diff options
| author | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-08 17:04:20 -0400 |
|---|---|---|
| committer | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-08 17:04:20 -0400 |
| commit | 685cbcb8b381a1a9ca2016417fbe7930eb6851c6 (patch) | |
| tree | fc7f53f1c5dcec8088ea16e7cc22fb7cea732a03 /scripts/windows_guest_automation.sh | |
| parent | eea5ea93fee533ab7492fa31f405399ee46a2467 (diff) | |
| parent | eac4cbac7943834487b5876809424eb6c1a2d56f (diff) | |
| download | open-iov-685cbcb8b381a1a9ca2016417fbe7930eb6851c6.tar.gz open-iov-685cbcb8b381a1a9ca2016417fbe7930eb6851c6.zip | |
vgsp: restore reproducible green CUDA (ISSUE-0001) - BAR1 exposure, walk-pin and publish fixes
- Revert m143's BAR1 exposure: report the physical aperture size (256MB)
again. Every post-m143 build reported BAR1=512MB and stalled the guest in
its first cudaMalloc; the m140 green reference reported 256MB. barprobe-r2
(this tree) is green: VADD n=32768 bad=0, STATUS OK ~10s.
- Revert the FORENSICS-grv6-ring walk-authoritative GR ring pin (fix A): it
gated the guest's GR pushes on every boot since (no-push class).
- Restore the end-of-pass GR counter publish with the pass-computed put
(m140 behavior); the H3-only early publish read put=0 from the USERD.
Probe branch: probe/no-walkpin (dc7fc0c3, 8d6254f3, eac4cbac).
Diffstat (limited to 'scripts/windows_guest_automation.sh')
0 files changed, 0 insertions, 0 deletions