- Enable new heap code and cleanup the heap, releasing unused memory, before starting reactos
- Gives our kernel about 4MB additional memory, which - at the current leak rate - means around 8 minutes longer before it runs out of memory :)
- Using the temp heap for all temporary allocations should increase the amount of saved space
svn path=/trunk/; revision=55518
- Use the parse context to store whether or not the desktop object was created. If the object was not created, NtUserCreateDesktop should return immediately (this can happen if it didn't exist)
- Before this , if the desktop already exited, we opened the existing desktop and initialized it again. We also created another desktop heap, desktop window, etc.. This fact confused threads using this desktop and caused problems like the assertion we see in the test suite
svn path=/trunk/; revision=55517
Use MiNonPagedSystemSize when creating the system NP memory area instead of making assumptions about the memory layout
Finally the amd64 port boots into usermode and usetup is partly working, formats the harddisk but freezes, when copying files.
svn path=/trunk/; revision=55513
[SMSS]: Turn off all code other than setting up the pagefile.
There's a lot of debug prints, as soon as things seem stable after a few days, SMSS will be gone and SMSS2 will take over and DPRINT1s will mostly be gone.
svn path=/trunk/; revision=55509
Slightly reorganize architecture dependent headers, so that ASSERT can be used
[CRT]
Use some definitions only for vista+, since 2k3 msvcrt doesn't have them
svn path=/trunk/; revision=55504
[CSRSRV]: Fix many bugs in ServerSbApiPortThread. This function never worked as SMSS never actually connects to the SB API Port. Since SMSS2 does, a few bugs were discovered. Also, hack-plement the one SB API that is required for SMSS2 (SbpCreateSession). All we do is resume the thread we receive (Winlogon)... normally we'd also register a hard-error port, etc.
[CSRSRV]: If connecting with SMDLL to SMSS did not work (such as on my system where SMSS is no longer there), try using SMLIB and SMSS2 instead. This way CSRSS still works with both environments.
[CSRSRV]: Required rbuild/cmake changes to build with SMLIB.
svn path=/trunk/; revision=55500
Fix a FIXME in IopCreateArcNames.
This allows ntoskrnl to find where it boots from and where it is when performing remote boot.
Now ReactOS should boot a bit farther using PXE :-).
svn path=/trunk/; revision=55485
[SMSS2]: Implement the main LPC loop and handle all the supported APIs (all stubs for now). Also handle new connection requests and implement SmpHandleConnectionRequest.
[SMSS2]: Implement subsystem helper functions.
[SMSS2]: Use SmApiPort2 instead of Sm2ApiPort.
[SMSS2]: Rename the SMSRV_APIs not to conflict with the function names, nor with the client functions in smlib.
svn path=/trunk/; revision=55478
- Implement MiIsUserP*e inline functions to efficiently determine if a pte is a pte/pde/ppe or pxe for a user mode page
- Make MiDetermineUserGlobalPteMask work for _MI_PAGING_LEVELS >= 3
- In MiResolveDemandZeroFault, use MI_MAKE_HARDWARE_PTE only, it does all the necessary work.
svn path=/trunk/; revision=55472
[SMSS2]: Implement SmpLoadSubsystemsForMuSession which correctly creates a session, loads win32k.sys inside of it, and then detaches from the session. Winlogon.exe is set as the initial command (but not launched), and CSRSS is fake-started.
[CSRSRV]: No longer start win32k.sys, that's not your job.
[SMSS2]: Kill most debug prints that aren't errors.
svn path=/trunk/; revision=55467
- Modify the logic in MmArmAccessFault, so that faults on kernel mode addresses including page table addresses are handled in the first part and user mode addresses (VADs) in the second part. This works, because for the special case of page table addresses, the user mode and kernel mode part of the code end up doing the same thing and this simplifies the code.
- In the user mode part call MiCheckVirtualAddress early and bail out if no VAD is found, since we do not care about any other cases any more.
svn path=/trunk/; revision=55459
- Cleanup Win32kProcessCallback and Win32kThreadCallback
- Even though it works, trying to post a message in Win32kProcessCallback after the THREADINFO is destroyed can only cause trouble, so move it in Win32kThreadCallback
- There is no need to try opening a desktop and windowstation before winlogon initializes
- Handle errors properly when we fail to get a default window station and desktop for the new process
- Enable win32k syscall hook callbacks
- Rewrite SetThreadDesktop to update THREADINFO properly and handle errors properly
- Do not initialize the THREADINFO in GetW32ThreadInfo as it is now done properly (actually leave a small hack that updates pci->dwTIFlags)
- Add UserDbgAssertThreadInfo that asserts the integrity of THREADINFO, CLIENTINFO and CLIENTTHREADINFO. This is called by GetW32ThreadInfo and the syscall hook callbacks
svn path=/trunk/; revision=55458