[Bug 38627] New: Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

Bug ID: 38627
           Summary: Holy Avatar vs. Maidens of the Dead hangs on exit
           Product: Wine
           Version: 1.5.31
          Hardware: x86
                OS: Linux
            Status: NEW
          Keywords: regression
          Severity: normal
          Priority: P2
         Component: kernel32
          Assignee: [e-mail]
          Reporter: [e-mail]
                CC: [e-mail]
   Regression SHA1: bfcdf6df8eea1230fb0e1835a809e49e5971e5d2
      Distribution: ---

Created attachment 51534
  --> https://bugs.winehq.org/attachment.cgi?id=51534
terminal output

I let the Desura client install the game to avoid bug #38614.

The game starts properly without any winetricks, but it hangs on exit. After
selecting quit from the menu, the screen is frozen and the game executable
constantly uses 100% CPU and it needs to be killed.

The problem is present since

bfcdf6df8eea1230fb0e1835a809e49e5971e5d2 is the first bad commit
commit bfcdf6df8eea1230fb0e1835a809e49e5971e5d2
Author: Dmitry Timoshkov 
Date:   Mon May 13 16:46:37 2013 +0900

kernel32: Terminate threads before unloading DLLs on process exit.

No demo version exists.
Let me know if you need debug logs or further testing.

Tested in wine-1.7.43-123-g90ed96a

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #1 from Dmitry Timoshkov  ---
Does installing native msvcrt/msvcp help? If not, please generate
+relay,+seh,+tid log (without wine startup by generating log from
a separate terminal and in append '>>' mode).

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #2 from Béla Gyebrószki  ---
Created attachment 51535
  --> https://bugs.winehq.org/attachment.cgi?id=51535
+relay,+seh,+tid partial log (uncompressed 98 MB)

(In reply to Dmitry Timoshkov from comment #1)
> Does installing native msvcrt/msvcp help? If not, please generate
> +relay,+seh,+tid log (without wine startup by generating log from
> a separate terminal and in append '>>' mode).

The game uses msvcp90,msvcr90 but installing native ones doesn't help. Native
msvcrt.dll doesn't help either.
The full +relay log is nearly 3 GB, the attached file contains the last 1.5
million lines of that log. Is that of any help to you?

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #3 from Béla Gyebrószki  ---
The vast majority of the full +relay log (2.9 GB) consists of the same
msvcp90/msvcr90 calls so I could squeeze it into a 30 MB .7z archive:
https://drive.google.com/open?id=0B-tTbLKBl-tONjJ5RWo1ZmlCYzQ

What if I create a +relay with native msvcp90/msvcr90 installed? The problem is
present with native too, and that would reduce the log size to a humble 200 MB?
Or would that make debugging more difficult?

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #4 from Dmitry Timoshkov  ---
(In reply to Béla Gyebrószki from comment #2)
> Created attachment 51535 [details]
> +relay,+seh,+tid partial log (uncompressed 98 MB)

Thanks. There is an access violation exception in Vision90.dll when it handles
PROCESS_DETACH event. An exception occurs right after the following calls:
wined3d.wined3d_buffer_decref()=0
wined3d.wined3d_mutex_unlock()=0
These calls should go from inside of Wine ddraw*/d3d* dlls, but there is no
appropriate relay calls for them in the log, so it's impossible to say what
is going on exactly. I'd guess that in order to investigate further one needs
to see what's wrong with ddraw*/d3d*, as a wild guess may be one of these
dlls invokes a callback on destruction which application doesn't expect at
this point, but that's a pure speculation.

(In reply to Béla Gyebrószki from comment #3)
> The vast majority of the full +relay log (2.9 GB) consists of the same
> msvcp90/msvcr90 calls so I could squeeze it into a 30 MB .7z archive:
> https://drive.google.com/open?id=0B-tTbLKBl-tONjJ5RWo1ZmlCYzQ
>
> What if I create a +relay with native msvcp90/msvcr90 installed? The problem
> is present with native too, and that would reduce the log size to a humble
> 200 MB?
> Or would that make debugging more difficult?

It would be more interesting to force ddraw*/d3d* calls appear in the log,
but I can't off the top of my head suggest how to do that, probably playing
with RelayExclude/RelayInclude keys in the registry would make the trick.
Although reducing the log using native msvc* dlls should still keep
interesting parts in the log.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #5 from Dmitry Timoshkov  ---
Another thing to investigate is a hang before Ctrl+C exception:

probably vBase90.dll:
thread_proc()
{
    for(;;) WaitForSingleObject(hevent, INFINITE);
}

hevent = CreateEvent(0,0,0,0);
CreateThread(thread_proc);

vBase90.dll,PROCESS_DETACH:
SetEvent(hevent); <= probably hangs here, there is no return before Ctrl+C

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #6 from Béla Gyebrószki  ---
Created attachment 51544
  --> https://bugs.winehq.org/attachment.cgi?id=51544
winedbg bt all

Not sure how I could help you resolve this bug..
I attached winedbg to the already hung process (only 1 thread was running) and
issued 'bt all'.
Looks like it hangs in vBase90.dll indeed.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #7 from Dmitry Timoshkov  ---
Is it possible to generate a log without the hang? (with either the offending
patch reverted in current git or before the commit point).

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #8 from Béla Gyebrószki  ---
Created attachment 51548
  --> https://bugs.winehq.org/attachment.cgi?id=51548
+relay,+seh,+tid (wine-1.5.30-128-g7ad43e4)

The last 1 million lines of the +relay log, created in wine-1.5.30-128-g7ad43e4
(one commit before the regression). The game exits cleanly.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #9 from Dmitry Timoshkov  ---
(In reply to Dmitry Timoshkov from comment #5)
> Another thing to investigate is a hang before Ctrl+C exception:
>
> probably vBase90.dll:
> thread_proc()
> {
>     for(;;) WaitForSingleObject(hevent, INFINITE);
> }
>
> hevent = CreateEvent(0,0,0,0);
> CreateThread(thread_proc);
>
> vBase90.dll,PROCESS_DETACH:
> SetEvent(hevent); <= probably hangs here, there is no return before Ctrl+C

After looking at the vBase90.dll disassembly what application actually does
is slightly different:

vBase90.dll:
int var_to_test;
thread_proc()
{
    for(;;) WaitForSingleObject(hevent, INFINITE);
    var_to_test = 1;
}

var_to_test = 0;
hevent = CreateEvent(0,0,0,0);
CreateThread(thread_proc);
...
ExitProcess(0);
...
vBase90.dll,PROCESS_DETACH:
SetEvent(hevent);
while (var_to_test != 1) /* wait*/; <= hangs here

The applications expects that variable var_to_test is set by the thread
proc after the event is signalled, but all the threads are already dead
at the point of PROCESS_DETACH event, so the app hangs forever waiting for
the variable.

So far I failed to reproduce this application's expectation with a test
under Windows7, the thread gets killed before the PROCESS_DETACH event.
My guess was that the event could get an abandoned state during thread
termination by ExitProcess, so that WaitForSingleObejct() would return
and allow setting the variable, but so far I couldn't confirm that theory.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #10 from Dmitry Timoshkov  ---
(In reply to Dmitry Timoshkov from comment #9)
> vBase90.dll:
> int var_to_test;
> thread_proc()
> {
>     for(;;) WaitForSingleObject(hevent, INFINITE);
>     var_to_test = 1;
> }

Of course the line
for(;;) WaitForSingleObject(hevent, INFINITE);
should be
WaitForSingleObject(hevent, INFINITE);

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #11 from Dmitry Timoshkov  ---
(In reply to Dmitry Timoshkov from comment #9)
> The applications expects that variable var_to_test is set by the thread
> proc after the event is signalled, but all the threads are already dead
> at the point of PROCESS_DETACH event, so the app hangs forever waiting for
> the variable.
>
> So far I failed to reproduce this application's expectation with a test
> under Windows7, the thread gets killed before the PROCESS_DETACH event.
> My guess was that the event could get an abandoned state during thread
> termination by ExitProcess, so that WaitForSingleObejct() would return
> and allow setting the variable, but so far I couldn't confirm that theory.

A probable explanation could be that variable var_to_test might be set from
some other place besides the thread_proc, and it's another bug revealed by
the thread termination on process exit.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #12 from Dmitry Timoshkov  ---
Béla, could you please test the game under Windows? Does it exit cleanly there?

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #13 from Béla Gyebrószki  ---
(In reply to Dmitry Timoshkov from comment #12)
> Béla, could you please test the game under Windows? Does it exit cleanly
> there?

I set up a Virtualbox VM running Windows XP SP3 in it because I don't have
Windows installed, but the game failed to start even when 3D was enabled in the
VM. The game complained about unsupported render target format and exited.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

Anastasius Focht  changed:

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

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

--- Comment #14 from Béla Gyebrószki  ---
Still present in 1.9.22.

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

joaopa  changed:

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

--- Comment #15 from joaopa  ---
Is still a bug in current wine(3.20)?

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

Béla Gyebrószki  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
           Keywords|regression                  |
          Component|kernel32                    |-unknown
             Status|NEW                         |RESOLVED
    Regression SHA1|bfcdf6df8eea1230fb0e1835a80 |
                   |9e49e5971e5d2               |

--- Comment #16 from Béla Gyebrószki  ---
This bug was reported against the Desura version of the game. The title screen
shows v.0.0.8.1 builddate Sep 7 2012 as the game version. Desura digital
distribution platform went offline years ago and apparently they've never
released a patch for this game.
Now I tried the Steam version (v1.0.0.0 builddate Feb 19 2014) and it doesn't
have the reported problem, the game quits properly.

Marking invalid (possibly a bug in the early version of the game).

[Bug 38627] Holy Avatar vs. Maidens of the Dead hangs on exit

https://bugs.winehq.org/show_bug.cgi?id=38627

Béla Gyebrószki  changed:

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

--- Comment #17 from Béla Gyebrószki  ---
Closing invalid bug.