[Bug 33331] New: When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

Bug #: 33331
           Summary: When DLL receives PROCESS_DETACH notification when a
                    process is exiting all the threads should be already
                    terminated
           Product: Wine
           Version: 1.5.27
          Platform: x86-64
        OS/Version: Linux
            Status: NEW
          Keywords: download, source, testcase
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified

Created attachment 44083
  --> http://bugs.winehq.org/attachment.cgi?id=44083
source and binary for the test

While debugging the problem described in
http://www.winehq.org/pipermail/wine-patches/2013-March/123127.html
(libiomp5md.dll waits until all the tracked threads terminate and doesn't
return control out of its PROCESS_DETACH handler when process is going to exit)
I decided to create a test which examines what's supposed to happen when
static (loaded implicitly as a part of PE imports) and dynamic (loaded
explicitly by LoadLibrary) DLLs create threads along with threads created
by main exe.

To make long story short: main difference between Windows and Wine is that Wine
doesn't terminate threads before calling PROCESS_DETACH for each loaded DLL.

Here are the snippets from results I see under Windows 7 (both 32 and 64 bit
builds behave same way) and Wine:

Win7:
...
main: call ExitProcess(0)
dynamic: 72FB0000, DLL_PROCESS_DETACH, 00000001
dynamic: GetExitCodeThread(0) => 1,0
dynamic: GetExitCodeThread(1) => 1,0
dynamic: GetExitCodeThread(2) => 1,0
static: 72FC0000, DLL_PROCESS_DETACH, 00000001
static: GetExitCodeThread(0) => 1,0
static: GetExitCodeThread(1) => 1,0
static: GetExitCodeThread(2) => 1,0
=====================================================
Wine:
...
main: call ExitProcess(0)
dynamic: 00330000, DLL_PROCESS_DETACH, 00000001
dynamic: GetExitCodeThread(0) => 1,259
dynamic: GetExitCodeThread(1) => 1,259
dynamic: GetExitCodeThread(2) => 1,259
static: 10000000, DLL_PROCESS_DETACH, 00000001
static: GetExitCodeThread(0) => 1,259
static: GetExitCodeThread(1) => 1,259
static: GetExitCodeThread(2) => 1,259

I.e. libiomp5md.dll when it receives PROCESS_DETACH notification at process
exit time shoudn't get STILL_ACTIVE (259) return value from GetExitCodeThread.

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

--- Comment #1 from Dmitry Timoshkov  2013-04-04 03:42:04 CDT ---
Created attachment 44084
  --> http://bugs.winehq.org/attachment.cgi?id=44084
Windows 7 output

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

--- Comment #2 from Dmitry Timoshkov  2013-04-04 03:43:02 CDT ---
Created attachment 44085
  --> http://bugs.winehq.org/attachment.cgi?id=44085
Wine output

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

--- Comment #3 from Rico  2013-04-04 04:01:54 CDT ---
A somewhat similar problem is bug 32951 .

Thus it seems, that all threads beside the main tread should be terminated
first, regardless where the thread originates from. Then all resources could be
unloaded.

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

--- Comment #4 from Dmitry Timoshkov  2013-04-04 10:10:48 CDT ---
(In reply to comment #3)
> A somewhat similar problem is bug 32951 .
>
> Thus it seems, that all threads beside the main tread should be terminated
> first, regardless where the thread originates from. Then all resources could be
> unloaded.

When a process terminates freeing resources doesn't make sense and is pure
waste of time since the process address space is going to be destroyed.

What libiomp5md.dll does and what is the source of this bug report doesn't
make sense either, but since it works under Windows and real world application
depends on this then Wine needs to be fixed.

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

Austin English  changed:

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

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
      Fixed by SHA1|                            |bfcdf6df8eea1230fb0e1835a80
                   |                            |9e49e5971e5d2
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #5 from Dmitry Timoshkov  2013-05-17 20:28:00 CDT ---
Should be fixed now.

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

http://bugs.winehq.org/show_bug.cgi?id=33331

Alexandre Julliard  changed:

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

--- Comment #6 from Alexandre Julliard  2013-05-24 13:33:25 CDT ---
Closing bugs fixed in 1.5.31.

[Bug 33331] When DLL receives PROCESS_DETACH notification when a process is exiting all the threads should be already terminated

https://bugs.winehq.org/show_bug.cgi?id=33331

Anastasius Focht  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
          Component|-unknown                    |kernel32