[Bug 60382] New: Diablo 2 Resurrected: game freezes after 5-60 mins of play (thread spins in RtlWakeAddressSingle on a corrupted address-wait queue)

[WineHQ Bugzilla]

http://bugs.winehq.org/show_bug.cgi?id=60382

            Bug ID: 60382
           Summary: Diablo 2 Resurrected: game freezes after 5-60 mins of
                    play (thread spins in RtlWakeAddressSingle on a
                    corrupted address-wait queue)
           Product: Wine
           Version: 11.18
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ntdll
          Assignee: [e-mail]
          Reporter: [e-mail]
  Target Milestone: ---
      Distribution: ---

== Symptom ==

Diablo II: Resurrected (Battle.net, 64-bit) freezes after minutes to hours
of play (e.g. observed at 6 and 43 minutes).

Before freeze: GPU ~20% CPU ~30% (not important)
After freeze: GPU 0% CPU 100% all cores (16 in my case)
Same with fsync enabled or disabled (tried both, thouhgt it's fsync, but it's
NOT fsync)

== Playing with gdb and perf ==

TLDR: The main thread loops at RtlWakeAddressSingle+140 while holding the queue
spinlock (lock == -1). Other threads spin in spin_lock()/NtYieldExecution().
The queue list is cyclic and leads into a neighbouring queue's head:
H.next = Q, Q.next = A, A.next = B, B.next = Q. The walk never returns to H.

TLDR2: from debugging I got into this code in RtlWaitOnAddress 

https://gitlab.winehq.org/wine/wine/-/blob/wine-11.18/dlls/ntdll/sync.c#L911...

    /* We may have already been removed by a call to RtlWakeAddressSingle() or
RtlWakeAddressAll(). */
    if (entry.addr)                 <---- check without the lock
    {
        spin_lock( &queue->lock );  <---- then locks
        if (entry.addr)             <---- and then checks again?? suspicious
right?
            list_remove( &entry.entry );
        spin_unlock( &queue->lock );
    }

First try removing just ourter `if (entry.addr)` it works perfectly for me:

    spin_lock( &queue->lock );
    if (entry.addr)
        list_remove( &entry.entry );
    spin_unlock( &queue->lock );


After this at least in D2R I was playing like many hours since 14 Sep 2026 till
now 25 Sep 2026 and not a single freeze occurred :)

Is it a proof? No. But it's good enough for me, and I hope it helps.

== How to reproduce ==

In other curcumstances I'd be happy to provide reproduction guide, but due to
the nature of the bug there's no deterministic way to reproduce to get a freeze
(at least not in D2R). Also, the game is pay to play.

I think it's possible to write a synthetic binary which uses this queue and
hits the same issue.

-- 
Do not reply to this email, post in Bugzilla using the
above URL to reply.
You are receiving this mail because:
You are watching all bug changes.