github/kata-containers
1
0
Fork 2
mirror of https://github.com/kata-containers/kata-containers.git synced 2026-07-26 07:30:19 +00:00
Code Issues Projects Releases Wiki Activity

hypervisor/qemu: document #13270 (4+ Blackwell GPU GPA cap) in known issues

Browse Source 
Explains the root cause (40-bit GPA without host-phys-bits=on), the
short-term fix in the Makefile, and the post-refactor path (Q35 Platform
builder should emit host-phys-bits=on unconditionally for x86_64/KVM).

Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Assisted-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Zvonko Kaiser
2026-07-14 01:00:20 +00:00
parent cdd81c7a42
commit c28ea4eace
1 changed files with 17 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
src/runtime-rs/crates/hypervisor/src/qemu/ARCHITECTURE.md
View File

@@ -775,6 +775,23 @@ are related to this refactor. Items marked **post-refactor** require the new
`Platform` emission path to be wired end-to-end before they can be addressed
cleanly.
### 4+ Blackwell (B200/B300) GPU passthrough fails due to 40-bit GPA cap (#13270)
`-cpu host` without `host-phys-bits=on` limits the guest physical address space
to ~40 bits (~1 TiB). Each B200/B300 GPU has a 128 GiB 64-bit prefetchable BAR;
three GPUs (384 GiB) fit within 1 TiB, but four GPUs (512 GiB) do not, causing
the fourth GPU to fail PCIe BAR assignment at VM boot.
**Fix:** set `cpu_features = "pmu=off,host-phys-bits=on"` in
`configuration-qemu-nvidia-gpu*.toml.in`. The `Makefile` now defaults
`CPUFEATURES` and `TDXCPUFEATURES` to `pmu=off,host-phys-bits=on` for `amd64`
builds so operator installs that regenerate configs from source pick up the fix;
existing installed configs must be updated manually.
**Post-refactor:** the `Platform` Q35 builder should emit `host-phys-bits=on`
unconditionally for x86_64/KVM so the fix is durable even if the config file
is overwritten by `kata-deploy`.
### QMP startup timeout is hard-coded and ignores the caller's deadline (#13343)
`QemuInner::start_vm(_timeout)` accepts the timeout argument from the