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 <linux/pfn.h>
23 #include <asm/sparsemem.h>
25 #include <asm/system.h>
28 static int numa_enabled = 1;
30 static char *cmdline __initdata;
32 static int numa_debug;
33 #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
35 int numa_cpu_lookup_table[NR_CPUS];
36 cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
37 struct pglist_data *node_data[MAX_NUMNODES];
39 EXPORT_SYMBOL(numa_cpu_lookup_table);
40 EXPORT_SYMBOL(numa_cpumask_lookup_table);
41 EXPORT_SYMBOL(node_data);
43 static int min_common_depth;
44 static int n_mem_addr_cells, n_mem_size_cells;
46 static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
49 unsigned long long mem;
51 static unsigned int fake_nid;
52 static unsigned long long curr_boundary;
55 * Modify node id, iff we started creating NUMA nodes
56 * We want to continue from where we left of the last time
61 * In case there are no more arguments to parse, the
62 * node_id should be the same as the last fake node id
63 * (we've handled this above).
68 mem = memparse(p, &p);
72 if (mem < curr_boundary)
77 if ((end_pfn << PAGE_SHIFT) > mem) {
79 * Skip commas and spaces
81 while (*p == ',' || *p == ' ' || *p == '\t')
87 dbg("created new fake_node with id %d\n", fake_nid);
94 * get_active_region_work_fn - A helper function for get_node_active_region
95 * Returns datax set to the start_pfn and end_pfn if they contain
96 * the initial value of datax->start_pfn between them
97 * @start_pfn: start page(inclusive) of region to check
98 * @end_pfn: end page(exclusive) of region to check
99 * @datax: comes in with ->start_pfn set to value to search for and
100 * goes out with active range if it contains it
101 * Returns 1 if search value is in range else 0
103 static int __init get_active_region_work_fn(unsigned long start_pfn,
104 unsigned long end_pfn, void *datax)
106 struct node_active_region *data;
107 data = (struct node_active_region *)datax;
109 if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) {
110 data->start_pfn = start_pfn;
111 data->end_pfn = end_pfn;
119 * get_node_active_region - Return active region containing start_pfn
120 * Active range returned is empty if none found.
121 * @start_pfn: The page to return the region for.
122 * @node_ar: Returned set to the active region containing start_pfn
124 static void __init get_node_active_region(unsigned long start_pfn,
125 struct node_active_region *node_ar)
127 int nid = early_pfn_to_nid(start_pfn);
130 node_ar->start_pfn = start_pfn;
131 node_ar->end_pfn = start_pfn;
132 work_with_active_regions(nid, get_active_region_work_fn, node_ar);
135 static void __cpuinit map_cpu_to_node(int cpu, int node)
137 numa_cpu_lookup_table[cpu] = node;
139 dbg("adding cpu %d to node %d\n", cpu, node);
141 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
142 cpu_set(cpu, numa_cpumask_lookup_table[node]);
145 #ifdef CONFIG_HOTPLUG_CPU
146 static void unmap_cpu_from_node(unsigned long cpu)
148 int node = numa_cpu_lookup_table[cpu];
150 dbg("removing cpu %lu from node %d\n", cpu, node);
152 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
153 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
155 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
159 #endif /* CONFIG_HOTPLUG_CPU */
161 static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
163 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
164 struct device_node *cpu_node = NULL;
165 const unsigned int *interrupt_server, *reg;
168 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
169 /* Try interrupt server first */
170 interrupt_server = of_get_property(cpu_node,
171 "ibm,ppc-interrupt-server#s", &len);
173 len = len / sizeof(u32);
175 if (interrupt_server && (len > 0)) {
177 if (interrupt_server[len] == hw_cpuid)
181 reg = of_get_property(cpu_node, "reg", &len);
182 if (reg && (len > 0) && (reg[0] == hw_cpuid))
190 /* must hold reference to node during call */
191 static const int *of_get_associativity(struct device_node *dev)
193 return of_get_property(dev, "ibm,associativity", NULL);
197 * Returns the property linux,drconf-usable-memory if
198 * it exists (the property exists only in kexec/kdump kernels,
199 * added by kexec-tools)
201 static const u32 *of_get_usable_memory(struct device_node *memory)
205 prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
206 if (!prop || len < sizeof(unsigned int))
211 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
214 static int of_node_to_nid_single(struct device_node *device)
217 const unsigned int *tmp;
219 if (min_common_depth == -1)
222 tmp = of_get_associativity(device);
226 if (tmp[0] >= min_common_depth)
227 nid = tmp[min_common_depth];
229 /* POWER4 LPAR uses 0xffff as invalid node */
230 if (nid == 0xffff || nid >= MAX_NUMNODES)
236 /* Walk the device tree upwards, looking for an associativity id */
237 int of_node_to_nid(struct device_node *device)
239 struct device_node *tmp;
244 nid = of_node_to_nid_single(device);
249 device = of_get_parent(tmp);
256 EXPORT_SYMBOL_GPL(of_node_to_nid);
259 * In theory, the "ibm,associativity" property may contain multiple
260 * associativity lists because a resource may be multiply connected
261 * into the machine. This resource then has different associativity
262 * characteristics relative to its multiple connections. We ignore
263 * this for now. We also assume that all cpu and memory sets have
264 * their distances represented at a common level. This won't be
265 * true for hierarchical NUMA.
267 * In any case the ibm,associativity-reference-points should give
268 * the correct depth for a normal NUMA system.
270 * - Dave Hansen <haveblue@us.ibm.com>
272 static int __init find_min_common_depth(void)
275 const unsigned int *ref_points;
276 struct device_node *rtas_root;
279 rtas_root = of_find_node_by_path("/rtas");
285 * this property is 2 32-bit integers, each representing a level of
286 * depth in the associativity nodes. The first is for an SMP
287 * configuration (should be all 0's) and the second is for a normal
288 * NUMA configuration.
290 ref_points = of_get_property(rtas_root,
291 "ibm,associativity-reference-points", &len);
293 if ((len >= 1) && ref_points) {
294 depth = ref_points[1];
296 dbg("NUMA: ibm,associativity-reference-points not found.\n");
299 of_node_put(rtas_root);
304 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
306 struct device_node *memory = NULL;
308 memory = of_find_node_by_type(memory, "memory");
310 panic("numa.c: No memory nodes found!");
312 *n_addr_cells = of_n_addr_cells(memory);
313 *n_size_cells = of_n_size_cells(memory);
317 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
319 unsigned long result = 0;
322 result = (result << 32) | **buf;
328 struct of_drconf_cell {
336 #define DRCONF_MEM_ASSIGNED 0x00000008
337 #define DRCONF_MEM_AI_INVALID 0x00000040
338 #define DRCONF_MEM_RESERVED 0x00000080
341 * Read the next lmb list entry from the ibm,dynamic-memory property
342 * and return the information in the provided of_drconf_cell structure.
344 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
348 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
351 drmem->drc_index = cp[0];
352 drmem->reserved = cp[1];
353 drmem->aa_index = cp[2];
354 drmem->flags = cp[3];
360 * Retreive and validate the ibm,dynamic-memory property of the device tree.
362 * The layout of the ibm,dynamic-memory property is a number N of lmb
363 * list entries followed by N lmb list entries. Each lmb list entry
364 * contains information as layed out in the of_drconf_cell struct above.
366 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
371 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
372 if (!prop || len < sizeof(unsigned int))
377 /* Now that we know the number of entries, revalidate the size
378 * of the property read in to ensure we have everything
380 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
388 * Retreive and validate the ibm,lmb-size property for drconf memory
389 * from the device tree.
391 static u64 of_get_lmb_size(struct device_node *memory)
396 prop = of_get_property(memory, "ibm,lmb-size", &len);
397 if (!prop || len < sizeof(unsigned int))
400 return read_n_cells(n_mem_size_cells, &prop);
403 struct assoc_arrays {
410 * Retreive and validate the list of associativity arrays for drconf
411 * memory from the ibm,associativity-lookup-arrays property of the
414 * The layout of the ibm,associativity-lookup-arrays property is a number N
415 * indicating the number of associativity arrays, followed by a number M
416 * indicating the size of each associativity array, followed by a list
417 * of N associativity arrays.
419 static int of_get_assoc_arrays(struct device_node *memory,
420 struct assoc_arrays *aa)
425 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
426 if (!prop || len < 2 * sizeof(unsigned int))
429 aa->n_arrays = *prop++;
430 aa->array_sz = *prop++;
432 /* Now that we know the number of arrrays and size of each array,
433 * revalidate the size of the property read in.
435 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
443 * This is like of_node_to_nid_single() for memory represented in the
444 * ibm,dynamic-reconfiguration-memory node.
446 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
447 struct assoc_arrays *aa)
450 int nid = default_nid;
453 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
454 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
455 drmem->aa_index < aa->n_arrays) {
456 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
457 nid = aa->arrays[index];
459 if (nid == 0xffff || nid >= MAX_NUMNODES)
467 * Figure out to which domain a cpu belongs and stick it there.
468 * Return the id of the domain used.
470 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
473 struct device_node *cpu = find_cpu_node(lcpu);
480 nid = of_node_to_nid_single(cpu);
482 if (nid < 0 || !node_online(nid))
483 nid = any_online_node(NODE_MASK_ALL);
485 map_cpu_to_node(lcpu, nid);
492 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
493 unsigned long action,
496 unsigned long lcpu = (unsigned long)hcpu;
497 int ret = NOTIFY_DONE;
501 case CPU_UP_PREPARE_FROZEN:
502 numa_setup_cpu(lcpu);
505 #ifdef CONFIG_HOTPLUG_CPU
507 case CPU_DEAD_FROZEN:
508 case CPU_UP_CANCELED:
509 case CPU_UP_CANCELED_FROZEN:
510 unmap_cpu_from_node(lcpu);
519 * Check and possibly modify a memory region to enforce the memory limit.
521 * Returns the size the region should have to enforce the memory limit.
522 * This will either be the original value of size, a truncated value,
523 * or zero. If the returned value of size is 0 the region should be
524 * discarded as it lies wholy above the memory limit.
526 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
530 * We use lmb_end_of_DRAM() in here instead of memory_limit because
531 * we've already adjusted it for the limit and it takes care of
532 * having memory holes below the limit. Also, in the case of
533 * iommu_is_off, memory_limit is not set but is implicitly enforced.
536 if (start + size <= lmb_end_of_DRAM())
539 if (start >= lmb_end_of_DRAM())
542 return lmb_end_of_DRAM() - start;
546 * Reads the counter for a given entry in
547 * linux,drconf-usable-memory property
549 static inline int __init read_usm_ranges(const u32 **usm)
552 * For each lmb in ibm,dynamic-memory a corresponding
553 * entry in linux,drconf-usable-memory property contains
554 * a counter followed by that many (base, size) duple.
555 * read the counter from linux,drconf-usable-memory
557 return read_n_cells(n_mem_size_cells, usm);
561 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
562 * node. This assumes n_mem_{addr,size}_cells have been set.
564 static void __init parse_drconf_memory(struct device_node *memory)
567 unsigned int n, rc, ranges, is_kexec_kdump = 0;
568 unsigned long lmb_size, base, size, sz;
570 struct assoc_arrays aa;
572 n = of_get_drconf_memory(memory, &dm);
576 lmb_size = of_get_lmb_size(memory);
580 rc = of_get_assoc_arrays(memory, &aa);
584 /* check if this is a kexec/kdump kernel */
585 usm = of_get_usable_memory(memory);
589 for (; n != 0; --n) {
590 struct of_drconf_cell drmem;
592 read_drconf_cell(&drmem, &dm);
594 /* skip this block if the reserved bit is set in flags (0x80)
595 or if the block is not assigned to this partition (0x8) */
596 if ((drmem.flags & DRCONF_MEM_RESERVED)
597 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
600 base = drmem.base_addr;
604 if (is_kexec_kdump) {
605 ranges = read_usm_ranges(&usm);
606 if (!ranges) /* there are no (base, size) duple */
610 if (is_kexec_kdump) {
611 base = read_n_cells(n_mem_addr_cells, &usm);
612 size = read_n_cells(n_mem_size_cells, &usm);
614 nid = of_drconf_to_nid_single(&drmem, &aa);
615 fake_numa_create_new_node(
616 ((base + size) >> PAGE_SHIFT),
618 node_set_online(nid);
619 sz = numa_enforce_memory_limit(base, size);
621 add_active_range(nid, base >> PAGE_SHIFT,
623 + (sz >> PAGE_SHIFT));
628 static int __init parse_numa_properties(void)
630 struct device_node *cpu = NULL;
631 struct device_node *memory = NULL;
635 if (numa_enabled == 0) {
636 printk(KERN_WARNING "NUMA disabled by user\n");
640 min_common_depth = find_min_common_depth();
642 if (min_common_depth < 0)
643 return min_common_depth;
645 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
648 * Even though we connect cpus to numa domains later in SMP
649 * init, we need to know the node ids now. This is because
650 * each node to be onlined must have NODE_DATA etc backing it.
652 for_each_present_cpu(i) {
655 cpu = find_cpu_node(i);
657 nid = of_node_to_nid_single(cpu);
661 * Don't fall back to default_nid yet -- we will plug
662 * cpus into nodes once the memory scan has discovered
667 node_set_online(nid);
670 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
672 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
677 const unsigned int *memcell_buf;
680 memcell_buf = of_get_property(memory,
681 "linux,usable-memory", &len);
682 if (!memcell_buf || len <= 0)
683 memcell_buf = of_get_property(memory, "reg", &len);
684 if (!memcell_buf || len <= 0)
688 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
690 /* these are order-sensitive, and modify the buffer pointer */
691 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
692 size = read_n_cells(n_mem_size_cells, &memcell_buf);
695 * Assumption: either all memory nodes or none will
696 * have associativity properties. If none, then
697 * everything goes to default_nid.
699 nid = of_node_to_nid_single(memory);
703 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
704 node_set_online(nid);
706 if (!(size = numa_enforce_memory_limit(start, size))) {
713 add_active_range(nid, start >> PAGE_SHIFT,
714 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
721 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
722 * property in the ibm,dynamic-reconfiguration-memory node.
724 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
726 parse_drconf_memory(memory);
731 static void __init setup_nonnuma(void)
733 unsigned long top_of_ram = lmb_end_of_DRAM();
734 unsigned long total_ram = lmb_phys_mem_size();
735 unsigned long start_pfn, end_pfn;
736 unsigned int i, nid = 0;
738 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
739 top_of_ram, total_ram);
740 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
741 (top_of_ram - total_ram) >> 20);
743 for (i = 0; i < lmb.memory.cnt; ++i) {
744 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
745 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
747 fake_numa_create_new_node(end_pfn, &nid);
748 add_active_range(nid, start_pfn, end_pfn);
749 node_set_online(nid);
753 void __init dump_numa_cpu_topology(void)
756 unsigned int cpu, count;
758 if (min_common_depth == -1 || !numa_enabled)
761 for_each_online_node(node) {
762 printk(KERN_DEBUG "Node %d CPUs:", node);
766 * If we used a CPU iterator here we would miss printing
767 * the holes in the cpumap.
769 for (cpu = 0; cpu < NR_CPUS; cpu++) {
770 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
776 printk("-%u", cpu - 1);
782 printk("-%u", NR_CPUS - 1);
787 static void __init dump_numa_memory_topology(void)
792 if (min_common_depth == -1 || !numa_enabled)
795 for_each_online_node(node) {
798 printk(KERN_DEBUG "Node %d Memory:", node);
802 for (i = 0; i < lmb_end_of_DRAM();
803 i += (1 << SECTION_SIZE_BITS)) {
804 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
822 * Allocate some memory, satisfying the lmb or bootmem allocator where
823 * required. nid is the preferred node and end is the physical address of
824 * the highest address in the node.
826 * Returns the virtual address of the memory.
828 static void __init *careful_zallocation(int nid, unsigned long size,
830 unsigned long end_pfn)
834 unsigned long ret_paddr;
836 ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
838 /* retry over all memory */
840 ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
843 panic("numa.c: cannot allocate %lu bytes for node %d",
846 ret = __va(ret_paddr);
849 * We initialize the nodes in numeric order: 0, 1, 2...
850 * and hand over control from the LMB allocator to the
851 * bootmem allocator. If this function is called for
852 * node 5, then we know that all nodes <5 are using the
853 * bootmem allocator instead of the LMB allocator.
855 * So, check the nid from which this allocation came
856 * and double check to see if we need to use bootmem
857 * instead of the LMB. We don't free the LMB memory
858 * since it would be useless.
860 new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
862 ret = __alloc_bootmem_node(NODE_DATA(new_nid),
865 dbg("alloc_bootmem %p %lx\n", ret, size);
868 memset(ret, 0, size);
872 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
873 .notifier_call = cpu_numa_callback,
874 .priority = 1 /* Must run before sched domains notifier. */
877 static void mark_reserved_regions_for_nid(int nid)
879 struct pglist_data *node = NODE_DATA(nid);
882 for (i = 0; i < lmb.reserved.cnt; i++) {
883 unsigned long physbase = lmb.reserved.region[i].base;
884 unsigned long size = lmb.reserved.region[i].size;
885 unsigned long start_pfn = physbase >> PAGE_SHIFT;
886 unsigned long end_pfn = PFN_UP(physbase + size);
887 struct node_active_region node_ar;
888 unsigned long node_end_pfn = node->node_start_pfn +
889 node->node_spanned_pages;
892 * Check to make sure that this lmb.reserved area is
893 * within the bounds of the node that we care about.
894 * Checking the nid of the start and end points is not
895 * sufficient because the reserved area could span the
898 if (end_pfn <= node->node_start_pfn ||
899 start_pfn >= node_end_pfn)
902 get_node_active_region(start_pfn, &node_ar);
903 while (start_pfn < end_pfn &&
904 node_ar.start_pfn < node_ar.end_pfn) {
905 unsigned long reserve_size = size;
907 * if reserved region extends past active region
908 * then trim size to active region
910 if (end_pfn > node_ar.end_pfn)
911 reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
914 * Only worry about *this* node, others may not
915 * yet have valid NODE_DATA().
917 if (node_ar.nid == nid) {
918 dbg("reserve_bootmem %lx %lx nid=%d\n",
919 physbase, reserve_size, node_ar.nid);
920 reserve_bootmem_node(NODE_DATA(node_ar.nid),
921 physbase, reserve_size,
925 * if reserved region is contained in the active region
928 if (end_pfn <= node_ar.end_pfn)
932 * reserved region extends past the active region
933 * get next active region that contains this
936 start_pfn = node_ar.end_pfn;
937 physbase = start_pfn << PAGE_SHIFT;
938 size = size - reserve_size;
939 get_node_active_region(start_pfn, &node_ar);
945 void __init do_init_bootmem(void)
950 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
951 max_pfn = max_low_pfn;
953 if (parse_numa_properties())
956 dump_numa_memory_topology();
958 register_cpu_notifier(&ppc64_numa_nb);
959 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
960 (void *)(unsigned long)boot_cpuid);
962 for_each_online_node(nid) {
963 unsigned long start_pfn, end_pfn;
965 unsigned long bootmap_pages;
967 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
970 * Allocate the node structure node local if possible
972 * Be careful moving this around, as it relies on all
973 * previous nodes' bootmem to be initialized and have
974 * all reserved areas marked.
976 NODE_DATA(nid) = careful_zallocation(nid,
977 sizeof(struct pglist_data),
978 SMP_CACHE_BYTES, end_pfn);
980 dbg("node %d\n", nid);
981 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
983 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
984 NODE_DATA(nid)->node_start_pfn = start_pfn;
985 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
987 if (NODE_DATA(nid)->node_spanned_pages == 0)
990 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
991 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
993 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
994 bootmem_vaddr = careful_zallocation(nid,
995 bootmap_pages << PAGE_SHIFT,
998 dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
1000 init_bootmem_node(NODE_DATA(nid),
1001 __pa(bootmem_vaddr) >> PAGE_SHIFT,
1002 start_pfn, end_pfn);
1004 free_bootmem_with_active_regions(nid, end_pfn);
1006 * Be very careful about moving this around. Future
1007 * calls to careful_zallocation() depend on this getting
1010 mark_reserved_regions_for_nid(nid);
1011 sparse_memory_present_with_active_regions(nid);
1015 void __init paging_init(void)
1017 unsigned long max_zone_pfns[MAX_NR_ZONES];
1018 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
1019 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
1020 free_area_init_nodes(max_zone_pfns);
1023 static int __init early_numa(char *p)
1028 if (strstr(p, "off"))
1031 if (strstr(p, "debug"))
1034 p = strstr(p, "fake=");
1036 cmdline = p + strlen("fake=");
1040 early_param("numa", early_numa);
1042 #ifdef CONFIG_MEMORY_HOTPLUG
1044 * Validate the node associated with the memory section we are
1047 int valid_hot_add_scn(int *nid, unsigned long start, u32 lmb_size,
1048 unsigned long scn_addr)
1052 if (*nid < 0 || !node_online(*nid))
1053 *nid = any_online_node(NODE_MASK_ALL);
1055 if ((scn_addr >= start) && (scn_addr < (start + lmb_size))) {
1056 nodes_setall(nodes);
1057 while (NODE_DATA(*nid)->node_spanned_pages == 0) {
1058 node_clear(*nid, nodes);
1059 *nid = any_online_node(nodes);
1069 * Find the node associated with a hot added memory section represented
1070 * by the ibm,dynamic-reconfiguration-memory node.
1072 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1073 unsigned long scn_addr)
1077 unsigned long lmb_size;
1078 int default_nid = any_online_node(NODE_MASK_ALL);
1080 struct assoc_arrays aa;
1082 n = of_get_drconf_memory(memory, &dm);
1084 return default_nid;;
1086 lmb_size = of_get_lmb_size(memory);
1090 rc = of_get_assoc_arrays(memory, &aa);
1094 for (; n != 0; --n) {
1095 struct of_drconf_cell drmem;
1097 read_drconf_cell(&drmem, &dm);
1099 /* skip this block if it is reserved or not assigned to
1101 if ((drmem.flags & DRCONF_MEM_RESERVED)
1102 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1105 nid = of_drconf_to_nid_single(&drmem, &aa);
1107 if (valid_hot_add_scn(&nid, drmem.base_addr, lmb_size,
1112 BUG(); /* section address should be found above */
1117 * Find the node associated with a hot added memory section. Section
1118 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1119 * sections are fully contained within a single LMB.
1121 int hot_add_scn_to_nid(unsigned long scn_addr)
1123 struct device_node *memory = NULL;
1126 if (!numa_enabled || (min_common_depth < 0))
1127 return any_online_node(NODE_MASK_ALL);
1129 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1131 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1132 of_node_put(memory);
1136 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1137 unsigned long start, size;
1139 const unsigned int *memcell_buf;
1142 memcell_buf = of_get_property(memory, "reg", &len);
1143 if (!memcell_buf || len <= 0)
1146 /* ranges in cell */
1147 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1149 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1150 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1151 nid = of_node_to_nid_single(memory);
1153 if (valid_hot_add_scn(&nid, start, size, scn_addr)) {
1154 of_node_put(memory);
1158 if (--ranges) /* process all ranges in cell */
1161 BUG(); /* section address should be found above */
1164 #endif /* CONFIG_MEMORY_HOTPLUG */