#include <linux/types.h>
 #include <linux/percpu.h>
+#include <linux/config.h>
+#include <linux/mmzone.h>
+#include <asm/atomic.h>
 
 /*
  * Global page accounting.  One instance per CPU.  Only unsigned longs are
 
 DECLARE_PER_CPU(struct page_state, page_states);
 
-#endif /* _LINUX_VMSTAT_H */
+/*
+ * Zone based page accounting with per cpu differentials.
+ */
+extern atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
+
+static inline void zone_page_state_add(long x, struct zone *zone,
+                                enum zone_stat_item item)
+{
+       atomic_long_add(x, &zone->vm_stat[item]);
+       atomic_long_add(x, &vm_stat[item]);
+}
+
+static inline unsigned long global_page_state(enum zone_stat_item item)
+{
+       long x = atomic_long_read(&vm_stat[item]);
+#ifdef CONFIG_SMP
+       if (x < 0)
+               x = 0;
+#endif
+       return x;
+}
+
+static inline unsigned long zone_page_state(struct zone *zone,
+                                       enum zone_stat_item item)
+{
+       long x = atomic_long_read(&zone->vm_stat[item]);
+#ifdef CONFIG_SMP
+       if (x < 0)
+               x = 0;
+#endif
+       return x;
+}
+
+#ifdef CONFIG_NUMA
+/*
+ * Determine the per node value of a stat item. This function
+ * is called frequently in a NUMA machine, so try to be as
+ * frugal as possible.
+ */
+static inline unsigned long node_page_state(int node,
+                                enum zone_stat_item item)
+{
+       struct zone *zones = NODE_DATA(node)->node_zones;
+
+       return
+#ifndef CONFIG_DMA_IS_NORMAL
+#if !defined(CONFIG_DMA_IS_DMA32) && BITS_PER_LONG >= 64
+               zone_page_state(&zones[ZONE_DMA32], item) +
+#endif
+               zone_page_state(&zones[ZONE_NORMAL], item) +
+#endif
+#ifdef CONFIG_HIGHMEM
+               zone_page_state(&zones[ZONE_HIGHMEM], item) +
+#endif
+               zone_page_state(&zones[ZONE_DMA], item);
+}
+#else
+#define node_page_state(node, item) global_page_state(item)
+#endif
+
+#define __add_zone_page_state(__z, __i, __d)   \
+               __mod_zone_page_state(__z, __i, __d)
+#define __sub_zone_page_state(__z, __i, __d)   \
+               __mod_zone_page_state(__z, __i,-(__d))
+
+#define add_zone_page_state(__z, __i, __d) mod_zone_page_state(__z, __i, __d)
+#define sub_zone_page_state(__z, __i, __d) mod_zone_page_state(__z, __i, -(__d))
+
+static inline void zap_zone_vm_stats(struct zone *zone)
+{
+       memset(zone->vm_stat, 0, sizeof(zone->vm_stat));
+}
+
+#ifdef CONFIG_SMP
+void __mod_zone_page_state(struct zone *, enum zone_stat_item item, int);
+void __inc_zone_page_state(struct page *, enum zone_stat_item);
+void __dec_zone_page_state(struct page *, enum zone_stat_item);
 
+void mod_zone_page_state(struct zone *, enum zone_stat_item, int);
+void inc_zone_page_state(struct page *, enum zone_stat_item);
+void dec_zone_page_state(struct page *, enum zone_stat_item);
+
+extern void inc_zone_state(struct zone *, enum zone_stat_item);
+
+void refresh_cpu_vm_stats(int);
+void refresh_vm_stats(void);
+
+#else /* CONFIG_SMP */
+
+/*
+ * We do not maintain differentials in a single processor configuration.
+ * The functions directly modify the zone and global counters.
+ */
+static inline void __mod_zone_page_state(struct zone *zone,
+                       enum zone_stat_item item, int delta)
+{
+       zone_page_state_add(delta, zone, item);
+}
+
+static inline void __inc_zone_page_state(struct page *page,
+                       enum zone_stat_item item)
+{
+       atomic_long_inc(&page_zone(page)->vm_stat[item]);
+       atomic_long_inc(&vm_stat[item]);
+}
+
+static inline void __dec_zone_page_state(struct page *page,
+                       enum zone_stat_item item)
+{
+       atomic_long_dec(&page_zone(page)->vm_stat[item]);
+       atomic_long_dec(&vm_stat[item]);
+}
+
+/*
+ * We only use atomic operations to update counters. So there is no need to
+ * disable interrupts.
+ */
+#define inc_zone_page_state __inc_zone_page_state
+#define dec_zone_page_state __dec_zone_page_state
+#define mod_zone_page_state __mod_zone_page_state
+
+static inline void refresh_cpu_vm_stats(int cpu) { }
+static inline void refresh_vm_stats(void) { }
+#endif
+
+#endif /* _LINUX_VMSTAT_H */
 
  *
  *  Manages VM statistics
  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  zoned VM statistics
+ *  Copyright (C) 2006 Silicon Graphics, Inc.,
+ *             Christoph Lameter <christoph@lameter.com>
  */
 
 #include <linux/config.h>
 #include <linux/mm.h>
+#include <linux/module.h>
 
 /*
  * Accumulate the page_state information across all CPUs.
        }
 }
 
+/*
+ * Manage combined zone based / global counters
+ *
+ * vm_stat contains the global counters
+ */
+atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
+EXPORT_SYMBOL(vm_stat);
+
+#ifdef CONFIG_SMP
+
+#define STAT_THRESHOLD 32
+
+/*
+ * Determine pointer to currently valid differential byte given a zone and
+ * the item number.
+ *
+ * Preemption must be off
+ */
+static inline s8 *diff_pointer(struct zone *zone, enum zone_stat_item item)
+{
+       return &zone_pcp(zone, smp_processor_id())->vm_stat_diff[item];
+}
+
+/*
+ * For use when we know that interrupts are disabled.
+ */
+void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
+                               int delta)
+{
+       s8 *p;
+       long x;
+
+       p = diff_pointer(zone, item);
+       x = delta + *p;
+
+       if (unlikely(x > STAT_THRESHOLD || x < -STAT_THRESHOLD)) {
+               zone_page_state_add(x, zone, item);
+               x = 0;
+       }
+
+       *p = x;
+}
+EXPORT_SYMBOL(__mod_zone_page_state);
+
+/*
+ * For an unknown interrupt state
+ */
+void mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
+                                       int delta)
+{
+       unsigned long flags;
+
+       local_irq_save(flags);
+       __mod_zone_page_state(zone, item, delta);
+       local_irq_restore(flags);
+}
+EXPORT_SYMBOL(mod_zone_page_state);
+
+/*
+ * Optimized increment and decrement functions.
+ *
+ * These are only for a single page and therefore can take a struct page *
+ * argument instead of struct zone *. This allows the inclusion of the code
+ * generated for page_zone(page) into the optimized functions.
+ *
+ * No overflow check is necessary and therefore the differential can be
+ * incremented or decremented in place which may allow the compilers to
+ * generate better code.
+ *
+ * The increment or decrement is known and therefore one boundary check can
+ * be omitted.
+ *
+ * Some processors have inc/dec instructions that are atomic vs an interrupt.
+ * However, the code must first determine the differential location in a zone
+ * based on the processor number and then inc/dec the counter. There is no
+ * guarantee without disabling preemption that the processor will not change
+ * in between and therefore the atomicity vs. interrupt cannot be exploited
+ * in a useful way here.
+ */
+void __inc_zone_page_state(struct page *page, enum zone_stat_item item)
+{
+       struct zone *zone = page_zone(page);
+       s8 *p = diff_pointer(zone, item);
+
+       (*p)++;
+
+       if (unlikely(*p > STAT_THRESHOLD)) {
+               zone_page_state_add(*p, zone, item);
+               *p = 0;
+       }
+}
+EXPORT_SYMBOL(__inc_zone_page_state);
+
+void __dec_zone_page_state(struct page *page, enum zone_stat_item item)
+{
+       struct zone *zone = page_zone(page);
+       s8 *p = diff_pointer(zone, item);
+
+       (*p)--;
+
+       if (unlikely(*p < -STAT_THRESHOLD)) {
+               zone_page_state_add(*p, zone, item);
+               *p = 0;
+       }
+}
+EXPORT_SYMBOL(__dec_zone_page_state);
+
+void inc_zone_page_state(struct page *page, enum zone_stat_item item)
+{
+       unsigned long flags;
+       struct zone *zone;
+       s8 *p;
+
+       zone = page_zone(page);
+       local_irq_save(flags);
+       p = diff_pointer(zone, item);
+
+       (*p)++;
+
+       if (unlikely(*p > STAT_THRESHOLD)) {
+               zone_page_state_add(*p, zone, item);
+               *p = 0;
+       }
+       local_irq_restore(flags);
+}
+EXPORT_SYMBOL(inc_zone_page_state);
+
+void dec_zone_page_state(struct page *page, enum zone_stat_item item)
+{
+       unsigned long flags;
+       struct zone *zone;
+       s8 *p;
+
+       zone = page_zone(page);
+       local_irq_save(flags);
+       p = diff_pointer(zone, item);
+
+       (*p)--;
+
+       if (unlikely(*p < -STAT_THRESHOLD)) {
+               zone_page_state_add(*p, zone, item);
+               *p = 0;
+       }
+       local_irq_restore(flags);
+}
+EXPORT_SYMBOL(dec_zone_page_state);
+
+/*
+ * Update the zone counters for one cpu.
+ */
+void refresh_cpu_vm_stats(int cpu)
+{
+       struct zone *zone;
+       int i;
+       unsigned long flags;
+
+       for_each_zone(zone) {
+               struct per_cpu_pageset *pcp;
+
+               pcp = zone_pcp(zone, cpu);
+
+               for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
+                       if (pcp->vm_stat_diff[i]) {
+                               local_irq_save(flags);
+                               zone_page_state_add(pcp->vm_stat_diff[i],
+                                       zone, i);
+                               pcp->vm_stat_diff[i] = 0;
+                               local_irq_restore(flags);
+                       }
+       }
+}
+
+static void __refresh_cpu_vm_stats(void *dummy)
+{
+       refresh_cpu_vm_stats(smp_processor_id());
+}
+
+/*
+ * Consolidate all counters.
+ *
+ * Note that the result is less inaccurate but still inaccurate
+ * if concurrent processes are allowed to run.
+ */
+void refresh_vm_stats(void)
+{
+       on_each_cpu(__refresh_cpu_vm_stats, NULL, 0, 1);
+}
+EXPORT_SYMBOL(refresh_vm_stats);
+
+#endif
+
 #ifdef CONFIG_PROC_FS
 
 #include <linux/seq_file.h>
 };
 
 static char *vmstat_text[] = {
+       /* Zoned VM counters */
+
+       /* Page state */
        "nr_dirty",
        "nr_writeback",
        "nr_unstable",
                           zone->nr_scan_active, zone->nr_scan_inactive,
                           zone->spanned_pages,
                           zone->present_pages);
+
+               for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
+                       seq_printf(m, "\n    %-12s %lu", vmstat_text[i],
+                                       zone_page_state(zone, i));
+
                seq_printf(m,
                           "\n        protection: (%lu",
                           zone->lowmem_reserve[0]);
 
 static void *vmstat_start(struct seq_file *m, loff_t *pos)
 {
+       unsigned long *v;
        struct page_state *ps;
+       int i;
 
        if (*pos >= ARRAY_SIZE(vmstat_text))
                return NULL;
 
-       ps = kmalloc(sizeof(*ps), GFP_KERNEL);
-       m->private = ps;
-       if (!ps)
+       v = kmalloc(NR_VM_ZONE_STAT_ITEMS * sizeof(unsigned long)
+                       + sizeof(*ps), GFP_KERNEL);
+       m->private = v;
+       if (!v)
                return ERR_PTR(-ENOMEM);
+       for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
+               v[i] = global_page_state(i);
+       ps = (struct page_state *)(v + NR_VM_ZONE_STAT_ITEMS);
        get_full_page_state(ps);
        ps->pgpgin /= 2;                /* sectors -> kbytes */
        ps->pgpgout /= 2;
-       return (unsigned long *)ps + *pos;
+       return v + *pos;
 }
 
 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)