[Bug 31883] New: regression/dogfood: PCSX2 locks up when/shortly after initializing MTGS

http://bugs.winehq.org/show_bug.cgi?id=31883

Bug #: 31883
           Summary: regression/dogfood: PCSX2 locks up when/shortly after
                    initializing MTGS
           Product: Wine
           Version: 1.5.14
          Platform: x86-64
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: winex11.drv
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified

This has been happening since mid/early August when work began to remove the
"Big X Lock."

Since then, attempting to initialise the GS has caused either one of two
results
1. The MTGS fails to initialise
2. it initialises and completely locks up after a few seconds of continually
degrading performance. Console output identifies the context_section critical
section as the point of failure

I have attached 3 logs, their names indicate the WIENDEBUG value used.
relay_wgl.log - failed to initialise (even after multiple attempts)
relay.log - failed to initialise
wgl.log - initialised and locked up shortly after

The first problematic build I have is dated August 14th, things get
increasingly worse going forwards.
The last problem free build is dated August 2nd

As for how things got worse (and isn't really relevant to the bug report),
since 8/14, additional work kept being done.
Initially I could scroll or move another wine window to fudge the locking a
bit, this would either resolve it temporarily to at least give me a fighting
chance to restart the renderer and play. It could still occasionally initialise
it properly at this stage.
Going forward, I could not longer used this trick and it started locking up
constantly and reliably, to the point that I could no longer get a lucky break
to run it for even a moment.
Also worth noting that in some of the worst cases, the window created by the GS
looks like it's initial update to a full black window is being done in 100x100
tiles over a few seconds.

[Bug 31883] regression/dogfood: PCSX2 locks up when/shortly after initializing MTGS

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #1 from Kevin Meyer  2012-10-04 08:02:10 CDT ---
Bluh... Attachment upload failed...
I've put it on my dropbox, it will take a little while to upload though.

https://dl.dropbox.com/u/23891252/logs.tar.bz2

[Bug 31883] regression/dogfood: PCSX2 locks up when/shortly after initializing MTGS

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #2 from Rico  2012-10-04 08:26:32 CDT ---
Please run a regression test (see http://wiki.winehq.org/RegressionTesting ).
and attach a backtrace of the hanging threads (See
http://wiki.winehq.org/Backtraces).

[Bug 31883] regression/dogfood: PCSX2 locks up when/shortly after initializing MTGS

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #3 from Kevin Meyer  2012-10-04 10:17:19 CDT ---
Created attachment 41972
  --> http://bugs.winehq.org/attachment.cgi?id=41972
backtrace taken ASAHP after locking up

Using the latest revision I can I have tried to get you the backtrace and it
seems it fell afoul of the "surfaces_section" in user32 this time. It's a
complete first as far as I'm concerned.

then I tried again and it got stuck in the "surface" critical section - it kind
of just hanged when I told it to backtrace. (this is also a first)

third time's the charm, I got a backtrace for most of the running threads via
bt all, but debug information was rather absent...

a bunch of long compile times later, here's a possibly incomplete backtrace...
I attached the debugger as soon as it froze, before it had a chance to say it
timed out.

I'll get onto the bisection stuff in the morning.

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |regression
            Summary|regression/dogfood: PCSX2   |PCSX2 locks up when/shortly
                   |locks up when/shortly after |after initializing MTGS
                   |initializing MTGS           |(dogfood)

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #4 from Rico  2012-10-04 15:37:43 CDT ---
Could you check, in one run if the blocking thread from the error:
"err:ntdll:RtlpWaitForCriticalSection section 0x7e265040
"../../../wine-git/dlls/winex11.drv/opengl.c: context_section" wait timed out
in thread 0009, blocked by 002d, retrying (60 sec)"
is equal the thread which called xcb_wait_for_reply:
"4 0x7dfb27b7 xcb_wait_for_reply+0x66() in libxcb.so.1 (0x7def7374)"

Though, make sure you use the values from the same run. If they are equal,
please have a look at bug 31406. Does current git from libxcb help for you?
(build instructions are here:
http://forum.winehq.org/viewtopic.php?f=2&t=17203, but you don't need the debug
patch anymore, so skip the line with "git apply ../<name>.patch").

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #5 from Kevin Meyer  2012-10-04 21:35:35 CDT ---
After reading the bug report, my initial response is: My distro already ships
1.8.1

Regardless, I have cloned xcb-git and xcb-proto-git and installed them.
(someone should make a note somewhere that it doesn't build with python
3.x...).
Amusingly there's a commit dated 25 sept which seems quite relevant..

Observations...
It now seems to initialise reliably...
It still freezes after several seconds, but the previously mentioned scrolling
trick lets it continue for a little while again. If I don't do that, it will
still print the context_section time out.

Also of note, with the freezes, I can now "pause" the VM thanks to the
scrolling trick. Resuming it at this point seems to set it into a functional
and playable state, which is almost what it was like back in 1.5.10.

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #6 from Kevin Meyer  2012-10-04 22:13:04 CDT ---
Oh bluh... I just realised I only built the 64bit version of the library.
I've now built the 32bit version as well...

The program now initialises to 7-8 FPS, I've been unable to make it
freeze/lock-up after multiple attempts.
After some pause/resume invocations, it runs at the intended 60 FPS.

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |DUPLICATE

--- Comment #7 from Alexandre Julliard  2012-10-10 11:20:14 CDT ---
Duplicate then.

*** This bug has been marked as a duplicate of bug 31406 ***

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

t800  changed:

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

--- Comment #8 from t800  2012-10-10 11:38:15 CDT ---
(In reply to comment #5)
> After reading the bug report, my initial response is: My distro already ships
> 1.8.1
>
> Regardless, I have cloned xcb-git and xcb-proto-git and installed them.
> (someone should make a note somewhere that it doesn't build with python
> 3.x...).
> Amusingly there's a commit dated 25 sept which seems quite relevant..
>
> Observations...
> It now seems to initialise reliably...
> It still freezes after several seconds, but the previously mentioned scrolling
> trick lets it continue for a little while again. If I don't do that, it will
> still print the context_section time out.
>
> Also of note, with the freezes, I can now "pause" the VM thanks to the
> scrolling trick. Resuming it at this point seems to set it into a functional
> and playable state, which is almost what it was like back in 1.5.10.

What distro you using ? It seems it was fixed for Ubuntu.

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

--- Comment #9 from Kevin Meyer  2012-10-10 18:48:45 CDT ---
(In reply to comment #8)
> What distro you using ? It seems it was fixed for Ubuntu.

Ubuntu is a distro I absolutely detest for various reasons not suitable for
discussion on bugzilla, in short I will never use it willingly or wish it upon
my enemies.

My current of choice is Arch, which is rather well known for shipping the
latest and greatest vanilla packages.

[Bug 31883] PCSX2 locks up when/shortly after initializing MTGS (dogfood)

http://bugs.winehq.org/show_bug.cgi?id=31883

Austin English  changed:

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

--- Comment #10 from Austin English  2012-10-10 19:53:12 CDT ---
(In reply to comment #9)
> (In reply to comment #8)
> > What distro you using ? It seems it was fixed for Ubuntu.
>
> Ubuntu is a distro I absolutely detest for various reasons not suitable for
> discussion on bugzilla, in short I will never use it willingly or wish it upon
> my enemies.
>
> My current of choice is Arch, which is rather well known for shipping the
> latest and greatest vanilla packages.

It's also known to ship broken packages, see,
http://bugs.winehq.org/show_bug.cgi?id=22316 et al.