ftp://plex86.org/source-tarballs/plex86-2001_0514a.tar.gz
>From ChangeLog:
- "Kevin P. Lawton" <[EMAIL PROTECTED]>: Mon May 14 14:03:20 EDT 2001
Integrated a lot of new DT architecture (from dt-testbed/proto4) into
plex86. Eliminated the previous Software Instruction Virtualization
layer (SBE). Parts that are now integrated/added:
- Tcode (translated code) storage/allocation.
- Guest instruction address to tcode address sparse tables. These
are on a per-page basis.
- Reverse tcode address to guest instruction address sparse tables
for exception handling.
- Maintenance of two hash tables for high speed lookups. One
for guest offset (EIP) to tcode address, and one for linear
page address to page meta info index. These will be more important
when I integrate the branch translations.
- Invalidation of tcode based on writes to managed tcode pages.
- Revalidation of tcode pages across guest OS context switches.
- Arbitrary code expansion. We're not locked into maintain the
same page offsets anymore.
- Updated the 'PERFORMANCE' file with more related ideas/todos.
Parts not integrated yet. The only thing that wasn't integrated is
the branch translations. To make debugging easier, for this phase,
_all_ branches (even static intra-page branches) are translated as
INT3 and emulated by the monitor. This means performance will suck
until I add them. The code is prototyped already in dt-testbed/proto4,
and now the hash table management logic is tested and works. So this
should be reasonable to add in short time.
FreeDOS / Pragma Linux (kernel 2.0.33) / Win95 all work again.
I used 'conf/freedos', 'conf/pragma', and 'conf/win95'.
This release is for testing only. To decrease noise on the developers
list, test it only with host = {single processor Linux kernel}, and
guest = {FreeDOS, Pragma Linux} and guest = {Win95} if you already
have an installed disk image.
Performance will be very slow for this phase. This is OK and expected,
as every guest branch instruction is monitored/emulated. First,
I need your help to test this code. After a small amount of time
to shake out bugs, I will integrate the branch translation logic
from the prototype (phase#2), and things should get fun!
OK plex86 testers, go nuts! You can get the Pragma boot floppy image
and FreeDOS boot disk image files from the ftp site in the guestOS
directory.
The quicker we shake out problems, the quicker I can get on to phase#2.
Also, I just remembered. On my Mdk8.0 system with Linux kernel 2.4.3-20mdk,
I have to go through the process of loading the kernel module and the
user application twice, before it works. The 1st time it returns an
error. Think because the paging bits are set to Present==0 the 1st time.
I've been too busy to debug this, since it consistently works after the
1st go. Be my guest...
-Kevin
--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Kevin Lawton [EMAIL PROTECTED]
MandrakeSoft, Inc. Plex86 developer
http://www.linux-mandrake.com/ http://www.plex86.org/
