| |
|
|
| |
GSP_BRIDGE_INLINE_BYTES/SVC_GR_INLINE_BYTES/GSP_PHYS_GR_INLINE_BYTES 256->2048 (+static_asserts; svc_gl_flush prs[] static); svc_grce_dst_newest harvests window's REAL captured upload from svc_grce_dst_ring, svc_gl_flush attaches real 1536B@0x4350900 + 1408B@0x4350f00 (w0=0x020a0481) as inline records at own dst VAs (page-crossing split, tail page candidate); svc_grce_inline_write dual-writes engine-visible injected-pin page (svc_grce_land_write page-clamped); SPH-BIND also emits SET_PIPELINE_SHADER(2)<-0x21/(5)<-0x51 - delivery WORKS: real-upload record va=0x4350900 len=1536 first=020a0481 (engine-visible delivery), INLINE-APPLY guest=1, INLINE-RB rb=020a0481 pa=0x64cfd6000, page[7] va=0x4350000 hpa=0x64cfd6000 (r1283); wall survives real image (9/11 + ESR 0xa0000a00 +14ms; V4 window still 9/11); bind+type+real-image combo never seen on executing launch (lane wedges 1-3 launches); stores 0/900-901; next: healthy boot >=3 bind-carrying launches then harvest client init-push pipeline state (20010810 00000011 SET_PIPELINE_SHADER(1)<-0x11 + 0x204c pair) into SPH-BIND prelude - SESSION 153 r1282-r1286
|
|
|
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).
|
