diff --git a/mm/percpu.c b/mm/percpu.c
index 3311c89..43d8cac 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -8,12 +8,13 @@
  *
  * This is percpu allocator which can handle both static and dynamic
  * areas.  Percpu areas are allocated in chunks in vmalloc area.  Each
- * chunk is consisted of nr_cpu_ids units and the first chunk is used
- * for static percpu variables in the kernel image (special boot time
- * alloc/init handling necessary as these areas need to be brought up
- * before allocation services are running).  Unit grows as necessary
- * and all units grow or shrink in unison.  When a chunk is filled up,
- * another chunk is allocated.  ie. in vmalloc area
+ * chunk is consisted of boot-time determined number of units and the
+ * first chunk is used for static percpu variables in the kernel image
+ * (special boot time alloc/init handling necessary as these areas
+ * need to be brought up before allocation services are running).
+ * Unit grows as necessary and all units grow or shrink in unison.
+ * When a chunk is filled up, another chunk is allocated.  ie. in
+ * vmalloc area
  *
  *  c0                           c1                         c2
  *  -------------------          -------------------        ------------
@@ -22,11 +23,13 @@
  *
  * Allocation is done in offset-size areas of single unit space.  Ie,
  * an area of 512 bytes at 6k in c1 occupies 512 bytes at 6k of c1:u0,
- * c1:u1, c1:u2 and c1:u3.  Percpu access can be done by configuring
- * percpu base registers pcpu_unit_size apart.
+ * c1:u1, c1:u2 and c1:u3.  On UMA, units corresponds directly to
+ * cpus.  On NUMA, the mapping can be non-linear and even sparse.
+ * Percpu access can be done by configuring percpu base registers
+ * according to cpu to unit mapping and pcpu_unit_size.
  *
- * There are usually many small percpu allocations many of them as
- * small as 4 bytes.  The allocator organizes chunks into lists
+ * There are usually many small percpu allocations many of them being
+ * as small as 4 bytes.  The allocator organizes chunks into lists
  * according to free size and tries to allocate from the fullest one.
  * Each chunk keeps the maximum contiguous area size hint which is
  * guaranteed to be eqaul to or larger than the maximum contiguous
@@ -43,7 +46,7 @@
  *
  * To use this allocator, arch code should do the followings.
  *
- * - define CONFIG_HAVE_DYNAMIC_PER_CPU_AREA
+ * - drop CONFIG_HAVE_LEGACY_PER_CPU_AREA
  *
  * - define __addr_to_pcpu_ptr() and __pcpu_ptr_to_addr() to translate
  *   regular address to percpu pointer and back if they need to be
@@ -55,7 +58,9 @@
 
 #include <linux/bitmap.h>
 #include <linux/bootmem.h>
+#include <linux/err.h>
 #include <linux/list.h>
+#include <linux/log2.h>
 #include <linux/mm.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
@@ -89,25 +94,38 @@
 	struct list_head	list;		/* linked to pcpu_slot lists */
 	int			free_size;	/* free bytes in the chunk */
 	int			contig_hint;	/* max contiguous size hint */
-	struct vm_struct	*vm;		/* mapped vmalloc region */
+	void			*base_addr;	/* base address of this chunk */
 	int			map_used;	/* # of map entries used */
 	int			map_alloc;	/* # of map entries allocated */
 	int			*map;		/* allocation map */
+	struct vm_struct	**vms;		/* mapped vmalloc regions */
 	bool			immutable;	/* no [de]population allowed */
-	struct page		**page;		/* points to page array */
-	struct page		*page_ar[];	/* #cpus * UNIT_PAGES */
+	unsigned long		populated[];	/* populated bitmap */
 };
 
 static int pcpu_unit_pages __read_mostly;
 static int pcpu_unit_size __read_mostly;
-static int pcpu_chunk_size __read_mostly;
+static int pcpu_nr_units __read_mostly;
+static int pcpu_atom_size __read_mostly;
 static int pcpu_nr_slots __read_mostly;
 static size_t pcpu_chunk_struct_size __read_mostly;
 
+/* cpus with the lowest and highest unit numbers */
+static unsigned int pcpu_first_unit_cpu __read_mostly;
+static unsigned int pcpu_last_unit_cpu __read_mostly;
+
 /* the address of the first chunk which starts with the kernel static area */
 void *pcpu_base_addr __read_mostly;
 EXPORT_SYMBOL_GPL(pcpu_base_addr);
 
+static const int *pcpu_unit_map __read_mostly;		/* cpu -> unit */
+const unsigned long *pcpu_unit_offsets __read_mostly;	/* cpu -> unit offset */
+
+/* group information, used for vm allocation */
+static int pcpu_nr_groups __read_mostly;
+static const unsigned long *pcpu_group_offsets __read_mostly;
+static const size_t *pcpu_group_sizes __read_mostly;
+
 /*
  * The first chunk which always exists.  Note that unlike other
  * chunks, this one can be allocated and mapped in several different
@@ -129,9 +147,9 @@
  * Synchronization rules.
  *
  * There are two locks - pcpu_alloc_mutex and pcpu_lock.  The former
- * protects allocation/reclaim paths, chunks and chunk->page arrays.
- * The latter is a spinlock and protects the index data structures -
- * chunk slots, chunks and area maps in chunks.
+ * protects allocation/reclaim paths, chunks, populated bitmap and
+ * vmalloc mapping.  The latter is a spinlock and protects the index
+ * data structures - chunk slots, chunks and area maps in chunks.
  *
  * During allocation, pcpu_alloc_mutex is kept locked all the time and
  * pcpu_lock is grabbed and released as necessary.  All actual memory
@@ -178,31 +196,23 @@
 
 static int pcpu_page_idx(unsigned int cpu, int page_idx)
 {
-	return cpu * pcpu_unit_pages + page_idx;
-}
-
-static struct page **pcpu_chunk_pagep(struct pcpu_chunk *chunk,
-				      unsigned int cpu, int page_idx)
-{
-	return &chunk->page[pcpu_page_idx(cpu, page_idx)];
+	return pcpu_unit_map[cpu] * pcpu_unit_pages + page_idx;
 }
 
 static unsigned long pcpu_chunk_addr(struct pcpu_chunk *chunk,
 				     unsigned int cpu, int page_idx)
 {
-	return (unsigned long)chunk->vm->addr +
-		(pcpu_page_idx(cpu, page_idx) << PAGE_SHIFT);
+	return (unsigned long)chunk->base_addr + pcpu_unit_offsets[cpu] +
+		(page_idx << PAGE_SHIFT);
 }
 
-static bool pcpu_chunk_page_occupied(struct pcpu_chunk *chunk,
-				     int page_idx)
+static struct page *pcpu_chunk_page(struct pcpu_chunk *chunk,
+				    unsigned int cpu, int page_idx)
 {
-	/*
-	 * Any possible cpu id can be used here, so there's no need to
-	 * worry about preemption or cpu hotplug.
-	 */
-	return *pcpu_chunk_pagep(chunk, raw_smp_processor_id(),
-				 page_idx) != NULL;
+	/* must not be used on pre-mapped chunk */
+	WARN_ON(chunk->immutable);
+
+	return vmalloc_to_page((void *)pcpu_chunk_addr(chunk, cpu, page_idx));
 }
 
 /* set the pointer to a chunk in a page struct */
@@ -217,6 +227,34 @@
 	return (struct pcpu_chunk *)page->index;
 }
 
+static void pcpu_next_unpop(struct pcpu_chunk *chunk, int *rs, int *re, int end)
+{
+	*rs = find_next_zero_bit(chunk->populated, end, *rs);
+	*re = find_next_bit(chunk->populated, end, *rs + 1);
+}
+
+static void pcpu_next_pop(struct pcpu_chunk *chunk, int *rs, int *re, int end)
+{
+	*rs = find_next_bit(chunk->populated, end, *rs);
+	*re = find_next_zero_bit(chunk->populated, end, *rs + 1);
+}
+
+/*
+ * (Un)populated page region iterators.  Iterate over (un)populated
+ * page regions betwen @start and @end in @chunk.  @rs and @re should
+ * be integer variables and will be set to start and end page index of
+ * the current region.
+ */
+#define pcpu_for_each_unpop_region(chunk, rs, re, start, end)		    \
+	for ((rs) = (start), pcpu_next_unpop((chunk), &(rs), &(re), (end)); \
+	     (rs) < (re);						    \
+	     (rs) = (re) + 1, pcpu_next_unpop((chunk), &(rs), &(re), (end)))
+
+#define pcpu_for_each_pop_region(chunk, rs, re, start, end)		    \
+	for ((rs) = (start), pcpu_next_pop((chunk), &(rs), &(re), (end));   \
+	     (rs) < (re);						    \
+	     (rs) = (re) + 1, pcpu_next_pop((chunk), &(rs), &(re), (end)))
+
 /**
  * pcpu_mem_alloc - allocate memory
  * @size: bytes to allocate
@@ -292,10 +330,10 @@
  */
 static struct pcpu_chunk *pcpu_chunk_addr_search(void *addr)
 {
-	void *first_start = pcpu_first_chunk->vm->addr;
+	void *first_start = pcpu_first_chunk->base_addr;
 
 	/* is it in the first chunk? */
-	if (addr >= first_start && addr < first_start + pcpu_chunk_size) {
+	if (addr >= first_start && addr < first_start + pcpu_unit_size) {
 		/* is it in the reserved area? */
 		if (addr < first_start + pcpu_reserved_chunk_limit)
 			return pcpu_reserved_chunk;
@@ -309,7 +347,7 @@
 	 * space.  Note that any possible cpu id can be used here, so
 	 * there's no need to worry about preemption or cpu hotplug.
 	 */
-	addr += raw_smp_processor_id() * pcpu_unit_size;
+	addr += pcpu_unit_offsets[raw_smp_processor_id()];
 	return pcpu_get_page_chunk(vmalloc_to_page(addr));
 }
 
@@ -558,42 +596,272 @@
 }
 
 /**
- * pcpu_unmap - unmap pages out of a pcpu_chunk
+ * pcpu_get_pages_and_bitmap - get temp pages array and bitmap
  * @chunk: chunk of interest
+ * @bitmapp: output parameter for bitmap
+ * @may_alloc: may allocate the array
+ *
+ * Returns pointer to array of pointers to struct page and bitmap,
+ * both of which can be indexed with pcpu_page_idx().  The returned
+ * array is cleared to zero and *@bitmapp is copied from
+ * @chunk->populated.  Note that there is only one array and bitmap
+ * and access exclusion is the caller's responsibility.
+ *
+ * CONTEXT:
+ * pcpu_alloc_mutex and does GFP_KERNEL allocation if @may_alloc.
+ * Otherwise, don't care.
+ *
+ * RETURNS:
+ * Pointer to temp pages array on success, NULL on failure.
+ */
+static struct page **pcpu_get_pages_and_bitmap(struct pcpu_chunk *chunk,
+					       unsigned long **bitmapp,
+					       bool may_alloc)
+{
+	static struct page **pages;
+	static unsigned long *bitmap;
+	size_t pages_size = pcpu_nr_units * pcpu_unit_pages * sizeof(pages[0]);
+	size_t bitmap_size = BITS_TO_LONGS(pcpu_unit_pages) *
+			     sizeof(unsigned long);
+
+	if (!pages || !bitmap) {
+		if (may_alloc && !pages)
+			pages = pcpu_mem_alloc(pages_size);
+		if (may_alloc && !bitmap)
+			bitmap = pcpu_mem_alloc(bitmap_size);
+		if (!pages || !bitmap)
+			return NULL;
+	}
+
+	memset(pages, 0, pages_size);
+	bitmap_copy(bitmap, chunk->populated, pcpu_unit_pages);
+
+	*bitmapp = bitmap;
+	return pages;
+}
+
+/**
+ * pcpu_free_pages - free pages which were allocated for @chunk
+ * @chunk: chunk pages were allocated for
+ * @pages: array of pages to be freed, indexed by pcpu_page_idx()
+ * @populated: populated bitmap
+ * @page_start: page index of the first page to be freed
+ * @page_end: page index of the last page to be freed + 1
+ *
+ * Free pages [@page_start and @page_end) in @pages for all units.
+ * The pages were allocated for @chunk.
+ */
+static void pcpu_free_pages(struct pcpu_chunk *chunk,
+			    struct page **pages, unsigned long *populated,
+			    int page_start, int page_end)
+{
+	unsigned int cpu;
+	int i;
+
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page *page = pages[pcpu_page_idx(cpu, i)];
+
+			if (page)
+				__free_page(page);
+		}
+	}
+}
+
+/**
+ * pcpu_alloc_pages - allocates pages for @chunk
+ * @chunk: target chunk
+ * @pages: array to put the allocated pages into, indexed by pcpu_page_idx()
+ * @populated: populated bitmap
+ * @page_start: page index of the first page to be allocated
+ * @page_end: page index of the last page to be allocated + 1
+ *
+ * Allocate pages [@page_start,@page_end) into @pages for all units.
+ * The allocation is for @chunk.  Percpu core doesn't care about the
+ * content of @pages and will pass it verbatim to pcpu_map_pages().
+ */
+static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
+			    struct page **pages, unsigned long *populated,
+			    int page_start, int page_end)
+{
+	const gfp_t gfp = GFP_KERNEL | __GFP_HIGHMEM | __GFP_COLD;
+	unsigned int cpu;
+	int i;
+
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page **pagep = &pages[pcpu_page_idx(cpu, i)];
+
+			*pagep = alloc_pages_node(cpu_to_node(cpu), gfp, 0);
+			if (!*pagep) {
+				pcpu_free_pages(chunk, pages, populated,
+						page_start, page_end);
+				return -ENOMEM;
+			}
+		}
+	}
+	return 0;
+}
+
+/**
+ * pcpu_pre_unmap_flush - flush cache prior to unmapping
+ * @chunk: chunk the regions to be flushed belongs to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages in [@page_start,@page_end) of @chunk are about to be
+ * unmapped.  Flush cache.  As each flushing trial can be very
+ * expensive, issue flush on the whole region at once rather than
+ * doing it for each cpu.  This could be an overkill but is more
+ * scalable.
+ */
+static void pcpu_pre_unmap_flush(struct pcpu_chunk *chunk,
+				 int page_start, int page_end)
+{
+	flush_cache_vunmap(
+		pcpu_chunk_addr(chunk, pcpu_first_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_last_unit_cpu, page_end));
+}
+
+static void __pcpu_unmap_pages(unsigned long addr, int nr_pages)
+{
+	unmap_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT);
+}
+
+/**
+ * pcpu_unmap_pages - unmap pages out of a pcpu_chunk
+ * @chunk: chunk of interest
+ * @pages: pages array which can be used to pass information to free
+ * @populated: populated bitmap
  * @page_start: page index of the first page to unmap
  * @page_end: page index of the last page to unmap + 1
- * @flush_tlb: whether to flush tlb or not
  *
  * For each cpu, unmap pages [@page_start,@page_end) out of @chunk.
- * If @flush is true, vcache is flushed before unmapping and tlb
- * after.
+ * Corresponding elements in @pages were cleared by the caller and can
+ * be used to carry information to pcpu_free_pages() which will be
+ * called after all unmaps are finished.  The caller should call
+ * proper pre/post flush functions.
  */
-static void pcpu_unmap(struct pcpu_chunk *chunk, int page_start, int page_end,
-		       bool flush_tlb)
+static void pcpu_unmap_pages(struct pcpu_chunk *chunk,
+			     struct page **pages, unsigned long *populated,
+			     int page_start, int page_end)
 {
-	unsigned int last = nr_cpu_ids - 1;
 	unsigned int cpu;
+	int i;
 
-	/* unmap must not be done on immutable chunk */
-	WARN_ON(chunk->immutable);
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page *page;
 
-	/*
-	 * Each flushing trial can be very expensive, issue flush on
-	 * the whole region at once rather than doing it for each cpu.
-	 * This could be an overkill but is more scalable.
-	 */
-	flush_cache_vunmap(pcpu_chunk_addr(chunk, 0, page_start),
-			   pcpu_chunk_addr(chunk, last, page_end));
+			page = pcpu_chunk_page(chunk, cpu, i);
+			WARN_ON(!page);
+			pages[pcpu_page_idx(cpu, i)] = page;
+		}
+		__pcpu_unmap_pages(pcpu_chunk_addr(chunk, cpu, page_start),
+				   page_end - page_start);
+	}
 
-	for_each_possible_cpu(cpu)
-		unmap_kernel_range_noflush(
-				pcpu_chunk_addr(chunk, cpu, page_start),
-				(page_end - page_start) << PAGE_SHIFT);
+	for (i = page_start; i < page_end; i++)
+		__clear_bit(i, populated);
+}
 
-	/* ditto as flush_cache_vunmap() */
-	if (flush_tlb)
-		flush_tlb_kernel_range(pcpu_chunk_addr(chunk, 0, page_start),
-				       pcpu_chunk_addr(chunk, last, page_end));
+/**
+ * pcpu_post_unmap_tlb_flush - flush TLB after unmapping
+ * @chunk: pcpu_chunk the regions to be flushed belong to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages [@page_start,@page_end) of @chunk have been unmapped.  Flush
+ * TLB for the regions.  This can be skipped if the area is to be
+ * returned to vmalloc as vmalloc will handle TLB flushing lazily.
+ *
+ * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
+ * for the whole region.
+ */
+static void pcpu_post_unmap_tlb_flush(struct pcpu_chunk *chunk,
+				      int page_start, int page_end)
+{
+	flush_tlb_kernel_range(
+		pcpu_chunk_addr(chunk, pcpu_first_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_last_unit_cpu, page_end));
+}
+
+static int __pcpu_map_pages(unsigned long addr, struct page **pages,
+			    int nr_pages)
+{
+	return map_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT,
+					PAGE_KERNEL, pages);
+}
+
+/**
+ * pcpu_map_pages - map pages into a pcpu_chunk
+ * @chunk: chunk of interest
+ * @pages: pages array containing pages to be mapped
+ * @populated: populated bitmap
+ * @page_start: page index of the first page to map
+ * @page_end: page index of the last page to map + 1
+ *
+ * For each cpu, map pages [@page_start,@page_end) into @chunk.  The
+ * caller is responsible for calling pcpu_post_map_flush() after all
+ * mappings are complete.
+ *
+ * This function is responsible for setting corresponding bits in
+ * @chunk->populated bitmap and whatever is necessary for reverse
+ * lookup (addr -> chunk).
+ */
+static int pcpu_map_pages(struct pcpu_chunk *chunk,
+			  struct page **pages, unsigned long *populated,
+			  int page_start, int page_end)
+{
+	unsigned int cpu, tcpu;
+	int i, err;
+
+	for_each_possible_cpu(cpu) {
+		err = __pcpu_map_pages(pcpu_chunk_addr(chunk, cpu, page_start),
+				       &pages[pcpu_page_idx(cpu, page_start)],
+				       page_end - page_start);
+		if (err < 0)
+			goto err;
+	}
+
+	/* mapping successful, link chunk and mark populated */
+	for (i = page_start; i < page_end; i++) {
+		for_each_possible_cpu(cpu)
+			pcpu_set_page_chunk(pages[pcpu_page_idx(cpu, i)],
+					    chunk);
+		__set_bit(i, populated);
+	}
+
+	return 0;
+
+err:
+	for_each_possible_cpu(tcpu) {
+		if (tcpu == cpu)
+			break;
+		__pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
+				   page_end - page_start);
+	}
+	return err;
+}
+
+/**
+ * pcpu_post_map_flush - flush cache after mapping
+ * @chunk: pcpu_chunk the regions to be flushed belong to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages [@page_start,@page_end) of @chunk have been mapped.  Flush
+ * cache.
+ *
+ * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
+ * for the whole region.
+ */
+static void pcpu_post_map_flush(struct pcpu_chunk *chunk,
+				int page_start, int page_end)
+{
+	flush_cache_vmap(
+		pcpu_chunk_addr(chunk, pcpu_first_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_last_unit_cpu, page_end));
 }
 
 /**
@@ -610,73 +878,45 @@
  * CONTEXT:
  * pcpu_alloc_mutex.
  */
-static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk, int off, int size,
-				  bool flush)
+static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk, int off, int size)
 {
 	int page_start = PFN_DOWN(off);
 	int page_end = PFN_UP(off + size);
-	int unmap_start = -1;
-	int uninitialized_var(unmap_end);
-	unsigned int cpu;
-	int i;
+	struct page **pages;
+	unsigned long *populated;
+	int rs, re;
 
-	for (i = page_start; i < page_end; i++) {
-		for_each_possible_cpu(cpu) {
-			struct page **pagep = pcpu_chunk_pagep(chunk, cpu, i);
-
-			if (!*pagep)
-				continue;
-
-			__free_page(*pagep);
-
-			/*
-			 * If it's partial depopulation, it might get
-			 * populated or depopulated again.  Mark the
-			 * page gone.
-			 */
-			*pagep = NULL;
-
-			unmap_start = unmap_start < 0 ? i : unmap_start;
-			unmap_end = i + 1;
-		}
+	/* quick path, check whether it's empty already */
+	pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
+		if (rs == page_start && re == page_end)
+			return;
+		break;
 	}
 
-	if (unmap_start >= 0)
-		pcpu_unmap(chunk, unmap_start, unmap_end, flush);
-}
-
-/**
- * pcpu_map - map pages into a pcpu_chunk
- * @chunk: chunk of interest
- * @page_start: page index of the first page to map
- * @page_end: page index of the last page to map + 1
- *
- * For each cpu, map pages [@page_start,@page_end) into @chunk.
- * vcache is flushed afterwards.
- */
-static int pcpu_map(struct pcpu_chunk *chunk, int page_start, int page_end)
-{
-	unsigned int last = nr_cpu_ids - 1;
-	unsigned int cpu;
-	int err;
-
-	/* map must not be done on immutable chunk */
+	/* immutable chunks can't be depopulated */
 	WARN_ON(chunk->immutable);
 
-	for_each_possible_cpu(cpu) {
-		err = map_kernel_range_noflush(
-				pcpu_chunk_addr(chunk, cpu, page_start),
-				(page_end - page_start) << PAGE_SHIFT,
-				PAGE_KERNEL,
-				pcpu_chunk_pagep(chunk, cpu, page_start));
-		if (err < 0)
-			return err;
-	}
+	/*
+	 * If control reaches here, there must have been at least one
+	 * successful population attempt so the temp pages array must
+	 * be available now.
+	 */
+	pages = pcpu_get_pages_and_bitmap(chunk, &populated, false);
+	BUG_ON(!pages);
 
-	/* flush at once, please read comments in pcpu_unmap() */
-	flush_cache_vmap(pcpu_chunk_addr(chunk, 0, page_start),
-			 pcpu_chunk_addr(chunk, last, page_end));
-	return 0;
+	/* unmap and free */
+	pcpu_pre_unmap_flush(chunk, page_start, page_end);
+
+	pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
+		pcpu_unmap_pages(chunk, pages, populated, rs, re);
+
+	/* no need to flush tlb, vmalloc will handle it lazily */
+
+	pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
+		pcpu_free_pages(chunk, pages, populated, rs, re);
+
+	/* commit new bitmap */
+	bitmap_copy(chunk->populated, populated, pcpu_unit_pages);
 }
 
 /**
@@ -693,58 +933,68 @@
  */
 static int pcpu_populate_chunk(struct pcpu_chunk *chunk, int off, int size)
 {
-	const gfp_t alloc_mask = GFP_KERNEL | __GFP_HIGHMEM | __GFP_COLD;
 	int page_start = PFN_DOWN(off);
 	int page_end = PFN_UP(off + size);
-	int map_start = -1;
-	int uninitialized_var(map_end);
+	int free_end = page_start, unmap_end = page_start;
+	struct page **pages;
+	unsigned long *populated;
 	unsigned int cpu;
-	int i;
+	int rs, re, rc;
 
-	for (i = page_start; i < page_end; i++) {
-		if (pcpu_chunk_page_occupied(chunk, i)) {
-			if (map_start >= 0) {
-				if (pcpu_map(chunk, map_start, map_end))
-					goto err;
-				map_start = -1;
-			}
-			continue;
-		}
-
-		map_start = map_start < 0 ? i : map_start;
-		map_end = i + 1;
-
-		for_each_possible_cpu(cpu) {
-			struct page **pagep = pcpu_chunk_pagep(chunk, cpu, i);
-
-			*pagep = alloc_pages_node(cpu_to_node(cpu),
-						  alloc_mask, 0);
-			if (!*pagep)
-				goto err;
-			pcpu_set_page_chunk(*pagep, chunk);
-		}
+	/* quick path, check whether all pages are already there */
+	pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end) {
+		if (rs == page_start && re == page_end)
+			goto clear;
+		break;
 	}
 
-	if (map_start >= 0 && pcpu_map(chunk, map_start, map_end))
-		goto err;
+	/* need to allocate and map pages, this chunk can't be immutable */
+	WARN_ON(chunk->immutable);
 
+	pages = pcpu_get_pages_and_bitmap(chunk, &populated, true);
+	if (!pages)
+		return -ENOMEM;
+
+	/* alloc and map */
+	pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
+		rc = pcpu_alloc_pages(chunk, pages, populated, rs, re);
+		if (rc)
+			goto err_free;
+		free_end = re;
+	}
+
+	pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
+		rc = pcpu_map_pages(chunk, pages, populated, rs, re);
+		if (rc)
+			goto err_unmap;
+		unmap_end = re;
+	}
+	pcpu_post_map_flush(chunk, page_start, page_end);
+
+	/* commit new bitmap */
+	bitmap_copy(chunk->populated, populated, pcpu_unit_pages);
+clear:
 	for_each_possible_cpu(cpu)
-		memset(chunk->vm->addr + cpu * pcpu_unit_size + off, 0,
-		       size);
-
+		memset((void *)pcpu_chunk_addr(chunk, cpu, 0) + off, 0, size);
 	return 0;
-err:
-	/* likely under heavy memory pressure, give memory back */
-	pcpu_depopulate_chunk(chunk, off, size, true);
-	return -ENOMEM;
+
+err_unmap:
+	pcpu_pre_unmap_flush(chunk, page_start, unmap_end);
+	pcpu_for_each_unpop_region(chunk, rs, re, page_start, unmap_end)
+		pcpu_unmap_pages(chunk, pages, populated, rs, re);
+	pcpu_post_unmap_tlb_flush(chunk, page_start, unmap_end);
+err_free:
+	pcpu_for_each_unpop_region(chunk, rs, re, page_start, free_end)
+		pcpu_free_pages(chunk, pages, populated, rs, re);
+	return rc;
 }
 
 static void free_pcpu_chunk(struct pcpu_chunk *chunk)
 {
 	if (!chunk)
 		return;
-	if (chunk->vm)
-		free_vm_area(chunk->vm);
+	if (chunk->vms)
+		pcpu_free_vm_areas(chunk->vms, pcpu_nr_groups);
 	pcpu_mem_free(chunk->map, chunk->map_alloc * sizeof(chunk->map[0]));
 	kfree(chunk);
 }
@@ -760,10 +1010,11 @@
 	chunk->map = pcpu_mem_alloc(PCPU_DFL_MAP_ALLOC * sizeof(chunk->map[0]));
 	chunk->map_alloc = PCPU_DFL_MAP_ALLOC;
 	chunk->map[chunk->map_used++] = pcpu_unit_size;
-	chunk->page = chunk->page_ar;
 
-	chunk->vm = get_vm_area(pcpu_chunk_size, VM_ALLOC);
-	if (!chunk->vm) {
+	chunk->vms = pcpu_get_vm_areas(pcpu_group_offsets, pcpu_group_sizes,
+				       pcpu_nr_groups, pcpu_atom_size,
+				       GFP_KERNEL);
+	if (!chunk->vms) {
 		free_pcpu_chunk(chunk);
 		return NULL;
 	}
@@ -771,6 +1022,7 @@
 	INIT_LIST_HEAD(&chunk->list);
 	chunk->free_size = pcpu_unit_size;
 	chunk->contig_hint = pcpu_unit_size;
+	chunk->base_addr = chunk->vms[0]->addr - pcpu_group_offsets[0];
 
 	return chunk;
 }
@@ -860,7 +1112,8 @@
 
 	mutex_unlock(&pcpu_alloc_mutex);
 
-	return __addr_to_pcpu_ptr(chunk->vm->addr + off);
+	/* return address relative to base address */
+	return __addr_to_pcpu_ptr(chunk->base_addr + off);
 
 fail_unlock:
 	spin_unlock_irq(&pcpu_lock);
@@ -938,12 +1191,13 @@
 	}
 
 	spin_unlock_irq(&pcpu_lock);
-	mutex_unlock(&pcpu_alloc_mutex);
 
 	list_for_each_entry_safe(chunk, next, &todo, list) {
-		pcpu_depopulate_chunk(chunk, 0, pcpu_unit_size, false);
+		pcpu_depopulate_chunk(chunk, 0, pcpu_unit_size);
 		free_pcpu_chunk(chunk);
 	}
+
+	mutex_unlock(&pcpu_alloc_mutex);
 }
 
 /**
@@ -968,7 +1222,7 @@
 	spin_lock_irqsave(&pcpu_lock, flags);
 
 	chunk = pcpu_chunk_addr_search(addr);
-	off = addr - chunk->vm->addr;
+	off = addr - chunk->base_addr;
 
 	pcpu_free_area(chunk, off);
 
@@ -987,30 +1241,295 @@
 }
 EXPORT_SYMBOL_GPL(free_percpu);
 
+static inline size_t pcpu_calc_fc_sizes(size_t static_size,
+					size_t reserved_size,
+					ssize_t *dyn_sizep)
+{
+	size_t size_sum;
+
+	size_sum = PFN_ALIGN(static_size + reserved_size +
+			     (*dyn_sizep >= 0 ? *dyn_sizep : 0));
+	if (*dyn_sizep != 0)
+		*dyn_sizep = size_sum - static_size - reserved_size;
+
+	return size_sum;
+}
+
 /**
- * pcpu_setup_first_chunk - initialize the first percpu chunk
- * @get_page_fn: callback to fetch page pointer
- * @static_size: the size of static percpu area in bytes
+ * pcpu_alloc_alloc_info - allocate percpu allocation info
+ * @nr_groups: the number of groups
+ * @nr_units: the number of units
+ *
+ * Allocate ai which is large enough for @nr_groups groups containing
+ * @nr_units units.  The returned ai's groups[0].cpu_map points to the
+ * cpu_map array which is long enough for @nr_units and filled with
+ * NR_CPUS.  It's the caller's responsibility to initialize cpu_map
+ * pointer of other groups.
+ *
+ * RETURNS:
+ * Pointer to the allocated pcpu_alloc_info on success, NULL on
+ * failure.
+ */
+struct pcpu_alloc_info * __init pcpu_alloc_alloc_info(int nr_groups,
+						      int nr_units)
+{
+	struct pcpu_alloc_info *ai;
+	size_t base_size, ai_size;
+	void *ptr;
+	int unit;
+
+	base_size = ALIGN(sizeof(*ai) + nr_groups * sizeof(ai->groups[0]),
+			  __alignof__(ai->groups[0].cpu_map[0]));
+	ai_size = base_size + nr_units * sizeof(ai->groups[0].cpu_map[0]);
+
+	ptr = alloc_bootmem_nopanic(PFN_ALIGN(ai_size));
+	if (!ptr)
+		return NULL;
+	ai = ptr;
+	ptr += base_size;
+
+	ai->groups[0].cpu_map = ptr;
+
+	for (unit = 0; unit < nr_units; unit++)
+		ai->groups[0].cpu_map[unit] = NR_CPUS;
+
+	ai->nr_groups = nr_groups;
+	ai->__ai_size = PFN_ALIGN(ai_size);
+
+	return ai;
+}
+
+/**
+ * pcpu_free_alloc_info - free percpu allocation info
+ * @ai: pcpu_alloc_info to free
+ *
+ * Free @ai which was allocated by pcpu_alloc_alloc_info().
+ */
+void __init pcpu_free_alloc_info(struct pcpu_alloc_info *ai)
+{
+	free_bootmem(__pa(ai), ai->__ai_size);
+}
+
+/**
+ * pcpu_build_alloc_info - build alloc_info considering distances between CPUs
  * @reserved_size: the size of reserved percpu area in bytes
  * @dyn_size: free size for dynamic allocation in bytes, -1 for auto
- * @unit_size: unit size in bytes, must be multiple of PAGE_SIZE, -1 for auto
- * @base_addr: mapped address, NULL for auto
- * @populate_pte_fn: callback to allocate pagetable, NULL if unnecessary
+ * @atom_size: allocation atom size
+ * @cpu_distance_fn: callback to determine distance between cpus, optional
+ *
+ * This function determines grouping of units, their mappings to cpus
+ * and other parameters considering needed percpu size, allocation
+ * atom size and distances between CPUs.
+ *
+ * Groups are always mutliples of atom size and CPUs which are of
+ * LOCAL_DISTANCE both ways are grouped together and share space for
+ * units in the same group.  The returned configuration is guaranteed
+ * to have CPUs on different nodes on different groups and >=75% usage
+ * of allocated virtual address space.
+ *
+ * RETURNS:
+ * On success, pointer to the new allocation_info is returned.  On
+ * failure, ERR_PTR value is returned.
+ */
+struct pcpu_alloc_info * __init pcpu_build_alloc_info(
+				size_t reserved_size, ssize_t dyn_size,
+				size_t atom_size,
+				pcpu_fc_cpu_distance_fn_t cpu_distance_fn)
+{
+	static int group_map[NR_CPUS] __initdata;
+	static int group_cnt[NR_CPUS] __initdata;
+	const size_t static_size = __per_cpu_end - __per_cpu_start;
+	int group_cnt_max = 0, nr_groups = 1, nr_units = 0;
+	size_t size_sum, min_unit_size, alloc_size;
+	int upa, max_upa, uninitialized_var(best_upa);	/* units_per_alloc */
+	int last_allocs, group, unit;
+	unsigned int cpu, tcpu;
+	struct pcpu_alloc_info *ai;
+	unsigned int *cpu_map;
+
+	/*
+	 * Determine min_unit_size, alloc_size and max_upa such that
+	 * alloc_size is multiple of atom_size and is the smallest
+	 * which can accomodate 4k aligned segments which are equal to
+	 * or larger than min_unit_size.
+	 */
+	size_sum = pcpu_calc_fc_sizes(static_size, reserved_size, &dyn_size);
+	min_unit_size = max_t(size_t, size_sum, PCPU_MIN_UNIT_SIZE);
+
+	alloc_size = roundup(min_unit_size, atom_size);
+	upa = alloc_size / min_unit_size;
+	while (alloc_size % upa || ((alloc_size / upa) & ~PAGE_MASK))
+		upa--;
+	max_upa = upa;
+
+	/* group cpus according to their proximity */
+	for_each_possible_cpu(cpu) {
+		group = 0;
+	next_group:
+		for_each_possible_cpu(tcpu) {
+			if (cpu == tcpu)
+				break;
+			if (group_map[tcpu] == group && cpu_distance_fn &&
+			    (cpu_distance_fn(cpu, tcpu) > LOCAL_DISTANCE ||
+			     cpu_distance_fn(tcpu, cpu) > LOCAL_DISTANCE)) {
+				group++;
+				nr_groups = max(nr_groups, group + 1);
+				goto next_group;
+			}
+		}
+		group_map[cpu] = group;
+		group_cnt[group]++;
+		group_cnt_max = max(group_cnt_max, group_cnt[group]);
+	}
+
+	/*
+	 * Expand unit size until address space usage goes over 75%
+	 * and then as much as possible without using more address
+	 * space.
+	 */
+	last_allocs = INT_MAX;
+	for (upa = max_upa; upa; upa--) {
+		int allocs = 0, wasted = 0;
+
+		if (alloc_size % upa || ((alloc_size / upa) & ~PAGE_MASK))
+			continue;
+
+		for (group = 0; group < nr_groups; group++) {
+			int this_allocs = DIV_ROUND_UP(group_cnt[group], upa);
+			allocs += this_allocs;
+			wasted += this_allocs * upa - group_cnt[group];
+		}
+
+		/*
+		 * Don't accept if wastage is over 25%.  The
+		 * greater-than comparison ensures upa==1 always
+		 * passes the following check.
+		 */
+		if (wasted > num_possible_cpus() / 3)
+			continue;
+
+		/* and then don't consume more memory */
+		if (allocs > last_allocs)
+			break;
+		last_allocs = allocs;
+		best_upa = upa;
+	}
+	upa = best_upa;
+
+	/* allocate and fill alloc_info */
+	for (group = 0; group < nr_groups; group++)
+		nr_units += roundup(group_cnt[group], upa);
+
+	ai = pcpu_alloc_alloc_info(nr_groups, nr_units);
+	if (!ai)
+		return ERR_PTR(-ENOMEM);
+	cpu_map = ai->groups[0].cpu_map;
+
+	for (group = 0; group < nr_groups; group++) {
+		ai->groups[group].cpu_map = cpu_map;
+		cpu_map += roundup(group_cnt[group], upa);
+	}
+
+	ai->static_size = static_size;
+	ai->reserved_size = reserved_size;
+	ai->dyn_size = dyn_size;
+	ai->unit_size = alloc_size / upa;
+	ai->atom_size = atom_size;
+	ai->alloc_size = alloc_size;
+
+	for (group = 0, unit = 0; group_cnt[group]; group++) {
+		struct pcpu_group_info *gi = &ai->groups[group];
+
+		/*
+		 * Initialize base_offset as if all groups are located
+		 * back-to-back.  The caller should update this to
+		 * reflect actual allocation.
+		 */
+		gi->base_offset = unit * ai->unit_size;
+
+		for_each_possible_cpu(cpu)
+			if (group_map[cpu] == group)
+				gi->cpu_map[gi->nr_units++] = cpu;
+		gi->nr_units = roundup(gi->nr_units, upa);
+		unit += gi->nr_units;
+	}
+	BUG_ON(unit != nr_units);
+
+	return ai;
+}
+
+/**
+ * pcpu_dump_alloc_info - print out information about pcpu_alloc_info
+ * @lvl: loglevel
+ * @ai: allocation info to dump
+ *
+ * Print out information about @ai using loglevel @lvl.
+ */
+static void pcpu_dump_alloc_info(const char *lvl,
+				 const struct pcpu_alloc_info *ai)
+{
+	int group_width = 1, cpu_width = 1, width;
+	char empty_str[] = "--------";
+	int alloc = 0, alloc_end = 0;
+	int group, v;
+	int upa, apl;	/* units per alloc, allocs per line */
+
+	v = ai->nr_groups;
+	while (v /= 10)
+		group_width++;
+
+	v = num_possible_cpus();
+	while (v /= 10)
+		cpu_width++;
+	empty_str[min_t(int, cpu_width, sizeof(empty_str) - 1)] = '\0';
+
+	upa = ai->alloc_size / ai->unit_size;
+	width = upa * (cpu_width + 1) + group_width + 3;
+	apl = rounddown_pow_of_two(max(60 / width, 1));
+
+	printk("%spcpu-alloc: s%zu r%zu d%zu u%zu alloc=%zu*%zu",
+	       lvl, ai->static_size, ai->reserved_size, ai->dyn_size,
+	       ai->unit_size, ai->alloc_size / ai->atom_size, ai->atom_size);
+
+	for (group = 0; group < ai->nr_groups; group++) {
+		const struct pcpu_group_info *gi = &ai->groups[group];
+		int unit = 0, unit_end = 0;
+
+		BUG_ON(gi->nr_units % upa);
+		for (alloc_end += gi->nr_units / upa;
+		     alloc < alloc_end; alloc++) {
+			if (!(alloc % apl)) {
+				printk("\n");
+				printk("%spcpu-alloc: ", lvl);
+			}
+			printk("[%0*d] ", group_width, group);
+
+			for (unit_end += upa; unit < unit_end; unit++)
+				if (gi->cpu_map[unit] != NR_CPUS)
+					printk("%0*d ", cpu_width,
+					       gi->cpu_map[unit]);
+				else
+					printk("%s ", empty_str);
+		}
+	}
+	printk("\n");
+}
+
+/**
+ * pcpu_setup_first_chunk - initialize the first percpu chunk
+ * @ai: pcpu_alloc_info describing how to percpu area is shaped
+ * @base_addr: mapped address
  *
  * Initialize the first percpu chunk which contains the kernel static
  * perpcu area.  This function is to be called from arch percpu area
- * setup path.  The first two parameters are mandatory.  The rest are
- * optional.
+ * setup path.
  *
- * @get_page_fn() should return pointer to percpu page given cpu
- * number and page number.  It should at least return enough pages to
- * cover the static area.  The returned pages for static area should
- * have been initialized with valid data.  If @unit_size is specified,
- * it can also return pages after the static area.  NULL return
- * indicates end of pages for the cpu.  Note that @get_page_fn() must
- * return the same number of pages for all cpus.
+ * @ai contains all information necessary to initialize the first
+ * chunk and prime the dynamic percpu allocator.
  *
- * @reserved_size, if non-zero, specifies the amount of bytes to
+ * @ai->static_size is the size of static percpu area.
+ *
+ * @ai->reserved_size, if non-zero, specifies the amount of bytes to
  * reserve after the static area in the first chunk.  This reserves
  * the first chunk such that it's available only through reserved
  * percpu allocation.  This is primarily used to serve module percpu
@@ -1018,22 +1537,29 @@
  * limited offset range for symbol relocations to guarantee module
  * percpu symbols fall inside the relocatable range.
  *
- * @dyn_size, if non-negative, determines the number of bytes
- * available for dynamic allocation in the first chunk.  Specifying
- * non-negative value makes percpu leave alone the area beyond
- * @static_size + @reserved_size + @dyn_size.
+ * @ai->dyn_size determines the number of bytes available for dynamic
+ * allocation in the first chunk.  The area between @ai->static_size +
+ * @ai->reserved_size + @ai->dyn_size and @ai->unit_size is unused.
  *
- * @unit_size, if non-negative, specifies unit size and must be
- * aligned to PAGE_SIZE and equal to or larger than @static_size +
- * @reserved_size + if non-negative, @dyn_size.
+ * @ai->unit_size specifies unit size and must be aligned to PAGE_SIZE
+ * and equal to or larger than @ai->static_size + @ai->reserved_size +
+ * @ai->dyn_size.
  *
- * Non-null @base_addr means that the caller already allocated virtual
- * region for the first chunk and mapped it.  percpu must not mess
- * with the chunk.  Note that @base_addr with 0 @unit_size or non-NULL
- * @populate_pte_fn doesn't make any sense.
+ * @ai->atom_size is the allocation atom size and used as alignment
+ * for vm areas.
  *
- * @populate_pte_fn is used to populate the pagetable.  NULL means the
- * caller already populated the pagetable.
+ * @ai->alloc_size is the allocation size and always multiple of
+ * @ai->atom_size.  This is larger than @ai->atom_size if
+ * @ai->unit_size is larger than @ai->atom_size.
+ *
+ * @ai->nr_groups and @ai->groups describe virtual memory layout of
+ * percpu areas.  Units which should be colocated are put into the
+ * same group.  Dynamic VM areas will be allocated according to these
+ * groupings.  If @ai->nr_groups is zero, a single group containing
+ * all units is assumed.
+ *
+ * The caller should have mapped the first chunk at @base_addr and
+ * copied static data to each unit.
  *
  * If the first chunk ends up with both reserved and dynamic areas, it
  * is served by two chunks - one to serve the core static and reserved
@@ -1043,49 +1569,83 @@
  * and available for dynamic allocation like any other chunks.
  *
  * RETURNS:
- * The determined pcpu_unit_size which can be used to initialize
- * percpu access.
+ * 0 on success, -errno on failure.
  */
-size_t __init pcpu_setup_first_chunk(pcpu_get_page_fn_t get_page_fn,
-				     size_t static_size, size_t reserved_size,
-				     ssize_t dyn_size, ssize_t unit_size,
-				     void *base_addr,
-				     pcpu_populate_pte_fn_t populate_pte_fn)
+int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
+				  void *base_addr)
 {
-	static struct vm_struct first_vm;
 	static int smap[2], dmap[2];
-	size_t size_sum = static_size + reserved_size +
-			  (dyn_size >= 0 ? dyn_size : 0);
+	size_t dyn_size = ai->dyn_size;
+	size_t size_sum = ai->static_size + ai->reserved_size + dyn_size;
 	struct pcpu_chunk *schunk, *dchunk = NULL;
+	unsigned long *group_offsets;
+	size_t *group_sizes;
+	unsigned long *unit_off;
 	unsigned int cpu;
-	int nr_pages;
-	int err, i;
+	int *unit_map;
+	int group, unit, i;
 
-	/* santiy checks */
+	/* sanity checks */
 	BUILD_BUG_ON(ARRAY_SIZE(smap) >= PCPU_DFL_MAP_ALLOC ||
 		     ARRAY_SIZE(dmap) >= PCPU_DFL_MAP_ALLOC);
-	BUG_ON(!static_size);
-	if (unit_size >= 0) {
-		BUG_ON(unit_size < size_sum);
-		BUG_ON(unit_size & ~PAGE_MASK);
-		BUG_ON(unit_size < PCPU_MIN_UNIT_SIZE);
-	} else
-		BUG_ON(base_addr);
-	BUG_ON(base_addr && populate_pte_fn);
+	BUG_ON(ai->nr_groups <= 0);
+	BUG_ON(!ai->static_size);
+	BUG_ON(!base_addr);
+	BUG_ON(ai->unit_size < size_sum);
+	BUG_ON(ai->unit_size & ~PAGE_MASK);
+	BUG_ON(ai->unit_size < PCPU_MIN_UNIT_SIZE);
 
-	if (unit_size >= 0)
-		pcpu_unit_pages = unit_size >> PAGE_SHIFT;
-	else
-		pcpu_unit_pages = max_t(int, PCPU_MIN_UNIT_SIZE >> PAGE_SHIFT,
-					PFN_UP(size_sum));
+	pcpu_dump_alloc_info(KERN_DEBUG, ai);
 
+	/* process group information and build config tables accordingly */
+	group_offsets = alloc_bootmem(ai->nr_groups * sizeof(group_offsets[0]));
+	group_sizes = alloc_bootmem(ai->nr_groups * sizeof(group_sizes[0]));
+	unit_map = alloc_bootmem(nr_cpu_ids * sizeof(unit_map[0]));
+	unit_off = alloc_bootmem(nr_cpu_ids * sizeof(unit_off[0]));
+
+	for (cpu = 0; cpu < nr_cpu_ids; cpu++)
+		unit_map[cpu] = NR_CPUS;
+	pcpu_first_unit_cpu = NR_CPUS;
+
+	for (group = 0, unit = 0; group < ai->nr_groups; group++, unit += i) {
+		const struct pcpu_group_info *gi = &ai->groups[group];
+
+		group_offsets[group] = gi->base_offset;
+		group_sizes[group] = gi->nr_units * ai->unit_size;
+
+		for (i = 0; i < gi->nr_units; i++) {
+			cpu = gi->cpu_map[i];
+			if (cpu == NR_CPUS)
+				continue;
+
+			BUG_ON(cpu > nr_cpu_ids || !cpu_possible(cpu));
+			BUG_ON(unit_map[cpu] != NR_CPUS);
+
+			unit_map[cpu] = unit + i;
+			unit_off[cpu] = gi->base_offset + i * ai->unit_size;
+
+			if (pcpu_first_unit_cpu == NR_CPUS)
+				pcpu_first_unit_cpu = cpu;
+		}
+	}
+	pcpu_last_unit_cpu = cpu;
+	pcpu_nr_units = unit;
+
+	for_each_possible_cpu(cpu)
+		BUG_ON(unit_map[cpu] == NR_CPUS);
+
+	pcpu_nr_groups = ai->nr_groups;
+	pcpu_group_offsets = group_offsets;
+	pcpu_group_sizes = group_sizes;
+	pcpu_unit_map = unit_map;
+	pcpu_unit_offsets = unit_off;
+
+	/* determine basic parameters */
+	pcpu_unit_pages = ai->unit_size >> PAGE_SHIFT;
 	pcpu_unit_size = pcpu_unit_pages << PAGE_SHIFT;
-	pcpu_chunk_size = nr_cpu_ids * pcpu_unit_size;
-	pcpu_chunk_struct_size = sizeof(struct pcpu_chunk)
-		+ nr_cpu_ids * pcpu_unit_pages * sizeof(struct page *);
-
-	if (dyn_size < 0)
-		dyn_size = pcpu_unit_size - static_size - reserved_size;
+	pcpu_atom_size = ai->atom_size;
+	pcpu_chunk_struct_size = sizeof(struct pcpu_chunk) +
+		BITS_TO_LONGS(pcpu_unit_pages) * sizeof(unsigned long);
 
 	/*
 	 * Allocate chunk slots.  The additional last slot is for
@@ -1105,189 +1665,351 @@
 	 */
 	schunk = alloc_bootmem(pcpu_chunk_struct_size);
 	INIT_LIST_HEAD(&schunk->list);
-	schunk->vm = &first_vm;
+	schunk->base_addr = base_addr;
 	schunk->map = smap;
 	schunk->map_alloc = ARRAY_SIZE(smap);
-	schunk->page = schunk->page_ar;
+	schunk->immutable = true;
+	bitmap_fill(schunk->populated, pcpu_unit_pages);
 
-	if (reserved_size) {
-		schunk->free_size = reserved_size;
+	if (ai->reserved_size) {
+		schunk->free_size = ai->reserved_size;
 		pcpu_reserved_chunk = schunk;
-		pcpu_reserved_chunk_limit = static_size + reserved_size;
+		pcpu_reserved_chunk_limit = ai->static_size + ai->reserved_size;
 	} else {
 		schunk->free_size = dyn_size;
 		dyn_size = 0;			/* dynamic area covered */
 	}
 	schunk->contig_hint = schunk->free_size;
 
-	schunk->map[schunk->map_used++] = -static_size;
+	schunk->map[schunk->map_used++] = -ai->static_size;
 	if (schunk->free_size)
 		schunk->map[schunk->map_used++] = schunk->free_size;
 
 	/* init dynamic chunk if necessary */
 	if (dyn_size) {
-		dchunk = alloc_bootmem(sizeof(struct pcpu_chunk));
+		dchunk = alloc_bootmem(pcpu_chunk_struct_size);
 		INIT_LIST_HEAD(&dchunk->list);
-		dchunk->vm = &first_vm;
+		dchunk->base_addr = base_addr;
 		dchunk->map = dmap;
 		dchunk->map_alloc = ARRAY_SIZE(dmap);
-		dchunk->page = schunk->page_ar;	/* share page map with schunk */
+		dchunk->immutable = true;
+		bitmap_fill(dchunk->populated, pcpu_unit_pages);
 
 		dchunk->contig_hint = dchunk->free_size = dyn_size;
 		dchunk->map[dchunk->map_used++] = -pcpu_reserved_chunk_limit;
 		dchunk->map[dchunk->map_used++] = dchunk->free_size;
 	}
 
-	/* allocate vm address */
-	first_vm.flags = VM_ALLOC;
-	first_vm.size = pcpu_chunk_size;
-
-	if (!base_addr)
-		vm_area_register_early(&first_vm, PAGE_SIZE);
-	else {
-		/*
-		 * Pages already mapped.  No need to remap into
-		 * vmalloc area.  In this case the first chunks can't
-		 * be mapped or unmapped by percpu and are marked
-		 * immutable.
-		 */
-		first_vm.addr = base_addr;
-		schunk->immutable = true;
-		if (dchunk)
-			dchunk->immutable = true;
-	}
-
-	/* assign pages */
-	nr_pages = -1;
-	for_each_possible_cpu(cpu) {
-		for (i = 0; i < pcpu_unit_pages; i++) {
-			struct page *page = get_page_fn(cpu, i);
-
-			if (!page)
-				break;
-			*pcpu_chunk_pagep(schunk, cpu, i) = page;
-		}
-
-		BUG_ON(i < PFN_UP(static_size));
-
-		if (nr_pages < 0)
-			nr_pages = i;
-		else
-			BUG_ON(nr_pages != i);
-	}
-
-	/* map them */
-	if (populate_pte_fn) {
-		for_each_possible_cpu(cpu)
-			for (i = 0; i < nr_pages; i++)
-				populate_pte_fn(pcpu_chunk_addr(schunk,
-								cpu, i));
-
-		err = pcpu_map(schunk, 0, nr_pages);
-		if (err)
-			panic("failed to setup static percpu area, err=%d\n",
-			      err);
-	}
-
 	/* link the first chunk in */
 	pcpu_first_chunk = dchunk ?: schunk;
 	pcpu_chunk_relocate(pcpu_first_chunk, -1);
 
 	/* we're done */
-	pcpu_base_addr = (void *)pcpu_chunk_addr(schunk, 0, 0);
-	return pcpu_unit_size;
+	pcpu_base_addr = base_addr;
+	return 0;
 }
 
-/*
- * Embedding first chunk setup helper.
- */
-static void *pcpue_ptr __initdata;
-static size_t pcpue_size __initdata;
-static size_t pcpue_unit_size __initdata;
+const char *pcpu_fc_names[PCPU_FC_NR] __initdata = {
+	[PCPU_FC_AUTO]	= "auto",
+	[PCPU_FC_EMBED]	= "embed",
+	[PCPU_FC_PAGE]	= "page",
+};
 
-static struct page * __init pcpue_get_page(unsigned int cpu, int pageno)
+enum pcpu_fc pcpu_chosen_fc __initdata = PCPU_FC_AUTO;
+
+static int __init percpu_alloc_setup(char *str)
 {
-	size_t off = (size_t)pageno << PAGE_SHIFT;
+	if (0)
+		/* nada */;
+#ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
+	else if (!strcmp(str, "embed"))
+		pcpu_chosen_fc = PCPU_FC_EMBED;
+#endif
+#ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
+	else if (!strcmp(str, "page"))
+		pcpu_chosen_fc = PCPU_FC_PAGE;
+#endif
+	else
+		pr_warning("PERCPU: unknown allocator %s specified\n", str);
 
-	if (off >= pcpue_size)
-		return NULL;
-
-	return virt_to_page(pcpue_ptr + cpu * pcpue_unit_size + off);
+	return 0;
 }
+early_param("percpu_alloc", percpu_alloc_setup);
 
+#if defined(CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK) || \
+	!defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
 /**
  * pcpu_embed_first_chunk - embed the first percpu chunk into bootmem
- * @static_size: the size of static percpu area in bytes
  * @reserved_size: the size of reserved percpu area in bytes
  * @dyn_size: free size for dynamic allocation in bytes, -1 for auto
- * @unit_size: unit size in bytes, must be multiple of PAGE_SIZE, -1 for auto
+ * @atom_size: allocation atom size
+ * @cpu_distance_fn: callback to determine distance between cpus, optional
+ * @alloc_fn: function to allocate percpu page
+ * @free_fn: funtion to free percpu page
  *
  * This is a helper to ease setting up embedded first percpu chunk and
  * can be called where pcpu_setup_first_chunk() is expected.
  *
  * If this function is used to setup the first chunk, it is allocated
- * as a contiguous area using bootmem allocator and used as-is without
- * being mapped into vmalloc area.  This enables the first chunk to
- * piggy back on the linear physical mapping which often uses larger
- * page size.
+ * by calling @alloc_fn and used as-is without being mapped into
+ * vmalloc area.  Allocations are always whole multiples of @atom_size
+ * aligned to @atom_size.
+ *
+ * This enables the first chunk to piggy back on the linear physical
+ * mapping which often uses larger page size.  Please note that this
+ * can result in very sparse cpu->unit mapping on NUMA machines thus
+ * requiring large vmalloc address space.  Don't use this allocator if
+ * vmalloc space is not orders of magnitude larger than distances
+ * between node memory addresses (ie. 32bit NUMA machines).
  *
  * When @dyn_size is positive, dynamic area might be larger than
- * specified to fill page alignment.  Also, when @dyn_size is auto,
- * @dyn_size does not fill the whole first chunk but only what's
- * necessary for page alignment after static and reserved areas.
+ * specified to fill page alignment.  When @dyn_size is auto,
+ * @dyn_size is just big enough to fill page alignment after static
+ * and reserved areas.
  *
  * If the needed size is smaller than the minimum or specified unit
- * size, the leftover is returned to the bootmem allocator.
+ * size, the leftover is returned using @free_fn.
  *
  * RETURNS:
- * The determined pcpu_unit_size which can be used to initialize
- * percpu access on success, -errno on failure.
+ * 0 on success, -errno on failure.
  */
-ssize_t __init pcpu_embed_first_chunk(size_t static_size, size_t reserved_size,
-				      ssize_t dyn_size, ssize_t unit_size)
+int __init pcpu_embed_first_chunk(size_t reserved_size, ssize_t dyn_size,
+				  size_t atom_size,
+				  pcpu_fc_cpu_distance_fn_t cpu_distance_fn,
+				  pcpu_fc_alloc_fn_t alloc_fn,
+				  pcpu_fc_free_fn_t free_fn)
 {
-	size_t chunk_size;
-	unsigned int cpu;
+	void *base = (void *)ULONG_MAX;
+	void **areas = NULL;
+	struct pcpu_alloc_info *ai;
+	size_t size_sum, areas_size;
+	int group, i, rc;
 
-	/* determine parameters and allocate */
-	pcpue_size = PFN_ALIGN(static_size + reserved_size +
-			       (dyn_size >= 0 ? dyn_size : 0));
-	if (dyn_size != 0)
-		dyn_size = pcpue_size - static_size - reserved_size;
+	ai = pcpu_build_alloc_info(reserved_size, dyn_size, atom_size,
+				   cpu_distance_fn);
+	if (IS_ERR(ai))
+		return PTR_ERR(ai);
 
-	if (unit_size >= 0) {
-		BUG_ON(unit_size < pcpue_size);
-		pcpue_unit_size = unit_size;
-	} else
-		pcpue_unit_size = max_t(size_t, pcpue_size, PCPU_MIN_UNIT_SIZE);
+	size_sum = ai->static_size + ai->reserved_size + ai->dyn_size;
+	areas_size = PFN_ALIGN(ai->nr_groups * sizeof(void *));
 
-	chunk_size = pcpue_unit_size * nr_cpu_ids;
-
-	pcpue_ptr = __alloc_bootmem_nopanic(chunk_size, PAGE_SIZE,
-					    __pa(MAX_DMA_ADDRESS));
-	if (!pcpue_ptr) {
-		pr_warning("PERCPU: failed to allocate %zu bytes for "
-			   "embedding\n", chunk_size);
-		return -ENOMEM;
+	areas = alloc_bootmem_nopanic(areas_size);
+	if (!areas) {
+		rc = -ENOMEM;
+		goto out_free;
 	}
 
-	/* return the leftover and copy */
-	for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
-		void *ptr = pcpue_ptr + cpu * pcpue_unit_size;
+	/* allocate, copy and determine base address */
+	for (group = 0; group < ai->nr_groups; group++) {
+		struct pcpu_group_info *gi = &ai->groups[group];
+		unsigned int cpu = NR_CPUS;
+		void *ptr;
 
-		if (cpu_possible(cpu)) {
-			free_bootmem(__pa(ptr + pcpue_size),
-				     pcpue_unit_size - pcpue_size);
-			memcpy(ptr, __per_cpu_load, static_size);
-		} else
-			free_bootmem(__pa(ptr), pcpue_unit_size);
+		for (i = 0; i < gi->nr_units && cpu == NR_CPUS; i++)
+			cpu = gi->cpu_map[i];
+		BUG_ON(cpu == NR_CPUS);
+
+		/* allocate space for the whole group */
+		ptr = alloc_fn(cpu, gi->nr_units * ai->unit_size, atom_size);
+		if (!ptr) {
+			rc = -ENOMEM;
+			goto out_free_areas;
+		}
+		areas[group] = ptr;
+
+		base = min(ptr, base);
+
+		for (i = 0; i < gi->nr_units; i++, ptr += ai->unit_size) {
+			if (gi->cpu_map[i] == NR_CPUS) {
+				/* unused unit, free whole */
+				free_fn(ptr, ai->unit_size);
+				continue;
+			}
+			/* copy and return the unused part */
+			memcpy(ptr, __per_cpu_load, ai->static_size);
+			free_fn(ptr + size_sum, ai->unit_size - size_sum);
+		}
+	}
+
+	/* base address is now known, determine group base offsets */
+	for (group = 0; group < ai->nr_groups; group++)
+		ai->groups[group].base_offset = areas[group] - base;
+
+	pr_info("PERCPU: Embedded %zu pages/cpu @%p s%zu r%zu d%zu u%zu\n",
+		PFN_DOWN(size_sum), base, ai->static_size, ai->reserved_size,
+		ai->dyn_size, ai->unit_size);
+
+	rc = pcpu_setup_first_chunk(ai, base);
+	goto out_free;
+
+out_free_areas:
+	for (group = 0; group < ai->nr_groups; group++)
+		free_fn(areas[group],
+			ai->groups[group].nr_units * ai->unit_size);
+out_free:
+	pcpu_free_alloc_info(ai);
+	if (areas)
+		free_bootmem(__pa(areas), areas_size);
+	return rc;
+}
+#endif /* CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK ||
+	  !CONFIG_HAVE_SETUP_PER_CPU_AREA */
+
+#ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
+/**
+ * pcpu_page_first_chunk - map the first chunk using PAGE_SIZE pages
+ * @reserved_size: the size of reserved percpu area in bytes
+ * @alloc_fn: function to allocate percpu page, always called with PAGE_SIZE
+ * @free_fn: funtion to free percpu page, always called with PAGE_SIZE
+ * @populate_pte_fn: function to populate pte
+ *
+ * This is a helper to ease setting up page-remapped first percpu
+ * chunk and can be called where pcpu_setup_first_chunk() is expected.
+ *
+ * This is the basic allocator.  Static percpu area is allocated
+ * page-by-page into vmalloc area.
+ *
+ * RETURNS:
+ * 0 on success, -errno on failure.
+ */
+int __init pcpu_page_first_chunk(size_t reserved_size,
+				 pcpu_fc_alloc_fn_t alloc_fn,
+				 pcpu_fc_free_fn_t free_fn,
+				 pcpu_fc_populate_pte_fn_t populate_pte_fn)
+{
+	static struct vm_struct vm;
+	struct pcpu_alloc_info *ai;
+	char psize_str[16];
+	int unit_pages;
+	size_t pages_size;
+	struct page **pages;
+	int unit, i, j, rc;
+
+	snprintf(psize_str, sizeof(psize_str), "%luK", PAGE_SIZE >> 10);
+
+	ai = pcpu_build_alloc_info(reserved_size, -1, PAGE_SIZE, NULL);
+	if (IS_ERR(ai))
+		return PTR_ERR(ai);
+	BUG_ON(ai->nr_groups != 1);
+	BUG_ON(ai->groups[0].nr_units != num_possible_cpus());
+
+	unit_pages = ai->unit_size >> PAGE_SHIFT;
+
+	/* unaligned allocations can't be freed, round up to page size */
+	pages_size = PFN_ALIGN(unit_pages * num_possible_cpus() *
+			       sizeof(pages[0]));
+	pages = alloc_bootmem(pages_size);
+
+	/* allocate pages */
+	j = 0;
+	for (unit = 0; unit < num_possible_cpus(); unit++)
+		for (i = 0; i < unit_pages; i++) {
+			unsigned int cpu = ai->groups[0].cpu_map[unit];
+			void *ptr;
+
+			ptr = alloc_fn(cpu, PAGE_SIZE, PAGE_SIZE);
+			if (!ptr) {
+				pr_warning("PERCPU: failed to allocate %s page "
+					   "for cpu%u\n", psize_str, cpu);
+				goto enomem;
+			}
+			pages[j++] = virt_to_page(ptr);
+		}
+
+	/* allocate vm area, map the pages and copy static data */
+	vm.flags = VM_ALLOC;
+	vm.size = num_possible_cpus() * ai->unit_size;
+	vm_area_register_early(&vm, PAGE_SIZE);
+
+	for (unit = 0; unit < num_possible_cpus(); unit++) {
+		unsigned long unit_addr =
+			(unsigned long)vm.addr + unit * ai->unit_size;
+
+		for (i = 0; i < unit_pages; i++)
+			populate_pte_fn(unit_addr + (i << PAGE_SHIFT));
+
+		/* pte already populated, the following shouldn't fail */
+		rc = __pcpu_map_pages(unit_addr, &pages[unit * unit_pages],
+				      unit_pages);
+		if (rc < 0)
+			panic("failed to map percpu area, err=%d\n", rc);
+
+		/*
+		 * FIXME: Archs with virtual cache should flush local
+		 * cache for the linear mapping here - something
+		 * equivalent to flush_cache_vmap() on the local cpu.
+		 * flush_cache_vmap() can't be used as most supporting
+		 * data structures are not set up yet.
+		 */
+
+		/* copy static data */
+		memcpy((void *)unit_addr, __per_cpu_load, ai->static_size);
 	}
 
 	/* we're ready, commit */
-	pr_info("PERCPU: Embedded %zu pages at %p, static data %zu bytes\n",
-		pcpue_size >> PAGE_SHIFT, pcpue_ptr, static_size);
+	pr_info("PERCPU: %d %s pages/cpu @%p s%zu r%zu d%zu\n",
+		unit_pages, psize_str, vm.addr, ai->static_size,
+		ai->reserved_size, ai->dyn_size);
 
-	return pcpu_setup_first_chunk(pcpue_get_page, static_size,
-				      reserved_size, dyn_size,
-				      pcpue_unit_size, pcpue_ptr, NULL);
+	rc = pcpu_setup_first_chunk(ai, vm.addr);
+	goto out_free_ar;
+
+enomem:
+	while (--j >= 0)
+		free_fn(page_address(pages[j]), PAGE_SIZE);
+	rc = -ENOMEM;
+out_free_ar:
+	free_bootmem(__pa(pages), pages_size);
+	pcpu_free_alloc_info(ai);
+	return rc;
 }
+#endif /* CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK */
+
+/*
+ * Generic percpu area setup.
+ *
+ * The embedding helper is used because its behavior closely resembles
+ * the original non-dynamic generic percpu area setup.  This is
+ * important because many archs have addressing restrictions and might
+ * fail if the percpu area is located far away from the previous
+ * location.  As an added bonus, in non-NUMA cases, embedding is
+ * generally a good idea TLB-wise because percpu area can piggy back
+ * on the physical linear memory mapping which uses large page
+ * mappings on applicable archs.
+ */
+#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
+unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
+EXPORT_SYMBOL(__per_cpu_offset);
+
+static void * __init pcpu_dfl_fc_alloc(unsigned int cpu, size_t size,
+				       size_t align)
+{
+	return __alloc_bootmem_nopanic(size, align, __pa(MAX_DMA_ADDRESS));
+}
+
+static void __init pcpu_dfl_fc_free(void *ptr, size_t size)
+{
+	free_bootmem(__pa(ptr), size);
+}
+
+void __init setup_per_cpu_areas(void)
+{
+	unsigned long delta;
+	unsigned int cpu;
+	int rc;
+
+	/*
+	 * Always reserve area for module percpu variables.  That's
+	 * what the legacy allocator did.
+	 */
+	rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
+				    PERCPU_DYNAMIC_RESERVE, PAGE_SIZE, NULL,
+				    pcpu_dfl_fc_alloc, pcpu_dfl_fc_free);
+	if (rc < 0)
+		panic("Failed to initialized percpu areas.");
+
+	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
+	for_each_possible_cpu(cpu)
+		__per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
+}
+#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */