From c3a295efd1b1132bbaa5d422cd05b3fa8e89a8bf Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Mon, 21 Sep 2026 13:06:51 -0400 Subject: kernel: Fix GSP emulation shadow binding, eventfd wiring, RPC livelocks + docs suite + repo reorg CRITICAL FIXES (guest driver bring-up): - gsp_emulation.c: emulation engine bound to the WRONG BAR0 - it polled the host physical BAR0 (bars[0].vaddr ioremap) for the doorbell and read RPCs from a local dma_alloc_wc buffer, while the guest writes land in the vzalloc'd bar0_shadow. The RPC path could never work; PBA writes even corrupted the host GPU's physical BAR0+0x3000. gsp_emu_init now binds exclusively to bar0_shadow (doorbell/mailbox/PBA aliases; shared_mem_dma==0 marks the alias so cleanup never dma_free's it). - main.c SET_IRQS: DATA_EVENTFD eventfds were DISCARDED ('TODO' + kfree) - the guest never received MSI-X completions. Full wiring now: eventfd_ctx_fdget per vector, fd -1 clears, dev_close releases refs. gsp_emu_inject_completion_interrupt looks up vgpu->irqs[0].trigger LIVE (init-time copy was permanently NULL since QEMU wires after gsp_emu_init). - gsp_rpc_client.c: BAR_TRAP/MIGRATE in emulation mode called gsp_rpc_submit_raw which rings the SAME doorbell the thread polls - self-triggered RPC livelock + multi-second stalls. Now fail fast with -ENOTSUPP + pr_warn_ratelimited (guest BAR access is direct MMIO). MORE FIXES: - main.c: nv_vgpu_config_write returned SIZE_MAX (while(count--) wrap) and was dead code while the VFIO write handler did unfiltered byte copies (guest could rewrite vendor ID/BARs). Return fixed; VFIO write handler now routes through the per-register filter. - kernel/Makefile: M=$(PWD) used the env var which is stale when invoked via the top-level make (pointed at repo root -> kbuild silently built NOTHING; explains every 'stale .ko' incident). Now M=$(CURDIR). - Warnings sweep: header prototypes for mdev/type functions, format casts in gsp_comm.c DRY-RUN logs, unused variable. ROBUSTNESS (harness/tooling): - vgpu_vm.py: run had no timeout (start could block forever); guest_shell recv could block forever on a wedged guest. Both now bounded; guest_shell pre-flights guest-ping and reports wedge diagnostics. - stack_manager.sh: qemu/qemu-debug now launch via the vgpu_vm.py harness (workflow rule: never raw QEMU). REPO REORGANIZATION: - driver/ (legacy guest-side gsp-vgpu.ko) moved to legacy-driver-guest/ (reference only, not built; README explains). 3 legacy un-buildable test files moved to legacy-driver-guest/kernel-tests/. Top-level Makefile cleaned of the dead DRIVER_DIR reference. DOCUMENTATION SUITE (new): - docs/GSP_EMULATION_ARCHITECTURE.md - emulator deep-dive: shadow BAR0 model, doorbell protocol, dispatch table, MSI-X path, lifecycle, memory- model caveats - docs/DEBUGGING.md - QEMU GDB stub workflow, serial console, QMP NMI, failure taxonomy, full bring-up validation checklist, EPT-livelock case study, version-matrix discipline - docs/ONBOARDING.md - repo tour, coding standards (incl. virt_to_phys / dma_addr_t / get-put rules), first build & test - docs/VERSION_UPGRADE.md - end-to-end new-driver-version procedure (firmware fetch, Ghidra analysis, configs, signatures, validation) - docs/PERFORMANCE.md - bare-metal A/B methodology + targets - README.md doc index; STATE.md refreshed; DEVELOPMENT_WORKFLOW.md corrected (load order: open-iov-vfio BEFORE nvidia; stock-driver blacklist note) --- docs/PERFORMANCE.md | 98 +++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 98 insertions(+) create mode 100644 docs/PERFORMANCE.md (limited to 'docs/PERFORMANCE.md') diff --git a/docs/PERFORMANCE.md b/docs/PERFORMANCE.md new file mode 100644 index 0000000..6077920 --- /dev/null +++ b/docs/PERFORMANCE.md @@ -0,0 +1,98 @@ +# Open-IOV Performance Testing & A/B Methodology + +> How we measure vGPU performance against bare metal, interpret the results, +> and keep overhead on the critical path honest. **Run correctness first** +> (`docs/DEBUGGING.md` §5) — numbers from a half-working stack are noise. + +--- + +## 1. What overhead is acceptable (targets) + +| Path | Target | Why | +|------|--------|-----| +| GSP RPC round-trip (doorbell→response→MSI-X) | < 50 µs (stretch: < 10 µs) | every driver op goes through here | +| MSI-X injection (eventfd → guest handler) | < 5 µs | completion latency dominates | +| Guest BAR0 MMIO read/write | < 2 µs (shadow RAM) | hot path for register polling | +| CUDA kernel launch (added overhead vs bare metal) | < 10% | user-visible | +| Memory alloc/map RPCs | < 100 µs | setup-time, amortized | +| CUDA compute throughput | ≥ 95% of bare metal | GPU-bound, should be ~free | +| Graphics (frame rate) | ≥ 90% of bare metal | swapchain + submission overhead | + +## 2. The A/B rule + +Every measurement runs **twice on the same machine, same boot if possible**: + +- **A (bare metal)**: workload on the physical GPU via the host's stock + driver (or our OpenRM nvidia.ko directly on the hardware). +- **B (vGPU)**: identical workload in the guest on the emulated vGPU. + +Same binary, same input sizes, same environment (governor, clocks, hugepages). +Report **B/A ratio %** — never absolute numbers alone. + +One variable at a time. If B regresses, you changed the stack, the workload, +or the machine — find which before optimizing. + +## 3. Microbenchmarks (build these first) + +1. **RPC round-trip**: from the guest, submit a DEBUG/NOOP-ish RPC in a loop; + time doorbell-write → response-visible. Host side: `gsp_emu_get_stats()` + gives `total_latency_ns` and counts — divide for per-RPC host cost. + (Add `trace_printk` in `gsp_emu_dispatch_rpc` entry/exit for distribution; + read via `/sys/kernel/debug/tracing/trace`.) +2. **MSI-X latency**: timestamp (guest TSC) at RPC submit → in the guest + driver's interrupt handler. Compare with the host eventfd_signal timestamp. +3. **BAR0 shadow read latency**: guest loops reading a shadow register; + expect ~memory speed (uncached, so ~100-300 ns) — anything ≥ 1 µs means + you're trapping somewhere you shouldn't. +4. **Doorbell polling cost**: host `top`/`perf` on `gsp_emu_N` — the adaptive + poll (busy → 50-100 µs sleep) should sit at ~0% CPU when idle, and react + within one sleep quantum under load. Tune `GSP_EMU_POLL_*` only with data. + +## 4. Workload benchmarks + +Runner: `tools/benchmark_infrastructure.sh` (extend it; keep the A/B pattern). +In-guest execution goes through the harness (`vgpu_vm.py exec`) or the guest +test agent (`tools/guest_agent/`). + +| Suite | What | Pass criterion | +|-------|------|----------------| +| `cuda-samples vectorAdd` | smoke + element-wise throughput | correct result; ≥ 95% B/A | +| `simpleCUBLAS` | SGEMM TFLOPS | ≥ 95% B/A | +| `nvbandwidth` (if available) | H2D/D2H/device copy GB/s | report; DMA path validation | +| `glxgears -info` / a real GL app | frame rate | ≥ 90% B/A | +| `nvenc`/`nvdec` sample | transcode fps | functional (perf later) | +| PyTorch CUDA op (matmul loop) | end-to-end framework | ≥ 90% B/A | + +Always log: guest/host driver versions, GPU clocks (`nvidia-smi -q -d CLOCK` +both sides — throttling skews everything), and the command lines. + +## 5. Where the overhead lives (profile in this order) + +1. **RPC dispatch** — the single biggest lever. Everything the guest driver + does funnels through the emulation thread. +2. **Doorbell polling cadence** — too slow = latency; too hot = host CPU burn. +3. **eventfd/MSI-X delivery** — measure, don't assume. +4. **BAR0 uncached semantics** — shadow is UC for the guest; that's + MMIO-correct but slow for large reads. If the driver bulk-reads shadow, + consider caching-safe sub-regions. +5. **Memory model gap** (ALLOC host-pages vs BAR1 direct VRAM) — see + `docs/GSP_EMULATION_ARCHITECTURE.md` §3. Until unified, DMA-heavy + workloads are not representative. + +## 6. Optimizing safely + +- Change one thing, re-run the A/B pair, keep the numbers with the commit. +- Never optimize off bare-metal-only data: the goal is B/A, not A. +- If a change makes B faster but breaks the §5 validation checklist in + `docs/DEBUGGING.md`, it doesn't count. +- Record results in `STATE.md` (date, versions, B/A ratios) so regressions + are visible across commits. + +## 7. Reporting template + +``` +Date / commit / versions (guest=host=firmware=emulation): +Machine / GPU / clocks locked?: +Test, unit, A (bare), B (vGPU), B/A %: +Notes (what changed vs last run): +``` -- cgit v1.2.3