On Tue, Jun 19, 2018 at 9:22 AM Johannes Weiner [off-list ref] wrote:

On Mon, Jun 18, 2018 at 10:13:25PM -0700, Shakeel Butt wrote:

quoted

@@ -248,6 +248,30 @@ static inline void memalloc_noreclaim_restore(unsigned int flags)
      current->flags = (current->flags & ~PF_MEMALLOC) | flags;
 }

+#ifdef CONFIG_MEMCG
+static inline struct mem_cgroup *memalloc_memcg_save(struct mem_cgroup *memcg)
+{
+     struct mem_cgroup *old_memcg = current->target_memcg;
+
+     current->target_memcg = memcg;
+     return old_memcg;
+}
+
+static inline void memalloc_memcg_restore(struct mem_cgroup *memcg)
+{
+     current->target_memcg = memcg;
+}

The use_mm() and friends naming scheme would be better here:
memalloc_use_memcg(), memalloc_unuse_memcg(), current->active_memcg

Ack. Though do you still think <linux/sched/mm.h> is the right place
for these functions?

quoted

@@ -375,6 +376,27 @@ static __always_inline void kfree_bulk(size_t size, void **p)
      kmem_cache_free_bulk(NULL, size, p);
 }

+/*
+ * Calling kmem_cache_alloc_memcg implicitly assumes that the caller wants
+ * a __GFP_ACCOUNT allocation. However if memcg is NULL then
+ * kmem_cache_alloc_memcg is same as kmem_cache_alloc.
+ */
+static __always_inline void *kmem_cache_alloc_memcg(struct kmem_cache *cachep,
+                                                 gfp_t flags,
+                                                 struct mem_cgroup *memcg)
+{
+     struct mem_cgroup *old_memcg;
+     void *ptr;
+
+     if (!memcg)
+             return kmem_cache_alloc(cachep, flags);
+
+     old_memcg = memalloc_memcg_save(memcg);
+     ptr = kmem_cache_alloc(cachep, flags | __GFP_ACCOUNT);
+     memalloc_memcg_restore(old_memcg);
+     return ptr;

I'm not a big fan of these functions as an interface because it
implies that kmem_cache_alloc() et al wouldn't charge a memcg - but
they do, just using current's memcg.

It's also a lot of churn to duplicate all the various slab functions.

Can you please inline the save/restore (or use/unuse) functions into
the callsites? If you make them handle NULL as parameters, it merely
adds two bracketing lines around the allocation call in the callsites,
which I think would be better to understand - in particular with a
comment on why we are charging *that* group instead of current's.

Ack.

quoted

+static __always_inline struct mem_cgroup *get_mem_cgroup(
+                             struct mem_cgroup *memcg, struct mm_struct *mm)
+{
+     if (unlikely(memcg)) {
+             rcu_read_lock();
+             if (css_tryget_online(&memcg->css)) {
+                     rcu_read_unlock();
+                     return memcg;
+             }
+             rcu_read_unlock();
+     }
+     return get_mem_cgroup_from_mm(mm);
+}
+
 /**
  * mem_cgroup_iter - iterate over memory cgroup hierarchy
  * @root: hierarchy root

@@ -2260,7 +2274,7 @@ struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
      if (current->memcg_kmem_skip_account)
              return cachep;

-     memcg = get_mem_cgroup_from_mm(current->mm);
+     memcg = get_mem_cgroup(current->target_memcg, current->mm);

get_mem_cgroup_from_current(), which uses current->active_memcg if set
and current->mm->memcg otherwise, would be a nicer abstraction IMO.

Ack.

thanks,
Shakeel