@@ -8,29 +8,67 @@
#define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
#define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
+/*
+ * struct encoded_va - a nonexistent type marking this pointer
+ *
+ * An 'encoded_va' pointer is a pointer to a aligned virtual address, which is
+ * at least aligned to PAGE_SIZE, that means there are at least 12 lower bits
+ * space available for other purposes.
+ *
+ * Currently we use the lower 8 bits and bit 9 for the order and PFMEMALLOC
+ * flag of the page this 'va' is corresponding to.
+ *
+ * Use the supplied helper functions to endcode/decode the pointer and bits.
+ */
+struct encoded_va;
+
+#define PAGE_FRAG_CACHE_ORDER_MASK GENMASK(7, 0)
+#define PAGE_FRAG_CACHE_PFMEMALLOC_BIT BIT(8)
+#define PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT 8
+
+static inline struct encoded_va *encode_aligned_va(void *va,
+ unsigned int order,
+ bool pfmemalloc)
+{
+ return (struct encoded_va *)((unsigned long)va | order |
+ pfmemalloc << PAGE_FRAG_CACHE_PFMEMALLOC_SHIFT);
+}
+
+static inline unsigned long encoded_page_order(struct encoded_va *encoded_va)
+{
+ return PAGE_FRAG_CACHE_ORDER_MASK & (unsigned long)encoded_va;
+}
+
+static inline bool encoded_page_pfmemalloc(struct encoded_va *encoded_va)
+{
+ return PAGE_FRAG_CACHE_PFMEMALLOC_BIT & (unsigned long)encoded_va;
+}
+
+static inline void *encoded_page_address(struct encoded_va *encoded_va)
+{
+ return (void *)((unsigned long)encoded_va & PAGE_MASK);
+}
+
struct page_frag_cache {
- void *va;
-#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
- __u16 offset;
- __u16 size;
+ struct encoded_va *encoded_va;
+
+#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) && (BITS_PER_LONG <= 32)
+ u16 pagecnt_bias;
+ u16 remaining;
#else
- __u32 offset;
+ u32 pagecnt_bias;
+ u32 remaining;
#endif
- /* we maintain a pagecount bias, so that we dont dirty cache line
- * containing page->_refcount every time we allocate a fragment.
- */
- unsigned int pagecnt_bias;
- bool pfmemalloc;
};
static inline void page_frag_cache_init(struct page_frag_cache *nc)
{
- nc->va = NULL;
+ memset(nc, 0, sizeof(*nc));
}
static inline bool page_frag_cache_is_pfmemalloc(struct page_frag_cache *nc)
{
- return !!nc->pfmemalloc;
+ return encoded_page_pfmemalloc(nc->encoded_va);
}
void page_frag_cache_drain(struct page_frag_cache *nc);
@@ -22,6 +22,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
gfp_t gfp_mask)
{
struct page *page = NULL;
+ unsigned int size, order;
gfp_t gfp = gfp_mask;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
@@ -32,23 +33,41 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
__GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC;
page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
PAGE_FRAG_CACHE_MAX_ORDER);
- nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
#endif
- if (unlikely(!page))
+ if (unlikely(!page)) {
page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
+ if (unlikely(!page))
+ goto alloc_failed;
+
+ size = PAGE_SIZE;
+ order = 0;
+ } else {
+ size = PAGE_FRAG_CACHE_MAX_SIZE;
+ order = PAGE_FRAG_CACHE_MAX_ORDER;
+ }
- nc->va = page ? page_address(page) : NULL;
+ nc->encoded_va = encode_aligned_va(page_address(page), order,
+ page_is_pfmemalloc(page));
+ nc->remaining = size;
+ page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
+ nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
return page;
+
+alloc_failed:
+ nc->encoded_va = NULL;
+ nc->remaining = 0;
+ return NULL;
}
void page_frag_cache_drain(struct page_frag_cache *nc)
{
- if (!nc->va)
+ if (!nc->encoded_va)
return;
- __page_frag_cache_drain(virt_to_head_page(nc->va), nc->pagecnt_bias);
- nc->va = NULL;
+ __page_frag_cache_drain(virt_to_head_page(nc->encoded_va),
+ nc->pagecnt_bias);
+ memset(nc, 0, sizeof(*nc));
}
EXPORT_SYMBOL(page_frag_cache_drain);
@@ -65,35 +84,32 @@ void *__page_frag_alloc_va_align(struct page_frag_cache *nc,
unsigned int fragsz, gfp_t gfp_mask,
unsigned int align_mask)
{
- unsigned int size, offset;
+ unsigned int remaining, page_size;
+ struct encoded_va *encoded_va;
+#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
+ unsigned long page_order;
+#endif
struct page *page;
- if (unlikely(!nc->va)) {
-refill:
- page = __page_frag_cache_refill(nc, gfp_mask);
- if (!page)
- return NULL;
-
- /* Even if we own the page, we do not use atomic_set().
- * This would break get_page_unless_zero() users.
- */
- page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
-
- /* reset page count bias and offset to start of new frag */
- nc->pfmemalloc = page_is_pfmemalloc(page);
- nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
- nc->offset = 0;
- }
-
+alloc_fragment:
+ remaining = nc->remaining & align_mask;
+ encoded_va = nc->encoded_va;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
- /* if size can vary use size else just use PAGE_SIZE */
- size = nc->size;
+ page_order = encoded_page_order(encoded_va);
+ page_size = PAGE_SIZE << page_order;
#else
- size = PAGE_SIZE;
+ page_size = PAGE_SIZE;
#endif
- offset = __ALIGN_KERNEL_MASK(nc->offset, ~align_mask);
- if (unlikely(offset + fragsz > size)) {
+ if (unlikely(fragsz > remaining)) {
+ if (unlikely(!encoded_va)) {
+ page = __page_frag_cache_refill(nc, gfp_mask);
+ if (page)
+ goto alloc_fragment;
+
+ return NULL;
+ }
+
/* fragsz is not supposed to be bigger than PAGE_SIZE as we are
* allowing order 3 page allocation to fail easily under low
* memory condition.
@@ -101,14 +117,22 @@ void *__page_frag_alloc_va_align(struct page_frag_cache *nc,
if (WARN_ON_ONCE(fragsz > PAGE_SIZE))
return NULL;
- page = virt_to_page(nc->va);
+ page = virt_to_page(encoded_va);
+ if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) {
+ page = __page_frag_cache_refill(nc, gfp_mask);
+ if (page)
+ goto alloc_fragment;
- if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
- goto refill;
+ return NULL;
+ }
- if (unlikely(nc->pfmemalloc)) {
+ if (unlikely(encoded_page_pfmemalloc(encoded_va))) {
free_unref_page(page, compound_order(page));
- goto refill;
+ page = __page_frag_cache_refill(nc, gfp_mask);
+ if (page)
+ goto alloc_fragment;
+
+ return NULL;
}
/* OK, page count is 0, we can safely set it */
@@ -116,13 +140,13 @@ void *__page_frag_alloc_va_align(struct page_frag_cache *nc,
/* reset page count bias and offset to start of new frag */
nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
- offset = 0;
+ remaining = page_size;
}
+ nc->remaining = remaining - fragsz;
nc->pagecnt_bias--;
- nc->offset = offset + fragsz;
- return nc->va + offset;
+ return encoded_page_address(encoded_va) + (page_size - remaining);
}
EXPORT_SYMBOL(__page_frag_alloc_va_align);
Currently there is one 'struct page_frag' for every 'struct sock' and 'struct task_struct', we are about to replace the 'struct page_frag' with 'struct page_frag_cache' for them. Before begin the replacing, we need to ensure the size of 'struct page_frag_cache' is not bigger than the size of 'struct page_frag', as there may be tens of thousands of 'struct sock' and 'struct task_struct' instances in the system. By or'ing the page order & pfmemalloc with lower bits of 'va' instead of using 'u16' or 'u32' for page size and 'u8' for pfmemalloc, we are able to avoid 3 or 5 bytes space waste. And page address & pfmemalloc & order is unchanged for the same page in the same 'page_frag_cache' instance, it makes sense to fit them together. Also, it is better to replace 'offset' with 'remaining', which is the remaining size for the cache in a 'page_frag_cache' instance, we are able to do a single 'fragsz > remaining' checking for the case of cache being enough, which should be the fast path if we ensure size is zoro when 'va' == NULL by memset'ing 'struct page_frag_cache' in page_frag_cache_init() and page_frag_cache_drain(). After this patch, the size of 'struct page_frag_cache' should be the same as the size of 'struct page_frag'. CC: Alexander Duyck <alexander.duyck@gmail.com> Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com> --- include/linux/page_frag_cache.h | 62 +++++++++++++++++---- mm/page_frag_cache.c | 98 ++++++++++++++++++++------------- 2 files changed, 111 insertions(+), 49 deletions(-)