Claude review: drm/vblank: timer: Avoid reading the vblank time unnecessarily

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

Patch Review

**Optimization and robustness improvement.** The change from double `drm_crtc_vblank_count_and_time()` to initial `drm_crtc_vblank_count()` + single `drm_crtc_vblank_count_and_time()` per iteration avoids one seqlock read per iteration.

**The retry limit of 10 with `drm_WARN_ON` is reasonable** as a backstop. If the vblank counter changes 10+ times during this read loop, something is fundamentally broken.

**Subtle loop behavior:** The condition `cur_count != lst_count && retries--` means the `retries--` side effect only happens when the count actually changed. If the count is stable on the first try, `retries` is never decremented and `cur_time` is set correctly. This is correct.

**Minor:** `cur_count` is read initially with `drm_crtc_vblank_count()` (which goes through `drm_vblank_count` → atomic64_read), while subsequent reads use `drm_crtc_vblank_count_and_time()` (which goes through the seqlock). The `drm_crtc_vblank_count` path doesn't have the seqlock barrier that `drm_crtc_vblank_count_and_time` has. This means the initial `cur_count` could theoretically be stale relative to the following `*vblank_time = READ_ONCE(vtimer->timer.node.expires)` read. However, since we're comparing with the post-read count, this is caught by the loop — the loop simply runs one extra iteration in that case. No correctness issue, just a theoretical efficiency consideration.

No functional issues in this patch.

---

**Summary of key issues requiring attention:**

1. **Clock domain mismatch (Patches 3, 5):** Replace `ktime_get_ns()` with `ktime_get_boottime_ns()` in `drm_crtc_vblank_start_timer()` and the disabled-timer estimation path, or keep `CLOCK_MONOTONIC` throughout. The comparison between timer expiry and vblank core timestamps also needs reconciliation.
2. **Missing vmwgfx update (Patch 1):** `vmw_vkms_get_vblank_timestamp()` must propagate the new return value.
3. **Division by zero risk (Patch 3):** Guard `roundup(..., vblank->framedur_ns)` against zero denominator.

---
Generated by Claude Code Patch Reviewer