4 * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 #include <linux/threads.h>
12 #include <linux/bootmem.h>
13 #include <linux/init.h>
15 #include <linux/mmzone.h>
16 #include <linux/module.h>
17 #include <linux/nodemask.h>
18 #include <linux/cpu.h>
19 #include <linux/notifier.h>
20 #include <linux/lmb.h>
22 #include <asm/sparsemem.h>
24 #include <asm/system.h>
27 static int numa_enabled = 1;
29 static char *cmdline __initdata;
31 static int numa_debug;
32 #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
34 int numa_cpu_lookup_table[NR_CPUS];
35 cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
36 struct pglist_data *node_data[MAX_NUMNODES];
38 EXPORT_SYMBOL(numa_cpu_lookup_table);
39 EXPORT_SYMBOL(numa_cpumask_lookup_table);
40 EXPORT_SYMBOL(node_data);
42 static int min_common_depth;
43 static int n_mem_addr_cells, n_mem_size_cells;
45 static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
48 unsigned long long mem;
50 static unsigned int fake_nid;
51 static unsigned long long curr_boundary;
54 * Modify node id, iff we started creating NUMA nodes
55 * We want to continue from where we left of the last time
60 * In case there are no more arguments to parse, the
61 * node_id should be the same as the last fake node id
62 * (we've handled this above).
67 mem = memparse(p, &p);
71 if (mem < curr_boundary)
76 if ((end_pfn << PAGE_SHIFT) > mem) {
78 * Skip commas and spaces
80 while (*p == ',' || *p == ' ' || *p == '\t')
86 dbg("created new fake_node with id %d\n", fake_nid);
93 * get_active_region_work_fn - A helper function for get_node_active_region
94 * Returns datax set to the start_pfn and end_pfn if they contain
95 * the initial value of datax->start_pfn between them
96 * @start_pfn: start page(inclusive) of region to check
97 * @end_pfn: end page(exclusive) of region to check
98 * @datax: comes in with ->start_pfn set to value to search for and
99 * goes out with active range if it contains it
100 * Returns 1 if search value is in range else 0
102 static int __init get_active_region_work_fn(unsigned long start_pfn,
103 unsigned long end_pfn, void *datax)
105 struct node_active_region *data;
106 data = (struct node_active_region *)datax;
108 if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) {
109 data->start_pfn = start_pfn;
110 data->end_pfn = end_pfn;
118 * get_node_active_region - Return active region containing start_pfn
119 * @start_pfn: The page to return the region for.
120 * @node_ar: Returned set to the active region containing start_pfn
122 static void __init get_node_active_region(unsigned long start_pfn,
123 struct node_active_region *node_ar)
125 int nid = early_pfn_to_nid(start_pfn);
128 node_ar->start_pfn = start_pfn;
129 work_with_active_regions(nid, get_active_region_work_fn, node_ar);
132 static void __cpuinit map_cpu_to_node(int cpu, int node)
134 numa_cpu_lookup_table[cpu] = node;
136 dbg("adding cpu %d to node %d\n", cpu, node);
138 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
139 cpu_set(cpu, numa_cpumask_lookup_table[node]);
142 #ifdef CONFIG_HOTPLUG_CPU
143 static void unmap_cpu_from_node(unsigned long cpu)
145 int node = numa_cpu_lookup_table[cpu];
147 dbg("removing cpu %lu from node %d\n", cpu, node);
149 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
150 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
152 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
156 #endif /* CONFIG_HOTPLUG_CPU */
158 static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
160 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
161 struct device_node *cpu_node = NULL;
162 const unsigned int *interrupt_server, *reg;
165 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
166 /* Try interrupt server first */
167 interrupt_server = of_get_property(cpu_node,
168 "ibm,ppc-interrupt-server#s", &len);
170 len = len / sizeof(u32);
172 if (interrupt_server && (len > 0)) {
174 if (interrupt_server[len] == hw_cpuid)
178 reg = of_get_property(cpu_node, "reg", &len);
179 if (reg && (len > 0) && (reg[0] == hw_cpuid))
187 /* must hold reference to node during call */
188 static const int *of_get_associativity(struct device_node *dev)
190 return of_get_property(dev, "ibm,associativity", NULL);
194 * Returns the property linux,drconf-usable-memory if
195 * it exists (the property exists only in kexec/kdump kernels,
196 * added by kexec-tools)
198 static const u32 *of_get_usable_memory(struct device_node *memory)
202 prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
203 if (!prop || len < sizeof(unsigned int))
208 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
211 static int of_node_to_nid_single(struct device_node *device)
214 const unsigned int *tmp;
216 if (min_common_depth == -1)
219 tmp = of_get_associativity(device);
223 if (tmp[0] >= min_common_depth)
224 nid = tmp[min_common_depth];
226 /* POWER4 LPAR uses 0xffff as invalid node */
227 if (nid == 0xffff || nid >= MAX_NUMNODES)
233 /* Walk the device tree upwards, looking for an associativity id */
234 int of_node_to_nid(struct device_node *device)
236 struct device_node *tmp;
241 nid = of_node_to_nid_single(device);
246 device = of_get_parent(tmp);
253 EXPORT_SYMBOL_GPL(of_node_to_nid);
256 * In theory, the "ibm,associativity" property may contain multiple
257 * associativity lists because a resource may be multiply connected
258 * into the machine. This resource then has different associativity
259 * characteristics relative to its multiple connections. We ignore
260 * this for now. We also assume that all cpu and memory sets have
261 * their distances represented at a common level. This won't be
262 * true for hierarchical NUMA.
264 * In any case the ibm,associativity-reference-points should give
265 * the correct depth for a normal NUMA system.
267 * - Dave Hansen <haveblue@us.ibm.com>
269 static int __init find_min_common_depth(void)
272 const unsigned int *ref_points;
273 struct device_node *rtas_root;
276 rtas_root = of_find_node_by_path("/rtas");
282 * this property is 2 32-bit integers, each representing a level of
283 * depth in the associativity nodes. The first is for an SMP
284 * configuration (should be all 0's) and the second is for a normal
285 * NUMA configuration.
287 ref_points = of_get_property(rtas_root,
288 "ibm,associativity-reference-points", &len);
290 if ((len >= 1) && ref_points) {
291 depth = ref_points[1];
293 dbg("NUMA: ibm,associativity-reference-points not found.\n");
296 of_node_put(rtas_root);
301 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
303 struct device_node *memory = NULL;
305 memory = of_find_node_by_type(memory, "memory");
307 panic("numa.c: No memory nodes found!");
309 *n_addr_cells = of_n_addr_cells(memory);
310 *n_size_cells = of_n_size_cells(memory);
314 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
316 unsigned long result = 0;
319 result = (result << 32) | **buf;
325 struct of_drconf_cell {
333 #define DRCONF_MEM_ASSIGNED 0x00000008
334 #define DRCONF_MEM_AI_INVALID 0x00000040
335 #define DRCONF_MEM_RESERVED 0x00000080
338 * Read the next lmb list entry from the ibm,dynamic-memory property
339 * and return the information in the provided of_drconf_cell structure.
341 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
345 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
348 drmem->drc_index = cp[0];
349 drmem->reserved = cp[1];
350 drmem->aa_index = cp[2];
351 drmem->flags = cp[3];
357 * Retreive and validate the ibm,dynamic-memory property of the device tree.
359 * The layout of the ibm,dynamic-memory property is a number N of lmb
360 * list entries followed by N lmb list entries. Each lmb list entry
361 * contains information as layed out in the of_drconf_cell struct above.
363 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
368 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
369 if (!prop || len < sizeof(unsigned int))
374 /* Now that we know the number of entries, revalidate the size
375 * of the property read in to ensure we have everything
377 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
385 * Retreive and validate the ibm,lmb-size property for drconf memory
386 * from the device tree.
388 static u64 of_get_lmb_size(struct device_node *memory)
393 prop = of_get_property(memory, "ibm,lmb-size", &len);
394 if (!prop || len < sizeof(unsigned int))
397 return read_n_cells(n_mem_size_cells, &prop);
400 struct assoc_arrays {
407 * Retreive and validate the list of associativity arrays for drconf
408 * memory from the ibm,associativity-lookup-arrays property of the
411 * The layout of the ibm,associativity-lookup-arrays property is a number N
412 * indicating the number of associativity arrays, followed by a number M
413 * indicating the size of each associativity array, followed by a list
414 * of N associativity arrays.
416 static int of_get_assoc_arrays(struct device_node *memory,
417 struct assoc_arrays *aa)
422 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
423 if (!prop || len < 2 * sizeof(unsigned int))
426 aa->n_arrays = *prop++;
427 aa->array_sz = *prop++;
429 /* Now that we know the number of arrrays and size of each array,
430 * revalidate the size of the property read in.
432 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
440 * This is like of_node_to_nid_single() for memory represented in the
441 * ibm,dynamic-reconfiguration-memory node.
443 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
444 struct assoc_arrays *aa)
447 int nid = default_nid;
450 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
451 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
452 drmem->aa_index < aa->n_arrays) {
453 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
454 nid = aa->arrays[index];
456 if (nid == 0xffff || nid >= MAX_NUMNODES)
464 * Figure out to which domain a cpu belongs and stick it there.
465 * Return the id of the domain used.
467 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
470 struct device_node *cpu = find_cpu_node(lcpu);
477 nid = of_node_to_nid_single(cpu);
479 if (nid < 0 || !node_online(nid))
480 nid = any_online_node(NODE_MASK_ALL);
482 map_cpu_to_node(lcpu, nid);
489 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
490 unsigned long action,
493 unsigned long lcpu = (unsigned long)hcpu;
494 int ret = NOTIFY_DONE;
498 case CPU_UP_PREPARE_FROZEN:
499 numa_setup_cpu(lcpu);
502 #ifdef CONFIG_HOTPLUG_CPU
504 case CPU_DEAD_FROZEN:
505 case CPU_UP_CANCELED:
506 case CPU_UP_CANCELED_FROZEN:
507 unmap_cpu_from_node(lcpu);
516 * Check and possibly modify a memory region to enforce the memory limit.
518 * Returns the size the region should have to enforce the memory limit.
519 * This will either be the original value of size, a truncated value,
520 * or zero. If the returned value of size is 0 the region should be
521 * discarded as it lies wholy above the memory limit.
523 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
527 * We use lmb_end_of_DRAM() in here instead of memory_limit because
528 * we've already adjusted it for the limit and it takes care of
529 * having memory holes below the limit.
535 if (start + size <= lmb_end_of_DRAM())
538 if (start >= lmb_end_of_DRAM())
541 return lmb_end_of_DRAM() - start;
545 * Reads the counter for a given entry in
546 * linux,drconf-usable-memory property
548 static inline int __init read_usm_ranges(const u32 **usm)
551 * For each lmb in ibm,dynamic-memory a corresponding
552 * entry in linux,drconf-usable-memory property contains
553 * a counter followed by that many (base, size) duple.
554 * read the counter from linux,drconf-usable-memory
556 return read_n_cells(n_mem_size_cells, usm);
560 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
561 * node. This assumes n_mem_{addr,size}_cells have been set.
563 static void __init parse_drconf_memory(struct device_node *memory)
566 unsigned int n, rc, ranges, is_kexec_kdump = 0;
567 unsigned long lmb_size, base, size, sz;
569 struct assoc_arrays aa;
571 n = of_get_drconf_memory(memory, &dm);
575 lmb_size = of_get_lmb_size(memory);
579 rc = of_get_assoc_arrays(memory, &aa);
583 /* check if this is a kexec/kdump kernel */
584 usm = of_get_usable_memory(memory);
588 for (; n != 0; --n) {
589 struct of_drconf_cell drmem;
591 read_drconf_cell(&drmem, &dm);
593 /* skip this block if the reserved bit is set in flags (0x80)
594 or if the block is not assigned to this partition (0x8) */
595 if ((drmem.flags & DRCONF_MEM_RESERVED)
596 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
599 base = drmem.base_addr;
603 if (is_kexec_kdump) {
604 ranges = read_usm_ranges(&usm);
605 if (!ranges) /* there are no (base, size) duple */
609 if (is_kexec_kdump) {
610 base = read_n_cells(n_mem_addr_cells, &usm);
611 size = read_n_cells(n_mem_size_cells, &usm);
613 nid = of_drconf_to_nid_single(&drmem, &aa);
614 fake_numa_create_new_node(
615 ((base + size) >> PAGE_SHIFT),
617 node_set_online(nid);
618 sz = numa_enforce_memory_limit(base, size);
620 add_active_range(nid, base >> PAGE_SHIFT,
622 + (sz >> PAGE_SHIFT));
627 static int __init parse_numa_properties(void)
629 struct device_node *cpu = NULL;
630 struct device_node *memory = NULL;
634 if (numa_enabled == 0) {
635 printk(KERN_WARNING "NUMA disabled by user\n");
639 min_common_depth = find_min_common_depth();
641 if (min_common_depth < 0)
642 return min_common_depth;
644 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
647 * Even though we connect cpus to numa domains later in SMP
648 * init, we need to know the node ids now. This is because
649 * each node to be onlined must have NODE_DATA etc backing it.
651 for_each_present_cpu(i) {
654 cpu = find_cpu_node(i);
656 nid = of_node_to_nid_single(cpu);
660 * Don't fall back to default_nid yet -- we will plug
661 * cpus into nodes once the memory scan has discovered
666 node_set_online(nid);
669 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
671 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
676 const unsigned int *memcell_buf;
679 memcell_buf = of_get_property(memory,
680 "linux,usable-memory", &len);
681 if (!memcell_buf || len <= 0)
682 memcell_buf = of_get_property(memory, "reg", &len);
683 if (!memcell_buf || len <= 0)
687 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
689 /* these are order-sensitive, and modify the buffer pointer */
690 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
691 size = read_n_cells(n_mem_size_cells, &memcell_buf);
694 * Assumption: either all memory nodes or none will
695 * have associativity properties. If none, then
696 * everything goes to default_nid.
698 nid = of_node_to_nid_single(memory);
702 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
703 node_set_online(nid);
705 if (!(size = numa_enforce_memory_limit(start, size))) {
712 add_active_range(nid, start >> PAGE_SHIFT,
713 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
720 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
721 * property in the ibm,dynamic-reconfiguration-memory node.
723 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
725 parse_drconf_memory(memory);
730 static void __init setup_nonnuma(void)
732 unsigned long top_of_ram = lmb_end_of_DRAM();
733 unsigned long total_ram = lmb_phys_mem_size();
734 unsigned long start_pfn, end_pfn;
735 unsigned int i, nid = 0;
737 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
738 top_of_ram, total_ram);
739 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
740 (top_of_ram - total_ram) >> 20);
742 for (i = 0; i < lmb.memory.cnt; ++i) {
743 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
744 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
746 fake_numa_create_new_node(end_pfn, &nid);
747 add_active_range(nid, start_pfn, end_pfn);
748 node_set_online(nid);
752 void __init dump_numa_cpu_topology(void)
755 unsigned int cpu, count;
757 if (min_common_depth == -1 || !numa_enabled)
760 for_each_online_node(node) {
761 printk(KERN_DEBUG "Node %d CPUs:", node);
765 * If we used a CPU iterator here we would miss printing
766 * the holes in the cpumap.
768 for (cpu = 0; cpu < NR_CPUS; cpu++) {
769 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
775 printk("-%u", cpu - 1);
781 printk("-%u", NR_CPUS - 1);
786 static void __init dump_numa_memory_topology(void)
791 if (min_common_depth == -1 || !numa_enabled)
794 for_each_online_node(node) {
797 printk(KERN_DEBUG "Node %d Memory:", node);
801 for (i = 0; i < lmb_end_of_DRAM();
802 i += (1 << SECTION_SIZE_BITS)) {
803 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
821 * Allocate some memory, satisfying the lmb or bootmem allocator where
822 * required. nid is the preferred node and end is the physical address of
823 * the highest address in the node.
825 * Returns the physical address of the memory.
827 static void __init *careful_allocation(int nid, unsigned long size,
829 unsigned long end_pfn)
832 unsigned long ret = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
834 /* retry over all memory */
836 ret = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
839 panic("numa.c: cannot allocate %lu bytes on node %d",
843 * If the memory came from a previously allocated node, we must
844 * retry with the bootmem allocator.
846 new_nid = early_pfn_to_nid(ret >> PAGE_SHIFT);
848 ret = (unsigned long)__alloc_bootmem_node(NODE_DATA(new_nid),
852 panic("numa.c: cannot allocate %lu bytes on node %d",
857 dbg("alloc_bootmem %lx %lx\n", ret, size);
863 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
864 .notifier_call = cpu_numa_callback,
865 .priority = 1 /* Must run before sched domains notifier. */
868 void __init do_init_bootmem(void)
874 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
875 max_pfn = max_low_pfn;
877 if (parse_numa_properties())
880 dump_numa_memory_topology();
882 register_cpu_notifier(&ppc64_numa_nb);
883 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
884 (void *)(unsigned long)boot_cpuid);
886 for_each_online_node(nid) {
887 unsigned long start_pfn, end_pfn;
888 unsigned long bootmem_paddr;
889 unsigned long bootmap_pages;
891 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
893 /* Allocate the node structure node local if possible */
894 NODE_DATA(nid) = careful_allocation(nid,
895 sizeof(struct pglist_data),
896 SMP_CACHE_BYTES, end_pfn);
897 NODE_DATA(nid) = __va(NODE_DATA(nid));
898 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
900 dbg("node %d\n", nid);
901 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
903 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
904 NODE_DATA(nid)->node_start_pfn = start_pfn;
905 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
907 if (NODE_DATA(nid)->node_spanned_pages == 0)
910 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
911 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
913 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
914 bootmem_paddr = (unsigned long)careful_allocation(nid,
915 bootmap_pages << PAGE_SHIFT,
917 memset(__va(bootmem_paddr), 0, bootmap_pages << PAGE_SHIFT);
919 dbg("bootmap_paddr = %lx\n", bootmem_paddr);
921 init_bootmem_node(NODE_DATA(nid), bootmem_paddr >> PAGE_SHIFT,
924 free_bootmem_with_active_regions(nid, end_pfn);
927 /* Mark reserved regions */
928 for (i = 0; i < lmb.reserved.cnt; i++) {
929 unsigned long physbase = lmb.reserved.region[i].base;
930 unsigned long size = lmb.reserved.region[i].size;
931 unsigned long start_pfn = physbase >> PAGE_SHIFT;
932 unsigned long end_pfn = ((physbase + size) >> PAGE_SHIFT);
933 struct node_active_region node_ar;
935 get_node_active_region(start_pfn, &node_ar);
936 while (start_pfn < end_pfn) {
938 * if reserved region extends past active region
939 * then trim size to active region
941 if (end_pfn > node_ar.end_pfn)
942 size = (node_ar.end_pfn << PAGE_SHIFT)
943 - (start_pfn << PAGE_SHIFT);
944 dbg("reserve_bootmem %lx %lx nid=%d\n", physbase, size,
946 reserve_bootmem_node(NODE_DATA(node_ar.nid), physbase,
947 size, BOOTMEM_DEFAULT);
949 * if reserved region is contained in the active region
952 if (end_pfn <= node_ar.end_pfn)
956 * reserved region extends past the active region
957 * get next active region that contains this
960 start_pfn = node_ar.end_pfn;
961 physbase = start_pfn << PAGE_SHIFT;
962 get_node_active_region(start_pfn, &node_ar);
967 for_each_online_node(nid)
968 sparse_memory_present_with_active_regions(nid);
971 void __init paging_init(void)
973 unsigned long max_zone_pfns[MAX_NR_ZONES];
974 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
975 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
976 free_area_init_nodes(max_zone_pfns);
979 static int __init early_numa(char *p)
984 if (strstr(p, "off"))
987 if (strstr(p, "debug"))
990 p = strstr(p, "fake=");
992 cmdline = p + strlen("fake=");
996 early_param("numa", early_numa);
998 #ifdef CONFIG_MEMORY_HOTPLUG
1000 * Validate the node associated with the memory section we are
1003 int valid_hot_add_scn(int *nid, unsigned long start, u32 lmb_size,
1004 unsigned long scn_addr)
1008 if (*nid < 0 || !node_online(*nid))
1009 *nid = any_online_node(NODE_MASK_ALL);
1011 if ((scn_addr >= start) && (scn_addr < (start + lmb_size))) {
1012 nodes_setall(nodes);
1013 while (NODE_DATA(*nid)->node_spanned_pages == 0) {
1014 node_clear(*nid, nodes);
1015 *nid = any_online_node(nodes);
1025 * Find the node associated with a hot added memory section represented
1026 * by the ibm,dynamic-reconfiguration-memory node.
1028 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1029 unsigned long scn_addr)
1033 unsigned long lmb_size;
1034 int default_nid = any_online_node(NODE_MASK_ALL);
1036 struct assoc_arrays aa;
1038 n = of_get_drconf_memory(memory, &dm);
1040 return default_nid;;
1042 lmb_size = of_get_lmb_size(memory);
1046 rc = of_get_assoc_arrays(memory, &aa);
1050 for (; n != 0; --n) {
1051 struct of_drconf_cell drmem;
1053 read_drconf_cell(&drmem, &dm);
1055 /* skip this block if it is reserved or not assigned to
1057 if ((drmem.flags & DRCONF_MEM_RESERVED)
1058 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1061 nid = of_drconf_to_nid_single(&drmem, &aa);
1063 if (valid_hot_add_scn(&nid, drmem.base_addr, lmb_size,
1068 BUG(); /* section address should be found above */
1073 * Find the node associated with a hot added memory section. Section
1074 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1075 * sections are fully contained within a single LMB.
1077 int hot_add_scn_to_nid(unsigned long scn_addr)
1079 struct device_node *memory = NULL;
1082 if (!numa_enabled || (min_common_depth < 0))
1083 return any_online_node(NODE_MASK_ALL);
1085 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1087 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1088 of_node_put(memory);
1092 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1093 unsigned long start, size;
1095 const unsigned int *memcell_buf;
1098 memcell_buf = of_get_property(memory, "reg", &len);
1099 if (!memcell_buf || len <= 0)
1102 /* ranges in cell */
1103 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1105 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1106 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1107 nid = of_node_to_nid_single(memory);
1109 if (valid_hot_add_scn(&nid, start, size, scn_addr)) {
1110 of_node_put(memory);
1114 if (--ranges) /* process all ranges in cell */
1117 BUG(); /* section address should be found above */
1120 #endif /* CONFIG_MEMORY_HOTPLUG */