HwRender: fix ARM64 crash when GPU is suspended during Present (DXGI_ERROR_DEVICE_REMOVED)

[etvorun]

Summary

On ARM64 devices, WPF crashes when the GPU is suspended during an active IDirect3DSwapChain9::Present call. The D3D9-over-DXGI compatibility layer on ARM64 surfaces DXGI_ERROR_DEVICE_REMOVED (0x887A0005) directly instead of translating it to D3DERR_DEVICELOST. WPF's PresentWithD3D did not recognize this code; it fell through to MIL_THR, which treats 0x887A0005 as fatal and crashes the process. HandlePresentFailure — and the device-lost recovery path — was never reached.

What changed

src/Microsoft.DotNet.Wpf/src/WpfGfx/include/wgx_error.h

• Added a local #define DXGI_ERROR_DEVICE_REMOVED ((HRESULT)0x887A0005L) guarded by #ifndef, avoiding a new dependency on dxgi.h. The value is a stable DirectX ABI constant.

src/Microsoft.DotNet.Wpf/src/WpfGfx/core/hw/d3ddevice.cpp

• PresentWithD3D: new else if (hr == DXGI_ERROR_DEVICE_REMOVED) branch placed before the MIL_THR call, converting the error to D3DERR_DEVICELOST so the existing device-lost recovery path handles it.
• HandlePresentFailure: added DXGI_ERROR_DEVICE_REMOVED to the device-lost if condition, as defense-in-depth for other callers such as PresentWithGDI.

Why

The fix is aligned with the previously validated internal source fix behavior. The change is surgical: no new recovery logic is introduced; WPF reuses the existing TDR / device-lost path (MarkUnusable() → WGXERR_DISPLAYSTATEINVALID → RENDERING_STATUS_DEVICE_LOST → resource recreation on GPU resume).

Recovery sequence

State	What happens
Present returns DXGI_ERROR_DEVICE_REMOVED	Converted to D3DERR_DEVICELOST; HandlePresentFailure calls MarkUnusable(), invalidates GPU resources, notifies device manager
Same frame — render thread	CRenderTargetManager::HandlePresentErrors swallows WGXERR_DISPLAYSTATEINVALID; fires RENDERING_STATUS_DEVICE_LOST to UI thread. No crash, no zombie.
Subsequent frames (GPU suspended)	UpdateDisplayState returns WGXERR_DISPLAYSTATEINVALID; render passes are skipped
GPU comes back	UpdateDisplayState succeeds; NotifyTierChange fires; render targets recreated; window repainted

Validation

Validated manually on ARM64 hardware by triggering GPU suspension events while a WPF application is actively rendering. The application recovered without crashing after applying this fix. No automated test infrastructure changes are included.

Fixes #11471

Microsoft Reviewers: Open in CodeFlow

[Copilot]

Pull request overview

This PR fixes a critical crash on ARM64 devices when the GPU is suspended during active rendering. On ARM64, the D3D9-over-DXGI compatibility layer returns DXGI_ERROR_DEVICE_REMOVED (0x887A0005) instead of the expected D3D9 error code D3DERR_DEVICELOST. WPF's PresentWithD3D did not recognize this code, causing MIL_THR to treat it as fatal and crash the process. The fix converts the DXGI error to the D3D9 equivalent before error processing, allowing the existing device-lost recovery mechanism to handle GPU suspension gracefully.

Changes:

• Defined DXGI_ERROR_DEVICE_REMOVED constant locally in wgx_error.h to avoid adding a dependency on dxgi.h
• Added error code translation in PresentWithD3D to convert DXGI_ERROR_DEVICE_REMOVED to D3DERR_DEVICELOST before MIL_THR processing
• Added DXGI_ERROR_DEVICE_REMOVED to the device-lost condition in HandlePresentFailure as defense-in-depth

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/Microsoft.DotNet.Wpf/src/WpfGfx/include/wgx_error.h	Adds local definition of DXGI_ERROR_DEVICE_REMOVED constant (0x887A0005L) with #ifndef guard
src/Microsoft.DotNet.Wpf/src/WpfGfx/core/hw/d3ddevice.cpp	Converts DXGI_ERROR_DEVICE_REMOVED to D3DERR_DEVICELOST in PresentWithD3D before MIL_THR processing; adds DXGI_ERROR_DEVICE_REMOVED to HandlePresentFailure's device-lost condition

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[lindexi]

LGTM