Claude review: udmabuf: Do not create malformed scatterlists

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

Patch Review

**Correctness of the core fix:**

The key change in `get_sg_table()` replaces:
```c
ret = sg_alloc_table(sg, ubuf->pagecount, GFP_KERNEL);
...
for_each_sg(sg->sgl, sgl, ubuf->pagecount, i)
    sg_set_folio(sgl, ubuf->folios[i], PAGE_SIZE, ubuf->offsets[i]);
```
with:
```c
ret = sg_alloc_table_from_pages(sg, ubuf->pages, ubuf->pagecount, 0,
                                ubuf->pagecount << PAGE_SHIFT, GFP_KERNEL);
```

This is correct. `sg_alloc_table_from_pages()` will merge contiguous pages into single scatterlist entries, which is the expected behavior. The old code allocated one SG entry per 4K page and set each individually, which produced a valid-looking but semantically broken scatterlist (no coalescing, and the `sg_set_folio()` offset bug on large folios).

**Data structure change:**

Replacing `struct folio **folios` + `pgoff_t *offsets` with `struct page **pages` is appropriate. The page pointers are obtained via `folio_page()` in `udmabuf_pin_folios()`:
```c
ubuf->pages[upgcnt] = folio_page(folios[cur_folio],
                    subpgoff >> PAGE_SHIFT);
```

This correctly extracts the tail page at the right sub-page offset within the folio. The folio pin tracking remains via the separate `pinned_folios` array, which is unchanged — this is fine since `unpin_folio()` works on the head folio.

**vmap_udmabuf simplification:**

The old code allocated a temporary `pages` array and populated it from `folios[]`+`offsets[]` just to call `vm_map_ram()`. Now it passes `ubuf->pages` directly. This is a nice cleanup that removes an allocation and avoids a potential `-ENOMEM` failure path.

**Comment nit (very minor):**

The comment update at line 31-32 says:
```c
 * Unlike pages, pinned_folios is only used for unpin.
```
The grammar "Unlike pages" is slightly awkward — it could be read as "pages are used for unpin too" when the intent is "pages is used for more than just unpin; pinned_folios is only for unpin." This is a pre-existing comment pattern that was just mechanically updated from "folios" to "pages", so not really an issue introduced by this patch.

**No issues found.** The patch is a clean, correct fix for a real bug with a good Fixes tag and proper attribution.

---
Generated by Claude Code Patch Reviewer