Possible critical section bug - lock by invalid thread

[James Liggett]

I also see this in PokerStars.NET (an online poker game) In my case, I
believe the threads are valid however. I'll check on that and get back
to you. I notice it after playing the game for about a half-hour,
sometimes a bit longer. When this happens, I also noticed that I can
barely move my mouse, CPU usage is maxed, and the hard drive is busy
too. I think your theory about a race condition is probably correct. The
odd thing is I see this in both ALSA and OSS sound modes, so I don't
think the problem is there, but is more likely to be somewhere the
server, because any critical section used by a running Windows program
seems to be affected. I get wait timeouts like what you're showing
everywhere in Wine (x11drv, sound, anything) Perhaps you should post a
full output the errors you get someplace on the web. (I imagine it'll be
too big for the list)

James

On Mon, 2005-08-29 at 14:07 -0500, Alex Villací­s Lasso wrote:
> In several RPG I have been testing, I have found deadlocks on the sound 
> thread - the sound loops on and on, and the program freezes. At the same 
> time, I see messages like the following:
> 
> fixme:ddraw:Main_DirectDraw_SetCooperativeLevel 
> (0x7fe32708)->(00010022,00000008)
> err:dsound:DSOUND_MixOne underrun on sound buffer 0x7fe82310
> err:dsound:DSOUND_MixOne underrun on sound buffer 0x7fe82880
> err:wave:DSDB_MMAPCopy mmap partially failed.
> err:ntdll:RtlpWaitForCriticalSection section 0x7fe3336c "?" wait timed 
> out in thread 000a, blocked by 0000, retrying (60 sec)
> err:ntdll:RtlpWaitForCriticalSection section 0x7fe3336c "?" wait timed 
> out in thread 0009, blocked by 0000, retrying (60 sec)
> err:ntdll:RtlpWaitForCriticalSection section 0x7fe3336c "?" wait timed 
> out in thread 000a, blocked by 0000, retrying (60 sec)
> err:ntdll:RtlpWaitForCriticalSection section 0x7fe3336c "?" wait timed 
> out in thread 0009, blocked by 0000, retrying (60 sec)
> 
> I thought that DirectSound was not performing locking properly, until I 
> noticed the following: all the times I see this deadlock, the thread ID 
> "responsible" for the deadlock is 0000. From a little digging in the 
> code, I see that 0 is the value used in the critical section when it is 
> not held by any thread. So these are potentially cases where a previous 
> thread frees the critical section, but somehow the critical section 
> remains locked. I think I saw a select() call in the wineserver, used as 
> an implementation for this wait, and this select() call times out, 
> resulting in the behavior shown above.
> 
> All these instances happen under heavy CPU load, so I suspect some sort 
> of race condition. If I really try and bog down the CPU, I can even 
> trigger this deadlock on the DirectSound tests (under ALSA).
> 
> Could anybody who knows about Wine critical sections comment on this? I 
> do not really understand the implementation of critical sections at the 
> wineserver side (other than the fact that a select() is used somewhere), 
> so I could take a long time to unravel this bug. This problem does not 
> seem specific to DirectSound as far as I could tell, so this might be 
> affecting a couple other apps under Wine. Also, comments on how I could 
> supply more information on this are welcome.
> 
> Alex Villacís Lasso
> 
>