Intravision/reactos
1
0
Fork 0
mirror of https://github.com/reactos/reactos.git synced 2025-04-22 13:10:39 +00:00
Code Issues Projects Releases Packages Wiki Activity
reactos/ntoskrnl/mm/ARM3/procsup.c

2146 lines
65 KiB
C
Raw Normal View History

- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
/*
* PROJECT: ReactOS Kernel
* LICENSE: BSD - See COPYING.ARM in the top level directory
* FILE: ntoskrnl/mm/ARM3/procsup.c
* PURPOSE: ARM Memory Manager Process Related Management
* PROGRAMMERS: ReactOS Portable Systems Group
*/
/* INCLUDES *******************************************************************/
#include <ntoskrnl.h>
#define NDEBUG
#include <debug.h>
#define MODULE_INVOLVED_IN_ARM3
#include "../ARM3/miarm.h"
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* GLOBALS ********************************************************************/
ULONG MmProcessColorSeed = 0x12345678;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
PMMWSL MmWorkingSetList;
[MEGAPERF]: This build introduces the following performance boosts: - Using a dead stack lookaside list for up to 5 dead kernel thread stacks. 1500% improvement when rapidly destroying/creating threads, such as during second stage setup and many winetests. - Using a free pool page lookaside list for up to 12 free non-paged or paged pool pages. 800% improvement when allocating big pages from the pool, as well as during pool expansion. - Using a bucketized per-processor and local list (in the KPRCB) for block sizes between 1 and 32 bytes. 1000% improvement when rapidly allocating/freeing small pool allocations, and 8x reduction in pool fragmentation. svn path=/trunk/; revision=56000
2012-03-04 17:56:00 +00:00
ULONG MmMaximumDeadKernelStacks = 5;
SLIST_HEADER MmDeadStackSListHead;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
/* PRIVATE FUNCTIONS **********************************************************/
[NTOS]: MiRosTakeOverPebTebRanges now creates a small ~1MB ARM3 memory range on top of the ReactOS per-process VA. This does a couple of things: First of all, it changes the default PEB address to another static address. Still not dynamic like it will be soon, but at least it changes it a bit so we can test if anything breaks due to that. It also likewise changes the addresses of the TEBs (Shifted down by 1MB, basically). Finally, it blocks off that part of address space, which nobody should be using now, to see if anyone does indeed touch it. [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
2010-07-22 03:22:43 +00:00
VOID
NTAPI
[NTOSKRNL] - Modify MiRosTakeOverPebTebRanges to only create a memory area for the shared user page and rename it to MiRosTakeOverSharedUserPage. Previously it was creating a memory area for the whole region from USER_SHARED_DATA up to MM_HIGHEST_USER_ADDRESS, which is the majority of the x64 user mode address space and doesn't even contain the PEB/TEB. Those are allocated below the shared user page and get their memory areas created in MiInsertNode. - Add amd64 versions of MmGetPageTableForProcess, MmUnmapPageTable, MmGetPageFileMapping svn path=/trunk/; revision=55438
2012-02-05 18:41:37 +00:00
MiRosTakeOverSharedUserPage(IN PEPROCESS Process)
[NTOS]: MiRosTakeOverPebTebRanges now creates a small ~1MB ARM3 memory range on top of the ReactOS per-process VA. This does a couple of things: First of all, it changes the default PEB address to another static address. Still not dynamic like it will be soon, but at least it changes it a bit so we can test if anything breaks due to that. It also likewise changes the addresses of the TEBs (Shifted down by 1MB, basically). Finally, it blocks off that part of address space, which nobody should be using now, to see if anyone does indeed touch it. [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
2010-07-22 03:22:43 +00:00
{
NTSTATUS Status;
PMEMORY_AREA MemoryArea;
PHYSICAL_ADDRESS BoundaryAddressMultiple;
[NTOSKRNL] - Add MM_HAL_VA_END and use it in MiAddHalIoMappings instead of making assumptions about the page tables - Add MM_SHARED_USER_DATA_VA for x86 - use MM_SHARED_USER_DATA_VA in Mm to avoid ambiguities - fix or comment out USER_SHARED_DATA in asm headers, as it should be the kernel mode address here - set Teb->ExceptionList to NULL on amd64, it is used as a link to the Wow64 TEB, if any svn path=/trunk/; revision=55445
2012-02-06 00:24:31 +00:00
PVOID AllocatedBase = (PVOID)MM_SHARED_USER_DATA_VA;
[NTOS]: MiRosTakeOverPebTebRanges now creates a small ~1MB ARM3 memory range on top of the ReactOS per-process VA. This does a couple of things: First of all, it changes the default PEB address to another static address. Still not dynamic like it will be soon, but at least it changes it a bit so we can test if anything breaks due to that. It also likewise changes the addresses of the TEBs (Shifted down by 1MB, basically). Finally, it blocks off that part of address space, which nobody should be using now, to see if anyone does indeed touch it. [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
2010-07-22 03:22:43 +00:00
BoundaryAddressMultiple.QuadPart = 0;
Status = MmCreateMemoryArea(&Process->Vm,
MEMORY_AREA_OWNED_BY_ARM3,
&AllocatedBase,
[NTOSKRNL] - Modify MiRosTakeOverPebTebRanges to only create a memory area for the shared user page and rename it to MiRosTakeOverSharedUserPage. Previously it was creating a memory area for the whole region from USER_SHARED_DATA up to MM_HIGHEST_USER_ADDRESS, which is the majority of the x64 user mode address space and doesn't even contain the PEB/TEB. Those are allocated below the shared user page and get their memory areas created in MiInsertNode. - Add amd64 versions of MmGetPageTableForProcess, MmUnmapPageTable, MmGetPageFileMapping svn path=/trunk/; revision=55438
2012-02-05 18:41:37 +00:00
PAGE_SIZE,
[NTOS]: MiRosTakeOverPebTebRanges now creates a small ~1MB ARM3 memory range on top of the ReactOS per-process VA. This does a couple of things: First of all, it changes the default PEB address to another static address. Still not dynamic like it will be soon, but at least it changes it a bit so we can test if anything breaks due to that. It also likewise changes the addresses of the TEBs (Shifted down by 1MB, basically). Finally, it blocks off that part of address space, which nobody should be using now, to see if anyone does indeed touch it. [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
2010-07-22 03:22:43 +00:00
PAGE_READWRITE,
&MemoryArea,
TRUE,
0,
BoundaryAddressMultiple);
ASSERT(NT_SUCCESS(Status));
}
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
NTSTATUS
NTAPI
MiCreatePebOrTeb(IN PEPROCESS Process,
IN ULONG Size,
OUT PULONG_PTR Base)
{
PETHREAD Thread = PsGetCurrentThread();
PMMVAD_LONG Vad;
NTSTATUS Status;
ULONG RandomCoeff;
ULONG_PTR StartAddress, EndAddress;
LARGE_INTEGER CurrentTime;
[NTOSKRNL] - Rewrite MiFindEmptyAddressRangeDownTree. The old implementation's "most awesome loop" duplicated both the initialization and the interation steps. It was also overcomplicated. The new implementation additionally returns the parent for the following table insertion, so this doesnt need to be done in an extra search. The return value is changed from NTSTATUS to TABLE_SEARCH_RESULT - Modify MiInsertNode to accept a parent and TABLE_SEARCH_RESULT instead of searching for a free location. - Modify MiCreatePebOrTeb to make use of the new features - Handle failed allocation of the PEB/TEB - Fixes a failed assertion that Olaf got - I tested this code quite some time and no problems were found svn path=/trunk/; revision=48606
2010-08-23 03:00:03 +00:00
TABLE_SEARCH_RESULT Result = TableFoundNode;
PMMADDRESS_NODE Parent;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Allocate a VAD */
Vad = ExAllocatePoolWithTag(NonPagedPool, sizeof(MMVAD_LONG), 'ldaV');
if (!Vad) return STATUS_NO_MEMORY;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Setup the primary flags with the size, and make it commited, private, RW */
Vad->u.LongFlags = 0;
Vad->u.VadFlags.CommitCharge = BYTES_TO_PAGES(Size);
Vad->u.VadFlags.MemCommit = TRUE;
Vad->u.VadFlags.PrivateMemory = TRUE;
Vad->u.VadFlags.Protection = MM_READWRITE;
Vad->u.VadFlags.NoChange = TRUE;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Setup the secondary flags to make it a secured, writable, long VAD */
Vad->u2.LongFlags2 = 0;
Vad->u2.VadFlags2.OneSecured = TRUE;
Vad->u2.VadFlags2.LongVad = TRUE;
Vad->u2.VadFlags2.ReadOnly = FALSE;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Lock the process address space */
KeAcquireGuardedMutex(&Process->AddressCreationLock);
/* Check if this is a PEB creation */
if (Size == sizeof(PEB))
{
/* Start at the highest valid address */
StartAddress = (ULONG_PTR)MM_HIGHEST_VAD_ADDRESS + 1;
/* Select the random coefficient */
KeQueryTickCount(&CurrentTime);
CurrentTime.LowPart &= ((64 * _1KB) >> PAGE_SHIFT) - 1;
if (CurrentTime.LowPart <= 1) CurrentTime.LowPart = 2;
RandomCoeff = CurrentTime.LowPart << PAGE_SHIFT;
/* Select the highest valid address minus the random coefficient */
StartAddress -= RandomCoeff;
EndAddress = StartAddress + ROUND_TO_PAGES(Size) - 1;
[NTOSKRNL] - Rewrite MiFindEmptyAddressRangeDownTree. The old implementation's "most awesome loop" duplicated both the initialization and the interation steps. It was also overcomplicated. The new implementation additionally returns the parent for the following table insertion, so this doesnt need to be done in an extra search. The return value is changed from NTSTATUS to TABLE_SEARCH_RESULT - Modify MiInsertNode to accept a parent and TABLE_SEARCH_RESULT instead of searching for a free location. - Modify MiCreatePebOrTeb to make use of the new features - Handle failed allocation of the PEB/TEB - Fixes a failed assertion that Olaf got - I tested this code quite some time and no problems were found svn path=/trunk/; revision=48606
2010-08-23 03:00:03 +00:00
/* Try to find something below the random upper margin */
Result = MiFindEmptyAddressRangeDownTree(ROUND_TO_PAGES(Size),
EndAddress,
PAGE_SIZE,
&Process->VadRoot,
Base,
&Parent);
}
/* Check for success. TableFoundNode means nothing free. */
if (Result == TableFoundNode)
{
/* For TEBs, or if a PEB location couldn't be found, scan the VAD root */
Result = MiFindEmptyAddressRangeDownTree(ROUND_TO_PAGES(Size),
(ULONG_PTR)MM_HIGHEST_VAD_ADDRESS + 1,
PAGE_SIZE,
&Process->VadRoot,
Base,
&Parent);
/* Bail out, if still nothing free was found */
[NTOSKRNL] - ExFreePool --> ExFreePoolWithTag in some places. - Remove dead code (section.c) - Free resource (sysldr.c) svn path=/trunk/; revision=57412
2012-09-28 12:17:23 +00:00
if (Result == TableFoundNode)
{
ExFreePoolWithTag(Vad, 'ldaV');
return STATUS_NO_MEMORY;
}
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Validate that it came from the VAD ranges */
ASSERT(*Base >= (ULONG_PTR)MI_LOWEST_VAD_ADDRESS);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Build the rest of the VAD now */
Vad->StartingVpn = (*Base) >> PAGE_SHIFT;
[NTOS]: Use the Spare flag in the VAD as a ReactOS/MemoryArea specific flag to signify that this VAD is associated with a MEMORY_AREA and should be unlinked at process exit, but not freed. This is because MemoryAreas themselves are cleaned up later, and in the future their associated VADs (not yet in Trunk) will also be parsed. In the process death scenario, those VADs will be freed, but not unlinked (since it would already have been unlinked). svn path=/trunk/; revision=48990
2010-10-05 05:00:19 +00:00
Vad->EndingVpn = ((*Base) + Size - 1) >> PAGE_SHIFT;
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
Vad->u3.Secured.StartVpn = *Base;
Vad->u3.Secured.EndVpn = (Vad->EndingVpn << PAGE_SHIFT) | (PAGE_SIZE - 1);
[NTOSKRNL] - Initialize the Parent member of the new Vad to NULL. This also initializes the Balance to 0 aka RtlBalancedAvlTree Should fix the failed assertion that randomly occurs. svn path=/trunk/; revision=48562
2010-08-19 05:10:16 +00:00
Vad->u1.Parent = NULL;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* FIXME: Should setup VAD bitmap */
Status = STATUS_SUCCESS;
/* Pretend as if we own the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiLockProcessWorkingSetUnsafe(Process, Thread);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Insert the VAD */
ASSERT(Vad->EndingVpn >= Vad->StartingVpn);
Process->VadRoot.NodeHint = Vad;
[NTOS]: Use MI_SET_PFN_DELETED where we missed it. [NTOS]: Implement support for deleting user-mode pageable VM addresses. Now when cleaning up the process address space, MiDeleteVirtualAddresses is called for the VADs, so this will now actually free the PEB/TEB pages that were previously getting leaked for each thread/process (a known regression I introduced when moving to VADs for PEB/TEB). svn path=/trunk/; revision=49187
2010-10-17 20:02:17 +00:00
Vad->ControlArea = NULL; // For Memory-Area hack
Vad->FirstPrototypePte = NULL;
[NTOS]: Complete the VAD-to-MAREA Synchronization hack by removing fake MAREAs that were added when inserting real VADs. To do this, we have to track the fake MAREA associated with a VAD, so we overload the FirstProtoTypePte field in the VAD, if this is NOT a section VAD (which we don't use yet). We'll figure something out for section VADs later. [NTOS]: Now that VAD and MAREA views are synchronized, remove the VAD limit and let VADs be created at any address. Also do not create an arbitrary 16MB VAD memory area anymore. This basically now allows for as many PEB/TEBs as can fit in the address space, fixing the recent known regression that limited the number of threads a process could have. svn path=/trunk/; revision=49193
2010-10-18 13:10:54 +00:00
DPRINT("VAD: %p\n", Vad);
DPRINT("Allocated PEB/TEB at: 0x%p for %16s\n", *Base, Process->ImageFileName);
[NTOSKRNL] - Rewrite MiFindEmptyAddressRangeDownTree. The old implementation's "most awesome loop" duplicated both the initialization and the interation steps. It was also overcomplicated. The new implementation additionally returns the parent for the following table insertion, so this doesnt need to be done in an extra search. The return value is changed from NTSTATUS to TABLE_SEARCH_RESULT - Modify MiInsertNode to accept a parent and TABLE_SEARCH_RESULT instead of searching for a free location. - Modify MiCreatePebOrTeb to make use of the new features - Handle failed allocation of the PEB/TEB - Fixes a failed assertion that Olaf got - I tested this code quite some time and no problems were found svn path=/trunk/; revision=48606
2010-08-23 03:00:03 +00:00
MiInsertNode(&Process->VadRoot, (PVOID)Vad, Parent, Result);
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Release the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiUnlockProcessWorkingSetUnsafe(Process, Thread);
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
/* Release the address space lock */
KeReleaseGuardedMutex(&Process->AddressCreationLock);
/* Return the status */
return Status;
}
VOID
NTAPI
MmDeleteTeb(IN PEPROCESS Process,
IN PTEB Teb)
{
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
ULONG_PTR TebEnd;
PETHREAD Thread = PsGetCurrentThread();
PMMVAD Vad;
PMM_AVL_TABLE VadTree = &Process->VadRoot;
DPRINT("Deleting TEB: %p in %16s\n", Teb, Process->ImageFileName);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* TEB is one page */
TebEnd = (ULONG_PTR)Teb + ROUND_TO_PAGES(sizeof(TEB)) - 1;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Attach to the process */
KeAttachProcess(&Process->Pcb);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Lock the process address space */
KeAcquireGuardedMutex(&Process->AddressCreationLock);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Find the VAD, make sure it's a TEB VAD */
Vad = MiLocateAddress(Teb);
DPRINT("Removing node for VAD: %lx %lx\n", Vad->StartingVpn, Vad->EndingVpn);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
ASSERT(Vad != NULL);
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
if (Vad->StartingVpn != ((ULONG_PTR)Teb >> PAGE_SHIFT))
{
/* Bug in the AVL code? */
DPRINT1("Corrupted VAD!\n");
}
else
{
/* Sanity checks for a valid TEB VAD */
ASSERT((Vad->StartingVpn == ((ULONG_PTR)Teb >> PAGE_SHIFT) &&
(Vad->EndingVpn == (TebEnd >> PAGE_SHIFT))));
ASSERT(Vad->u.VadFlags.NoChange == TRUE);
[NTOS]: Sanity checks when dealing with creating/cleaning process address space. [NTOS]: Set the VmDeleted flag when the process address space is being nuked. svn path=/trunk/; revision=49434
2010-11-02 14:41:03 +00:00
ASSERT(Vad->u2.VadFlags2.OneSecured == TRUE);
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
ASSERT(Vad->u2.VadFlags2.MultipleSecured == FALSE);
/* Lock the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiLockProcessWorkingSetUnsafe(Process, Thread);
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Remove this VAD from the tree */
ASSERT(VadTree->NumberGenericTableElements >= 1);
MiRemoveNode((PMMADDRESS_NODE)Vad, VadTree);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
/* Delete the pages */
MiDeleteVirtualAddresses((ULONG_PTR)Teb, TebEnd, NULL);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Release the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiUnlockProcessWorkingSetUnsafe(Process, Thread);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Remove the VAD */
ExFreePool(Vad);
}
/* Release the address space lock */
KeReleaseGuardedMutex(&Process->AddressCreationLock);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmDeleteTeb, VADs are now deleted/freed on thread exit as well (but the underlying page is still leaked). Should fix the advapi32 security crash. [NTOS]: Sometimes it seems we hit some bad VADs due to bugs? in the AVL tree implementation. I'm going on vacation for a month and can't look at this, so I've hacked the code to ignore such VADs for now, in the interest of fixing the winetest regression. svn path=/trunk/; revision=48235
2010-07-24 16:28:51 +00:00
/* Detach */
KeDetachProcess();
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
}
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
VOID
NTAPI
MmDeleteKernelStack(IN PVOID StackBase,
IN BOOLEAN GuiStack)
{
PMMPTE PointerPte;
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
PFN_NUMBER PageFrameNumber, PageTableFrameNumber;
[HAL/NDK] - Make Vector parameter in HalEnableSystemInterrupt, HalDisableSystemInterrupt and HalBeginSystemInterrupt an ULONG, not an UCHAR [NDK] - 64bit fixes for HANDLE_TABLE, KPROCESS, SECTION_IMAGE_INFORMATION, MMADDRESS_LIST, MMVAD_FLAGS, MMVAD, MMVAD_LONG, MMVAD_SHORT, MEMORY_DESCRIPTOR, MEMORY_ALLOCATION_DESCRIPTOR, LdrVerifyMappedImageMatchesChecksum - KDPC_DATA::DpcQueueDepth is signed on amd64, unsigned on x86 [NTOSKRNL] - Fix hundreds of MSVC and amd64 warnings - add a pragma message to FstubFixupEfiPartition, since it looks broken - Move portable Ke constants from <arch>/cpu.c to krnlinit.c - Fixed a bug in amd64 KiGeneralProtectionFaultHandler svn path=/trunk/; revision=53734
2011-09-18 13:11:45 +00:00
PFN_COUNT StackPages;
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
PMMPFN Pfn1, Pfn2;
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
ULONG i;
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
KIRQL OldIrql;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// This should be the guard page, so decrement by one
//
PointerPte = MiAddressToPte(StackBase);
PointerPte--;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[MEGAPERF]: This build introduces the following performance boosts: - Using a dead stack lookaside list for up to 5 dead kernel thread stacks. 1500% improvement when rapidly destroying/creating threads, such as during second stage setup and many winetests. - Using a free pool page lookaside list for up to 12 free non-paged or paged pool pages. 800% improvement when allocating big pages from the pool, as well as during pool expansion. - Using a bucketized per-processor and local list (in the KPRCB) for block sizes between 1 and 32 bytes. 1000% improvement when rapidly allocating/freeing small pool allocations, and 8x reduction in pool fragmentation. svn path=/trunk/; revision=56000
2012-03-04 17:56:00 +00:00
//
// If this is a small stack, just push the stack onto the dead stack S-LIST
//
if (!GuiStack)
{
if (ExQueryDepthSList(&MmDeadStackSListHead) < MmMaximumDeadKernelStacks)
{
Pfn1 = MiGetPfnEntry(PointerPte->u.Hard.PageFrameNumber);
[NTOSKRNL] - Make file offsets 64 bit - Fix lots of MSVC/64 bit warnings svn path=/trunk/; revision=56264
2012-03-28 19:41:40 +00:00
InterlockedPushEntrySList(&MmDeadStackSListHead,
(PSLIST_ENTRY)&Pfn1->u1.NextStackPfn);
[MEGAPERF]: This build introduces the following performance boosts: - Using a dead stack lookaside list for up to 5 dead kernel thread stacks. 1500% improvement when rapidly destroying/creating threads, such as during second stage setup and many winetests. - Using a free pool page lookaside list for up to 12 free non-paged or paged pool pages. 800% improvement when allocating big pages from the pool, as well as during pool expansion. - Using a bucketized per-processor and local list (in the KPRCB) for block sizes between 1 and 32 bytes. 1000% improvement when rapidly allocating/freeing small pool allocations, and 8x reduction in pool fragmentation. svn path=/trunk/; revision=56000
2012-03-04 17:56:00 +00:00
return;
}
}
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Calculate pages used
//
StackPages = BYTES_TO_PAGES(GuiStack ?
KERNEL_LARGE_STACK_SIZE : KERNEL_STACK_SIZE);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Acquire the PFN lock */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Loop them
//
for (i = 0; i < StackPages; i++)
{
//
// Check if this is a valid PTE
//
if (PointerPte->u.Hard.Valid == 1)
{
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Get the PTE's page */
PageFrameNumber = PFN_FROM_PTE(PointerPte);
Pfn1 = MiGetPfnEntry(PageFrameNumber);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Now get the page of the page table mapping it */
PageTableFrameNumber = Pfn1->u4.PteFrame;
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
Pfn2 = MiGetPfnEntry(PageTableFrameNumber);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Remove a shared reference, since the page is going away */
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
MiDecrementShareCount(Pfn2, PageTableFrameNumber);
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Set the special pending delete marker */
[NTOS]: Use MI_SET_PFN_DELETED where we missed it. [NTOS]: Implement support for deleting user-mode pageable VM addresses. Now when cleaning up the process address space, MiDeleteVirtualAddresses is called for the VADs, so this will now actually free the PEB/TEB pages that were previously getting leaked for each thread/process (a known regression I introduced when moving to VADs for PEB/TEB). svn path=/trunk/; revision=49187
2010-10-17 20:02:17 +00:00
MI_SET_PFN_DELETED(Pfn1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* And now delete the actual stack page */
MiDecrementShareCount(Pfn1, PageFrameNumber);
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Next one
//
PointerPte--;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// We should be at the guard page now
//
ASSERT(PointerPte->u.Hard.Valid == 0);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Release the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Release the PTEs
//
MiReleaseSystemPtes(PointerPte, StackPages + 1, SystemPteSpace);
}
PVOID
NTAPI
MmCreateKernelStack(IN BOOLEAN GuiStack,
IN UCHAR Node)
{
[HAL/NDK] - Make Vector parameter in HalEnableSystemInterrupt, HalDisableSystemInterrupt and HalBeginSystemInterrupt an ULONG, not an UCHAR [NDK] - 64bit fixes for HANDLE_TABLE, KPROCESS, SECTION_IMAGE_INFORMATION, MMADDRESS_LIST, MMVAD_FLAGS, MMVAD, MMVAD_LONG, MMVAD_SHORT, MEMORY_DESCRIPTOR, MEMORY_ALLOCATION_DESCRIPTOR, LdrVerifyMappedImageMatchesChecksum - KDPC_DATA::DpcQueueDepth is signed on amd64, unsigned on x86 [NTOSKRNL] - Fix hundreds of MSVC and amd64 warnings - add a pragma message to FstubFixupEfiPartition, since it looks broken - Move portable Ke constants from <arch>/cpu.c to krnlinit.c - Fixed a bug in amd64 KiGeneralProtectionFaultHandler svn path=/trunk/; revision=53734
2011-09-18 13:11:45 +00:00
PFN_COUNT StackPtes, StackPages;
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
PMMPTE PointerPte, StackPte;
PVOID BaseAddress;
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
MMPTE TempPte, InvalidPte;
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
KIRQL OldIrql;
PFN_NUMBER PageFrameIndex;
ULONG i;
[MEGAPERF]: This build introduces the following performance boosts: - Using a dead stack lookaside list for up to 5 dead kernel thread stacks. 1500% improvement when rapidly destroying/creating threads, such as during second stage setup and many winetests. - Using a free pool page lookaside list for up to 12 free non-paged or paged pool pages. 800% improvement when allocating big pages from the pool, as well as during pool expansion. - Using a bucketized per-processor and local list (in the KPRCB) for block sizes between 1 and 32 bytes. 1000% improvement when rapidly allocating/freeing small pool allocations, and 8x reduction in pool fragmentation. svn path=/trunk/; revision=56000
2012-03-04 17:56:00 +00:00
PMMPFN Pfn1;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Calculate pages needed
//
if (GuiStack)
{
//
// We'll allocate 64KB stack, but only commit 12K
//
StackPtes = BYTES_TO_PAGES(KERNEL_LARGE_STACK_SIZE);
StackPages = BYTES_TO_PAGES(KERNEL_LARGE_STACK_COMMIT);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
}
else
{
[MEGAPERF]: This build introduces the following performance boosts: - Using a dead stack lookaside list for up to 5 dead kernel thread stacks. 1500% improvement when rapidly destroying/creating threads, such as during second stage setup and many winetests. - Using a free pool page lookaside list for up to 12 free non-paged or paged pool pages. 800% improvement when allocating big pages from the pool, as well as during pool expansion. - Using a bucketized per-processor and local list (in the KPRCB) for block sizes between 1 and 32 bytes. 1000% improvement when rapidly allocating/freeing small pool allocations, and 8x reduction in pool fragmentation. svn path=/trunk/; revision=56000
2012-03-04 17:56:00 +00:00
//
// If the dead stack S-LIST has a stack on it, use it instead of allocating
// new system PTEs for this stack
//
if (ExQueryDepthSList(&MmDeadStackSListHead))
{
Pfn1 = (PMMPFN)InterlockedPopEntrySList(&MmDeadStackSListHead);
if (Pfn1)
{
PointerPte = Pfn1->PteAddress;
BaseAddress = MiPteToAddress(++PointerPte);
return BaseAddress;
}
}
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// We'll allocate 12K and that's it
//
StackPtes = BYTES_TO_PAGES(KERNEL_STACK_SIZE);
StackPages = StackPtes;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Reserve stack pages, plus a guard page
//
StackPte = MiReserveSystemPtes(StackPtes + 1, SystemPteSpace);
if (!StackPte) return NULL;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Get the stack address
//
BaseAddress = MiPteToAddress(StackPte + StackPtes + 1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Select the right PTE address where we actually start committing pages
//
PointerPte = StackPte;
if (GuiStack) PointerPte += BYTES_TO_PAGES(KERNEL_LARGE_STACK_SIZE -
KERNEL_LARGE_STACK_COMMIT);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
/* Setup the temporary invalid PTE */
MI_MAKE_SOFTWARE_PTE(&InvalidPte, MM_NOACCESS);
/* Setup the template stack PTE */
[NTOS]: Learn to build User PTEs as well, with MI_MAKE_HARDWARE_PTE_USER. [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
2010-07-22 02:20:27 +00:00
MI_MAKE_HARDWARE_PTE_KERNEL(&TempPte, PointerPte + 1, MM_READWRITE, 0);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Acquire the PFN DB lock
//
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Loop each stack page
//
for (i = 0; i < StackPages; i++)
{
//
// Next PTE
//
PointerPte++;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
/* Get a page and write the current invalid PTE */
[NTOS]: Add page-tracking macro calls wherever needed. MI_TRACE_PFNS is still off, though, so this code won't be called. svn path=/trunk/; revision=49444
2010-11-02 15:16:22 +00:00
MI_SET_USAGE(MI_USAGE_KERNEL_STACK);
MI_SET_PROCESS2(PsGetCurrentProcess()->ImageFileName);
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
PageFrameIndex = MiRemoveAnyPage(MI_GET_NEXT_COLOR());
[NTOS] Inline writing to PTEs through MI_WRITE_VALID/INVALID_PTE. svn path=/trunk/; revision=47633
2010-06-06 18:45:46 +00:00
MI_WRITE_INVALID_PTE(PointerPte, InvalidPte);
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Initialize the PFN entry for this page */
MiInitializePfn(PageFrameIndex, PointerPte, 1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Write the valid PTE */
TempPte.u.Hard.PageFrameNumber = PageFrameIndex;
[NTOS] Inline writing to PTEs through MI_WRITE_VALID/INVALID_PTE. svn path=/trunk/; revision=47633
2010-06-06 18:45:46 +00:00
MI_WRITE_VALID_PTE(PointerPte, TempPte);
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
}
Compensate for continuous memory usage decrease (bug #4835). svn path=/trunk/; revision=44899
2010-01-02 20:41:04 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Release the PFN lock
//
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Return the stack address
//
return BaseAddress;
}
NTSTATUS
NTAPI
MmGrowKernelStackEx(IN PVOID StackPointer,
IN ULONG GrowSize)
{
PKTHREAD Thread = KeGetCurrentThread();
PMMPTE LimitPte, NewLimitPte, LastPte;
KIRQL OldIrql;
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
MMPTE TempPte, InvalidPte;
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
PFN_NUMBER PageFrameIndex;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Make sure the stack did not overflow
//
ASSERT(((ULONG_PTR)Thread->StackBase - (ULONG_PTR)Thread->StackLimit) <=
(KERNEL_LARGE_STACK_SIZE + PAGE_SIZE));
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Get the current stack limit
//
LimitPte = MiAddressToPte(Thread->StackLimit);
ASSERT(LimitPte->u.Hard.Valid == 1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Get the new one and make sure this isn't a retarded request
//
NewLimitPte = MiAddressToPte((PVOID)((ULONG_PTR)StackPointer - GrowSize));
if (NewLimitPte == LimitPte) return STATUS_SUCCESS;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Now make sure you're not going past the reserved space
//
LastPte = MiAddressToPte((PVOID)((ULONG_PTR)Thread->StackBase -
KERNEL_LARGE_STACK_SIZE));
if (NewLimitPte < LastPte)
{
//
// Sorry!
//
DPRINT1("Thread wants too much stack\n");
return STATUS_STACK_OVERFLOW;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Calculate the number of new pages
//
LimitPte--;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
/* Setup the temporary invalid PTE */
MI_MAKE_SOFTWARE_PTE(&InvalidPte, MM_NOACCESS);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Acquire the PFN DB lock
//
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
[NTOS]: Fix definition of unused MI_MAKE_SOFTWARE_PTE macro. [NTOS]: Correctly setup the PFN entries for freshly allocated paged pool pages. Fixes a problem where the page could've still had stale/garbage data. [NTOS]: Add some extra assertions in the code to catch memory corruption and detect invalid logic. [NTOS]: Fix some typos in the code (comments/whitespace). [NTOS]: Make the dreaded page fault message that breaks paged pool on some systems more verbose for future debugging. svn path=/trunk/; revision=47189
2010-05-12 22:47:46 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Loop each stack page
//
while (LimitPte >= NewLimitPte)
{
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
/* Get a page and write the current invalid PTE */
[NTOS]: Add page-tracking macro calls wherever needed. MI_TRACE_PFNS is still off, though, so this code won't be called. svn path=/trunk/; revision=49444
2010-11-02 15:16:22 +00:00
MI_SET_USAGE(MI_USAGE_KERNEL_STACK_EXPANSION);
MI_SET_PROCESS2(PsGetCurrentProcess()->ImageFileName);
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
PageFrameIndex = MiRemoveAnyPage(MI_GET_NEXT_COLOR());
[NTOS] Inline writing to PTEs through MI_WRITE_VALID/INVALID_PTE. svn path=/trunk/; revision=47633
2010-06-06 18:45:46 +00:00
MI_WRITE_INVALID_PTE(LimitPte, InvalidPte);
Testers: Please pay attention to this build and test it fully: [NTOS]: Implement MiDecrementShareCount (to start replacing MmReleasePageMemoryConsumer calls for pages that were grabbed through ARM3, not Mm). [NTOS]: Implement MiInitializePfn (to initialize pages grabbed through ARM3/MiRemoveAnyPage instead of Mm/MmAllocPage). [NTOS]: For stack pages, use new ARM3 PFN alloc/free routines, as a first test/beginning of the new ARM3 ABI. [NTOS]: Implement and start using the Pending-Deletion PFN flag. [NTOS]: As a result, for stack pages, the Transition page state will now be seen, and the new routine for re-inserting pages into the free list will now be used. Tracking of page table references is also done now for these pages (but we don't free the PT since this doesn't seem safe yet). svn path=/trunk/; revision=47424
2010-05-29 18:33:50 +00:00
/* Initialize the PFN entry for this page */
MiInitializePfn(PageFrameIndex, LimitPte, 1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
Testers: Please test this build. [NTOS]: Implement a MI_MAKE_HARDWARE_PTE macro for the generation of valid kernel PTEs instead of always taking the ValidKernelPte and changing its flags. This macro will take into account the protection mask (up until now ignored) and use the array previously implemented to determine the correct hardware PTE settings. Assertions are also added to validate correct usage of the macro, and later revisions will fill out NT-specific fields to help deal with transition PTEs, page faults, etc. [NTOS]: Make the stack code the first user of this macro, for the stack PTEs. Good testing base as we create kernel stacks very often. [NTOS]: The NT MM ABI specifies that in between the allocation of a new PTE and its initialization as a valid PFN, the PTE entry should be an invalid PTE, and should only be marked valid after the PFN has been initialized. For stack PTEs, do this -- first allocating the page, making it invalid, then initializing the PFN, and then writing the valid page. svn path=/trunk/; revision=47571
2010-06-04 17:49:36 +00:00
/* Setup the template stack PTE */
[NTOS]: Learn to build User PTEs as well, with MI_MAKE_HARDWARE_PTE_USER. [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
2010-07-22 02:20:27 +00:00
MI_MAKE_HARDWARE_PTE_KERNEL(&TempPte, LimitPte, MM_READWRITE, PageFrameIndex);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Fix definition of unused MI_MAKE_SOFTWARE_PTE macro. [NTOS]: Correctly setup the PFN entries for freshly allocated paged pool pages. Fixes a problem where the page could've still had stale/garbage data. [NTOS]: Add some extra assertions in the code to catch memory corruption and detect invalid logic. [NTOS]: Fix some typos in the code (comments/whitespace). [NTOS]: Make the dreaded page fault message that breaks paged pool on some systems more verbose for future debugging. svn path=/trunk/; revision=47189
2010-05-12 22:47:46 +00:00
/* Write the valid PTE */
[NTOS] Inline writing to PTEs through MI_WRITE_VALID/INVALID_PTE. svn path=/trunk/; revision=47633
2010-06-06 18:45:46 +00:00
MI_WRITE_VALID_PTE(LimitPte--, TempPte);
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Release the PFN lock
//
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
//
// Set the new limit
//
Thread->StackLimit = (ULONG_PTR)MiPteToAddress(NewLimitPte);
return STATUS_SUCCESS;
}
NTSTATUS
NTAPI
MmGrowKernelStack(IN PVOID StackPointer)
{
//
// Call the extended version
//
return MmGrowKernelStackEx(StackPointer, KERNEL_LARGE_STACK_COMMIT);
}
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
NTSTATUS
NTAPI
MmSetMemoryPriorityProcess(IN PEPROCESS Process,
IN UCHAR MemoryPriority)
{
UCHAR OldPriority;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Check if we have less then 16MB of Physical Memory
//
if ((MmSystemSize == MmSmallSystem) &&
Get rid of MmStats -- most of the fields weren't used anymore and we have duplicates for most of the ones that are still in use; NrTotalPages -> MmNumberOfPhysicalPages NrFreePages -> MmAvailablePages (new) Get rid of NrSystemPages. Its value was only respected in MmInitializePageList at boot, even though it got updated later. Use a local variable in MmInitializePageList instead. Fix SystemBasicInformation to use the correct variables for physical page information. Also, don't set ResidentSystemCodePage in the SystemPerformanceInformation query to some random incorrect Mm value. We don't depend this value anywhere in ReactOS currently, so just set it to 0. svn path=/trunk/; revision=43550
2009-10-18 13:55:44 +00:00
(MmNumberOfPhysicalPages < ((15 * 1024 * 1024) / PAGE_SIZE)))
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
{
//
// Always use background priority
//
MemoryPriority = MEMORY_PRIORITY_BACKGROUND;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Save the old priority and update it
//
OldPriority = (UCHAR)Process->Vm.Flags.MemoryPriority;
Process->Vm.Flags.MemoryPriority = MemoryPriority;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Return the old priority
//
return OldPriority;
}
LCID
NTAPI
MmGetSessionLocaleId(VOID)
{
PEPROCESS Process;
PAGED_CODE();
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Get the current process
//
Process = PsGetCurrentProcess();
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Check if it's the Session Leader
//
if (Process->Vm.Flags.SessionLeader)
{
//
// Make sure it has a valid Session
//
if (Process->Session)
{
//
// Get the Locale ID
//
return ((PMM_SESSION_SPACE)Process->Session)->LocaleId;
}
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Not a session leader, return the default
//
return PsDefaultThreadLocaleId;
}
NTSTATUS
NTAPI
MmCreatePeb(IN PEPROCESS Process,
IN PINITIAL_PEB InitialPeb,
OUT PPEB *BasePeb)
{
PPEB Peb = NULL;
LARGE_INTEGER SectionOffset;
SIZE_T ViewSize = 0;
PVOID TableBase = NULL;
PIMAGE_NT_HEADERS NtHeaders;
PIMAGE_LOAD_CONFIG_DIRECTORY ImageConfigData;
NTSTATUS Status;
USHORT Characteristics;
KAFFINITY ProcessAffinityMask = 0;
SectionOffset.QuadPart = (ULONGLONG)0;
*BasePeb = NULL;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Attach to Process
//
KeAttachProcess(&Process->Pcb);
[HAL/NDK] - Make Vector parameter in HalEnableSystemInterrupt, HalDisableSystemInterrupt and HalBeginSystemInterrupt an ULONG, not an UCHAR [NDK] - 64bit fixes for HANDLE_TABLE, KPROCESS, SECTION_IMAGE_INFORMATION, MMADDRESS_LIST, MMVAD_FLAGS, MMVAD, MMVAD_LONG, MMVAD_SHORT, MEMORY_DESCRIPTOR, MEMORY_ALLOCATION_DESCRIPTOR, LdrVerifyMappedImageMatchesChecksum - KDPC_DATA::DpcQueueDepth is signed on amd64, unsigned on x86 [NTOSKRNL] - Fix hundreds of MSVC and amd64 warnings - add a pragma message to FstubFixupEfiPartition, since it looks broken - Move portable Ke constants from <arch>/cpu.c to krnlinit.c - Fixed a bug in amd64 KiGeneralProtectionFaultHandler svn path=/trunk/; revision=53734
2011-09-18 13:11:45 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Map NLS Tables
//
[NTOS]: Go back to using RosMm for the NLS section since supposedly this causes issues on some systems. This is the only actual change in r206/208/209, so if Caemyr's data is correct, this fill fix it. svn path=/trunk/; revision=49211
2010-10-19 20:23:37 +00:00
Status = MmMapViewOfSection(ExpNlsSectionPointer,
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
(PEPROCESS)Process,
&TableBase,
0,
0,
&SectionOffset,
&ViewSize,
ViewShare,
MEM_TOP_DOWN,
PAGE_READONLY);
- Downgrade some DPRINT1s to DPRINTs. svn path=/trunk/; revision=53635
2011-09-08 08:28:41 +00:00
DPRINT("NLS Tables at: %p\n", TableBase);
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
if (!NT_SUCCESS(Status))
{
/* Cleanup and exit */
KeDetachProcess();
return Status;
}
//
// Allocate the PEB
//
Status = MiCreatePebOrTeb(Process, sizeof(PEB), (PULONG_PTR)&Peb);
- Downgrade some DPRINT1s to DPRINTs. svn path=/trunk/; revision=53635
2011-09-08 08:28:41 +00:00
DPRINT("PEB at: %p\n", Peb);
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
if (!NT_SUCCESS(Status))
{
/* Cleanup and exit */
KeDetachProcess();
return Status;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Use SEH in case we can't load the PEB
//
_SEH2_TRY
{
//
// Initialize the PEB
//
RtlZeroMemory(Peb, sizeof(PEB));
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Set up data
//
Peb->ImageBaseAddress = Process->SectionBaseAddress;
Peb->InheritedAddressSpace = InitialPeb->InheritedAddressSpace;
Peb->Mutant = InitialPeb->Mutant;
Peb->ImageUsesLargePages = InitialPeb->ImageUsesLargePages;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// NLS
//
Peb->AnsiCodePageData = (PCHAR)TableBase + ExpAnsiCodePageDataOffset;
Peb->OemCodePageData = (PCHAR)TableBase + ExpOemCodePageDataOffset;
Peb->UnicodeCaseTableData = (PCHAR)TableBase + ExpUnicodeCaseTableDataOffset;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Default Version Data (could get changed below)
//
Peb->OSMajorVersion = NtMajorVersion;
Peb->OSMinorVersion = NtMinorVersion;
Peb->OSBuildNumber = (USHORT)(NtBuildNumber & 0x3FFF);
[KERNEL32/NTDLL] - Move the ReactOS-specific workaround described in http://jira.reactos.org/browse/CORE-6611 and http://jira.reactos.org/browse/CORE-4620 from kernel32 to ntdll for using it by RtlGetVersion, in order to : * still having it available in user-mode only, * and having VerifyVersionInfo (and its Rtl counter-part) working properly. [RTL/NTOSKRNL] - Code formatting. - Use a defined-constant instead of a hard-coded value. svn path=/trunk/; revision=57662
2012-11-01 12:52:34 +00:00
Peb->OSPlatformId = VER_PLATFORM_WIN32_NT;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
Peb->OSCSDVersion = (USHORT)CmNtCSDVersion;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Heap and Debug Data
//
Peb->NumberOfProcessors = KeNumberProcessors;
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
Peb->BeingDebugged = (BOOLEAN)(Process->DebugPort != NULL);
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
Peb->NtGlobalFlag = NtGlobalFlag;
/*Peb->HeapSegmentReserve = MmHeapSegmentReserve;
Peb->HeapSegmentCommit = MmHeapSegmentCommit;
Peb->HeapDeCommitTotalFreeThreshold = MmHeapDeCommitTotalFreeThreshold;
Peb->HeapDeCommitFreeBlockThreshold = MmHeapDeCommitFreeBlockThreshold;
Peb->CriticalSectionTimeout = MmCriticalSectionTimeout;
Peb->MinimumStackCommit = MmMinimumStackCommitInBytes;
*/
Peb->MaximumNumberOfHeaps = (PAGE_SIZE - sizeof(PEB)) / sizeof(PVOID);
Peb->ProcessHeaps = (PVOID*)(Peb + 1);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Session ID
//
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
MmGetSessionId(Process);
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
}
_SEH2_EXCEPT(EXCEPTION_EXECUTE_HANDLER)
{
//
// Fail
//
KeDetachProcess();
_SEH2_YIELD(return _SEH2_GetExceptionCode());
}
_SEH2_END;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Use SEH in case we can't load the image
//
_SEH2_TRY
{
//
// Get NT Headers
//
NtHeaders = RtlImageNtHeader(Peb->ImageBaseAddress);
Characteristics = NtHeaders->FileHeader.Characteristics;
}
_SEH2_EXCEPT(EXCEPTION_EXECUTE_HANDLER)
{
//
// Fail
//
KeDetachProcess();
_SEH2_YIELD(return STATUS_INVALID_IMAGE_PROTECT);
}
_SEH2_END;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Parse the headers
//
if (NtHeaders)
{
//
// Use SEH in case we can't load the headers
//
_SEH2_TRY
{
//
// Get the Image Config Data too
//
ImageConfigData = RtlImageDirectoryEntryToData(Peb->ImageBaseAddress,
TRUE,
IMAGE_DIRECTORY_ENTRY_LOAD_CONFIG,
Merge 36852, 37322, 37333, 37334, 43428, 43451, 44259, 46404 from amd64 branch. - Multiple ULONG/ULONG_PTR/SIZE_T issues - Use SLIST_ENTRY instead of SINGLE_LIST_ENTRY with InterlockedPushEntrySList - Use %p to print pointer sized values svn path=/trunk/; revision=48073
2010-07-16 00:34:26 +00:00
(PULONG)&ViewSize);
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
if (ImageConfigData)
{
//
// Probe it
//
ProbeForRead(ImageConfigData,
sizeof(IMAGE_LOAD_CONFIG_DIRECTORY),
sizeof(ULONG));
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Write subsystem data
//
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Peb->ImageSubsystem = NtHeaders->OptionalHeader.Subsystem;
Peb->ImageSubsystemMajorVersion = NtHeaders->OptionalHeader.MajorSubsystemVersion;
Peb->ImageSubsystemMinorVersion = NtHeaders->OptionalHeader.MinorSubsystemVersion;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Check for version data
//
if (NtHeaders->OptionalHeader.Win32VersionValue)
{
//
// Extract values and write them
//
Peb->OSMajorVersion = NtHeaders->OptionalHeader.Win32VersionValue & 0xFF;
Peb->OSMinorVersion = (NtHeaders->OptionalHeader.Win32VersionValue >> 8) & 0xFF;
Peb->OSBuildNumber = (NtHeaders->OptionalHeader.Win32VersionValue >> 16) & 0x3FFF;
Peb->OSPlatformId = (NtHeaders->OptionalHeader.Win32VersionValue >> 30) ^ 2;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
/* Process CSD version override */
if ((ImageConfigData) && (ImageConfigData->CSDVersion))
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
{
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
/* Take the value from the image configuration directory */
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
Peb->OSCSDVersion = ImageConfigData->CSDVersion;
}
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
/* Process optional process affinity mask override */
if ((ImageConfigData) && (ImageConfigData->ProcessAffinityMask))
{
/* Take the value from the image configuration directory */
ProcessAffinityMask = ImageConfigData->ProcessAffinityMask;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Check if this is a UP image
if (Characteristics & IMAGE_FILE_UP_SYSTEM_ONLY)
{
//
// Force it to use CPU 0
//
[NTOSKRNL]: NLS tables should be mapped before PEB. [NTOSKRNL]: Handle failure of NLS and PEB mapping correctly. [NTOSKRNL]: Only set Peb->OsCSDVersion if NtHeader->Win32VersionValue is set. svn path=/trunk/; revision=53059
2011-08-04 13:10:57 +00:00
/* FIXME: this should use the MmRotatingUniprocessorNumber */
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
Peb->ImageProcessAffinityMask = 0;
}
else
{
//
// Whatever was configured
//
Peb->ImageProcessAffinityMask = ProcessAffinityMask;
}
}
_SEH2_EXCEPT(EXCEPTION_EXECUTE_HANDLER)
{
//
// Fail
//
KeDetachProcess();
_SEH2_YIELD(return STATUS_INVALID_IMAGE_PROTECT);
}
_SEH2_END;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Detach from the Process
//
KeDetachProcess();
*BasePeb = Peb;
return STATUS_SUCCESS;
}
NTSTATUS
NTAPI
MmCreateTeb(IN PEPROCESS Process,
IN PCLIENT_ID ClientId,
IN PINITIAL_TEB InitialTeb,
OUT PTEB *BaseTeb)
{
PTEB Teb;
NTSTATUS Status = STATUS_SUCCESS;
*BaseTeb = NULL;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Attach to Target
//
KeAttachProcess(&Process->Pcb);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Allocate the TEB
//
[NTOS]: Allocate the PEB at a pseudo-random address just like Windows XP SP2 and later, to prevent certain kinds of exploits. [NTOS]: Allocate PEB and TEB using VADs! The user-mode fault handler seems to work fine, and I could find no regressions. This is the beginning of the end for MAREAs for VM allocations (they will remain for sections). svn path=/trunk/; revision=48194
2010-07-22 19:08:45 +00:00
Status = MiCreatePebOrTeb(Process, sizeof(TEB), (PULONG_PTR)&Teb);
ASSERT(NT_SUCCESS(Status));
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Use SEH in case we can't load the TEB
//
_SEH2_TRY
{
//
// Initialize the PEB
//
RtlZeroMemory(Teb, sizeof(TEB));
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Set TIB Data
//
[NTOSKRNL] - Add MM_HAL_VA_END and use it in MiAddHalIoMappings instead of making assumptions about the page tables - Add MM_SHARED_USER_DATA_VA for x86 - use MM_SHARED_USER_DATA_VA in Mm to avoid ambiguities - fix or comment out USER_SHARED_DATA in asm headers, as it should be the kernel mode address here - set Teb->ExceptionList to NULL on amd64, it is used as a link to the Wow64 TEB, if any svn path=/trunk/; revision=55445
2012-02-06 00:24:31 +00:00
#ifdef _M_AMD64
Teb->NtTib.ExceptionList = NULL;
#else
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Teb->NtTib.ExceptionList = EXCEPTION_CHAIN_END;
[NTOSKRNL] - Add MM_HAL_VA_END and use it in MiAddHalIoMappings instead of making assumptions about the page tables - Add MM_SHARED_USER_DATA_VA for x86 - use MM_SHARED_USER_DATA_VA in Mm to avoid ambiguities - fix or comment out USER_SHARED_DATA in asm headers, as it should be the kernel mode address here - set Teb->ExceptionList to NULL on amd64, it is used as a link to the Wow64 TEB, if any svn path=/trunk/; revision=55445
2012-02-06 00:24:31 +00:00
#endif
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Teb->NtTib.Self = (PNT_TIB)Teb;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Identify this as an OS/2 V3.0 ("Cruiser") TIB
//
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Teb->NtTib.Version = 30 << 8;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Set TEB Data
//
Teb->ClientId = *ClientId;
Teb->RealClientId = *ClientId;
Teb->ProcessEnvironmentBlock = Process->Peb;
Teb->CurrentLocale = PsDefaultThreadLocaleId;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Check if we have a grandparent TEB
//
if ((InitialTeb->PreviousStackBase == NULL) &&
(InitialTeb->PreviousStackLimit == NULL))
{
//
- Fix inverted if condition in TEB initialization, spotted by Stefan. svn path=/trunk/; revision=43479
2009-10-15 16:47:26 +00:00
// Use initial TEB values
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Teb->NtTib.StackBase = InitialTeb->StackBase;
Teb->NtTib.StackLimit = InitialTeb->StackLimit;
- Fix inverted if condition in TEB initialization, spotted by Stefan. svn path=/trunk/; revision=43479
2009-10-15 16:47:26 +00:00
Teb->DeallocationStack = InitialTeb->AllocatedStackBase;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
}
else
{
//
- Fix inverted if condition in TEB initialization, spotted by Stefan. svn path=/trunk/; revision=43479
2009-10-15 16:47:26 +00:00
// Use grandparent TEB values
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
[NDK] - Add CLIENT_ID32/64 - Create explicit 32/64 bit versions of PEB, GDI_TEB_BATCH and TEB, using macros. Thanks to Alex for feedback and suggestions. - Fix some member names. - Note to self: fix rostests, too! svn path=/branches/ros-amd64-bringup/; revision=44501
2009-12-10 00:35:12 +00:00
Teb->NtTib.StackBase = InitialTeb->PreviousStackBase;
Teb->NtTib.StackLimit = InitialTeb->PreviousStackLimit;
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
}
- Fix inverted if condition in TEB initialization, spotted by Stefan. svn path=/trunk/; revision=43479
2009-10-15 16:47:26 +00:00
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
//
// Initialize the static unicode string
//
Teb->StaticUnicodeString.MaximumLength = sizeof(Teb->StaticUnicodeBuffer);
Teb->StaticUnicodeString.Buffer = Teb->StaticUnicodeBuffer;
}
_SEH2_EXCEPT(EXCEPTION_EXECUTE_HANDLER)
{
//
// Get error code
//
Status = _SEH2_GetExceptionCode();
}
_SEH2_END;
//
// Return
//
KeDetachProcess();
*BaseTeb = Teb;
return Status;
}
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
VOID
NTAPI
MiInitializeWorkingSetList(IN PEPROCESS CurrentProcess)
{
PMMPFN Pfn1;
[NTOSKRNL/MM] - Make MmWorkingSetList PTE local - Map MmWorkingSetList to Vm.VmWorkingSetList when creating process. svn path=/trunk/; revision=55761
2012-02-20 20:50:36 +00:00
PMMPTE sysPte;
MMPTE tempPte;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
/* Setup some bogus list data */
MmWorkingSetList->LastEntry = CurrentProcess->Vm.MinimumWorkingSetSize;
MmWorkingSetList->HashTable = NULL;
MmWorkingSetList->HashTableSize = 0;
MmWorkingSetList->NumberOfImageWaiters = 0;
MmWorkingSetList->Wsle = (PVOID)0xDEADBABE;
MmWorkingSetList->VadBitMapHint = 1;
MmWorkingSetList->HashTableStart = (PVOID)0xBADAB00B;
MmWorkingSetList->HighestPermittedHashAddress = (PVOID)0xCAFEBABE;
MmWorkingSetList->FirstFree = 1;
MmWorkingSetList->FirstDynamic = 2;
MmWorkingSetList->NextSlot = 3;
MmWorkingSetList->LastInitializedWsle = 4;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
/* The rule is that the owner process is always in the FLINK of the PDE's PFN entry */
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
Pfn1 = MiGetPfnEntry(CurrentProcess->Pcb.DirectoryTableBase[0] >> PAGE_SHIFT);
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
ASSERT(Pfn1->u4.PteFrame == MiGetPfnEntryIndex(Pfn1));
Pfn1->u1.Event = (PKEVENT)CurrentProcess;
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL/MM] - Make MmWorkingSetList PTE local - Map MmWorkingSetList to Vm.VmWorkingSetList when creating process. svn path=/trunk/; revision=55761
2012-02-20 20:50:36 +00:00
/* Map the process working set in kernel space */
sysPte = MiReserveSystemPtes(1, SystemPteSpace);
MI_MAKE_HARDWARE_PTE_KERNEL(&tempPte, sysPte, MM_READWRITE, CurrentProcess->WorkingSetPage);
MI_WRITE_VALID_PTE(sysPte, tempPte);
CurrentProcess->Vm.VmWorkingSetList = MiPteToAddress(sysPte);
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
}
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
NTSTATUS
NTAPI
MmInitializeProcessAddressSpace(IN PEPROCESS Process,
IN PEPROCESS ProcessClone OPTIONAL,
IN PVOID Section OPTIONAL,
IN OUT PULONG Flags,
IN POBJECT_NAME_INFORMATION *AuditName OPTIONAL)
{
NTSTATUS Status = STATUS_SUCCESS;
SIZE_T ViewSize = 0;
PVOID ImageBase = 0;
PROS_SECTION_OBJECT SectionObject = Section;
PMMPTE PointerPte;
KIRQL OldIrql;
PMMPDE PointerPde;
PFN_NUMBER PageFrameNumber;
UNICODE_STRING FileName;
PWCHAR Source;
PCHAR Destination;
USHORT Length = 0;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
MMPTE TempPte;
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* We should have a PDE */
ASSERT(Process->Pcb.DirectoryTableBase[0] != 0);
ASSERT(Process->PdeUpdateNeeded == FALSE);
/* Attach to the process */
KeAttachProcess(&Process->Pcb);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* The address space should now been in phase 1 or 0 */
ASSERT(Process->AddressSpaceInitialized <= 1);
Process->AddressSpaceInitialized = 2;
/* Initialize the Addresss Space lock */
KeInitializeGuardedMutex(&Process->AddressCreationLock);
Process->Vm.WorkingSetExpansionLinks.Flink = NULL;
/* Initialize AVL tree */
ASSERT(Process->VadRoot.NumberGenericTableElements == 0);
Process->VadRoot.BalancedRoot.u1.Parent = &Process->VadRoot.BalancedRoot;
/* Lock PFN database */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
/* Setup the PFN for the PDE base of this process */
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
#ifdef _M_AMD64
PointerPte = MiAddressToPte(PXE_BASE);
#else
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
PointerPte = MiAddressToPte(PDE_BASE);
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
#endif
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
PageFrameNumber = PFN_FROM_PTE(PointerPte);
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
ASSERT(Process->Pcb.DirectoryTableBase[0] == PageFrameNumber * PAGE_SIZE);
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
MiInitializePfn(PageFrameNumber, PointerPte, TRUE);
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* Do the same for hyperspace */
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
#ifdef _M_AMD64
PointerPde = MiAddressToPxe((PVOID)HYPER_SPACE);
#else
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
PointerPde = MiAddressToPde(HYPER_SPACE);
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
#endif
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
PageFrameNumber = PFN_FROM_PTE(PointerPde);
[NTOSKRNL] - Fix a bug in CmpSetSystemValues, where an uninitialized handle would be closed in the failure path. - Add a hack on top of the MI_GET_ROS_DATA(x) hack so that we can squeeze a pointer into a 32 bit field. Make MmInitializeProcessAddressSpace amd64 ready and use a portable way to determine the page table base pfn in MiInitializeWorkingSetList - Make MmProbeAndLockPages ready for 3 and 4 level page tables add MiIsPteOnP*eBoundary macros - use these in MmProbeAndLockPages - Raise IRQL to SYNCH_LEVEL not DISPATCH_LEVEL in KiAcquireDispatcherLock - Add MiNonPagedSystemSize for all architectures - Fix amd64 definition of KERNEL_HANDLE_FLAG - Fix definition of PrototypePte - Fix KiGetLinkedTrapFrame() - Make MmProtectTpPteMask 64 bit wide - Fix definition of MI_PTE_LOOKUP_NEEDED for amd64 - Impllement KiSendEOI() to be able to send an EOI from C code. - Fix some MSVC/amd64 warnings svn path=/trunk/; revision=55423
2012-02-04 23:08:20 +00:00
//ASSERT(Process->Pcb.DirectoryTableBase[0] == PageFrameNumber * PAGE_SIZE); // we're not lucky
[NTOS]: Fix incorrect assumptions that a PDE == PTE which have crept up throughout new ARM3 code. [NTOS]: Fix a couple more unportable directly "Hard" PTE field accesses, instead of using portable MI macros. [NTOS]: Cleanup some of the pool code that had really messed up PDE vs PTE. [NTOS]: Define the number of page directoty levels for ARM. [NTOS]: Comment out some ARM3 code that won't be needed now that most of the x86 code is in C. Fixes all but one build error. svn path=/trunk/; revision=49774
2010-11-24 15:21:45 +00:00
MiInitializePfn(PageFrameNumber, (PMMPTE)PointerPde, TRUE);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
/* Setup the PFN for the PTE for the working set */
PointerPte = MiAddressToPte(MI_WORKING_SET_LIST);
MI_MAKE_HARDWARE_PTE(&TempPte, PointerPte, MM_READWRITE, 0);
ASSERT(PointerPte->u.Long != 0);
PageFrameNumber = PFN_FROM_PTE(PointerPte);
MI_WRITE_INVALID_PTE(PointerPte, DemandZeroPte);
MiInitializePfn(PageFrameNumber, PointerPte, TRUE);
TempPte.u.Hard.PageFrameNumber = PageFrameNumber;
MI_WRITE_VALID_PTE(PointerPte, TempPte);
/* Now initialize the working set list */
MiInitializeWorkingSetList(Process);
/* Sanity check */
ASSERT(Process->PhysicalVadRoot == NULL);
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* Release PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
/* Lock the VAD, ARM3-owned ranges away */
[NTOSKRNL] - Modify MiRosTakeOverPebTebRanges to only create a memory area for the shared user page and rename it to MiRosTakeOverSharedUserPage. Previously it was creating a memory area for the whole region from USER_SHARED_DATA up to MM_HIGHEST_USER_ADDRESS, which is the majority of the x64 user mode address space and doesn't even contain the PEB/TEB. Those are allocated below the shared user page and get their memory areas created in MiInsertNode. - Add amd64 versions of MmGetPageTableForProcess, MmUnmapPageTable, MmGetPageFileMapping svn path=/trunk/; revision=55438
2012-02-05 18:41:37 +00:00
MiRosTakeOverSharedUserPage(Process);
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* Check if there's a Section Object */
if (SectionObject)
{
/* Determine the image file name and save it to EPROCESS */
FileName = SectionObject->FileObject->FileName;
Source = (PWCHAR)((PCHAR)FileName.Buffer + FileName.Length);
if (FileName.Buffer)
{
/* Loop the file name*/
while (Source > FileName.Buffer)
{
/* Make sure this isn't a backslash */
if (*--Source == OBJ_NAME_PATH_SEPARATOR)
{
/* If so, stop it here */
Source++;
break;
}
else
{
/* Otherwise, keep going */
Length++;
}
}
}
/* Copy the to the process and truncate it to 15 characters if necessary */
Destination = Process->ImageFileName;
Length = min(Length, sizeof(Process->ImageFileName) - 1);
while (Length--) *Destination++ = (UCHAR)*Source++;
*Destination = ANSI_NULL;
/* Check if caller wants an audit name */
if (AuditName)
{
/* Setup the audit name */
Status = SeInitializeProcessAuditName(SectionObject->FileObject,
FALSE,
AuditName);
if (!NT_SUCCESS(Status))
{
/* Fail */
KeDetachProcess();
return Status;
}
}
/* Map the section */
Status = MmMapViewOfSection(Section,
Process,
(PVOID*)&ImageBase,
0,
0,
NULL,
&ViewSize,
0,
MEM_COMMIT,
PAGE_READWRITE);
/* Save the pointer */
Process->SectionBaseAddress = ImageBase;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* Be nice and detach */
KeDetachProcess();
/* Return status to caller */
return Status;
}
NTSTATUS
NTAPI
[NTOS]: Mark all initialization functions INIT_FUNCTION. [NTOS]: This will save ~280KB of RAM once we dump INIT from RAM in a future patch. [NTOS]: Display current free RAM at certain checkpoints. svn path=/trunk/; revision=49445
2010-11-02 16:29:06 +00:00
INIT_FUNCTION
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
MmInitializeHandBuiltProcess(IN PEPROCESS Process,
IN PULONG_PTR DirectoryTableBase)
{
/* Share the directory base with the idle process */
DirectoryTableBase[0] = PsGetCurrentProcess()->Pcb.DirectoryTableBase[0];
DirectoryTableBase[1] = PsGetCurrentProcess()->Pcb.DirectoryTableBase[1];
/* Initialize the Addresss Space */
KeInitializeGuardedMutex(&Process->AddressCreationLock);
KeInitializeSpinLock(&Process->HyperSpaceLock);
Process->Vm.WorkingSetExpansionLinks.Flink = NULL;
ASSERT(Process->VadRoot.NumberGenericTableElements == 0);
Process->VadRoot.BalancedRoot.u1.Parent = &Process->VadRoot.BalancedRoot;
[NTOSKRNL/MM] - initialize VmWorkingSet pointer for system process. Should fix random crashes when paging out. svn path=/trunk/; revision=56012
2012-03-04 20:30:42 +00:00
/* Use idle process Working set */
Process->Vm.VmWorkingSetList = PsGetCurrentProcess()->Vm.VmWorkingSetList;
[NTOSKRNL] - Make file offsets 64 bit - Fix lots of MSVC/64 bit warnings svn path=/trunk/; revision=56264
2012-03-28 19:41:40 +00:00
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
/* Done */
Process->HasAddressSpace = TRUE;//??
return STATUS_SUCCESS;
}
NTSTATUS
NTAPI
[NTOS]: Mark all initialization functions INIT_FUNCTION. [NTOS]: This will save ~280KB of RAM once we dump INIT from RAM in a future patch. [NTOS]: Display current free RAM at certain checkpoints. svn path=/trunk/; revision=49445
2010-11-02 16:29:06 +00:00
INIT_FUNCTION
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
MmInitializeHandBuiltProcess2(IN PEPROCESS Process)
{
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
/* Lock the VAD, ARM3-owned ranges away */
[NTOSKRNL] - Modify MiRosTakeOverPebTebRanges to only create a memory area for the shared user page and rename it to MiRosTakeOverSharedUserPage. Previously it was creating a memory area for the whole region from USER_SHARED_DATA up to MM_HIGHEST_USER_ADDRESS, which is the majority of the x64 user mode address space and doesn't even contain the PEB/TEB. Those are allocated below the shared user page and get their memory areas created in MiInsertNode. - Add amd64 versions of MmGetPageTableForProcess, MmUnmapPageTable, MmGetPageFileMapping svn path=/trunk/; revision=55438
2012-02-05 18:41:37 +00:00
MiRosTakeOverSharedUserPage(Process);
[NTOS]: Take over MmInitializeProcessAddressSpace, MmInitializeHandBuiltProcess, MmInitializeHandBuiltProcess2 into ARM3, and cleanup the code. [NTOS]: Prepare to take over MmCreateProcessAddressSpace. svn path=/trunk/; revision=48231
2010-07-24 15:01:05 +00:00
return STATUS_SUCCESS;
}
[NTOSKRNL] - Fix MiAddressToPti and implement MiAddressToPxi for amd64 - Replace #error with DPRINT and ASSERT(FALSE) in MiInitializeLargePageSupport - Implement amd64 specific MmCreateProcessAddressSpace - Add MmProtectToPteMask for amd64 (copied from x86) - Remove amd64 version of MmInitializeHandBuiltProcess svn path=/trunk/; revision=48249
2010-07-25 12:00:26 +00:00
#ifdef _M_IX86
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
/* FIXME: Evaluate ways to make this portable yet arch-specific */
BOOLEAN
NTAPI
MmCreateProcessAddressSpace(IN ULONG MinWs,
IN PEPROCESS Process,
OUT PULONG_PTR DirectoryTableBase)
{
KIRQL OldIrql;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
PFN_NUMBER PdeIndex, HyperIndex, WsListIndex;
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
PMMPTE PointerPte;
MMPTE TempPte, PdePte;
ULONG PdeOffset;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
PMMPTE SystemTable, HyperTable;
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
ULONG Color;
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
PMMPFN Pfn1;
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* Choose a process color */
[HAL/NDK] - Make Vector parameter in HalEnableSystemInterrupt, HalDisableSystemInterrupt and HalBeginSystemInterrupt an ULONG, not an UCHAR [NDK] - 64bit fixes for HANDLE_TABLE, KPROCESS, SECTION_IMAGE_INFORMATION, MMADDRESS_LIST, MMVAD_FLAGS, MMVAD, MMVAD_LONG, MMVAD_SHORT, MEMORY_DESCRIPTOR, MEMORY_ALLOCATION_DESCRIPTOR, LdrVerifyMappedImageMatchesChecksum - KDPC_DATA::DpcQueueDepth is signed on amd64, unsigned on x86 [NTOSKRNL] - Fix hundreds of MSVC and amd64 warnings - add a pragma message to FstubFixupEfiPartition, since it looks broken - Move portable Ke constants from <arch>/cpu.c to krnlinit.c - Fixed a bug in amd64 KiGeneralProtectionFaultHandler svn path=/trunk/; revision=53734
2011-09-18 13:11:45 +00:00
Process->NextPageColor = (USHORT)RtlRandom(&MmProcessColorSeed);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
/* Setup the hyperspace lock */
KeInitializeSpinLock(&Process->HyperSpaceLock);
/* Lock PFN database */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* Get a zero page for the PDE, if possible */
Color = MI_GET_NEXT_PROCESS_COLOR(Process);
[NTOS]: Add page-tracking macro calls wherever needed. MI_TRACE_PFNS is still off, though, so this code won't be called. svn path=/trunk/; revision=49444
2010-11-02 15:16:22 +00:00
MI_SET_USAGE(MI_USAGE_PAGE_DIRECTORY);
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
PdeIndex = MiRemoveZeroPageSafe(Color);
if (!PdeIndex)
{
/* No zero pages, grab a free one */
PdeIndex = MiRemoveAnyPage(Color);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(PdeIndex);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* Get a zero page for hyperspace, if possible */
[NTOS]: Add page-tracking macro calls wherever needed. MI_TRACE_PFNS is still off, though, so this code won't be called. svn path=/trunk/; revision=49444
2010-11-02 15:16:22 +00:00
MI_SET_USAGE(MI_USAGE_PAGE_DIRECTORY);
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
Color = MI_GET_NEXT_PROCESS_COLOR(Process);
HyperIndex = MiRemoveZeroPageSafe(Color);
if (!HyperIndex)
{
/* No zero pages, grab a free one */
HyperIndex = MiRemoveAnyPage(Color);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(HyperIndex);
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Get a zero page for the woring set list, if possible */
MI_SET_USAGE(MI_USAGE_PAGE_TABLE);
Color = MI_GET_NEXT_PROCESS_COLOR(Process);
WsListIndex = MiRemoveZeroPageSafe(Color);
if (!WsListIndex)
{
/* No zero pages, grab a free one */
WsListIndex = MiRemoveAnyPage(Color);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(WsListIndex);
[NTOS]: Add MiRemoveZeroPageSafe helper function, when a zero page is required, but the inline zeroing of MiRemoveZeroPage is not. This function will only try grabbing a zero page if one exists, otherwise a free page will be grabbed and zeroed with custom code of the caller's choosing. [NTOS]: Add concept of process color and system color. Compute correct color to use whenever requesting a page. [NTOS]: Uncondtionally enable the color code when inserting/removing pages. For now, when requesting a page, colors are still ignored, and the global PFN lists are scanned instead. If there are no regressions, we are one patch away from that. svn path=/trunk/; revision=48927
2010-09-29 01:10:28 +00:00
}
else
{
/* Release the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
}
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
/* Switch to phase 1 initialization */
ASSERT(Process->AddressSpaceInitialized == 0);
Process->AddressSpaceInitialized = 1;
/* Set the base directory pointers */
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
Process->WorkingSetPage = WsListIndex;
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
DirectoryTableBase[0] = PdeIndex << PAGE_SHIFT;
DirectoryTableBase[1] = HyperIndex << PAGE_SHIFT;
/* Make sure we don't already have a page directory setup */
ASSERT(Process->Pcb.DirectoryTableBase[0] == 0);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
/* Get a PTE to map hyperspace */
PointerPte = MiReserveSystemPtes(1, SystemPteSpace);
ASSERT(PointerPte != NULL);
/* Build it */
MI_MAKE_HARDWARE_PTE_KERNEL(&PdePte,
PointerPte,
MM_READWRITE,
HyperIndex);
/* Set it dirty and map it */
[NTOS]: Fix incorrect assumptions that a PDE == PTE which have crept up throughout new ARM3 code. [NTOS]: Fix a couple more unportable directly "Hard" PTE field accesses, instead of using portable MI macros. [NTOS]: Cleanup some of the pool code that had really messed up PDE vs PTE. [NTOS]: Define the number of page directoty levels for ARM. [NTOS]: Comment out some ARM3 code that won't be needed now that most of the x86 code is in C. Fixes all but one build error. svn path=/trunk/; revision=49774
2010-11-24 15:21:45 +00:00
MI_MAKE_DIRTY_PAGE(&PdePte);
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
MI_WRITE_VALID_PTE(PointerPte, PdePte);
/* Now get hyperspace's page table */
HyperTable = MiPteToAddress(PointerPte);
/* Now write the PTE/PDE entry for the working set list index itself */
TempPte = ValidKernelPte;
TempPte.u.Hard.PageFrameNumber = WsListIndex;
[NTOSKRNL/MM] - Make MmWorkingSetList PTE local - Map MmWorkingSetList to Vm.VmWorkingSetList when creating process. svn path=/trunk/; revision=55761
2012-02-20 20:50:36 +00:00
/* Hyperspace is local */
MI_MAKE_LOCAL_PAGE(&TempPte);
[NTOS]: Assign a working set to the system process and correctly initialize its address space. [NTOS]: Assign the working set list address, system-wide, but per-process (in hyperspace). [NTOS]: Give every process its working set page, and store it. Build a bogus working set list (MMWSL). [NTOS]: Use the process working set list (MMWSL) to track page table references during faults, just as Windows does. [NTOS]: Correctly initialize the colored page list heads and assert their validity. svn path=/trunk/; revision=49525
2010-11-08 12:35:50 +00:00
PdeOffset = MiAddressToPteOffset(MmWorkingSetList);
HyperTable[PdeOffset] = TempPte;
/* Let go of the system PTE */
MiReleaseSystemPtes(PointerPte, 1, SystemPteSpace);
/* Save the PTE address of the page directory itself */
Pfn1 = MiGetPfnEntry(PdeIndex);
Pfn1->PteAddress = (PMMPTE)PDE_BASE;
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
/* Insert us into the Mm process list */
InsertTailList(&MmProcessList, &Process->MmProcessLinks);
/* Get a PTE to map the page directory */
PointerPte = MiReserveSystemPtes(1, SystemPteSpace);
ASSERT(PointerPte != NULL);
/* Build it */
MI_MAKE_HARDWARE_PTE_KERNEL(&PdePte,
PointerPte,
MM_READWRITE,
PdeIndex);
/* Set it dirty and map it */
[NTOS]: Fix incorrect assumptions that a PDE == PTE which have crept up throughout new ARM3 code. [NTOS]: Fix a couple more unportable directly "Hard" PTE field accesses, instead of using portable MI macros. [NTOS]: Cleanup some of the pool code that had really messed up PDE vs PTE. [NTOS]: Define the number of page directoty levels for ARM. [NTOS]: Comment out some ARM3 code that won't be needed now that most of the x86 code is in C. Fixes all but one build error. svn path=/trunk/; revision=49774
2010-11-24 15:21:45 +00:00
MI_MAKE_DIRTY_PAGE(&PdePte);
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
MI_WRITE_VALID_PTE(PointerPte, PdePte);
/* Now get the page directory (which we'll double map, so call it a page table */
SystemTable = MiPteToAddress(PointerPte);
/* Copy all the kernel mappings */
PdeOffset = MiGetPdeOffset(MmSystemRangeStart);
RtlCopyMemory(&SystemTable[PdeOffset],
MiAddressToPde(MmSystemRangeStart),
PAGE_SIZE - PdeOffset * sizeof(MMPTE));
/* Now write the PTE/PDE entry for hyperspace itself */
TempPte = ValidKernelPte;
TempPte.u.Hard.PageFrameNumber = HyperIndex;
PdeOffset = MiGetPdeOffset(HYPER_SPACE);
SystemTable[PdeOffset] = TempPte;
/* Sanity check */
PdeOffset++;
ASSERT(MiGetPdeOffset(MmHyperSpaceEnd) >= PdeOffset);
/* Now do the x86 trick of making the PDE a page table itself */
PdeOffset = MiGetPdeOffset(PTE_BASE);
TempPte.u.Hard.PageFrameNumber = PdeIndex;
SystemTable[PdeOffset] = TempPte;
/* Let go of the system PTE */
MiReleaseSystemPtes(PointerPte, 1, SystemPteSpace);
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
/* Add the process to the session */
MiSessionAddProcess(Process);
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
return TRUE;
}
[NTOSKRNL] - Fix MiAddressToPti and implement MiAddressToPxi for amd64 - Replace #error with DPRINT and ASSERT(FALSE) in MiInitializeLargePageSupport - Implement amd64 specific MmCreateProcessAddressSpace - Add MmProtectToPteMask for amd64 (copied from x86) - Remove amd64 version of MmInitializeHandBuiltProcess svn path=/trunk/; revision=48249
2010-07-25 12:00:26 +00:00
#endif
[NTOS]: Reimplement MmCreateProcessAddressSpace in ARM3. Basically the same as before but using the ARM3 PFN list APIs, the ARM3 macros, etc. Once change is processes are now populating the MmProcessList, and the "MmGlobalKernelPageTable" isn't used anymore for new processes. Also the mappings come from SysPTE space, not hyperspace. [NTOS]: More work will be needed in this area, but this gets rid of another ReactOS dinosaur. svn path=/trunk/; revision=48233
2010-07-24 15:30:24 +00:00
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
VOID
NTAPI
MmCleanProcessAddressSpace(IN PEPROCESS Process)
{
PMMVAD Vad;
PMM_AVL_TABLE VadTree;
PETHREAD Thread = PsGetCurrentThread();
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Sanity checks when dealing with creating/cleaning process address space. [NTOS]: Set the VmDeleted flag when the process address space is being nuked. svn path=/trunk/; revision=49434
2010-11-02 14:41:03 +00:00
/* Only support this */
ASSERT(Process->AddressSpaceInitialized == 2);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
/* Remove from the session */
MiSessionRemoveProcess();
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
/* Lock the process address space from changes */
MmLockAddressSpace(&Process->Vm);
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiLockProcessWorkingSetUnsafe(Process, Thread);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Sanity checks when dealing with creating/cleaning process address space. [NTOS]: Set the VmDeleted flag when the process address space is being nuked. svn path=/trunk/; revision=49434
2010-11-02 14:41:03 +00:00
/* VM is deleted now */
Process->VmDeleted = TRUE;
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiUnlockProcessWorkingSetUnsafe(Process, Thread);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
/* Enumerate the VADs */
VadTree = &Process->VadRoot;
while (VadTree->NumberGenericTableElements)
{
/* Grab the current VAD */
Vad = (PMMVAD)VadTree->BalancedRoot.RightChild;
/* Lock the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiLockProcessWorkingSetUnsafe(Process, Thread);
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
/* Remove this VAD from the tree */
ASSERT(VadTree->NumberGenericTableElements >= 1);
MiRemoveNode((PMMADDRESS_NODE)Vad, VadTree);
[NTOS]: Add support for unmapping ARM3 sections, destroying segments and control areas, and clearing out subsection PTEs. [NTOS]: Add support to MiDeletePte/MiDeleteVirtualAddresses to handle Section VADs. [NTOS]: Add support to MiDeletePte to handle valid, prototype PTEs. [NTOS]: Add MEM_TOP_DOWN support to ARM3 section code. [NTOS]: Add support for unmapping currently mapped ARM3 section views at process termination. [NTOS]: Use the new ARM3 section code for mapping the NLS section in the system (tests the system-view mapping code) and in each new process (tests the data-mapping code). Section is correctly unmapped at process termination time! svn path=/trunk/; revision=49206
2010-10-19 17:07:11 +00:00
/* Only regular VADs supported for now */
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
ASSERT(Vad->u.VadFlags.VadType == VadNone);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add support for unmapping ARM3 sections, destroying segments and control areas, and clearing out subsection PTEs. [NTOS]: Add support to MiDeletePte/MiDeleteVirtualAddresses to handle Section VADs. [NTOS]: Add support to MiDeletePte to handle valid, prototype PTEs. [NTOS]: Add MEM_TOP_DOWN support to ARM3 section code. [NTOS]: Add support for unmapping currently mapped ARM3 section views at process termination. [NTOS]: Use the new ARM3 section code for mapping the NLS section in the system (tests the system-view mapping code) and in each new process (tests the data-mapping code). Section is correctly unmapped at process termination time! svn path=/trunk/; revision=49206
2010-10-19 17:07:11 +00:00
/* Check if this is a section VAD */
if (!(Vad->u.VadFlags.PrivateMemory) && (Vad->ControlArea))
{
/* Remove the view */
MiRemoveMappedView(Process, Vad);
}
else
{
/* Delete the addresses */
MiDeleteVirtualAddresses(Vad->StartingVpn << PAGE_SHIFT,
(Vad->EndingVpn << PAGE_SHIFT) | (PAGE_SIZE - 1),
Vad);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Add support for unmapping ARM3 sections, destroying segments and control areas, and clearing out subsection PTEs. [NTOS]: Add support to MiDeletePte/MiDeleteVirtualAddresses to handle Section VADs. [NTOS]: Add support to MiDeletePte to handle valid, prototype PTEs. [NTOS]: Add MEM_TOP_DOWN support to ARM3 section code. [NTOS]: Add support for unmapping currently mapped ARM3 section views at process termination. [NTOS]: Use the new ARM3 section code for mapping the NLS section in the system (tests the system-view mapping code) and in each new process (tests the data-mapping code). Section is correctly unmapped at process termination time! svn path=/trunk/; revision=49206
2010-10-19 17:07:11 +00:00
/* Release the working set */
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
MiUnlockProcessWorkingSetUnsafe(Process, Thread);
[NTOS]: Add support for unmapping ARM3 sections, destroying segments and control areas, and clearing out subsection PTEs. [NTOS]: Add support to MiDeletePte/MiDeleteVirtualAddresses to handle Section VADs. [NTOS]: Add support to MiDeletePte to handle valid, prototype PTEs. [NTOS]: Add MEM_TOP_DOWN support to ARM3 section code. [NTOS]: Add support for unmapping currently mapped ARM3 section views at process termination. [NTOS]: Use the new ARM3 section code for mapping the NLS section in the system (tests the system-view mapping code) and in each new process (tests the data-mapping code). Section is correctly unmapped at process termination time! svn path=/trunk/; revision=49206
2010-10-19 17:07:11 +00:00
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Use the Spare flag in the VAD as a ReactOS/MemoryArea specific flag to signify that this VAD is associated with a MEMORY_AREA and should be unlinked at process exit, but not freed. This is because MemoryAreas themselves are cleaned up later, and in the future their associated VADs (not yet in Trunk) will also be parsed. In the process death scenario, those VADs will be freed, but not unlinked (since it would already have been unlinked). svn path=/trunk/; revision=48990
2010-10-05 05:00:19 +00:00
/* Skip ARM3 fake VADs, they'll be freed by MmDeleteProcessAddresSpace */
if (Vad->u.VadFlags.Spare == 1)
{
/* Set a flag so MmDeleteMemoryArea knows to free, but not to remove */
Vad->u.VadFlags.Spare = 2;
continue;
}
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
/* Free the VAD memory */
ExFreePool(Vad);
}
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
/* Lock the working set */
MiLockProcessWorkingSetUnsafe(Process, Thread);
ASSERT(Process->CloneRoot == NULL);
ASSERT(Process->PhysicalVadRoot == NULL);
[NTOSKRNL/MM] - finally, release user shared data at process address space cleanup. - release PDE pages that might not be freed at process end. - Let the caller handle PDE release when deleting a PTE - restore Richard's ASSERT : All user PDE pages are now freed! svn path=/trunk/; revision=55770
2012-02-21 00:32:24 +00:00
/* Delete the shared user data section */
MiDeleteVirtualAddresses(USER_SHARED_DATA, USER_SHARED_DATA, NULL);
[NTOSKRNL] Remove all #line 15 "ARM³::BLA" Reasons: - It doesn't provide any benefits, its only purpose was to "look cool" - It never looked cool, instead a character mess appeared - It makes finding the related file harder, especially when the file is named differently then the description or when multiple files have the same tag - It effectively breaks Coverity scans svn path=/trunk/; revision=50149
2010-12-26 15:23:03 +00:00
[NTOSKRNL]: Implement correct locking and unlocking of the working set, one of the biggest blunders in ARM3 so far. [NTOSKRNL]: Implement MiDereferenceControlArea to avoid leaking CAs in failure cases. svn path=/trunk/; revision=57228
2012-09-03 06:23:31 +00:00
/* Release the working set */
MiUnlockProcessWorkingSetUnsafe(Process, Thread);
[NTOS]: Implement MmCleanProcessAddressSpace in ARM3, now the PEB/TEB VADs are removed when the process exits (although the pages are still leaking, for now), and the pool allocation for the VAD is also freed. [NTOS]: Use ARM3 paged pool up until smss.exe starts. There's a last bug in the expansion code before we can get rid of the old paged pool. svn path=/trunk/; revision=48234
2010-07-24 16:12:39 +00:00
/* Release the address space */
MmUnlockAddressSpace(&Process->Vm);
}
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
VOID
NTAPI
MmDeleteProcessAddressSpace2(IN PEPROCESS Process)
{
PMMPFN Pfn1, Pfn2;
KIRQL OldIrql;
PFN_NUMBER PageFrameIndex;
//ASSERT(Process->CommitCharge == 0);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
/* Acquire the PFN lock */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
/* Check for fully initialized process */
if (Process->AddressSpaceInitialized == 2)
{
/* Map the working set page and its page table */
Pfn1 = MiGetPfnEntry(Process->WorkingSetPage);
Pfn2 = MiGetPfnEntry(Pfn1->u4.PteFrame);
/* Nuke it */
MI_SET_PFN_DELETED(Pfn1);
MiDecrementShareCount(Pfn2, Pfn1->u4.PteFrame);
MiDecrementShareCount(Pfn1, Process->WorkingSetPage);
ASSERT((Pfn1->u3.e2.ReferenceCount == 0) || (Pfn1->u3.e1.WriteInProgress));
[NTOSKRNL/MM] - Make MmWorkingSetList PTE local - Map MmWorkingSetList to Vm.VmWorkingSetList when creating process. svn path=/trunk/; revision=55761
2012-02-20 20:50:36 +00:00
MiReleaseSystemPtes(MiAddressToPte(Process->Vm.VmWorkingSetList), 1, SystemPteSpace);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
/* Now map hyperspace and its page table */
PageFrameIndex = Process->Pcb.DirectoryTableBase[1] >> PAGE_SHIFT;
Pfn1 = MiGetPfnEntry(PageFrameIndex);
Pfn2 = MiGetPfnEntry(Pfn1->u4.PteFrame);
/* Nuke it */
MI_SET_PFN_DELETED(Pfn1);
MiDecrementShareCount(Pfn2, Pfn1->u4.PteFrame);
MiDecrementShareCount(Pfn1, PageFrameIndex);
ASSERT((Pfn1->u3.e2.ReferenceCount == 0) || (Pfn1->u3.e1.WriteInProgress));
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
/* Finally, nuke the PDE itself */
PageFrameIndex = Process->Pcb.DirectoryTableBase[0] >> PAGE_SHIFT;
Pfn1 = MiGetPfnEntry(PageFrameIndex);
MI_SET_PFN_DELETED(Pfn1);
MiDecrementShareCount(Pfn1, PageFrameIndex);
MiDecrementShareCount(Pfn1, PageFrameIndex);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL/MM] - finally, release user shared data at process address space cleanup. - release PDE pages that might not be freed at process end. - Let the caller handle PDE release when deleting a PTE - restore Richard's ASSERT : All user PDE pages are now freed! svn path=/trunk/; revision=55770
2012-02-21 00:32:24 +00:00
/* Page table is now dead. Bye bye... */
ASSERT((Pfn1->u3.e2.ReferenceCount == 0) || (Pfn1->u3.e1.WriteInProgress));
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
}
else
{
/* A partly-initialized process should never exit through here */
ASSERT(FALSE);
}
/* Release the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
/* Drop a reference on the session */
if (Process->Session) MiReleaseProcessReferenceToSessionDataPage(Process->Session);
[NTOSKRNL]: Drop the reference on the PDE for kernel stack pages being deleted, now that zekflop has made all PDEs ARM3-owned. svn path=/trunk/; revision=55882
2012-02-27 16:10:30 +00:00
[NTOSKRNL]: This is kind of embarssing, but after doing a local grep for some of cgutman and zekflop's changes (respectively the patch that had added MmDeleteProcessPageDirectory, and the mshtml fix patch), I fell upon a .patch file. It included 3 memory-leak-fixing patches from richard that had been sent to me during his last days -- which, had I paid attention, would've fixed the MSHTML bug and the memory leaks months ago! I've tried my best to now re-integrate a portion of the patch (the other two portions deal with freeing the loader block, and freeing .INIT sections, and I will commit them later) with the current state of trunk. Some things that patch did, no longer seem to work, and I've commented where appropriate. But in general it does seem to dereference/delete some PTEs that had been left behind before (such as deleted TEBs, and deleted VA mappings). It no longer seems to be able to delete the root PDE nor the shared data page however (this worked in the original patch's timeframe). Zekflop, tkreuzer, please take a look. I hope this will set us on a better path! svn path=/trunk/; revision=55666
2012-02-17 07:07:47 +00:00
/* Clear out the PDE pages */
Process->Pcb.DirectoryTableBase[0] = 0;
Process->Pcb.DirectoryTableBase[1] = 0;
}
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
/* SESSION CODE TO MOVE TO SESSION.C ******************************************/
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
PMM_SESSION_SPACE MmSessionSpace;
PFN_NUMBER MiSessionDataPages, MiSessionTagPages, MiSessionTagSizePages;
PFN_NUMBER MiSessionBigPoolPages, MiSessionCreateCharge;
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
KGUARDED_MUTEX MiSessionIdMutex;
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
LONG MmSessionDataPages;
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
PRTL_BITMAP MiSessionIdBitmap;
volatile LONG MiSessionLeaderExists;
VOID
NTAPI
MiInitializeSessionIds(VOID)
{
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
ULONG Size, BitmapSize;
PFN_NUMBER TotalPages;
/* Setup the total number of data pages needed for the structure */
TotalPages = MI_SESSION_DATA_PAGES_MAXIMUM;
MiSessionDataPages = ROUND_TO_PAGES(sizeof(MM_SESSION_SPACE)) >> PAGE_SHIFT;
ASSERT(MiSessionDataPages <= MI_SESSION_DATA_PAGES_MAXIMUM - 3);
TotalPages -= MiSessionDataPages;
/* Setup the number of pages needed for session pool tags */
MiSessionTagSizePages = 2;
MiSessionBigPoolPages = 1;
MiSessionTagPages = MiSessionTagSizePages + MiSessionBigPoolPages;
ASSERT(MiSessionTagPages <= TotalPages);
ASSERT(MiSessionTagPages < MI_SESSION_TAG_PAGES_MAXIMUM);
/* Total pages needed for a session (FIXME: Probably different on PAE/x64) */
MiSessionCreateCharge = 1 + MiSessionDataPages + MiSessionTagPages;
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
/* Initialize the lock */
KeInitializeGuardedMutex(&MiSessionIdMutex);
/* Allocate the bitmap */
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
Size = MI_INITIAL_SESSION_IDS;
BitmapSize = ((Size + 31) / 32) * sizeof(ULONG);
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
MiSessionIdBitmap = ExAllocatePoolWithTag(PagedPool,
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
sizeof(RTL_BITMAP) + BitmapSize,
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
' mM');
if (MiSessionIdBitmap)
{
/* Free all the bits */
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
RtlInitializeBitMap(MiSessionIdBitmap,
(PVOID)(MiSessionIdBitmap + 1),
Size);
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
RtlClearAllBits(MiSessionIdBitmap);
}
else
{
/* Die if we couldn't allocate the bitmap */
KeBugCheckEx(INSTALL_MORE_MEMORY,
MmNumberOfPhysicalPages,
MmLowestPhysicalPage,
MmHighestPhysicalPage,
0x200);
}
}
VOID
NTAPI
MiSessionLeader(IN PEPROCESS Process)
{
KIRQL OldIrql;
/* Set the flag while under the expansion lock */
OldIrql = KeAcquireQueuedSpinLock(LockQueueExpansionLock);
Process->Vm.Flags.SessionLeader = TRUE;
KeReleaseQueuedSpinLock(LockQueueExpansionLock, OldIrql);
}
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
ULONG
NTAPI
MmGetSessionId(IN PEPROCESS Process)
{
PMM_SESSION_SPACE SessionGlobal;
/* The session leader is always session zero */
if (Process->Vm.Flags.SessionLeader == 1) return 0;
/* Otherwise, get the session global, and read the session ID from it */
SessionGlobal = (PMM_SESSION_SPACE)Process->Session;
if (!SessionGlobal) return 0;
return SessionGlobal->SessionId;
}
VOID
NTAPI
MiReleaseProcessReferenceToSessionDataPage(IN PMM_SESSION_SPACE SessionGlobal)
{
ULONG i, SessionId;
PMMPTE PointerPte;
PFN_NUMBER PageFrameIndex[MI_SESSION_DATA_PAGES_MAXIMUM];
PMMPFN Pfn1;
KIRQL OldIrql;
/* Is there more than just this reference? If so, bail out */
if (InterlockedDecrement(&SessionGlobal->ProcessReferenceToSession)) return;
/* Get the session ID */
SessionId = SessionGlobal->SessionId;
DPRINT1("Last process in sessino %d going down!!!\n", SessionId);
/* Free the session page tables */
ExFreePool(SessionGlobal->PageTables);
ASSERT(!MI_IS_PHYSICAL_ADDRESS(SessionGlobal));
/* Capture the data page PFNs */
PointerPte = MiAddressToPte(SessionGlobal);
for (i = 0; i < MiSessionDataPages; i++)
{
PageFrameIndex[i] = PFN_FROM_PTE(PointerPte + i);
}
/* Release them */
MiReleaseSystemPtes(PointerPte, MiSessionDataPages, SystemPteSpace);
/* Mark them as deleted */
for (i = 0; i < MiSessionDataPages; i++)
{
Pfn1 = MI_PFN_ELEMENT(PageFrameIndex[i]);
MI_SET_PFN_DELETED(Pfn1);
}
/* Loop every data page and drop a reference count */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
for (i = 0; i < MiSessionDataPages; i++)
{
/* Sanity check that the page is correct, then decrement it */
Pfn1 = MI_PFN_ELEMENT(PageFrameIndex[i]);
ASSERT(Pfn1->u2.ShareCount == 1);
ASSERT(Pfn1->u3.e2.ReferenceCount == 1);
MiDecrementShareCount(Pfn1, PageFrameIndex[i]);
}
/* Done playing with pages, release the lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
/* Decrement the number of data pages */
InterlockedDecrement(&MmSessionDataPages);
/* Free this session ID from the session bitmap */
KeAcquireGuardedMutex(&MiSessionIdMutex);
ASSERT(RtlCheckBit(MiSessionIdBitmap, SessionId));
RtlClearBit(MiSessionIdBitmap, SessionId);
KeReleaseGuardedMutex(&MiSessionIdMutex);
}
VOID
NTAPI
MiSessionRemoveProcess(VOID)
{
PEPROCESS CurrentProcess = PsGetCurrentProcess();
/* If the process isn't already in a session, or if it's the leader... */
if (!(CurrentProcess->Flags & PSF_PROCESS_IN_SESSION_BIT) ||
(CurrentProcess->Vm.Flags.SessionLeader))
{
/* Then there's nothing to do */
return;
}
/* Sanity check */
ASSERT(MmIsAddressValid(MmSessionSpace) == TRUE);
/* Remove the process from the list ,and dereference the session */
RemoveEntryList(&CurrentProcess->SessionProcessLinks);
//MiDereferenceSession();
}
VOID
NTAPI
MiSessionAddProcess(IN PEPROCESS NewProcess)
{
PMM_SESSION_SPACE SessionGlobal;
/* The current process must already be in a session */
if (!(PsGetCurrentProcess()->Flags & PSF_PROCESS_IN_SESSION_BIT)) return;
/* Sanity check */
ASSERT(MmIsAddressValid(MmSessionSpace) == TRUE);
/* Get the global session */
SessionGlobal = MmSessionSpace->GlobalVirtualAddress;
/* Increment counters */
InterlockedIncrement((PLONG)&SessionGlobal->ReferenceCount);
InterlockedIncrement(&SessionGlobal->ResidentProcessCount);
InterlockedIncrement(&SessionGlobal->ProcessReferenceToSession);
/* Set the session pointer */
ASSERT(NewProcess->Session == NULL);
NewProcess->Session = SessionGlobal;
/* Insert it into the process list */
InsertTailList(&SessionGlobal->ProcessList, &NewProcess->SessionProcessLinks);
/* Set the flag */
PspSetProcessFlag(NewProcess, PSF_PROCESS_IN_SESSION_BIT);
}
NTSTATUS
NTAPI
MiSessionInitializeWorkingSetList(VOID)
{
KIRQL OldIrql;
PMMPTE PointerPte, PointerPde;
MMPTE TempPte;
ULONG Color, Index;
PFN_NUMBER PageFrameIndex;
PMM_SESSION_SPACE SessionGlobal;
BOOLEAN AllocatedPageTable;
PMMWSL WorkingSetList;
/* Get pointers to session global and the session working set list */
SessionGlobal = MmSessionSpace->GlobalVirtualAddress;
WorkingSetList = (PMMWSL)MiSessionSpaceWs;
/* Fill out the two pointers */
MmSessionSpace->Vm.VmWorkingSetList = WorkingSetList;
MmSessionSpace->Wsle = (PMMWSLE)WorkingSetList->UsedPageTableEntries;
/* Get the PDE for the working set, and check if it's already allocated */
PointerPde = MiAddressToPde(WorkingSetList);
if (PointerPde->u.Hard.Valid == 1)
{
/* Nope, we'll have to do it */
ASSERT(PointerPde->u.Hard.Global == 0);
AllocatedPageTable = FALSE;
}
else
{
/* Yep, that makes our job easier */
AllocatedPageTable = TRUE;
}
/* Get the PTE for the working set */
PointerPte = MiAddressToPte(WorkingSetList);
/* Initialize the working set lock, and lock the PFN database */
ExInitializePushLock(&SessionGlobal->Vm.WorkingSetMutex);
//MmLockPageableSectionByHandle(ExPageLockHandle);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
/* Check if we need a page table */
if (AllocatedPageTable == TRUE)
{
/* Get a zeroed colored zero page */
Color = MI_GET_NEXT_COLOR();
PageFrameIndex = MiRemoveZeroPageSafe(Color);
if (!PageFrameIndex)
{
/* No zero pages, grab a free one */
PageFrameIndex = MiRemoveAnyPage(Color);
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(PageFrameIndex);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Write a valid PDE for it */
TempPte.u.Long = ValidKernelPdeLocal.u.Long;
TempPte.u.Hard.PageFrameNumber = PageFrameIndex;
MI_WRITE_VALID_PTE(PointerPde, TempPte);
/* Add this into the list */
Index = ((ULONG_PTR)WorkingSetList - (ULONG_PTR)MmSessionBase) >> 22;
MmSessionSpace->PageTables[Index] = TempPte;
/* Initialize the page directory page, and now zero the working set list itself */
MiInitializePfnForOtherProcess(PageFrameIndex,
PointerPde,
MmSessionSpace->SessionPageDirectoryIndex);
KeZeroPages(PointerPte, PAGE_SIZE);
}
/* Get a zeroed colored zero page */
Color = MI_GET_NEXT_COLOR();
PageFrameIndex = MiRemoveZeroPageSafe(Color);
if (!PageFrameIndex)
{
/* No zero pages, grab a free one */
PageFrameIndex = MiRemoveAnyPage(Color);
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(PageFrameIndex);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Write a valid PTE for it */
TempPte.u.Long = ValidKernelPteLocal.u.Long;
TempPte.u.Hard.Dirty = TRUE;
TempPte.u.Hard.PageFrameNumber = PageFrameIndex;
/* Initialize the working set list page */
MiInitializePfnAndMakePteValid(PageFrameIndex, PointerPte, TempPte);
/* Now we can release the PFN database lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
/* Fill out the working set structure */
MmSessionSpace->Vm.Flags.SessionSpace = 1;
MmSessionSpace->Vm.MinimumWorkingSetSize = 20;
MmSessionSpace->Vm.MaximumWorkingSetSize = 384;
WorkingSetList->LastEntry = 20;
WorkingSetList->HashTable = NULL;
WorkingSetList->HashTableSize = 0;
WorkingSetList->Wsle = MmSessionSpace->Wsle;
/* FIXME: Handle list insertions */
ASSERT(SessionGlobal->WsListEntry.Flink == NULL);
ASSERT(SessionGlobal->WsListEntry.Blink == NULL);
ASSERT(SessionGlobal->Vm.WorkingSetExpansionLinks.Flink == NULL);
ASSERT(SessionGlobal->Vm.WorkingSetExpansionLinks.Blink == NULL);
/* All done, return */
//MmUnlockPageableImageSection(ExPageLockHandle);
return STATUS_SUCCESS;
}
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
NTSTATUS
NTAPI
MiSessionCreateInternal(OUT PULONG SessionId)
{
PEPROCESS Process = PsGetCurrentProcess();
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
ULONG NewFlags, Flags, Size, i, Color;
KIRQL OldIrql;
PMMPTE PointerPte, PageTables, SessionPte;
PMMPDE PointerPde;
PMM_SESSION_SPACE SessionGlobal;
MMPTE TempPte;
NTSTATUS Status;
BOOLEAN Result;
PFN_NUMBER SessionPageDirIndex;
PFN_NUMBER TagPage[MI_SESSION_TAG_PAGES_MAXIMUM];
PFN_NUMBER DataPage[MI_SESSION_DATA_PAGES_MAXIMUM];
/* This should not exist yet */
ASSERT(MmIsAddressValid(MmSessionSpace) == FALSE);
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
/* Loop so we can set the session-is-creating flag */
Flags = Process->Flags;
while (TRUE)
{
/* Check if it's already set */
if (Flags & PSF_SESSION_CREATION_UNDERWAY_BIT)
{
/* Bail out */
DPRINT1("Lost session race\n");
return STATUS_ALREADY_COMMITTED;
}
/* Now try to set it */
NewFlags = InterlockedCompareExchange((PLONG)&Process->Flags,
Flags | PSF_SESSION_CREATION_UNDERWAY_BIT,
Flags);
if (NewFlags == Flags) break;
/* It changed, try again */
Flags = NewFlags;
}
/* Now we should own the flag */
ASSERT(Process->Flags & PSF_SESSION_CREATION_UNDERWAY_BIT);
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/*
* Session space covers everything from 0xA0000000 to 0xC0000000.
* Allocate enough page tables to describe the entire region
*/
Size = (0x20000000 / PDE_MAPPED_VA) * sizeof(MMPTE);
PageTables = ExAllocatePoolWithTag(NonPagedPool, Size, 'tHmM');
ASSERT(PageTables != NULL);
RtlZeroMemory(PageTables, Size);
/* Lock the session ID creation mutex */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
KeAcquireGuardedMutex(&MiSessionIdMutex);
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/* Allocate a new Session ID */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
*SessionId = RtlFindClearBitsAndSet(MiSessionIdBitmap, 1, 0);
if (*SessionId == 0xFFFFFFFF)
{
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/* We ran out of session IDs, we should expand */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
DPRINT1("Too many sessions created. Expansion not yet supported\n");
[NTOSKRNL] - ExFreePool --> ExFreePoolWithTag in some places. - Remove dead code (section.c) - Free resource (sysldr.c) svn path=/trunk/; revision=57412
2012-09-28 12:17:23 +00:00
ExFreePoolWithTag(PageTables, 'tHmM');
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
return STATUS_NO_MEMORY;
}
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/* Unlock the session ID creation mutex */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
KeReleaseGuardedMutex(&MiSessionIdMutex);
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/* Reserve the global PTEs */
SessionPte = MiReserveSystemPtes(MiSessionDataPages, SystemPteSpace);
ASSERT(SessionPte != NULL);
/* Acquire the PFN lock while we set everything up */
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
/* Loop the global PTEs */
TempPte.u.Long = ValidKernelPte.u.Long;
for (i = 0; i < MiSessionDataPages; i++)
{
/* Get a zeroed colored zero page */
Color = MI_GET_NEXT_COLOR();
DataPage[i] = MiRemoveZeroPageSafe(Color);
if (!DataPage[i])
{
/* No zero pages, grab a free one */
DataPage[i] = MiRemoveAnyPage(Color);
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(DataPage[i]);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Fill the PTE out */
TempPte.u.Hard.PageFrameNumber = DataPage[i];
MI_WRITE_VALID_PTE(SessionPte + i, TempPte);
}
/* Set the pointer to global space */
SessionGlobal = MiPteToAddress(SessionPte);
/* Get a zeroed colored zero page */
Color = MI_GET_NEXT_COLOR();
SessionPageDirIndex = MiRemoveZeroPageSafe(Color);
if (!SessionPageDirIndex)
{
/* No zero pages, grab a free one */
SessionPageDirIndex = MiRemoveAnyPage(Color);
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(SessionPageDirIndex);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Fill the PTE out */
TempPte.u.Long = ValidKernelPdeLocal.u.Long;
TempPte.u.Hard.PageFrameNumber = SessionPageDirIndex;
/* Setup, allocate, fill out the MmSessionSpace PTE */
PointerPde = MiAddressToPde(MmSessionSpace);
ASSERT(PointerPde->u.Long == 0);
MI_WRITE_VALID_PTE(PointerPde, TempPte);
MiInitializePfnForOtherProcess(SessionPageDirIndex,
PointerPde,
SessionPageDirIndex);
ASSERT(MI_PFN_ELEMENT(SessionPageDirIndex)->u1.WsIndex == 0);
/* Loop all the local PTEs for it */
TempPte.u.Long = ValidKernelPteLocal.u.Long;
PointerPte = MiAddressToPte(MmSessionSpace);
for (i = 0; i < MiSessionDataPages; i++)
{
/* And fill them out */
TempPte.u.Hard.PageFrameNumber = DataPage[i];
MiInitializePfnAndMakePteValid(DataPage[i], PointerPte + i, TempPte);
ASSERT(MI_PFN_ELEMENT(DataPage[i])->u1.WsIndex == 0);
}
/* Finally loop all of the session pool tag pages */
for (i = 0; i < MiSessionTagPages; i++)
{
/* Grab a zeroed colored page */
Color = MI_GET_NEXT_COLOR();
TagPage[i] = MiRemoveZeroPageSafe(Color);
if (!TagPage[i])
{
/* No zero pages, grab a free one */
TagPage[i] = MiRemoveAnyPage(Color);
/* Zero it outside the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
MiZeroPhysicalPage(TagPage[i]);
OldIrql = KeAcquireQueuedSpinLock(LockQueuePfnLock);
}
/* Fill the PTE out */
TempPte.u.Hard.PageFrameNumber = TagPage[i];
MiInitializePfnAndMakePteValid(TagPage[i],
PointerPte + MiSessionDataPages + i,
TempPte);
}
/* PTEs have been setup, release the PFN lock */
KeReleaseQueuedSpinLock(LockQueuePfnLock, OldIrql);
/* Fill out the session space structure now */
MmSessionSpace->GlobalVirtualAddress = SessionGlobal;
MmSessionSpace->ReferenceCount = 1;
MmSessionSpace->ResidentProcessCount = 1;
MmSessionSpace->u.LongFlags = 0;
MmSessionSpace->SessionId = *SessionId;
MmSessionSpace->LocaleId = PsDefaultSystemLocaleId;
MmSessionSpace->SessionPageDirectoryIndex = SessionPageDirIndex;
MmSessionSpace->Color = Color;
MmSessionSpace->NonPageablePages = MiSessionCreateCharge;
MmSessionSpace->CommittedPages = MiSessionCreateCharge;
MmSessionSpace->PageTables = PageTables;
MmSessionSpace->PageTables[PointerPde - MiAddressToPde(MmSessionBase)] = *PointerPde;
InitializeListHead(&MmSessionSpace->ImageList);
DPRINT1("Session %d is ready to go: 0x%p 0x%p, %lx 0x%p\n",
*SessionId, MmSessionSpace, SessionGlobal, SessionPageDirIndex, PageTables);
/* Initialize session pool */
//Status = MiInitializeSessionPool();
Status = STATUS_SUCCESS;
ASSERT(NT_SUCCESS(Status) == TRUE);
/* Initialize system space */
Result = MiInitializeSystemSpaceMap(&SessionGlobal->Session);
ASSERT(Result == TRUE);
/* Initialize the process list, make sure the workign set list is empty */
ASSERT(SessionGlobal->WsListEntry.Flink == NULL);
ASSERT(SessionGlobal->WsListEntry.Blink == NULL);
InitializeListHead(&SessionGlobal->ProcessList);
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
/* We're done, clear the flag */
ASSERT(Process->Flags & PSF_SESSION_CREATION_UNDERWAY_BIT);
PspClearProcessFlag(Process, PSF_SESSION_CREATION_UNDERWAY_BIT);
[NTOSKRNL]: Implement support for session pool (not yet enabled) and session space (implemented and enabled, but nobody is calling the APIs yet). [NTOSKRNL]: Implement MmMapViewOInSessionSpace, MmUnmapViewInSessionSpace. Win32k needs to use these to we can test them. svn path=/trunk/; revision=56906
2012-07-15 23:42:27 +00:00
/* Insert the process into the session */
ASSERT(Process->Session == NULL);
ASSERT(SessionGlobal->ProcessReferenceToSession == 0);
SessionGlobal->ProcessReferenceToSession = 1;
/* We're done */
InterlockedIncrement(&MmSessionDataPages);
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
return STATUS_SUCCESS;
}
NTSTATUS
NTAPI
MmSessionCreate(OUT PULONG SessionId)
{
PEPROCESS Process = PsGetCurrentProcess();
ULONG SessionLeaderExists;
NTSTATUS Status;
/* Fail if the process is already in a session */
if (Process->Flags & PSF_PROCESS_IN_SESSION_BIT)
{
DPRINT1("Process already in session\n");
return STATUS_ALREADY_COMMITTED;
}
/* Check if the process is already the session leader */
if (!Process->Vm.Flags.SessionLeader)
{
/* Atomically set it as the leader */
SessionLeaderExists = InterlockedCompareExchange(&MiSessionLeaderExists, 1, 0);
if (SessionLeaderExists)
{
DPRINT1("Session leader race\n");
return STATUS_INVALID_SYSTEM_SERVICE;
}
/* Do the work required to upgrade him */
MiSessionLeader(Process);
}
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
/* Create the session */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
KeEnterCriticalRegion();
Status = MiSessionCreateInternal(SessionId);
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
if (!NT_SUCCESS(Status))
{
KeLeaveCriticalRegion();
return Status;
}
/* Set up the session working set */
Status = MiSessionInitializeWorkingSetList();
if (!NT_SUCCESS(Status))
{
/* Fail */
//MiDereferenceSession();
ASSERT(FALSE);
KeLeaveCriticalRegion();
return Status;
}
/* All done */
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
KeLeaveCriticalRegion();
/* Set and assert the flags, and return */
[NTOSKRNL]: Implement last bits of session space support (minus session pool). Processes now have a ->Session pointer! Implement and enable MmSessionId in a bunch of places. [NTOSKRNL]: Set MiSessionSpaceWs address and use where needed. [NTOSKRNL]: Initialize the session working set, and add support to MiLockWorkingSet and MiUnlockWorkingSet for session WS. [NTOSKRNL]: Implement MiSessionAddProcess and MiSessionRemoveProcess which does the work required for this. [NTOSKRNL]: Fix IoGetIrpRequestorProcess. In my private branch, MmMapViewInSessionSpace is working fine for the user and GDI heaps. Will enable after TestBot tests. svn path=/trunk/; revision=57011
2012-08-01 07:54:37 +00:00
MmSessionSpace->u.Flags.Initialized = 1;
[NTOSKRNL]: Stubplement SystemSessionCreateInformation (it allocates a real Session ID and sets the right EPROCESS flags). Nobody uses this yet (future SMSS2 will). [SMSS2]: More attempts at fixing KVM. svn path=/trunk/; revision=55464
2012-02-06 17:41:49 +00:00
PspSetProcessFlag(Process, PSF_PROCESS_IN_SESSION_BIT);
ASSERT(MiSessionLeaderExists == 1);
return Status;
}
[SMSS2/SMSS]: Put SMSS back in charge of the pagefile as yet another attempt to fix KVM. [NTOSKRNL]: Stub-plement SystemSessionDetachInformation. svn path=/trunk/; revision=55465
2012-02-06 18:14:36 +00:00
NTSTATUS
NTAPI
MmSessionDelete(IN ULONG SessionId)
{
PEPROCESS Process = PsGetCurrentProcess();
/* Process must be in a session */
if (!(Process->Flags & PSF_PROCESS_IN_SESSION_BIT))
{
DPRINT1("Not in a session!\n");
return STATUS_UNABLE_TO_FREE_VM;
}
/* It must be the session leader */
if (!Process->Vm.Flags.SessionLeader)
{
DPRINT1("Not a session leader!\n");
return STATUS_UNABLE_TO_FREE_VM;
}
/* Remove one reference count */
KeEnterCriticalRegion();
/* FIXME: Do it */
KeLeaveCriticalRegion();
/* All done */
return STATUS_SUCCESS;
}
- Multiple PEB/TEB creation fixes: - Set up PEB and TEB under SEH to handle possible paging errors (which are legitimate). - Also touch the image under SEH to handle image paging errors (also legitimate). - Should avoid kernel panics in cases where the executable is damaged, invalid, or impossible to page in. - Initialize more PEB fields. - Add new INITIAL_PEB structure to support fork() in the future. Also fixes the fact we don't create a PEB for the system process anymore. - Create PEB while attached to the process. - Handle UP-only images, and set correct affinity mask in the PEB. - Set session ID instead of session structure (currently zero). - Set correct TIB version in the TEB. Due to a historical oddity, the NT TIB identifies itself as '0x1e00', which is 30 shifted 8 bits. 30 is the version identifier of OS/2 3.0, the original 32-bit target of the Windows NT effort. - Handle initial TEB correctly for fork(). - Move AWE APIs to procsup.c instea of section.c since they are unrelated. - Move the rewritten PEB/TEB functions to ARM3's procsup.c. svn path=/trunk/; revision=43476
2009-10-15 05:56:41 +00:00
/* SYSTEM CALLS ***************************************************************/
NTSTATUS
NTAPI
NtAllocateUserPhysicalPages(IN HANDLE ProcessHandle,
IN OUT PULONG_PTR NumberOfPages,
IN OUT PULONG_PTR UserPfnArray)
{
UNIMPLEMENTED;
return STATUS_NOT_IMPLEMENTED;
}
NTSTATUS
NTAPI
NtMapUserPhysicalPages(IN PVOID VirtualAddresses,
IN ULONG_PTR NumberOfPages,
IN OUT PULONG_PTR UserPfnArray)
{
UNIMPLEMENTED;
return STATUS_NOT_IMPLEMENTED;
}
NTSTATUS
NTAPI
NtMapUserPhysicalPagesScatter(IN PVOID *VirtualAddresses,
IN ULONG_PTR NumberOfPages,
IN OUT PULONG_PTR UserPfnArray)
{
UNIMPLEMENTED;
return STATUS_NOT_IMPLEMENTED;
}
NTSTATUS
NTAPI
NtFreeUserPhysicalPages(IN HANDLE ProcessHandle,
IN OUT PULONG_PTR NumberOfPages,
IN OUT PULONG_PTR UserPfnArray)
{
UNIMPLEMENTED;
return STATUS_NOT_IMPLEMENTED;
}
- Reimplement kernel stack allocation (MmCreateKernelStack, MmDeleteKernelStack, MmGrowKernelStack, MmGrowKernelStackEx): - Use System PTEs. - Allocate 12KB stacks with a 4KB guard page, and expand them up to 60KB as required. - Previous implementation always allocated 60KB, without any guard page. - The result is that on a minimal ReactOS install, simply booting up now requires an average of 0.5MB less of physical memory than before. - And once again, optimizations to the system PTE allocation code should significantly improve performance. - Should also analyze benefits of using a dead stack slist as done on Windows. (Assembla Ticket #39). svn path=/trunk/; revision=41636
2009-06-27 08:41:45 +00:00
/* EOF */
Reference in a new issue Copy permalink