[PATCH v2 2/6] mm/sparse-vmemmap: support device DAX in common vmemmap path

[Muchun Song]

The common vmemmap population path cannot yet handle optimized Device DAX
mappings on its own. It uses pfn_to_zone() to find the shared tail page,
but Device DAX populates its vmemmap at runtime before the ZONE_DEVICE span
is initialized.

Teach the common path to use device_zone() for runtime optimized vmemmap
population while retaining pfn_to_zone() for early boot. This allows the
same path to support both early boot mappings and Device DAX.

The backing PFN supplied by the Device DAX-specific population path is no
longer used, allowing the redundant lookup and population code to be
removed later.

Signed-off-by: Muchun Song <songmuchun@xxxxxxxxxxxxx>
---
v2:
- Expand comments around slab initialization to explain zone lookup and
  page refcounting (suggested by Qi Zheng)
---
 mm/sparse-vmemmap.c | 56 +++++++++++++++++++++++++--------------------
 1 file changed, 31 insertions(+), 25 deletions(-)

diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 919c94a36346..cdcc93714a9a 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -208,18 +208,39 @@ static __meminit void *vmemmap_alloc_pte(unsigned long pfn, int node,
 	struct page *page;
 	const unsigned int order = pfn_to_section_compound_order(pfn);
 
-	/*
-	 * Device DAX still relies on vmemmap_populate_compound_pages() for
-	 * head/first-tail allocation and tail-page reuse.
-	 */
 	if (!vmemmap_optimizable_pfn(pfn))
 		return vmemmap_alloc_block_buf(PAGE_SIZE, node, altmap);
 
-	zone = pfn_to_zone(pfn, node);
+	/*
+	 * Before slab is available, vmemmap optimization is used for early
+	 * system RAM, whose zone can be determined from the PFN.
+	 *
+	 * Once slab is available, only ZONE_DEVICE memory reaches this
+	 * optimized population path. Its zone span has not been initialized
+	 * while its vmemmap is being populated, so pfn_to_zone() cannot be
+	 * used. Obtain ZONE_DEVICE directly from the node instead.
+	 */
+	zone = slab_is_available() ? device_zone(node) : pfn_to_zone(pfn, node);
 	page = vmemmap_shared_tail_page(order, zone);
 	if (!page)
 		return NULL;
 
+	/*
+	 * During early vmemmap population, the shared tail vmemmap backing
+	 * page is allocated from memblock before its struct page can safely
+	 * participate in page refcounting. Therefore, no reference can be
+	 * held for each shared PTE mapping, and the mappings must be unshared
+	 * before the vmemmap is depopulated.
+	 *
+	 * Once slab is available, the shared backing page is allocated from
+	 * the buddy allocator and can be refcounted. Hold one reference for
+	 * each shared PTE mapping. The architecture vmemmap teardown drops
+	 * the reference through __free_pages() when removing the mapping,
+	 * preventing the backing page from being freed while it is shared.
+	 */
+	if (slab_is_available())
+		get_page(page);
+
 	return page_address(page);
 }
 
@@ -231,27 +252,12 @@ static pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, in
 
 	if (pte_none(ptep_get(pte))) {
 		pte_t entry;
+		void *p = vmemmap_alloc_pte(pfn, node, altmap);
 
-		if (ptpfn == (unsigned long)-1) {
-			void *p = vmemmap_alloc_pte(pfn, node, altmap);
-
-			if (!p)
-				return NULL;
-			ptpfn = PHYS_PFN(__pa(p));
-		} else {
-			/*
-			 * When a PTE/PMD entry is freed from the init_mm
-			 * there's a free_pages() call to this page allocated
-			 * above. Thus this get_page() is paired with the
-			 * put_page_testzero() on the freeing path.
-			 * This can only called by certain ZONE_DEVICE path,
-			 * and through vmemmap_populate_compound_pages() when
-			 * slab is available.
-			 */
-			if (slab_is_available())
-				get_page(pfn_to_page(ptpfn));
-		}
-		entry = pfn_pte(ptpfn, PAGE_KERNEL);
+		if (!p)
+			return NULL;
+
+		entry = pfn_pte(PHYS_PFN(__pa(p)), PAGE_KERNEL);
 		set_pte_at(&init_mm, addr, pte, entry);
 	} else if (WARN_ON_ONCE(vmemmap_optimizable_pfn(pfn)))
 		return NULL;
-- 
2.54.0