Claude review: Revert "accel/amdxdna: Support read-only user-pointer BO mappings"

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

Patch Review

**Commit message issues:**

The message "The read-only feature requires further consideration" is insufficiently detailed for a revert. It should explain the specific problems. The original `readonly_va_entry()` function had at least two notable issues:

1. **Race condition / TOCTOU:** The function checks VMA flags (`vm_flags & VM_WRITE`) under `mmap_read_lock`, but by the time `pin_user_pages_fast()` is called (which does its own VMA lookup internally), the VMA could have changed — it could have been split, unmapped, or had its permissions changed via `mprotect()`. This is a classic time-of-check-to-time-of-use race. The `pin_user_pages_fast` GUP path already handles write permission checking internally when `FOLL_WRITE` is (or isn't) passed, so duplicating that logic at a different point in time is fragile.

2. **Spanning multiple VMAs:** The check `vma->vm_end - va_ent->vaddr < va_ent->len` assumes the entire range is covered by a single VMA, but a user-pointer range could span multiple VMAs with different permissions. `find_vma()` only returns the first one. This means the read-only detection could be incorrect for multi-VMA ranges.

These would be good things to mention in the commit message.

**Code changes — correctness:**

The diff itself is a clean revert:

- Removes `readonly_va_entry()` entirely — correct.
- Removes `bool readonly = true;` declaration — correct.
- Removes the `readonly_va_entry()` call in the validation loop — correct.
- Changes `(readonly ? 0 : FOLL_WRITE) | FOLL_LONGTERM` to `FOLL_WRITE | FOLL_LONGTERM` — correct, always pins as writable.
- Changes `(readonly ? O_RDONLY : O_RDWR) | O_CLOEXEC` to `O_RDWR | O_CLOEXEC` — correct, always exports as read-write.

The post-revert state matches what existed before `f649e63d4a64` was applied.

**Minor nit:** The revert message doesn't include a `Cc: stable@` tag. If the original commit went into a released kernel and the TOCTOU race is considered a security-relevant bug, a stable backport annotation might be appropriate. But this is at the maintainer's discretion.

**Recommendation:** Request an improved commit message that explains the specific technical concerns motivating the revert (TOCTOU race, single-VMA assumption, or whatever the actual driver is). The code change itself is correct and can be accepted as-is once the message is improved.

---
Generated by Claude Code Patch Reviewer