4 * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 #include <linux/threads.h>
12 #include <linux/bootmem.h>
13 #include <linux/init.h>
15 #include <linux/mmzone.h>
16 #include <linux/module.h>
17 #include <linux/nodemask.h>
18 #include <linux/cpu.h>
19 #include <linux/notifier.h>
20 #include <linux/lmb.h>
22 #include <asm/sparsemem.h>
24 #include <asm/system.h>
27 static int numa_enabled = 1;
29 static char *cmdline __initdata;
31 static int numa_debug;
32 #define dbg(args...) if (numa_debug) { printk(KERN_INFO args); }
34 int numa_cpu_lookup_table[NR_CPUS];
35 cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
36 struct pglist_data *node_data[MAX_NUMNODES];
38 EXPORT_SYMBOL(numa_cpu_lookup_table);
39 EXPORT_SYMBOL(numa_cpumask_lookup_table);
40 EXPORT_SYMBOL(node_data);
42 static int min_common_depth;
43 static int n_mem_addr_cells, n_mem_size_cells;
45 static int __cpuinit fake_numa_create_new_node(unsigned long end_pfn,
48 unsigned long long mem;
50 static unsigned int fake_nid;
51 static unsigned long long curr_boundary;
54 * Modify node id, iff we started creating NUMA nodes
55 * We want to continue from where we left of the last time
60 * In case there are no more arguments to parse, the
61 * node_id should be the same as the last fake node id
62 * (we've handled this above).
67 mem = memparse(p, &p);
71 if (mem < curr_boundary)
76 if ((end_pfn << PAGE_SHIFT) > mem) {
78 * Skip commas and spaces
80 while (*p == ',' || *p == ' ' || *p == '\t')
86 dbg("created new fake_node with id %d\n", fake_nid);
93 * get_active_region_work_fn - A helper function for get_node_active_region
94 * Returns datax set to the start_pfn and end_pfn if they contain
95 * the initial value of datax->start_pfn between them
96 * @start_pfn: start page(inclusive) of region to check
97 * @end_pfn: end page(exclusive) of region to check
98 * @datax: comes in with ->start_pfn set to value to search for and
99 * goes out with active range if it contains it
100 * Returns 1 if search value is in range else 0
102 static int __init get_active_region_work_fn(unsigned long start_pfn,
103 unsigned long end_pfn, void *datax)
105 struct node_active_region *data;
106 data = (struct node_active_region *)datax;
108 if (start_pfn <= data->start_pfn && end_pfn > data->start_pfn) {
109 data->start_pfn = start_pfn;
110 data->end_pfn = end_pfn;
118 * get_node_active_region - Return active region containing start_pfn
119 * Active range returned is empty if none found.
120 * @start_pfn: The page to return the region for.
121 * @node_ar: Returned set to the active region containing start_pfn
123 static void __init get_node_active_region(unsigned long start_pfn,
124 struct node_active_region *node_ar)
126 int nid = early_pfn_to_nid(start_pfn);
129 node_ar->start_pfn = start_pfn;
130 node_ar->end_pfn = start_pfn;
131 work_with_active_regions(nid, get_active_region_work_fn, node_ar);
134 static void __cpuinit map_cpu_to_node(int cpu, int node)
136 numa_cpu_lookup_table[cpu] = node;
138 dbg("adding cpu %d to node %d\n", cpu, node);
140 if (!(cpu_isset(cpu, numa_cpumask_lookup_table[node])))
141 cpu_set(cpu, numa_cpumask_lookup_table[node]);
144 #ifdef CONFIG_HOTPLUG_CPU
145 static void unmap_cpu_from_node(unsigned long cpu)
147 int node = numa_cpu_lookup_table[cpu];
149 dbg("removing cpu %lu from node %d\n", cpu, node);
151 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
152 cpu_clear(cpu, numa_cpumask_lookup_table[node]);
154 printk(KERN_ERR "WARNING: cpu %lu not found in node %d\n",
158 #endif /* CONFIG_HOTPLUG_CPU */
160 static struct device_node * __cpuinit find_cpu_node(unsigned int cpu)
162 unsigned int hw_cpuid = get_hard_smp_processor_id(cpu);
163 struct device_node *cpu_node = NULL;
164 const unsigned int *interrupt_server, *reg;
167 while ((cpu_node = of_find_node_by_type(cpu_node, "cpu")) != NULL) {
168 /* Try interrupt server first */
169 interrupt_server = of_get_property(cpu_node,
170 "ibm,ppc-interrupt-server#s", &len);
172 len = len / sizeof(u32);
174 if (interrupt_server && (len > 0)) {
176 if (interrupt_server[len] == hw_cpuid)
180 reg = of_get_property(cpu_node, "reg", &len);
181 if (reg && (len > 0) && (reg[0] == hw_cpuid))
189 /* must hold reference to node during call */
190 static const int *of_get_associativity(struct device_node *dev)
192 return of_get_property(dev, "ibm,associativity", NULL);
196 * Returns the property linux,drconf-usable-memory if
197 * it exists (the property exists only in kexec/kdump kernels,
198 * added by kexec-tools)
200 static const u32 *of_get_usable_memory(struct device_node *memory)
204 prop = of_get_property(memory, "linux,drconf-usable-memory", &len);
205 if (!prop || len < sizeof(unsigned int))
210 /* Returns nid in the range [0..MAX_NUMNODES-1], or -1 if no useful numa
213 static int of_node_to_nid_single(struct device_node *device)
216 const unsigned int *tmp;
218 if (min_common_depth == -1)
221 tmp = of_get_associativity(device);
225 if (tmp[0] >= min_common_depth)
226 nid = tmp[min_common_depth];
228 /* POWER4 LPAR uses 0xffff as invalid node */
229 if (nid == 0xffff || nid >= MAX_NUMNODES)
235 /* Walk the device tree upwards, looking for an associativity id */
236 int of_node_to_nid(struct device_node *device)
238 struct device_node *tmp;
243 nid = of_node_to_nid_single(device);
248 device = of_get_parent(tmp);
255 EXPORT_SYMBOL_GPL(of_node_to_nid);
258 * In theory, the "ibm,associativity" property may contain multiple
259 * associativity lists because a resource may be multiply connected
260 * into the machine. This resource then has different associativity
261 * characteristics relative to its multiple connections. We ignore
262 * this for now. We also assume that all cpu and memory sets have
263 * their distances represented at a common level. This won't be
264 * true for hierarchical NUMA.
266 * In any case the ibm,associativity-reference-points should give
267 * the correct depth for a normal NUMA system.
269 * - Dave Hansen <haveblue@us.ibm.com>
271 static int __init find_min_common_depth(void)
274 const unsigned int *ref_points;
275 struct device_node *rtas_root;
278 rtas_root = of_find_node_by_path("/rtas");
284 * this property is 2 32-bit integers, each representing a level of
285 * depth in the associativity nodes. The first is for an SMP
286 * configuration (should be all 0's) and the second is for a normal
287 * NUMA configuration.
289 ref_points = of_get_property(rtas_root,
290 "ibm,associativity-reference-points", &len);
292 if ((len >= 1) && ref_points) {
293 depth = ref_points[1];
295 dbg("NUMA: ibm,associativity-reference-points not found.\n");
298 of_node_put(rtas_root);
303 static void __init get_n_mem_cells(int *n_addr_cells, int *n_size_cells)
305 struct device_node *memory = NULL;
307 memory = of_find_node_by_type(memory, "memory");
309 panic("numa.c: No memory nodes found!");
311 *n_addr_cells = of_n_addr_cells(memory);
312 *n_size_cells = of_n_size_cells(memory);
316 static unsigned long __devinit read_n_cells(int n, const unsigned int **buf)
318 unsigned long result = 0;
321 result = (result << 32) | **buf;
327 struct of_drconf_cell {
335 #define DRCONF_MEM_ASSIGNED 0x00000008
336 #define DRCONF_MEM_AI_INVALID 0x00000040
337 #define DRCONF_MEM_RESERVED 0x00000080
340 * Read the next lmb list entry from the ibm,dynamic-memory property
341 * and return the information in the provided of_drconf_cell structure.
343 static void read_drconf_cell(struct of_drconf_cell *drmem, const u32 **cellp)
347 drmem->base_addr = read_n_cells(n_mem_addr_cells, cellp);
350 drmem->drc_index = cp[0];
351 drmem->reserved = cp[1];
352 drmem->aa_index = cp[2];
353 drmem->flags = cp[3];
359 * Retreive and validate the ibm,dynamic-memory property of the device tree.
361 * The layout of the ibm,dynamic-memory property is a number N of lmb
362 * list entries followed by N lmb list entries. Each lmb list entry
363 * contains information as layed out in the of_drconf_cell struct above.
365 static int of_get_drconf_memory(struct device_node *memory, const u32 **dm)
370 prop = of_get_property(memory, "ibm,dynamic-memory", &len);
371 if (!prop || len < sizeof(unsigned int))
376 /* Now that we know the number of entries, revalidate the size
377 * of the property read in to ensure we have everything
379 if (len < (entries * (n_mem_addr_cells + 4) + 1) * sizeof(unsigned int))
387 * Retreive and validate the ibm,lmb-size property for drconf memory
388 * from the device tree.
390 static u64 of_get_lmb_size(struct device_node *memory)
395 prop = of_get_property(memory, "ibm,lmb-size", &len);
396 if (!prop || len < sizeof(unsigned int))
399 return read_n_cells(n_mem_size_cells, &prop);
402 struct assoc_arrays {
409 * Retreive and validate the list of associativity arrays for drconf
410 * memory from the ibm,associativity-lookup-arrays property of the
413 * The layout of the ibm,associativity-lookup-arrays property is a number N
414 * indicating the number of associativity arrays, followed by a number M
415 * indicating the size of each associativity array, followed by a list
416 * of N associativity arrays.
418 static int of_get_assoc_arrays(struct device_node *memory,
419 struct assoc_arrays *aa)
424 prop = of_get_property(memory, "ibm,associativity-lookup-arrays", &len);
425 if (!prop || len < 2 * sizeof(unsigned int))
428 aa->n_arrays = *prop++;
429 aa->array_sz = *prop++;
431 /* Now that we know the number of arrrays and size of each array,
432 * revalidate the size of the property read in.
434 if (len < (aa->n_arrays * aa->array_sz + 2) * sizeof(unsigned int))
442 * This is like of_node_to_nid_single() for memory represented in the
443 * ibm,dynamic-reconfiguration-memory node.
445 static int of_drconf_to_nid_single(struct of_drconf_cell *drmem,
446 struct assoc_arrays *aa)
449 int nid = default_nid;
452 if (min_common_depth > 0 && min_common_depth <= aa->array_sz &&
453 !(drmem->flags & DRCONF_MEM_AI_INVALID) &&
454 drmem->aa_index < aa->n_arrays) {
455 index = drmem->aa_index * aa->array_sz + min_common_depth - 1;
456 nid = aa->arrays[index];
458 if (nid == 0xffff || nid >= MAX_NUMNODES)
466 * Figure out to which domain a cpu belongs and stick it there.
467 * Return the id of the domain used.
469 static int __cpuinit numa_setup_cpu(unsigned long lcpu)
472 struct device_node *cpu = find_cpu_node(lcpu);
479 nid = of_node_to_nid_single(cpu);
481 if (nid < 0 || !node_online(nid))
482 nid = any_online_node(NODE_MASK_ALL);
484 map_cpu_to_node(lcpu, nid);
491 static int __cpuinit cpu_numa_callback(struct notifier_block *nfb,
492 unsigned long action,
495 unsigned long lcpu = (unsigned long)hcpu;
496 int ret = NOTIFY_DONE;
500 case CPU_UP_PREPARE_FROZEN:
501 numa_setup_cpu(lcpu);
504 #ifdef CONFIG_HOTPLUG_CPU
506 case CPU_DEAD_FROZEN:
507 case CPU_UP_CANCELED:
508 case CPU_UP_CANCELED_FROZEN:
509 unmap_cpu_from_node(lcpu);
518 * Check and possibly modify a memory region to enforce the memory limit.
520 * Returns the size the region should have to enforce the memory limit.
521 * This will either be the original value of size, a truncated value,
522 * or zero. If the returned value of size is 0 the region should be
523 * discarded as it lies wholy above the memory limit.
525 static unsigned long __init numa_enforce_memory_limit(unsigned long start,
529 * We use lmb_end_of_DRAM() in here instead of memory_limit because
530 * we've already adjusted it for the limit and it takes care of
531 * having memory holes below the limit. Also, in the case of
532 * iommu_is_off, memory_limit is not set but is implicitly enforced.
535 if (start + size <= lmb_end_of_DRAM())
538 if (start >= lmb_end_of_DRAM())
541 return lmb_end_of_DRAM() - start;
545 * Reads the counter for a given entry in
546 * linux,drconf-usable-memory property
548 static inline int __init read_usm_ranges(const u32 **usm)
551 * For each lmb in ibm,dynamic-memory a corresponding
552 * entry in linux,drconf-usable-memory property contains
553 * a counter followed by that many (base, size) duple.
554 * read the counter from linux,drconf-usable-memory
556 return read_n_cells(n_mem_size_cells, usm);
560 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
561 * node. This assumes n_mem_{addr,size}_cells have been set.
563 static void __init parse_drconf_memory(struct device_node *memory)
566 unsigned int n, rc, ranges, is_kexec_kdump = 0;
567 unsigned long lmb_size, base, size, sz;
569 struct assoc_arrays aa;
571 n = of_get_drconf_memory(memory, &dm);
575 lmb_size = of_get_lmb_size(memory);
579 rc = of_get_assoc_arrays(memory, &aa);
583 /* check if this is a kexec/kdump kernel */
584 usm = of_get_usable_memory(memory);
588 for (; n != 0; --n) {
589 struct of_drconf_cell drmem;
591 read_drconf_cell(&drmem, &dm);
593 /* skip this block if the reserved bit is set in flags (0x80)
594 or if the block is not assigned to this partition (0x8) */
595 if ((drmem.flags & DRCONF_MEM_RESERVED)
596 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
599 base = drmem.base_addr;
603 if (is_kexec_kdump) {
604 ranges = read_usm_ranges(&usm);
605 if (!ranges) /* there are no (base, size) duple */
609 if (is_kexec_kdump) {
610 base = read_n_cells(n_mem_addr_cells, &usm);
611 size = read_n_cells(n_mem_size_cells, &usm);
613 nid = of_drconf_to_nid_single(&drmem, &aa);
614 fake_numa_create_new_node(
615 ((base + size) >> PAGE_SHIFT),
617 node_set_online(nid);
618 sz = numa_enforce_memory_limit(base, size);
620 add_active_range(nid, base >> PAGE_SHIFT,
622 + (sz >> PAGE_SHIFT));
627 static int __init parse_numa_properties(void)
629 struct device_node *cpu = NULL;
630 struct device_node *memory = NULL;
634 if (numa_enabled == 0) {
635 printk(KERN_WARNING "NUMA disabled by user\n");
639 min_common_depth = find_min_common_depth();
641 if (min_common_depth < 0)
642 return min_common_depth;
644 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth);
647 * Even though we connect cpus to numa domains later in SMP
648 * init, we need to know the node ids now. This is because
649 * each node to be onlined must have NODE_DATA etc backing it.
651 for_each_present_cpu(i) {
654 cpu = find_cpu_node(i);
656 nid = of_node_to_nid_single(cpu);
660 * Don't fall back to default_nid yet -- we will plug
661 * cpus into nodes once the memory scan has discovered
666 node_set_online(nid);
669 get_n_mem_cells(&n_mem_addr_cells, &n_mem_size_cells);
671 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
676 const unsigned int *memcell_buf;
679 memcell_buf = of_get_property(memory,
680 "linux,usable-memory", &len);
681 if (!memcell_buf || len <= 0)
682 memcell_buf = of_get_property(memory, "reg", &len);
683 if (!memcell_buf || len <= 0)
687 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
689 /* these are order-sensitive, and modify the buffer pointer */
690 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
691 size = read_n_cells(n_mem_size_cells, &memcell_buf);
694 * Assumption: either all memory nodes or none will
695 * have associativity properties. If none, then
696 * everything goes to default_nid.
698 nid = of_node_to_nid_single(memory);
702 fake_numa_create_new_node(((start + size) >> PAGE_SHIFT), &nid);
703 node_set_online(nid);
705 if (!(size = numa_enforce_memory_limit(start, size))) {
712 add_active_range(nid, start >> PAGE_SHIFT,
713 (start >> PAGE_SHIFT) + (size >> PAGE_SHIFT));
720 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
721 * property in the ibm,dynamic-reconfiguration-memory node.
723 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
725 parse_drconf_memory(memory);
730 static void __init setup_nonnuma(void)
732 unsigned long top_of_ram = lmb_end_of_DRAM();
733 unsigned long total_ram = lmb_phys_mem_size();
734 unsigned long start_pfn, end_pfn;
735 unsigned int i, nid = 0;
737 printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
738 top_of_ram, total_ram);
739 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
740 (top_of_ram - total_ram) >> 20);
742 for (i = 0; i < lmb.memory.cnt; ++i) {
743 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
744 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
746 fake_numa_create_new_node(end_pfn, &nid);
747 add_active_range(nid, start_pfn, end_pfn);
748 node_set_online(nid);
752 void __init dump_numa_cpu_topology(void)
755 unsigned int cpu, count;
757 if (min_common_depth == -1 || !numa_enabled)
760 for_each_online_node(node) {
761 printk(KERN_DEBUG "Node %d CPUs:", node);
765 * If we used a CPU iterator here we would miss printing
766 * the holes in the cpumap.
768 for (cpu = 0; cpu < NR_CPUS; cpu++) {
769 if (cpu_isset(cpu, numa_cpumask_lookup_table[node])) {
775 printk("-%u", cpu - 1);
781 printk("-%u", NR_CPUS - 1);
786 static void __init dump_numa_memory_topology(void)
791 if (min_common_depth == -1 || !numa_enabled)
794 for_each_online_node(node) {
797 printk(KERN_DEBUG "Node %d Memory:", node);
801 for (i = 0; i < lmb_end_of_DRAM();
802 i += (1 << SECTION_SIZE_BITS)) {
803 if (early_pfn_to_nid(i >> PAGE_SHIFT) == node) {
821 * Allocate some memory, satisfying the lmb or bootmem allocator where
822 * required. nid is the preferred node and end is the physical address of
823 * the highest address in the node.
825 * Returns the virtual address of the memory.
827 static void __init *careful_zallocation(int nid, unsigned long size,
829 unsigned long end_pfn)
833 unsigned long ret_paddr;
835 ret_paddr = __lmb_alloc_base(size, align, end_pfn << PAGE_SHIFT);
837 /* retry over all memory */
839 ret_paddr = __lmb_alloc_base(size, align, lmb_end_of_DRAM());
842 panic("numa.c: cannot allocate %lu bytes for node %d",
845 ret = __va(ret_paddr);
848 * We initialize the nodes in numeric order: 0, 1, 2...
849 * and hand over control from the LMB allocator to the
850 * bootmem allocator. If this function is called for
851 * node 5, then we know that all nodes <5 are using the
852 * bootmem allocator instead of the LMB allocator.
854 * So, check the nid from which this allocation came
855 * and double check to see if we need to use bootmem
856 * instead of the LMB. We don't free the LMB memory
857 * since it would be useless.
859 new_nid = early_pfn_to_nid(ret_paddr >> PAGE_SHIFT);
861 ret = __alloc_bootmem_node(NODE_DATA(new_nid),
864 dbg("alloc_bootmem %p %lx\n", ret, size);
867 memset(ret, 0, size);
871 static struct notifier_block __cpuinitdata ppc64_numa_nb = {
872 .notifier_call = cpu_numa_callback,
873 .priority = 1 /* Must run before sched domains notifier. */
876 static void mark_reserved_regions_for_nid(int nid)
878 struct pglist_data *node = NODE_DATA(nid);
881 for (i = 0; i < lmb.reserved.cnt; i++) {
882 unsigned long physbase = lmb.reserved.region[i].base;
883 unsigned long size = lmb.reserved.region[i].size;
884 unsigned long start_pfn = physbase >> PAGE_SHIFT;
885 unsigned long end_pfn = ((physbase + size) >> PAGE_SHIFT);
886 struct node_active_region node_ar;
887 unsigned long node_end_pfn = node->node_start_pfn +
888 node->node_spanned_pages;
891 * Check to make sure that this lmb.reserved area is
892 * within the bounds of the node that we care about.
893 * Checking the nid of the start and end points is not
894 * sufficient because the reserved area could span the
897 if (end_pfn <= node->node_start_pfn ||
898 start_pfn >= node_end_pfn)
901 get_node_active_region(start_pfn, &node_ar);
902 while (start_pfn < end_pfn &&
903 node_ar.start_pfn < node_ar.end_pfn) {
904 unsigned long reserve_size = size;
906 * if reserved region extends past active region
907 * then trim size to active region
909 if (end_pfn > node_ar.end_pfn)
910 reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
911 - (start_pfn << PAGE_SHIFT);
913 * Only worry about *this* node, others may not
914 * yet have valid NODE_DATA().
916 if (node_ar.nid == nid) {
917 dbg("reserve_bootmem %lx %lx nid=%d\n",
918 physbase, reserve_size, node_ar.nid);
919 reserve_bootmem_node(NODE_DATA(node_ar.nid),
920 physbase, reserve_size,
924 * if reserved region is contained in the active region
927 if (end_pfn <= node_ar.end_pfn)
931 * reserved region extends past the active region
932 * get next active region that contains this
935 start_pfn = node_ar.end_pfn;
936 physbase = start_pfn << PAGE_SHIFT;
937 size = size - reserve_size;
938 get_node_active_region(start_pfn, &node_ar);
944 void __init do_init_bootmem(void)
949 max_low_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
950 max_pfn = max_low_pfn;
952 if (parse_numa_properties())
955 dump_numa_memory_topology();
957 register_cpu_notifier(&ppc64_numa_nb);
958 cpu_numa_callback(&ppc64_numa_nb, CPU_UP_PREPARE,
959 (void *)(unsigned long)boot_cpuid);
961 for_each_online_node(nid) {
962 unsigned long start_pfn, end_pfn;
964 unsigned long bootmap_pages;
966 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
969 * Allocate the node structure node local if possible
971 * Be careful moving this around, as it relies on all
972 * previous nodes' bootmem to be initialized and have
973 * all reserved areas marked.
975 NODE_DATA(nid) = careful_zallocation(nid,
976 sizeof(struct pglist_data),
977 SMP_CACHE_BYTES, end_pfn);
979 dbg("node %d\n", nid);
980 dbg("NODE_DATA() = %p\n", NODE_DATA(nid));
982 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
983 NODE_DATA(nid)->node_start_pfn = start_pfn;
984 NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
986 if (NODE_DATA(nid)->node_spanned_pages == 0)
989 dbg("start_paddr = %lx\n", start_pfn << PAGE_SHIFT);
990 dbg("end_paddr = %lx\n", end_pfn << PAGE_SHIFT);
992 bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
993 bootmem_vaddr = careful_zallocation(nid,
994 bootmap_pages << PAGE_SHIFT,
997 dbg("bootmap_vaddr = %p\n", bootmem_vaddr);
999 init_bootmem_node(NODE_DATA(nid),
1000 __pa(bootmem_vaddr) >> PAGE_SHIFT,
1001 start_pfn, end_pfn);
1003 free_bootmem_with_active_regions(nid, end_pfn);
1005 * Be very careful about moving this around. Future
1006 * calls to careful_zallocation() depend on this getting
1009 mark_reserved_regions_for_nid(nid);
1010 sparse_memory_present_with_active_regions(nid);
1014 void __init paging_init(void)
1016 unsigned long max_zone_pfns[MAX_NR_ZONES];
1017 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
1018 max_zone_pfns[ZONE_DMA] = lmb_end_of_DRAM() >> PAGE_SHIFT;
1019 free_area_init_nodes(max_zone_pfns);
1022 static int __init early_numa(char *p)
1027 if (strstr(p, "off"))
1030 if (strstr(p, "debug"))
1033 p = strstr(p, "fake=");
1035 cmdline = p + strlen("fake=");
1039 early_param("numa", early_numa);
1041 #ifdef CONFIG_MEMORY_HOTPLUG
1043 * Validate the node associated with the memory section we are
1046 int valid_hot_add_scn(int *nid, unsigned long start, u32 lmb_size,
1047 unsigned long scn_addr)
1051 if (*nid < 0 || !node_online(*nid))
1052 *nid = any_online_node(NODE_MASK_ALL);
1054 if ((scn_addr >= start) && (scn_addr < (start + lmb_size))) {
1055 nodes_setall(nodes);
1056 while (NODE_DATA(*nid)->node_spanned_pages == 0) {
1057 node_clear(*nid, nodes);
1058 *nid = any_online_node(nodes);
1068 * Find the node associated with a hot added memory section represented
1069 * by the ibm,dynamic-reconfiguration-memory node.
1071 static int hot_add_drconf_scn_to_nid(struct device_node *memory,
1072 unsigned long scn_addr)
1076 unsigned long lmb_size;
1077 int default_nid = any_online_node(NODE_MASK_ALL);
1079 struct assoc_arrays aa;
1081 n = of_get_drconf_memory(memory, &dm);
1083 return default_nid;;
1085 lmb_size = of_get_lmb_size(memory);
1089 rc = of_get_assoc_arrays(memory, &aa);
1093 for (; n != 0; --n) {
1094 struct of_drconf_cell drmem;
1096 read_drconf_cell(&drmem, &dm);
1098 /* skip this block if it is reserved or not assigned to
1100 if ((drmem.flags & DRCONF_MEM_RESERVED)
1101 || !(drmem.flags & DRCONF_MEM_ASSIGNED))
1104 nid = of_drconf_to_nid_single(&drmem, &aa);
1106 if (valid_hot_add_scn(&nid, drmem.base_addr, lmb_size,
1111 BUG(); /* section address should be found above */
1116 * Find the node associated with a hot added memory section. Section
1117 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1118 * sections are fully contained within a single LMB.
1120 int hot_add_scn_to_nid(unsigned long scn_addr)
1122 struct device_node *memory = NULL;
1125 if (!numa_enabled || (min_common_depth < 0))
1126 return any_online_node(NODE_MASK_ALL);
1128 memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1130 nid = hot_add_drconf_scn_to_nid(memory, scn_addr);
1131 of_node_put(memory);
1135 while ((memory = of_find_node_by_type(memory, "memory")) != NULL) {
1136 unsigned long start, size;
1138 const unsigned int *memcell_buf;
1141 memcell_buf = of_get_property(memory, "reg", &len);
1142 if (!memcell_buf || len <= 0)
1145 /* ranges in cell */
1146 ranges = (len >> 2) / (n_mem_addr_cells + n_mem_size_cells);
1148 start = read_n_cells(n_mem_addr_cells, &memcell_buf);
1149 size = read_n_cells(n_mem_size_cells, &memcell_buf);
1150 nid = of_node_to_nid_single(memory);
1152 if (valid_hot_add_scn(&nid, start, size, scn_addr)) {
1153 of_node_put(memory);
1157 if (--ranges) /* process all ranges in cell */
1160 BUG(); /* section address should be found above */
1163 #endif /* CONFIG_MEMORY_HOTPLUG */