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/memblock.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_var_t node_to_cpumask_map[MAX_NUMNODES];
37 struct pglist_data *node_data[MAX_NUMNODES];
39 EXPORT_SYMBOL(numa_cpu_lookup_table);
40 EXPORT_SYMBOL(node_to_cpumask_map);
41 EXPORT_SYMBOL(node_data);
43 static int min_common_depth;
44 static int n_mem_addr_cells, n_mem_size_cells;
47 * Allocate node_to_cpumask_map based on number of available nodes
48 * Requires node_possible_map to be valid.
50 * Note: node_to_cpumask() is not valid until after this is done.
52 static void __init setup_node_to_cpumask_map(void)
54 unsigned int node, num = 0;
56 /* setup nr_node_ids if not done yet */
57 if (nr_node_ids == MAX_NUMNODES) {
58 for_each_node_mask(node, node_possible_map)
60 nr_node_ids = num + 1;
63 /* allocate the map */
64 for (node = 0; node < nr_node_ids; node++)
65 alloc_bootmem_cpumask_var(&node_to_cpumask_map[node]);
67 /* cpumask_of_node() will now work */
68 dbg("Node to cpumask map for %d nodes\n", nr_node_ids);
71 static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
74 unsigned long long mem;
76 static unsigned int fake_nid;
77 static unsigned long long curr_boundary;
80 * Modify node id, iff we started creating NUMA nodes
81 * We want to continue from where we left of the last time
86 * In case there are no more arguments to parse, the
87 * node_id should be the same as the last fake node id
88 * (we've handled this above).
93 mem = memparse(p, &p);
97 if (mem < curr_boundary)
102 if ((end_pfn << PAGE_SHIFT) > mem) {
104 * Skip commas and spaces
106 while (*p == ',' || *p == ' ' || *p == '\t')
112 dbg("created new fake_node with id %d\n", fake_nid);
119 * get_active_region_work_fn - A helper function for get_node_active_region
120 * Returns datax set to the start_pfn and end_pfn if they contain
121 * the initial value of datax->start_pfn between them
122 * @start_pfn: start page(inclusive) of region to check
123 * @end_pfn: end page(exclusive) of region to check
124 * @datax: comes in with ->start_pfn set to value to search for and
125 * goes out with active range if it contains it
126 * Returns 1 if search value is in range else 0
128 static int __init get_active_region_work_fn(unsigned long start_pfn,
129 unsigned long end_pfn, void *datax)
131 struct node_active_region *data;
132 data = (struct node_active_region *)datax;
134 if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) {
135 data->start_pfn = start_pfn;
136 data->end_pfn = end_pfn;
144 * get_node_active_region - Return active region containing start_pfn
145 * Active range returned is empty if none found.
146 * @start_pfn: The page to return the region for.
147 * @node_ar: Returned set to the active region containing start_pfn
149 static void __init get_node_active_region(unsigned long start_pfn,
150 struct node_active_region *node_ar)
152 int nid = early_pfn_to_nid(start_pfn);
155 node_ar->start_pfn = start_pfn;
156 node_ar->end_pfn = start_pfn;
157 work_with_active_regions(nid, get_active_region_work_fn, node_ar);
160 static void __cpuinit map_cpu_to_node(int cpu, int node)
162 numa_cpu_lookup_table[cpu] = node;
164 dbg("adding cpu %d to node %d\n", cpu, node);
166 if (!(cpumask_test_cpu(cpu, node_to_cpumask_map[node])))
167 cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
170 #ifdef CONFIG_HOTPLUG_CPU
171 static void unmap_cpu_from_node(unsigned long cpu)
173 int node = numa_cpu_lookup_table[cpu];
175 dbg("removing cpu %lu from node %d\n", cpu, node);
177 if (cpumask_test_cpu(cpu, node_to_cpumask_map[node])) {
178 cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
180 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
184 #endif /* CONFIG_HOTPLUG_CPU */
186 /* must hold reference to node during call */
187 static const int *of_get_associativity(struct device_node *dev)
189 return of_get_property(dev, "ibm,associativity", NULL);
193 * Returns the property linux,drconf-usable-memory if
194 * it exists (the property exists only in kexec/kdump kernels,
195 * added by kexec-tools)
197 static const u32 *of_get_usable_memory(struct device_node *memory)
201 prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
202 if (!prop || len < sizeof(unsigned int))
207 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
210 static int of_node_to_nid_single(struct device_node *device)
213 const unsigned int *tmp;
215 if (min_common_depth == -1)
218 tmp = of_get_associativity(device);
222 if (tmp[0] >= min_common_depth)
223 nid = tmp[min_common_depth];
225 /* POWER4 LPAR uses 0xffff as invalid node */
226 if (nid == 0xffff || nid >= MAX_NUMNODES)
232 /* Walk the device tree upwards, looking for an associativity id */
233 int of_node_to_nid(struct device_node *device)
235 struct device_node *tmp;
240 nid = of_node_to_nid_single(device);
245 device = of_get_parent(tmp);
252 EXPORT_SYMBOL_GPL(of_node_to_nid);
255 * In theory, the "ibm,associativity" property may contain multiple
256 * associativity lists because a resource may be multiply connected
257 * into the machine. This resource then has different associativity
258 * characteristics relative to its multiple connections. We ignore
259 * this for now. We also assume that all cpu and memory sets have
260 * their distances represented at a common level. This won't be
261 * true for hierarchical NUMA.
263 * In any case the ibm,associativity-reference-points should give
264 * the correct depth for a normal NUMA system.
266 * - Dave Hansen <haveblue@us.ibm.com>
268 static int __init find_min_common_depth(void)
271 const unsigned int *ref_points;
272 struct device_node *rtas_root;
274 struct device_node *chosen;
277 rtas_root = of_find_node_by_path("/rtas");
283 * this property is 2 32-bit integers, each representing a level of
284 * depth in the associativity nodes. The first is for an SMP
285 * configuration (should be all 0's) and the second is for a normal
286 * NUMA configuration.
289 ref_points = of_get_property(rtas_root,
290 "ibm,associativity-reference-points", &len);
293 * For form 1 affinity information we want the first field
295 #define VEC5_AFFINITY_BYTE 5
296 #define VEC5_AFFINITY 0x80
297 chosen = of_find_node_by_path("/chosen");
299 vec5 = of_get_property(chosen, "ibm,architecture-vec-5", NULL);
300 if (vec5 && (vec5[VEC5_AFFINITY_BYTE] & VEC5_AFFINITY)) {
301 dbg("Using form 1 affinity\n");
306 if ((len >= 2 * sizeof(unsigned int)) && ref_points) {
307 depth = ref_points[index];
309 dbg("NUMA: ibm,associativity-reference-points not found.\n");
312 of_node_put(rtas_root);
317 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
319 struct device_node *memory = NULL;
321 memory = of_find_node_by_type(memory, "memory");
323 panic("numa.c: No memory nodes found!");
325 *n_addr_cells = of_n_addr_cells(memory);
326 *n_size_cells = of_n_size_cells(memory);
330 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
332 unsigned long result = 0;
335 result = (result << 32) | **buf;
341 struct of_drconf_cell {
349 #define DRCONF_MEM_ASSIGNED 0x00000008
350 #define DRCONF_MEM_AI_INVALID 0x00000040
351 #define DRCONF_MEM_RESERVED 0x00000080
354 * Read the next memblock list entry from the ibm,dynamic-memory property
355 * and return the information in the provided of_drconf_cell structure.
357 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
361 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
364 drmem->drc_index = cp[0];
365 drmem->reserved = cp[1];
366 drmem->aa_index = cp[2];
367 drmem->flags = cp[3];
373 * Retreive and validate the ibm,dynamic-memory property of the device tree.
375 * The layout of the ibm,dynamic-memory property is a number N of memblock
376 * list entries followed by N memblock list entries. Each memblock list entry
377 * contains information as layed out in the of_drconf_cell struct above.
379 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
384 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
385 if (!prop || len < sizeof(unsigned int))
390 /* Now that we know the number of entries, revalidate the size
391 * of the property read in to ensure we have everything
393 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
401 * Retreive and validate the ibm,memblock-size property for drconf memory
402 * from the device tree.
404 static u64 of_get_memblock_size(struct device_node *memory)
409 prop = of_get_property(memory, "ibm,memblock-size", &len);
410 if (!prop || len < sizeof(unsigned int))
413 return read_n_cells(n_mem_size_cells, &prop);
416 struct assoc_arrays {
423 * Retreive and validate the list of associativity arrays for drconf
424 * memory from the ibm,associativity-lookup-arrays property of the
427 * The layout of the ibm,associativity-lookup-arrays property is a number N
428 * indicating the number of associativity arrays, followed by a number M
429 * indicating the size of each associativity array, followed by a list
430 * of N associativity arrays.
432 static int of_get_assoc_arrays(struct device_node *memory,
433 struct assoc_arrays *aa)
438 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
439 if (!prop || len < 2 * sizeof(unsigned int))
442 aa->n_arrays = *prop++;
443 aa->array_sz = *prop++;
445 /* Now that we know the number of arrrays and size of each array,
446 * revalidate the size of the property read in.
448 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
456 * This is like of_node_to_nid_single() for memory represented in the
457 * ibm,dynamic-reconfiguration-memory node.
459 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
460 struct assoc_arrays *aa)
463 int nid = default_nid;
466 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
467 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
468 drmem->aa_index < aa->n_arrays) {
469 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
470 nid = aa->arrays[index];
472 if (nid == 0xffff || nid >= MAX_NUMNODES)
480 * Figure out to which domain a cpu belongs and stick it there.
481 * Return the id of the domain used.
483 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
486 struct device_node *cpu = of_get_cpu_node(lcpu, NULL);
493 nid = of_node_to_nid_single(cpu);
495 if (nid < 0 || !node_online(nid))
496 nid = first_online_node;
498 map_cpu_to_node(lcpu, nid);
505 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
506 unsigned long action,
509 unsigned long lcpu = (unsigned long)hcpu;
510 int ret = NOTIFY_DONE;
514 case CPU_UP_PREPARE_FROZEN:
515 numa_setup_cpu(lcpu);
518 #ifdef CONFIG_HOTPLUG_CPU
520 case CPU_DEAD_FROZEN:
521 case CPU_UP_CANCELED:
522 case CPU_UP_CANCELED_FROZEN:
523 unmap_cpu_from_node(lcpu);
532 * Check and possibly modify a memory region to enforce the memory limit.
534 * Returns the size the region should have to enforce the memory limit.
535 * This will either be the original value of size, a truncated value,
536 * or zero. If the returned value of size is 0 the region should be
537 * discarded as it lies wholy above the memory limit.
539 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
543 * We use memblock_end_of_DRAM() in here instead of memory_limit because
544 * we've already adjusted it for the limit and it takes care of
545 * having memory holes below the limit. Also, in the case of
546 * iommu_is_off, memory_limit is not set but is implicitly enforced.
549 if (start + size <= memblock_end_of_DRAM())
552 if (start >= memblock_end_of_DRAM())
555 return memblock_end_of_DRAM() - start;
559 * Reads the counter for a given entry in
560 * linux,drconf-usable-memory property
562 static inline int __init read_usm_ranges(const u32 **usm)
565 * For each memblock in ibm,dynamic-memory a corresponding
566 * entry in linux,drconf-usable-memory property contains
567 * a counter followed by that many (base, size) duple.
568 * read the counter from linux,drconf-usable-memory
570 return read_n_cells(n_mem_size_cells, usm);
574 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
575 * node. This assumes n_mem_{addr,size}_cells have been set.
577 static void __init parse_drconf_memory(struct device_node *memory)
580 unsigned int n, rc, ranges, is_kexec_kdump = 0;
581 unsigned long memblock_size, base, size, sz;
583 struct assoc_arrays aa;
585 n = of_get_drconf_memory(memory, &dm);
589 memblock_size = of_get_memblock_size(memory);
593 rc = of_get_assoc_arrays(memory, &aa);
597 /* check if this is a kexec/kdump kernel */
598 usm = of_get_usable_memory(memory);
602 for (; n != 0; --n) {
603 struct of_drconf_cell drmem;
605 read_drconf_cell(&drmem, &dm);
607 /* skip this block if the reserved bit is set in flags (0x80)
608 or if the block is not assigned to this partition (0x8) */
609 if ((drmem.flags & DRCONF_MEM_RESERVED)
610 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
613 base = drmem.base_addr;
614 size = memblock_size;
617 if (is_kexec_kdump) {
618 ranges = read_usm_ranges(&usm);
619 if (!ranges) /* there are no (base, size) duple */
623 if (is_kexec_kdump) {
624 base = read_n_cells(n_mem_addr_cells, &usm);
625 size = read_n_cells(n_mem_size_cells, &usm);
627 nid = of_drconf_to_nid_single(&drmem, &aa);
628 fake_numa_create_new_node(
629 ((base + size) >> PAGE_SHIFT),
631 node_set_online(nid);
632 sz = numa_enforce_memory_limit(base, size);
634 add_active_range(nid, base >> PAGE_SHIFT,
636 + (sz >> PAGE_SHIFT));
641 static int __init parse_numa_properties(void)
643 struct device_node *cpu = NULL;
644 struct device_node *memory = NULL;
648 if (numa_enabled == 0) {
649 printk(KERN_WARNING "NUMA disabled by user\n");
653 min_common_depth = find_min_common_depth();
655 if (min_common_depth < 0)
656 return min_common_depth;
658 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
661 * Even though we connect cpus to numa domains later in SMP
662 * init, we need to know the node ids now. This is because
663 * each node to be onlined must have NODE_DATA etc backing it.
665 for_each_present_cpu(i) {
668 cpu = of_get_cpu_node(i, NULL);
670 nid = of_node_to_nid_single(cpu);
674 * Don't fall back to default_nid yet -- we will plug
675 * cpus into nodes once the memory scan has discovered
680 node_set_online(nid);
683 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
685 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
690 const unsigned int *memcell_buf;
693 memcell_buf = of_get_property(memory,
694 "linux,usable-memory", &len);
695 if (!memcell_buf || len <= 0)
696 memcell_buf = of_get_property(memory, "reg", &len);
697 if (!memcell_buf || len <= 0)
701 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
703 /* these are order-sensitive, and modify the buffer pointer */
704 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
705 size = read_n_cells(n_mem_size_cells, &memcell_buf);
708 * Assumption: either all memory nodes or none will
709 * have associativity properties. If none, then
710 * everything goes to default_nid.
712 nid = of_node_to_nid_single(memory);
716 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
717 node_set_online(nid);
719 if (!(size = numa_enforce_memory_limit(start, size))) {
726 add_active_range(nid, start >> PAGE_SHIFT,
727 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
734 * Now do the same thing for each MEMBLOCK listed in the ibm,dynamic-memory
735 * property in the ibm,dynamic-reconfiguration-memory node.
737 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
739 parse_drconf_memory(memory);
744 static void __init setup_nonnuma(void)
746 unsigned long top_of_ram = memblock_end_of_DRAM();
747 unsigned long total_ram = memblock_phys_mem_size();
748 unsigned long start_pfn, end_pfn;
749 unsigned int i, nid = 0;
751 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
752 top_of_ram, total_ram);
753 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
754 (top_of_ram - total_ram) >> 20);
756 for (i = 0; i < memblock.memory.cnt; ++i) {
757 start_pfn = memblock.memory.region[i].base >> PAGE_SHIFT;
758 end_pfn = start_pfn + memblock_size_pages(&memblock.memory, i);
760 fake_numa_create_new_node(end_pfn, &nid);
761 add_active_range(nid, start_pfn, end_pfn);
762 node_set_online(nid);
766 void __init dump_numa_cpu_topology(void)
769 unsigned int cpu, count;
771 if (min_common_depth == -1 || !numa_enabled)
774 for_each_online_node(node) {
775 printk(KERN_DEBUG "Node %d CPUs:", node);
779 * If we used a CPU iterator here we would miss printing
780 * the holes in the cpumap.
782 for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
783 if (cpumask_test_cpu(cpu,
784 node_to_cpumask_map[node])) {
790 printk("-%u", cpu - 1);
796 printk("-%u", nr_cpu_ids - 1);
801 static void __init dump_numa_memory_topology(void)
806 if (min_common_depth == -1 || !numa_enabled)
809 for_each_online_node(node) {
812 printk(KERN_DEBUG "Node %d Memory:", node);
816 for (i = 0; i < memblock_end_of_DRAM();
817 i += (1 << SECTION_SIZE_BITS)) {
818 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
836 * Allocate some memory, satisfying the memblock or bootmem allocator where
837 * required. nid is the preferred node and end is the physical address of
838 * the highest address in the node.
840 * Returns the virtual address of the memory.
842 static void __init *careful_zallocation(int nid, unsigned long size,
844 unsigned long end_pfn)
848 unsigned long ret_paddr;
850 ret_paddr = __memblock_alloc_base(size, align, end_pfn << PAGE_SHIFT);
852 /* retry over all memory */
854 ret_paddr = __memblock_alloc_base(size, align, memblock_end_of_DRAM());
857 panic("numa.c: cannot allocate %lu bytes for node %d",
860 ret = __va(ret_paddr);
863 * We initialize the nodes in numeric order: 0, 1, 2...
864 * and hand over control from the MEMBLOCK allocator to the
865 * bootmem allocator. If this function is called for
866 * node 5, then we know that all nodes <5 are using the
867 * bootmem allocator instead of the MEMBLOCK allocator.
869 * So, check the nid from which this allocation came
870 * and double check to see if we need to use bootmem
871 * instead of the MEMBLOCK. We don't free the MEMBLOCK memory
872 * since it would be useless.
874 new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
876 ret = __alloc_bootmem_node(NODE_DATA(new_nid),
879 dbg("alloc_bootmem %p %lx\n", ret, size);
882 memset(ret, 0, size);
886 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
887 .notifier_call = cpu_numa_callback,
888 .priority = 1 /* Must run before sched domains notifier. */
891 static void mark_reserved_regions_for_nid(int nid)
893 struct pglist_data *node = NODE_DATA(nid);
896 for (i = 0; i < memblock.reserved.cnt; i++) {
897 unsigned long physbase = memblock.reserved.region[i].base;
898 unsigned long size = memblock.reserved.region[i].size;
899 unsigned long start_pfn = physbase >> PAGE_SHIFT;
900 unsigned long end_pfn = PFN_UP(physbase + size);
901 struct node_active_region node_ar;
902 unsigned long node_end_pfn = node->node_start_pfn +
903 node->node_spanned_pages;
906 * Check to make sure that this memblock.reserved area is
907 * within the bounds of the node that we care about.
908 * Checking the nid of the start and end points is not
909 * sufficient because the reserved area could span the
912 if (end_pfn <= node->node_start_pfn ||
913 start_pfn >= node_end_pfn)
916 get_node_active_region(start_pfn, &node_ar);
917 while (start_pfn < end_pfn &&
918 node_ar.start_pfn < node_ar.end_pfn) {
919 unsigned long reserve_size = size;
921 * if reserved region extends past active region
922 * then trim size to active region
924 if (end_pfn > node_ar.end_pfn)
925 reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
928 * Only worry about *this* node, others may not
929 * yet have valid NODE_DATA().
931 if (node_ar.nid == nid) {
932 dbg("reserve_bootmem %lx %lx nid=%d\n",
933 physbase, reserve_size, node_ar.nid);
934 reserve_bootmem_node(NODE_DATA(node_ar.nid),
935 physbase, reserve_size,
939 * if reserved region is contained in the active region
942 if (end_pfn <= node_ar.end_pfn)
946 * reserved region extends past the active region
947 * get next active region that contains this
950 start_pfn = node_ar.end_pfn;
951 physbase = start_pfn << PAGE_SHIFT;
952 size = size - reserve_size;
953 get_node_active_region(start_pfn, &node_ar);
959 void __init do_init_bootmem(void)
964 max_low_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
965 max_pfn = max_low_pfn;
967 if (parse_numa_properties())
970 dump_numa_memory_topology();
972 for_each_online_node(nid) {
973 unsigned long start_pfn, end_pfn;
975 unsigned long bootmap_pages;
977 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
980 * Allocate the node structure node local if possible
982 * Be careful moving this around, as it relies on all
983 * previous nodes' bootmem to be initialized and have
984 * all reserved areas marked.
986 NODE_DATA(nid) = careful_zallocation(nid,
987 sizeof(struct pglist_data),
988 SMP_CACHE_BYTES, end_pfn);
990 dbg("node %d\n", nid);
991 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
993 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
994 NODE_DATA(nid)->node_start_pfn = start_pfn;
995 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
997 if (NODE_DATA(nid)->node_spanned_pages == 0)
1000 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
1001 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
1003 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
1004 bootmem_vaddr = careful_zallocation(nid,
1005 bootmap_pages << PAGE_SHIFT,
1006 PAGE_SIZE, end_pfn);
1008 dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
1010 init_bootmem_node(NODE_DATA(nid),
1011 __pa(bootmem_vaddr) >> PAGE_SHIFT,
1012 start_pfn, end_pfn);
1014 free_bootmem_with_active_regions(nid, end_pfn);
1016 * Be very careful about moving this around. Future
1017 * calls to careful_zallocation() depend on this getting
1020 mark_reserved_regions_for_nid(nid);
1021 sparse_memory_present_with_active_regions(nid);
1024 init_bootmem_done = 1;
1027 * Now bootmem is initialised we can create the node to cpumask
1028 * lookup tables and setup the cpu callback to populate them.
1030 setup_node_to_cpumask_map();
1032 register_cpu_notifier(&ppc64_numa_nb);
1033 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
1034 (void *)(unsigned long)boot_cpuid);
1037 void __init paging_init(void)
1039 unsigned long max_zone_pfns[MAX_NR_ZONES];
1040 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
1041 max_zone_pfns[ZONE_DMA] = memblock_end_of_DRAM() >> PAGE_SHIFT;
1042 free_area_init_nodes(max_zone_pfns);
1045 static int __init early_numa(char *p)
1050 if (strstr(p, "off"))
1053 if (strstr(p, "debug"))
1056 p = strstr(p, "fake=");
1058 cmdline = p + strlen("fake=");
1062 early_param("numa", early_numa);
1064 #ifdef CONFIG_MEMORY_HOTPLUG
1066 * Find the node associated with a hot added memory section for
1067 * memory represented in the device tree by the property
1068 * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
1070 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1071 unsigned long scn_addr)
1074 unsigned int drconf_cell_cnt, rc;
1075 unsigned long memblock_size;
1076 struct assoc_arrays aa;
1079 drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
1080 if (!drconf_cell_cnt)
1083 memblock_size = of_get_memblock_size(memory);
1087 rc = of_get_assoc_arrays(memory, &aa);
1091 for (; drconf_cell_cnt != 0; --drconf_cell_cnt) {
1092 struct of_drconf_cell drmem;
1094 read_drconf_cell(&drmem, &dm);
1096 /* skip this block if it is reserved or not assigned to
1098 if ((drmem.flags & DRCONF_MEM_RESERVED)
1099 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1102 if ((scn_addr < drmem.base_addr)
1103 || (scn_addr >= (drmem.base_addr + memblock_size)))
1106 nid = of_drconf_to_nid_single(&drmem, &aa);
1114 * Find the node associated with a hot added memory section for memory
1115 * represented in the device tree as a node (i.e. memory@XXXX) for
1118 int hot_add_node_scn_to_nid(unsigned long scn_addr)
1120 struct device_node *memory = NULL;
1123 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1124 unsigned long start, size;
1126 const unsigned int *memcell_buf;
1129 memcell_buf = of_get_property(memory, "reg", &len);
1130 if (!memcell_buf || len <= 0)
1133 /* ranges in cell */
1134 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1137 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1138 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1140 if ((scn_addr < start) || (scn_addr >= (start + size)))
1143 nid = of_node_to_nid_single(memory);
1147 of_node_put(memory);
1156 * Find the node associated with a hot added memory section. Section
1157 * corresponds to a SPARSEMEM section, not an MEMBLOCK. It is assumed that
1158 * sections are fully contained within a single MEMBLOCK.
1160 int hot_add_scn_to_nid(unsigned long scn_addr)
1162 struct device_node *memory = NULL;
1165 if (!numa_enabled || (min_common_depth < 0))
1166 return first_online_node;
1168 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1170 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1171 of_node_put(memory);
1173 nid = hot_add_node_scn_to_nid(scn_addr);
1176 if (nid < 0 || !node_online(nid))
1177 nid = first_online_node;
1179 if (NODE_DATA(nid)->node_spanned_pages)
1182 for_each_online_node(nid) {
1183 if (NODE_DATA(nid)->node_spanned_pages) {
1193 #endif /* CONFIG_MEMORY_HOTPLUG */