]> git.openfabrics.org - ~shefty/rdma-dev.git/blobdiff - mm/page_alloc.c
mm: avoid waking kswapd for THP allocations when compaction is deferred or contended
[~shefty/rdma-dev.git] / mm / page_alloc.c
index 8193809f3de0e683b1bcfb95954f37316768c851..a8f2c87792c33eb264cde47f1bb42db2e8a6c935 100644 (file)
@@ -2378,6 +2378,15 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
        return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
 }
 
+/* Returns true if the allocation is likely for THP */
+static bool is_thp_alloc(gfp_t gfp_mask, unsigned int order)
+{
+       if (order == pageblock_order &&
+           (gfp_mask & (__GFP_MOVABLE|__GFP_REPEAT)) == __GFP_MOVABLE)
+               return true;
+       return false;
+}
+
 static inline struct page *
 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
        struct zonelist *zonelist, enum zone_type high_zoneidx,
@@ -2416,7 +2425,9 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
                goto nopage;
 
 restart:
-       wake_all_kswapd(order, zonelist, high_zoneidx,
+       /* The decision whether to wake kswapd for THP is made later */
+       if (!is_thp_alloc(gfp_mask, order))
+               wake_all_kswapd(order, zonelist, high_zoneidx,
                                        zone_idx(preferred_zone));
 
        /*
@@ -2487,15 +2498,21 @@ rebalance:
                goto got_pg;
        sync_migration = true;
 
-       /*
-        * If compaction is deferred for high-order allocations, it is because
-        * sync compaction recently failed. In this is the case and the caller
-        * requested a movable allocation that does not heavily disrupt the
-        * system then fail the allocation instead of entering direct reclaim.
-        */
-       if ((deferred_compaction || contended_compaction) &&
-           (gfp_mask & (__GFP_MOVABLE|__GFP_REPEAT)) == __GFP_MOVABLE)
-               goto nopage;
+       if (is_thp_alloc(gfp_mask, order)) {
+               /*
+                * If compaction is deferred for high-order allocations, it is
+                * because sync compaction recently failed. If this is the case
+                * and the caller requested a movable allocation that does not
+                * heavily disrupt the system then fail the allocation instead
+                * of entering direct reclaim.
+                */
+               if (deferred_compaction || contended_compaction)
+                       goto nopage;
+
+               /* If process is willing to reclaim/compact then wake kswapd */
+               wake_all_kswapd(order, zonelist, high_zoneidx,
+                                       zone_idx(preferred_zone));
+       }
 
        /* Try direct reclaim and then allocating */
        page = __alloc_pages_direct_reclaim(gfp_mask, order,