Bochs/bochs/TODO

This is the "roadmap" posted in the mailing list, augmented by 
comments from the mailing list and the irc chat. 
Anybody is welcome to work on any of these issues. Some of
these items are rather simple and can be implemented by single 
individuals. Other items are quite complex and development needs 
to be coordonated. So, if you want to contribute, please drop 
us a note in the mailing list, so you can get help or exchange 
ideas.
Christophe Bothamy.

0. Donations
Source Forge recently set up a donation system for hosted projects.
Should we accept donations ? What could we do with the money ? 
- give to EFF, FSF or other 
- fund Kevin to continue the work on plex86 so we can use it
- bounties for somebody write optimized win9x/NT/XFree/linux/*BSD 
  drivers for our vga/net/ide cards
- other ?

1. GPL source file in Bochs
Daniel Gimpelevich pointed out that rfbproto.h is released 
under the GPL.  This is incompatible with our license, so 
we need to rewrite/find a replacement this.
Also, we should make sure that contributed code / patches
are LGPL compatible.

2. PCI
2.1 I'd like to implement a PCI framework, that would take care 
of handling pci config regs (mmio enables/remaps, port io 
enables/remaps), offer entry points for the simulated pci chipset
(for example pci irq routing), handle embedded bios, and so on. 
I guess it will also need to take over the ISA I/O and
memory management. Ideally, it would be keeping compatibility
when pci is not compiled in.
(see http://marc.theaimsgroup.com/?l=bochs-dev&m=107165515315420&w=2)
2.2 The System BIOS will also be impacted. Although it has
basic BIOS32 service directory functions, we will also have to 
identify and initialize the connected cards, including optional
roms (copy to main memory and initialize), and build the correct 
PIR table in memory.
2.3 configuration file for pci
It would be nice to have a generic way to set the bus/device numbers
of pci devices in the configuration file.
2.4 host<->guest pci proxy
Being able to use a real pci device from inside Bochs would be a 
great feature of Bochs. It would ease reverse engineering of non
documented cards, or one could even use a real spare vga card.
Frank Cornellis has done a great job on this subject, and we began
integrating his changes. 
2.5 existing PCI device will need a slight rework to use the framework

3. Subdirectories in iodev
The iodev directory contains the various implemented iodevice.
With the new pci devices, new harddrives and new net access methods, 
it could be interesting to add new subdirectories like :
iodev/video/... --> for standard vga and new card emulation
iodev/disks/... --> for the ata/atapi classes, hd/cd classes and host accesses
iodev/net/...   --> for ne2k and host net access
isa and pci devices would be mixed in the directories, but this should
be manageable.

4. Speed
Speed (well lack of) is one of the biggest criticism made by users
who'd like to see Bochs run as fast as Virtual PC.
Paths we can explore to get more speed :
4.1 virtualization : plex86
4.2 dynamic translation : qemu, patch from h.johansson
4.3 optimizations of instructions path / host asm (Conn Clark has been
working on this, and an anonymous patch for win32 
http://sourceforge.net/tracker/index.php?func=detail&aid=867044&group_id=12580&atid=312580
)
4.4 multithreading. Conn Clark wrote :
Threading might be nice too, for those of us who have SMP/SMT machines. 
I have a patch from Mathis (who hangs out on the IRC channel all the 
time) that puts the video card interface in its own thread. It has 
troubles though that I have not resolved. It may also be easier to debug 
a threaded peripheral.
I also think that it might be possible to thread a chunk of the CPU 
emulation to improve performance on a SMP/SMT machine. Specificaly 
write_virtual_dword, write_virtual_word, write_virtual_byte, etc... 
might just be able to be threaded. I think the threading overhead might 
be less than the protection and address translation code. We would have 
to try it to find out. I'm also sure there can be some nasty hurdles to 
overcome.

5. CPU
5.1 This was asked in the ml, and I believe it's a good idea to provide
a configure switch to set the cpu model, for example :
--with-cpu-386
--with-cpu-486dx
--with-cpu-pentium
--with-cpu-pentium-mmx
--with-cpu-k6-iii
--with-cpu-amd64
and so on. The main difficulty here is to set up the list of features
by cpu model. I started such a list, available at 
http://cbothamy.free.fr/projects/bochs/CPU_Features.sxc
The configure script will then set up constants on features to compile in,
ISA, FPU, MMX, 3DNOW, SSE, etc... Most of the feature flags already 
exists in config.h, so this should be easy. It would also be a good
idea to clean up the cpuid function beased on those flags.
We also have to keep in mind that some features are also enablable
by the guest os.
Please note the all features are still not supported/complete in Bochs.
5.2 Stanislav thinks that configure --with-cpu-pentium-mmx --enable-cpu-level-4 
would create lots of conflicts in the generated config.h. He suggests
that we should write an external GUI configure script that would propose 
standard or custom cpus and would detect conflicts.

6. VGA
Several thigs we could use to improve SVGA emulation :
6.1 framebuffer by Christopher Nelson
6.2 S3 emulation in qemu / patch by tld 
6.3 Look at Trident emulation of dosemu 
6.4 patch for PCI CirrusLogic CLGD54xx by m-suzu
6.5 voodoo3 (specs http://v3tv.sourceforge.net/docs.php )

7. Random thoughts on disk emulation improvements :
7.0 IDE-Busmaster support (needs pci)
7.1 lba48 support
7.2 >32GiB disks
7.3 autodetection of disk size / geometry
7.4 uml cow disk image support
7.5 compressed disk image support
7.6 extend redolog-disk specification to add coherency check of the flat 
image file, by storing its fstat-mtime field in the redolog.

8. net
8.1 New devices ? are specs available anywhere ?
8.2 bootable ethernet rom ? 
see etherboot, Micheal Brown wrote :
This already works; you can build an Etherboot rom image with the pnic 
driver, specify it as an option ROM in bochsrc and it will boot.  I'm 
using this extensively at the moment in Etherboot development.
In the Etherboot project's CVS, in the contrib/bochs directory, you can
find a working bochsrc file and an up-to-date README with step-by-step 
instructions on getting this working.

9. Bios
9.1 our bios is quite heavy on stack space (notably during int13 functions).
This should be fixed. A partial patch exists 
see
http://sourceforge.net/tracker/index.php?func=detail&aid=832330&group_id=12580&atid=312580
Some parts can be rewritten in assembler, to save stack space and 
make code smaller.
We could define a reserved area in the BIOS where we can fill in PCI and SMP
specific data before starting the simulation or during POST. The same could
be done in the VGABIOS with the VBE mode list.
9.2 with pci and smp support we'll need to include several tables in the bios. 
We have two possibilities : dynamically build them during POST 
or statically build a collection of bioses, which can be selected 
at run time.
9.3 add "jump table placeholder" and log missing function calls in the bios. 
Check completness with Ralf Brown interrupt list.

10. LGPL VGABios
10.1 Stack space
As for the system bios, the vgabios is heavy on stack space, and 
should be fixed. Some parts can be rewritten in assembler, to save 
stack space and make code smaller. The size of the debug version 
of the VGABIOS is currently 30k and it will exceed 32k if we 
fill in some of the unimplemented features.
10.2 Video parameters table
There is a very nice parameter table in 3dfx banshee document 
http://www2.lm-sensors.nu/~lm78/pdfs/Banshee_2d_spec.PDF
see also http://www.xyzzy.claranet.de/dos/vgacrt.c

11. Optimized Guest drivers still needed : VGA, IDE, NET
We have a specific VGA driver for winNT/2K, but still
lack drivers for other OSes.

12. Plugin architecture
12.1 The plugin architecture can be reworked if we want to support
multiple similar devices like seral net or vga cards.
We currently have two "types" of plugins: "core" and "optional". 
Maybe we could add "classes" of plugins. The current version of 
Bochs supports the classes "display_library" and "io_device". 
New classes can be "config_interface", "net_lowlevel" and 
"sound_lowlevel"
12.2 Stanislav wrote :
Plugin architecture should be rewritten like real plugin architecture s.t.
Bochs VGA plugin for example will be real plugin. I mean that replacement 
of plugin dll in already compiled Bochs will replace Bochs VGA card and 
the new card will be detected automatically.
This will allow for example developing of plugins separately from Bochs.
12.3 Michael Brow wrote :
If the configuration interface is to be reworked, could we also make it so 
that plugins are self-contained, rather than needing to pollute main.cc 
with code for defining and parsing plugin-specific options

13. Save/Restore
We already have a specific branch in the cvs with a save/restore
framework. The work still needs to be completed.
I guess it would be useful to be able to stop and
restart the emulation, or create "checkpoints"  to restart the
emulation from a known state. We can also give a look on how
integrate commitable harddisks with the save/restore strategy.

14. USB support
Ben Lunt has been working on USB support. We should check with him
what is the current state of his work and if we can merge his code.
How can we add USB device, for example a mouse ?
The current version of Bochs supports only a PS/2 mouse. If the 
USB support is more complete the user should be able to select 
the mouse type (ps2, serial, usb).
Ben Lunt wrote :
I have, but it is not complete yet.  Currently, my goal and the
way it is coded, you simply have to increment the "number of devices"
and add a new USB_DEVICE struct.  Then you fill in the struct
with the values needed for that device: 18 bytes of descriptor, 255
bytes of config, strings, interface, etc.  One of the members of
this struct will be a pointer to a driver for that device.
I have two structs filled in.  One is a USB mouse.  Win98SE, as
a guest, recognizes the mouse, but does random movement and
clicks.  I haven't tracked this down yet.
The other device is a USB numpad.  Win98SE also detects the 
USB Composite Device under USB devices, two HID's, and 
a System Control and User Control device under HID's.  However,
there is a yellow mark on the HID's.  I think it has to do
with my config binary data.  I am still working on it.
However, the USB stack, command phase, data IO phase, and
status phase is complete.  You can successfully retrieve
the descriptor(s) and string(s), and set the devices address.

15. Patches / Bug reports
There are dozens of patches floating around. Some are outdated,
don't apply cleanly, are obsolete/unneeded. We could try to do 
some clean-up, and keep only relevent ones.
We should also clean up the SF bug tracker. Some bugreports are 
very old and we asked for more information with no response.

16. Small others things
16.1 add support for multiple boot devices 
16.2 add keyboard leds to GUIs
16.3 add activity leds (net, disks, floppies) to GUIs
Volker added a status bar to the win32 simulation window.
It can be used to display leds. The status bar still needs
to be implemented by other display libraries.

17. Positions
If you want to help without coding, here are available positions :
17.1 Webmaster : update website (Jan Bruun Andersen offered to help)
17.2 patch coordonator : look at incoming patches (sourceforge and 
mailing list) and upload / update in the cvs patches directory.
17.3 platform maintainers for macos / win32 
17.4 disk image maintainer : create and maintain our collection
of disk images. Usually, only the configuration file needs to be
updated, and old bios files have to be removed.  Some packages 
still contain very old bios files, they should definitely have
to be removed.

18. Bochs demo cd/dvd
With version 2.1, it is now technically possible to use disk images 
on a read-only media, with a journal files on a read/write media. 
It would be great to create a demo cd/dvd with executables for 
supported platforms, configuration files and read-only disk 
images, the journal files would be written in a temporary 
directory on the harddisk.

19. Other CPU architectures : arm, ppc 
This has been asked in the mailing list. I'm not really
interested, but other people might be. Should we propose to 
host the new CPUs code in our source tree, or should we let 
people fork ?

20. Config file and dynamic menu
20.1 Benjamen R. Meyer wrote :
I think we should rework the .bochsrc file to be more standard across all
devices. I like how the USB configuration is done in it, and think we should
put something similar together for everything else. In otherwords, create
something that can be easily used for everything, and make it easier to
configure in the process.
From what I can tell right now, most of the configuration lines are randomly
thrown together as each gets implemented or added, instead of having
something that is based on a standard approach to the configuration.
The result should be something that would be able to easily auto-configured
by another program (a configuration editor?) with minimal changes necessary
when new devices/features are added.
20.2  Franck Cornelis wrote : the config system needs some work... 
e.g. the main menu is static while it could be generated at run-time...
the main menu text lives somewhere in a file... while it should be generated 
at run-time by iterating the main menu objects

21. APIC and windows NT SMP.  Stanislav wrote :
Current Bochs SMP is almost unusable. I not succeed to boot any SMP OS I
have due to APIC problems. Bochs APIC implementation should be reviewed. 
In the patches folder we already have at least 3 patches for APIC, I hope 
one of them or their combination will fix the problem or approach my 
WINNT SMP boot tries.

22. add to the roadmap that we should find some ideas to check 
correctness of instructions emulation, especially system instructions.  
arithmetic instructions could be validated with simply random testing

23. lowlevel serial support for Windows.
Volker is currently working on this.

24. runtime configuration dialog box for Windows
Volker is currently working on this.

25. Parallel port
Conn Clark wrote :
I would like to see better parallel port support so I can use a dongle. 
This is something I would find very useful as it would mean I wouldn't 
have to boot back into windows ever again. I also recognize that this 
may require a kernel module be written, which is beyond my current 
skills.  I know others will find this useful as I have had to tell a 
few people that their parallel port driven peripherals that require a 
bidirectional parallel port won't work.