mm: swap: swap ratio support

Add support to receive a static ratio from userspace to
divide the swap pages between ZRAM and disk based swap
devices. The existing infrastructure allows to keep
same priority for multiple swap devices, which results
in round robin distribution of pages. With this patch,
the ratio can be defined.

Change-Id: I54f54489db84cabb206569dd62d61a8a7a898991
Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>
[swatsrid@codeaurora.org: Fix trivial merge conflicts]
Signed-off-by: Swathi Sridhar <swatsrid@codeaurora.org>

include/linux/plist.h

@@ -266,6 +266,9 @@ static inline int plist_node_empty(const struct plist_node *node)
 #define plist_next(pos) \
 	list_next_entry(pos, node_list)
 
+#define plist_next_entry(pos, type, member)   \
+	container_of(plist_next(pos), type, member)
+
 /**
  * plist_prev - get the prev entry in list
  * @pos:	the type * to cursor

include/linux/swap.h

@@ -178,6 +178,7 @@ enum {
 
 #define SWAP_CLUSTER_MAX 32UL
 #define COMPACT_CLUSTER_MAX SWAP_CLUSTER_MAX
+#define SWAPFILE_CLUSTER	256
 
 #define SWAP_MAP_MAX	0x3e	/* Max duplication count, in first swap_map */
 #define SWAP_MAP_BAD	0x3f	/* Note pageblock is bad, in first swap_map */
@@ -273,6 +274,8 @@ struct swap_info_struct {
 					 */
 	struct work_struct discard_work; /* discard worker */
 	struct swap_cluster_list discard_clusters; /* discard clusters list */
+	unsigned int write_pending;
+	unsigned int max_writes;
 };
 
 #ifdef CONFIG_64BIT
@@ -355,6 +358,8 @@ extern unsigned long mem_cgroup_shrink_node(struct mem_cgroup *mem,
 						unsigned long *nr_scanned);
 extern unsigned long shrink_all_memory(unsigned long nr_pages);
 extern int vm_swappiness;
+extern int sysctl_swap_ratio;
+extern int sysctl_swap_ratio_enable;
 extern int remove_mapping(struct address_space *mapping, struct page *page);
 extern unsigned long vm_total_pages;
 

include/linux/swapfile.h

@@ -8,9 +8,14 @@
  */
 extern spinlock_t swap_lock;
 extern struct plist_head swap_active_head;
+extern spinlock_t swap_avail_lock;
+extern struct plist_head *swap_avail_heads;
 extern struct swap_info_struct *swap_info[];
 extern int try_to_unuse(unsigned int, bool, unsigned long);
 extern unsigned long generic_max_swapfile_size(void);
 extern unsigned long max_swapfile_size(void);
+extern int swap_ratio(struct swap_info_struct **si, int node);
+extern void setup_swap_ratio(struct swap_info_struct *p, int prio);
+extern bool is_swap_ratio_group(int prio);
 
 #endif /* _LINUX_SWAPFILE_H */

kernel/sysctl.c

@@ -1672,6 +1672,22 @@ static struct ctl_table vm_table[] = {
 		.extra1		= (void *)&mmap_rnd_compat_bits_min,
 		.extra2		= (void *)&mmap_rnd_compat_bits_max,
 	},
+#endif
+#ifdef CONFIG_SWAP
+	{
+		.procname	= "swap_ratio",
+		.data		= &sysctl_swap_ratio,
+		.maxlen		= sizeof(sysctl_swap_ratio),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+	},
+	{
+		.procname	= "swap_ratio_enable",
+		.data		= &sysctl_swap_ratio_enable,
+		.maxlen		= sizeof(sysctl_swap_ratio_enable),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+	},
 #endif
 	{ }
 };

mm/Makefile

@@ -54,7 +54,7 @@ ifdef CONFIG_MMU
 endif
 obj-$(CONFIG_HAVE_MEMBLOCK) += memblock.o
 
-obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o swap_slots.o
+obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o swap_slots.o swap_ratio.o
 obj-$(CONFIG_FRONTSWAP)	+= frontswap.o
 obj-$(CONFIG_ZSWAP)	+= zswap.o
 obj-$(CONFIG_HAS_DMA)	+= dmapool.o

mm/swap_ratio.c

@@ -0,0 +1,182 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2015-2018, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/mm_types.h>
+#include <linux/swapfile.h>
+#include <linux/swap.h>
+
+#define SWAP_RATIO_GROUP_START (SWAP_FLAG_PRIO_MASK - 9) /* 32758 */
+#define SWAP_RATIO_GROUP_END (SWAP_FLAG_PRIO_MASK) /* 32767 */
+#define SWAP_FAST_WRITES (SWAPFILE_CLUSTER * (SWAP_CLUSTER_MAX / 8))
+#define SWAP_SLOW_WRITES SWAPFILE_CLUSTER
+
+/*
+ * The fast/slow swap write ratio.
+ * 100 indicates that all writes should
+ * go to fast swap device.
+ */
+int sysctl_swap_ratio = 100;
+
+/* Enable the swap ratio feature */
+int sysctl_swap_ratio_enable;
+
+static bool is_same_group(struct swap_info_struct *a,
+		struct swap_info_struct *b)
+{
+	if (!sysctl_swap_ratio_enable)
+		return false;
+
+	if (!is_swap_ratio_group(a->prio))
+		return false;
+
+	if (a->prio == b->prio)
+		return true;
+
+	return false;
+}
+
+/* Caller must hold swap_avail_lock */
+static int calculate_write_pending(struct swap_info_struct *si,
+			struct swap_info_struct *n)
+{
+	int ratio = sysctl_swap_ratio;
+
+	if ((ratio < 0) || (ratio > 100))
+		return -EINVAL;
+
+	if (WARN_ON(!(si->flags & SWP_SYNCHRONOUS_IO)))
+		return -ENODEV;
+
+	if ((n->flags & SWP_SYNCHRONOUS_IO) || !is_same_group(si, n))
+		return -ENODEV;
+
+	si->max_writes = ratio ? SWAP_FAST_WRITES : 0;
+	n->max_writes  = ratio ? (SWAP_FAST_WRITES * 100) /
+			ratio - SWAP_FAST_WRITES : SWAP_SLOW_WRITES;
+
+	si->write_pending = si->max_writes;
+	n->write_pending = n->max_writes;
+
+	return 0;
+}
+
+static int swap_ratio_slow(struct swap_info_struct **si, int node)
+{
+	struct swap_info_struct *n = NULL;
+	int ret = 0;
+
+	spin_lock(&(*si)->lock);
+	spin_lock(&swap_avail_lock);
+	if (&(*si)->avail_lists[node] == plist_last(&swap_avail_heads[node])) {
+		/* just to make skip work */
+		n = *si;
+		ret = -ENODEV;
+		goto skip;
+	}
+	n = plist_next_entry(&(*si)->avail_lists[node],
+			struct swap_info_struct,
+			avail_lists[node]);
+	spin_unlock(&swap_avail_lock);
+	spin_lock(&n->lock);
+	spin_lock(&swap_avail_lock);
+
+	if ((*si)->flags & SWP_SYNCHRONOUS_IO) {
+		if ((*si)->write_pending) {
+			(*si)->write_pending--;
+			goto exit;
+		} else {
+			if ((n->flags & SWP_SYNCHRONOUS_IO) ||
+			     !is_same_group(*si, n)) {
+				/* Should never happen */
+				ret = -ENODEV;
+			} else if (n->write_pending) {
+				/*
+				 * Requeue fast device, since there are pending
+				 * writes for slow device.
+				 */
+				plist_requeue(&(*si)->avail_lists[node],
+					&swap_avail_heads[node]);
+				n->write_pending--;
+				spin_unlock(&(*si)->lock);
+				*si = n;
+				goto skip;
+			} else {
+				if (calculate_write_pending(*si, n) < 0) {
+					ret = -ENODEV;
+					goto exit;
+				}
+				/* Restart from fast device */
+				(*si)->write_pending--;
+			}
+		}
+	} else {
+		if (!(n->flags & SWP_SYNCHRONOUS_IO) ||
+		      !is_same_group(*si, n)) {
+			/* Should never happen */
+			ret = -ENODEV;
+		} else if (n->write_pending) {
+			/*
+			 * Pending writes for fast device.
+			 * We reach here when slow device is swapped on first,
+			 * before fast device.
+			 */
+			/* requeue slow device to the end */
+			plist_requeue(&(*si)->avail_lists[node],
+					&swap_avail_heads[node]);
+			n->write_pending--;
+			spin_unlock(&(*si)->lock);
+			*si = n;
+			goto skip;
+		} else {
+			if ((*si)->write_pending) {
+				(*si)->write_pending--;
+			} else {
+				if (calculate_write_pending(n, *si) < 0) {
+					ret = -ENODEV;
+					goto exit;
+				}
+				n->write_pending--;
+				plist_requeue(&(*si)->avail_lists[node],
+					&swap_avail_heads[node]);
+				spin_unlock(&(*si)->lock);
+				*si = n;
+				goto skip;
+			}
+		}
+	}
+exit:
+	spin_unlock(&(*si)->lock);
+skip:
+	spin_unlock(&swap_avail_lock);
+	/* n and si would have got interchanged */
+	spin_unlock(&n->lock);
+	return ret;
+}
+
+bool is_swap_ratio_group(int prio)
+{
+	return ((prio >= SWAP_RATIO_GROUP_START) &&
+		(prio <= SWAP_RATIO_GROUP_END)) ? true : false;
+}
+
+void setup_swap_ratio(struct swap_info_struct *p, int prio)
+{
+	/* Used only if sysctl_swap_ratio_enable is set */
+	if (is_swap_ratio_group(prio)) {
+		if (p->flags & SWP_SYNCHRONOUS_IO)
+			p->write_pending = SWAP_FAST_WRITES;
+		else
+			p->write_pending = SWAP_SLOW_WRITES;
+		p->max_writes =  p->write_pending;
+	}
+}
+
+int swap_ratio(struct swap_info_struct **si, int node)
+{
+	if (is_swap_ratio_group((*si)->prio))
+		return swap_ratio_slow(si, node);
+	else
+		return -ENODEV;
+}

mm/swapfile.c

@@ -85,8 +85,8 @@ PLIST_HEAD(swap_active_head);
  * is held and the locking order requires swap_lock to be taken
  * before any swap_info_struct->lock.
  */
-static struct plist_head *swap_avail_heads;
-static DEFINE_SPINLOCK(swap_avail_lock);
+struct plist_head *swap_avail_heads;
+DEFINE_SPINLOCK(swap_avail_lock);
 
 struct swap_info_struct *swap_info[MAX_SWAPFILES];
 
@@ -947,6 +947,7 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
 	long avail_pgs;
 	int n_ret = 0;
 	int node;
+	int swap_ratio_off = 0;
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
@@ -963,14 +964,34 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
 
 	atomic_long_sub(n_goal * size, &nr_swap_pages);
 
+lock_and_start:
 	spin_lock(&swap_avail_lock);
 
 start_over:
 	node = numa_node_id();
 	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
+
+		if (sysctl_swap_ratio && !swap_ratio_off) {
+			int ret;
+
+			spin_unlock(&swap_avail_lock);
+			ret = swap_ratio(&si, node);
+			if (ret < 0) {
+				/*
+				 * Error. Start again with swap
+				 * ratio disabled.
+				 */
+				swap_ratio_off = 1;
+				goto lock_and_start;
+			} else {
+				goto start;
+			}
+		}
+
 		/* requeue si to after same-priority siblings */
 		plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
 		spin_unlock(&swap_avail_lock);
+start:
 		spin_lock(&si->lock);
 		if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
 			spin_lock(&swap_avail_lock);
@@ -3270,9 +3291,11 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 
 	mutex_lock(&swapon_mutex);
 	prio = -1;
-	if (swap_flags & SWAP_FLAG_PREFER)
+	if (swap_flags & SWAP_FLAG_PREFER) {
 		prio =
 		  (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
+		setup_swap_ratio(p, prio);
+	}
 	enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
 
 	pr_info("Adding %uk swap on %s.  Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",