Running the Make/CreateBuildResultsFolders script after a clone, or
after clobbering BuildResults/, was a pain. Now a clean build can be
guaranteed with:
rm -rf BuildResults
git checkout BuildResults
The names of the NanoKernel MP calls made by this 68k ROM library were
found by inspecting the PowerPC MPLibrary, which makes these and many
other MP calls. The document "Adding Multitasking Capability to
Applications Using Multiprocessing Services" gave some background
information. It also helped to determine the function signatures of the
MP calls and of these ROM wrappers.
The sources needed no changes (how embarrassing), except to remove an
unnecessary dependence on a nonexistent SoundPrivate.a. Lastly.lib had
an incorrect position for SIDESOUNDBUFFER, which has been fixed.
The unlinker's new ISLANDGUESS HIDE' option removes autogenerated BSR.L
from the binary dumps and makes tweaks that will cause them to be
regenerated by the linker. MPW's linker seems to have the correct branch
island placement policy, because round-tripping is preserved. When
building from edited sources, there is a remote possibility that the
arrangement of branch islands inside a dumped file will be changed,
disrupting references within that file.
Previously these were lumped in with Lastly. Now they are unlinked and
correctly labelled, and Lastly contains only a small amount of PowerPC-
specific code.
The unlink tool now REFGUESSes branches between object files, with some
manual guidance. If you patch the ROM enough to change its function
offsets, it should still work fine.
(One potential exception: if the linker needs to put a new "BSR.L"
island module *within* a pre-built object file, that module file's
internal references *may* be disrupted. This situation is fairly
unlikely, but with improved reversal it will be addressed.)
"DeVectorize" was a hideous several-thousand-line Python script that
split 70% of the ROM into barely passable MPW object files. The new
unlinker, "unlink", produces much nicer files and enjoys a clean
separation of code from data. Using DumpObj, you can get a very readable
listing of a given object file (GoNativeROMLib.o has improved a great
deal). With a bit more work, unlink will be superior to FDisasm for all
68k Mac ROM reversal.
That is to day, put *something* in its place with the right length and
symbol locations. Now we can be sure that no incorrect bytes are due to
confused branch instructions, and therefore that they must all be fixed.
The ExpansionBusDispatch trap (AAF3) is implemented in ROM using a
DispatchHelper table. Using a header from the embryonic Expansion Bus Mgr in
CubeE (rather suspect) and from the Universal Interfaces (solid), the names of
24 of the 47 dispatch selectors have been found. This commit updates the
dumped ExpansionBusMgr.lib with new symbol names.
All but two of the 47 dispatch selectors are Vectorized, so this commit also
updates VectorTable.a with selector numbers and, where possible, names. The
two non-Vectorized dispatch selectors (27 and 28) both point to the same 'nop'
function immediately before ExpansionBusDispatch.
