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_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
;
275 rtas_root
= of_find_node_by_path("/rtas");
281 * this property is 2 32-bit integers, each representing a level of
282 * depth in the associativity nodes. The first is for an SMP
283 * configuration (should be all 0's) and the second is for a normal
284 * NUMA configuration.
286 ref_points
= of_get_property(rtas_root
,
287 "ibm,associativity-reference-points", &len
);
289 if ((len
>= 2 * sizeof(unsigned int)) && ref_points
) {
290 depth
= ref_points
[1];
292 dbg("NUMA: ibm,associativity-reference-points not found.\n");
295 of_node_put(rtas_root
);
300 static void __init
get_n_mem_cells(int *n_addr_cells
, int *n_size_cells
)
302 struct device_node
*memory
= NULL
;
304 memory
= of_find_node_by_type(memory
, "memory");
306 panic("numa.c: No memory nodes found!");
308 *n_addr_cells
= of_n_addr_cells(memory
);
309 *n_size_cells
= of_n_size_cells(memory
);
313 static unsigned long __devinit
read_n_cells(int n
, const unsigned int **buf
)
315 unsigned long result
= 0;
318 result
= (result
<< 32) | **buf
;
324 struct of_drconf_cell
{
332 #define DRCONF_MEM_ASSIGNED 0x00000008
333 #define DRCONF_MEM_AI_INVALID 0x00000040
334 #define DRCONF_MEM_RESERVED 0x00000080
337 * Read the next lmb list entry from the ibm,dynamic-memory property
338 * and return the information in the provided of_drconf_cell structure.
340 static void read_drconf_cell(struct of_drconf_cell
*drmem
, const u32
**cellp
)
344 drmem
->base_addr
= read_n_cells(n_mem_addr_cells
, cellp
);
347 drmem
->drc_index
= cp
[0];
348 drmem
->reserved
= cp
[1];
349 drmem
->aa_index
= cp
[2];
350 drmem
->flags
= cp
[3];
356 * Retreive and validate the ibm,dynamic-memory property of the device tree.
358 * The layout of the ibm,dynamic-memory property is a number N of lmb
359 * list entries followed by N lmb list entries. Each lmb list entry
360 * contains information as layed out in the of_drconf_cell struct above.
362 static int of_get_drconf_memory(struct device_node
*memory
, const u32
**dm
)
367 prop
= of_get_property(memory
, "ibm,dynamic-memory", &len
);
368 if (!prop
|| len
< sizeof(unsigned int))
373 /* Now that we know the number of entries, revalidate the size
374 * of the property read in to ensure we have everything
376 if (len
< (entries
* (n_mem_addr_cells
+ 4) + 1) * sizeof(unsigned int))
384 * Retreive and validate the ibm,lmb-size property for drconf memory
385 * from the device tree.
387 static u64
of_get_lmb_size(struct device_node
*memory
)
392 prop
= of_get_property(memory
, "ibm,lmb-size", &len
);
393 if (!prop
|| len
< sizeof(unsigned int))
396 return read_n_cells(n_mem_size_cells
, &prop
);
399 struct assoc_arrays
{
406 * Retreive and validate the list of associativity arrays for drconf
407 * memory from the ibm,associativity-lookup-arrays property of the
410 * The layout of the ibm,associativity-lookup-arrays property is a number N
411 * indicating the number of associativity arrays, followed by a number M
412 * indicating the size of each associativity array, followed by a list
413 * of N associativity arrays.
415 static int of_get_assoc_arrays(struct device_node
*memory
,
416 struct assoc_arrays
*aa
)
421 prop
= of_get_property(memory
, "ibm,associativity-lookup-arrays", &len
);
422 if (!prop
|| len
< 2 * sizeof(unsigned int))
425 aa
->n_arrays
= *prop
++;
426 aa
->array_sz
= *prop
++;
428 /* Now that we know the number of arrrays and size of each array,
429 * revalidate the size of the property read in.
431 if (len
< (aa
->n_arrays
* aa
->array_sz
+ 2) * sizeof(unsigned int))
439 * This is like of_node_to_nid_single() for memory represented in the
440 * ibm,dynamic-reconfiguration-memory node.
442 static int of_drconf_to_nid_single(struct of_drconf_cell
*drmem
,
443 struct assoc_arrays
*aa
)
446 int nid
= default_nid
;
449 if (min_common_depth
> 0 && min_common_depth
<= aa
->array_sz
&&
450 !(drmem
->flags
& DRCONF_MEM_AI_INVALID
) &&
451 drmem
->aa_index
< aa
->n_arrays
) {
452 index
= drmem
->aa_index
* aa
->array_sz
+ min_common_depth
- 1;
453 nid
= aa
->arrays
[index
];
455 if (nid
== 0xffff || nid
>= MAX_NUMNODES
)
463 * Figure out to which domain a cpu belongs and stick it there.
464 * Return the id of the domain used.
466 static int __cpuinit
numa_setup_cpu(unsigned long lcpu
)
469 struct device_node
*cpu
= of_get_cpu_node(lcpu
, NULL
);
476 nid
= of_node_to_nid_single(cpu
);
478 if (nid
< 0 || !node_online(nid
))
479 nid
= first_online_node
;
481 map_cpu_to_node(lcpu
, nid
);
488 static int __cpuinit
cpu_numa_callback(struct notifier_block
*nfb
,
489 unsigned long action
,
492 unsigned long lcpu
= (unsigned long)hcpu
;
493 int ret
= NOTIFY_DONE
;
497 case CPU_UP_PREPARE_FROZEN
:
498 numa_setup_cpu(lcpu
);
501 #ifdef CONFIG_HOTPLUG_CPU
503 case CPU_DEAD_FROZEN
:
504 case CPU_UP_CANCELED
:
505 case CPU_UP_CANCELED_FROZEN
:
506 unmap_cpu_from_node(lcpu
);
515 * Check and possibly modify a memory region to enforce the memory limit.
517 * Returns the size the region should have to enforce the memory limit.
518 * This will either be the original value of size, a truncated value,
519 * or zero. If the returned value of size is 0 the region should be
520 * discarded as it lies wholy above the memory limit.
522 static unsigned long __init
numa_enforce_memory_limit(unsigned long start
,
526 * We use lmb_end_of_DRAM() in here instead of memory_limit because
527 * we've already adjusted it for the limit and it takes care of
528 * having memory holes below the limit. Also, in the case of
529 * iommu_is_off, memory_limit is not set but is implicitly enforced.
532 if (start
+ size
<= lmb_end_of_DRAM())
535 if (start
>= lmb_end_of_DRAM())
538 return lmb_end_of_DRAM() - start
;
542 * Reads the counter for a given entry in
543 * linux,drconf-usable-memory property
545 static inline int __init
read_usm_ranges(const u32
**usm
)
548 * For each lmb in ibm,dynamic-memory a corresponding
549 * entry in linux,drconf-usable-memory property contains
550 * a counter followed by that many (base, size) duple.
551 * read the counter from linux,drconf-usable-memory
553 return read_n_cells(n_mem_size_cells
, usm
);
557 * Extract NUMA information from the ibm,dynamic-reconfiguration-memory
558 * node. This assumes n_mem_{addr,size}_cells have been set.
560 static void __init
parse_drconf_memory(struct device_node
*memory
)
563 unsigned int n
, rc
, ranges
, is_kexec_kdump
= 0;
564 unsigned long lmb_size
, base
, size
, sz
;
566 struct assoc_arrays aa
;
568 n
= of_get_drconf_memory(memory
, &dm
);
572 lmb_size
= of_get_lmb_size(memory
);
576 rc
= of_get_assoc_arrays(memory
, &aa
);
580 /* check if this is a kexec/kdump kernel */
581 usm
= of_get_usable_memory(memory
);
585 for (; n
!= 0; --n
) {
586 struct of_drconf_cell drmem
;
588 read_drconf_cell(&drmem
, &dm
);
590 /* skip this block if the reserved bit is set in flags (0x80)
591 or if the block is not assigned to this partition (0x8) */
592 if ((drmem
.flags
& DRCONF_MEM_RESERVED
)
593 || !(drmem
.flags
& DRCONF_MEM_ASSIGNED
))
596 base
= drmem
.base_addr
;
600 if (is_kexec_kdump
) {
601 ranges
= read_usm_ranges(&usm
);
602 if (!ranges
) /* there are no (base, size) duple */
606 if (is_kexec_kdump
) {
607 base
= read_n_cells(n_mem_addr_cells
, &usm
);
608 size
= read_n_cells(n_mem_size_cells
, &usm
);
610 nid
= of_drconf_to_nid_single(&drmem
, &aa
);
611 fake_numa_create_new_node(
612 ((base
+ size
) >> PAGE_SHIFT
),
614 node_set_online(nid
);
615 sz
= numa_enforce_memory_limit(base
, size
);
617 add_active_range(nid
, base
>> PAGE_SHIFT
,
619 + (sz
>> PAGE_SHIFT
));
624 static int __init
parse_numa_properties(void)
626 struct device_node
*cpu
= NULL
;
627 struct device_node
*memory
= NULL
;
631 if (numa_enabled
== 0) {
632 printk(KERN_WARNING
"NUMA disabled by user\n");
636 min_common_depth
= find_min_common_depth();
638 if (min_common_depth
< 0)
639 return min_common_depth
;
641 dbg("NUMA associativity depth for CPU/Memory: %d\n", min_common_depth
);
644 * Even though we connect cpus to numa domains later in SMP
645 * init, we need to know the node ids now. This is because
646 * each node to be onlined must have NODE_DATA etc backing it.
648 for_each_present_cpu(i
) {
651 cpu
= of_get_cpu_node(i
, NULL
);
653 nid
= of_node_to_nid_single(cpu
);
657 * Don't fall back to default_nid yet -- we will plug
658 * cpus into nodes once the memory scan has discovered
663 node_set_online(nid
);
666 get_n_mem_cells(&n_mem_addr_cells
, &n_mem_size_cells
);
668 while ((memory
= of_find_node_by_type(memory
, "memory")) != NULL
) {
673 const unsigned int *memcell_buf
;
676 memcell_buf
= of_get_property(memory
,
677 "linux,usable-memory", &len
);
678 if (!memcell_buf
|| len
<= 0)
679 memcell_buf
= of_get_property(memory
, "reg", &len
);
680 if (!memcell_buf
|| len
<= 0)
684 ranges
= (len
>> 2) / (n_mem_addr_cells
+ n_mem_size_cells
);
686 /* these are order-sensitive, and modify the buffer pointer */
687 start
= read_n_cells(n_mem_addr_cells
, &memcell_buf
);
688 size
= read_n_cells(n_mem_size_cells
, &memcell_buf
);
691 * Assumption: either all memory nodes or none will
692 * have associativity properties. If none, then
693 * everything goes to default_nid.
695 nid
= of_node_to_nid_single(memory
);
699 fake_numa_create_new_node(((start
+ size
) >> PAGE_SHIFT
), &nid
);
700 node_set_online(nid
);
702 if (!(size
= numa_enforce_memory_limit(start
, size
))) {
709 add_active_range(nid
, start
>> PAGE_SHIFT
,
710 (start
>> PAGE_SHIFT
) + (size
>> PAGE_SHIFT
));
717 * Now do the same thing for each LMB listed in the ibm,dynamic-memory
718 * property in the ibm,dynamic-reconfiguration-memory node.
720 memory
= of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
722 parse_drconf_memory(memory
);
727 static void __init
setup_nonnuma(void)
729 unsigned long top_of_ram
= lmb_end_of_DRAM();
730 unsigned long total_ram
= lmb_phys_mem_size();
731 unsigned long start_pfn
, end_pfn
;
732 unsigned int i
, nid
= 0;
734 printk(KERN_DEBUG
"Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
735 top_of_ram
, total_ram
);
736 printk(KERN_DEBUG
"Memory hole size: %ldMB\n",
737 (top_of_ram
- total_ram
) >> 20);
739 for (i
= 0; i
< lmb
.memory
.cnt
; ++i
) {
740 start_pfn
= lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
;
741 end_pfn
= start_pfn
+ lmb_size_pages(&lmb
.memory
, i
);
743 fake_numa_create_new_node(end_pfn
, &nid
);
744 add_active_range(nid
, start_pfn
, end_pfn
);
745 node_set_online(nid
);
749 void __init
dump_numa_cpu_topology(void)
752 unsigned int cpu
, count
;
754 if (min_common_depth
== -1 || !numa_enabled
)
757 for_each_online_node(node
) {
758 printk(KERN_DEBUG
"Node %d CPUs:", node
);
762 * If we used a CPU iterator here we would miss printing
763 * the holes in the cpumap.
765 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
++) {
766 if (cpumask_test_cpu(cpu
,
767 node_to_cpumask_map
[node
])) {
773 printk("-%u", cpu
- 1);
779 printk("-%u", nr_cpu_ids
- 1);
784 static void __init
dump_numa_memory_topology(void)
789 if (min_common_depth
== -1 || !numa_enabled
)
792 for_each_online_node(node
) {
795 printk(KERN_DEBUG
"Node %d Memory:", node
);
799 for (i
= 0; i
< lmb_end_of_DRAM();
800 i
+= (1 << SECTION_SIZE_BITS
)) {
801 if (early_pfn_to_nid(i
>> PAGE_SHIFT
) == node
) {
819 * Allocate some memory, satisfying the lmb or bootmem allocator where
820 * required. nid is the preferred node and end is the physical address of
821 * the highest address in the node.
823 * Returns the virtual address of the memory.
825 static void __init
*careful_zallocation(int nid
, unsigned long size
,
827 unsigned long end_pfn
)
831 unsigned long ret_paddr
;
833 ret_paddr
= __lmb_alloc_base(size
, align
, end_pfn
<< PAGE_SHIFT
);
835 /* retry over all memory */
837 ret_paddr
= __lmb_alloc_base(size
, align
, lmb_end_of_DRAM());
840 panic("numa.c: cannot allocate %lu bytes for node %d",
843 ret
= __va(ret_paddr
);
846 * We initialize the nodes in numeric order: 0, 1, 2...
847 * and hand over control from the LMB allocator to the
848 * bootmem allocator. If this function is called for
849 * node 5, then we know that all nodes <5 are using the
850 * bootmem allocator instead of the LMB allocator.
852 * So, check the nid from which this allocation came
853 * and double check to see if we need to use bootmem
854 * instead of the LMB. We don't free the LMB memory
855 * since it would be useless.
857 new_nid
= early_pfn_to_nid(ret_paddr
>> PAGE_SHIFT
);
859 ret
= __alloc_bootmem_node(NODE_DATA(new_nid
),
862 dbg("alloc_bootmem %p %lx\n", ret
, size
);
865 memset(ret
, 0, size
);
869 static struct notifier_block __cpuinitdata ppc64_numa_nb
= {
870 .notifier_call
= cpu_numa_callback
,
871 .priority
= 1 /* Must run before sched domains notifier. */
874 static void mark_reserved_regions_for_nid(int nid
)
876 struct pglist_data
*node
= NODE_DATA(nid
);
879 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++) {
880 unsigned long physbase
= lmb
.reserved
.region
[i
].base
;
881 unsigned long size
= lmb
.reserved
.region
[i
].size
;
882 unsigned long start_pfn
= physbase
>> PAGE_SHIFT
;
883 unsigned long end_pfn
= PFN_UP(physbase
+ size
);
884 struct node_active_region node_ar
;
885 unsigned long node_end_pfn
= node
->node_start_pfn
+
886 node
->node_spanned_pages
;
889 * Check to make sure that this lmb.reserved area is
890 * within the bounds of the node that we care about.
891 * Checking the nid of the start and end points is not
892 * sufficient because the reserved area could span the
895 if (end_pfn
<= node
->node_start_pfn
||
896 start_pfn
>= node_end_pfn
)
899 get_node_active_region(start_pfn
, &node_ar
);
900 while (start_pfn
< end_pfn
&&
901 node_ar
.start_pfn
< node_ar
.end_pfn
) {
902 unsigned long reserve_size
= size
;
904 * if reserved region extends past active region
905 * then trim size to active region
907 if (end_pfn
> node_ar
.end_pfn
)
908 reserve_size
= (node_ar
.end_pfn
<< PAGE_SHIFT
)
911 * Only worry about *this* node, others may not
912 * yet have valid NODE_DATA().
914 if (node_ar
.nid
== nid
) {
915 dbg("reserve_bootmem %lx %lx nid=%d\n",
916 physbase
, reserve_size
, node_ar
.nid
);
917 reserve_bootmem_node(NODE_DATA(node_ar
.nid
),
918 physbase
, reserve_size
,
922 * if reserved region is contained in the active region
925 if (end_pfn
<= node_ar
.end_pfn
)
929 * reserved region extends past the active region
930 * get next active region that contains this
933 start_pfn
= node_ar
.end_pfn
;
934 physbase
= start_pfn
<< PAGE_SHIFT
;
935 size
= size
- reserve_size
;
936 get_node_active_region(start_pfn
, &node_ar
);
942 void __init
do_init_bootmem(void)
947 max_low_pfn
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
948 max_pfn
= max_low_pfn
;
950 if (parse_numa_properties())
953 dump_numa_memory_topology();
955 for_each_online_node(nid
) {
956 unsigned long start_pfn
, end_pfn
;
958 unsigned long bootmap_pages
;
960 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
963 * Allocate the node structure node local if possible
965 * Be careful moving this around, as it relies on all
966 * previous nodes' bootmem to be initialized and have
967 * all reserved areas marked.
969 NODE_DATA(nid
) = careful_zallocation(nid
,
970 sizeof(struct pglist_data
),
971 SMP_CACHE_BYTES
, end_pfn
);
973 dbg("node %d\n", nid
);
974 dbg("NODE_DATA() = %p\n", NODE_DATA(nid
));
976 NODE_DATA(nid
)->bdata
= &bootmem_node_data
[nid
];
977 NODE_DATA(nid
)->node_start_pfn
= start_pfn
;
978 NODE_DATA(nid
)->node_spanned_pages
= end_pfn
- start_pfn
;
980 if (NODE_DATA(nid
)->node_spanned_pages
== 0)
983 dbg("start_paddr = %lx\n", start_pfn
<< PAGE_SHIFT
);
984 dbg("end_paddr = %lx\n", end_pfn
<< PAGE_SHIFT
);
986 bootmap_pages
= bootmem_bootmap_pages(end_pfn
- start_pfn
);
987 bootmem_vaddr
= careful_zallocation(nid
,
988 bootmap_pages
<< PAGE_SHIFT
,
991 dbg("bootmap_vaddr = %p\n", bootmem_vaddr
);
993 init_bootmem_node(NODE_DATA(nid
),
994 __pa(bootmem_vaddr
) >> PAGE_SHIFT
,
997 free_bootmem_with_active_regions(nid
, end_pfn
);
999 * Be very careful about moving this around. Future
1000 * calls to careful_zallocation() depend on this getting
1003 mark_reserved_regions_for_nid(nid
);
1004 sparse_memory_present_with_active_regions(nid
);
1007 init_bootmem_done
= 1;
1010 * Now bootmem is initialised we can create the node to cpumask
1011 * lookup tables and setup the cpu callback to populate them.
1013 setup_node_to_cpumask_map();
1015 register_cpu_notifier(&ppc64_numa_nb
);
1016 cpu_numa_callback(&ppc64_numa_nb
, CPU_UP_PREPARE
,
1017 (void *)(unsigned long)boot_cpuid
);
1020 void __init
paging_init(void)
1022 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
1023 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
1024 max_zone_pfns
[ZONE_DMA
] = lmb_end_of_DRAM() >> PAGE_SHIFT
;
1025 free_area_init_nodes(max_zone_pfns
);
1028 static int __init
early_numa(char *p
)
1033 if (strstr(p
, "off"))
1036 if (strstr(p
, "debug"))
1039 p
= strstr(p
, "fake=");
1041 cmdline
= p
+ strlen("fake=");
1045 early_param("numa", early_numa
);
1047 #ifdef CONFIG_MEMORY_HOTPLUG
1049 * Find the node associated with a hot added memory section for
1050 * memory represented in the device tree by the property
1051 * ibm,dynamic-reconfiguration-memory/ibm,dynamic-memory.
1053 static int hot_add_drconf_scn_to_nid(struct device_node
*memory
,
1054 unsigned long scn_addr
)
1057 unsigned int drconf_cell_cnt
, rc
;
1058 unsigned long lmb_size
;
1059 struct assoc_arrays aa
;
1062 drconf_cell_cnt
= of_get_drconf_memory(memory
, &dm
);
1063 if (!drconf_cell_cnt
)
1066 lmb_size
= of_get_lmb_size(memory
);
1070 rc
= of_get_assoc_arrays(memory
, &aa
);
1074 for (; drconf_cell_cnt
!= 0; --drconf_cell_cnt
) {
1075 struct of_drconf_cell drmem
;
1077 read_drconf_cell(&drmem
, &dm
);
1079 /* skip this block if it is reserved or not assigned to
1081 if ((drmem
.flags
& DRCONF_MEM_RESERVED
)
1082 || !(drmem
.flags
& DRCONF_MEM_ASSIGNED
))
1085 if ((scn_addr
< drmem
.base_addr
)
1086 || (scn_addr
>= (drmem
.base_addr
+ lmb_size
)))
1089 nid
= of_drconf_to_nid_single(&drmem
, &aa
);
1097 * Find the node associated with a hot added memory section for memory
1098 * represented in the device tree as a node (i.e. memory@XXXX) for
1101 int hot_add_node_scn_to_nid(unsigned long scn_addr
)
1103 struct device_node
*memory
= NULL
;
1106 while ((memory
= of_find_node_by_type(memory
, "memory")) != NULL
) {
1107 unsigned long start
, size
;
1109 const unsigned int *memcell_buf
;
1112 memcell_buf
= of_get_property(memory
, "reg", &len
);
1113 if (!memcell_buf
|| len
<= 0)
1116 /* ranges in cell */
1117 ranges
= (len
>> 2) / (n_mem_addr_cells
+ n_mem_size_cells
);
1120 start
= read_n_cells(n_mem_addr_cells
, &memcell_buf
);
1121 size
= read_n_cells(n_mem_size_cells
, &memcell_buf
);
1123 if ((scn_addr
< start
) || (scn_addr
>= (start
+ size
)))
1126 nid
= of_node_to_nid_single(memory
);
1130 of_node_put(memory
);
1139 * Find the node associated with a hot added memory section. Section
1140 * corresponds to a SPARSEMEM section, not an LMB. It is assumed that
1141 * sections are fully contained within a single LMB.
1143 int hot_add_scn_to_nid(unsigned long scn_addr
)
1145 struct device_node
*memory
= NULL
;
1148 if (!numa_enabled
|| (min_common_depth
< 0))
1149 return first_online_node
;
1151 memory
= of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
1153 nid
= hot_add_drconf_scn_to_nid(memory
, scn_addr
);
1154 of_node_put(memory
);
1156 nid
= hot_add_node_scn_to_nid(scn_addr
);
1159 if (nid
< 0 || !node_online(nid
))
1160 nid
= first_online_node
;
1162 if (NODE_DATA(nid
)->node_spanned_pages
)
1165 for_each_online_node(nid
) {
1166 if (NODE_DATA(nid
)->node_spanned_pages
) {
1176 #endif /* CONFIG_MEMORY_HOTPLUG */