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;
249 struct device_node *options;
251 rtas_root = of_find_node_by_path("/rtas");
257 * this property is 2 32-bit integers, each representing a level of
258 * depth in the associativity nodes. The first is for an SMP
259 * configuration (should be all 0's) and the second is for a normal
260 * NUMA configuration.
263 ref_points = of_get_property(rtas_root,
264 "ibm,associativity-reference-points", &len);
267 * For type 1 affinity information we want the first field
269 options = of_find_node_by_path("/options");
272 str = of_get_property(options, "ibm,associativity-form", NULL);
273 if (str && !strcmp(str, "1"))
277 if ((len >= 2 * sizeof(unsigned int)) && ref_points) {
278 depth = ref_points[index];
280 dbg("NUMA: ibm,associativity-reference-points not found.\n");
283 of_node_put(rtas_root);
288 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
290 struct device_node *memory = NULL;
292 memory = of_find_node_by_type(memory, "memory");
294 panic("numa.c: No memory nodes found!");
296 *n_addr_cells = of_n_addr_cells(memory);
297 *n_size_cells = of_n_size_cells(memory);
301 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
303 unsigned long result = 0;
306 result = (result << 32) | **buf;
312 struct of_drconf_cell {
320 #define DRCONF_MEM_ASSIGNED 0x00000008
321 #define DRCONF_MEM_AI_INVALID 0x00000040
322 #define DRCONF_MEM_RESERVED 0x00000080
325 * Read the next lmb list entry from the ibm,dynamic-memory property
326 * and return the information in the provided of_drconf_cell structure.
328 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
332 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
335 drmem->drc_index = cp[0];
336 drmem->reserved = cp[1];
337 drmem->aa_index = cp[2];
338 drmem->flags = cp[3];
344 * Retreive and validate the ibm,dynamic-memory property of the device tree.
346 * The layout of the ibm,dynamic-memory property is a number N of lmb
347 * list entries followed by N lmb list entries. Each lmb list entry
348 * contains information as layed out in the of_drconf_cell struct above.
350 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
355 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
356 if (!prop || len < sizeof(unsigned int))
361 /* Now that we know the number of entries, revalidate the size
362 * of the property read in to ensure we have everything
364 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
372 * Retreive and validate the ibm,lmb-size property for drconf memory
373 * from the device tree.
375 static u64 of_get_lmb_size(struct device_node *memory)
380 prop = of_get_property(memory, "ibm,lmb-size", &len);
381 if (!prop || len < sizeof(unsigned int))
384 return read_n_cells(n_mem_size_cells, &prop);
387 struct assoc_arrays {
394 * Retreive and validate the list of associativity arrays for drconf
395 * memory from the ibm,associativity-lookup-arrays property of the
398 * The layout of the ibm,associativity-lookup-arrays property is a number N
399 * indicating the number of associativity arrays, followed by a number M
400 * indicating the size of each associativity array, followed by a list
401 * of N associativity arrays.
403 static int of_get_assoc_arrays(struct device_node *memory,
404 struct assoc_arrays *aa)
409 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
410 if (!prop || len < 2 * sizeof(unsigned int))
413 aa->n_arrays = *prop++;
414 aa->array_sz = *prop++;
416 /* Now that we know the number of arrrays and size of each array,
417 * revalidate the size of the property read in.
419 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
427 * This is like of_node_to_nid_single() for memory represented in the
428 * ibm,dynamic-reconfiguration-memory node.
430 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
431 struct assoc_arrays *aa)
434 int nid = default_nid;
437 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
438 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
439 drmem->aa_index < aa->n_arrays) {
440 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
441 nid = aa->arrays[index];
443 if (nid == 0xffff || nid >= MAX_NUMNODES)
451 * Figure out to which domain a cpu belongs and stick it there.
452 * Return the id of the domain used.
454 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
457 struct device_node *cpu = of_get_cpu_node(lcpu, NULL);
464 nid = of_node_to_nid_single(cpu);
466 if (nid < 0 || !node_online(nid))
467 nid = first_online_node;
469 map_cpu_to_node(lcpu, nid);
476 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
477 unsigned long action,
480 unsigned long lcpu = (unsigned long)hcpu;
481 int ret = NOTIFY_DONE;
485 case CPU_UP_PREPARE_FROZEN:
486 numa_setup_cpu(lcpu);
489 #ifdef CONFIG_HOTPLUG_CPU
491 case CPU_DEAD_FROZEN:
492 case CPU_UP_CANCELED:
493 case CPU_UP_CANCELED_FROZEN:
494 unmap_cpu_from_node(lcpu);
503 * Check and possibly modify a memory region to enforce the memory limit.
505 * Returns the size the region should have to enforce the memory limit.
506 * This will either be the original value of size, a truncated value,
507 * or zero. If the returned value of size is 0 the region should be
508 * discarded as it lies wholy above the memory limit.
510 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
514 * We use lmb_end_of_DRAM() in here instead of memory_limit because
515 * we've already adjusted it for the limit and it takes care of
516 * having memory holes below the limit. Also, in the case of
517 * iommu_is_off, memory_limit is not set but is implicitly enforced.
520 if (start + size <= lmb_end_of_DRAM())
523 if (start >= lmb_end_of_DRAM())
526 return lmb_end_of_DRAM() - start;
530 * Reads the counter for a given entry in
531 * linux,drconf-usable-memory property
533 static inline int __init read_usm_ranges(const u32 **usm)
536 * For each lmb in ibm,dynamic-memory a corresponding
537 * entry in linux,drconf-usable-memory property contains
538 * a counter followed by that many (base, size) duple.
539 * read the counter from linux,drconf-usable-memory
541 return read_n_cells(n_mem_size_cells, usm);
545 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
546 * node. This assumes n_mem_{addr,size}_cells have been set.
548 static void __init parse_drconf_memory(struct device_node *memory)
551 unsigned int n, rc, ranges, is_kexec_kdump = 0;
552 unsigned long lmb_size, base, size, sz;
554 struct assoc_arrays aa;
556 n = of_get_drconf_memory(memory, &dm);
560 lmb_size = of_get_lmb_size(memory);
564 rc = of_get_assoc_arrays(memory, &aa);
568 /* check if this is a kexec/kdump kernel */
569 usm = of_get_usable_memory(memory);
573 for (; n != 0; --n) {
574 struct of_drconf_cell drmem;
576 read_drconf_cell(&drmem, &dm);
578 /* skip this block if the reserved bit is set in flags (0x80)
579 or if the block is not assigned to this partition (0x8) */
580 if ((drmem.flags & DRCONF_MEM_RESERVED)
581 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
584 base = drmem.base_addr;
588 if (is_kexec_kdump) {
589 ranges = read_usm_ranges(&usm);
590 if (!ranges) /* there are no (base, size) duple */
594 if (is_kexec_kdump) {
595 base = read_n_cells(n_mem_addr_cells, &usm);
596 size = read_n_cells(n_mem_size_cells, &usm);
598 nid = of_drconf_to_nid_single(&drmem, &aa);
599 fake_numa_create_new_node(
600 ((base + size) >> PAGE_SHIFT),
602 node_set_online(nid);
603 sz = numa_enforce_memory_limit(base, size);
605 add_active_range(nid, base >> PAGE_SHIFT,
607 + (sz >> PAGE_SHIFT));
612 static int __init parse_numa_properties(void)
614 struct device_node *cpu = NULL;
615 struct device_node *memory = NULL;
619 if (numa_enabled == 0) {
620 printk(KERN_WARNING "NUMA disabled by user\n");
624 min_common_depth = find_min_common_depth();
626 if (min_common_depth < 0)
627 return min_common_depth;
629 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
632 * Even though we connect cpus to numa domains later in SMP
633 * init, we need to know the node ids now. This is because
634 * each node to be onlined must have NODE_DATA etc backing it.
636 for_each_present_cpu(i) {
639 cpu = of_get_cpu_node(i, NULL);
641 nid = of_node_to_nid_single(cpu);
645 * Don't fall back to default_nid yet -- we will plug
646 * cpus into nodes once the memory scan has discovered
651 node_set_online(nid);
654 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
656 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
661 const unsigned int *memcell_buf;
664 memcell_buf = of_get_property(memory,
665 "linux,usable-memory", &len);
666 if (!memcell_buf || len <= 0)
667 memcell_buf = of_get_property(memory, "reg", &len);
668 if (!memcell_buf || len <= 0)
672 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
674 /* these are order-sensitive, and modify the buffer pointer */
675 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
676 size = read_n_cells(n_mem_size_cells, &memcell_buf);
679 * Assumption: either all memory nodes or none will
680 * have associativity properties. If none, then
681 * everything goes to default_nid.
683 nid = of_node_to_nid_single(memory);
687 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
688 node_set_online(nid);
690 if (!(size = numa_enforce_memory_limit(start, size))) {
697 add_active_range(nid, start >> PAGE_SHIFT,
698 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
705 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
706 * property in the ibm,dynamic-reconfiguration-memory node.
708 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
710 parse_drconf_memory(memory);
715 static void __init setup_nonnuma(void)
717 unsigned long top_of_ram = lmb_end_of_DRAM();
718 unsigned long total_ram = lmb_phys_mem_size();
719 unsigned long start_pfn, end_pfn;
720 unsigned int i, nid = 0;
722 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
723 top_of_ram, total_ram);
724 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
725 (top_of_ram - total_ram) >> 20);
727 for (i = 0; i < lmb.memory.cnt; ++i) {
728 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
729 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
731 fake_numa_create_new_node(end_pfn, &nid);
732 add_active_range(nid, start_pfn, end_pfn);
733 node_set_online(nid);
737 void __init dump_numa_cpu_topology(void)
740 unsigned int cpu, count;
742 if (min_common_depth == -1 || !numa_enabled)
745 for_each_online_node(node) {
746 printk(KERN_DEBUG "Node %d CPUs:", node);
750 * If we used a CPU iterator here we would miss printing
751 * the holes in the cpumap.
753 for (cpu = 0; cpu < NR_CPUS; cpu++) {
754 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
760 printk("-%u", cpu - 1);
766 printk("-%u", NR_CPUS - 1);
771 static void __init dump_numa_memory_topology(void)
776 if (min_common_depth == -1 || !numa_enabled)
779 for_each_online_node(node) {
782 printk(KERN_DEBUG "Node %d Memory:", node);
786 for (i = 0; i < lmb_end_of_DRAM();
787 i += (1 << SECTION_SIZE_BITS)) {
788 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
806 * Allocate some memory, satisfying the lmb or bootmem allocator where
807 * required. nid is the preferred node and end is the physical address of
808 * the highest address in the node.
810 * Returns the virtual address of the memory.
812 static void __init *careful_zallocation(int nid, unsigned long size,
814 unsigned long end_pfn)
818 unsigned long ret_paddr;
820 ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
822 /* retry over all memory */
824 ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
827 panic("numa.c: cannot allocate %lu bytes for node %d",
830 ret = __va(ret_paddr);
833 * We initialize the nodes in numeric order: 0, 1, 2...
834 * and hand over control from the LMB allocator to the
835 * bootmem allocator. If this function is called for
836 * node 5, then we know that all nodes <5 are using the
837 * bootmem allocator instead of the LMB allocator.
839 * So, check the nid from which this allocation came
840 * and double check to see if we need to use bootmem
841 * instead of the LMB. We don't free the LMB memory
842 * since it would be useless.
844 new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
846 ret = __alloc_bootmem_node(NODE_DATA(new_nid),
849 dbg("alloc_bootmem %p %lx\n", ret, size);
852 memset(ret, 0, size);
856 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
857 .notifier_call = cpu_numa_callback,
858 .priority = 1 /* Must run before sched domains notifier. */
861 static void mark_reserved_regions_for_nid(int nid)
863 struct pglist_data *node = NODE_DATA(nid);
866 for (i = 0; i < lmb.reserved.cnt; i++) {
867 unsigned long physbase = lmb.reserved.region[i].base;
868 unsigned long size = lmb.reserved.region[i].size;
869 unsigned long start_pfn = physbase >> PAGE_SHIFT;
870 unsigned long end_pfn = PFN_UP(physbase + size);
871 struct node_active_region node_ar;
872 unsigned long node_end_pfn = node->node_start_pfn +
873 node->node_spanned_pages;
876 * Check to make sure that this lmb.reserved area is
877 * within the bounds of the node that we care about.
878 * Checking the nid of the start and end points is not
879 * sufficient because the reserved area could span the
882 if (end_pfn <= node->node_start_pfn ||
883 start_pfn >= node_end_pfn)
886 get_node_active_region(start_pfn, &node_ar);
887 while (start_pfn < end_pfn &&
888 node_ar.start_pfn < node_ar.end_pfn) {
889 unsigned long reserve_size = size;
891 * if reserved region extends past active region
892 * then trim size to active region
894 if (end_pfn > node_ar.end_pfn)
895 reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
898 * Only worry about *this* node, others may not
899 * yet have valid NODE_DATA().
901 if (node_ar.nid == nid) {
902 dbg("reserve_bootmem %lx %lx nid=%d\n",
903 physbase, reserve_size, node_ar.nid);
904 reserve_bootmem_node(NODE_DATA(node_ar.nid),
905 physbase, reserve_size,
909 * if reserved region is contained in the active region
912 if (end_pfn <= node_ar.end_pfn)
916 * reserved region extends past the active region
917 * get next active region that contains this
920 start_pfn = node_ar.end_pfn;
921 physbase = start_pfn << PAGE_SHIFT;
922 size = size - reserve_size;
923 get_node_active_region(start_pfn, &node_ar);
929 void __init do_init_bootmem(void)
934 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
935 max_pfn = max_low_pfn;
937 if (parse_numa_properties())
940 dump_numa_memory_topology();
942 register_cpu_notifier(&ppc64_numa_nb);
943 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
944 (void *)(unsigned long)boot_cpuid);
946 for_each_online_node(nid) {
947 unsigned long start_pfn, end_pfn;
949 unsigned long bootmap_pages;
951 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
954 * Allocate the node structure node local if possible
956 * Be careful moving this around, as it relies on all
957 * previous nodes' bootmem to be initialized and have
958 * all reserved areas marked.
960 NODE_DATA(nid) = careful_zallocation(nid,
961 sizeof(struct pglist_data),
962 SMP_CACHE_BYTES, end_pfn);
964 dbg("node %d\n", nid);
965 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
967 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
968 NODE_DATA(nid)->node_start_pfn = start_pfn;
969 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
971 if (NODE_DATA(nid)->node_spanned_pages == 0)
974 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
975 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
977 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
978 bootmem_vaddr = careful_zallocation(nid,
979 bootmap_pages << PAGE_SHIFT,
982 dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
984 init_bootmem_node(NODE_DATA(nid),
985 __pa(bootmem_vaddr) >> PAGE_SHIFT,
988 free_bootmem_with_active_regions(nid, end_pfn);
990 * Be very careful about moving this around. Future
991 * calls to careful_zallocation() depend on this getting
994 mark_reserved_regions_for_nid(nid);
995 sparse_memory_present_with_active_regions(nid);
998 init_bootmem_done = 1;
1001 void __init paging_init(void)
1003 unsigned long max_zone_pfns[MAX_NR_ZONES];
1004 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
1005 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
1006 free_area_init_nodes(max_zone_pfns);
1009 static int __init early_numa(char *p)
1014 if (strstr(p, "off"))
1017 if (strstr(p, "debug"))
1020 p = strstr(p, "fake=");
1022 cmdline = p + strlen("fake=");
1026 early_param("numa", early_numa);
1028 #ifdef CONFIG_MEMORY_HOTPLUG
1030 * Find the node associated with a hot added memory section for
1031 * memory represented in the device tree by the property
1032 * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
1034 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1035 unsigned long scn_addr)
1038 unsigned int drconf_cell_cnt, rc;
1039 unsigned long lmb_size;
1040 struct assoc_arrays aa;
1043 drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
1044 if (!drconf_cell_cnt)
1047 lmb_size = of_get_lmb_size(memory);
1051 rc = of_get_assoc_arrays(memory, &aa);
1055 for (; drconf_cell_cnt != 0; --drconf_cell_cnt) {
1056 struct of_drconf_cell drmem;
1058 read_drconf_cell(&drmem, &dm);
1060 /* skip this block if it is reserved or not assigned to
1062 if ((drmem.flags & DRCONF_MEM_RESERVED)
1063 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1066 if ((scn_addr < drmem.base_addr)
1067 || (scn_addr >= (drmem.base_addr + lmb_size)))
1070 nid = of_drconf_to_nid_single(&drmem, &aa);
1078 * Find the node associated with a hot added memory section for memory
1079 * represented in the device tree as a node (i.e. memory@XXXX) for
1082 int hot_add_node_scn_to_nid(unsigned long scn_addr)
1084 struct device_node *memory = NULL;
1087 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1088 unsigned long start, size;
1090 const unsigned int *memcell_buf;
1093 memcell_buf = of_get_property(memory, "reg", &len);
1094 if (!memcell_buf || len <= 0)
1097 /* ranges in cell */
1098 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1101 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1102 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1104 if ((scn_addr < start) || (scn_addr >= (start + size)))
1107 nid = of_node_to_nid_single(memory);
1111 of_node_put(memory);
1120 * Find the node associated with a hot added memory section. Section
1121 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1122 * sections are fully contained within a single LMB.
1124 int hot_add_scn_to_nid(unsigned long scn_addr)
1126 struct device_node *memory = NULL;
1129 if (!numa_enabled || (min_common_depth < 0))
1130 return first_online_node;
1132 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1134 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1135 of_node_put(memory);
1137 nid = hot_add_node_scn_to_nid(scn_addr);
1140 if (nid < 0 || !node_online(nid))
1141 nid = first_online_node;
1143 if (NODE_DATA(nid)->node_spanned_pages)
1146 for_each_online_node(nid) {
1147 if (NODE_DATA(nid)->node_spanned_pages) {
1157 #endif /* CONFIG_MEMORY_HOTPLUG */