Add overlap check between MMIO passthrough regions and MMIO intercept handler regions

[Solicey]

Summary

MMIO passthrough regions (MEM_TYPE_IO, mapped in stage-2 page table) and MMIO intercept handler regions (registered via mmio_region_register()) have no cross-overlap validation. If both occupy overlapping GPA ranges, the handler silently becomes dead code — the stage-2 mapping prevents any fault, so the handler is never invoked.

Background

hvisor uses two mechanisms for guest MMIO:

Mechanism	Registration	Behavior
Passthrough	pt_init() → gpm.insert()	Direct hardware access, no stage-2 fault
Intercept	mmio_region_register() → ZoneInner.mmio	Stage-2 fault triggers handler

Each mechanism checks overlaps internally (gpm checks gpm-vs-gpm; mmio checks exact start-address duplicates), but no cross-check exists between them. The zone creation flow processes them independently:

zone_create()
  pt_init()      → gpm.insert()          (checks gpm-vs-gpm only)
  mmio_init()    → mmio_region_register() (checks exact duplicate only)
  PCI init       → mmio_region_register() (checks exact duplicate only)

Fix

Added bidirectional overlap checks at two insertion points:

1. mmio_region_register() (src/zone.rs) — When registering a new intercept handler, panic if:

• It overlaps with any existing MMIO handler region (not just exact duplicates).
• It overlaps with any passthrough region already in the stage-2 page table.

2. pt_init() (all four architectures) — Before inserting a passthrough region into gpm, check whether it overlaps with any already-registered MMIO handler. Overlap triggers a panic.

New utility methods:

• MMIORegion::is_overlap_with() in src/memory/mmio.rs — range overlap test
• MemorySet::is_range_overlap() in src/memory/mm.rs — public O(log n) query against stage-2 mappings
• ZoneInner::is_mmio_handler_overlap() in src/zone.rs — helper for pt_init checks

On loongarch64, the VIRTIO branch in pt_init() was reordered (mmio_region_register before gpm.insert) to avoid a false positive from the VIRTIO trap page.

Coverage

Scenario	Caught by
Two intercept handler regions overlap	mmio_region_register() handler-vs-handler
Passthrough inserted before intercept handler	mmio_region_register() handler-vs-gpm
Intercept handler registered before passthrough	pt_init() gpm-vs-handler
Same loop: IO entry before VIRTIO entry	mmio_region_register() handler-vs-gpm
Same loop: VIRTIO entry before IO entry	pt_init() gpm-vs-handler

Files Changed

• src/memory/mmio.rs — add MMIORegion::is_overlap_with()
• src/memory/mm.rs — add MemorySet::is_range_overlap()
• src/zone.rs — add checks in mmio_region_register() + is_mmio_handler_overlap() helper
• src/arch/x86_64/zone.rs — add handler overlap check before gpm.insert()
• src/arch/aarch64/zone.rs — same
• src/arch/riscv64/zone.rs — same
• src/arch/loongarch64/zone.rs — same + swap VIRTIO order

Verification

Tested on x86_64 QEMU by adding a VIRTIO region at the same address as the HPET IO passthrough (0xfed0_0000). The fix correctly panics at zone creation:

New MMIO handler region MMIORegion { start: 0xfed00000, size: 0x1000 }
overlaps with passthrough region in stage-2 page table

All four architectures (x86_64, aarch64, riscv64, loongarch64) compile successfully.