From 31594b263cc4fbe30ebbe412a4a788f726f03a02 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Sun, 4 Oct 2026 16:19:31 -0400 Subject: vgsp: end-to-end CUDA acceptance in the guest The no-claim emulation path now runs a stock CUDA program end to end: device properties, cudaMalloc, a GPU-authored string kernel read back byte-for-byte, two 4 MB host-to-device copies, a 1 M-element vectorAdd executed on the physical GPU through the host bridge, and the device-to-host copy-back verified element by element. Three consecutive supervised cycles report MAIL + VADD bad=0 checksum=392344090.500000 + STATUS OK with identical input/output hashes. Two data-path correctness fixes make the transfers sound: - Once-only op execution per GPFIFO entry. The pump replays recently consumed ring entries when translation batches are injected so missed releases can land; re-executing a CE copy is not idempotent because its source is a client staging buffer that may have been refilled. Slots now record their raw entry words and a replayed entry matching that identity suppresses operations while still processing releases. - Per-page translation of virtual operands. svc_copy resolved only the first page of a VA range and walked the PA linearly; client staging ranges are frequently split (observed 195 + 61 pages), so the tail of alternating chunks was read from unrelated memory. Multi-page copies now re-resolve each 4 KB page through the trusted resolver. Also carries the earlier uncommitted emulation fixes that got the client to the launch (QMD address-window gate, counter registration persistence, 512 MiB emulated FB, bridge slot hygiene) and bounded diagnostics (sema-bigop, sema-bigcopy, bridge-in/out hashes, daemon INCHECK/OUTCHECK/dumps). --- docs/04-implementation/01-known-knowns-and-unknowns.md | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'docs/04-implementation') diff --git a/docs/04-implementation/01-known-knowns-and-unknowns.md b/docs/04-implementation/01-known-knowns-and-unknowns.md index 0c8e50f..084afc8 100644 --- a/docs/04-implementation/01-known-knowns-and-unknowns.md +++ b/docs/04-implementation/01-known-knowns-and-unknowns.md @@ -67,6 +67,18 @@ listed as an unknown-unknown has a *method* by which we will discover it. * The client's page-table contract for server-owned ranges is explicit (`COPY_SERVER_RESERVED_PDES`) — and its leaf levels are *empty*, i.e. ours to populate. +* **End-to-end CUDA works in emulation mode.** A stock CUDA program in the + guest (device properties, `cudaMalloc`, a GPU-authored string kernel, two + 4 MB host-to-device copies, a 1 M-element `vectorAdd`, and the + device-to-host copy-back) completes with `bad=0` against the host-computed + reference and the expected checksum. The host-side bridge executes the + client's kernels on the physical GPU, and the emulated CE path stages the + data in and out. Two correctness invariants are required and verified: + a decoded copy executes at most once per GPFIFO entry identity, and every + virtual operand is translated per 4 KB page (see + [`../03-architecture/06-host-module-and-hybrid-model.md`](../03-architecture/06-host-module-and-hybrid-model.md) §4.4). + The acceptance is deterministic across consecutive supervised cycles + (identical input/output hashes and checksum). ### 1.4 Operational -- cgit v1.2.3