[Bug 43748] New: Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Bug ID: 43748
           Summary: Dead or Alive Last Round ntdll crash on Wine newer
                    than 1.8.7 under Linux, but not in OS X
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ntdll
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

Created attachment 59244
  --> https://bugs.winehq.org/attachment.cgi?id=59244
staging-2.17 console/debug output

I got Dead or Alive Last Round (free on Steam) working without issues under OS
X, with multiple Wine versions tested: 2.0.2, 2.13, 2.16-staging. However,
under Linux, no version of Wine newer than 1.8.7 even proceeds beyond a brief
white window.

I attached a log with the default debug settings, please let me know what
further to enable as I'm not sure what the issue is.

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #1 from Tod Jackson  ---
A notable difference from the working Wine version appears that Wine didn't
recognize my GPU yet (an Intel HD 4000):err:d3d:init_driver_info No driver
version info found for device 8086:0166, driver model 0x2. I gathered that the
next stable release added more device info.

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |regression
            Version|unspecified                 |2.0.2

--- Comment #2 from Austin English  ---
Could you run a regression test?
https://wiki.winehq.org/Regression_Testing

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #3 from Tod Jackson  ---
Yeah, working on it. :) The game works on Wine 1.9.1 (commit
91dd155e9bc13c1bc1be860cd67fe3e48c20feae in master) as well.

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Tod Jackson  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
    Regression SHA1|                            |b1429f931e816feee2f7b7fe065
                   |                            |a393c4caa7960
          Component|ntdll                       |directx-d3d

--- Comment #4 from Tod Jackson  ---
The offending commit is:

commit b1429f931e816feee2f7b7fe065a393c4caa7960
Author: Matteo Bruni 
Date:   Thu Jan 28 00:15:43 2016 +0100

wined3d: Add a real implementation of
wined3d_check_device_multisample_type().

Signed-off-by: Matteo Bruni 
    Signed-off-by: Henri Verbeet 
    Signed-off-by: Alexandre Julliard

winetricks multisampling=enabled or multisampling=disabled doesn't help.

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #5 from Tod Jackson  ---
Mesa bug? MESA_EXTENSION_MAX_YEAR=2010 allows the game to start on
Wine-staging. Hooray!

[Bug 43748] Dead or Alive Last Round ntdll crash on Wine newer than 1.8.7 under Linux , but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #6 from Tod Jackson  ---
Well, after further reading I think this is a duplicate of bug 25362 (OpenGL
extension string truncation/overflow), and the 'offending' commit added more
available extensions. Also, it works on OS X because their OpenGL version has
fewer extensions.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Matteo Bruni  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Dead or Alive Last Round    |Dead or Alive Last Round
                   |ntdll crash on Wine newer   |crashes on Wine newer than
                   |than 1.8.7 under Linux, but |1.8.7 under Linux, but not
                   |not in OS X                 |in OS X

--- Comment #7 from Matteo Bruni  ---
(In reply to Tod Jackson from comment #6)
> Well, after further reading I think this is a duplicate of bug 25362 (OpenGL
> extension string truncation/overflow), and the 'offending' commit added more
> available extensions. Also, it works on OS X because their OpenGL version
> has fewer extensions.

No, that only makes sense for some old OpenGL games while this is a d3d9 game.

I can't reproduce the issue here, can you please attach a
+tid,d3d9,d3d,d3d_shader log?

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #8 from Tod Jackson  ---
Created attachment 59260
  --> https://bugs.winehq.org/attachment.cgi?id=59260
+tid,d3d9,d3d,d3d_shader failing

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Matteo Bruni  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
            Version|2.0.2                       |1.9.3

--- Comment #9 from Matteo Bruni  ---
(In reply to Tod Jackson from comment #8)
> Created attachment 59260 [details]
> +tid,d3d9,d3d,d3d_shader failing

Thanks, that helped. It looks like the game checks for 2x multisampling and
then proceeds to crash on a NULL dereference if it's not supported. I managed
to reproduce your crash by hacking wined3d_check_device_multisample_type() to
return WINED3DERR_NOTAVAILABLE for 2x multisampling.

I don't know if it's true that your GPU doesn't support 2x multisampling but I
seem to recall some kind of limitation in Mesa Intel drivers with specifically
2x multisampling. I have no idea if that changes in any way with a newer kernel
/ Mesa.

I'd say this is an application bug but we might want to add some workaround if
it turns out that there are more applications "surprised" by this specific
behavior.

BTW, there was this in the log:
> 0009:trace:d3d:init_driver_info GPU override ffff:ffff.

Do you really need that?

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #10 from Tod Jackson  ---
When I saw it was a multisampling issue this came to mind, something I've
always seen in chrome://gpu: Multisampling is buggy in Intel IvyBridge: 116370
Applied Workarounds: disable_chromium_framebuffer_multisample
http://crbug.com/116370

Sorry, what's this GPU override mean? I probably didn't do it on purpose.

Also, any clue why the game works with MESA_EXTENSION_MAX_YEAR=2010?

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #11 from Matteo Bruni  ---
(In reply to Tod Jackson from comment #10)
> When I saw it was a multisampling issue this came to mind, something I've
> always seen in chrome://gpu: Multisampling is buggy in Intel IvyBridge:
> 116370
> Applied Workarounds: disable_chromium_framebuffer_multisample
> http://crbug.com/116370

Unclear but probably not related.
> Sorry, what's this GPU override mean? I probably didn't do it on purpose.

I looked that up again in the older Wine sources and it looks like, contrarily
to what I thought previously, that trace was always printed at the time.
Nothing to see here, sorry for the noise.
> Also, any clue why the game works with MESA_EXTENSION_MAX_YEAR=2010?

Right, I forgot about that part. The patch causing this regression essentially
started asking the driver for detailed per-format, per-sample-count
multisampling support. For that it uses the ARB_internalformat_query extension
which is marked in the Mesa sources with the 2011 year. So by setting the
environment variable you're practically disabling the new "detailed" codepath
and falling back to the previous behavior, which is to advertise all the
multisample levels up to MAX_SAMPLES. If the game later on wants to create a
texture with sample count 2, which is apparently unsupported by your GPU, that
is automatically "upgraded" to the lowest supported count (probably 4) and
everything SHOULD keep working.

So good catch and probably a good workaround for this bug.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #12 from Henri Verbeet  ---
So, NOTOURBUG? Or is there something we should do about this?

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #13 from Matteo Bruni  ---
(In reply to Henri Verbeet from comment #12)
> So, NOTOURBUG? Or is there something we should do about this?

I'm not sure. AFAIU it's a specific combination of:

- GPU driver advertising 4x+ multisampling but not 2x multisampling
- game expecting 2x multisampling once 4x is supported

I guess closing this bug (either NOTOURBUG or INVALID, depending on who you
fault more between the driver and the game :P) is fine. It seems like we
haven't seen other instances of this bug yet so it's probably not worthwhile to
do anything on our part at this point.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Józef Kucia  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #14 from Józef Kucia  ---
(In reply to Matteo Bruni from comment #13)
> I'm not sure. AFAIU it's a specific combination of:
> - GPU driver advertising 4x+ multisampling but not 2x multisampling
> - game expecting 2x multisampling once 4x is supported

If Windows drivers support 2x multisampling once 4x is supported, then one
could argue that it's our bug.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #15 from Henri Verbeet  ---
Yeah, I guess. I'd be curious what the Windows driver for that hardware does.
I'm inclined to mostly blame this on the application, but wouldn't necessarily
be opposed to some kind of quirk for 2xMSAA. At the same time, such a quirk
would somewhat defeat the point of having a per sample count query, as opposed
to just a maximum like classic GL.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #16 from Tod Jackson  ---
I'm on Windows at the moment. Under Steam, going to Help > System Information
indicates that 2x MSAA is supported:

Primary Bus Type Not Detected
    Primary VRAM: -1984 MB
    Supported MSAA Modes:  2x 4x 8x

Under Wine, 2x is not listed.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #17 from Tod Jackson  ---
I should clarify, with Mesa specifically, Steam doesn't show 2x.

https://cgit.freedesktop.org/mesa/mesa/tree/src/mesa/drivers/dri/i965/intel_...
The Intel HD 4000 is Ivybridge (generation 3), but I don't know what genX_modes
that actually corresponds to above. It seems clear that Mesa doesn't support 2x
for this hardware though.

"A speciality of the Ivy Bridge GPUs is that 4x MSAA is supported in hardware
now. However, 2x is only supported through software.  The algorithm to support
2x is going through the 4x pipeline with a software algorithm, so performance
is similar to 4x MSAA."
https://www.notebookcheck.net/Intel-HD-Graphics-4000.69168.0.html

Whatever you guys decide to do, I indicated the MESA_EXTENSION_MAX_YEAR=2010
workaround on the AppDB and have never personally seen this issue anywhere else
under Wine.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #18 from Henri Verbeet  ---
(In reply to Tod Jackson from comment #17)
> Whatever you guys decide to do, I indicated the MESA_EXTENSION_MAX_YEAR=2010
> workaround on the AppDB and have never personally seen this issue anywhere
> else under Wine.

For what it's worth, I think
MESA_EXTENSION_OVERRIDE=-GL_ARB_internalformat_query should also work as a
workaround, and is a little more targeted.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

--- Comment #19 from Tod Jackson  ---
(In reply to Henri Verbeet from comment #18)
> (In reply to Tod Jackson from comment #17)
> > Whatever you guys decide to do, I indicated the MESA_EXTENSION_MAX_YEAR=2010
> > workaround on the AppDB and have never personally seen this issue anywhere
> > else under Wine.
>
> For what it's worth, I think
> MESA_EXTENSION_OVERRIDE=-GL_ARB_internalformat_query should also work as a
> workaround, and is a little more targeted.

Gotcha, thanks.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

Henri Verbeet  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |INVALID

--- Comment #20 from Henri Verbeet  ---
I'm going to blame this on the application. It wouldn't be hard to put in a
workaround, but I don't think that would add much over the
MESA_EXTENSION_OVERRIDE workaround. We can certainly revisit that if this issue
turns out to be more widespread.

[Bug 43748] Dead or Alive Last Round crashes on Wine newer than 1.8.7 under Linux, but not in OS X

https://bugs.winehq.org/show_bug.cgi?id=43748

André H.  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED
                 CC|                            |[e-mail]

--- Comment #21 from André H.  ---
closing invalid