Claude review: drm/vmwgfx: Convert to DRM vblank timers

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

Patch Review

**Purpose:** The actual conversion — removes custom hrtimer code and wires up the DRM vblank timer infrastructure.

**Assessment: Looks good overall, with one style nit.**

**Key changes reviewed:**

1. **`vmwgfx_kms.h`**: Removal of `hrtimer timer` and `ktime_t period_ns` from the `vkms` struct in `vmw_display_unit` — correct, these are now managed by the DRM vblank timer infrastructure inside `drm_vblank_crtc`.

2. **LDU/SCRN/STDU helper_funcs**: All three display backends wire up `.handle_vblank_timeout = vmw_vkms_handle_vblank_timeout`. This is the DRM callback that gets invoked by the shared timer.

3. **`vmw_vkms_get_vblank_timestamp()`**: Replaced the manual timestamp calculation with `drm_crtc_vblank_get_vblank_timeout(crtc, vblank_time)`. The shared implementation in `drm_vblank.c:2285` handles the `!vblank->enabled` case, the race-safe count+time read loop, and the one-frame correction — all of which were done (less robustly) inline before. This is a clear improvement.

4. **`vmw_vkms_enable_vblank()`**: Replaced `hrtimer_setup()` + `hrtimer_start()` with `drm_crtc_vblank_start_timer(crtc)`. The shared function calls `drm_calc_timestamping_constants()` internally, so the explicit call is correctly removed.

5. **`vmw_vkms_disable_vblank()`**: Replaced `hrtimer_cancel()` with `drm_crtc_vblank_cancel_timer()`. Note: the shared cancel uses `hrtimer_try_to_cancel()` (non-blocking) plus an interval-zeroing approach to avoid deadlocks, which is actually better than the old `hrtimer_cancel()` (blocking). The `du->vkms.period_ns = ktime_set(0, 0)` clear is correctly removed since the period is now tracked internally.

6. **`vmw_vkms_crtc_cleanup()`**: The v2 fix: only cancels the timer when `vmw->vkms_enabled` is true. This prevents calling into the vblank timer infrastructure when it was never initialized. Good fix addressing Zack's review feedback.

7. **`vmwgfx_vkms.h`**: Removal of `#include <linux/hrtimer_types.h>` — correct, no longer needed.

**Style nit:** The `.handle_vblank_timeout` lines in ldu/scrn/stdu use double space before `=`:
```c
.handle_vblank_timeout  = vmw_vkms_handle_vblank_timeout,
```
This appears to be intentional alignment with the other fields, but the surrounding fields use single-space (e.g., `.atomic_flush = ...`). This is very minor and not worth a respin.

**No functional issues found.**

---

**Summary:** Series is well-structured, the conversion is correct, and the v2 fixes address the previous review feedback. The shared DRM vblank timer provides better race handling in the timestamp calculation than the old open-coded version. Reviewed-by worthy.

---
Generated by Claude Code Patch Reviewer