Claude review: drm/amdgpu: use bitmap_clear() in amdgpu_amdkfd_device_init()

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

Patch Review

**Correctness:**

The transformation from:
```c
for (i = last_valid_bit; i < AMDGPU_MAX_QUEUES; ++i)
    clear_bit(i, gpu_resources.cp_queue_bitmap);
```
to:
```c
bitmap_clear(gpu_resources.cp_queue_bitmap, last_valid_bit,
             AMDGPU_MAX_QUEUES - last_valid_bit);
```
is semantically correct. Both clear the same range of bits `[last_valid_bit, AMDGPU_MAX_QUEUES)`.

**Atomicity change is safe:** The old code used `clear_bit()` (atomic), while `bitmap_clear()` uses `__bitmap_clear()` (non-atomic) for the runtime-nbits path. This is fine because `gpu_resources` is a local variable on the stack — it was just initialized a few lines above and no other thread can observe it. There is no need for atomic operations on stack-local data.

**Removal of misleading comment:** Correct to remove. The comment:
```c
/* According to linux/bitmap.h we shouldn't use bitmap_clear if
 * nbits is not compile time constant
 */
```
was wrong. `bitmap_clear()` at `include/linux/bitmap.h:547` handles non-constant `nbits` via `__bitmap_clear()`.

**Removal of `int i`:** Clean — the loop variable is no longer needed, and `last_valid_bit` is the only remaining local.

**Alignment nit (cosmetic only):** The continuation line aligns `AMDGPU_MAX_QUEUES - last_valid_bit` with a tab+spaces to match the opening paren, which results in slightly unusual indentation but is consistent with the surrounding code style (e.g., the `bitmap_complement` call just above uses the same pattern).

**No functional concerns.** This is a clean, correct simplification.

**Reviewed-by recommendation: Yes.**

---
Generated by Claude Code Patch Reviewer