Claude review: iommu: introduce iova_to_phys_length in iommu_domain_ops

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

The core infrastructure patch. The v1 version uses 0 as error return and the v2 switches to PHYS_ADDR_MAX -- I'll review the v1 as submitted.

**v1 issues:**

The IDENTITY domain case doesn't set `*mapped_length`:
```c
if (domain->type == IOMMU_DOMAIN_IDENTITY)
    return iova;
```
If a caller passes a non-NULL `mapped_length` and gets an IDENTITY domain, `mapped_length` stays 0 (from the initialization at the top). This is arguably fine since IDENTITY domains shouldn't be used in the VFIO/iommufd paths, but it's an inconsistency -- the caller cannot distinguish "no translation" from "identity" by looking at `mapped_length`.

The BLOCKED domain check was removed in v1 but the fallback handles it implicitly (both `iova_to_phys_length` and `iova_to_phys` would be NULL for BLOCKED domains, falling through to `return 0`). This is fine.

**v2 improvement**: The v2 version checks `!domain->ops->iova_to_phys_length` instead of `domain->type`, which is cleaner since it handles BLOCKED domains naturally and doesn't need to know about domain type internals. However, the v2 fallback for legacy `iova_to_phys` returns `PHYS_ADDR_MAX` when `phys == 0`, which maps "no translation" to "error" -- correct but worth documenting clearly.

---
Generated by Claude Code Patch Reviewer