Skip to content

WDF-001: stability qualification is incomplete

Two real guest bugchecks interrupted separate controller campaigns. Two DS4 series passed the per-run validation before the next repetition crashed. In a later Xbox One/GIP campaign, two producer/receiver sequences completed before another crash; those filter series remain excluded from latency comparisons because the adapter captured extra protocol reports without one-to-one generated-state correlation. Both minidumps have the same WDF queue-purge signature at Vdevice+0x651e. Successful latency rows do not erase this incident and do not establish safe repeated teardown.

Evidence Observation
Bugcheck WDF_VIOLATION 0x10D, argument 1 0xA ? fatal queue operation
Stack boundary Wdf01000!imp_WdfIoQueuePurge ? Vdevice+0x651e
Driver under investigation Vdevice 1.0.5.0, SHA-256 3977be4dd7811eb4e0f0bc0ad9e2e2df3f63facc514d5484a08c78fd9b1bc5c8
Test contexts Virtual DS4 and virtual Xbox One/GIP with Interceptor Gamepad & Controller 1.0.13.0 in Windows 11 Hyper-V
Driver Verifier Flags 0x00000000; no verified drivers configured
Host impact None observed; all tested drivers and the crash remained in the dedicated VM

What is established

Windows recorded the bugcheck and saved a minidump and kernel dump. Microsoft CDB analysis reaches Vdevice at an asynchronous WDF queue-purge call. The WDF in-flight log explicitly reports that the queue is already deleted, with 0xC0000056 (STATUS_DELETE_PENDING). The public source has WdfIoQueuePurge calls in VdEndpointPurge and queue deletion in DeleteVirtual. This establishes the invalid queue state at the failing call; it does not yet establish the complete ordering of the initiating callbacks.

The exact lifetime race and initiating sequence are still under investigation. Private matching driver symbols were not loaded in the first analysis. Stack involvement does not prove that every Vdevice configuration fails or that another component is irrelevant. No driver was patched or rebuilt to change the benchmark outcome.

Required follow-up

Preserve the dump and exact signed binary; correlate WDF queue state with endpoint purge and device deletion. Build an isolated teardown reproducer, review callback/queue lifetime ordering, and test the eventual fix under cancellation, repeated removal, restart and concurrent requests. Any changed kernel image needs its normal signing and qualification process before distribution.

The dumps remain local because a kernel dump contains broad system state. Their SHA-256 fingerprints and a sanitized incident record are published in reports/incident-wdf-001.json. Raw completed and interrupted runs remain in the dataset. This is an open stability limitation, not a resolved bug or a production certification.