diff --git a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
index 8dc8a554f774..dd0cf4834eaa 100644
--- a/arch/ia64/mm/discontig.c
+++ b/arch/ia64/mm/discontig.c
@@ -608,11 +608,11 @@ void __init paging_init(void)
 	zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
 }
 
-pg_data_t *arch_alloc_nodedata(int nid)
+pg_data_t * __init arch_alloc_nodedata(int nid)
 {
 	unsigned long size = compute_pernodesize(nid);
 
-	return kzalloc(size, GFP_KERNEL);
+	return memblock_alloc(size, SMP_CACHE_BYTES);
 }
 
 void arch_free_nodedata(pg_data_t *pgdat)
diff --git a/include/linux/memory_hotplug.h b/include/linux/memory_hotplug.h
index 4355983b364d..cdd66bfdf855 100644
--- a/include/linux/memory_hotplug.h
+++ b/include/linux/memory_hotplug.h
@@ -44,7 +44,7 @@ extern void arch_refresh_nodedata(int nid, pg_data_t *pgdat);
  */
 #define generic_alloc_nodedata(nid)				\
 ({								\
-	kzalloc(sizeof(pg_data_t), GFP_KERNEL);			\
+	memblock_alloc(sizeof(*pgdat), SMP_CACHE_BYTES);	\
 })
 /*
  * This definition is just for error path in node hotadd.
diff --git a/mm/internal.h b/mm/internal.h
index 9c298afb9688..f1554a4e249e 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -707,4 +707,6 @@ void vunmap_range_noflush(unsigned long start, unsigned long end);
 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);
 
+DECLARE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
+
 #endif	/* __MM_INTERNAL_H */
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 0139b77c51d5..11f39d0e76ec 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1162,19 +1162,21 @@ static void reset_node_present_pages(pg_data_t *pgdat)
 }
 
 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
-static pg_data_t __ref *hotadd_new_pgdat(int nid)
+static pg_data_t __ref *hotadd_init_pgdat(int nid)
 {
 	struct pglist_data *pgdat;
 
 	pgdat = NODE_DATA(nid);
-	if (!pgdat) {
-		pgdat = arch_alloc_nodedata(nid);
-		if (!pgdat)
-			return NULL;
 
+	/*
+	 * NODE_DATA is preallocated (free_area_init) but its internal
+	 * state is not allocated completely. Add missing pieces.
+	 * Completely offline nodes stay around and they just need
+	 * reintialization.
+	 */
+	if (pgdat->per_cpu_nodestats == &boot_nodestats) {
 		pgdat->per_cpu_nodestats =
 			alloc_percpu(struct per_cpu_nodestat);
-		arch_refresh_nodedata(nid, pgdat);
 	} else {
 		int cpu;
 		/*
@@ -1193,8 +1195,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid)
 		}
 	}
 
-	/* we can use NODE_DATA(nid) from here */
-	pgdat->node_id = nid;
 	pgdat->node_start_pfn = 0;
 
 	/* init node's zones as empty zones, we don't have any present pages.*/
@@ -1246,7 +1246,7 @@ static int __try_online_node(int nid, bool set_node_online)
 	if (node_online(nid))
 		return 0;
 
-	pgdat = hotadd_new_pgdat(nid);
+	pgdat = hotadd_init_pgdat(nid);
 	if (!pgdat) {
 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
 		ret = -ENOMEM;
@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
 
 	return ret;
 error:
-	/* rollback pgdat allocation and others */
-	if (new_node)
-		rollback_node_hotadd(nid);
 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 		memblock_remove(start, size);
 error_mem_hotplug_end:
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 8b18a077c409..4f141a4e5b64 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6341,7 +6341,7 @@ static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonesta
 #define BOOT_PAGESET_BATCH	1
 static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset);
 static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats);
-static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
+DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
 
 static void __build_all_zonelists(void *data)
 {
@@ -6363,7 +6363,11 @@ static void __build_all_zonelists(void *data)
 	if (self && !node_online(self->node_id)) {
 		build_zonelists(self);
 	} else {
-		for_each_online_node(nid) {
+		/*
+		 * All possible nodes have pgdat preallocated
+		 * in free_area_init
+		 */
+		for_each_node(nid) {
 			pg_data_t *pgdat = NODE_DATA(nid);
 
 			build_zonelists(pgdat);
@@ -8063,8 +8067,36 @@ void __init free_area_init(unsigned long *max_zone_pfn)
 	/* Initialise every node */
 	mminit_verify_pageflags_layout();
 	setup_nr_node_ids();
-	for_each_online_node(nid) {
-		pg_data_t *pgdat = NODE_DATA(nid);
+	for_each_node(nid) {
+		pg_data_t *pgdat;
+
+		if (!node_online(nid)) {
+			pr_info("Initializing node %d as memoryless\n", nid);
+
+			/* Allocator not initialized yet */
+			pgdat = arch_alloc_nodedata(nid);
+			if (!pgdat) {
+				pr_err("Cannot allocate %zuB for node %d.\n",
+						sizeof(*pgdat), nid);
+				continue;
+			}
+			arch_refresh_nodedata(nid, pgdat);
+			free_area_init_memoryless_node(nid);
+
+			/*
+			 * We do not want to confuse userspace by sysfs
+			 * files/directories for node without any memory
+			 * attached to it, so this node is not marked as
+			 * N_MEMORY and not marked online so that no sysfs
+			 * hierarchy will be created via register_one_node for
+			 * it. The pgdat will get fully initialized by
+			 * hotadd_init_pgdat() when memory is hotplugged into
+			 * this node.
+			 */
+			continue;
+		}
+
+		pgdat = NODE_DATA(nid);
 		free_area_init_node(nid);
 
 		/* Any memory on that node */