Re: [ros-dev] [ros-diffs] [fireball] 34232: Yuri Sidorov <jura@cp-lab.com> - Prevent boolean flags from being optimized away by compiler (due to PSEH usage) by making them volatile. See issue #3408 for more details.
1 Jul
2008
1 Jul
'08
6:28 p.m.
What?
Why would the compiler optimize these away? This doesn't make any sense.
If something is happening to these flags, PSEH should be fixed -- not
making every flag in the source base a volatile.
On 1-Jul-08, at 3:48 AM, fireball(a)svn.reactos.org wrote:
Author: fireball
Date: Tue Jul 1 05:48:50 2008
New Revision: 34232
URL: http://svn.reactos.org/svn/reactos?rev=34232&view=rev
Log:
Yuri Sidorov <jura(a)cp-lab.com>
- Prevent boolean flags from being optimized away by compiler (due
to PSEH usage) by making them volatile.
See issue #3408 for more details.
Modified:
trunk/reactos/ntoskrnl/mm/virtual.c
Modified: trunk/reactos/ntoskrnl/mm/virtual.c
URL:
http://svn.reactos.org/svn/reactos/trunk/reactos/ntoskrnl/mm/virtual.c?rev=…
=
=
=
=
=
=
=
=
======================================================================
--- trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] (original)
+++ trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] Tue Jul 1
05:48:50 2008
@@ -70,7 +70,7 @@
PFN_NUMBER MdlBuffer[(sizeof(MDL) / sizeof(PFN_NUMBER)) +
MI_MAPPED_COPY_PAGES + 1];
PMDL Mdl = (PMDL)MdlBuffer;
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMapping = FALSE,
FailedInMoving;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMapping =
FALSE, FailedInMoving;
BOOLEAN PagesLocked;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID MdlAddress;
@@ -225,7 +225,7 @@
{
UCHAR StackBuffer[MI_POOL_COPY_BYTES];
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMoving, HavePoolAddress
= FALSE;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMoving,
HavePoolAddress = FALSE;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID PoolAddress;
KAPC_STATE ApcState;
Best regards,
Alex Ionescu
1 Jul
1 Jul
7:03 p.m.
New subject: [ros-diffs] [fireball] 34232: Yuri Sidorov<jura@cp-lab.com> - Prevent boolean flags from beingoptimized away by compiler (due to PSEH usage) by making themvolatile. See issue #3408 for more details.
Hi,
It's corner case problem and it occurs only in these functions.
Code:
BOOLEAN f = FALSE;
f = TRUE;
call_some_func();
f = FALSE;
gcc optimizer removes TRUE/FALSE assignments, since it knows that f
was FALSE before this code block and remains FALSE after. gcc knows
nothing about exceptions and SEH and performs such optimizations
here...
Yury Sidorov.
----- Original Message -----
From: "Alex Ionescu" <ionucu(a)videotron.ca>
To: <ros-dev(a)reactos.org>
Sent: Tuesday, July 01, 2008 9:28 PM
Subject: Re: [ros-dev] [ros-diffs] [fireball] 34232: Yuri
Sidorov<jura(a)cp-lab.com> - Prevent boolean flags from beingoptimized
away by compiler (due to PSEH usage) by making themvolatile. See issue
#3408 for more details.
What?
Why would the compiler optimize these away? This doesn't make any
sense.
If something is happening to these flags, PSEH should be fixed --
not
making every flag in the source base a volatile.
On 1-Jul-08, at 3:48 AM, fireball(a)svn.reactos.org wrote:
Author: fireball
Date: Tue Jul 1 05:48:50 2008
New Revision: 34232
URL: http://svn.reactos.org/svn/reactos?rev=34232&view=rev
Log:
Yuri Sidorov <jura(a)cp-lab.com>
- Prevent boolean flags from being optimized away by compiler (due
to PSEH usage) by making them volatile.
See issue #3408 for more details.
Modified:
trunk/reactos/ntoskrnl/mm/virtual.c
Modified: trunk/reactos/ntoskrnl/mm/virtual.c
URL:
http://svn.reactos.org/svn/reactos/trunk/reactos/ntoskrnl/mm/virtual.c?rev=…
=
=
=
=
=
=
=
=
======================================================================
--- trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] (original)
+++ trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] Tue Jul 1
05:48:50 2008
@@ -70,7 +70,7 @@
PFN_NUMBER MdlBuffer[(sizeof(MDL) / sizeof(PFN_NUMBER)) +
MI_MAPPED_COPY_PAGES + 1];
PMDL Mdl = (PMDL)MdlBuffer;
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMapping = FALSE,
FailedInMoving;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMapping =
FALSE, FailedInMoving;
BOOLEAN PagesLocked;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID MdlAddress;
@@ -225,7 +225,7 @@
{
UCHAR StackBuffer[MI_POOL_COPY_BYTES];
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMoving, HavePoolAddress
= FALSE;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMoving,
HavePoolAddress = FALSE;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID PoolAddress;
KAPC_STATE ApcState;
Best regards,
Alex Ionescu
_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://www.reactos.org/mailman/listinfo/ros-dev
9:45 p.m.
New subject: [ros-diffs] [fireball] 34232: Yuri Sidorov<jura@cp-lab.com> - Prevent boolean flags from beingoptimized away by compiler (due to PSEH usage) by making themvolatile. See issue #3408 for more details.
Then gcc or PSEH should be fixed, not worked-around.
On 1-Jul-08, at 12:03 PM, Yury Sidorov wrote:
Hi,
It's corner case problem and it occurs only in these functions.
Code:
BOOLEAN f = FALSE;
f = TRUE;
call_some_func();
f = FALSE;
gcc optimizer removes TRUE/FALSE assignments, since it knows that f
was FALSE before this code block and remains FALSE after. gcc knows
nothing about exceptions and SEH and performs such optimizations
here...
Yury Sidorov.
----- Original Message -----
From: "Alex Ionescu" <ionucu(a)videotron.ca>
To: <ros-dev(a)reactos.org>
Sent: Tuesday, July 01, 2008 9:28 PM
Subject: Re: [ros-dev] [ros-diffs] [fireball] 34232: Yuri
Sidorov<jura(a)cp-lab.com> - Prevent boolean flags from beingoptimized
away by compiler (due to PSEH usage) by making themvolatile. See issue
#3408 for more details.
Best regards,
Alex Ionescu
What?
Why would the compiler optimize these away? This doesn't make any
sense.
If something is happening to these flags, PSEH should be fixed --
not
making every flag in the source base a volatile.
On 1-Jul-08, at 3:48 AM, fireball(a)svn.reactos.org wrote:
_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://www.reactos.org/mailman/listinfo/ros-dev Author: fireball
Date: Tue Jul 1 05:48:50 2008
New Revision: 34232
URL: http://svn.reactos.org/svn/reactos?rev=34232&view=rev
Log:
Yuri Sidorov <jura(a)cp-lab.com>
- Prevent boolean flags from being optimized away by compiler (due
to PSEH usage) by making them volatile.
See issue #3408 for more details.
Modified:
trunk/reactos/ntoskrnl/mm/virtual.c
Modified: trunk/reactos/ntoskrnl/mm/virtual.c
URL:
http://svn.reactos.org/svn/reactos/trunk/reactos/ntoskrnl/mm/virtual.c?rev=…
=
=
=
=
=
=
=
=
=
=
====================================================================
--- trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] (original)
+++ trunk/reactos/ntoskrnl/mm/virtual.c [iso-8859-1] Tue Jul 1
05:48:50 2008
@@ -70,7 +70,7 @@
PFN_NUMBER MdlBuffer[(sizeof(MDL) / sizeof(PFN_NUMBER)) +
MI_MAPPED_COPY_PAGES + 1];
PMDL Mdl = (PMDL)MdlBuffer;
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMapping = FALSE,
FailedInMoving;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMapping =
FALSE, FailedInMoving;
BOOLEAN PagesLocked;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID MdlAddress;
@@ -225,7 +225,7 @@
{
UCHAR StackBuffer[MI_POOL_COPY_BYTES];
ULONG TotalSize, CurrentSize, RemainingSize;
- BOOLEAN FailedInProbe = FALSE, FailedInMoving, HavePoolAddress
= FALSE;
+ volatile BOOLEAN FailedInProbe = FALSE, FailedInMoving,
HavePoolAddress = FALSE;
PVOID CurrentAddress = SourceAddress, CurrentTargetAddress =
TargetAddress;
PVOID PoolAddress;
KAPC_STATE ApcState;
Best regards,
Alex Ionescu
_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://www.reactos.org/mailman/listinfo/ros-dev 
2 Jul
2 Jul
8:16 a.m.
New subject: [ros-diffs] [fireball] 34232: Yuri Sidorov <jura@cp-lab.com> - Prevent boolean flags from being optimized away by compiler (due to PSEH usage) by making them volatile. See issue #3408 for more details.
Alex Ionescu ha scritto:
If something is happening to these flags, PSEH should
be fixed -- not
making every flag in the source base a volatile.
PSEH is based setjmp/longjmp, some local variables WILL be reset when an
exception is caught and they MUST be marked as "volatile". I documented
this a long time ago
11:39 a.m.
New subject: [ros-diffs] [fireball] 34232: Yuri Sidorov <jura@cp-lab.com> - Prevent boolean flags from being optimized away by compiler (due to PSEH usage) by making them volatile. See issue #3408 for more details.
What about wrapping the _SEH_TRY part in a local inline function? In my
tests this successfully prevented the variables from being optimized
away. At least under normal optimisation settings.
_SEH_TRY
{
void inline tryblock()
{
...
}
tryblock();
}
...
KJK::Hyperion schrieb:
Alex Ionescu ha scritto:
If something is happening to these flags, PSEH
should be fixed -- not
making every flag in the source base a volatile.
PSEH is based setjmp/longjmp, some local variables WILL be reset when an
exception is caught and they MUST be marked as "volatile". I documented
this a long time ago
_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://www.reactos.org/mailman/listinfo/ros-dev
1:08 p.m.
New subject: [ros-diffs] [fireball] 34232: Yuri Sidorov <jura@cp-lab.com> - Prevent boolean flags from being optimized away by compiler (due to PSEH usage) by making them volatile. See issue #3408 for more details.
Timo Kreuzer ha scritto:
What about wrapping the _SEH_TRY part in a local
inline function? In my
tests this successfully prevented the variables from being optimized
away. At least under normal optimisation settings.
I find this a terrible, terrible idea. An unreliable side effect of an
unspecified behavior of a non-ANSI C feature. It will break unoptimized
builds as soon as we add support for DEP, too, because local functions
require an executable stack. I am at a loss of words to convey how bad
of an idea this seems to me
6020
days inactive
6021
days old
5 comments
4 participants
participants (4)
-
Alex Ionescu -
KJK::Hyperion -
Timo Kreuzer -
Yury Sidorov