diff options
| -rw-r--r-- | STATE-CURRENT.md | 29 |
1 files changed, 22 insertions, 7 deletions
diff --git a/STATE-CURRENT.md b/STATE-CURRENT.md index 61d7217..2f11b57 100644 --- a/STATE-CURRENT.md +++ b/STATE-CURRENT.md @@ -1,6 +1,6 @@ # Open-IOV Current State -> **Snapshot: 2026-10-01 15:45.** This is a dated public status page, not live +> **Snapshot: 2026-10-01 15:47.** 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 known-knowns/unknowns document > is authoritative here: @@ -43,8 +43,14 @@ meant to execute on the physical GPU. populated from live values. Init now reaches the server's **static-config collection** and fails there (`0x59`); the device-id override key was exercised with no change, and a display-disable experiment showed display is - not the first error. Each of cycles 4-10 released the device cleanly with no - reboot. | + not the first error. A source-level review of the failing collector has now + closed the host side: it reads none of the remaining host-supplied fields + (the one registry key on its call graph can only clear an error), so the + ACPI/workspace/zeroed-field leads are retired. The one hard dependency left + on the failing path is a **PMU-provisioned encrypted device identity** (no + message-queue field or host path supplies it), so the next leads are the + PMU initialization/ucode stage and deeper server-log visibility. Each of + cycles 4-10 released the device cleanly with no reboot. | | Graphics (3D), video | ❌ | | Faults, multi-version, Windows, migration, multi-tenancy, performance | ❌ | @@ -107,10 +113,19 @@ chain succeeds. a display-disable experiment showed display is not the first error. The collector's failing step is not visible in the server journal (its log ring is enabled only after the collector, and the protected region blocks host - reads). Next leads: enumerate the collector's registry reads (in flight), - port the ACPI method-data path if indicated, and gain deeper server - visibility (enable the log ring; a server memory read if the protected - region allows). + reads). A source-level review has now **closed the host-side input + campaign**: the collector reads none of the remaining host-supplied fields, + all earlier leads (device-id override, display key, ACPI method data, + video-memory workspace branch, zeroed fields) are exercised or retired, and + the only hard dependency on the failing path is a server-PMU-provisioned + encrypted chip identity that no message-queue field or host RPC supplies. + Next leads, ranked: (1) the PMU initialization/ucode stage - determine + whether a host-driven boot must stage it or the server provisions it + itself; (2) server-log visibility - repair the resource-manager log ring + mapping (enabled only after the collector) so the server's first error + becomes readable, with a memory-read fallback if the protected region + allows; (3) if neither, treat it as a server-side firmware-instrumentation + problem and record it as such rather than adding more host fields. - **The full CUDA driver-API smoke test passes with correct data**: the client completes `cuCtxCreate -> cuMemAlloc -> cuMemcpyHtoD -> cuMemcpyDtoH -> cuMemFree -> cuCtxDestroy` and prints `ALL OK`, with the 64-byte payload |