I agree with Alex's reasoning, plus in a multiprocessor environment such page zeroing
may be done asynchronously to other cores or processors attempting to acquire use of that
page while its state is inconsistent. Removing it counts as a security hole, in other
words, except for single core processors that have interrupts disabled, that I see.
Sent from Outlook Mobile<https://aka.ms/blhgte>
________________________________
From: Ros-dev <ros-dev-bounces(a)reactos.org> on behalf of Alex Ionescu
<ionucu(a)videotron.ca>
Sent: Sunday, May 10, 2020 12:37:09 PM
To: ReactOS Development List <ros-dev(a)reactos.org>rg>; Thomas Faber
<thomas.faber(a)reactos.org>
Cc: Linda Wang <ros-diffs(a)reactos.org>
Subject: Re: [ros-dev] [ros-diffs] [reactos] 02/02: [NTOS:MM] Get rid of unnecessary
MmZeroingPageThreadActive.
I know I probably don't have much weight anymore but, I re-iterate
that one of the goals of the project should be to make things as
similar as possible for debugging/troubleshooting and learning as
Server 2003... which does have this variable. I don't think it's the
end of the world to keep it around. Stuff like this also tends to
break kdexts.dll.
Best regards,
Alex Ionescu
On Mon, Apr 6, 2020 at 5:15 AM Thomas Faber <thomas.faber(a)reactos.org> wrote:
https://git.reactos.org/?p=reactos.git;a=commitdiff;h=25a5aee86f43e9f526d4d…
commit 25a5aee86f43e9f526d4d360e4e4f35feb9d22b4
Author: Thomas Faber <thomas.faber(a)reactos.org>
AuthorDate: Sat Feb 22 12:31:54 2020 +0100
Commit: Thomas Faber <thomas.faber(a)reactos.org>
CommitDate: Mon Apr 6 11:13:55 2020 +0200
[NTOS:MM] Get rid of unnecessary MmZeroingPageThreadActive.
---
ntoskrnl/mm/ARM3/miarm.h | 1 -
ntoskrnl/mm/ARM3/mminit.c | 5 ++---
ntoskrnl/mm/ARM3/pfnlist.c | 3 +--
ntoskrnl/mm/ARM3/zeropage.c | 3 +--
4 files changed, 4 insertions(+), 8 deletions(-)
diff --git a/ntoskrnl/mm/ARM3/miarm.h b/ntoskrnl/mm/ARM3/miarm.h
index c242b795af6..9d85c42afff 100644
--- a/ntoskrnl/mm/ARM3/miarm.h
+++ b/ntoskrnl/mm/ARM3/miarm.h
@@ -639,7 +639,6 @@ extern PMMPDE MiHighestUserPde;
extern PFN_NUMBER MmSystemPageDirectory[PPE_PER_PAGE];
extern PMMPTE MmSharedUserDataPte;
extern LIST_ENTRY MmProcessList;
-extern BOOLEAN MmZeroingPageThreadActive;
extern KEVENT MmZeroingPageEvent;
extern ULONG MmSystemPageColor;
extern ULONG MmProcessColorSeed;
diff --git a/ntoskrnl/mm/ARM3/mminit.c b/ntoskrnl/mm/ARM3/mminit.c
index 7cd7af10838..d72e7186193 100644
--- a/ntoskrnl/mm/ARM3/mminit.c
+++ b/ntoskrnl/mm/ARM3/mminit.c
@@ -2181,9 +2181,8 @@ MmArmInitSystem(IN ULONG Phase,
/* Initialize the Loader Lock */
KeInitializeMutant(&MmSystemLoadLock, FALSE);
- /* Set the zero page event */
- KeInitializeEvent(&MmZeroingPageEvent, SynchronizationEvent, FALSE);
- MmZeroingPageThreadActive = FALSE;
+ /* Set up the zero page event */
+ KeInitializeEvent(&MmZeroingPageEvent, NotificationEvent, FALSE);
/* Initialize the dead stack S-LIST */
InitializeSListHead(&MmDeadStackSListHead);
diff --git a/ntoskrnl/mm/ARM3/pfnlist.c b/ntoskrnl/mm/ARM3/pfnlist.c
index eac3a043418..f175541a500 100644
--- a/ntoskrnl/mm/ARM3/pfnlist.c
+++ b/ntoskrnl/mm/ARM3/pfnlist.c
@@ -701,10 +701,9 @@ MiInsertPageInFreeList(IN PFN_NUMBER PageFrameIndex)
ColorTable->Count++;
/* Notify zero page thread if enough pages are on the free list now */
- if ((ListHead->Total >= 8) && !(MmZeroingPageThreadActive))
+ if (ListHead->Total >= 8)
{
/* Set the event */
- MmZeroingPageThreadActive = TRUE;
KeSetEvent(&MmZeroingPageEvent, IO_NO_INCREMENT, FALSE);
}
diff --git a/ntoskrnl/mm/ARM3/zeropage.c b/ntoskrnl/mm/ARM3/zeropage.c
index 6d859f6806f..5fe22294199 100644
--- a/ntoskrnl/mm/ARM3/zeropage.c
+++ b/ntoskrnl/mm/ARM3/zeropage.c
@@ -17,7 +17,6 @@
/* GLOBALS ********************************************************************/
-BOOLEAN MmZeroingPageThreadActive;
KEVENT MmZeroingPageEvent;
/* PRIVATE FUNCTIONS **********************************************************/
@@ -73,7 +72,7 @@ MmZeroPageThread(VOID)
{
if (!MmFreePageListHead.Total)
{
- MmZeroingPageThreadActive = FALSE;
+ KeClearEvent(&MmZeroingPageEvent);
MiReleasePfnLock(OldIrql);
break;
}
_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://reactos.org/mailman/listinfo/ros-dev