Software information

Watch Dogs Legion (Ubisoft Connect), DX12 mode, no mods.

System information

• GPU: AMD Radeon RX 9060 XT (RDNA4, Navi 44, SOC 0x7590, PCI 0000:0a:00.0)
• Driver: amdgpu (kernel 7.1.5-1-cachyos), Mesa RADV
• Wine/Proton: CachyOS native Proton
• vkd3d-proton version: 3.1.0

Issue

GPU hang during gameplay. The crash follows the same pattern as #3130 (Starfield on RDNA4):

ring gfx_0.0.0 timeout, signaled seq=108457621, emitted seq=108457624
Process WatchDogsLegion pid 832939 thread vkd3d_queue pid 833084
Ring gfx_0.0.0 reset failed
GPU reset begin!. Source: 1
MES(1) failed to respond to msg=REMOVE_QUEUE  (x5)
failed to unmap legacy queue
gfx_v12_0_hw_fini failed to halt cp gfx
GPU mode1 reset
VRAM is lost due to GPU reset!
Failed to initialize parser -125
device wedged, but no recovery needed

Devcoredump analysis

Key findings from the devcoredump:

[gfxhub] Page fault observed
Faulty page starting at address: 0x0000000000000000
Protection fault status register: 0x0

regCP_CPF_STALLED_STAT1    0x00000001
regCP_CPF_BUSY_STAT        0x00000006
regCP_CPF_STATUS           0xb4000063
regGCVM_L2_PROTECTION_FAULT_STATUS_LO32  0x00701031
regCP_MES_HEADER_DUMP      0xdef0def0 through 0xdef7def7

The fault is a NULL page dereference in GPU virtual address space (address 0x0000000000000000). The L2 page fault handler correctly identified the unmapped address. MES was mid-operation when the GFX ring timed out.

This appears to be VKD3D-Proton passing a NULL GPU pointer to the command processor. The GPU is doing the right thing by faulting — the issue is in the DX12-to-Vulkan translation layer.

Notes

• DXVK (DX11) does not exhibit this issue — only VKD3D-Proton (DX12) path is affected
• Same hardware/driver combo works fine with DXVK games (Starfield DX11 mode, Division 2 DX11 mode)
• Related: #3130 (Starfield, same RDNA4 ring timeout pattern)
• GPU recovers via MODE1 reset but VRAM is lost, requiring game restart