Claude review: accel/ethosu: reject NPU_OP_RESIZE commands from userspace

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

Patch Review

**Verdict: Correct fix, one minor style observation.**

The change replaces:
```c
case NPU_OP_RESIZE: // U85 only
    WARN_ON(1); // TODO
    break;
```
with:
```c
case NPU_OP_RESIZE: // U85 only
    return -EINVAL;
```

This is the right approach. The `WARN_ON(1)` was unconditionally user-triggerable from unprivileged userspace via `DRM_IOCTL_ETHOSU_GEM_CREATE`, making it a DoS vector (kernel log spam, or panic if `panic_on_warn` is set). Returning `-EINVAL` properly rejects the command at the ioctl level.

**Minor observation:** The `default:` case at line 650 silently ignores unknown opcodes (`break` with no error). This means `NPU_OP_RESIZE` is now treated *more strictly* than completely unknown/garbage opcodes — it returns an error, while unknown commands are silently accepted and passed through. This is arguably the right behavior for a known-but-unimplemented opcode, but the maintainer may want to consider whether `default:` should also return `-EINVAL` in a follow-up, since silently ignoring unrecognized commands seems fragile for a security-sensitive parser.

**No issues with fall-through:** The `return -EINVAL` exits the function entirely, so the absence of `break` before `case NPU_SET_KERNEL_WIDTH_M1:` is correct — there's no fall-through path.

---
Generated by Claude Code Patch Reviewer