[Bug 54510] New: d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

Bug ID: 54510
           Summary: d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the
                    debian11 VM
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: d3d
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM:

d3d11.c:25778: Test failed: Got 0, expected 33 (index 1).
d3d11.c:25778: Test failed: Got 0, expected 66 (index 2).
...
d3d11.c:25810: Test failed: Got 0, expected 32 (index 0).
d3d11.c:25810: Test failed: Got 0, expected 96 (index 1).
...
d3d11.c:25852: Test failed: Got 0.00000000e+000, expected 1 (index 32).
d3d11.c:25854: Test failed: Got 0, expected 1 (index 32).
d3d11.c:25852: Test failed: Got 0.00000000e+000, expected 2 (index 33).
d3d11.c:25854: Test failed: Got 0, expected 2 (index 33).
...

See https://test.winehq.org/data/patterns.html#d3d11:d3d11

These failures are close to 100% reproducible (but not quite) and started on
2023-02-14. But note that the test also has a preexisting crash which masks
these failures in the test pattern.

Interestingly these failures do not happen:
* In the wow32 and wow64 tests (note: debian11b is a clone of debian11).
* On my fgtb-debian11 VM.
* On the debiant VM.

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |source, testcase

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

Zeb Figura  changed:

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

--- Comment #1 from Zeb Figura  ---
(In reply to François Gouget from comment #0)
> These failures are close to 100% reproducible (but not quite) and started on
> 2023-02-14. But note that the test also has a preexisting crash which masks
> these failures in the test pattern.

And that crash was fixed on the same date, so it's likely that these failures
are older than that.

If I had to guess, I'd say these failures are *also* due to address space
exhaustion? I don't know why wow64 would affect that, but I guess wow64 does
take up a *bit* of extra address space (dlls, extra TEB), so if we're that
close it could maybe make a difference...

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
    Regression SHA1|                            |97b836a938928ce714fd7cf430a
                   |                            |97d5fbefd7df0
           Keywords|                            |regression

--- Comment #2 from François Gouget  ---
I checked the logs and these failures were *not* present before 2023-02-14, and
they are still present in the 2023-02-17 build.

A bisect shows that these failures started with the commit below which means
they are related to the patch that was meant to fix bug 53217 (i.e. the crash):

commit 97b836a938928ce714fd7cf430a97d5fbefd7df0
Author:     Zebediah Figura 
AuthorDate: Thu Feb 2 16:59:14 2023 -0600

wined3d: Destroy objects belonging to wined3d_device_gl.completed_fence_id.

Not just those belonging to earlier completed fences.

completed_fence_id names the most recent fence to be completed. In essence,
this fixes an off-by-one error.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=53217

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

--- Comment #3 from François Gouget  ---
Correction: still present in the 2023-02-16 build (as checked in the 2023-02-17
TestBot results).

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

--- Comment #4 from François Gouget  ---
It did happen on debian11b-wow32 today:
https://test.winehq.org/data/d0ce5a77c60e0a9613f1af03d67d13b89816441b/linux_...

So what I said about it not happening in wow32 was wrong. That config just got
lucky in the previous couple of days.

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

--- Comment #5 from Zeb Figura  ---
Whatever the cause, I'm not seeing this anywhere on the patterns page. It was
intermittent, but not rare, so should we assume it's fixed?

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

Zeb Figura  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |WORKSFORME
             Status|NEW                         |RESOLVED

--- Comment #6 from Zeb Figura  ---
This hasn't happened recently either. I'm going to call it RESOLVED WORKSFORME,
and of course we can look into it again if it ever does come up.

[Bug 54510] d3d11:d3d11 - test_tgsm() fails in 32-bit tests on the debian11 VM

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54510

Austin English  changed:

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

--- Comment #7 from Austin English  ---
Closing.