Difference between revisions of "Previous GSoC Projects"
|Line 1:||Line 1:|
= Projects 2008 =
= Projects 2008 =
Revision as of 18:37, 20 February 2010
VGA BIOS for Geode LX
This project's goal is to write a VGA BIOS (PCI option rom) for AMD Geode LX systems (such as the Linutop, Thincan or XO). There exists a free VGA BIOS but it knows nothing about real hardware. If you really want to kick the iron, this project could be enhanced to contain a complete infrastructure for including hardware initialization code for many different graphics cards.
USB Option ROM for SeaBIOS
SeaBIOS is our latest and greatest way to boot all kinds of different operating systems. It is a coreboot payload that implements 16bit BIOS interrupts as they are needed by nearly all boot loaders today. In the last year, SeaBIOS learned how to cope with coreboot ACPI, and how to boot off SCSI drives. One major feature that we're desperately lacking is USB stick/disk/cdrom booting from SeaBIOS. USB support for SeaBIOS should be implemented as a PCI option rom, using the libpayload USB stack. The USB stack currently supports UHCI controllers. Part of this project could also be to add OHCI and EHCI support to the USB stack in libpayload (not a requirement for participation, but would sure be nice!)
- BIOS Boot Specification
- BIOS Enhanced Disk Drive Specification
- Low Cost Embedded x86 Teaching Tool by Darmawan Salihun contains an Option ROM example
- Another Option ROM example
- USB Option ROM progess
AVATT part 2
AVATT is coreboot+Linux+KVM as hypervisor in the ROM. A first version was done during last GSoC, but there is still a pretty big TODO list.
- make the kvm userspace tool not to crash anymore. A possible solution would be to fix the TLS issues from the version of uClibc we currently use (daily snapshots from their SVN tree). This could even mean porting the uClibc-nptl branch to x86 if both of the x86 linuxthreads branches prove to be too hard to fix.
- user-friendly tool that can create and run virtual machines.
- automatically starting the virtual machines at boot.
- get the network to work in qemu since it fails with both coreboot v2 and v3.
- integrate the virt-manager daemon inside the ROM image, if it and its dependencies fit the remaining free space. This needs network support, to really be useful.
- fix compilation on x86_64 boxes by compiling everything in 64bit mode. We need a 64bit hardware anyway since the SVM instructions are available only on recent 64 bit boxes so this shouldn't matter too much, except for some extra wasted ROM space caused by the 64bit code. We can't cross-compile because we're not using a full toolchain, like buildroot does.
- keep the versions as up-to-date as possible but also compatible with each other
SCSI booting in coreboot
Currently coreboot can not boot from an arbitrary SCSI controller. There are two solutions for the problem:
- Use Linux and Kexec. This requires to keep the SCSI driver in the flash chip.
- Use x86emu/vm86/ADLO and the int13 method. This would allow to use the PCI option rom available on all modern SCSI controllers.
So we obviously need a solution based on the later. This could as well be implemented as a Linux program, as an intermediate payload, or as a shared library.
The code you are going to write needs to catch the int13 interrupt vector that the SCSI option rom installs and make it available to arbitrary (firmware/payload) code trying to load something from disk.
This is a coreboot v3 project.
All Virtual All The Time (AVATT)
The goal here is to build a system that comes up running Linux and KVM from power on reset. From that point the system could boot anything -- Windows, Linux, *BSD, Plan 9 -- anything that runs on x86 32 and 64-bit architectures. Coreboot would be integrated with a Linux kernel and initrd that had KVM built in. The initrd would include the minimal set of tools needed for starting new KVM virtual machine guests. Note that Linux booting from coreboot is a solved problem, using the buildrom tool, so the main effort here is to develop a minimal KVM infrastructure that can fit in a 2Mbyte FLASH part. Linux + X11 have been demonstrated in a 1Mbyte part, so we feel that this task is not impossible, but will be a terrific learning experience for a student, and will provide the community with a valuable resource when it is finished. The Xen and KVM communities have both asked for this capability for some time now, so there is a group of people ready to use this system when it is finished.
coreboot graphical port creator
In coreboot v2, every port to a new mainboard requires that you touch a lot of source files with only minimal changes. In version 3 we try to fix this issue and pack all mainboard specific information into a configuration file that we call the Device Tree Source (DTS). This Device Tree config file is a simple text file describing what static (non-detectable and/or soldered on) devices are used on the mainbard and how they are wired (SPD-ROM, Interrupt Routing, SuperIO, Northbridge, Southbridge, Hypertransport,..). It is mostly organized as a tree (with some special cases, Hypertransport allows cycles for instance)
The idea is to create a tool, based on the [www.eclipse.org/ Eclipse IDE], Swing, or your favourite portable toolkit, which allows you to drag and drop those components together and describe how they are wired.
This would be a great help for mainboard vendors that build mainboards of already supported components. No more reading of coreboot code would be required, but rather only the understanding of the hardware, and probably the mainboard schematics.
This is a coreboot v3 project. It requires good Java and/or Eclipse skills (or whatever toolkit/language you choose)
There are many, many "payloads" for coreboot these days: Linux, FILO, GRUB2, Tiano Core, Open Firmware, etherboot, and some more to count. All these payloads have a few functions in common that they use to read information from coreboot or change coreboot settings in NVRAM. It would be incredibly useful to unify all this code and enhance it, so that not every coreboot payload has to keep reinventing the wheel.
Booting Windows and other Operating Systems in LinuxBIOS 
The goal of this sub project is to figure out how to boot Windows Vista/XP/2003. There are three approaches that might proof successful:
- using a dedicated LinuxBIOS loader (ie. adapting ReactOS FREELDR)
- booting Windows on top of Linux using kexec/kboot
- fixing ADLO so that it boots Vista/XP and removing the mainboard dependencies in it's code.
- Some information on usage of bios services in Windows can be found here and here.
Port GRUB2 to work in LinuxBIOS
GRUB2 is going to be _the_ bootloader of choice in the forseeable future. As such, it could replace both Grub legacy and FILO, the LinuxBIOS hack for grub compatibility. FILO lacks many features that come with GRUB2 with no extra effort.
This task splits into four sub-problems:
- Add a target i386-linuxbios, next to i386-pc and i386-efi to the configuration process
- Add an IDE driver that does direct access instead of intXX calls
- Make the build process generate a single static ELF image, like it is done on Sparc
- Add support for reading the memory size from the LinuxBIOS table.
CMOS Config / Device Tree Browser Payload
Unlike other BIOSes, Linux has no such thing as a "CMOS setup". This does not mean that you can not configure it. There is a nice and small Linux command line utility called lxbios for that purpose. But people are often asking for a builtin config tool. Such a config tool could feature VGA graphics (maybe even VESA?), it should be easy to use, allow to browse information from LinuxBIOS' central structure: the device tree, and provide lxbios functionality with some sex appeal.
This is a LinuxBIOSv3 project.
Open Firmware payload for LinuxBIOS
Mitch Bradley from Firmworks, Inc. released the Open Firmware sources under a BSD license. The released code does work in LinuxBIOS, but could use some proper integration and testing on some hardware or in Qemu.
- The released Open Firmware code is very much optimized towards the OLPC. A lot of things don't work yet on other systems, such as using a graphical framebuffer. Therefore things in LinuxBIOS need to be changed. For example, if LinuxBIOS initializes a graphics mode, it should add a LinuxBIOS table entry that specifies the address of the framebuffer and the depth and resolution.
- Add words to view the LinuxBIOS table in OFW
- Add words to change LinuxBIOS CMOS settings from OFW
- For LinuxBIOSv3, the start address of the payload can be variable. This is a fundamental change to v2, and will make life a lot easier and LinuxBIOS a lot more flexible. OFW requires to know its in-rom address at build time. This needs to be fixed to a dynamic behavior
- Also, there's no good documentation on what features can be used and how they can be used. Like the graphical OLPC menu, the built-in web server.
- Get a STOP-A like behavior working in Linux
- Get Smart Firmware working with new compilers
- Get Smart Firmware working as a payload
- Enhance the Distributed and Automated Testsystem to work with FILO and OpenFirmware payloads, and add The Hayes ANS Forth testsuite.
Some parts might require cooperation with other GSoC projects:
- Boot from Non-OFW SCSI controllers by running their int13 handler
- Boot Windows XP/2003/Vista in OFW
This project might benefit from Forth skills.
GNUFI or TianoCore payloads
This project requires no hardware skills, but especially in case of TianoCore might require knowledge of Windows compilers (VC2005?)
Boot OpenSolaris, FreeBSD, NetBSD, OpenBSD or other free OSes
LinuxBIOS has (despite its name) been a little Linux centric. A nice project would be to analyze what it takes to get OpenSolaris, the BSDs or other free operating systems to work in LinuxBIOS, without the need for legacy emulation (ie. no ADLO)
Improve Linux as a BIOS 
There's a small project called  which creates a payload from a Linux kernel and some user space utilities. It has been written to work with the OLPC. This project could be enhanced to work on all supported LinuxBIOS motherboards.
Porting Flashrom utility to Windows 2000/XP
Flashrom is used to burn LinuxBIOS binary to flash chips in the target motherboards. It runs on Unix/Linux. In this project the flashrom utility is ported from Linux/Unix to Windows 2000/XP. The Windows port is called Winflashrom. The project is divided into two tasks. The first task is to port most of the user space flashrom source code to MinGW and the second task is to code a Windows device driver to provide direct hardware access for the user space code. The difficulty of this task is in providing a reliable Windows device driver for the user space code.