diff options
Diffstat (limited to 'STATE-CURRENT.md')
| -rw-r--r-- | STATE-CURRENT.md | 59 |
1 files changed, 32 insertions, 27 deletions
diff --git a/STATE-CURRENT.md b/STATE-CURRENT.md index 0289f77..2eb02d61 100644 --- a/STATE-CURRENT.md +++ b/STATE-CURRENT.md @@ -1,32 +1,37 @@ # Open-IOV Current State -> **Snapshot: 2026-10-24 ~18:45 (the r632 first-client hang is -> root-caused and fixed — a held GR launch's push-level release records are -> per-decode state and the two hold paths dropped them at the pass -> boundary, so the eventual forward engaged the release hold empty and the -> client's tracking word never advanced; r633 parks the records with the -> launch and delivers them on the forward. Mail is green on two consecutive -> fresh boots (`VADD bad=0 checksum=12204516.000000`), and the first -> zero-failure battery landed (r633b: 9 pass, 0 fail, 1 hang — streams2 -> mid-battery). The release-target question is answered: the -> `release0 could not target the guest page` redirect is the deliberate -> `gsp_phys_rmcall_gr_rel_mod=1` default, and the release page resolves -> (`rel_hpa=0x2fc495000`, `map_ret=0`); a live probe module confirms the -> GPA->HPA path. The r632 synthetic shadow caps are gated (anchor + -> per-episode budgets; shadow-table peak 2048/8192 vs 8192/8192 full) and -> the r631 math red draw is green (`badf=0 badd=0 badi=0`). The remaining -> top blocker is the deterministic wrongness class (streams2 / -> streams_events / memset2d / roundtrip) — traced to the per-client operand -> resolution shortfall: when the guest scanner's kcore walks miss a -> client's device VAs (`sems=1/7`), the fill drops its argument bases and -> the engine runs with zero operands. The mid-battery hangs (reduce, -> streams2) share that same resolution variable. GL runs on the guest GPU -> (next wall = the FBO class); the Looking Glass frame path is proven -> end-to-end (guest CUDA render -> KVMFR ivshmem -> host PNG frame); two -> guests with 512M + 1GiB emulated VRAM coexist at boot level; the 0.9.9 -> release-preview demo video v1 is recorded and synced to files.vgpu.io. -> Frontier = operand resolution + wrongness, then 15/15 + soak, events A/B, -> GZDOOM (GL/Vulkan), two-guest concurrent CUDA, Windows guests).** +> **Snapshot: 2026-10-24 ~20:40 (the host-side operand-resolution gap +> is closed and the shadow anchor is content-gated — r635 tracks REAL +> written content per shadow page (`r->content` on every landed write) +> and uses it as the operand anchor: a client's own H2D bytes that took +> the shadow lane anchor their span and arm a launch-level content +> anchor, so an answerless output candidate (the buffer the client never +> wrote) is served from a bounded synthetic page set (1024/candidate, +> 4096/launch, three-quarter-table guard) instead of failing the launch +> closed; trusted pin/walk/ledger answers keep the r633 budgets and no +> longer arm the cross mint, and a synthetic page can never re-anchor. +> The first draft of the change self-fed on its own minted pages and +> filled the shared shadow table on a green boot (6144/8192 measured, +> later clients crawled behind it); the content gate restored zero +> minted pages on a green boot and the historical green shape. Fresh +> r636 boot battery: 9 OK / 2 clean FAIL / 1 HANG — the best pass count +> repeated with strictly better failure modes: streams_events converted +> its historical hang into a clean client-visible `CUDA_ERROR` via the +> auto-fault, and streams2's red chain is now measured end-to-end (its +> H2D destinations are replay-denied, the GR forward defers on the CE +> stream-order watermark, the launch is refused as degenerate and the +> refusal faults cleanly; `STREAMS n=32768 bad=65534`). Remaining: the +> streams2 CE-order chain (a copy-pipeline class, not the fill), the +> first-wait crawl class (smiley/memset2d spin in libcuda before +> submission), used-boot allocation exhaustion (roundtrip/inplace_scale +> `CUDA_ERROR out of memory` after thirteen clients), then 15/15 + soak. +> GL runs on the guest GPU (next wall = the FBO class); the Looking +> Glass frame path is proven end-to-end (guest CUDA render -> KVMFR +> ivshmem -> host PNG frame); two guests with 512M + 1GiB emulated VRAM +> coexist at boot level; the 0.9.9 release-preview demo video v1 is +> recorded and synced to files.vgpu.io. Frontier = 15/15 + multi-round +> soak, events A/B, GZDOOM (GL/Vulkan), two-guest concurrent CUDA, +> longevity/video v2, Windows guests).** > This is a dated public > status page, not live state; it is refreshed on every progress check. Live > working notes are maintained outside this repository; the public |