[PATCH] memcg: effective memory.high reclaim for remote charging

[Shakeel Butt]

Currently the reclaim of excessive usage over memory.high is scheduled
to run on returning to the userland. The main reason behind this
approach was simplicity i.e. always reclaim with GFP_KERNEL context.
However the underlying assumptions behind this approach are: the current
task shares the memcg hierarchy with the given memcg and the memcg of
the current task most probably will not change on return to userland.

With the remote charging, the first assumption breaks and it allows the
usage to grow way beyond the memory.high as the reclaim and the
throttling becomes ineffective.

This patch forces the synchronous reclaim and potentially throttling for
the callers with context that allows blocking. For unblockable callers
or whose synch high reclaim is still not successful, a high reclaim is
scheduled either to return-to-userland if current task shares the
hierarchy with the given memcg or to system work queue.

Signed-off-by: Shakeel Butt <shakeelb@xxxxxxxxxx>
---
 mm/memcontrol.c | 63 +++++++++++++++++++++++++++++--------------------
 1 file changed, 37 insertions(+), 26 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 317dbbaac603..7abb762f26cd 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2387,23 +2387,13 @@ static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
 	return min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES);
 }
 
-/*
- * Scheduled by try_charge() to be executed from the userland return path
- * and reclaims memory over the high limit.
- */
-void mem_cgroup_handle_over_high(void)
+static void reclaim_over_high(struct mem_cgroup *memcg, gfp_t gfp_mask,
+			      unsigned long nr_pages)
 {
 	unsigned long penalty_jiffies;
 	unsigned long pflags;
-	unsigned int nr_pages = current->memcg_nr_pages_over_high;
-	struct mem_cgroup *memcg;
 
-	if (likely(!nr_pages))
-		return;
-
-	memcg = get_mem_cgroup_from_mm(current->mm);
-	reclaim_high(memcg, nr_pages, GFP_KERNEL);
-	current->memcg_nr_pages_over_high = 0;
+	reclaim_high(memcg, nr_pages, gfp_mask);
 
 	/*
 	 * memory.high is breached and reclaim is unable to keep up. Throttle
@@ -2418,7 +2408,7 @@ void mem_cgroup_handle_over_high(void)
 	 * been aggressively reclaimed enough yet.
 	 */
 	if (penalty_jiffies <= HZ / 100)
-		goto out;
+		return;
 
 	/*
 	 * If we exit early, we're guaranteed to die (since
@@ -2428,8 +2418,23 @@ void mem_cgroup_handle_over_high(void)
 	psi_memstall_enter(&pflags);
 	schedule_timeout_killable(penalty_jiffies);
 	psi_memstall_leave(&pflags);
+}
 
-out:
+/*
+ * Scheduled by try_charge() to be executed from the userland return path
+ * and reclaims memory over the high limit.
+ */
+void mem_cgroup_handle_over_high(void)
+{
+	unsigned int nr_pages = current->memcg_nr_pages_over_high;
+	struct mem_cgroup *memcg;
+
+	if (likely(!nr_pages))
+		return;
+
+	memcg = get_mem_cgroup_from_mm(current->mm);
+	reclaim_over_high(memcg, GFP_KERNEL, nr_pages);
+	current->memcg_nr_pages_over_high = 0;
 	css_put(&memcg->css);
 }
 
@@ -2584,15 +2589,6 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 	if (batch > nr_pages)
 		refill_stock(memcg, batch - nr_pages);
 
-	/*
-	 * If the hierarchy is above the normal consumption range, schedule
-	 * reclaim on returning to userland.  We can perform reclaim here
-	 * if __GFP_RECLAIM but let's always punt for simplicity and so that
-	 * GFP_KERNEL can consistently be used during reclaim.  @memcg is
-	 * not recorded as it most likely matches current's and won't
-	 * change in the meantime.  As high limit is checked again before
-	 * reclaim, the cost of mismatch is negligible.
-	 */
 	do {
 		if (page_counter_read(&memcg->memory) > READ_ONCE(memcg->high)) {
 			/* Don't bother a random interrupted task */
@@ -2600,8 +2596,23 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 				schedule_work(&memcg->high_work);
 				break;
 			}
-			current->memcg_nr_pages_over_high += batch;
-			set_notify_resume(current);
+
+			if (gfpflags_allow_blocking(gfp_mask))
+				reclaim_over_high(memcg, gfp_mask, batch);
+
+			if (page_counter_read(&memcg->memory) <=
+			    READ_ONCE(memcg->high))
+				break;
+			/*
+			 * The above reclaim might not be able to do much. Punt
+			 * the high reclaim to return to userland if the current
+			 * task shares the hierarchy.
+			 */
+			if (current->mm && mm_match_cgroup(current->mm, memcg)) {
+				current->memcg_nr_pages_over_high += batch;
+				set_notify_resume(current);
+			} else
+				schedule_work(&memcg->high_work);
 			break;
 		}
 	} while ((memcg = parent_mem_cgroup(memcg)));
-- 
2.26.2.526.g744177e7f7-goog