[Bug 54411] New: ntoskrnl.exe:ntoskrnl - test_pnp_devices() sometimes gets an unexpected success in Wine

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54411

Bug ID: 54411
           Summary: ntoskrnl.exe:ntoskrnl - test_pnp_devices() sometimes
                    gets an unexpected success in Wine
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ntoskrnl
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

ntoskrnl.exe:ntoskrnl - test_pnp_devices() sometimes gets an unexpected success
in Wine:

ntoskrnl.c:1715: Test succeeded inside todo block: got error 997

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

Where 997 == ERROR_IO_PENDING

The test happens about once per month but has been known to impact merge
requests (e.g. MR2039).

This looks like a race condition. Mark the test as flaky in Wine?

[Bug 54411] ntoskrnl.exe:ntoskrnl - test_pnp_devices() sometimes gets an unexpected success in Wine

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54411

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
           Keywords|                            |source, testcase

[Bug 54411] ntoskrnl.exe:ntoskrnl - test_pnp_devices() sometimes gets an unexpected success in Wine

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=54411

Zeb Figura  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
            Version|unspecified                 |8.0

--- Comment #1 from Zeb Figura  ---
Yes, it's a race condition. The device is deleted on IoDeleteDevice(), which
causes further server calls to return STATUS_FILE_DELETED in the server, but
sometimes that ioctl makes it through before we hit IoDeleteDevice(). The
correct thing to do is not delete the device until the last reference is
released, and not force calls to return STATUS_FILE_DELETED. Ideally we should
fix the code, although it's a bit tricky, because the refcounting is currently
kind of backwards in the server.