r679: 0x2080a097 handler with runtime-tunable reply words (wall-2 session 75)

The guest Vulkan client (libnvidia-eglcore) issues RM control 0x2080a097 in a
~1.8 s retry loop while its engine-readiness check stays negative; the
bare-metal oracle (same client on the host with the real driver) never issues
it, so it is a vGPU-only path (engine type 0xe00101). The handler echoes the
1168-byte in/out record block and applies three module parameters -
wall2_a097_flag/v488/v48c - so one boot can sweep reply variants without
rebuilding (the client reads params +0x204, +0x488, +0x48c back; file offsets
0xecd45f..0xecd475 in libnvidia-eglcore.so.570.144).

Sweep result so far: flag=1 with v488=0 or 1 does not stop the retry; the
readiness wait lives in the client's engine loop (0xd6ef40..0xd6f0ab polling
byte +0x1b038 of the engine object, recomputed by a91ea0 from engine-state
fields), so the missing state is deeper than the reply words.

0 files changed, 0 insertions, 0 deletions