powerpc/fsl-booke: Add support for tlbilx instructions
[safe/jmp/linux-2.6] / arch / powerpc / mm / numa.c
index da09ba0..9047145 100644 (file)
 #include <linux/nodemask.h>
 #include <linux/cpu.h>
 #include <linux/notifier.h>
-#include <asm/lmb.h>
-#include <asm/machdep.h>
-#include <asm/abs_addr.h>
+#include <linux/lmb.h>
+#include <linux/of.h>
+#include <linux/pfn.h>
+#include <asm/sparsemem.h>
+#include <asm/prom.h>
 #include <asm/system.h>
 #include <asm/smp.h>
 
 static int numa_enabled = 1;
 
+static char *cmdline __initdata;
+
 static int numa_debug;
 #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
 
-#ifdef DEBUG_NUMA
-#define ARRAY_INITIALISER -1
-#else
-#define ARRAY_INITIALISER 0
-#endif
-
-int numa_cpu_lookup_table[NR_CPUS] = { [ 0 ... (NR_CPUS - 1)] =
-       ARRAY_INITIALISER};
-char *numa_memory_lookup_table;
+int numa_cpu_lookup_table[NR_CPUS];
 cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
-int nr_cpus_in_node[MAX_NUMNODES] = { [0 ... (MAX_NUMNODES -1)] = 0};
-
 struct pglist_data *node_data[MAX_NUMNODES];
-bootmem_data_t __initdata plat_node_bdata[MAX_NUMNODES];
+
+EXPORT_SYMBOL(numa_cpu_lookup_table);
+EXPORT_SYMBOL(numa_cpumask_lookup_table);
+EXPORT_SYMBOL(node_data);
+
 static int min_common_depth;
+static int n_mem_addr_cells, n_mem_size_cells;
+
+static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
+                                               unsigned int *nid)
+{
+       unsigned long long mem;
+       char *p = cmdline;
+       static unsigned int fake_nid;
+       static unsigned long long curr_boundary;
+
+       /*
+        * Modify node id, iff we started creating NUMA nodes
+        * We want to continue from where we left of the last time
+        */
+       if (fake_nid)
+               *nid = fake_nid;
+       /*
+        * In case there are no more arguments to parse, the
+        * node_id should be the same as the last fake node id
+        * (we've handled this above).
+        */
+       if (!p)
+               return 0;
+
+       mem = memparse(p, &p);
+       if (!mem)
+               return 0;
+
+       if (mem < curr_boundary)
+               return 0;
+
+       curr_boundary = mem;
+
+       if ((end_pfn << PAGE_SHIFT) > mem) {
+               /*
+                * Skip commas and spaces
+                */
+               while (*p == ',' || *p == ' ' || *p == '\t')
+                       p++;
+
+               cmdline = p;
+               fake_nid++;
+               *nid = fake_nid;
+               dbg("created new fake_node with id %d\n", fake_nid);
+               return 1;
+       }
+       return 0;
+}
 
 /*
- * We need somewhere to store start/span for each node until we have
- * allocated the real node_data structures.
+ * get_active_region_work_fn - A helper function for get_node_active_region
+ *     Returns datax set to the start_pfn and end_pfn if they contain
+ *     the initial value of datax->start_pfn between them
+ * @start_pfn: start page(inclusive) of region to check
+ * @end_pfn: end page(exclusive) of region to check
+ * @datax: comes in with ->start_pfn set to value to search for and
+ *     goes out with active range if it contains it
+ * Returns 1 if search value is in range else 0
  */
-static struct {
-       unsigned long node_start_pfn;
-       unsigned long node_end_pfn;
-       unsigned long node_present_pages;
-} init_node_data[MAX_NUMNODES] __initdata;
+static int __init get_active_region_work_fn(unsigned long start_pfn,
+                                       unsigned long end_pfn, void *datax)
+{
+       struct node_active_region *data;
+       data = (struct node_active_region *)datax;
 
-EXPORT_SYMBOL(node_data);
-EXPORT_SYMBOL(numa_cpu_lookup_table);
-EXPORT_SYMBOL(numa_memory_lookup_table);
-EXPORT_SYMBOL(numa_cpumask_lookup_table);
-EXPORT_SYMBOL(nr_cpus_in_node);
+       if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) {
+               data->start_pfn = start_pfn;
+               data->end_pfn = end_pfn;
+               return 1;
+       }
+       return 0;
+
+}
+
+/*
+ * get_node_active_region - Return active region containing start_pfn
+ * Active range returned is empty if none found.
+ * @start_pfn: The page to return the region for.
+ * @node_ar: Returned set to the active region containing start_pfn
+ */
+static void __init get_node_active_region(unsigned long start_pfn,
+                      struct node_active_region *node_ar)
+{
+       int nid = early_pfn_to_nid(start_pfn);
 
-static inline void map_cpu_to_node(int cpu, int node)
+       node_ar->nid = nid;
+       node_ar->start_pfn = start_pfn;
+       node_ar->end_pfn = start_pfn;
+       work_with_active_regions(nid, get_active_region_work_fn, node_ar);
+}
+
+static void __cpuinit map_cpu_to_node(int cpu, int node)
 {
        numa_cpu_lookup_table[cpu] = node;
-       if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node]))) {
+
+       dbg("adding cpu %d to node %d\n", cpu, node);
+
+       if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
                cpu_set(cpu, numa_cpumask_lookup_table[node]);
-               nr_cpus_in_node[node]++;
-       }
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
@@ -78,7 +151,6 @@ static void unmap_cpu_from_node(unsigned long cpu)
 
        if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
                cpu_clear(cpu, numa_cpumask_lookup_table[node]);
-               nr_cpus_in_node[node]--;
        } else {
                printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
                       cpu, node);
@@ -86,60 +158,73 @@ static void unmap_cpu_from_node(unsigned long cpu)
 }
 #endif /* CONFIG_HOTPLUG_CPU */
 
-static struct device_node * __devinit find_cpu_node(unsigned int cpu)
+/* must hold reference to node during call */
+static const int *of_get_associativity(struct device_node *dev)
 {
-       unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
-       struct device_node *cpu_node = NULL;
-       unsigned int *interrupt_server, *reg;
-       int len;
-
-       while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
-               /* Try interrupt server first */
-               interrupt_server = (unsigned int *)get_property(cpu_node,
-                                       "ibm,ppc-interrupt-server#s", &len);
-
-               len = len / sizeof(u32);
-
-               if (interrupt_server && (len > 0)) {
-                       while (len--) {
-                               if (interrupt_server[len] == hw_cpuid)
-                                       return cpu_node;
-                       }
-               } else {
-                       reg = (unsigned int *)get_property(cpu_node,
-                                                          "reg", &len);
-                       if (reg && (len > 0) && (reg[0] == hw_cpuid))
-                               return cpu_node;
-               }
-       }
-
-       return NULL;
+       return of_get_property(dev, "ibm,associativity", NULL);
 }
 
-/* must hold reference to node during call */
-static int *of_get_associativity(struct device_node *dev)
+/*
+ * Returns the property linux,drconf-usable-memory if
+ * it exists (the property exists only in kexec/kdump kernels,
+ * added by kexec-tools)
+ */
+static const u32 *of_get_usable_memory(struct device_node *memory)
 {
-       return (unsigned int *)get_property(dev, "ibm,associativity", NULL);
+       const u32 *prop;
+       u32 len;
+       prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
+       if (!prop || len < sizeof(unsigned int))
+               return 0;
+       return prop;
 }
 
-static int of_node_numa_domain(struct device_node *device)
+/* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
+ * info is found.
+ */
+static int of_node_to_nid_single(struct device_node *device)
 {
-       int numa_domain;
-       unsigned int *tmp;
+       int nid = -1;
+       const unsigned int *tmp;
 
        if (min_common_depth == -1)
-               return 0;
+               goto out;
 
        tmp = of_get_associativity(device);
-       if (tmp && (tmp[0] >= min_common_depth)) {
-               numa_domain = tmp[min_common_depth];
-       } else {
-               dbg("WARNING: no NUMA information for %s\n",
-                   device->full_name);
-               numa_domain = 0;
+       if (!tmp)
+               goto out;
+
+       if (tmp[0] >= min_common_depth)
+               nid = tmp[min_common_depth];
+
+       /* POWER4 LPAR uses 0xffff as invalid node */
+       if (nid == 0xffff || nid >= MAX_NUMNODES)
+               nid = -1;
+out:
+       return nid;
+}
+
+/* Walk the device tree upwards, looking for an associativity id */
+int of_node_to_nid(struct device_node *device)
+{
+       struct device_node *tmp;
+       int nid = -1;
+
+       of_node_get(device);
+       while (device) {
+               nid = of_node_to_nid_single(device);
+               if (nid != -1)
+                       break;
+
+               tmp = device;
+               device = of_get_parent(tmp);
+               of_node_put(tmp);
        }
-       return numa_domain;
+       of_node_put(device);
+
+       return nid;
 }
+EXPORT_SYMBOL_GPL(of_node_to_nid);
 
 /*
  * In theory, the "ibm,associativity" property may contain multiple
@@ -148,7 +233,7 @@ static int of_node_numa_domain(struct device_node *device)
  * characteristics relative to its multiple connections.  We ignore
  * this for now.  We also assume that all cpu and memory sets have
  * their distances represented at a common level.  This won't be
- * true for heirarchical NUMA.
+ * true for hierarchical NUMA.
  *
  * In any case the ibm,associativity-reference-points should give
  * the correct depth for a normal NUMA system.
@@ -158,7 +243,7 @@ static int of_node_numa_domain(struct device_node *device)
 static int __init find_min_common_depth(void)
 {
        int depth;
-       unsigned int *ref_points;
+       const unsigned int *ref_points;
        struct device_node *rtas_root;
        unsigned int len;
 
@@ -173,14 +258,13 @@ static int __init find_min_common_depth(void)
         * configuration (should be all 0's) and the second is for a normal
         * NUMA configuration.
         */
-       ref_points = (unsigned int *)get_property(rtas_root,
+       ref_points = of_get_property(rtas_root,
                        "ibm,associativity-reference-points", &len);
 
-       if ((len >= 1) && ref_points) {
+       if ((len >= 2 * sizeof(unsigned int)) && ref_points) {
                depth = ref_points[1];
        } else {
-               dbg("WARNING: could not find NUMA "
-                   "associativity reference point\n");
+               dbg("NUMA: ibm,associativity-reference-points not found.\n");
                depth = -1;
        }
        of_node_put(rtas_root);
@@ -188,32 +272,20 @@ static int __init find_min_common_depth(void)
        return depth;
 }
 
-static int __init get_mem_addr_cells(void)
+static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
 {
        struct device_node *memory = NULL;
-       int rc;
 
        memory = of_find_node_by_type(memory, "memory");
        if (!memory)
-               return 0; /* it won't matter */
+               panic("numa.c: No memory nodes found!");
 
-       rc = prom_n_addr_cells(memory);
-       return rc;
+       *n_addr_cells = of_n_addr_cells(memory);
+       *n_size_cells = of_n_size_cells(memory);
+       of_node_put(memory);
 }
 
-static int __init get_mem_size_cells(void)
-{
-       struct device_node *memory = NULL;
-       int rc;
-
-       memory = of_find_node_by_type(memory, "memory");
-       if (!memory)
-               return 0; /* it won't matter */
-       rc = prom_n_size_cells(memory);
-       return rc;
-}
-
-static unsigned long read_n_cells(int n, unsigned int **buf)
+static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
 {
        unsigned long result = 0;
 
@@ -224,44 +296,171 @@ static unsigned long read_n_cells(int n, unsigned int **buf)
        return result;
 }
 
+struct of_drconf_cell {
+       u64     base_addr;
+       u32     drc_index;
+       u32     reserved;
+       u32     aa_index;
+       u32     flags;
+};
+
+#define DRCONF_MEM_ASSIGNED    0x00000008
+#define DRCONF_MEM_AI_INVALID  0x00000040
+#define DRCONF_MEM_RESERVED    0x00000080
+
+/*
+ * Read the next lmb list entry from the ibm,dynamic-memory property
+ * and return the information in the provided of_drconf_cell structure.
+ */
+static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
+{
+       const u32 *cp;
+
+       drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
+
+       cp = *cellp;
+       drmem->drc_index = cp[0];
+       drmem->reserved = cp[1];
+       drmem->aa_index = cp[2];
+       drmem->flags = cp[3];
+
+       *cellp = cp + 4;
+}
+
+/*
+ * Retreive and validate the ibm,dynamic-memory property of the device tree.
+ *
+ * The layout of the ibm,dynamic-memory property is a number N of lmb
+ * list entries followed by N lmb list entries.  Each lmb list entry
+ * contains information as layed out in the of_drconf_cell struct above.
+ */
+static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
+{
+       const u32 *prop;
+       u32 len, entries;
+
+       prop = of_get_property(memory, "ibm,dynamic-memory", &len);
+       if (!prop || len < sizeof(unsigned int))
+               return 0;
+
+       entries = *prop++;
+
+       /* Now that we know the number of entries, revalidate the size
+        * of the property read in to ensure we have everything
+        */
+       if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
+               return 0;
+
+       *dm = prop;
+       return entries;
+}
+
+/*
+ * Retreive and validate the ibm,lmb-size property for drconf memory
+ * from the device tree.
+ */
+static u64 of_get_lmb_size(struct device_node *memory)
+{
+       const u32 *prop;
+       u32 len;
+
+       prop = of_get_property(memory, "ibm,lmb-size", &len);
+       if (!prop || len < sizeof(unsigned int))
+               return 0;
+
+       return read_n_cells(n_mem_size_cells, &prop);
+}
+
+struct assoc_arrays {
+       u32     n_arrays;
+       u32     array_sz;
+       const u32 *arrays;
+};
+
+/*
+ * Retreive and validate the list of associativity arrays for drconf
+ * memory from the ibm,associativity-lookup-arrays property of the
+ * device tree..
+ *
+ * The layout of the ibm,associativity-lookup-arrays property is a number N
+ * indicating the number of associativity arrays, followed by a number M
+ * indicating the size of each associativity array, followed by a list
+ * of N associativity arrays.
+ */
+static int of_get_assoc_arrays(struct device_node *memory,
+                              struct assoc_arrays *aa)
+{
+       const u32 *prop;
+       u32 len;
+
+       prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
+       if (!prop || len < 2 * sizeof(unsigned int))
+               return -1;
+
+       aa->n_arrays = *prop++;
+       aa->array_sz = *prop++;
+
+       /* Now that we know the number of arrrays and size of each array,
+        * revalidate the size of the property read in.
+        */
+       if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
+               return -1;
+
+       aa->arrays = prop;
+       return 0;
+}
+
+/*
+ * This is like of_node_to_nid_single() for memory represented in the
+ * ibm,dynamic-reconfiguration-memory node.
+ */
+static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
+                                  struct assoc_arrays *aa)
+{
+       int default_nid = 0;
+       int nid = default_nid;
+       int index;
+
+       if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
+           !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
+           drmem->aa_index < aa->n_arrays) {
+               index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
+               nid = aa->arrays[index];
+
+               if (nid == 0xffff || nid >= MAX_NUMNODES)
+                       nid = default_nid;
+       }
+
+       return nid;
+}
+
 /*
  * Figure out to which domain a cpu belongs and stick it there.
  * Return the id of the domain used.
  */
-static int numa_setup_cpu(unsigned long lcpu)
+static int __cpuinit numa_setup_cpu(unsigned long lcpu)
 {
-       int numa_domain = 0;
-       struct device_node *cpu = find_cpu_node(lcpu);
+       int nid = 0;
+       struct device_node *cpu = of_get_cpu_node(lcpu, NULL);
 
        if (!cpu) {
                WARN_ON(1);
                goto out;
        }
 
-       numa_domain = of_node_numa_domain(cpu);
+       nid = of_node_to_nid_single(cpu);
 
-       if (numa_domain >= num_online_nodes()) {
-               /*
-                * POWER4 LPAR uses 0xffff as invalid node,
-                * dont warn in this case.
-                */
-               if (numa_domain != 0xffff)
-                       printk(KERN_ERR "WARNING: cpu %ld "
-                              "maps to invalid NUMA node %d\n",
-                              lcpu, numa_domain);
-               numa_domain = 0;
-       }
+       if (nid < 0 || !node_online(nid))
+               nid = any_online_node(NODE_MASK_ALL);
 out:
-       node_set_online(numa_domain);
-
-       map_cpu_to_node(lcpu, numa_domain);
+       map_cpu_to_node(lcpu, nid);
 
        of_node_put(cpu);
 
-       return numa_domain;
+       return nid;
 }
 
-static int cpu_numa_callback(struct notifier_block *nfb,
+static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
                             unsigned long action,
                             void *hcpu)
 {
@@ -270,15 +469,15 @@ static int cpu_numa_callback(struct notifier_block *nfb,
 
        switch (action) {
        case CPU_UP_PREPARE:
-               if (min_common_depth == -1 || !numa_enabled)
-                       map_cpu_to_node(lcpu, 0);
-               else
-                       numa_setup_cpu(lcpu);
+       case CPU_UP_PREPARE_FROZEN:
+               numa_setup_cpu(lcpu);
                ret = NOTIFY_OK;
                break;
 #ifdef CONFIG_HOTPLUG_CPU
        case CPU_DEAD:
+       case CPU_DEAD_FROZEN:
        case CPU_UP_CANCELED:
+       case CPU_UP_CANCELED_FROZEN:
                unmap_cpu_from_node(lcpu);
                break;
                ret = NOTIFY_OK;
@@ -295,17 +494,16 @@ static int cpu_numa_callback(struct notifier_block *nfb,
  * or zero. If the returned value of size is 0 the region should be
  * discarded as it lies wholy above the memory limit.
  */
-static unsigned long __init numa_enforce_memory_limit(unsigned long start, unsigned long size)
+static unsigned long __init numa_enforce_memory_limit(unsigned long start,
+                                                     unsigned long size)
 {
        /*
         * We use lmb_end_of_DRAM() in here instead of memory_limit because
         * we've already adjusted it for the limit and it takes care of
-        * having memory holes below the limit.
+        * having memory holes below the limit.  Also, in the case of
+        * iommu_is_off, memory_limit is not set but is implicitly enforced.
         */
 
-       if (! memory_limit)
-               return size;
-
        if (start + size <= lmb_end_of_DRAM())
                return size;
 
@@ -315,13 +513,94 @@ static unsigned long __init numa_enforce_memory_limit(unsigned long start, unsig
        return lmb_end_of_DRAM() - start;
 }
 
+/*
+ * Reads the counter for a given entry in
+ * linux,drconf-usable-memory property
+ */
+static inline int __init read_usm_ranges(const u32 **usm)
+{
+       /*
+        * For each lmb in ibm,dynamic-memory a corresponding
+        * entry in linux,drconf-usable-memory property contains
+        * a counter followed by that many (base, size) duple.
+        * read the counter from linux,drconf-usable-memory
+        */
+       return read_n_cells(n_mem_size_cells, usm);
+}
+
+/*
+ * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
+ * node.  This assumes n_mem_{addr,size}_cells have been set.
+ */
+static void __init parse_drconf_memory(struct device_node *memory)
+{
+       const u32 *dm, *usm;
+       unsigned int n, rc, ranges, is_kexec_kdump = 0;
+       unsigned long lmb_size, base, size, sz;
+       int nid;
+       struct assoc_arrays aa;
+
+       n = of_get_drconf_memory(memory, &dm);
+       if (!n)
+               return;
+
+       lmb_size = of_get_lmb_size(memory);
+       if (!lmb_size)
+               return;
+
+       rc = of_get_assoc_arrays(memory, &aa);
+       if (rc)
+               return;
+
+       /* check if this is a kexec/kdump kernel */
+       usm = of_get_usable_memory(memory);
+       if (usm != NULL)
+               is_kexec_kdump = 1;
+
+       for (; n != 0; --n) {
+               struct of_drconf_cell drmem;
+
+               read_drconf_cell(&drmem, &dm);
+
+               /* skip this block if the reserved bit is set in flags (0x80)
+                  or if the block is not assigned to this partition (0x8) */
+               if ((drmem.flags & DRCONF_MEM_RESERVED)
+                   || !(drmem.flags & DRCONF_MEM_ASSIGNED))
+                       continue;
+
+               base = drmem.base_addr;
+               size = lmb_size;
+               ranges = 1;
+
+               if (is_kexec_kdump) {
+                       ranges = read_usm_ranges(&usm);
+                       if (!ranges) /* there are no (base, size) duple */
+                               continue;
+               }
+               do {
+                       if (is_kexec_kdump) {
+                               base = read_n_cells(n_mem_addr_cells, &usm);
+                               size = read_n_cells(n_mem_size_cells, &usm);
+                       }
+                       nid = of_drconf_to_nid_single(&drmem, &aa);
+                       fake_numa_create_new_node(
+                               ((base + size) >> PAGE_SHIFT),
+                                          &nid);
+                       node_set_online(nid);
+                       sz = numa_enforce_memory_limit(base, size);
+                       if (sz)
+                               add_active_range(nid, base >> PAGE_SHIFT,
+                                                (base >> PAGE_SHIFT)
+                                                + (sz >> PAGE_SHIFT));
+               } while (--ranges);
+       }
+}
+
 static int __init parse_numa_properties(void)
 {
        struct device_node *cpu = NULL;
        struct device_node *memory = NULL;
-       int addr_cells, size_cells;
-       int max_domain = 0;
-       long entries = lmb_end_of_DRAM() >> MEMORY_INCREMENT_SHIFT;
+       int default_nid = 0;
        unsigned long i;
 
        if (numa_enabled == 0) {
@@ -329,125 +608,93 @@ static int __init parse_numa_properties(void)
                return -1;
        }
 
-       numa_memory_lookup_table =
-               (char *)abs_to_virt(lmb_alloc(entries * sizeof(char), 1));
-       memset(numa_memory_lookup_table, 0, entries * sizeof(char));
-
-       for (i = 0; i < entries ; i++)
-               numa_memory_lookup_table[i] = ARRAY_INITIALISER;
-
        min_common_depth = find_min_common_depth();
 
-       dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
        if (min_common_depth < 0)
                return min_common_depth;
 
-       max_domain = numa_setup_cpu(boot_cpuid);
+       dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
 
        /*
-        * Even though we connect cpus to numa domains later in SMP init,
-        * we need to know the maximum node id now. This is because each
-        * node id must have NODE_DATA etc backing it.
-        * As a result of hotplug we could still have cpus appear later on
-        * with larger node ids. In that case we force the cpu into node 0.
+        * Even though we connect cpus to numa domains later in SMP
+        * init, we need to know the node ids now. This is because
+        * each node to be onlined must have NODE_DATA etc backing it.
         */
-       for_each_cpu(i) {
-               int numa_domain;
+       for_each_present_cpu(i) {
+               int nid;
 
-               cpu = find_cpu_node(i);
+               cpu = of_get_cpu_node(i, NULL);
+               BUG_ON(!cpu);
+               nid = of_node_to_nid_single(cpu);
+               of_node_put(cpu);
 
-               if (cpu) {
-                       numa_domain = of_node_numa_domain(cpu);
-                       of_node_put(cpu);
-
-                       if (numa_domain < MAX_NUMNODES &&
-                           max_domain < numa_domain)
-                               max_domain = numa_domain;
-               }
+               /*
+                * Don't fall back to default_nid yet -- we will plug
+                * cpus into nodes once the memory scan has discovered
+                * the topology.
+                */
+               if (nid < 0)
+                       continue;
+               node_set_online(nid);
        }
 
-       addr_cells = get_mem_addr_cells();
-       size_cells = get_mem_size_cells();
+       get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
        memory = NULL;
        while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
                unsigned long start;
                unsigned long size;
-               int numa_domain;
+               int nid;
                int ranges;
-               unsigned int *memcell_buf;
+               const unsigned int *memcell_buf;
                unsigned int len;
 
-               memcell_buf = (unsigned int *)get_property(memory, "reg", &len);
+               memcell_buf = of_get_property(memory,
+                       "linux,usable-memory", &len);
+               if (!memcell_buf || len <= 0)
+                       memcell_buf = of_get_property(memory, "reg", &len);
                if (!memcell_buf || len <= 0)
                        continue;
 
-               ranges = memory->n_addrs;
+               /* ranges in cell */
+               ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
 new_range:
                /* these are order-sensitive, and modify the buffer pointer */
-               start = read_n_cells(addr_cells, &memcell_buf);
-               size = read_n_cells(size_cells, &memcell_buf);
-
-               start = _ALIGN_DOWN(start, MEMORY_INCREMENT);
-               size = _ALIGN_UP(size, MEMORY_INCREMENT);
-
-               numa_domain = of_node_numa_domain(memory);
+               start = read_n_cells(n_mem_addr_cells, &memcell_buf);
+               size = read_n_cells(n_mem_size_cells, &memcell_buf);
 
-               if (numa_domain >= MAX_NUMNODES) {
-                       if (numa_domain != 0xffff)
-                               printk(KERN_ERR "WARNING: memory at %lx maps "
-                                      "to invalid NUMA node %d\n", start,
-                                      numa_domain);
-                       numa_domain = 0;
-               }
+               /*
+                * Assumption: either all memory nodes or none will
+                * have associativity properties.  If none, then
+                * everything goes to default_nid.
+                */
+               nid = of_node_to_nid_single(memory);
+               if (nid < 0)
+                       nid = default_nid;
 
-               if (max_domain < numa_domain)
-                       max_domain = numa_domain;
+               fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
+               node_set_online(nid);
 
-               if (! (size = numa_enforce_memory_limit(start, size))) {
+               if (!(size = numa_enforce_memory_limit(start, size))) {
                        if (--ranges)
                                goto new_range;
                        else
                                continue;
                }
 
-               /*
-                * Initialize new node struct, or add to an existing one.
-                */
-               if (init_node_data[numa_domain].node_end_pfn) {
-                       if ((start / PAGE_SIZE) <
-                           init_node_data[numa_domain].node_start_pfn)
-                               init_node_data[numa_domain].node_start_pfn =
-                                       start / PAGE_SIZE;
-                       if (((start / PAGE_SIZE) + (size / PAGE_SIZE)) >
-                           init_node_data[numa_domain].node_end_pfn)
-                               init_node_data[numa_domain].node_end_pfn =
-                                       (start / PAGE_SIZE) +
-                                       (size / PAGE_SIZE);
-
-                       init_node_data[numa_domain].node_present_pages +=
-                               size / PAGE_SIZE;
-               } else {
-                       node_set_online(numa_domain);
-
-                       init_node_data[numa_domain].node_start_pfn =
-                               start / PAGE_SIZE;
-                       init_node_data[numa_domain].node_end_pfn =
-                               init_node_data[numa_domain].node_start_pfn +
-                               size / PAGE_SIZE;
-                       init_node_data[numa_domain].node_present_pages =
-                               size / PAGE_SIZE;
-               }
-
-               for (i = start ; i < (start+size); i += MEMORY_INCREMENT)
-                       numa_memory_lookup_table[i >> MEMORY_INCREMENT_SHIFT] =
-                               numa_domain;
+               add_active_range(nid, start >> PAGE_SHIFT,
+                               (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
 
                if (--ranges)
                        goto new_range;
        }
 
-       for (i = 0; i <= max_domain; i++)
-               node_set_online(i);
+       /*
+        * Now do the same thing for each LMB listed in the ibm,dynamic-memory
+        * property in the ibm,dynamic-reconfiguration-memory node.
+        */
+       memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
+       if (memory)
+               parse_drconf_memory(memory);
 
        return 0;
 }
@@ -456,35 +703,59 @@ static void __init setup_nonnuma(void)
 {
        unsigned long top_of_ram = lmb_end_of_DRAM();
        unsigned long total_ram = lmb_phys_mem_size();
-       unsigned long i;
+       unsigned long start_pfn, end_pfn;
+       unsigned int i, nid = 0;
 
-       printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
+       printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
               top_of_ram, total_ram);
-       printk(KERN_INFO "Memory hole size: %ldMB\n",
+       printk(KERN_DEBUG "Memory hole size: %ldMB\n",
               (top_of_ram - total_ram) >> 20);
 
-       if (!numa_memory_lookup_table) {
-               long entries = top_of_ram >> MEMORY_INCREMENT_SHIFT;
-               numa_memory_lookup_table =
-                       (char *)abs_to_virt(lmb_alloc(entries * sizeof(char), 1));
-               memset(numa_memory_lookup_table, 0, entries * sizeof(char));
-               for (i = 0; i < entries ; i++)
-                       numa_memory_lookup_table[i] = ARRAY_INITIALISER;
+       for (i = 0; i < lmb.memory.cnt; ++i) {
+               start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
+               end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
+
+               fake_numa_create_new_node(end_pfn, &nid);
+               add_active_range(nid, start_pfn, end_pfn);
+               node_set_online(nid);
        }
+}
 
-       map_cpu_to_node(boot_cpuid, 0);
+void __init dump_numa_cpu_topology(void)
+{
+       unsigned int node;
+       unsigned int cpu, count;
 
-       node_set_online(0);
+       if (min_common_depth == -1 || !numa_enabled)
+               return;
 
-       init_node_data[0].node_start_pfn = 0;
-       init_node_data[0].node_end_pfn = lmb_end_of_DRAM() / PAGE_SIZE;
-       init_node_data[0].node_present_pages = total_ram / PAGE_SIZE;
+       for_each_online_node(node) {
+               printk(KERN_DEBUG "Node %d CPUs:", node);
 
-       for (i = 0 ; i < top_of_ram; i += MEMORY_INCREMENT)
-               numa_memory_lookup_table[i >> MEMORY_INCREMENT_SHIFT] = 0;
+               count = 0;
+               /*
+                * If we used a CPU iterator here we would miss printing
+                * the holes in the cpumap.
+                */
+               for (cpu = 0; cpu < NR_CPUS; cpu++) {
+                       if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
+                               if (count == 0)
+                                       printk(" %u", cpu);
+                               ++count;
+                       } else {
+                               if (count > 1)
+                                       printk("-%u", cpu - 1);
+                               count = 0;
+                       }
+               }
+
+               if (count > 1)
+                       printk("-%u", NR_CPUS - 1);
+               printk("\n");
+       }
 }
 
-static void __init dump_numa_topology(void)
+static void __init dump_numa_memory_topology(void)
 {
        unsigned int node;
        unsigned int count;
@@ -495,12 +766,13 @@ static void __init dump_numa_topology(void)
        for_each_online_node(node) {
                unsigned long i;
 
-               printk(KERN_INFO "Node %d Memory:", node);
+               printk(KERN_DEBUG "Node %d Memory:", node);
 
                count = 0;
 
-               for (i = 0; i < lmb_end_of_DRAM(); i += MEMORY_INCREMENT) {
-                       if (numa_memory_lookup_table[i >> MEMORY_INCREMENT_SHIFT] == node) {
+               for (i = 0; i < lmb_end_of_DRAM();
+                    i += (1 << SECTION_SIZE_BITS)) {
+                       if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
                                if (count == 0)
                                        printk(" 0x%lx", i);
                                ++count;
@@ -515,7 +787,6 @@ static void __init dump_numa_topology(void)
                        printk("-0x%lx", i);
                printk("\n");
        }
-       return;
 }
 
 /*
@@ -523,51 +794,128 @@ static void __init dump_numa_topology(void)
  * required. nid is the preferred node and end is the physical address of
  * the highest address in the node.
  *
- * Returns the physical address of the memory.
+ * Returns the virtual address of the memory.
  */
-static unsigned long careful_allocation(int nid, unsigned long size,
-                                       unsigned long align, unsigned long end)
+static void __init *careful_zallocation(int nid, unsigned long size,
+                                      unsigned long align,
+                                      unsigned long end_pfn)
 {
-       unsigned long ret = lmb_alloc_base(size, align, end);
+       void *ret;
+       int new_nid;
+       unsigned long ret_paddr;
+
+       ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
 
        /* retry over all memory */
-       if (!ret)
-               ret = lmb_alloc_base(size, align, lmb_end_of_DRAM());
+       if (!ret_paddr)
+               ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
 
-       if (!ret)
-               panic("numa.c: cannot allocate %lu bytes on node %d",
+       if (!ret_paddr)
+               panic("numa.c: cannot allocate %lu bytes for node %d",
                      size, nid);
 
+       ret = __va(ret_paddr);
+
        /*
-        * If the memory came from a previously allocated node, we must
-        * retry with the bootmem allocator.
+        * We initialize the nodes in numeric order: 0, 1, 2...
+        * and hand over control from the LMB allocator to the
+        * bootmem allocator.  If this function is called for
+        * node 5, then we know that all nodes <5 are using the
+        * bootmem allocator instead of the LMB allocator.
+        *
+        * So, check the nid from which this allocation came
+        * and double check to see if we need to use bootmem
+        * instead of the LMB.  We don't free the LMB memory
+        * since it would be useless.
         */
-       if (pa_to_nid(ret) < nid) {
-               nid = pa_to_nid(ret);
-               ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(nid),
+       new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
+       if (new_nid < nid) {
+               ret = __alloc_bootmem_node(NODE_DATA(new_nid),
                                size, align, 0);
 
-               if (!ret)
-                       panic("numa.c: cannot allocate %lu bytes on node %d",
-                             size, nid);
-
-               ret = virt_to_abs(ret);
-
-               dbg("alloc_bootmem %lx %lx\n", ret, size);
+               dbg("alloc_bootmem %p %lx\n", ret, size);
        }
 
+       memset(ret, 0, size);
        return ret;
 }
 
+static struct notifier_block __cpuinitdata ppc64_numa_nb = {
+       .notifier_call = cpu_numa_callback,
+       .priority = 1 /* Must run before sched domains notifier. */
+};
+
+static void mark_reserved_regions_for_nid(int nid)
+{
+       struct pglist_data *node = NODE_DATA(nid);
+       int i;
+
+       for (i = 0; i < lmb.reserved.cnt; i++) {
+               unsigned long physbase = lmb.reserved.region[i].base;
+               unsigned long size = lmb.reserved.region[i].size;
+               unsigned long start_pfn = physbase >> PAGE_SHIFT;
+               unsigned long end_pfn = PFN_UP(physbase + size);
+               struct node_active_region node_ar;
+               unsigned long node_end_pfn = node->node_start_pfn +
+                                            node->node_spanned_pages;
+
+               /*
+                * Check to make sure that this lmb.reserved area is
+                * within the bounds of the node that we care about.
+                * Checking the nid of the start and end points is not
+                * sufficient because the reserved area could span the
+                * entire node.
+                */
+               if (end_pfn <= node->node_start_pfn ||
+                   start_pfn >= node_end_pfn)
+                       continue;
+
+               get_node_active_region(start_pfn, &node_ar);
+               while (start_pfn < end_pfn &&
+                       node_ar.start_pfn < node_ar.end_pfn) {
+                       unsigned long reserve_size = size;
+                       /*
+                        * if reserved region extends past active region
+                        * then trim size to active region
+                        */
+                       if (end_pfn > node_ar.end_pfn)
+                               reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
+                                       - physbase;
+                       /*
+                        * Only worry about *this* node, others may not
+                        * yet have valid NODE_DATA().
+                        */
+                       if (node_ar.nid == nid) {
+                               dbg("reserve_bootmem %lx %lx nid=%d\n",
+                                       physbase, reserve_size, node_ar.nid);
+                               reserve_bootmem_node(NODE_DATA(node_ar.nid),
+                                               physbase, reserve_size,
+                                               BOOTMEM_DEFAULT);
+                       }
+                       /*
+                        * if reserved region is contained in the active region
+                        * then done.
+                        */
+                       if (end_pfn <= node_ar.end_pfn)
+                               break;
+
+                       /*
+                        * reserved region extends past the active region
+                        *   get next active region that contains this
+                        *   reserved region
+                        */
+                       start_pfn = node_ar.end_pfn;
+                       physbase = start_pfn << PAGE_SHIFT;
+                       size = size - reserve_size;
+                       get_node_active_region(start_pfn, &node_ar);
+               }
+       }
+}
+
+
 void __init do_init_bootmem(void)
 {
        int nid;
-       int addr_cells, size_cells;
-       struct device_node *memory = NULL;
-       static struct notifier_block ppc64_numa_nb = {
-               .notifier_call = cpu_numa_callback,
-               .priority = 1 /* Must run before sched domains notifier. */
-       };
 
        min_low_pfn = 0;
        max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
@@ -576,205 +924,219 @@ void __init do_init_bootmem(void)
        if (parse_numa_properties())
                setup_nonnuma();
        else
-               dump_numa_topology();
+               dump_numa_memory_topology();
 
        register_cpu_notifier(&ppc64_numa_nb);
+       cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
+                         (void *)(unsigned long)boot_cpuid);
 
        for_each_online_node(nid) {
-               unsigned long start_paddr, end_paddr;
-               int i;
-               unsigned long bootmem_paddr;
+               unsigned long start_pfn, end_pfn;
+               void *bootmem_vaddr;
                unsigned long bootmap_pages;
 
-               start_paddr = init_node_data[nid].node_start_pfn * PAGE_SIZE;
-               end_paddr = init_node_data[nid].node_end_pfn * PAGE_SIZE;
+               get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
 
-               /* Allocate the node structure node local if possible */
-               NODE_DATA(nid) = (struct pglist_data *)careful_allocation(nid,
+               /*
+                * Allocate the node structure node local if possible
+                *
+                * Be careful moving this around, as it relies on all
+                * previous nodes' bootmem to be initialized and have
+                * all reserved areas marked.
+                */
+               NODE_DATA(nid) = careful_zallocation(nid,
                                        sizeof(struct pglist_data),
-                                       SMP_CACHE_BYTES, end_paddr);
-               NODE_DATA(nid) = abs_to_virt(NODE_DATA(nid));
-               memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
+                                       SMP_CACHE_BYTES, end_pfn);
 
                dbg("node %d\n", nid);
                dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
 
-               NODE_DATA(nid)->bdata = &plat_node_bdata[nid];
-               NODE_DATA(nid)->node_start_pfn =
-                       init_node_data[nid].node_start_pfn;
-               NODE_DATA(nid)->node_spanned_pages =
-                       end_paddr - start_paddr;
+               NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
+               NODE_DATA(nid)->node_start_pfn = start_pfn;
+               NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
 
                if (NODE_DATA(nid)->node_spanned_pages == 0)
                        continue;
 
-               dbg("start_paddr = %lx\n", start_paddr);
-               dbg("end_paddr = %lx\n", end_paddr);
+               dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
+               dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
 
-               bootmap_pages = bootmem_bootmap_pages((end_paddr - start_paddr) >> PAGE_SHIFT);
+               bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
+               bootmem_vaddr = careful_zallocation(nid,
+                                       bootmap_pages << PAGE_SHIFT,
+                                       PAGE_SIZE, end_pfn);
 
-               bootmem_paddr = careful_allocation(nid,
-                               bootmap_pages << PAGE_SHIFT,
-                               PAGE_SIZE, end_paddr);
-               memset(abs_to_virt(bootmem_paddr), 0,
-                      bootmap_pages << PAGE_SHIFT);
-               dbg("bootmap_paddr = %lx\n", bootmem_paddr);
+               dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
 
-               init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
-                                 start_paddr >> PAGE_SHIFT,
-                                 end_paddr >> PAGE_SHIFT);
+               init_bootmem_node(NODE_DATA(nid),
+                                 __pa(bootmem_vaddr) >> PAGE_SHIFT,
+                                 start_pfn, end_pfn);
 
+               free_bootmem_with_active_regions(nid, end_pfn);
                /*
-                * We need to do another scan of all memory sections to
-                * associate memory with the correct node.
+                * Be very careful about moving this around.  Future
+                * calls to careful_zallocation() depend on this getting
+                * done correctly.
                 */
-               addr_cells = get_mem_addr_cells();
-               size_cells = get_mem_size_cells();
-               memory = NULL;
-               while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
-                       unsigned long mem_start, mem_size;
-                       int numa_domain, ranges;
-                       unsigned int *memcell_buf;
-                       unsigned int len;
-
-                       memcell_buf = (unsigned int *)get_property(memory, "reg", &len);
-                       if (!memcell_buf || len <= 0)
-                               continue;
+               mark_reserved_regions_for_nid(nid);
+               sparse_memory_present_with_active_regions(nid);
+       }
+}
 
-                       ranges = memory->n_addrs;       /* ranges in cell */
-new_range:
-                       mem_start = read_n_cells(addr_cells, &memcell_buf);
-                       mem_size = read_n_cells(size_cells, &memcell_buf);
-                       if (numa_enabled) {
-                               numa_domain = of_node_numa_domain(memory);
-                               if (numa_domain  >= MAX_NUMNODES)
-                                       numa_domain = 0;
-                       } else
-                               numa_domain =  0;
-
-                       if (numa_domain != nid)
-                               continue;
+void __init paging_init(void)
+{
+       unsigned long max_zone_pfns[MAX_NR_ZONES];
+       memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
+       max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
+       free_area_init_nodes(max_zone_pfns);
+}
 
-                       mem_size = numa_enforce_memory_limit(mem_start, mem_size);
-                       if (mem_size) {
-                               dbg("free_bootmem %lx %lx\n", mem_start, mem_size);
-                               free_bootmem_node(NODE_DATA(nid), mem_start, mem_size);
-                       }
+static int __init early_numa(char *p)
+{
+       if (!p)
+               return 0;
 
-                       if (--ranges)           /* process all ranges in cell */
-                               goto new_range;
-               }
+       if (strstr(p, "off"))
+               numa_enabled = 0;
 
-               /*
-                * Mark reserved regions on this node
-                */
-               for (i = 0; i < lmb.reserved.cnt; i++) {
-                       unsigned long physbase = lmb.reserved.region[i].base;
-                       unsigned long size = lmb.reserved.region[i].size;
+       if (strstr(p, "debug"))
+               numa_debug = 1;
 
-                       if (pa_to_nid(physbase) != nid &&
-                           pa_to_nid(physbase+size-1) != nid)
-                               continue;
+       p = strstr(p, "fake=");
+       if (p)
+               cmdline = p + strlen("fake=");
 
-                       if (physbase < end_paddr &&
-                           (physbase+size) > start_paddr) {
-                               /* overlaps */
-                               if (physbase < start_paddr) {
-                                       size -= start_paddr - physbase;
-                                       physbase = start_paddr;
-                               }
-
-                               if (size > end_paddr - physbase)
-                                       size = end_paddr - physbase;
-
-                               dbg("reserve_bootmem %lx %lx\n", physbase,
-                                   size);
-                               reserve_bootmem_node(NODE_DATA(nid), physbase,
-                                                    size);
-                       }
-               }
-               /*
-                * This loop may look famaliar, but we have to do it again
-                * after marking our reserved memory to mark memory present
-                * for sparsemem.
-                */
-               addr_cells = get_mem_addr_cells();
-               size_cells = get_mem_size_cells();
-               memory = NULL;
-               while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
-                       unsigned long mem_start, mem_size;
-                       int numa_domain, ranges;
-                       unsigned int *memcell_buf;
-                       unsigned int len;
-
-                       memcell_buf = (unsigned int *)get_property(memory, "reg", &len);
-                       if (!memcell_buf || len <= 0)
-                               continue;
+       return 0;
+}
+early_param("numa", early_numa);
 
-                       ranges = memory->n_addrs;       /* ranges in cell */
-new_range2:
-                       mem_start = read_n_cells(addr_cells, &memcell_buf);
-                       mem_size = read_n_cells(size_cells, &memcell_buf);
-                       if (numa_enabled) {
-                               numa_domain = of_node_numa_domain(memory);
-                               if (numa_domain  >= MAX_NUMNODES)
-                                       numa_domain = 0;
-                       } else
-                               numa_domain =  0;
-
-                       if (numa_domain != nid)
-                               continue;
+#ifdef CONFIG_MEMORY_HOTPLUG
+/*
+ * Find the node associated with a hot added memory section for
+ * memory represented in the device tree by the property
+ * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
+ */
+static int hot_add_drconf_scn_to_nid(struct device_node *memory,
+                                    unsigned long scn_addr)
+{
+       const u32 *dm;
+       unsigned int drconf_cell_cnt, rc;
+       unsigned long lmb_size;
+       struct assoc_arrays aa;
+       int nid = -1;
+
+       drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
+       if (!drconf_cell_cnt)
+               return -1;
 
-                       mem_size = numa_enforce_memory_limit(mem_start, mem_size);
-                       memory_present(numa_domain, mem_start >> PAGE_SHIFT,
-                                      (mem_start + mem_size) >> PAGE_SHIFT);
+       lmb_size = of_get_lmb_size(memory);
+       if (!lmb_size)
+               return -1;
 
-                       if (--ranges)           /* process all ranges in cell */
-                               goto new_range2;
-               }
+       rc = of_get_assoc_arrays(memory, &aa);
+       if (rc)
+               return -1;
+
+       for (; drconf_cell_cnt != 0; --drconf_cell_cnt) {
+               struct of_drconf_cell drmem;
+
+               read_drconf_cell(&drmem, &dm);
+
+               /* skip this block if it is reserved or not assigned to
+                * this partition */
+               if ((drmem.flags & DRCONF_MEM_RESERVED)
+                   || !(drmem.flags & DRCONF_MEM_ASSIGNED))
+                       continue;
+
+               if ((scn_addr < drmem.base_addr)
+                   || (scn_addr >= (drmem.base_addr + lmb_size)))
+                       continue;
 
+               nid = of_drconf_to_nid_single(&drmem, &aa);
+               break;
        }
+
+       return nid;
 }
 
-void __init paging_init(void)
+/*
+ * Find the node associated with a hot added memory section for memory
+ * represented in the device tree as a node (i.e. memory@XXXX) for
+ * each lmb.
+ */
+int hot_add_node_scn_to_nid(unsigned long scn_addr)
 {
-       unsigned long zones_size[MAX_NR_ZONES];
-       unsigned long zholes_size[MAX_NR_ZONES];
-       int nid;
+       struct device_node *memory = NULL;
+       int nid = -1;
 
-       memset(zones_size, 0, sizeof(zones_size));
-       memset(zholes_size, 0, sizeof(zholes_size));
+       while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
+               unsigned long start, size;
+               int ranges;
+               const unsigned int *memcell_buf;
+               unsigned int len;
 
-       for_each_online_node(nid) {
-               unsigned long start_pfn;
-               unsigned long end_pfn;
+               memcell_buf = of_get_property(memory, "reg", &len);
+               if (!memcell_buf || len <= 0)
+                       continue;
 
-               start_pfn = init_node_data[nid].node_start_pfn;
-               end_pfn = init_node_data[nid].node_end_pfn;
+               /* ranges in cell */
+               ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
 
-               zones_size[ZONE_DMA] = end_pfn - start_pfn;
-               zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] -
-                       init_node_data[nid].node_present_pages;
+               while (ranges--) {
+                       start = read_n_cells(n_mem_addr_cells, &memcell_buf);
+                       size = read_n_cells(n_mem_size_cells, &memcell_buf);
 
-               dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid,
-                   zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]);
+                       if ((scn_addr < start) || (scn_addr >= (start + size)))
+                               continue;
 
-               free_area_init_node(nid, NODE_DATA(nid), zones_size,
-                                                       start_pfn, zholes_size);
+                       nid = of_node_to_nid_single(memory);
+                       break;
+               }
+
+               of_node_put(memory);
+               if (nid >= 0)
+                       break;
        }
+
+       return nid;
 }
 
-static int __init early_numa(char *p)
+/*
+ * Find the node associated with a hot added memory section.  Section
+ * corresponds to a SPARSEMEM section, not an LMB.  It is assumed that
+ * sections are fully contained within a single LMB.
+ */
+int hot_add_scn_to_nid(unsigned long scn_addr)
 {
-       if (!p)
-               return 0;
+       struct device_node *memory = NULL;
+       int nid, found = 0;
 
-       if (strstr(p, "off"))
-               numa_enabled = 0;
+       if (!numa_enabled || (min_common_depth < 0))
+               return any_online_node(NODE_MASK_ALL);
 
-       if (strstr(p, "debug"))
-               numa_debug = 1;
+       memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
+       if (memory) {
+               nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
+               of_node_put(memory);
+       } else {
+               nid = hot_add_node_scn_to_nid(scn_addr);
+       }
 
-       return 0;
+       if (nid < 0 || !node_online(nid))
+               nid = any_online_node(NODE_MASK_ALL);
+
+       if (NODE_DATA(nid)->node_spanned_pages)
+               return nid;
+
+       for_each_online_node(nid) {
+               if (NODE_DATA(nid)->node_spanned_pages) {
+                       found = 1;
+                       break;
+               }
+       }
+
+       BUG_ON(!found);
+       return nid;
 }
-early_param("numa", early_numa);
+
+#endif /* CONFIG_MEMORY_HOTPLUG */