i386: prepare shared mm/discontig.c
authorThomas Gleixner <tglx@linutronix.de>
Thu, 11 Oct 2007 09:13:46 +0000 (11:13 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 11 Oct 2007 09:13:46 +0000 (11:13 +0200)
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/i386/mm/Makefile
arch/i386/mm/discontig.c [deleted file]
arch/i386/mm/discontig_32.c [new file with mode: 0644]

index f4ff3aca1a905f549a8e58442accb394a2f1f2ee..643b57b82f3b5a764fed593ff8d1888e4e7d1994 100644 (file)
@@ -4,7 +4,7 @@
 
 obj-y  := init.o pgtable.o fault.o ioremap_32.o extable.o pageattr.o mmap.o
 
-obj-$(CONFIG_NUMA) += discontig.o
+obj-$(CONFIG_NUMA) += discontig_32.o
 obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o
 obj-$(CONFIG_HIGHMEM) += highmem.o
 obj-$(CONFIG_BOOT_IOREMAP) += boot_ioremap_32.o
diff --git a/arch/i386/mm/discontig.c b/arch/i386/mm/discontig.c
deleted file mode 100644 (file)
index 860e912..0000000
+++ /dev/null
@@ -1,431 +0,0 @@
-/*
- * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
- * August 2002: added remote node KVA remap - Martin J. Bligh 
- *
- * Copyright (C) 2002, IBM Corp.
- *
- * All rights reserved.          
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
- * NON INFRINGEMENT.  See the GNU General Public License for more
- * details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
- */
-
-#include <linux/mm.h>
-#include <linux/bootmem.h>
-#include <linux/mmzone.h>
-#include <linux/highmem.h>
-#include <linux/initrd.h>
-#include <linux/nodemask.h>
-#include <linux/module.h>
-#include <linux/kexec.h>
-#include <linux/pfn.h>
-#include <linux/swap.h>
-
-#include <asm/e820.h>
-#include <asm/setup.h>
-#include <asm/mmzone.h>
-#include <bios_ebda.h>
-
-struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
-EXPORT_SYMBOL(node_data);
-bootmem_data_t node0_bdata;
-
-/*
- * numa interface - we expect the numa architecture specific code to have
- *                  populated the following initialisation.
- *
- * 1) node_online_map  - the map of all nodes configured (online) in the system
- * 2) node_start_pfn   - the starting page frame number for a node
- * 3) node_end_pfn     - the ending page fram number for a node
- */
-unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
-unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
-
-
-#ifdef CONFIG_DISCONTIGMEM
-/*
- * 4) physnode_map     - the mapping between a pfn and owning node
- * physnode_map keeps track of the physical memory layout of a generic
- * numa node on a 256Mb break (each element of the array will
- * represent 256Mb of memory and will be marked by the node id.  so,
- * if the first gig is on node 0, and the second gig is on node 1
- * physnode_map will contain:
- *
- *     physnode_map[0-3] = 0;
- *     physnode_map[4-7] = 1;
- *     physnode_map[8- ] = -1;
- */
-s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
-EXPORT_SYMBOL(physnode_map);
-
-void memory_present(int nid, unsigned long start, unsigned long end)
-{
-       unsigned long pfn;
-
-       printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
-                       nid, start, end);
-       printk(KERN_DEBUG "  Setting physnode_map array to node %d for pfns:\n", nid);
-       printk(KERN_DEBUG "  ");
-       for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
-               physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
-               printk("%ld ", pfn);
-       }
-       printk("\n");
-}
-
-unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
-                                             unsigned long end_pfn)
-{
-       unsigned long nr_pages = end_pfn - start_pfn;
-
-       if (!nr_pages)
-               return 0;
-
-       return (nr_pages + 1) * sizeof(struct page);
-}
-#endif
-
-extern unsigned long find_max_low_pfn(void);
-extern void add_one_highpage_init(struct page *, int, int);
-extern unsigned long highend_pfn, highstart_pfn;
-
-#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
-
-unsigned long node_remap_start_pfn[MAX_NUMNODES];
-unsigned long node_remap_size[MAX_NUMNODES];
-unsigned long node_remap_offset[MAX_NUMNODES];
-void *node_remap_start_vaddr[MAX_NUMNODES];
-void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
-
-void *node_remap_end_vaddr[MAX_NUMNODES];
-void *node_remap_alloc_vaddr[MAX_NUMNODES];
-static unsigned long kva_start_pfn;
-static unsigned long kva_pages;
-/*
- * FLAT - support for basic PC memory model with discontig enabled, essentially
- *        a single node with all available processors in it with a flat
- *        memory map.
- */
-int __init get_memcfg_numa_flat(void)
-{
-       printk("NUMA - single node, flat memory mode\n");
-
-       /* Run the memory configuration and find the top of memory. */
-       find_max_pfn();
-       node_start_pfn[0] = 0;
-       node_end_pfn[0] = max_pfn;
-       memory_present(0, 0, max_pfn);
-
-        /* Indicate there is one node available. */
-       nodes_clear(node_online_map);
-       node_set_online(0);
-       return 1;
-}
-
-/*
- * Find the highest page frame number we have available for the node
- */
-static void __init find_max_pfn_node(int nid)
-{
-       if (node_end_pfn[nid] > max_pfn)
-               node_end_pfn[nid] = max_pfn;
-       /*
-        * if a user has given mem=XXXX, then we need to make sure 
-        * that the node _starts_ before that, too, not just ends
-        */
-       if (node_start_pfn[nid] > max_pfn)
-               node_start_pfn[nid] = max_pfn;
-       BUG_ON(node_start_pfn[nid] > node_end_pfn[nid]);
-}
-
-/* 
- * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
- * method.  For node zero take this from the bottom of memory, for
- * subsequent nodes place them at node_remap_start_vaddr which contains
- * node local data in physically node local memory.  See setup_memory()
- * for details.
- */
-static void __init allocate_pgdat(int nid)
-{
-       if (nid && node_has_online_mem(nid))
-               NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
-       else {
-               NODE_DATA(nid) = (pg_data_t *)(pfn_to_kaddr(min_low_pfn));
-               min_low_pfn += PFN_UP(sizeof(pg_data_t));
-       }
-}
-
-void *alloc_remap(int nid, unsigned long size)
-{
-       void *allocation = node_remap_alloc_vaddr[nid];
-
-       size = ALIGN(size, L1_CACHE_BYTES);
-
-       if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
-               return 0;
-
-       node_remap_alloc_vaddr[nid] += size;
-       memset(allocation, 0, size);
-
-       return allocation;
-}
-
-void __init remap_numa_kva(void)
-{
-       void *vaddr;
-       unsigned long pfn;
-       int node;
-
-       for_each_online_node(node) {
-               for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
-                       vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
-                       set_pmd_pfn((ulong) vaddr, 
-                               node_remap_start_pfn[node] + pfn, 
-                               PAGE_KERNEL_LARGE);
-               }
-       }
-}
-
-static unsigned long calculate_numa_remap_pages(void)
-{
-       int nid;
-       unsigned long size, reserve_pages = 0;
-       unsigned long pfn;
-
-       for_each_online_node(nid) {
-               unsigned old_end_pfn = node_end_pfn[nid];
-
-               /*
-                * The acpi/srat node info can show hot-add memroy zones
-                * where memory could be added but not currently present.
-                */
-               if (node_start_pfn[nid] > max_pfn)
-                       continue;
-               if (node_end_pfn[nid] > max_pfn)
-                       node_end_pfn[nid] = max_pfn;
-
-               /* ensure the remap includes space for the pgdat. */
-               size = node_remap_size[nid] + sizeof(pg_data_t);
-
-               /* convert size to large (pmd size) pages, rounding up */
-               size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
-               /* now the roundup is correct, convert to PAGE_SIZE pages */
-               size = size * PTRS_PER_PTE;
-
-               /*
-                * Validate the region we are allocating only contains valid
-                * pages.
-                */
-               for (pfn = node_end_pfn[nid] - size;
-                    pfn < node_end_pfn[nid]; pfn++)
-                       if (!page_is_ram(pfn))
-                               break;
-
-               if (pfn != node_end_pfn[nid])
-                       size = 0;
-
-               printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
-                               size, nid);
-               node_remap_size[nid] = size;
-               node_remap_offset[nid] = reserve_pages;
-               reserve_pages += size;
-               printk("Shrinking node %d from %ld pages to %ld pages\n",
-                       nid, node_end_pfn[nid], node_end_pfn[nid] - size);
-
-               if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
-                       /*
-                        * Align node_end_pfn[] and node_remap_start_pfn[] to
-                        * pmd boundary. remap_numa_kva will barf otherwise.
-                        */
-                       printk("Shrinking node %d further by %ld pages for proper alignment\n",
-                               nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
-                       size +=  node_end_pfn[nid] & (PTRS_PER_PTE-1);
-               }
-
-               node_end_pfn[nid] -= size;
-               node_remap_start_pfn[nid] = node_end_pfn[nid];
-               shrink_active_range(nid, old_end_pfn, node_end_pfn[nid]);
-       }
-       printk("Reserving total of %ld pages for numa KVA remap\n",
-                       reserve_pages);
-       return reserve_pages;
-}
-
-extern void setup_bootmem_allocator(void);
-unsigned long __init setup_memory(void)
-{
-       int nid;
-       unsigned long system_start_pfn, system_max_low_pfn;
-
-       /*
-        * When mapping a NUMA machine we allocate the node_mem_map arrays
-        * from node local memory.  They are then mapped directly into KVA
-        * between zone normal and vmalloc space.  Calculate the size of
-        * this space and use it to adjust the boundry between ZONE_NORMAL
-        * and ZONE_HIGHMEM.
-        */
-       find_max_pfn();
-       get_memcfg_numa();
-
-       kva_pages = calculate_numa_remap_pages();
-
-       /* partially used pages are not usable - thus round upwards */
-       system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
-
-       kva_start_pfn = find_max_low_pfn() - kva_pages;
-
-#ifdef CONFIG_BLK_DEV_INITRD
-       /* Numa kva area is below the initrd */
-       if (LOADER_TYPE && INITRD_START)
-               kva_start_pfn = PFN_DOWN(INITRD_START)  - kva_pages;
-#endif
-       kva_start_pfn -= kva_start_pfn & (PTRS_PER_PTE-1);
-
-       system_max_low_pfn = max_low_pfn = find_max_low_pfn();
-       printk("kva_start_pfn ~ %ld find_max_low_pfn() ~ %ld\n",
-               kva_start_pfn, max_low_pfn);
-       printk("max_pfn = %ld\n", max_pfn);
-#ifdef CONFIG_HIGHMEM
-       highstart_pfn = highend_pfn = max_pfn;
-       if (max_pfn > system_max_low_pfn)
-               highstart_pfn = system_max_low_pfn;
-       printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
-              pages_to_mb(highend_pfn - highstart_pfn));
-       num_physpages = highend_pfn;
-       high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
-#else
-       num_physpages = system_max_low_pfn;
-       high_memory = (void *) __va(system_max_low_pfn * PAGE_SIZE - 1) + 1;
-#endif
-       printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
-                       pages_to_mb(system_max_low_pfn));
-       printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n", 
-                       min_low_pfn, max_low_pfn, highstart_pfn);
-
-       printk("Low memory ends at vaddr %08lx\n",
-                       (ulong) pfn_to_kaddr(max_low_pfn));
-       for_each_online_node(nid) {
-               node_remap_start_vaddr[nid] = pfn_to_kaddr(
-                               kva_start_pfn + node_remap_offset[nid]);
-               /* Init the node remap allocator */
-               node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
-                       (node_remap_size[nid] * PAGE_SIZE);
-               node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
-                       ALIGN(sizeof(pg_data_t), PAGE_SIZE);
-
-               allocate_pgdat(nid);
-               printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
-                       (ulong) node_remap_start_vaddr[nid],
-                       (ulong) pfn_to_kaddr(highstart_pfn
-                          + node_remap_offset[nid] + node_remap_size[nid]));
-       }
-       printk("High memory starts at vaddr %08lx\n",
-                       (ulong) pfn_to_kaddr(highstart_pfn));
-       for_each_online_node(nid)
-               find_max_pfn_node(nid);
-
-       memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
-       NODE_DATA(0)->bdata = &node0_bdata;
-       setup_bootmem_allocator();
-       return max_low_pfn;
-}
-
-void __init numa_kva_reserve(void)
-{
-       reserve_bootmem(PFN_PHYS(kva_start_pfn),PFN_PHYS(kva_pages));
-}
-
-void __init zone_sizes_init(void)
-{
-       int nid;
-       unsigned long max_zone_pfns[MAX_NR_ZONES];
-       memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
-       max_zone_pfns[ZONE_DMA] =
-               virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
-       max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
-#ifdef CONFIG_HIGHMEM
-       max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
-#endif
-
-       /* If SRAT has not registered memory, register it now */
-       if (find_max_pfn_with_active_regions() == 0) {
-               for_each_online_node(nid) {
-                       if (node_has_online_mem(nid))
-                               add_active_range(nid, node_start_pfn[nid],
-                                                       node_end_pfn[nid]);
-               }
-       }
-
-       free_area_init_nodes(max_zone_pfns);
-       return;
-}
-
-void __init set_highmem_pages_init(int bad_ppro) 
-{
-#ifdef CONFIG_HIGHMEM
-       struct zone *zone;
-       struct page *page;
-
-       for_each_zone(zone) {
-               unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
-
-               if (!is_highmem(zone))
-                       continue;
-
-               zone_start_pfn = zone->zone_start_pfn;
-               zone_end_pfn = zone_start_pfn + zone->spanned_pages;
-
-               printk("Initializing %s for node %d (%08lx:%08lx)\n",
-                               zone->name, zone_to_nid(zone),
-                               zone_start_pfn, zone_end_pfn);
-
-               for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
-                       if (!pfn_valid(node_pfn))
-                               continue;
-                       page = pfn_to_page(node_pfn);
-                       add_one_highpage_init(page, node_pfn, bad_ppro);
-               }
-       }
-       totalram_pages += totalhigh_pages;
-#endif
-}
-
-#ifdef CONFIG_MEMORY_HOTPLUG
-int paddr_to_nid(u64 addr)
-{
-       int nid;
-       unsigned long pfn = PFN_DOWN(addr);
-
-       for_each_node(nid)
-               if (node_start_pfn[nid] <= pfn &&
-                   pfn < node_end_pfn[nid])
-                       return nid;
-
-       return -1;
-}
-
-/*
- * This function is used to ask node id BEFORE memmap and mem_section's
- * initialization (pfn_to_nid() can't be used yet).
- * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
- */
-int memory_add_physaddr_to_nid(u64 addr)
-{
-       int nid = paddr_to_nid(addr);
-       return (nid >= 0) ? nid : 0;
-}
-
-EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
-#endif
diff --git a/arch/i386/mm/discontig_32.c b/arch/i386/mm/discontig_32.c
new file mode 100644 (file)
index 0000000..860e912
--- /dev/null
@@ -0,0 +1,431 @@
+/*
+ * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
+ * August 2002: added remote node KVA remap - Martin J. Bligh 
+ *
+ * Copyright (C) 2002, IBM Corp.
+ *
+ * All rights reserved.          
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/mm.h>
+#include <linux/bootmem.h>
+#include <linux/mmzone.h>
+#include <linux/highmem.h>
+#include <linux/initrd.h>
+#include <linux/nodemask.h>
+#include <linux/module.h>
+#include <linux/kexec.h>
+#include <linux/pfn.h>
+#include <linux/swap.h>
+
+#include <asm/e820.h>
+#include <asm/setup.h>
+#include <asm/mmzone.h>
+#include <bios_ebda.h>
+
+struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
+EXPORT_SYMBOL(node_data);
+bootmem_data_t node0_bdata;
+
+/*
+ * numa interface - we expect the numa architecture specific code to have
+ *                  populated the following initialisation.
+ *
+ * 1) node_online_map  - the map of all nodes configured (online) in the system
+ * 2) node_start_pfn   - the starting page frame number for a node
+ * 3) node_end_pfn     - the ending page fram number for a node
+ */
+unsigned long node_start_pfn[MAX_NUMNODES] __read_mostly;
+unsigned long node_end_pfn[MAX_NUMNODES] __read_mostly;
+
+
+#ifdef CONFIG_DISCONTIGMEM
+/*
+ * 4) physnode_map     - the mapping between a pfn and owning node
+ * physnode_map keeps track of the physical memory layout of a generic
+ * numa node on a 256Mb break (each element of the array will
+ * represent 256Mb of memory and will be marked by the node id.  so,
+ * if the first gig is on node 0, and the second gig is on node 1
+ * physnode_map will contain:
+ *
+ *     physnode_map[0-3] = 0;
+ *     physnode_map[4-7] = 1;
+ *     physnode_map[8- ] = -1;
+ */
+s8 physnode_map[MAX_ELEMENTS] __read_mostly = { [0 ... (MAX_ELEMENTS - 1)] = -1};
+EXPORT_SYMBOL(physnode_map);
+
+void memory_present(int nid, unsigned long start, unsigned long end)
+{
+       unsigned long pfn;
+
+       printk(KERN_INFO "Node: %d, start_pfn: %ld, end_pfn: %ld\n",
+                       nid, start, end);
+       printk(KERN_DEBUG "  Setting physnode_map array to node %d for pfns:\n", nid);
+       printk(KERN_DEBUG "  ");
+       for (pfn = start; pfn < end; pfn += PAGES_PER_ELEMENT) {
+               physnode_map[pfn / PAGES_PER_ELEMENT] = nid;
+               printk("%ld ", pfn);
+       }
+       printk("\n");
+}
+
+unsigned long node_memmap_size_bytes(int nid, unsigned long start_pfn,
+                                             unsigned long end_pfn)
+{
+       unsigned long nr_pages = end_pfn - start_pfn;
+
+       if (!nr_pages)
+               return 0;
+
+       return (nr_pages + 1) * sizeof(struct page);
+}
+#endif
+
+extern unsigned long find_max_low_pfn(void);
+extern void add_one_highpage_init(struct page *, int, int);
+extern unsigned long highend_pfn, highstart_pfn;
+
+#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
+
+unsigned long node_remap_start_pfn[MAX_NUMNODES];
+unsigned long node_remap_size[MAX_NUMNODES];
+unsigned long node_remap_offset[MAX_NUMNODES];
+void *node_remap_start_vaddr[MAX_NUMNODES];
+void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
+
+void *node_remap_end_vaddr[MAX_NUMNODES];
+void *node_remap_alloc_vaddr[MAX_NUMNODES];
+static unsigned long kva_start_pfn;
+static unsigned long kva_pages;
+/*
+ * FLAT - support for basic PC memory model with discontig enabled, essentially
+ *        a single node with all available processors in it with a flat
+ *        memory map.
+ */
+int __init get_memcfg_numa_flat(void)
+{
+       printk("NUMA - single node, flat memory mode\n");
+
+       /* Run the memory configuration and find the top of memory. */
+       find_max_pfn();
+       node_start_pfn[0] = 0;
+       node_end_pfn[0] = max_pfn;
+       memory_present(0, 0, max_pfn);
+
+        /* Indicate there is one node available. */
+       nodes_clear(node_online_map);
+       node_set_online(0);
+       return 1;
+}
+
+/*
+ * Find the highest page frame number we have available for the node
+ */
+static void __init find_max_pfn_node(int nid)
+{
+       if (node_end_pfn[nid] > max_pfn)
+               node_end_pfn[nid] = max_pfn;
+       /*
+        * if a user has given mem=XXXX, then we need to make sure 
+        * that the node _starts_ before that, too, not just ends
+        */
+       if (node_start_pfn[nid] > max_pfn)
+               node_start_pfn[nid] = max_pfn;
+       BUG_ON(node_start_pfn[nid] > node_end_pfn[nid]);
+}
+
+/* 
+ * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
+ * method.  For node zero take this from the bottom of memory, for
+ * subsequent nodes place them at node_remap_start_vaddr which contains
+ * node local data in physically node local memory.  See setup_memory()
+ * for details.
+ */
+static void __init allocate_pgdat(int nid)
+{
+       if (nid && node_has_online_mem(nid))
+               NODE_DATA(nid) = (pg_data_t *)node_remap_start_vaddr[nid];
+       else {
+               NODE_DATA(nid) = (pg_data_t *)(pfn_to_kaddr(min_low_pfn));
+               min_low_pfn += PFN_UP(sizeof(pg_data_t));
+       }
+}
+
+void *alloc_remap(int nid, unsigned long size)
+{
+       void *allocation = node_remap_alloc_vaddr[nid];
+
+       size = ALIGN(size, L1_CACHE_BYTES);
+
+       if (!allocation || (allocation + size) >= node_remap_end_vaddr[nid])
+               return 0;
+
+       node_remap_alloc_vaddr[nid] += size;
+       memset(allocation, 0, size);
+
+       return allocation;
+}
+
+void __init remap_numa_kva(void)
+{
+       void *vaddr;
+       unsigned long pfn;
+       int node;
+
+       for_each_online_node(node) {
+               for (pfn=0; pfn < node_remap_size[node]; pfn += PTRS_PER_PTE) {
+                       vaddr = node_remap_start_vaddr[node]+(pfn<<PAGE_SHIFT);
+                       set_pmd_pfn((ulong) vaddr, 
+                               node_remap_start_pfn[node] + pfn, 
+                               PAGE_KERNEL_LARGE);
+               }
+       }
+}
+
+static unsigned long calculate_numa_remap_pages(void)
+{
+       int nid;
+       unsigned long size, reserve_pages = 0;
+       unsigned long pfn;
+
+       for_each_online_node(nid) {
+               unsigned old_end_pfn = node_end_pfn[nid];
+
+               /*
+                * The acpi/srat node info can show hot-add memroy zones
+                * where memory could be added but not currently present.
+                */
+               if (node_start_pfn[nid] > max_pfn)
+                       continue;
+               if (node_end_pfn[nid] > max_pfn)
+                       node_end_pfn[nid] = max_pfn;
+
+               /* ensure the remap includes space for the pgdat. */
+               size = node_remap_size[nid] + sizeof(pg_data_t);
+
+               /* convert size to large (pmd size) pages, rounding up */
+               size = (size + LARGE_PAGE_BYTES - 1) / LARGE_PAGE_BYTES;
+               /* now the roundup is correct, convert to PAGE_SIZE pages */
+               size = size * PTRS_PER_PTE;
+
+               /*
+                * Validate the region we are allocating only contains valid
+                * pages.
+                */
+               for (pfn = node_end_pfn[nid] - size;
+                    pfn < node_end_pfn[nid]; pfn++)
+                       if (!page_is_ram(pfn))
+                               break;
+
+               if (pfn != node_end_pfn[nid])
+                       size = 0;
+
+               printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
+                               size, nid);
+               node_remap_size[nid] = size;
+               node_remap_offset[nid] = reserve_pages;
+               reserve_pages += size;
+               printk("Shrinking node %d from %ld pages to %ld pages\n",
+                       nid, node_end_pfn[nid], node_end_pfn[nid] - size);
+
+               if (node_end_pfn[nid] & (PTRS_PER_PTE-1)) {
+                       /*
+                        * Align node_end_pfn[] and node_remap_start_pfn[] to
+                        * pmd boundary. remap_numa_kva will barf otherwise.
+                        */
+                       printk("Shrinking node %d further by %ld pages for proper alignment\n",
+                               nid, node_end_pfn[nid] & (PTRS_PER_PTE-1));
+                       size +=  node_end_pfn[nid] & (PTRS_PER_PTE-1);
+               }
+
+               node_end_pfn[nid] -= size;
+               node_remap_start_pfn[nid] = node_end_pfn[nid];
+               shrink_active_range(nid, old_end_pfn, node_end_pfn[nid]);
+       }
+       printk("Reserving total of %ld pages for numa KVA remap\n",
+                       reserve_pages);
+       return reserve_pages;
+}
+
+extern void setup_bootmem_allocator(void);
+unsigned long __init setup_memory(void)
+{
+       int nid;
+       unsigned long system_start_pfn, system_max_low_pfn;
+
+       /*
+        * When mapping a NUMA machine we allocate the node_mem_map arrays
+        * from node local memory.  They are then mapped directly into KVA
+        * between zone normal and vmalloc space.  Calculate the size of
+        * this space and use it to adjust the boundry between ZONE_NORMAL
+        * and ZONE_HIGHMEM.
+        */
+       find_max_pfn();
+       get_memcfg_numa();
+
+       kva_pages = calculate_numa_remap_pages();
+
+       /* partially used pages are not usable - thus round upwards */
+       system_start_pfn = min_low_pfn = PFN_UP(init_pg_tables_end);
+
+       kva_start_pfn = find_max_low_pfn() - kva_pages;
+
+#ifdef CONFIG_BLK_DEV_INITRD
+       /* Numa kva area is below the initrd */
+       if (LOADER_TYPE && INITRD_START)
+               kva_start_pfn = PFN_DOWN(INITRD_START)  - kva_pages;
+#endif
+       kva_start_pfn -= kva_start_pfn & (PTRS_PER_PTE-1);
+
+       system_max_low_pfn = max_low_pfn = find_max_low_pfn();
+       printk("kva_start_pfn ~ %ld find_max_low_pfn() ~ %ld\n",
+               kva_start_pfn, max_low_pfn);
+       printk("max_pfn = %ld\n", max_pfn);
+#ifdef CONFIG_HIGHMEM
+       highstart_pfn = highend_pfn = max_pfn;
+       if (max_pfn > system_max_low_pfn)
+               highstart_pfn = system_max_low_pfn;
+       printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
+              pages_to_mb(highend_pfn - highstart_pfn));
+       num_physpages = highend_pfn;
+       high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
+#else
+       num_physpages = system_max_low_pfn;
+       high_memory = (void *) __va(system_max_low_pfn * PAGE_SIZE - 1) + 1;
+#endif
+       printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
+                       pages_to_mb(system_max_low_pfn));
+       printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n", 
+                       min_low_pfn, max_low_pfn, highstart_pfn);
+
+       printk("Low memory ends at vaddr %08lx\n",
+                       (ulong) pfn_to_kaddr(max_low_pfn));
+       for_each_online_node(nid) {
+               node_remap_start_vaddr[nid] = pfn_to_kaddr(
+                               kva_start_pfn + node_remap_offset[nid]);
+               /* Init the node remap allocator */
+               node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
+                       (node_remap_size[nid] * PAGE_SIZE);
+               node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
+                       ALIGN(sizeof(pg_data_t), PAGE_SIZE);
+
+               allocate_pgdat(nid);
+               printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
+                       (ulong) node_remap_start_vaddr[nid],
+                       (ulong) pfn_to_kaddr(highstart_pfn
+                          + node_remap_offset[nid] + node_remap_size[nid]));
+       }
+       printk("High memory starts at vaddr %08lx\n",
+                       (ulong) pfn_to_kaddr(highstart_pfn));
+       for_each_online_node(nid)
+               find_max_pfn_node(nid);
+
+       memset(NODE_DATA(0), 0, sizeof(struct pglist_data));
+       NODE_DATA(0)->bdata = &node0_bdata;
+       setup_bootmem_allocator();
+       return max_low_pfn;
+}
+
+void __init numa_kva_reserve(void)
+{
+       reserve_bootmem(PFN_PHYS(kva_start_pfn),PFN_PHYS(kva_pages));
+}
+
+void __init zone_sizes_init(void)
+{
+       int nid;
+       unsigned long max_zone_pfns[MAX_NR_ZONES];
+       memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
+       max_zone_pfns[ZONE_DMA] =
+               virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
+       max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
+#ifdef CONFIG_HIGHMEM
+       max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
+#endif
+
+       /* If SRAT has not registered memory, register it now */
+       if (find_max_pfn_with_active_regions() == 0) {
+               for_each_online_node(nid) {
+                       if (node_has_online_mem(nid))
+                               add_active_range(nid, node_start_pfn[nid],
+                                                       node_end_pfn[nid]);
+               }
+       }
+
+       free_area_init_nodes(max_zone_pfns);
+       return;
+}
+
+void __init set_highmem_pages_init(int bad_ppro) 
+{
+#ifdef CONFIG_HIGHMEM
+       struct zone *zone;
+       struct page *page;
+
+       for_each_zone(zone) {
+               unsigned long node_pfn, zone_start_pfn, zone_end_pfn;
+
+               if (!is_highmem(zone))
+                       continue;
+
+               zone_start_pfn = zone->zone_start_pfn;
+               zone_end_pfn = zone_start_pfn + zone->spanned_pages;
+
+               printk("Initializing %s for node %d (%08lx:%08lx)\n",
+                               zone->name, zone_to_nid(zone),
+                               zone_start_pfn, zone_end_pfn);
+
+               for (node_pfn = zone_start_pfn; node_pfn < zone_end_pfn; node_pfn++) {
+                       if (!pfn_valid(node_pfn))
+                               continue;
+                       page = pfn_to_page(node_pfn);
+                       add_one_highpage_init(page, node_pfn, bad_ppro);
+               }
+       }
+       totalram_pages += totalhigh_pages;
+#endif
+}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+int paddr_to_nid(u64 addr)
+{
+       int nid;
+       unsigned long pfn = PFN_DOWN(addr);
+
+       for_each_node(nid)
+               if (node_start_pfn[nid] <= pfn &&
+                   pfn < node_end_pfn[nid])
+                       return nid;
+
+       return -1;
+}
+
+/*
+ * This function is used to ask node id BEFORE memmap and mem_section's
+ * initialization (pfn_to_nid() can't be used yet).
+ * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
+ */
+int memory_add_physaddr_to_nid(u64 addr)
+{
+       int nid = paddr_to_nid(addr);
+       return (nid >= 0) ? nid : 0;
+}
+
+EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
+#endif