mm, swap: avoid over reclaim of full clusters

When running low on usable slots, cluster allocator will try to reclaim
the full clusters aggressively to reclaim HAS_CACHE slots.  This
guarantees that as long as there are any usable slots, HAS_CACHE or not,
the swap device will be usable and workload won't go OOM early.

Before the cluster allocator, swap allocator fails easily if device is
filled up with reclaimable HAS_CACHE slots.  Which can be easily
reproduced with following simple program:

    #include <stdio.h>
    #include <string.h>
    #include <linux/mman.h>
    #include <sys/mman.h>
    #define SIZE 8192UL * 1024UL * 1024UL
    int main(int argc, char **argv) {
        long tmp;
        char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE,
               MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
        memset(p, 0, SIZE);
        madvise(p, SIZE, MADV_PAGEOUT);
        for (unsigned long i = 0; i < SIZE; ++i)
            tmp += p[i];
        getchar(); /* Pause */
        return 0;
    }

Setup an 8G non ramdisk swap, the first run of the program will swapout 8G
ram successfully.  But run same program again after the first run paused,
the second run can't swapout all 8G memory as now half of the swap device
is pinned by HAS_CACHE.  There was a random scan in the old allocator that
may reclaim part of the HAS_CACHE by luck, but it's unreliable.

The new allocator's added reclaim of full clusters when device is low on
usable slots.  But when multiple CPUs are seeing the device is low on
usable slots at the same time, they ran into a thundering herd problem.

This is an observable problem on large machine with mass parallel
workload, as full cluster reclaim is slower on large swap device and
higher number of CPUs will also make things worse.

Testing using a 128G ZRAM on a 48c96t system.  When the swap device is
very close to full (eg.  124G / 128G), running build linux kernel with
make -j96 in a 1G memory cgroup will hung (not a softlockup though)
spinning in full cluster reclaim for about ~5min before go OOM.

To solve this, split the full reclaim into two parts:

- Instead of do a synchronous aggressively reclaim when device is low,
  do only one aggressively reclaim when device is strictly full with a
  kworker. This still ensures in worst case the device won't be unusable
  because of HAS_CACHE slots.

- To avoid allocation (especially higher order) suffer from HAS_CACHE
  filling up clusters and kworker not responsive enough, do one synchronous
  scan every time the free list is drained, and only scan one cluster. This
  is kind of similar to the random reclaim before, keeps the full clusters
  rotated and has a minimal latency. This should provide a fair reclaim
  strategy suitable for most workloads.

Link: https://lkml.kernel.org/r/[email protected]
Fixes: 2cacbdf ("mm: swap: add a adaptive full cluster cache reclaim")
Signed-off-by: Kairui Song <[email protected]>
Cc: Barry Song <[email protected]>
Cc: Chris Li <[email protected]>
Cc: "Huang, Ying" <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Kalesh Singh <[email protected]>
Cc: Ryan Roberts <[email protected]>
Cc: Yosry Ahmed <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

include/linux/swap.h

@@ -335,6 +335,7 @@ struct swap_info_struct {
 					 * list.
 					 */
 	struct work_struct discard_work; /* discard worker */
+	struct work_struct reclaim_work; /* reclaim worker */
 	struct list_head discard_clusters; /* discard clusters list */
 	struct plist_node avail_lists[]; /*
 					   * entries in swap_avail_heads, one

mm/swapfile.c

@@ -731,15 +731,16 @@ static unsigned int alloc_swap_scan_cluster(struct swap_info_struct *si, unsigne
 	return offset;
 }
 
-static void swap_reclaim_full_clusters(struct swap_info_struct *si)
+/* Return true if reclaimed a whole cluster */
+static void swap_reclaim_full_clusters(struct swap_info_struct *si, bool force)
 {
 	long to_scan = 1;
 	unsigned long offset, end;
 	struct swap_cluster_info *ci;
 	unsigned char *map = si->swap_map;
-	int nr_reclaim, total_reclaimed = 0;
+	int nr_reclaim;
 
-	if (atomic_long_read(&nr_swap_pages) <= SWAPFILE_CLUSTER)
+	if (force)
 		to_scan = si->inuse_pages / SWAPFILE_CLUSTER;
 
 	while (!list_empty(&si->full_clusters)) {
@@ -749,28 +750,36 @@ static void swap_reclaim_full_clusters(struct swap_info_struct *si)
 		end = min(si->max, offset + SWAPFILE_CLUSTER);
 		to_scan--;
 
+		spin_unlock(&si->lock);
 		while (offset < end) {
 			if (READ_ONCE(map[offset]) == SWAP_HAS_CACHE) {
-				spin_unlock(&si->lock);
 				nr_reclaim = __try_to_reclaim_swap(si, offset,
 								   TTRS_ANYWAY | TTRS_DIRECT);
-				spin_lock(&si->lock);
-				if (nr_reclaim > 0) {
-					offset += nr_reclaim;
-					total_reclaimed += nr_reclaim;
-					continue;
-				} else if (nr_reclaim < 0) {
-					offset += -nr_reclaim;
+				if (nr_reclaim) {
+					offset += abs(nr_reclaim);
 					continue;
 				}
 			}
 			offset++;
 		}
-		if (to_scan <= 0 || total_reclaimed)
+		spin_lock(&si->lock);
+
+		if (to_scan <= 0)
 			break;
 	}
 }
 
+static void swap_reclaim_work(struct work_struct *work)
+{
+	struct swap_info_struct *si;
+
+	si = container_of(work, struct swap_info_struct, reclaim_work);
+
+	spin_lock(&si->lock);
+	swap_reclaim_full_clusters(si, true);
+	spin_unlock(&si->lock);
+}
+
 /*
  * Try to get swap entries with specified order from current cpu's swap entry
  * pool (a cluster). This might involve allocating a new cluster for current CPU
@@ -800,6 +809,10 @@ static unsigned long cluster_alloc_swap_entry(struct swap_info_struct *si, int o
 		goto done;
 	}
 
+	/* Try reclaim from full clusters if free clusters list is drained */
+	if (vm_swap_full())
+		swap_reclaim_full_clusters(si, false);
+
 	if (order < PMD_ORDER) {
 		unsigned int frags = 0;
 
@@ -881,13 +894,6 @@ static unsigned long cluster_alloc_swap_entry(struct swap_info_struct *si, int o
 	}
 
 done:
-	/* Try reclaim from full clusters if device is nearfull */
-	if (vm_swap_full() && (!found || (si->pages - si->inuse_pages) < SWAPFILE_CLUSTER)) {
-		swap_reclaim_full_clusters(si);
-		if (!found && !order && si->pages != si->inuse_pages)
-			goto new_cluster;
-	}
-
 	cluster->next[order] = offset;
 	return found;
 }
@@ -922,6 +928,9 @@ static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
 		si->lowest_bit = si->max;
 		si->highest_bit = 0;
 		del_from_avail_list(si);
+
+		if (vm_swap_full())
+			schedule_work(&si->reclaim_work);
 	}
 }
 
@@ -2816,6 +2825,7 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
 	wait_for_completion(&p->comp);
 
 	flush_work(&p->discard_work);
+	flush_work(&p->reclaim_work);
 
 	destroy_swap_extents(p);
 	if (p->flags & SWP_CONTINUED)
@@ -3376,6 +3386,7 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 		return PTR_ERR(si);
 
 	INIT_WORK(&si->discard_work, swap_discard_work);
+	INIT_WORK(&si->reclaim_work, swap_reclaim_work);
 
 	name = getname(specialfile);
 	if (IS_ERR(name)) {