Graphics Execution

From Open-IOV
Revision as of 00:06, 10 November 2026 by WikiBot (talk | contribs) (2026-11-09 evening refresh 2: forced GLX selects libnvidia-glcore; windowed glxgears hangs in the glcore kick wait; kernel-visible present completion is the wall)
Jump to navigation Jump to search

An absence of critical technical documentation has historically slowed growth and adoption of developer ecosystems for GPU virtualization.

This CC-BY-4.0 licensed content can either be used with attribution, or used as inspiration for new documentation, created by GPU vendors for public commercial distribution as developer documentation.

Where possible, this documentation will clearly label dates and versions of observed-but-not-guaranteed behaviour vs. vendor-documented stable interfaces/behaviour with guarantees of forward or backward compatibility.

Graphics Execution

Scope. How the emulated vGPU runs 3D graphics work — the OpenGL and Vulkan command streams a stock guest submits — on the physical GPU, and what is proven about it. This is the shared machinery behind OpenGL.md, Vulkan.md and the Windows/DirectX target (DirectX.md); CUDA compute uses the same submission and completion path for a different engine workload (Execution-Frontier.md). Canonical sources: ../docs/04-implementation/05-graphics-execution-plan.md (the plan and milestone ladder) and ../docs/04-implementation/04-real-render-walls.md (the measured walls); the docs tree wins on disagreement.

Status: 2026-11-09, tree master @ 21e0d611. Tags: [live] observed end to end; [wip] partially built; [wall] current blocker; [target] defined success state; [planned] designed, not built.

1. What "real pixels" means

The acceptance target is strict: a stock guest GL or Vulkan application's draw commands are executed by the physical GPU's fixed-function 3D pipeline (vertex fetch, primitive assembly, rasterisation, fragment work, ROP), the render target is readable by the guest, and that image reaches the capture path through the emulated mdev. The pixels must not come from another GPU, a software rasteriser, or a host display round-trip.

(The project has also published two deliberately-labelled two-VM demo videos that use the standard virtio-gpu/virgl virtual GPU on a host integrated GPU, and a CUDA-animation video that does run on the emulated vGPU through the compute path. The virgl videos are honest demonstrations of the guest experience; they are not this execution path, and their captions say so.)

2. The path a graphics command takes

The graphics execution paths: guest OpenGL/Vulkan/CUDA commands to the physical engines, and completions back.
guest apps (GL / Vulkan / CUDA)
  -> guest user space (GLX client + GL core, Vulkan ICD, libcuda)
  -> guest X server (GLX module + NVIDIA DDX)           [OpenGL]
  -> nvidia.ko: channels, GPFIFO rings, doorbells       [all APIs]
  -> vgsp.ko: capture sections, decode pushes, forward
  -> physical GPU front-end -> GR engine (3D + compute) / copy engines
  <- engine-authored releases written into guest memory
  <- completion relay (MSI-X + wakes) back to the guest
ASCII fallback above; the diagram shows the three execution lanes.

3. The three execution lanes

Graphics work reaches the engines today in one of three shapes, in increasing fidelity:

  1. Verbatim 3D method-stream replay [live]. The emulation captures the X server's graphics-class (0xC597, TURING_A) section from its channel, relocates the operand addresses into host-built shadow pages the engine can reach, and submits the stream unchanged. The engine retires it: the uploads (LINE_LENGTH_IN/LAUNCH_DMA), the state, program and texture binds, and the release all execute with no MMU fault. This proves the mechanics of 3D execution — object bind, window placement, operand relocation, completion — on real hardware.
  2. Synthesized launch from a captured section [live]. A compute-shaped launch built from the captured section (operand list, program, a 64-word launch descriptor) reaches the engine and retires real sm_75 SASS; the Turing shader-program-header form (V3) is accepted with zero header faults across repeated launches. This proved engine-side dispatch before the verbatim lane existed. The program itself is still a development stand-in at this rung — execution proof, not pixels.
  3. Forwarding the client's own pushes [wip]. The GL client's own channels now register, submit and retire: the direct-GL GR channel (0xbeef0100) and its CE siblings (0xbeef0101/0xbeef0102) are discovered, their rings and releases are decoded, and the X server's DDX kick channel (0xbfef0002) is decoded from its ring head with its class-2D notification authored. The X DDX waits are fixed (class-aware display arm, 0x8000 -> 0 notification completion, decoded GL release applied), and with GLX forced (SDL_VIDEO_X11_FORCE_EGL=0) the windowed GL path selects libnvidia-glcore. What remains is the kernel-visible GL present/kick completion, the DDX get-word/notifier recycle, and physical dispatch of the client's draws (§6).

4. The channel and object picture

  • Every submission lives on a channel: a GPFIFO ring plus engine objects, with a USERD the driver advances its put pointer in and a doorbell the host module traps. Channels are typed by engine: a graphics channel carries 3D state and draws; copy-class channels carry memory movements.
  • The graphics object class the guest's 3D stream binds is the Turing graphics class, 0xC597. Streams now come from both the X server's channel and the direct GL client's own channels (0xbeef0100 GR plus the 0xbeef0101/0xbeef0102 CE siblings); the 2D present/readback work rides lighter 2D-class and copy-engine shapes.
  • Display and modeset [live]. The guest's modeset path is served by an emulated display engine: the display push-buffer channels (core 0xc57d, window 0xc57e, immediate/window 0xc57b) are registered, their push buffers parsed, and the notifier/UPDATE completion authored. The arm is class-aware (the window class has no NOTIFY bit; a bound notifier is the arm), and the X DDX's 0x8000 -> 0 notification completion is authored. This is what lets nvidia-modeset/NvKms register (/dev/dri), X settle and GLX clients obtain a real GL context. The display-capability seam is parameterised in the campaign configuration.
  • Cells of state: 3D streams are long—hundreds of method words per draw — and mix state binds (viewport, render target, ZS, blend/ROP), program and constant binds, vertex/index streams and draw-issuing methods. The capture machinery recognises the init-shaped operands today; the per-draw state vocabulary is being extended ([G-B in the plan]( ../docs/04-implementation/05-graphics-execution-plan.md)).
  • Synchronisation: tracking semaphores, notifications and fence values written by the engine; the guest's swap/present waits sit on top of those.

5. Completion, fences and presents

The acceptance signal is the engine-authored release: a semaphore the engine writes when the work completes. For GL, completion flows through the channel's semaphores/notifications to the X server and the app's swap wait; for Vulkan, the client's fence value is released; for CUDA, the launch's release group wakes the client. The emulation relays engine completions into guest-observable memory (MSI-X plus a poller), and the guest resumes from there.

For the display path, completion is the notifier write and UPDATE -> FINISHED; a completion event deferred by the reply path is force-retried until it posts, so the X server's mode set finishes. For the X DDX's own 2D push buffers, completion also requires the GPU-written engine get word to advance and the notifier to be recycled to 0xffff after a kick; both are host-side levers now, and a diagnostic write of the get word released the DDX's reserve loop. For a windowed GL client the present is a kick whose completion the guest kernel observes by polling its per-slot /dev/nvidia-modeset fd; the host must complete the exact notifier page the kernel watches and signal that event, not only write the notifier memory.

During bring-up the module also carries a diagnostics-only mirror for apps that spin on a shared progress word; it exists so a run can proceed while the natural completion path is being finished. The 60-fps milestone explicitly requires it switched off, along with any external poke.

6. Where the frontier is (ranked walls)

  1. Kernel-visible GL present completion [wall]. With GLX forced the windowed client uses libnvidia-glcore; the minimal glxgears then hangs in the glcore kick wait: the client polls its per-slot /dev/nvidia-modeset fd (poll(POLLIN, 1 s)). The host's sweep completes the beef notifier memory, but the guest kernel never signals the event the poll waits on. The exact notifier page the kernel watches must be completed and the guest event signalled (the kick-notify executor), not just the memory written.
  2. The X DDX push-buffer completion [wall]. The DDX's 2D reserve loop exits only when the GPU-written engine get word advances (get <= put) and the kick requires the notifier to be recycled to 0xffff after completion. A diagnostic host write released the loop; the kernel levers to publish/advance the get word and recycle the notifier are next.
  3. The X DDX footprint [wall]. The DDX allocates 4 GB of system pages at startup; an 8 GB guest has no headroom once GZDoom's own footprint lands and both are OOM-killed. Current runs use a 12 GB guest; the allocation is a suspected emulation gap (a 512 MiB emulated VRAM falling back to a 4 GB sysmem allocation) to fix.
  4. Physical dispatch of the GL draws [wall]. The direct-GL channels submit and get host-completed releases, but the forwarding path reports no dispatch to the physical GR engine (forwarded=0); the draws are decoded and host-completed, not executed on hardware.
  5. Non-black frames and sustained rate [wall]/[target]. Once the kernel-visible present completion and the DDX get-word/notifier completion land, the capture must show motion, then 60 fps in one VM with no forcing, then two VMs on two mediated devices with non-black frames in both.

7. What is proven today

Capability State
X server initialises on the emulated vGPU (driver DMA sanity push, ScreenInit) [live]
Display engine serves the modeset push buffers; notifier plus UPDATE -> FINISHED [live] (campaign configuration)
nvidia-modeset/NvKms registers (/dev/dri/card0, renderD128, fb0); nvidia-smi -L enumerates [live]
glxinfo reports the emulated GPU with the NVIDIA renderer (OpenGL 4.6.0, direct rendering) [live]
GZDoom obtains a real GL context; windowed reaches renderer startup [live] (pixels frontier)
GZDoom reaches MAP01 on Mesa llvmpipe (not the emulated vGPU path) [wip]
Forced GLX (SDL_VIDEO_X11_FORCE_EGL=0) selects libnvidia-glcore [live]
Windowed GL present/kick completion (minimal glxgears) [wall]
X server survives the GZDoom run on a 12 GB guest (the DDX's 4 GB sysmem allocation forces the size) [live]
GL section capture: the X server's 0xC597 stream decoded [live]
Verbatim 3D method-stream replay retired by the engine (no MMU fault) [live]
Synthesized launch executes real sm_75 SASS; Turing V3 program header accepted [live]
2D screen fills and copy-engine present ops execute [live]
Direct GL client: device-init gate handled, channels allocate, readback path returns clean [live]
Direct GL client channels submit and retire (GR 0xbeef0100, CE 0xbeef0101/0xbeef0102); DDX kick channel decoded [live]
Guest kernel observes the kick completion and signals the modeset event [wall]
NVIDIA Vulkan ICD creates a device [wall]
DDX push-buffer get-word publish and notifier recycle [wall]
Physical dispatch of the direct-GL draws (forwarded=0) [wall]
Non-black pixels from the emulated vGPU's 3D pipeline [wall]
Windows/DirectX guests [planned]

8. What is not claimed

  • No non-black pixel has been produced by the emulated vGPU's 3D pipeline yet.
  • Display/KMS bring-up (NvKms, /dev/dri, X, a real GL context) is not a rendered-frame claim; GZDoom reaching MAP01 on Mesa llvmpipe is not the emulated-vGPU path, and the later forced-GLX run selects libnvidia-glcore but hangs in the present/kick wait. The kernel-visible present completion, the DDX get-word/notifier completion and the physical dispatch of the GL draws are the current walls.
  • No frame-rate or latency numbers: the scheduled measurement is defined (60 fps over 60 s, no forcing) but not yet met.
  • The virgl-based demo videos are a different, labelled path — not this one.

9. Related pages