OpenVSA: Difference between revisions
Whiterocker (talk | contribs) mNo edit summary |
|||
Line 14: | Line 14: | ||
$ svn co svn://coreboot.org/openvsa/trunk/openvsa | $ svn co svn://coreboot.org/openvsa/trunk/openvsa | ||
You can browse the source via http://www.coreboot.org/viewvc/trunk/openvsa/?root=OpenVSA | |||
== Differences Between VSA and OpenVSA == | == Differences Between VSA and OpenVSA == |
Revision as of 19:19, 14 May 2008
VSA, or Virtual System Architecture is a low-level software library included in the bootloader/BIOS for system using AMD Geode-series CPUs and companion chips.
AMD released VSA sources under the name geode-vsa in 2006 (licensed under the terms of the GNU LGPL). Those sources were hosted by the OLPC project, and can be pulled with git from git://dev.laptop.org/geode-vsa
. The OpenVSA sources include modified geode-vsa sources, as well as some new components also released under the GNU LGPL.
The VSA code runs under x86 SMM (System Management Mode) which is like "real mode" with some extra opcodes, priviledges, and side-effects.
As originally published, the VSA code compiled and assembled with older, commercially unavailable versions of Microsoft tools. The OpenVSA code has been modified in order to build under a GNU toolchain so that it may be maintained and enhanced by a wider group of users.
Status
The current status of the OpenVSA is: experimental. Please do not use OpenVSA in your system unless you have access to a low-level re-flashing tools.
Download
$ svn co svn://coreboot.org/openvsa/trunk/openvsa
You can browse the source via http://www.coreboot.org/viewvc/trunk/openvsa/?root=OpenVSA
Differences Between VSA and OpenVSA
Summary
Category | VSA | OpenVSA |
---|---|---|
Assembler | MASM 6.11c or greater | GNU gas (part of binutils) |
Make | NMAKE.EXE Version 1.40 or greater | GNU make |
C-compiler | MSVC Version 1.52 | GNU gcc |
Final binary output | exe2bin.exe | GNU objcopy (part of binutils) |
Assembly syntax | Microsoft/Intel | GNU gas/AT&T |
Code Generation | 16-bit, inherent to the toolchain commands used during build | 16-bit assembly, generated by using .code16 in assembly files; 32-bit from C, prefixes generated by using .code16gcc in C files
|
Memory Model | "tiny": merges CS and DS, inherent to the toolchain commands used during build | "tiny" model accomplished with specific section names and linker script statements |
SMM-only opcode assembly | MASM macros | Perl script smimac.pl pre-processes to constant-sequences
|
Internal assembly functions: calling convention | custom/random, no apparent fixed pattern | unchanged |
Assembly functions called from C: calling convention | Microsoft pascal
|
GNU __attribute__((fastcall))
|
C header file translation to assembly include | h2inc.exe | manual/static translation |
Calling Conventions And Stack
VSA: MASM and Microsoft C
The original VSA sources used 16-bit MS PASCAL
calling convention (__pascal
):
- stack parameter order: left-to-right
- called function cleans up the stack
- all parameters are pushed onto the stack
- return values to 16 bits returned in
AX
- return values 17 to 32 bits returned in
DX:AX
- scratch registers
AX, BX, CX, DX, ES
^ |<- 16 bits ->| | higher | address | (pop) ------------- | first param | ------------- ... ------------- | last param | 4(BP) ------------- | return IP | 2(BP) ------------- | old BP | <-- BP ------------- (locals) -2(BP) ------------- (saved regs) ------------- | (saved reg) | <-- SP ------------- | lower | address | (push) v
OpenVSA: GNU
With GNU on the Intel 386, the fastcall attribute (__attribute__(fastcall)
) calling convention:
- stack parameter order: right-to-left
- called function cleans up the stack
- if the first one or two arguments are integers/pointers (<=32bit), they are passed via
ECX
andEDX
- all other parameters are pushed onto the stack
- return values to 32 bits are returned in
EAX
- scratch registers
EAX, ECX, EDX
^ |<- 32 bits ->| | higher | address | (pop) (uses (no frame) frame) ------------- | last param | ------------- ... ------------- | 3rd param | 8(EBP) 4(ESP) ------------- | return EIP | <-- ESP ------------- | saved EBP | <-- EBP ------------- (locals) -4(EBP) ------------- (saved regs) ------------- (saved reg) <-- ESP | lower | address | (push) v
Notable C-Code Differences
Microsoft C appears to accept a statement like:
ULONG Data; ... (UCHAR)Data = 0;
This is not a portable construction, GCC rejects is, so has been changed to:
*((UCHAR *) &Data) = 0;
Feature Wish List
TODO.
Hacking Notes
TODO.
I, the copyright holder of this work, hereby release it into the public domain. This applies worldwide.
In case this is not legally possible: |