[1/3] aarch64: PCIe host-bridge (ECAM) transport

[maxandrews]

[1/3] aarch64: PCIe host-bridge (ECAM) transport

Introduces a feature-gated PCIe host bridge so an aarch64 libkrun guest can
enumerate a PCI Express bus. This is the foundation for VFIO device/GPU
passthrough, which lands in the two follow-up PRs of this stack.

What's in it

• Ports Cloud Hypervisor's pci crate — ECAM (memory-mapped config) only, since
  aarch64 has no port I/O — adapted to libkrun's BusDevice model (which has no
  cross-vCPU Barrier return): PciConfiguration, PciBus, PciConfigMmio
  (ECAM), PciRoot host bridge, and the PciDevice trait.
• fdt/aarch64.rs: emits a pci-host-ecam-generic node (ECAM reg + 32/64-bit
  ranges).
• Registers the ECAM window on the MMIO bus + attaches the host bridge at 00:00.0.
• New pci cargo feature (fanned out to msb_krun + vmm/devices), off by
  default — no change to existing builds.

Result: with --features pci, the guest enumerates an initially-empty PCIe bus
(host bridge at 0000:00:00.0, ECAM at the aperture defined in layout.rs).

Commits (5)

1. aarch64: PCIe host-bridge memory-map constants (ECAM + 32/64-bit apertures)
2. pci: port CH PCIe host-bridge (ECAM config, PciBus, PciConfiguration)
3. pci: wire PCIe host bridge into aarch64 VMM (FDT node + ECAM on MMIO bus)
4. pci: silence dead-code warning on DeviceIdState::Reserved
5. pci: expose pci/vfio features on the msb_krun crate (src/krun)

Testing

Builds green with the pci feature; guest boots and enumerates the empty bus.

Stacked series (3 PRs, all against krun, reviewed/merged in order)

1. this PR — PCIe host-bridge / ECAM transport
2. VFIO device passthrough (config / BARs / DMA / KVM bind) — next
3. MSI-X interrupt delivery + guest-driver enablement (→ nvidia-smi on a T4G)

Note: a historically-adjacent virtio-gpu Venus blob-mapping fix that sat at
the old branch base is excluded here — it's a separate track on
feat/gpu-venus-mapping-fixes. This series is rebased to contain only the
passthrough work.

🤖 Generated with Claude Code