]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blobdiff - mm/vmscan.c
memcg: add mem_cgroup_zone_nr_pages()
[mirror_ubuntu-jammy-kernel.git] / mm / vmscan.c
index 5daf606e0a35cb6111b668a584d52b4de8cd940c..d958d624d3ae22daffdd7bd32e824169ad6cd4d5 100644 (file)
@@ -130,6 +130,22 @@ static DECLARE_RWSEM(shrinker_rwsem);
 #define scan_global_lru(sc)    (1)
 #endif
 
+static struct zone_reclaim_stat *get_reclaim_stat(struct zone *zone,
+                                                 struct scan_control *sc)
+{
+       return &zone->reclaim_stat;
+}
+
+static unsigned long zone_nr_pages(struct zone *zone, struct scan_control *sc,
+                                  enum lru_list lru)
+{
+       if (!scan_global_lru(sc))
+               return mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone, lru);
+
+       return zone_page_state(zone, NR_LRU_BASE + lru);
+}
+
+
 /*
  * Add a shrinker callback to be called from the vm
  */
@@ -512,7 +528,6 @@ redo:
                lru = LRU_UNEVICTABLE;
                add_page_to_unevictable_list(page);
        }
-       mem_cgroup_move_lists(page, lru);
 
        /*
         * page's status can change while we move it among lru. If an evictable
@@ -547,7 +562,6 @@ void putback_lru_page(struct page *page)
 
        lru = !!TestClearPageActive(page) + page_is_file_cache(page);
        lru_cache_add_lru(page, lru);
-       mem_cgroup_move_lists(page, lru);
        put_page(page);
 }
 #endif /* CONFIG_UNEVICTABLE_LRU */
@@ -813,6 +827,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
                return ret;
 
        ret = -EBUSY;
+
        if (likely(get_page_unless_zero(page))) {
                /*
                 * Be careful not to clear PageLRU until after we're
@@ -821,6 +836,7 @@ int __isolate_lru_page(struct page *page, int mode, int file)
                 */
                ClearPageLRU(page);
                ret = 0;
+               mem_cgroup_del_lru(page);
        }
 
        return ret;
@@ -1029,6 +1045,7 @@ static unsigned long shrink_inactive_list(unsigned long max_scan,
        struct pagevec pvec;
        unsigned long nr_scanned = 0;
        unsigned long nr_reclaimed = 0;
+       struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
 
        pagevec_init(&pvec, 1);
 
@@ -1072,10 +1089,14 @@ static unsigned long shrink_inactive_list(unsigned long max_scan,
 
                if (scan_global_lru(sc)) {
                        zone->pages_scanned += nr_scan;
-                       zone->recent_scanned[0] += count[LRU_INACTIVE_ANON];
-                       zone->recent_scanned[0] += count[LRU_ACTIVE_ANON];
-                       zone->recent_scanned[1] += count[LRU_INACTIVE_FILE];
-                       zone->recent_scanned[1] += count[LRU_ACTIVE_FILE];
+                       reclaim_stat->recent_scanned[0] +=
+                                                     count[LRU_INACTIVE_ANON];
+                       reclaim_stat->recent_scanned[0] +=
+                                                     count[LRU_ACTIVE_ANON];
+                       reclaim_stat->recent_scanned[1] +=
+                                                     count[LRU_INACTIVE_FILE];
+                       reclaim_stat->recent_scanned[1] +=
+                                                     count[LRU_ACTIVE_FILE];
                }
                spin_unlock_irq(&zone->lru_lock);
 
@@ -1134,10 +1155,9 @@ static unsigned long shrink_inactive_list(unsigned long max_scan,
                        SetPageLRU(page);
                        lru = page_lru(page);
                        add_page_to_lru_list(zone, page, lru);
-                       mem_cgroup_move_lists(page, lru);
                        if (PageActive(page) && scan_global_lru(sc)) {
                                int file = !!page_is_file_cache(page);
-                               zone->recent_rotated[file]++;
+                               reclaim_stat->recent_rotated[file]++;
                        }
                        if (!pagevec_add(&pvec, page)) {
                                spin_unlock_irq(&zone->lru_lock);
@@ -1197,6 +1217,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
        struct page *page;
        struct pagevec pvec;
        enum lru_list lru;
+       struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
 
        lru_add_drain();
        spin_lock_irq(&zone->lru_lock);
@@ -1209,7 +1230,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
         */
        if (scan_global_lru(sc)) {
                zone->pages_scanned += pgscanned;
-               zone->recent_scanned[!!file] += pgmoved;
+               reclaim_stat->recent_scanned[!!file] += pgmoved;
        }
 
        if (file)
@@ -1252,7 +1273,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
         * pages in get_scan_ratio.
         */
        if (scan_global_lru(sc))
-               zone->recent_rotated[!!file] += pgmoved;
+               reclaim_stat->recent_rotated[!!file] += pgmoved;
 
        while (!list_empty(&l_inactive)) {
                page = lru_to_page(&l_inactive);
@@ -1263,7 +1284,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
                ClearPageActive(page);
 
                list_move(&page->lru, &zone->lru[lru].list);
-               mem_cgroup_move_lists(page, lru);
+               mem_cgroup_add_lru_list(page, lru);
                pgmoved++;
                if (!pagevec_add(&pvec, page)) {
                        __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
@@ -1292,6 +1313,38 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
        pagevec_release(&pvec);
 }
 
+static int inactive_anon_is_low_global(struct zone *zone)
+{
+       unsigned long active, inactive;
+
+       active = zone_page_state(zone, NR_ACTIVE_ANON);
+       inactive = zone_page_state(zone, NR_INACTIVE_ANON);
+
+       if (inactive * zone->inactive_ratio < active)
+               return 1;
+
+       return 0;
+}
+
+/**
+ * inactive_anon_is_low - check if anonymous pages need to be deactivated
+ * @zone: zone to check
+ * @sc:   scan control of this context
+ *
+ * Returns true if the zone does not have enough inactive anon pages,
+ * meaning some active anon pages need to be deactivated.
+ */
+static int inactive_anon_is_low(struct zone *zone, struct scan_control *sc)
+{
+       int low;
+
+       if (scan_global_lru(sc))
+               low = inactive_anon_is_low_global(zone);
+       else
+               low = mem_cgroup_inactive_anon_is_low(sc->mem_cgroup, zone);
+       return low;
+}
+
 static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
        struct zone *zone, struct scan_control *sc, int priority)
 {
@@ -1302,8 +1355,7 @@ static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
                return 0;
        }
 
-       if (lru == LRU_ACTIVE_ANON &&
-           (!scan_global_lru(sc) || inactive_anon_is_low(zone))) {
+       if (lru == LRU_ACTIVE_ANON && inactive_anon_is_low(zone, sc)) {
                shrink_active_list(nr_to_scan, zone, sc, priority, file);
                return 0;
        }
@@ -1325,6 +1377,7 @@ static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
        unsigned long anon, file, free;
        unsigned long anon_prio, file_prio;
        unsigned long ap, fp;
+       struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
 
        /* If we have no swap space, do not bother scanning anon pages. */
        if (nr_swap_pages <= 0) {
@@ -1333,17 +1386,20 @@ static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
                return;
        }
 
-       anon  = zone_page_state(zone, NR_ACTIVE_ANON) +
-               zone_page_state(zone, NR_INACTIVE_ANON);
-       file  = zone_page_state(zone, NR_ACTIVE_FILE) +
-               zone_page_state(zone, NR_INACTIVE_FILE);
-       free  = zone_page_state(zone, NR_FREE_PAGES);
+       anon  = zone_nr_pages(zone, sc, LRU_ACTIVE_ANON) +
+               zone_nr_pages(zone, sc, LRU_INACTIVE_ANON);
+       file  = zone_nr_pages(zone, sc, LRU_ACTIVE_FILE) +
+               zone_nr_pages(zone, sc, LRU_INACTIVE_FILE);
 
-       /* If we have very few page cache pages, force-scan anon pages. */
-       if (unlikely(file + free <= zone->pages_high)) {
-               percent[0] = 100;
-               percent[1] = 0;
-               return;
+       if (scan_global_lru(sc)) {
+               free  = zone_page_state(zone, NR_FREE_PAGES);
+               /* If we have very few page cache pages,
+                  force-scan anon pages. */
+               if (unlikely(file + free <= zone->pages_high)) {
+                       percent[0] = 100;
+                       percent[1] = 0;
+                       return;
+               }
        }
 
        /*
@@ -1357,17 +1413,17 @@ static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
         *
         * anon in [0], file in [1]
         */
-       if (unlikely(zone->recent_scanned[0] > anon / 4)) {
+       if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
                spin_lock_irq(&zone->lru_lock);
-               zone->recent_scanned[0] /= 2;
-               zone->recent_rotated[0] /= 2;
+               reclaim_stat->recent_scanned[0] /= 2;
+               reclaim_stat->recent_rotated[0] /= 2;
                spin_unlock_irq(&zone->lru_lock);
        }
 
-       if (unlikely(zone->recent_scanned[1] > file / 4)) {
+       if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
                spin_lock_irq(&zone->lru_lock);
-               zone->recent_scanned[1] /= 2;
-               zone->recent_rotated[1] /= 2;
+               reclaim_stat->recent_scanned[1] /= 2;
+               reclaim_stat->recent_rotated[1] /= 2;
                spin_unlock_irq(&zone->lru_lock);
        }
 
@@ -1383,11 +1439,11 @@ static void get_scan_ratio(struct zone *zone, struct scan_control *sc,
         * proportional to the fraction of recently scanned pages on
         * each list that were recently referenced and in active use.
         */
-       ap = (anon_prio + 1) * (zone->recent_scanned[0] + 1);
-       ap /= zone->recent_rotated[0] + 1;
+       ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
+       ap /= reclaim_stat->recent_rotated[0] + 1;
 
-       fp = (file_prio + 1) * (zone->recent_scanned[1] + 1);
-       fp /= zone->recent_rotated[1] + 1;
+       fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
+       fp /= reclaim_stat->recent_rotated[1] + 1;
 
        /* Normalize to percentages */
        percent[0] = 100 * ap / (ap + fp + 1);
@@ -1467,9 +1523,7 @@ static void shrink_zone(int priority, struct zone *zone,
         * Even if we did not try to evict anon pages at all, we want to
         * rebalance the anon lru active/inactive ratio.
         */
-       if (!scan_global_lru(sc) || inactive_anon_is_low(zone))
-               shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
-       else if (!scan_global_lru(sc))
+       if (inactive_anon_is_low(zone, sc))
                shrink_active_list(SWAP_CLUSTER_MAX, zone, sc, priority, 0);
 
        throttle_vm_writeout(sc->gfp_mask);
@@ -1661,7 +1715,8 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
 
 unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
-                                               gfp_t gfp_mask)
+                                               gfp_t gfp_mask,
+                                          bool noswap)
 {
        struct scan_control sc = {
                .may_writepage = !laptop_mode,
@@ -1674,6 +1729,9 @@ unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *mem_cont,
        };
        struct zonelist *zonelist;
 
+       if (noswap)
+               sc.may_swap = 0;
+
        sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
                        (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
        zonelist = NODE_DATA(numa_node_id())->node_zonelists;
@@ -1761,7 +1819,7 @@ loop_again:
                         * Do some background aging of the anon list, to give
                         * pages a chance to be referenced before reclaiming.
                         */
-                       if (inactive_anon_is_low(zone))
+                       if (inactive_anon_is_low(zone, &sc))
                                shrink_active_list(SWAP_CLUSTER_MAX, zone,
                                                        &sc, priority, 0);
 
@@ -1867,6 +1925,23 @@ out:
 
                try_to_freeze();
 
+               /*
+                * Fragmentation may mean that the system cannot be
+                * rebalanced for high-order allocations in all zones.
+                * At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
+                * it means the zones have been fully scanned and are still
+                * not balanced. For high-order allocations, there is
+                * little point trying all over again as kswapd may
+                * infinite loop.
+                *
+                * Instead, recheck all watermarks at order-0 as they
+                * are the most important. If watermarks are ok, kswapd will go
+                * back to sleep. High-order users can still perform direct
+                * reclaim if they wish.
+                */
+               if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
+                       order = sc.order = 0;
+
                goto loop_again;
        }
 
@@ -2387,6 +2462,7 @@ retry:
 
                __dec_zone_state(zone, NR_UNEVICTABLE);
                list_move(&page->lru, &zone->lru[l].list);
+               mem_cgroup_move_lists(page, LRU_UNEVICTABLE, l);
                __inc_zone_state(zone, NR_INACTIVE_ANON + l);
                __count_vm_event(UNEVICTABLE_PGRESCUED);
        } else {
@@ -2395,6 +2471,7 @@ retry:
                 */
                SetPageUnevictable(page);
                list_move(&page->lru, &zone->lru[LRU_UNEVICTABLE].list);
+               mem_cgroup_rotate_lru_list(page, LRU_UNEVICTABLE);
                if (page_evictable(page, NULL))
                        goto retry;
        }