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 /* must hold reference to node during call */
162 static const int *of_get_associativity(struct device_node *dev)
164 return of_get_property(dev, "ibm,associativity", NULL);
168 * Returns the property linux,drconf-usable-memory if
169 * it exists (the property exists only in kexec/kdump kernels,
170 * added by kexec-tools)
172 static const u32 *of_get_usable_memory(struct device_node *memory)
176 prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
177 if (!prop || len < sizeof(unsigned int))
182 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
185 static int of_node_to_nid_single(struct device_node *device)
188 const unsigned int *tmp;
190 if (min_common_depth == -1)
193 tmp = of_get_associativity(device);
197 if (tmp[0] >= min_common_depth)
198 nid = tmp[min_common_depth];
200 /* POWER4 LPAR uses 0xffff as invalid node */
201 if (nid == 0xffff || nid >= MAX_NUMNODES)
207 /* Walk the device tree upwards, looking for an associativity id */
208 int of_node_to_nid(struct device_node *device)
210 struct device_node *tmp;
215 nid = of_node_to_nid_single(device);
220 device = of_get_parent(tmp);
227 EXPORT_SYMBOL_GPL(of_node_to_nid);
230 * In theory, the "ibm,associativity" property may contain multiple
231 * associativity lists because a resource may be multiply connected
232 * into the machine. This resource then has different associativity
233 * characteristics relative to its multiple connections. We ignore
234 * this for now. We also assume that all cpu and memory sets have
235 * their distances represented at a common level. This won't be
236 * true for hierarchical NUMA.
238 * In any case the ibm,associativity-reference-points should give
239 * the correct depth for a normal NUMA system.
241 * - Dave Hansen <haveblue@us.ibm.com>
243 static int __init find_min_common_depth(void)
246 const unsigned int *ref_points;
247 struct device_node *rtas_root;
250 rtas_root = of_find_node_by_path("/rtas");
256 * this property is 2 32-bit integers, each representing a level of
257 * depth in the associativity nodes. The first is for an SMP
258 * configuration (should be all 0's) and the second is for a normal
259 * NUMA configuration.
261 ref_points = of_get_property(rtas_root,
262 "ibm,associativity-reference-points", &len);
264 if ((len >= 2 * sizeof(unsigned int)) && ref_points) {
265 depth = ref_points[1];
267 dbg("NUMA: ibm,associativity-reference-points not found.\n");
270 of_node_put(rtas_root);
275 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
277 struct device_node *memory = NULL;
279 memory = of_find_node_by_type(memory, "memory");
281 panic("numa.c: No memory nodes found!");
283 *n_addr_cells = of_n_addr_cells(memory);
284 *n_size_cells = of_n_size_cells(memory);
288 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
290 unsigned long result = 0;
293 result = (result << 32) | **buf;
299 struct of_drconf_cell {
307 #define DRCONF_MEM_ASSIGNED 0x00000008
308 #define DRCONF_MEM_AI_INVALID 0x00000040
309 #define DRCONF_MEM_RESERVED 0x00000080
312 * Read the next lmb list entry from the ibm,dynamic-memory property
313 * and return the information in the provided of_drconf_cell structure.
315 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
319 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
322 drmem->drc_index = cp[0];
323 drmem->reserved = cp[1];
324 drmem->aa_index = cp[2];
325 drmem->flags = cp[3];
331 * Retreive and validate the ibm,dynamic-memory property of the device tree.
333 * The layout of the ibm,dynamic-memory property is a number N of lmb
334 * list entries followed by N lmb list entries. Each lmb list entry
335 * contains information as layed out in the of_drconf_cell struct above.
337 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
342 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
343 if (!prop || len < sizeof(unsigned int))
348 /* Now that we know the number of entries, revalidate the size
349 * of the property read in to ensure we have everything
351 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
359 * Retreive and validate the ibm,lmb-size property for drconf memory
360 * from the device tree.
362 static u64 of_get_lmb_size(struct device_node *memory)
367 prop = of_get_property(memory, "ibm,lmb-size", &len);
368 if (!prop || len < sizeof(unsigned int))
371 return read_n_cells(n_mem_size_cells, &prop);
374 struct assoc_arrays {
381 * Retreive and validate the list of associativity arrays for drconf
382 * memory from the ibm,associativity-lookup-arrays property of the
385 * The layout of the ibm,associativity-lookup-arrays property is a number N
386 * indicating the number of associativity arrays, followed by a number M
387 * indicating the size of each associativity array, followed by a list
388 * of N associativity arrays.
390 static int of_get_assoc_arrays(struct device_node *memory,
391 struct assoc_arrays *aa)
396 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
397 if (!prop || len < 2 * sizeof(unsigned int))
400 aa->n_arrays = *prop++;
401 aa->array_sz = *prop++;
403 /* Now that we know the number of arrrays and size of each array,
404 * revalidate the size of the property read in.
406 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
414 * This is like of_node_to_nid_single() for memory represented in the
415 * ibm,dynamic-reconfiguration-memory node.
417 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
418 struct assoc_arrays *aa)
421 int nid = default_nid;
424 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
425 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
426 drmem->aa_index < aa->n_arrays) {
427 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
428 nid = aa->arrays[index];
430 if (nid == 0xffff || nid >= MAX_NUMNODES)
438 * Figure out to which domain a cpu belongs and stick it there.
439 * Return the id of the domain used.
441 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
444 struct device_node *cpu = of_get_cpu_node(lcpu, NULL);
451 nid = of_node_to_nid_single(cpu);
453 if (nid < 0 || !node_online(nid))
454 nid = any_online_node(NODE_MASK_ALL);
456 map_cpu_to_node(lcpu, nid);
463 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
464 unsigned long action,
467 unsigned long lcpu = (unsigned long)hcpu;
468 int ret = NOTIFY_DONE;
472 case CPU_UP_PREPARE_FROZEN:
473 numa_setup_cpu(lcpu);
476 #ifdef CONFIG_HOTPLUG_CPU
478 case CPU_DEAD_FROZEN:
479 case CPU_UP_CANCELED:
480 case CPU_UP_CANCELED_FROZEN:
481 unmap_cpu_from_node(lcpu);
490 * Check and possibly modify a memory region to enforce the memory limit.
492 * Returns the size the region should have to enforce the memory limit.
493 * This will either be the original value of size, a truncated value,
494 * or zero. If the returned value of size is 0 the region should be
495 * discarded as it lies wholy above the memory limit.
497 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
501 * We use lmb_end_of_DRAM() in here instead of memory_limit because
502 * we've already adjusted it for the limit and it takes care of
503 * having memory holes below the limit. Also, in the case of
504 * iommu_is_off, memory_limit is not set but is implicitly enforced.
507 if (start + size <= lmb_end_of_DRAM())
510 if (start >= lmb_end_of_DRAM())
513 return lmb_end_of_DRAM() - start;
517 * Reads the counter for a given entry in
518 * linux,drconf-usable-memory property
520 static inline int __init read_usm_ranges(const u32 **usm)
523 * For each lmb in ibm,dynamic-memory a corresponding
524 * entry in linux,drconf-usable-memory property contains
525 * a counter followed by that many (base, size) duple.
526 * read the counter from linux,drconf-usable-memory
528 return read_n_cells(n_mem_size_cells, usm);
532 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
533 * node. This assumes n_mem_{addr,size}_cells have been set.
535 static void __init parse_drconf_memory(struct device_node *memory)
538 unsigned int n, rc, ranges, is_kexec_kdump = 0;
539 unsigned long lmb_size, base, size, sz;
541 struct assoc_arrays aa;
543 n = of_get_drconf_memory(memory, &dm);
547 lmb_size = of_get_lmb_size(memory);
551 rc = of_get_assoc_arrays(memory, &aa);
555 /* check if this is a kexec/kdump kernel */
556 usm = of_get_usable_memory(memory);
560 for (; n != 0; --n) {
561 struct of_drconf_cell drmem;
563 read_drconf_cell(&drmem, &dm);
565 /* skip this block if the reserved bit is set in flags (0x80)
566 or if the block is not assigned to this partition (0x8) */
567 if ((drmem.flags & DRCONF_MEM_RESERVED)
568 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
571 base = drmem.base_addr;
575 if (is_kexec_kdump) {
576 ranges = read_usm_ranges(&usm);
577 if (!ranges) /* there are no (base, size) duple */
581 if (is_kexec_kdump) {
582 base = read_n_cells(n_mem_addr_cells, &usm);
583 size = read_n_cells(n_mem_size_cells, &usm);
585 nid = of_drconf_to_nid_single(&drmem, &aa);
586 fake_numa_create_new_node(
587 ((base + size) >> PAGE_SHIFT),
589 node_set_online(nid);
590 sz = numa_enforce_memory_limit(base, size);
592 add_active_range(nid, base >> PAGE_SHIFT,
594 + (sz >> PAGE_SHIFT));
599 static int __init parse_numa_properties(void)
601 struct device_node *cpu = NULL;
602 struct device_node *memory = NULL;
606 if (numa_enabled == 0) {
607 printk(KERN_WARNING "NUMA disabled by user\n");
611 min_common_depth = find_min_common_depth();
613 if (min_common_depth < 0)
614 return min_common_depth;
616 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
619 * Even though we connect cpus to numa domains later in SMP
620 * init, we need to know the node ids now. This is because
621 * each node to be onlined must have NODE_DATA etc backing it.
623 for_each_present_cpu(i) {
626 cpu = of_get_cpu_node(i, NULL);
628 nid = of_node_to_nid_single(cpu);
632 * Don't fall back to default_nid yet -- we will plug
633 * cpus into nodes once the memory scan has discovered
638 node_set_online(nid);
641 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
643 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
648 const unsigned int *memcell_buf;
651 memcell_buf = of_get_property(memory,
652 "linux,usable-memory", &len);
653 if (!memcell_buf || len <= 0)
654 memcell_buf = of_get_property(memory, "reg", &len);
655 if (!memcell_buf || len <= 0)
659 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
661 /* these are order-sensitive, and modify the buffer pointer */
662 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
663 size = read_n_cells(n_mem_size_cells, &memcell_buf);
666 * Assumption: either all memory nodes or none will
667 * have associativity properties. If none, then
668 * everything goes to default_nid.
670 nid = of_node_to_nid_single(memory);
674 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
675 node_set_online(nid);
677 if (!(size = numa_enforce_memory_limit(start, size))) {
684 add_active_range(nid, start >> PAGE_SHIFT,
685 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
692 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
693 * property in the ibm,dynamic-reconfiguration-memory node.
695 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
697 parse_drconf_memory(memory);
702 static void __init setup_nonnuma(void)
704 unsigned long top_of_ram = lmb_end_of_DRAM();
705 unsigned long total_ram = lmb_phys_mem_size();
706 unsigned long start_pfn, end_pfn;
707 unsigned int i, nid = 0;
709 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
710 top_of_ram, total_ram);
711 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
712 (top_of_ram - total_ram) >> 20);
714 for (i = 0; i < lmb.memory.cnt; ++i) {
715 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
716 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
718 fake_numa_create_new_node(end_pfn, &nid);
719 add_active_range(nid, start_pfn, end_pfn);
720 node_set_online(nid);
724 void __init dump_numa_cpu_topology(void)
727 unsigned int cpu, count;
729 if (min_common_depth == -1 || !numa_enabled)
732 for_each_online_node(node) {
733 printk(KERN_DEBUG "Node %d CPUs:", node);
737 * If we used a CPU iterator here we would miss printing
738 * the holes in the cpumap.
740 for (cpu = 0; cpu < NR_CPUS; cpu++) {
741 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
747 printk("-%u", cpu - 1);
753 printk("-%u", NR_CPUS - 1);
758 static void __init dump_numa_memory_topology(void)
763 if (min_common_depth == -1 || !numa_enabled)
766 for_each_online_node(node) {
769 printk(KERN_DEBUG "Node %d Memory:", node);
773 for (i = 0; i < lmb_end_of_DRAM();
774 i += (1 << SECTION_SIZE_BITS)) {
775 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
793 * Allocate some memory, satisfying the lmb or bootmem allocator where
794 * required. nid is the preferred node and end is the physical address of
795 * the highest address in the node.
797 * Returns the virtual address of the memory.
799 static void __init *careful_zallocation(int nid, unsigned long size,
801 unsigned long end_pfn)
805 unsigned long ret_paddr;
807 ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
809 /* retry over all memory */
811 ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
814 panic("numa.c: cannot allocate %lu bytes for node %d",
817 ret = __va(ret_paddr);
820 * We initialize the nodes in numeric order: 0, 1, 2...
821 * and hand over control from the LMB allocator to the
822 * bootmem allocator. If this function is called for
823 * node 5, then we know that all nodes <5 are using the
824 * bootmem allocator instead of the LMB allocator.
826 * So, check the nid from which this allocation came
827 * and double check to see if we need to use bootmem
828 * instead of the LMB. We don't free the LMB memory
829 * since it would be useless.
831 new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
833 ret = __alloc_bootmem_node(NODE_DATA(new_nid),
836 dbg("alloc_bootmem %p %lx\n", ret, size);
839 memset(ret, 0, size);
843 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
844 .notifier_call = cpu_numa_callback,
845 .priority = 1 /* Must run before sched domains notifier. */
848 static void mark_reserved_regions_for_nid(int nid)
850 struct pglist_data *node = NODE_DATA(nid);
853 for (i = 0; i < lmb.reserved.cnt; i++) {
854 unsigned long physbase = lmb.reserved.region[i].base;
855 unsigned long size = lmb.reserved.region[i].size;
856 unsigned long start_pfn = physbase >> PAGE_SHIFT;
857 unsigned long end_pfn = PFN_UP(physbase + size);
858 struct node_active_region node_ar;
859 unsigned long node_end_pfn = node->node_start_pfn +
860 node->node_spanned_pages;
863 * Check to make sure that this lmb.reserved area is
864 * within the bounds of the node that we care about.
865 * Checking the nid of the start and end points is not
866 * sufficient because the reserved area could span the
869 if (end_pfn <= node->node_start_pfn ||
870 start_pfn >= node_end_pfn)
873 get_node_active_region(start_pfn, &node_ar);
874 while (start_pfn < end_pfn &&
875 node_ar.start_pfn < node_ar.end_pfn) {
876 unsigned long reserve_size = size;
878 * if reserved region extends past active region
879 * then trim size to active region
881 if (end_pfn > node_ar.end_pfn)
882 reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
885 * Only worry about *this* node, others may not
886 * yet have valid NODE_DATA().
888 if (node_ar.nid == nid) {
889 dbg("reserve_bootmem %lx %lx nid=%d\n",
890 physbase, reserve_size, node_ar.nid);
891 reserve_bootmem_node(NODE_DATA(node_ar.nid),
892 physbase, reserve_size,
896 * if reserved region is contained in the active region
899 if (end_pfn <= node_ar.end_pfn)
903 * reserved region extends past the active region
904 * get next active region that contains this
907 start_pfn = node_ar.end_pfn;
908 physbase = start_pfn << PAGE_SHIFT;
909 size = size - reserve_size;
910 get_node_active_region(start_pfn, &node_ar);
916 void __init do_init_bootmem(void)
921 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
922 max_pfn = max_low_pfn;
924 if (parse_numa_properties())
927 dump_numa_memory_topology();
929 register_cpu_notifier(&ppc64_numa_nb);
930 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
931 (void *)(unsigned long)boot_cpuid);
933 for_each_online_node(nid) {
934 unsigned long start_pfn, end_pfn;
936 unsigned long bootmap_pages;
938 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
941 * Allocate the node structure node local if possible
943 * Be careful moving this around, as it relies on all
944 * previous nodes' bootmem to be initialized and have
945 * all reserved areas marked.
947 NODE_DATA(nid) = careful_zallocation(nid,
948 sizeof(struct pglist_data),
949 SMP_CACHE_BYTES, end_pfn);
951 dbg("node %d\n", nid);
952 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
954 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
955 NODE_DATA(nid)->node_start_pfn = start_pfn;
956 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
958 if (NODE_DATA(nid)->node_spanned_pages == 0)
961 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
962 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
964 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
965 bootmem_vaddr = careful_zallocation(nid,
966 bootmap_pages << PAGE_SHIFT,
969 dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
971 init_bootmem_node(NODE_DATA(nid),
972 __pa(bootmem_vaddr) >> PAGE_SHIFT,
975 free_bootmem_with_active_regions(nid, end_pfn);
977 * Be very careful about moving this around. Future
978 * calls to careful_zallocation() depend on this getting
981 mark_reserved_regions_for_nid(nid);
982 sparse_memory_present_with_active_regions(nid);
986 void __init paging_init(void)
988 unsigned long max_zone_pfns[MAX_NR_ZONES];
989 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
990 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
991 free_area_init_nodes(max_zone_pfns);
994 static int __init early_numa(char *p)
999 if (strstr(p, "off"))
1002 if (strstr(p, "debug"))
1005 p = strstr(p, "fake=");
1007 cmdline = p + strlen("fake=");
1011 early_param("numa", early_numa);
1013 #ifdef CONFIG_MEMORY_HOTPLUG
1015 * Find the node associated with a hot added memory section for
1016 * memory represented in the device tree by the property
1017 * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
1019 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1020 unsigned long scn_addr)
1023 unsigned int drconf_cell_cnt, rc;
1024 unsigned long lmb_size;
1025 struct assoc_arrays aa;
1028 drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
1029 if (!drconf_cell_cnt)
1032 lmb_size = of_get_lmb_size(memory);
1036 rc = of_get_assoc_arrays(memory, &aa);
1040 for (; drconf_cell_cnt != 0; --drconf_cell_cnt) {
1041 struct of_drconf_cell drmem;
1043 read_drconf_cell(&drmem, &dm);
1045 /* skip this block if it is reserved or not assigned to
1047 if ((drmem.flags & DRCONF_MEM_RESERVED)
1048 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1051 if ((scn_addr < drmem.base_addr)
1052 || (scn_addr >= (drmem.base_addr + lmb_size)))
1055 nid = of_drconf_to_nid_single(&drmem, &aa);
1063 * Find the node associated with a hot added memory section for memory
1064 * represented in the device tree as a node (i.e. memory@XXXX) for
1067 int hot_add_node_scn_to_nid(unsigned long scn_addr)
1069 struct device_node *memory = NULL;
1072 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1073 unsigned long start, size;
1075 const unsigned int *memcell_buf;
1078 memcell_buf = of_get_property(memory, "reg", &len);
1079 if (!memcell_buf || len <= 0)
1082 /* ranges in cell */
1083 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1086 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1087 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1089 if ((scn_addr < start) || (scn_addr >= (start + size)))
1092 nid = of_node_to_nid_single(memory);
1096 of_node_put(memory);
1105 * Find the node associated with a hot added memory section. Section
1106 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1107 * sections are fully contained within a single LMB.
1109 int hot_add_scn_to_nid(unsigned long scn_addr)
1111 struct device_node *memory = NULL;
1114 if (!numa_enabled || (min_common_depth < 0))
1115 return any_online_node(NODE_MASK_ALL);
1117 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1119 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1120 of_node_put(memory);
1122 nid = hot_add_node_scn_to_nid(scn_addr);
1125 if (nid < 0 || !node_online(nid))
1126 nid = any_online_node(NODE_MASK_ALL);
1128 if (NODE_DATA(nid)->node_spanned_pages)
1131 for_each_online_node(nid) {
1132 if (NODE_DATA(nid)->node_spanned_pages) {
1142 #endif /* CONFIG_MEMORY_HOTPLUG */