[Bug 38178] New: Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Bug ID: 38178
           Summary: Crash with Heroes VI at very begin
           Product: Wine
           Version: 1.7.37
          Hardware: x86
                OS: Mac OS X
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: directx-d3dx9
          Assignee: [e-mail]
          Reporter: [e-mail]

Created attachment 50937
  --> https://bugs.winehq.org/attachment.cgi?id=50937
Heroes VI crashes with any engine > 1.7.7. Worked with 1.6.2

Assertion failed: (iface->lpVtbl == (const IDirect3DBaseTexture9Vtbl
*)&d3d9_texture_2d_vtbl || iface->lpVtbl == (const IDirect3DBaseTexture9Vtbl
*)&d3d9_texture_cube_vtbl || iface->lpVtbl == (const IDirect3DBaseTexture9Vtbl
*)&d3d9_texture_3d_vtbl), function unsafe_impl_from_IDirect3DBaseTexture9, file
texture.c, line 1257.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #1 from Sergey  ---
This is the bug of surface_init

~~~~
surface->IDirect3DSurface9_iface.lpVtbl = &d3d9_surface_vtbl;
...
surface->texture = unsafe_impl_from_IDirect3DBaseTexture9(texture); --> assert
~~~~

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |regression

--- Comment #2 from Austin English  ---
Please run a regression test:
http://wiki.winehq.org/RegressionTesting

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #3 from Sergey  ---
I will search a reason but
1. The bug is not stable. It may be caused by non-initialized variables.
2. There are no good and bad versions because last working version was 1.7.2.
Up to 1.7.37 all versions crashed at startup the game. This log appears in
recent update when I reported immediately.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #4 from Sergey  ---
this assertion may not occurred with revision 1.7.7-1.7.36 because the game
crashes before the procedure.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #5 from Sergey  ---
Log with 1.7.36 attached

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #6 from Sergey  ---
Created attachment 50952
  --> https://bugs.winehq.org/attachment.cgi?id=50952
Log with 1.7.36 very short

err:seh:raise_exception
Unhandled exception code e06d7363 flags 11 addr 0x7b82d0e2

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #7 from Sergey  ---
Now I tested exactly several times.
With engine 1.7.2 it works stable.
With engine 1.7.3 crash at begin. As well as with any later engines 1.7.4,
1.7.7, 1.7.18, 1.7.36, 1.7.38, Stage, Crossover etc.
But wine-1.7.3.diff is very huge file....

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #8 from Sergey  ---
I did the regression test and the results:
----
556e3e0c764ee5eff1eb02b97a01d4141d6ec58f is the first bad commit
commit 556e3e0c764ee5eff1eb02b97a01d4141d6ec58f
Author: Stefan Dösinger 
Date:   Mon Sep 23 13:29:16 2013 +0200

wined3d: Use GL_APPLE_client_storage for volumes if available.

:040000 040000 eb1811e2fa7dbf4906820b55361c409e7a102f5e
595e986ad37405438ecaf76760fc1d1f796724cb M    dlls
----

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #9 from Sergey Isakov  ---
Still crashes with 1.7.39

err:seh:raise_exception Unhandled exception code e06d7363 flags 11 addr
0x7b82d112

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Sebastian Lackner  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail],
                   |                            |[e-mail]
    Regression SHA1|                            |556e3e0c764ee5eff1eb02b97a0
                   |                            |1d4141d6ec58f

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Sergey Isakov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
    Regression SHA1|556e3e0c764ee5eff1eb02b97a0 |
                   |1d4141d6ec58f               |

--- Comment #10 from Sergey Isakov  ---
I did tracing

-----
trace:d3d:resource_cleanup Cleaning up resource 0x21c370.
trace:d3d:device_resource_released device 0x1e8090, resource 0x21c370, type
WINED3D_RTYPE_TEXTURE.
trace:d3d:device_resource_remove device 0x1e8090, resource 0x21c370.
trace:d3d:device_resource_released Resource released.
trace:d3d:device_resource_add device 0x1e8090, resource 0x21c370.
err:seh:raise_exception Unhandled exception code e06d7363 flags 11 addr
0x7b82d112

-----

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Sebastian Lackner  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
    Regression SHA1|                            |556e3e0c764ee5eff1eb02b97a0
                   |                            |1d4141d6ec58f

--- Comment #11 from Sebastian Lackner  ---
Why did you remove the regression SHA1 field? That is the information you
provided in comment #8. If you didn't change your mind please leave this
information in there, it makes it easier to find this regression for other
people.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Stefan Dösinger  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|directx-d3dx9               |directx-d3d

--- Comment #12 from Stefan Dösinger  ---
It seems odd that 556e3e0c764ee5eff1eb02b97a01d4141d6ec58f would cause a wrong
vtable assertion, but it's theoretically possible if there's a random memory
corruption.

Which OSX version are you running and which graphics card do you have?

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #13 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #12)
> It seems odd that 556e3e0c764ee5eff1eb02b97a01d4141d6ec58f would cause a
> wrong vtable assertion, but it's theoretically possible if there's a random
> memory corruption.
>
> Which OSX version are you running and which graphics card do you have?

OSX 10.9.5
Radeon Turks XT (as OpelGL see it) == HD6670 2Gb DDR3

wrong vtable assertion appears when I manually changed codes to enumerate
display.
Original codes lead to error like in comment #9.

I will make more tracing to catch exact place where is the memory leak.
This commit was just first commit with new methods for memory
allocation/release.

I have to confirm again that with any my patches/workaround all engines after
1.7.2 crashes at the game started.
Wine 1.7.2 and all previous work almost fine.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #14 from Stefan Dösinger  ---
The error in comment #9 is very generic. Did this ever work on an unmodified
Wine?

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #15 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #14)
> The error in comment #9 is very generic. Did this ever work on an unmodified
> Wine?

Yes, comment #9 is about unmodified Wine.
As I did regression test I worked with unmodified versions of different
commits.
The game works with unmodified 1.7.2 and doesn't work with 1.7.3 and later.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #16 from Stefan Dösinger  ---
Is the full name of this game "Might & Magic Heroes VI"? I'm currently
downloading a demo, but the download is horribly slow.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #17 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #16)
> Is the full name of this game "Might & Magic Heroes VI"? I'm currently
> downloading a demo, but the download is horribly slow.

Might and Magic Heroes VI - Shades of Darkness v2.1.0
My version from purchased DVD.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #18 from Stefan Dösinger  ---
Created attachment 51099
  --> https://bugs.winehq.org/attachment.cgi?id=51099
wined3d: hack: Don't allow dxt5 textures.

The game tries to use DXT5 textures. We do not support this, and even on
Windows only some drivers have this feature. device_update_volume calculates
the size wrong and the game crashes.

Failing the texture creation call fixes the problem, but unfortunately we
pretend that the call succeeds. The attached hack fixes the crash, but a proper
fix needs more work.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Stefan Dösinger  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #19 from Stefan Dösinger  ---
I can reproduce the bug, confirming it.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #20 from Stefan Dösinger  ---
I forgot to clearly state that the game tries to use dxtn VOLUME textures. 2D
and Cube dxtn textures work just fine in Wine.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #21 from Sergey Isakov  ---
Thanks you for the attention to my problem. I was going to inform about my
tracing but looks like too late
----
trace:d3d_texture:wined3d_texture_init texture 0x21c458, texture_ops
0x43ccef14, layer_count 1, level_count 3, resource_type
WINED3D_RTYPE_VOLUME_TEXTURE, format WINED3DFMT_DXT1, multisample_type 0,
multisample_quality 0, usage 0, pool WINED3D_POOL_SYSTEM_MEM, width 4, height
4, depth 4, surface_flags 0, device 0x1e7fc8, parent 0x21c3b8, parent_ops
0x43b924f8, resource_ops 0x43ccef08.
trace:d3d:resource_init  device 0x1e7fc8, resource 0x21c458
trace:d3d:device_resource_add device 0x1e7fc8, resource 0x21c458.
trace:d3d_surface:wined3d_volume_create container 0x21c458, width 4, height 4,
depth 4, level 0, format WINED3DFMT_DXT1, usage 0x10000000, pool
WINED3D_POOL_SYSTEM_MEM, volume 0x3eff06c.

----
You are quite right. This is the problem of DXT1
I applied your hack and... the program crashed
----
trace:d3d_surface:wined3d_volume_create container 0x6f30b48, width 128, height
128, depth 16, level 0, format WINED3DFMT_B8G8R8A8_UNORM, usage 0x10000000,
pool WINED3D_POOL_SYSTEM_MEM, volume 0x3eff06c.
err:seh:raise_exception Unhandled exception code e06d7363 flags 11 addr
0x7b82d112
----
This format also should be excluded?

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #22 from Stefan Dösinger  ---
No, WINED3DFMT_B8G8R8A8_UNORM is as standard as it can be. If you blacklist
this the game will probably think you're trying to run it on a GPU from 1995.

Do you have any other patches applied to your Wine tree? With attachment 51099
the  demo works without any additional patches for me (Early 2013 macbook pro
retina, OSX 10.9, Nvidia GPU)

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #23 from Sergey Isakov  ---
I have AMD GPU specific patches and some initialization of variables.
OK, I will do
git stash
git pull
and try again.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #24 from Stefan Dösinger  ---
You may have to set the video memory size registry key to e.g. 1024. The game
doesn't use d3d9ex (so it is subject to video memory accounting) and is quite
video memory heavy. If your GPU isn't detected properly it will crash in a way
that looks very similar to the update_volume crash.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #25 from Sergey Isakov  ---
OK,
now it works.
Revision 5e5adcf + your hack.

I hope some day there will be more correct support for the DXT1 but I can
play... no, I have another bug that I should report in another place:
 - black textures
 - white water.

Thank you for this hack.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #26 from Sergey Isakov  ---
Meanwhile the Wine shouldn't crash at unsupported feature. The hack must be
implemented into trunk until DXTn support will come.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #27 from Sergey Isakov  ---
Still crashes if without this hack
----
fixme:d3d:device_update_volume Source and destination formats do not match.
err:seh:raise_exception Unhandled exception code e06d7363 flags 11 addr
0x7b827d0b
----
wine-1.7.39-279-g56ab12c

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #28 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #18)
> Created attachment 51099 [details]
> wined3d: hack: Don't allow dxt5 textures.
>
> The game tries to use DXT5 textures. We do not support this, and even on
> Windows only some drivers have this feature. device_update_volume calculates
> the size wrong and the game crashes.
>
> Failing the texture creation call fixes the problem, but unfortunately we
> pretend that the call succeeds. The attached hack fixes the crash, but a
> proper fix needs more work.

Can you please show me a place where this crash occurred?

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #29 from Stefan Dösinger  ---
If I remember correctly the crash occurred outside of Wine's code because the
update_volume call corrupts memory. A backtrace should give some clues though.

I have patches that fix this properly, by keeping track of format capabilities
for each GL resource type separately. I will send them in the next days. Right
now I'm busy with some more regressions caused by my focus loss handling
patches.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #30 from Sergey Isakov  ---
I am just testing experimental DXTn support hoping it may resolve some of my
issues.
But no.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #31 from Stefan Dösinger  ---
"experimental DXTn support"?

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #32 from Sergey Isakov  ---
The patch from wine-stage

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #33 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #20)
> I forgot to clearly state that the game tries to use dxtn VOLUME textures.
> 2D and Cube dxtn textures work just fine in Wine.

Comment from MSDN

The runtime will not allow an application to create a surface using a DXTn
format unless the surface dimensions are multiples of 4. This applies to
offscreen-plain surfaces, render targets, 2D textures, cube textures, and
volume textures.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #34 from Sergey Isakov  ---
Created attachment 51273
  --> https://bugs.winehq.org/attachment.cgi?id=51273
An MS utility from DirectXSDK to reproduce the bug

This is utility from free distributed DirectxSDK that able to create volume
textures of format DXT5.
This utility requires mfc100.dll and works under wine.

Other bug with this utility - can't save the file with the texture.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #35 from Austin English  ---
The content of attachment 51273 has been deleted for the following reason:

copyrighted binary

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #36 from Sergey Isakov  ---
(In reply to Austin English from comment #35)
> The content of attachment 51273 [details] has been deleted for the following
> reason:
>
> copyrighted binary

Sorry!
There is an original URL
http://www.microsoft.com/en-us/download/details.aspx?id=6812

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #37 from Sergey Isakov  ---
Wine-1.7.41 unmodified
Still the same crash
----
fixme:d3d:device_update_volume Source and destination formats do not match.
err:seh:raise_exception Unhandled exception code e06d7363 flags 11 addr
0x7b82b078
---

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #38 from Sergey Isakov  ---
(In reply to Stefan Dösinger from comment #31)
> "experimental DXTn support"?

see attachment

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #39 from Sergey Isakov  ---
Created attachment 51336
  --> https://bugs.winehq.org/attachment.cgi?id=51336
texture created by DxTex.exe under wine with DXTn support

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #40 from Stefan Dösinger  ---
The patches I sent on Friday should fix this bug, but the last patch
(http://source.winehq.org/patches/data/110915) caused test failures. I'll
resend a fixed version on Monday.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Stefan Dösinger  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
      Fixed by SHA1|                            |7109bebd024ee68ec2b3ef0a6b2
                   |                            |9e72ded08abc0
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #41 from Stefan Dösinger  ---
This bug should be fixed by 7109bebd024ee68ec2b3ef0a6b29e72ded08abc0. Please
retest with Wine git from today.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #42 from Sergey Isakov  ---
Tested with wine-1.7.41-117-gd15ca4e
OK, it doesn't crash.
But it is the same hack, it just disables using DXTn volumes instead of support
them.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #43 from Stefan Dösinger  ---
Yeah, good luck implementing them without an equivalent GL functionality.
There's GL_NV_texture_compression_vtc, but it has an incompatible block layout
(4x4x4 blocks instead of 4x4x1). This functionality is optional on Windows, so
games have to deal with it not being available (my older nvidia GPUs don't have
it, not sure about newer ones. Radeon GPUs have it). Not advertising it is
better than trying to emulate it somehow and either unpack the data to
uncompressed RGBA8 or do expensive repacking to make it match
GL_NV_texture_compression_vtc.

It may become an issue for d3d10 though.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #44 from Sergey Isakov  ---
But I think it related to bug 38274.
Yes, GL functionality differ from DX so it may be expensive solution. :(

I want to understand why "device_update_volume calculates the size wrong and
the game crashes".

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #45 from Stefan Dösinger  ---
I'll reply about those concerns on the other bug report. Yes, it is possible
that this game really needs compressed volumes.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #46 from Sergey Isakov  ---
Please, don't close this bug because the game crashes even if full dxtn support
implemented.
There is another bug with volume textures.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #47 from Stefan Dösinger  ---
I think closing this is the right thing, and tracking the lack of support in
the other bug.

Wrt update_volume crashing: The size calculation is different for compressed
textures, in particular for mipmaps that are smaller than one block.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #48 from Sergey Isakov  ---
Should I report this as a separate bug?

fixme:d3d_texture:volumetexture_init create dxtn texture 0x31545844.
fixme:d3dx:D3DXLoadSurfaceFromMemory decode dxtn from pitch=512 to 1024
fixme:d3dx:D3DXLoadSurfaceFromMemory Unhandled filter 0x4.
err:d3d:context_bind_fbo >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION (0x502) from
glBindFramebuffer() @ context.c / 72

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #49 from Alexandre Julliard  ---
Closing bugs fixed in 1.7.42.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #50 from Stefan Dösinger  ---
(In reply to Sergey Isakov from comment #48)
> Should I report this as a separate bug?
>
> fixme:d3d_texture:volumetexture_init create dxtn texture 0x31545844.
> fixme:d3dx:D3DXLoadSurfaceFromMemory decode dxtn from pitch=512 to 1024
> fixme:d3dx:D3DXLoadSurfaceFromMemory Unhandled filter 0x4.

I think that's a bug in builtin d3dx9. It should use an ARGB8 texture in
response to volume dxtn failing. Try native d3dx9 and see if it helps.
> err:d3d:context_bind_fbo >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION (0x502) from
> glBindFramebuffer() @ context.c / 72

This looks unrelated.

[Bug 38178] Crash with Heroes VI at very begin

https://bugs.winehq.org/show_bug.cgi?id=38178

--- Comment #51 from Sergey Isakov  ---
I understand the crash was occurred with texture release.
The game tries to use compressed volume textures with mipmaps. Releasing them
causes crash same as in bug 34480.