
        .edata = .;
}

The SECTIONS command controls how to make the mapping and how to place the output sections in memory. Directives follow the syntax: 

.output-section : [optional-args]
                  { input-section, input-section, ... }

The kernel16.lds script sets the current location at offset 0x0. Then, the output .text section will start there and will contain the contents of any .init and .text input sections.

Next, we align the current location to a 4KB boundary and create the .data and .bss output sections. Use kernel16.lds to generate the kernel image as shown in Listing 6.

Listing 6. Building a 16-Bit Kernel Image

$ gcc -nostdlib -Wl,-T,kernel16.lds kernel1.S -o kernel1
$ ls -oh kernel1
-rwxr-xr-x 1 djprotti 64K 2008-10-17 19:09 kernel1

The -nostdlib flag avoids linking the standard system startup files and libraries (these will not be available inside our virtual machines). After this, we have our 64Kb 16-bit real-address kernel image.

How to Test It All

The Makefile in Listing 7 contains the commands to build both the kernel and the launcher.

Listing 7. Makefile

# If KVM was compiled from sources and you have errors about
# missing asm/kvm*.h files, copy them from
# kvm-XX/kernel/include/asm/* to {prefix}/include/asm/
CC=gcc
KERNEL16_CFLAGS=-nostdlib -ffreestanding -Wl,-T,kernel16.lds

all:    launcher kernel1

launcher: launcher.o
        $(CC) launcher.o /usr/lib/libkvm.a -o launcher

launcher.o:

kernel1: kernel1.S
        $(CC) $(KERNEL16_CFLAGS) kernel1.S -o kernel1

clean:
        rm *.o launcher kernel1

Launch the virtual machine with kernel1 as guest with the following command: 

$ ./launcher kernel1

If everything goes well, you will see no output, and the guest kernel should be consuming all of its available CPU. If you run the top command in another console, and you see output similar to that of Listing 8 (100% CPU usage for the launcher process), you have your kernel running in your first KVM virtual machine!

Listing 8. Output of top While Our Launcher Is Running

 PID USER     S %CPU %MEM    TIME+  COMMAND
8002 djprotti R  100  0.8   1:53.19 launcher
7428 djprotti S    0  0.8   0:04.45 gnome-terminal
8005 djprotti R    0  0.0   0:00.02 top
   1 root     S    0  0.0   0:03.92 init
   2 root     S    0  0.0   0:00.00 kthreadd
   3 root     S    0  0.0   0:00.12 migration/0
   4 root     S    0  0.0   0:02.76 ksoftirqd/0
   5 root     S    0  0.0   0:00.01 watchdog/0
An Improved Kernel

Now, let's build a kernel that communicates with the world. First, choose one of the I/O ports and use it to implement a “serial port”. Name the chosen port as IO_PORT_PSEUDO_SERIAL (as shown in Listing 10), then modify the outb callback in the launcher to interpret bytes sent to this port as characters printed to a serial console, and redirect them to launcher's standard output as shown in Listing 9.

Listing 9. Pseudo-Serial Port Implementation in launcher.c

#include "runtime.h"

static int my_outb (void *opaque, uint16_t addr, uint8_t data)
{
    if (addr == IO_PORT_PSEUDO_SERIAL)
        if (isprint(data) || data == '\n')
            putchar(data);
        else
            putchar('.');
    else
        printf("outb: %x, %d\n", addr, data);
    fflush (NULL);

    return 0;
}

Then, build a second kernel (kernel2) whose only task is to print “Hello\n” to its pseudo-serial port and then halt, as shown in Listing 10.

Listing 10. kernel2.S

#include "runtime.h"

.code16
start:
    mov    $0x48,%al    // H
    outb   %al,$IO_PORT_PSEUDO_SERIAL
    mov    $0x65,%al    // e
    outb   %al,$IO_PORT_PSEUDO_SERIAL
    mov    $0x6c,%al    // l
    outb   %al,$IO_PORT_PSEUDO_SERIAL
    mov    $0x6c,%al    // l
    outb   %al,$IO_PORT_PSEUDO_SERIAL
    mov    $0x6f,%al    // o
    outb   %al,$IO_PORT_PSEUDO_SERIAL
    mov    $0x0a,%al    // new_line
    outb   %al,$IO_PORT_PSEUDO_SERIAL

    hlt                 // halt the processor

. = 0xfff0
    ljmp    $0xf000, $start
______________________

Comments

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Exit instead of hlt

Curious's picture
Submitted by Curious (not verified) on Mon, 12/14/2009 - 16:44.

Thanks for this excellent article. I've tested out the sample but the program never returns after the kvm_run() call. I guess this is because the vcpu is halted in the last instruction of the test program. But how do I exit the KVM altogether and resume execution from kvm_run() onwards? Is there any documentation for libkvm somewhere I can consult?

This appears to be the libkvm in question, no?

Ben Scherrey's picture
Submitted by Ben Scherrey (not verified) on Mon, 11/02/2009 - 13:50.

http://www.linux-kvm.org/page/Code

See the userspace git tree. Is this the lib we need to be building to follow this article? I've been looking to do something like this with KVM for a while to create my own forth-like environment without having to screw around with low level hardware (or at least put it off til something interesting already works). Look forward to the follow up article. When is it scheduled?

-- Ben Scherrey

LibKVM on Ubuntu? Nonesuch...

basicman's picture
Submitted by basicman on Fri, 09/25/2009 - 19:08.

My attempt to follow this tutorial died on page 2, when it announced that all of the examples would be using the LibKVM library. Several hours of searching on Google and I've found nothing; no way to install or use the LibKVM library unless I'm on BSD.

If anyone has a workaround for this, I'd love to hear it. I was really looking forward to following the tutorial.

Different libkvm

Mitch Frazier's picture
Submitted by Mitch Frazier on Sat, 09/26/2009 - 12:40.

That's a different library. You need to install the qemu-kvm-devel package. Note you will probably have to get it directly from sourceforge since it does not appear to be in the Ubuntu repos. I don't use Ubuntu much so maybe I'm overlooking something, however, I know that openSUSE does not have a package for it either (at least in the standard places at 11.0). Make sure that you get the version that corresponds to the version of kvm that you have installed. For example, here is a link to the -devel package for release 88.

If you have to install it from sourceforge some fiddling around will probably be required. First, you'll have to build it, then install it, and then potentially modify your include/library paths to find the needed items.

Mitch Frazier is an Associate Editor for Linux Journal.

I used REHL5.5 installed devel package for release 88

jniu's picture
Submitted by jniu on Thu, 08/19/2010 - 04:59.

I used REHL5.5 installed devel package for release 88, but still cannot find libkvm.h and libkvm.a, the installation was successful, can you give me some suggestion? thanks.

Webcast
More Resources
How to Build an Optimal Hadoop Cluster to Store and Maintain Unlimited Amounts of Data Using Microservers

Realizing the promise of Apache® Hadoop® requires the effective deployment of compute, memory, storage and networking to achieve optimal results. With its flexibility and multitude of options, it is easy to over or under provision the server infrastructure, resulting in poor performance and high TCO. Join us for an in depth, technical discussion with industry experts from leading Hadoop and server companies who will provide insights into the key considerations for designing and deploying an optimal Hadoop cluster.

Learn More

Sponsored by AMD

White Paper
More Resources
Private PaaS for the Agile Enterprise

If you already use virtualized infrastructure, you are well on your way to leveraging the power of the cloud. Virtualization offers the promise of limitless resources, but how do you manage that scalability when your DevOps team doesn’t scale? In today’s hypercompetitive markets, fast results can make a difference between leading the pack vs. obsolescence. Organizations need more benefits from cloud computing than just raw resources. They need agility, flexibility, convenience, ROI, and control.

Stackato private Platform-as-a-Service technology from ActiveState extends your private cloud infrastructure by creating a private PaaS to provide on-demand availability, flexibility, control, and ultimately, faster time-to-market for your enterprise.

Learn More

Sponsored by ActiveState