mm: introduce and export pgdat peer_node

euleros inclusion
category: feature
feature: etmem
bugzilla: 48246

-------------------------------------------------

Each CPU socket can have 1 DRAM and 1 PMEM node, we call them "peer nodes".
Migration between DRAM and PMEM will by default happen between peer nodes.

It's a temp solution. In multiple memory layers, a node can have both
promotion and demotion targets instead of a single peer node. User space
may also be able to infer promotion/demotion targets based on future
HMAT info.
Signed-off-by: NFan Du <fan.du@intel.com>
Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NShijie Luo <luoshijie1@huawei.com>
Signed-off-by: NKemeng Shi <shikemeng@huawei.com>
Reviewed-by: Nlouhongxiang <louhongxiang@huawei.com>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

drivers/base/node.c

@@ -558,6 +558,16 @@ static ssize_t type_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(type);
 
+static ssize_t peer_node_show(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	int nid = dev->id;
+	struct pglist_data *pgdat = NODE_DATA(nid);
+
+	return sprintf(buf, "%d\n", pgdat->peer_node);
+}
+static DEVICE_ATTR_RO(peer_node);
+
 static struct attribute *node_dev_attrs[] = {
 	&dev_attr_cpumap.attr,
 	&dev_attr_cpulist.attr,
@@ -566,6 +576,7 @@ static struct attribute *node_dev_attrs[] = {
 	&dev_attr_distance.attr,
 	&dev_attr_vmstat.attr,
 	&dev_attr_type.attr,
+	&dev_attr_peer_node.attr,
 	NULL
 };
 ATTRIBUTE_GROUPS(node_dev);

include/linux/mm.h

@@ -2414,6 +2414,7 @@ static inline unsigned long get_num_physpages(void)
 	return phys_pages;
 }
 
+int find_best_peer_node(int nid);
 /*
  * Using memblock node mappings, an architecture may initialise its
  * zones, allocate the backing mem_map and account for memory holes in an

include/linux/mmzone.h

@@ -806,6 +806,7 @@ typedef struct pglist_data {
 	/* Per-node vmstats */
 	struct per_cpu_nodestat __percpu *per_cpu_nodestats;
 	atomic_long_t		vm_stat[NR_VM_NODE_STAT_ITEMS];
+	int peer_node;
 } pg_data_t;
 
 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)

mm/page_alloc.c

@@ -7400,6 +7400,34 @@ static void check_for_memory(pg_data_t *pgdat, int nid)
 	}
 }
 
+/*
+ * Return the nearest peer node in terms of *locality*
+ * E.g. peer of node 0 is node 2 per SLIT
+ * node distances:
+ * node   0   1   2   3
+ *   0:  10  21  17  28
+ *   1:  21  10  28  17
+ *   2:  17  28  10  28
+ *   3:  28  17  28  10
+ */
+int find_best_peer_node(int nid)
+{
+	int n, val;
+	int min_val = INT_MAX;
+	int peer = NUMA_NO_NODE;
+
+	for_each_online_node(n) {
+		if (n == nid)
+			continue;
+		val = node_distance(nid, n);
+		if (val < min_val) {
+			min_val = val;
+			peer = n;
+		}
+	}
+	return peer;
+}
+
 /*
  * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
  * such cases we allow max_zone_pfn sorted in the descending order
@@ -7506,6 +7534,7 @@ void __init free_area_init(unsigned long *max_zone_pfn)
 		if (pgdat->node_present_pages)
 			node_set_state(nid, N_MEMORY);
 		check_for_memory(pgdat, nid);
+		pgdat->peer_node = find_best_peer_node(nid);
 	}
 }