This brings the iommu page table allocator into the modern world of having
its own private page descriptor and not re-using fields from struct page
for its own purpose. It follows the basic pattern of struct ptdesc which
did this transformation for the CPU page table allocator.
Currently iommu-pages is pretty basic so this isn't a huge benefit,
however I see a coming need for features that CPU allocator has, like sub
PAGE_SIZE allocations, and RCU freeing. This provides the base
infrastructure to implement those cleanly.
Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
---
drivers/iommu/iommu-pages.c | 54 ++++++++++++++++++++++++++-----------
drivers/iommu/iommu-pages.h | 39 ++++++++++++++++++++++++++-
2 files changed, 76 insertions(+), 17 deletions(-)
diff --git a/drivers/iommu/iommu-pages.c b/drivers/iommu/iommu-pages.c
index af86c487ba100b..f7def716997af4 100644
--- a/drivers/iommu/iommu-pages.c
+++ b/drivers/iommu/iommu-pages.c
@@ -7,6 +7,21 @@
#include <linux/gfp.h>
#include <linux/mm.h>
+#define IOPTDESC_MATCH(pg_elm, elm) \
+ static_assert(offsetof(struct page, pg_elm) == \
+ offsetof(struct ioptdesc, elm))
+IOPTDESC_MATCH(flags, __page_flags);
+IOPTDESC_MATCH(lru, iopt_freelist_elm); /* Ensure bit 0 is clear */
+IOPTDESC_MATCH(mapping, __page_mapping);
+IOPTDESC_MATCH(private, _private);
+IOPTDESC_MATCH(page_type, __page_type);
+IOPTDESC_MATCH(_refcount, __page_refcount);
+#ifdef CONFIG_MEMCG
+IOPTDESC_MATCH(memcg_data, memcg_data);
+#endif
+#undef IOPTDESC_MATCH
+static_assert(sizeof(struct ioptdesc) <= sizeof(struct page));
+
/**
* iommu_alloc_pages_node - Allocate a zeroed page of a given order from
* specific NUMA node
@@ -20,10 +35,17 @@
void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order)
{
const unsigned long pgcnt = 1UL << order;
- struct page *page;
+ struct folio *folio;
- page = alloc_pages_node(nid, gfp | __GFP_ZERO | __GFP_COMP, order);
- if (unlikely(!page))
+ /*
+ * __folio_alloc_node() does not handle NUMA_NO_NODE like
+ * alloc_pages_node() did.
+ */
+ if (nid == NUMA_NO_NODE)
+ nid = numa_mem_id();
+
+ folio = __folio_alloc_node(gfp | __GFP_ZERO, order, nid);
+ if (unlikely(!folio))
return NULL;
/*
@@ -35,21 +57,21 @@ void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order)
* This is necessary for the proper accounting as IOMMU state can be
* rather large, i.e. multiple gigabytes in size.
*/
- mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, pgcnt);
- mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, pgcnt);
+ mod_node_page_state(folio_pgdat(folio), NR_IOMMU_PAGES, pgcnt);
+ lruvec_stat_mod_folio(folio, NR_SECONDARY_PAGETABLE, pgcnt);
- return page_address(page);
+ return folio_address(folio);
}
EXPORT_SYMBOL_GPL(iommu_alloc_pages_node);
-static void __iommu_free_page(struct page *page)
+static void __iommu_free_page(struct ioptdesc *iopt)
{
- unsigned int order = folio_order(page_folio(page));
- const unsigned long pgcnt = 1UL << order;
+ struct folio *folio = ioptdesc_folio(iopt);
+ const unsigned long pgcnt = 1UL << folio_order(folio);
- mod_node_page_state(page_pgdat(page), NR_IOMMU_PAGES, -pgcnt);
- mod_lruvec_page_state(page, NR_SECONDARY_PAGETABLE, -pgcnt);
- put_page(page);
+ mod_node_page_state(folio_pgdat(folio), NR_IOMMU_PAGES, -pgcnt);
+ lruvec_stat_mod_folio(folio, NR_SECONDARY_PAGETABLE, -pgcnt);
+ folio_put(folio);
}
/**
@@ -62,7 +84,7 @@ void iommu_free_page(void *virt)
{
if (!virt)
return;
- __iommu_free_page(virt_to_page(virt));
+ __iommu_free_page(virt_to_ioptdesc(virt));
}
EXPORT_SYMBOL_GPL(iommu_free_page);
@@ -74,9 +96,9 @@ EXPORT_SYMBOL_GPL(iommu_free_page);
*/
void iommu_put_pages_list(struct iommu_pages_list *list)
{
- struct page *p, *tmp;
+ struct ioptdesc *iopt, *tmp;
- list_for_each_entry_safe(p, tmp, &list->pages, lru)
- __iommu_free_page(p);
+ list_for_each_entry_safe(iopt, tmp, &list->pages, iopt_freelist_elm)
+ __iommu_free_page(iopt);
}
EXPORT_SYMBOL_GPL(iommu_put_pages_list);
diff --git a/drivers/iommu/iommu-pages.h b/drivers/iommu/iommu-pages.h
index d35d9f99f320dc..f36103f591e91b 100644
--- a/drivers/iommu/iommu-pages.h
+++ b/drivers/iommu/iommu-pages.h
@@ -9,6 +9,43 @@
#include <linux/iommu.h>
+/**
+ * struct ioptdesc - Memory descriptor for IOMMU page tables
+ * @iopt_freelist_elm: List element for a struct iommu_pages_list
+ *
+ * This struct overlays struct page for now. Do not modify without a good
+ * understanding of the issues.
+ */
+struct ioptdesc {
+ unsigned long __page_flags;
+
+ struct list_head iopt_freelist_elm;
+ unsigned long __page_mapping;
+ pgoff_t __index;
+ void *_private;
+
+ unsigned int __page_type;
+ atomic_t __page_refcount;
+#ifdef CONFIG_MEMCG
+ unsigned long memcg_data;
+#endif
+};
+
+static inline struct ioptdesc *folio_ioptdesc(struct folio *folio)
+{
+ return (struct ioptdesc *)folio;
+}
+
+static inline struct folio *ioptdesc_folio(struct ioptdesc *iopt)
+{
+ return (struct folio *)iopt;
+}
+
+static inline struct ioptdesc *virt_to_ioptdesc(void *virt)
+{
+ return folio_ioptdesc(virt_to_folio(virt));
+}
+
void *iommu_alloc_pages_node(int nid, gfp_t gfp, unsigned int order);
void iommu_free_page(void *virt);
void iommu_put_pages_list(struct iommu_pages_list *list);
@@ -21,7 +58,7 @@ void iommu_put_pages_list(struct iommu_pages_list *list);
static inline void iommu_pages_list_add(struct iommu_pages_list *list,
void *virt)
{
- list_add_tail(&virt_to_page(virt)->lru, &list->pages);
+ list_add_tail(&virt_to_ioptdesc(virt)->iopt_freelist_elm, &list->pages);
}
/**
--
2.43.0
