Claude review: drm/vmwgfx: Reserve ttm object before resv usage

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

Patch Review

**Status: Has issues in error handling**

The intent is correct — `vmw_resource_mob_detach()` calls `dma_resv_assert_held()`, so the buffer object's reservation must be held.

**Issue in `vmw_bo_free` (vmwgfx_bo.c)**: If `ttm_bo_reserve()` fails, the code warns but continues to call both `vmw_resource_mob_detach(res)` and `ttm_bo_unreserve(bo)`:
```c
ret = ttm_bo_reserve(bo, false, false, NULL);
if (ret != 0)
    drm_warn(&res->dev_priv->drm,
             "%s: failed reserve\n", __func__);
vmw_resource_mob_detach(res);   /* asserts lock held — will fire */
ttm_bo_unreserve(bo);           /* unreserves without holding reservation */
```

If the reserve fails:
1. `vmw_resource_mob_detach` will hit the `dma_resv_assert_held` assertion — the very thing this patch is meant to fix.
2. `ttm_bo_unreserve` on a non-reserved BO is undefined behavior (likely corrupts the ww_mutex state).

In practice, `ttm_bo_reserve(bo, false, false, NULL)` (non-interruptible, blocking) should essentially never fail, but since v5 explicitly added this check per review feedback, the error path should be correct. A simple approach would be to gate the unreserve:
```c
ret = ttm_bo_reserve(bo, false, false, NULL);
if (ret != 0)
    drm_warn(...);
vmw_resource_mob_detach(res);
if (ret == 0)
    ttm_bo_unreserve(bo);
```
Or skip the detach on failure if that's safe for the destroy path.

**The `crc_generate_worker` change (vmwgfx_vkms.c)** looks correct — the `goto done` properly skips `compute_crc` and falls through to `vmw_surface_unreference(&surf)`.

**Missing tags**: This patch is also missing `Fixes:` and `Cc: stable` tags despite fixing lockdep/dma_resv assertion warnings.

---

---
Generated by Claude Code Patch Reviewer