[NTOS]: Knowing if this change causes issues will help later in determining regressions due to TEB/PEBs mapped as VADs by ARM3, and regressions simply due to the change in VA layout.
[NTOS]: When implemented, the VAD mapping for PEB/TEB will only use that ~1MB, which yes, will limit ReactOS processes to each have only 256 threads. That is obviously a temporary limitation, one I doubt we'll even hit, but I'm putting it out here so you know.
svn path=/trunk/; revision=48176
[NTOS]: MI_MAKE_HARDWARE_PTE becomes MI_MAKE_HARDWARE_PTE_KERNEL, since it assumed this. MI_MAKE_HARDWARE_PTE is now a "generic" you can use when you don't know what the PTE should be. It uses MiDetermineUserGlobalMask to set the right bits.
[NTOS]: Add two more helpers: MI_IS_PAGE_TABLE_ADDRESS and MI_IS_SYSTEM_PAGE_TABLE_ADDDRESS. One is in the symbols, the other I made up to make things clearer.
[NTOS]: MiResolveDemandZeroFault now knnows how to resolve user-demand-zero-faults.
[NTOS]: Implement MiZeroPfn to do the actual zeroing during user-demand-zero-faults (also later for VAD faults).
svn path=/trunk/; revision=48175
[RTL]: Uncouple generic table from AVL table implementation into its own avltable.c
[RTL]: Get rid of "Austin" and fix prototypes of AVL table functions.
[RTL]: Re-implement AVL table functions, sharing as much code as possible with the SPLAY tree implementation which is pretty decent. Lookup, insert, enumeration are implemented, but not delete.
[RTL]: Make large part of the RTL AVL package into its own "support" file that can work both with MMADDRESS_NODE and RTL_BALANCED_LINKS structures. The former is used by ARM3 for VADs.
[NTOS]: Implement basic VAD AVL tree routines (Insert, LookupEmpty, GetPrevious, CheckForConflict, Locate). This is enough to insert VADs, find a free address range, and locate a VAD by address. No delete yet
Thanks to Timo Kreuzer for some clever definitions, Knuth for his genius, several online C implementations for ideas, the HPI kernel blog for insight on how Windows does it, and others.
svn path=/trunk/; revision=48173
- Add a number of amd64 specific math functions
- add generic c versions of cos and sin
- Remove leading underscores from amd64 symbols
svn path=/trunk/; revision=48161
WRC doesn't like (x+x+x+2+x+x+x+x+x+x), use (x+x+x+x+x+x+x+x+x+2) instead. To figure out the reason is left as an exercise for the reader.
svn path=/trunk/; revision=48140
- reactos.diff was pushed upstream, so it's not needed anymore.
- Add a missing section.
- Include sect_attribs.h into tlssup.c
svn path=/trunk/; revision=48130
For amd64 builds, invoke wrc directly, instead of using gcc to preprocess the files. The reason is that newer gcc versions (like 4.5.0) don't support multiline continuation (backslash-newline) except for preprocessor directives, but that breaks wine resource strings. This might have to be done for i386 builds too, sooner or later.
svn path=/trunk/; revision=48128
Add a tag to NDR_SCONTEXT struct. MS headers don't have it, but gcc 4.5.x requires it, because without the tag, the struct would be from an unnamed namespace and can't be shared by different compilation units in C++. Yeah whatever...
svn path=/trunk/; revision=48125
Convert freeldr and setupldr to PE format.
Previously freeldr was a raw binary file, which made handling by the bootsector very easy, but it disqualified it from proper debugging with gdb using symbols. This is possible with having unstripped PE files.
As we don't have any space to do proper PE loading from the bootsector (I already had to trim some strings to get enough space for the new jump code), we need to make sure, that the PE file doesn't contain a .bss section, which is achieved by a linker script. The next thing is to make sure, we don't have any symbols in the output file, because they would make freeldr too big to be loaded into memory and they are useless anyway. On the other hand we like to keep the symbols in the .nostrip.sys files if requested, as this is the primary purpose. This is in theory not a problem, as we could simply strip the file at the end, but binutils throw a monkey wrench in our plans: both strip and objcopy disrespect the file alignment and create unaligned sections, that don't naturally match their VAs. This is solved by hacking rbuild to do invoke ld 2 times, one time without and one time with the symbols (if requested).
Now the bootsectors also got some changes: instead of jumping to the loading address (0x8000) they get the address of the entry point from the image optional header. This is slightly simplified, by assuming the NtHeader begins at offset 0xE0. This finally allows source level debugging of freeldr with gdb.
svn path=/trunk/; revision=48124
- rewrite UserDrawIconEx, taken from yarotows, with same small modifications.
Fixes a whole bunch of user32:cursoricon tests
svn path=/trunk/; revision=48123
