[RFC PATCH 3/3] zsmalloc: charge the zspage's object to obj_cgroup

* [RFC PATCH 3/3] zsmalloc: charge the zspage's object to obj_cgroup
@ 2023-06-15  3:49 Zhongkun He
  2023-06-15  7:46 ` kernel test robot
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Zhongkun He @ 2023-06-15  3:49 UTC (permalink / raw)
  To: minchan, senozhatsky, mhocko; +Cc: linux-mm, linux-kernel, Zhongkun He

The zspage's object is not charge to any memory cgroup.The
memory of a task will have no limit if the zram swap device
is large enough even though the task is limited by memory
cgroup.

So, it should be necessary to charge the zspage's object to
obj_cgroup,just like slub.

Signed-off-by: Zhongkun He <hezhongkun.hzk@bytedance.com>
---
 mm/zsmalloc.c | 196 ++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 191 insertions(+), 5 deletions(-)

diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
index 02f7f414aade..a5d1c484dd8f 100644
--- a/mm/zsmalloc.c
+++ b/mm/zsmalloc.c
@@ -62,6 +62,7 @@
 #include <linux/pagemap.h>
 #include <linux/fs.h>
 #include <linux/local_lock.h>
+#include <linux/memcontrol.h>
 
 #define ZSPAGE_MAGIC	0x58
 
@@ -311,6 +312,7 @@ static bool ZsHugePage(struct zspage *zspage)
 static void migrate_lock_init(struct zspage *zspage);
 static void migrate_read_lock(struct zspage *zspage);
 static void migrate_read_unlock(struct zspage *zspage);
+static void zs_free_page_obj_cgroups(struct zspage *zspage);
 
 #ifdef CONFIG_COMPACTION
 static void migrate_write_lock(struct zspage *zspage);
@@ -973,6 +975,7 @@ static void __free_zspage(struct zs_pool *pool, struct size_class *class,
 	free_handles(pool, class, zspage);
 
 	next = page = get_first_page(zspage);
+	zs_free_page_obj_cgroups(zspage);
 	do {
 		VM_BUG_ON_PAGE(!PageLocked(page), page);
 		next = get_next_page(page);
@@ -1420,6 +1423,170 @@ size_t zs_huge_class_size(struct zs_pool *pool)
 }
 EXPORT_SYMBOL_GPL(zs_huge_class_size);
 
+
+#ifdef CONFIG_MEMCG_KMEM
+
+static inline struct obj_cgroup **page_objcgs_check(struct page *page)
+{
+	unsigned long memcg_data = READ_ONCE(page->memcg_data);
+
+	if (!memcg_data || !(memcg_data & MEMCG_DATA_OBJCGS))
+		return NULL;
+
+	VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, page);
+
+	return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK);
+}
+
+
+static void zs_free_page_obj_cgroups(struct zspage *zspage)
+{
+	struct obj_cgroup **objcgs;
+	struct page *page = zspage->first_page;
+
+	objcgs = page_objcgs_check(page);
+	if (!objcgs)
+		return;
+
+	kfree(objcgs);
+	page->memcg_data = 0;
+}
+
+static inline size_t zs_obj_full_size(struct size_class *class)
+{
+	return class->size + sizeof(struct obj_cgroup *);
+}
+
+static inline void zs_objcg_exchange(struct page *s_page, unsigned int s_index,
+				struct page *d_page, unsigned int d_index)
+{
+	struct page *sf_page, *df_page;
+	struct obj_cgroup **s_objcgs, **d_objcgs;
+
+	sf_page = get_first_page(get_zspage(s_page));
+	df_page = get_first_page(get_zspage(d_page));
+
+	s_objcgs =  page_objcgs_check(sf_page);
+	d_objcgs =  page_objcgs_check(df_page);
+
+	if (!s_objcgs || !d_objcgs)
+		return;
+
+	d_objcgs[d_index] = s_objcgs[s_index];
+	s_objcgs[s_index] = NULL;
+}
+
+void zs_alloc_obj_cgroups(struct zspage *zspage, struct size_class *class)
+{
+	struct page *page = zspage->first_page;
+	int objects = class->objs_per_zspage;
+	unsigned long memcg_data = 0;
+	void *vec;
+
+	vec = kcalloc_node(objects, sizeof(struct obj_cgroup *),
+			GFP_NOWAIT|__GFP_NOWARN|
+			__GFP_ZERO|__GFP_MOVABLE,
+			page_to_nid(page));
+
+	if (vec)
+		memcg_data = (unsigned long)vec | MEMCG_DATA_OBJCGS;
+	page->memcg_data = memcg_data;
+}
+
+static inline bool zs_obj_pre_alloc_hook(struct size_class *class,
+				struct obj_cgroup **objcgp, gfp_t flags)
+{
+	struct obj_cgroup *objcg;
+
+	*objcgp = NULL;
+
+	objcg = get_obj_cgroup_from_current();
+	if (!objcg)
+		return true;
+
+	if (obj_cgroup_charge(objcg, flags, zs_obj_full_size(class))) {
+		obj_cgroup_put(objcg);
+		return false;
+	}
+
+	*objcgp = objcg;
+	return true;
+}
+
+static inline void zs_obj_post_alloc_hook(struct size_class *class,
+				struct zspage *zspage,
+				unsigned int idx,
+				struct obj_cgroup *objcg)
+{
+	struct obj_cgroup **objcgs = NULL;
+
+	if (!objcg)
+		return;
+
+	if (zspage)
+		objcgs = page_objcgs_check(zspage->first_page);
+
+	if (!objcgs) {
+		obj_cgroup_uncharge(objcg, zs_obj_full_size(class));
+		obj_cgroup_put(objcg);
+		return;
+	}
+
+	objcgs[idx] = objcg;
+}
+
+static inline void zs_obj_free_hook(int class_size, struct zspage *zspage,
+				unsigned int idx)
+{
+	struct obj_cgroup **objcgs, *objcg;
+
+	objcgs = page_objcgs_check(zspage->first_page);
+
+	if (WARN_ON(!objcgs))
+		return;
+
+	objcg = objcgs[idx];
+	if (!objcg)
+		return;
+
+	objcgs[idx] = NULL;
+	obj_cgroup_uncharge(objcg, class_size + sizeof(struct obj_cgroup *));
+	obj_cgroup_put(objcg);
+}
+#else
+
+static void zs_free_page_obj_cgroups(struct zspage *zspage)
+{
+}
+
+static inline void zs_objcg_exchange(struct page *s_page, unsigned int s_index,
+				struct page *d_page, unsigned int d_index)
+{
+}
+
+int zs_alloc_obj_cgroups(struct zspage *zspage, struct size_class *class)
+{
+	return 0;
+}
+
+static inline bool zs_obj_pre_alloc_hook(struct size_class *class,
+				struct obj_cgroup **objcgp, gfp_t flags)
+{
+	return true;
+}
+
+static inline void zs_obj_post_alloc_hook(struct size_class *class,
+				struct page *page,
+				unsigned int idx,
+				struct obj_cgroup *objcg)
+{
+}
+static inline void zs_obj_free_hook(int class_size, struct zspage *zspage,
+				unsigned int idx)
+{
+}
+#endif
+
 static unsigned long obj_malloc(struct zs_pool *pool,
 				struct zspage *zspage, unsigned long handle)
 {
@@ -1475,7 +1642,8 @@ static unsigned long obj_malloc(struct zs_pool *pool,
  */
 unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp)
 {
-	unsigned long handle, obj;
+	unsigned long handle, obj, index;
+	struct obj_cgroup *objcg;
 	struct size_class *class;
 	int newfg;
 	struct zspage *zspage;
@@ -1491,17 +1659,23 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp)
 	size += ZS_HANDLE_SIZE;
 	class = pool->size_class[get_size_class_index(size)];
 
+	if (!zs_obj_pre_alloc_hook(class, &objcg, gfp)) {
+		cache_free_handle(pool, handle);
+		return (unsigned long)ERR_PTR(-ENOMEM);
+	}
+
 	/* pool->lock effectively protects the zpage migration */
 	spin_lock(&pool->lock);
 	zspage = find_get_zspage(class);
 	if (likely(zspage)) {
+		index = get_freeobj(zspage);
 		obj = obj_malloc(pool, zspage, handle);
 		/* Now move the zspage to another fullness group, if required */
 		fix_fullness_group(class, zspage);
 		record_obj(handle, obj);
 		class_stat_inc(class, ZS_OBJS_INUSE, 1);
 
-		goto out;
+		goto out_unlock;
 	}
 
 	spin_unlock(&pool->lock);
@@ -1509,9 +1683,12 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp)
 	zspage = alloc_zspage(pool, class, gfp);
 	if (!zspage) {
 		cache_free_handle(pool, handle);
-		return (unsigned long)ERR_PTR(-ENOMEM);
+		handle =  (unsigned long)ERR_PTR(-ENOMEM);
+		goto out;
 	}
 
+	zs_alloc_obj_cgroups(zspage, class);
+	index = get_freeobj(zspage);
 	spin_lock(&pool->lock);
 	obj = obj_malloc(pool, zspage, handle);
 	newfg = get_fullness_group(class, zspage);
@@ -1524,7 +1701,7 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp)
 
 	/* We completely set up zspage so mark them as movable */
 	SetZsPageMovable(pool, zspage);
-out:
+out_unlock:
 #ifdef CONFIG_ZPOOL
 	/* Add/move zspage to beginning of LRU */
 	if (!list_empty(&zspage->lru))
@@ -1533,6 +1710,8 @@ unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp)
 #endif
 
 	spin_unlock(&pool->lock);
+out:
+	zs_obj_post_alloc_hook(class, zspage, index, objcg);
 
 	return handle;
 }
@@ -1573,7 +1752,7 @@ static void obj_free(int class_size, unsigned long obj, unsigned long *handle)
 			f_page->index = 0;
 		set_freeobj(zspage, f_objidx);
 	}
-
+	zs_obj_free_hook(class_size, zspage, f_objidx);
 	kunmap_atomic(vaddr);
 	mod_zspage_inuse(zspage, -1);
 }
@@ -1640,6 +1819,7 @@ static void zs_object_copy(struct size_class *class, unsigned long dst,
 	obj_to_location(src, &s_page, &s_objidx);
 	obj_to_location(dst, &d_page, &d_objidx);
 
+	zs_objcg_exchange(s_page, s_objidx, d_page, d_objidx);
 	s_off = (class->size * s_objidx) & ~PAGE_MASK;
 	d_off = (class->size * d_objidx) & ~PAGE_MASK;
 
@@ -1960,6 +2140,12 @@ static void replace_sub_page(struct size_class *class, struct zspage *zspage,
 	int idx = 0;
 
 	page = get_first_page(zspage);
+#ifdef CONFIG_MEMCG_KMEM
+	if (page == oldpage) {
+		newpage->memcg_data = page->memcg_data;
+		page->memcg_data = 0;
+	}
+#endif
 	do {
 		if (page == oldpage)
 			pages[idx] = newpage;
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 4+ messages in thread