[PATCH] kernel32: force 4096 byte page alignment

[Gerald Pfeifer]

FreeBSD prefers to use a 0x200000 (super) page alignment. This causes 
the image base to be displaced from the requested location. Forcing a 
smaller page size ensures the image can be placed there.

This should not have negative impact on other, e.g. Linux-based, 
platforms I believe.

The patch was originally contributed to the FreeBSD emulators/wine(-devel) 
port by David Naylor in February 2015. We have carried it as part of those 
two ever since, with occasional adjustments of mine. I am now submitting 
this here with David's consent.

Signed-off-by: Gerald Pfeifer 
Signed-off-by: David Naylor , 
---
 dlls/kernel32/Makefile.in | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/dlls/kernel32/Makefile.in b/dlls/kernel32/Makefile.in
index 832f2365b15..f932f83c6b6 100644
--- a/dlls/kernel32/Makefile.in
+++ b/dlls/kernel32/Makefile.in
@@ -2,7 +2,7 @@ EXTRADEFS = -D_KERNEL32_ -D_NORMALIZE_
 MODULE    = kernel32.dll
 IMPORTLIB = kernel32
 IMPORTS   = kernelbase ntdll winecrt0
-EXTRADLLFLAGS = -mno-cygwin -nodefaultlibs -Wb,-F,KERNEL32.dll -Wl,--image-base,0x7b600000
+EXTRADLLFLAGS = -mno-cygwin -nodefaultlibs -Wb,-F,KERNEL32.dll -Wl,--image-base,0x7b600000 -Wl,-z,max-page-size=0x1000

C_SRCS = \
    atom.c \

[Stefan Dösinger]

> Am 12.11.2020 um 21:58 schrieb Gerald Pfeifer :
>
> FreeBSD prefers to use a 0x200000 (super) page alignment. This causes
> the image base to be displaced from the requested location. Forcing a
> smaller page size ensures the image can be placed there.

Is kernel32 built as a dll.so without mingw in this case? If it is a PE DLL as it should be then the problem should be fixed in the PE loader I think - it would also affect DLLs shipped by programs etc.

[Gerald Pfeifer]

On Fri, 13 Nov 2020, Stefan Dösinger wrote:
> Is kernel32 built as a dll.so without mingw in this case? If it is a PE 
> DLL as it should be then the problem should be fixed in the PE loader I 
> think - it would also affect DLLs shipped by programs etc.

Ah, good point.  And, yes, the FreeBSD ports don't use MinGW at
this point.

(Per Damjan and Alexandre the patch turns out to have been a noop the 
last while FreeBSD has been carrying it; I'll see to remove it there.)

Thank you,
Gerald

[Damjan Jovanovic]

On Thu, Nov 12, 2020 at 8:58 PM Gerald Pfeifer  wrote:
> FreeBSD prefers to use a 0x200000 (super) page alignment. This causes
> the image base to be displaced from the requested location. Forcing a
> smaller page size ensures the image can be placed there.
>
> This should not have negative impact on other, e.g. Linux-based,
> platforms I believe.
>
> The patch was originally contributed to the FreeBSD emulators/wine(-devel)
> port by David Naylor in February 2015. We have carried it as part of those
> two ever since, with occasional adjustments of mine. I am now submitting
> this here with David's consent.
>
> Signed-off-by: Gerald Pfeifer 
> Signed-off-by: David Naylor ,
>
> dlls/kernel32/Makefile.in | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/dlls/kernel32/Makefile.in b/dlls/kernel32/Makefile.in
> index 832f2365b15..f932f83c6b6 100644
> --- a/dlls/kernel32/Makefile.in
> +++ b/dlls/kernel32/Makefile.in
> @@ -2,7 +2,7 @@ EXTRADEFS = -D_KERNEL32_ -D_NORMALIZE_
>  MODULE    = kernel32.dll
>  IMPORTLIB = kernel32
>  IMPORTS   = kernelbase ntdll winecrt0
> -EXTRADLLFLAGS = -mno-cygwin -nodefaultlibs -Wb,-F,KERNEL32.dll
> -Wl,--image-base,0x7b600000
> +EXTRADLLFLAGS = -mno-cygwin -nodefaultlibs -Wb,-F,KERNEL32.dll
> -Wl,--image-base,0x7b600000 -Wl,-z,max-page-size=0x1000

Hi

Is this patch still necessary? It looks like I already have the correct
base address for kernel32, on both the ELF build, and my hacked-up PE build
(using mingw from Debian running on FreeBSD's Linuxulator).

PE build:
$ cat /proc/23365/map  | grep kernel32
*0x7b600000* 0x7b601000 1 0 0xfffff8026db42000 r-- 72 27 0x1000 COW NC
vnode /compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b601000 0x7b631000 48 0 0xfffff8026db42000 r-x 72 27 0x1000 COW NC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b631000 0x7b635000 4 0 0xfffff8039bd70900 rw- 1 0 0x3000 COW NNC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b635000 0x7b639000 4 0 0xfffff8026db42000 r-- 72 27 0x1000 COW NC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b63a000 0x7b63b000 1 0 0xfffff800b7497d00 rw- 1 0 0x3000 COW NNC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b63b000 0x7b64a000 14 0 0xfffff8026db42000 r-- 72 27 0x1000 COW NC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b64a000 0x7b6a7000 77 0 0xfffff8038bd8c100 rw- 1 0 0x3000 COW NC vnode
/compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003
0x7b6a7000 0x7b78c000 213 0 0xfffff8026db42000 r-- 72 27 0x1000 COW NC
vnode /compat/i386/usr/home/dj/.local/lib/wine/kernel32.dll CH 1003

ELF build:
$ cat /proc/14134/map | grep kernel32
*0x7b600000* 0x7b620000 32 0 0xfffff800a17c1600 r-x 1 0 0x2000 COW NC vnode
/compat/i386/usr/home/dj/.local/opt/wine-elf/lib/wine/kernel32.dll.so CH
1003
0x7b621000 0x7b676000 63 0 0xfffff802666dd500 r-x 1 0 0x3000 COW NC vnode
/compat/i386/usr/home/dj/.local/opt/wine-elf/lib/wine/kernel32.dll.so CH
1003
0x7b676000 0x7b6de000 104 0 0xfffff802664f8000 rw- 1 0 0x3000 COW NNC vnode
/compat/i386/usr/home/dj/.local/opt/wine-elf/lib/wine/kernel32.dll.so CH
1003

Maybe Wine already loads kernel32.dll(.so) at the right base address, after
55ba3648379d90642f174e74809b84130d6d1ddc and
e0138c4a6a7a87f1d25badd43e05490e706bc666 fixed mmap() usage on FreeBSD,
and/or other recent commits?

Damjan

[Marvin]

Hi,

While running your changed tests, I think I found new failures.
Being a bot and all I'm not very good at pattern recognition, so I might be
wrong, but could you please double-check?

Full results can be found at:
https://testbot.winehq.org/JobDetails.pl?Key=81852

Your paranoid android.

=== debiant (build log) ===

Task: Patch failed to apply

=== debiant (build log) ===

Task: Patch failed to apply

[Alexandre Julliard]

Damjan Jovanovic  writes:
> Maybe Wine already loads kernel32.dll(.so) at the right base address,
> after 55ba3648379d90642f174e74809b84130d6d1ddc and
> e0138c4a6a7a87f1d25badd43e05490e706bc666 fixed mmap() usage on
> FreeBSD, and/or other recent commits?

winegcc passes that flag already, so the patch should make no difference.

[Damjan Jovanovic]

On Fri, Nov 13, 2020 at 10:45 PM Alexandre Julliard 
wrote:
> Damjan Jovanovic  writes:
> > Maybe Wine already loads kernel32.dll(.so) at the right base address,
> > after 55ba3648379d90642f174e74809b84130d6d1ddc and
> > e0138c4a6a7a87f1d25badd43e05490e706bc666 fixed mmap() usage on
> > FreeBSD, and/or other recent commits?
>
> winegcc passes that flag already, so the patch should make no difference.

And has done so since wine-1.9.7:

commit e59395208d527d1bffb2725efc85fb798aee7610
Author: Alexandre Julliard 
Date:   Mon Mar 28 15:47:19 2016 +0900

makefiles: Reduce the maximum page size for modules that have a base
address.

It avoid bumping into the next allocated base address on 64-bit.

Signed-off-by: Alexandre Julliard

[Gerald Pfeifer]

On Sat, 14 Nov 2020, Damjan Jovanovic wrote:
> > winegcc passes that flag already, so the patch should make no difference.
>
> And has done so since wine-1.9.7:
>
> commit e59395208d527d1bffb2725efc85fb798aee7610
> Author: Alexandre Julliard 
> Date:   Mon Mar 28 15:47:19 2016 +0900
> makefiles: Reduce the maximum page size for modules that have a base
>
> address.
> It avoid bumping into the next allocated base address on 64-bit.

Thanks, Alexandre and Damjan!  I removed that patch from the
FreeBSD ports (and package) of Wine.

Gerald