2 * Generic VM initialization for x86-64 NUMA setups.
3 * Copyright 2002,2003 Andi Kleen, SuSE Labs.
5 #include <linux/kernel.h>
7 #include <linux/string.h>
8 #include <linux/init.h>
9 #include <linux/bootmem.h>
10 #include <linux/memblock.h>
11 #include <linux/mmzone.h>
12 #include <linux/ctype.h>
13 #include <linux/module.h>
14 #include <linux/nodemask.h>
15 #include <linux/sched.h>
16 #include <linux/acpi.h>
19 #include <asm/proto.h>
23 #include <asm/amd_nb.h>
33 struct numa_memblk blk
[NR_NODE_MEMBLKS
];
36 struct pglist_data
*node_data
[MAX_NUMNODES
] __read_mostly
;
37 EXPORT_SYMBOL(node_data
);
39 nodemask_t cpu_nodes_parsed __initdata
;
40 nodemask_t mem_nodes_parsed __initdata
;
42 struct memnode memnode
;
44 static unsigned long __initdata nodemap_addr
;
45 static unsigned long __initdata nodemap_size
;
47 static struct numa_meminfo numa_meminfo __initdata
;
49 struct bootnode numa_nodes
[MAX_NUMNODES
] __initdata
;
52 * Given a shift value, try to populate memnodemap[]
55 * 0 if memnodmap[] too small (of shift too small)
56 * -1 if node overlap or lost ram (shift too big)
58 static int __init
populate_memnodemap(const struct numa_meminfo
*mi
, int shift
)
60 unsigned long addr
, end
;
63 memset(memnodemap
, 0xff, sizeof(s16
)*memnodemapsize
);
64 for (i
= 0; i
< mi
->nr_blks
; i
++) {
65 addr
= mi
->blk
[i
].start
;
69 if ((end
>> shift
) >= memnodemapsize
)
72 if (memnodemap
[addr
>> shift
] != NUMA_NO_NODE
)
74 memnodemap
[addr
>> shift
] = mi
->blk
[i
].nid
;
75 addr
+= (1UL << shift
);
82 static int __init
allocate_cachealigned_memnodemap(void)
86 memnodemap
= memnode
.embedded_map
;
87 if (memnodemapsize
<= ARRAY_SIZE(memnode
.embedded_map
))
91 nodemap_size
= roundup(sizeof(s16
) * memnodemapsize
, L1_CACHE_BYTES
);
92 nodemap_addr
= memblock_find_in_range(addr
, get_max_mapped(),
93 nodemap_size
, L1_CACHE_BYTES
);
94 if (nodemap_addr
== MEMBLOCK_ERROR
) {
96 "NUMA: Unable to allocate Memory to Node hash map\n");
97 nodemap_addr
= nodemap_size
= 0;
100 memnodemap
= phys_to_virt(nodemap_addr
);
101 memblock_x86_reserve_range(nodemap_addr
, nodemap_addr
+ nodemap_size
, "MEMNODEMAP");
103 printk(KERN_DEBUG
"NUMA: Allocated memnodemap from %lx - %lx\n",
104 nodemap_addr
, nodemap_addr
+ nodemap_size
);
109 * The LSB of all start and end addresses in the node map is the value of the
110 * maximum possible shift.
112 static int __init
extract_lsb_from_nodes(const struct numa_meminfo
*mi
)
114 int i
, nodes_used
= 0;
115 unsigned long start
, end
;
116 unsigned long bitfield
= 0, memtop
= 0;
118 for (i
= 0; i
< mi
->nr_blks
; i
++) {
119 start
= mi
->blk
[i
].start
;
120 end
= mi
->blk
[i
].end
;
131 i
= find_first_bit(&bitfield
, sizeof(unsigned long)*8);
132 memnodemapsize
= (memtop
>> i
)+1;
136 static int __init
compute_hash_shift(const struct numa_meminfo
*mi
)
140 shift
= extract_lsb_from_nodes(mi
);
141 if (allocate_cachealigned_memnodemap())
143 printk(KERN_DEBUG
"NUMA: Using %d for the hash shift.\n",
146 if (populate_memnodemap(mi
, shift
) != 1) {
147 printk(KERN_INFO
"Your memory is not aligned you need to "
148 "rebuild your kernel with a bigger NODEMAPSIZE "
149 "shift=%d\n", shift
);
155 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
157 return phys_to_nid(pfn
<< PAGE_SHIFT
);
160 static void * __init
early_node_mem(int nodeid
, unsigned long start
,
161 unsigned long end
, unsigned long size
,
167 * put it on high as possible
168 * something will go with NODE_DATA
170 if (start
< (MAX_DMA_PFN
<<PAGE_SHIFT
))
171 start
= MAX_DMA_PFN
<<PAGE_SHIFT
;
172 if (start
< (MAX_DMA32_PFN
<<PAGE_SHIFT
) &&
173 end
> (MAX_DMA32_PFN
<<PAGE_SHIFT
))
174 start
= MAX_DMA32_PFN
<<PAGE_SHIFT
;
175 mem
= memblock_x86_find_in_range_node(nodeid
, start
, end
, size
, align
);
176 if (mem
!= MEMBLOCK_ERROR
)
179 /* extend the search scope */
180 end
= max_pfn_mapped
<< PAGE_SHIFT
;
181 start
= MAX_DMA_PFN
<< PAGE_SHIFT
;
182 mem
= memblock_find_in_range(start
, end
, size
, align
);
183 if (mem
!= MEMBLOCK_ERROR
)
186 printk(KERN_ERR
"Cannot find %lu bytes in node %d\n",
192 static __init
int conflicting_memblks(unsigned long start
, unsigned long end
)
194 struct numa_meminfo
*mi
= &numa_meminfo
;
197 for (i
= 0; i
< mi
->nr_blks
; i
++) {
198 struct numa_memblk
*blk
= &mi
->blk
[i
];
200 if (blk
->start
== blk
->end
)
202 if (blk
->end
> start
&& blk
->start
< end
)
204 if (blk
->end
== end
&& blk
->start
== start
)
210 int __init
numa_add_memblk(int nid
, u64 start
, u64 end
)
212 struct numa_meminfo
*mi
= &numa_meminfo
;
215 i
= conflicting_memblks(start
, end
);
217 printk(KERN_WARNING
"NUMA: Warning: node %d (%Lx-%Lx) overlaps with itself (%Lx-%Lx)\n",
218 nid
, start
, end
, numa_nodes
[i
].start
, numa_nodes
[i
].end
);
220 printk(KERN_ERR
"NUMA: node %d (%Lx-%Lx) overlaps with node %d (%Lx-%Lx)\n",
222 numa_nodes
[i
].start
, numa_nodes
[i
].end
);
226 mi
->blk
[mi
->nr_blks
].start
= start
;
227 mi
->blk
[mi
->nr_blks
].end
= end
;
228 mi
->blk
[mi
->nr_blks
].nid
= nid
;
233 static __init
void cutoff_node(int i
, unsigned long start
, unsigned long end
)
235 struct bootnode
*nd
= &numa_nodes
[i
];
237 if (nd
->start
< start
) {
239 if (nd
->end
< nd
->start
)
244 if (nd
->start
> nd
->end
)
249 /* Initialize bootmem allocator for a node */
251 setup_node_bootmem(int nodeid
, unsigned long start
, unsigned long end
)
253 unsigned long start_pfn
, last_pfn
, nodedata_phys
;
254 const int pgdat_size
= roundup(sizeof(pg_data_t
), PAGE_SIZE
);
261 * Don't confuse VM with a node that doesn't have the
262 * minimum amount of memory:
264 if (end
&& (end
- start
) < NODE_MIN_SIZE
)
267 start
= roundup(start
, ZONE_ALIGN
);
269 printk(KERN_INFO
"Initmem setup node %d %016lx-%016lx\n", nodeid
,
272 start_pfn
= start
>> PAGE_SHIFT
;
273 last_pfn
= end
>> PAGE_SHIFT
;
275 node_data
[nodeid
] = early_node_mem(nodeid
, start
, end
, pgdat_size
,
277 if (node_data
[nodeid
] == NULL
)
279 nodedata_phys
= __pa(node_data
[nodeid
]);
280 memblock_x86_reserve_range(nodedata_phys
, nodedata_phys
+ pgdat_size
, "NODE_DATA");
281 printk(KERN_INFO
" NODE_DATA [%016lx - %016lx]\n", nodedata_phys
,
282 nodedata_phys
+ pgdat_size
- 1);
283 nid
= phys_to_nid(nodedata_phys
);
285 printk(KERN_INFO
" NODE_DATA(%d) on node %d\n", nodeid
, nid
);
287 memset(NODE_DATA(nodeid
), 0, sizeof(pg_data_t
));
288 NODE_DATA(nodeid
)->node_id
= nodeid
;
289 NODE_DATA(nodeid
)->node_start_pfn
= start_pfn
;
290 NODE_DATA(nodeid
)->node_spanned_pages
= last_pfn
- start_pfn
;
292 node_set_online(nodeid
);
295 static int __init
numa_cleanup_meminfo(struct numa_meminfo
*mi
)
300 * Join together blocks on the same node, holes between
301 * which don't overlap with memory on other nodes.
303 for (i
= 0; i
< mi
->nr_blks
; ++i
) {
304 struct numa_memblk
*bi
= &mi
->blk
[i
];
307 for (j
= i
+ 1; j
< mi
->nr_blks
; ++j
) {
308 struct numa_memblk
*bj
= &mi
->blk
[j
];
309 unsigned long start
, end
;
311 if (bi
->nid
!= bj
->nid
)
313 start
= min(bi
->end
, bj
->end
);
314 end
= max(bi
->start
, bj
->start
);
315 for (k
= 0; k
< mi
->nr_blks
; ++k
) {
316 struct numa_memblk
*bk
= &mi
->blk
[k
];
318 if (bi
->nid
== bk
->nid
)
320 if (start
< bk
->end
&& end
> bk
->start
)
325 start
= min(bi
->start
, bj
->start
);
326 end
= max(bi
->end
, bj
->end
);
327 printk(KERN_INFO
"NUMA: Node %d [%Lx,%Lx) + [%Lx,%Lx) -> [%lx,%lx)\n",
328 bi
->nid
, bi
->start
, bi
->end
, bj
->start
, bj
->end
,
332 k
= --mi
->nr_blks
- j
;
333 memmove(mi
->blk
+ j
, mi
->blk
+ j
+ 1,
334 k
* sizeof(mi
->blk
[0]));
343 * Sanity check to catch more bad NUMA configurations (they are amazingly
344 * common). Make sure the nodes cover all memory.
346 static int __init
nodes_cover_memory(const struct bootnode
*nodes
)
348 unsigned long numaram
, e820ram
;
352 for_each_node_mask(i
, mem_nodes_parsed
) {
353 unsigned long s
= nodes
[i
].start
>> PAGE_SHIFT
;
354 unsigned long e
= nodes
[i
].end
>> PAGE_SHIFT
;
356 numaram
-= __absent_pages_in_range(i
, s
, e
);
357 if ((long)numaram
< 0)
361 e820ram
= max_pfn
- (memblock_x86_hole_size(0,
362 max_pfn
<< PAGE_SHIFT
) >> PAGE_SHIFT
);
363 /* We seem to lose 3 pages somewhere. Allow 1M of slack. */
364 if ((long)(e820ram
- numaram
) >= (1 << (20 - PAGE_SHIFT
))) {
365 printk(KERN_ERR
"NUMA: nodes only cover %luMB of your %luMB e820 RAM. Not used.\n",
366 (numaram
<< PAGE_SHIFT
) >> 20,
367 (e820ram
<< PAGE_SHIFT
) >> 20);
373 static int __init
numa_register_memblks(struct numa_meminfo
*mi
)
377 /* Account for nodes with cpus and no memory */
378 nodes_or(node_possible_map
, mem_nodes_parsed
, cpu_nodes_parsed
);
379 if (WARN_ON(nodes_empty(node_possible_map
)))
382 memnode_shift
= compute_hash_shift(mi
);
383 if (memnode_shift
< 0) {
384 printk(KERN_ERR
"NUMA: No NUMA node hash function found. Contact maintainer\n");
388 for (i
= 0; i
< mi
->nr_blks
; i
++)
389 memblock_x86_register_active_regions(mi
->blk
[i
].nid
,
390 mi
->blk
[i
].start
>> PAGE_SHIFT
,
391 mi
->blk
[i
].end
>> PAGE_SHIFT
);
393 /* for out of order entries */
395 if (!nodes_cover_memory(numa_nodes
))
398 init_memory_mapping_high();
400 /* Finally register nodes. */
401 for_each_node_mask(i
, node_possible_map
)
402 setup_node_bootmem(i
, numa_nodes
[i
].start
, numa_nodes
[i
].end
);
405 * Try again in case setup_node_bootmem missed one due to missing
408 for_each_node_mask(i
, node_possible_map
)
410 setup_node_bootmem(i
, numa_nodes
[i
].start
,
416 #ifdef CONFIG_NUMA_EMU
418 static struct bootnode nodes
[MAX_NUMNODES
] __initdata
;
419 static struct bootnode physnodes
[MAX_NUMNODES
] __cpuinitdata
;
420 static char *cmdline __initdata
;
422 void __init
numa_emu_cmdline(char *str
)
427 static int __init
setup_physnodes(unsigned long start
, unsigned long end
)
432 memset(physnodes
, 0, sizeof(physnodes
));
434 for_each_node_mask(i
, mem_nodes_parsed
) {
435 physnodes
[i
].start
= numa_nodes
[i
].start
;
436 physnodes
[i
].end
= numa_nodes
[i
].end
;
440 * Basic sanity checking on the physical node map: there may be errors
441 * if the SRAT or AMD code incorrectly reported the topology or the mem=
442 * kernel parameter is used.
444 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
445 if (physnodes
[i
].start
== physnodes
[i
].end
)
447 if (physnodes
[i
].start
> end
) {
448 physnodes
[i
].end
= physnodes
[i
].start
;
451 if (physnodes
[i
].end
< start
) {
452 physnodes
[i
].start
= physnodes
[i
].end
;
455 if (physnodes
[i
].start
< start
)
456 physnodes
[i
].start
= start
;
457 if (physnodes
[i
].end
> end
)
458 physnodes
[i
].end
= end
;
463 * If no physical topology was detected, a single node is faked to cover
464 * the entire address space.
467 physnodes
[ret
].start
= start
;
468 physnodes
[ret
].end
= end
;
474 static void __init
fake_physnodes(int acpi
, int amd
, int nr_nodes
)
479 #ifdef CONFIG_ACPI_NUMA
481 acpi_fake_nodes(nodes
, nr_nodes
);
483 #ifdef CONFIG_AMD_NUMA
485 amd_fake_nodes(nodes
, nr_nodes
);
488 for (i
= 0; i
< nr_cpu_ids
; i
++)
493 * Setups up nid to range from addr to addr + size. If the end
494 * boundary is greater than max_addr, then max_addr is used instead.
495 * The return value is 0 if there is additional memory left for
496 * allocation past addr and -1 otherwise. addr is adjusted to be at
497 * the end of the node.
499 static int __init
setup_node_range(int nid
, u64
*addr
, u64 size
, u64 max_addr
)
502 nodes
[nid
].start
= *addr
;
504 if (*addr
>= max_addr
) {
508 nodes
[nid
].end
= *addr
;
509 node_set(nid
, node_possible_map
);
510 printk(KERN_INFO
"Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid
,
511 nodes
[nid
].start
, nodes
[nid
].end
,
512 (nodes
[nid
].end
- nodes
[nid
].start
) >> 20);
517 * Sets up nr_nodes fake nodes interleaved over physical nodes ranging from addr
518 * to max_addr. The return value is the number of nodes allocated.
520 static int __init
split_nodes_interleave(u64 addr
, u64 max_addr
, int nr_nodes
)
522 nodemask_t physnode_mask
= NODE_MASK_NONE
;
530 if (nr_nodes
> MAX_NUMNODES
) {
531 pr_info("numa=fake=%d too large, reducing to %d\n",
532 nr_nodes
, MAX_NUMNODES
);
533 nr_nodes
= MAX_NUMNODES
;
536 size
= (max_addr
- addr
- memblock_x86_hole_size(addr
, max_addr
)) / nr_nodes
;
538 * Calculate the number of big nodes that can be allocated as a result
539 * of consolidating the remainder.
541 big
= ((size
& ~FAKE_NODE_MIN_HASH_MASK
) * nr_nodes
) /
544 size
&= FAKE_NODE_MIN_HASH_MASK
;
546 pr_err("Not enough memory for each node. "
547 "NUMA emulation disabled.\n");
551 for (i
= 0; i
< MAX_NUMNODES
; i
++)
552 if (physnodes
[i
].start
!= physnodes
[i
].end
)
553 node_set(i
, physnode_mask
);
556 * Continue to fill physical nodes with fake nodes until there is no
557 * memory left on any of them.
559 while (nodes_weight(physnode_mask
)) {
560 for_each_node_mask(i
, physnode_mask
) {
561 u64 end
= physnodes
[i
].start
+ size
;
562 u64 dma32_end
= PFN_PHYS(MAX_DMA32_PFN
);
565 end
+= FAKE_NODE_MIN_SIZE
;
568 * Continue to add memory to this fake node if its
569 * non-reserved memory is less than the per-node size.
571 while (end
- physnodes
[i
].start
-
572 memblock_x86_hole_size(physnodes
[i
].start
, end
) < size
) {
573 end
+= FAKE_NODE_MIN_SIZE
;
574 if (end
> physnodes
[i
].end
) {
575 end
= physnodes
[i
].end
;
581 * If there won't be at least FAKE_NODE_MIN_SIZE of
582 * non-reserved memory in ZONE_DMA32 for the next node,
583 * this one must extend to the boundary.
585 if (end
< dma32_end
&& dma32_end
- end
-
586 memblock_x86_hole_size(end
, dma32_end
) < FAKE_NODE_MIN_SIZE
)
590 * If there won't be enough non-reserved memory for the
591 * next node, this one must extend to the end of the
594 if (physnodes
[i
].end
- end
-
595 memblock_x86_hole_size(end
, physnodes
[i
].end
) < size
)
596 end
= physnodes
[i
].end
;
599 * Avoid allocating more nodes than requested, which can
600 * happen as a result of rounding down each node's size
601 * to FAKE_NODE_MIN_SIZE.
603 if (nodes_weight(physnode_mask
) + ret
>= nr_nodes
)
604 end
= physnodes
[i
].end
;
606 if (setup_node_range(ret
++, &physnodes
[i
].start
,
607 end
- physnodes
[i
].start
,
608 physnodes
[i
].end
) < 0)
609 node_clear(i
, physnode_mask
);
616 * Returns the end address of a node so that there is at least `size' amount of
617 * non-reserved memory or `max_addr' is reached.
619 static u64 __init
find_end_of_node(u64 start
, u64 max_addr
, u64 size
)
621 u64 end
= start
+ size
;
623 while (end
- start
- memblock_x86_hole_size(start
, end
) < size
) {
624 end
+= FAKE_NODE_MIN_SIZE
;
625 if (end
> max_addr
) {
634 * Sets up fake nodes of `size' interleaved over physical nodes ranging from
635 * `addr' to `max_addr'. The return value is the number of nodes allocated.
637 static int __init
split_nodes_size_interleave(u64 addr
, u64 max_addr
, u64 size
)
639 nodemask_t physnode_mask
= NODE_MASK_NONE
;
647 * The limit on emulated nodes is MAX_NUMNODES, so the size per node is
648 * increased accordingly if the requested size is too small. This
649 * creates a uniform distribution of node sizes across the entire
650 * machine (but not necessarily over physical nodes).
652 min_size
= (max_addr
- addr
- memblock_x86_hole_size(addr
, max_addr
)) /
654 min_size
= max(min_size
, FAKE_NODE_MIN_SIZE
);
655 if ((min_size
& FAKE_NODE_MIN_HASH_MASK
) < min_size
)
656 min_size
= (min_size
+ FAKE_NODE_MIN_SIZE
) &
657 FAKE_NODE_MIN_HASH_MASK
;
658 if (size
< min_size
) {
659 pr_err("Fake node size %LuMB too small, increasing to %LuMB\n",
660 size
>> 20, min_size
>> 20);
663 size
&= FAKE_NODE_MIN_HASH_MASK
;
665 for (i
= 0; i
< MAX_NUMNODES
; i
++)
666 if (physnodes
[i
].start
!= physnodes
[i
].end
)
667 node_set(i
, physnode_mask
);
669 * Fill physical nodes with fake nodes of size until there is no memory
670 * left on any of them.
672 while (nodes_weight(physnode_mask
)) {
673 for_each_node_mask(i
, physnode_mask
) {
674 u64 dma32_end
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
677 end
= find_end_of_node(physnodes
[i
].start
,
678 physnodes
[i
].end
, size
);
680 * If there won't be at least FAKE_NODE_MIN_SIZE of
681 * non-reserved memory in ZONE_DMA32 for the next node,
682 * this one must extend to the boundary.
684 if (end
< dma32_end
&& dma32_end
- end
-
685 memblock_x86_hole_size(end
, dma32_end
) < FAKE_NODE_MIN_SIZE
)
689 * If there won't be enough non-reserved memory for the
690 * next node, this one must extend to the end of the
693 if (physnodes
[i
].end
- end
-
694 memblock_x86_hole_size(end
, physnodes
[i
].end
) < size
)
695 end
= physnodes
[i
].end
;
698 * Setup the fake node that will be allocated as bootmem
699 * later. If setup_node_range() returns non-zero, there
700 * is no more memory available on this physical node.
702 if (setup_node_range(ret
++, &physnodes
[i
].start
,
703 end
- physnodes
[i
].start
,
704 physnodes
[i
].end
) < 0)
705 node_clear(i
, physnode_mask
);
712 * Sets up the system RAM area from start_pfn to last_pfn according to the
713 * numa=fake command-line option.
715 static int __init
numa_emulation(unsigned long start_pfn
,
716 unsigned long last_pfn
, int acpi
, int amd
)
718 static struct numa_meminfo ei __initdata
;
719 u64 addr
= start_pfn
<< PAGE_SHIFT
;
720 u64 max_addr
= last_pfn
<< PAGE_SHIFT
;
725 * If the numa=fake command-line contains a 'M' or 'G', it represents
726 * the fixed node size. Otherwise, if it is just a single number N,
727 * split the system RAM into N fake nodes.
729 if (strchr(cmdline
, 'M') || strchr(cmdline
, 'G')) {
732 size
= memparse(cmdline
, &cmdline
);
733 num_nodes
= split_nodes_size_interleave(addr
, max_addr
, size
);
737 n
= simple_strtoul(cmdline
, NULL
, 0);
738 num_nodes
= split_nodes_interleave(addr
, max_addr
, n
);
744 ei
.nr_blks
= num_nodes
;
745 for (i
= 0; i
< ei
.nr_blks
; i
++) {
746 ei
.blk
[i
].start
= nodes
[i
].start
;
747 ei
.blk
[i
].end
= nodes
[i
].end
;
751 memnode_shift
= compute_hash_shift(&ei
);
752 if (memnode_shift
< 0) {
754 printk(KERN_ERR
"No NUMA hash function found. NUMA emulation "
760 * We need to vacate all active ranges that may have been registered for
761 * the e820 memory map.
763 remove_all_active_ranges();
764 for_each_node_mask(i
, node_possible_map
)
765 memblock_x86_register_active_regions(i
, nodes
[i
].start
>> PAGE_SHIFT
,
766 nodes
[i
].end
>> PAGE_SHIFT
);
767 init_memory_mapping_high();
768 for_each_node_mask(i
, node_possible_map
)
769 setup_node_bootmem(i
, nodes
[i
].start
, nodes
[i
].end
);
770 setup_physnodes(addr
, max_addr
);
771 fake_physnodes(acpi
, amd
, num_nodes
);
775 #endif /* CONFIG_NUMA_EMU */
777 static int dummy_numa_init(void)
779 printk(KERN_INFO
"%s\n",
780 numa_off
? "NUMA turned off" : "No NUMA configuration found");
781 printk(KERN_INFO
"Faking a node at %016lx-%016lx\n",
782 0LU, max_pfn
<< PAGE_SHIFT
);
784 node_set(0, cpu_nodes_parsed
);
785 node_set(0, mem_nodes_parsed
);
786 numa_add_memblk(0, 0, (u64
)max_pfn
<< PAGE_SHIFT
);
787 numa_nodes
[0].start
= 0;
788 numa_nodes
[0].end
= (u64
)max_pfn
<< PAGE_SHIFT
;
793 void __init
initmem_init(void)
795 int (*numa_init
[])(void) = { [2] = dummy_numa_init
};
799 #ifdef CONFIG_ACPI_NUMA
800 numa_init
[0] = x86_acpi_numa_init
;
802 #ifdef CONFIG_AMD_NUMA
803 numa_init
[1] = amd_numa_init
;
807 for (i
= 0; i
< ARRAY_SIZE(numa_init
); i
++) {
811 for (j
= 0; j
< MAX_LOCAL_APIC
; j
++)
812 set_apicid_to_node(j
, NUMA_NO_NODE
);
814 nodes_clear(cpu_nodes_parsed
);
815 nodes_clear(mem_nodes_parsed
);
816 nodes_clear(node_possible_map
);
817 nodes_clear(node_online_map
);
818 memset(&numa_meminfo
, 0, sizeof(numa_meminfo
));
819 memset(numa_nodes
, 0, sizeof(numa_nodes
));
820 remove_all_active_ranges();
822 if (numa_init
[i
]() < 0)
825 /* clean up the node list */
826 for (j
= 0; j
< MAX_NUMNODES
; j
++)
827 cutoff_node(j
, 0, max_pfn
<< PAGE_SHIFT
);
829 #ifdef CONFIG_NUMA_EMU
830 setup_physnodes(0, max_pfn
<< PAGE_SHIFT
);
831 if (cmdline
&& !numa_emulation(0, max_pfn
, i
== 0, i
== 1))
833 setup_physnodes(0, max_pfn
<< PAGE_SHIFT
);
834 nodes_clear(node_possible_map
);
835 nodes_clear(node_online_map
);
837 if (numa_cleanup_meminfo(&numa_meminfo
) < 0)
840 if (numa_register_memblks(&numa_meminfo
) < 0)
843 for (j
= 0; j
< nr_cpu_ids
; j
++) {
844 int nid
= early_cpu_to_node(j
);
846 if (nid
== NUMA_NO_NODE
)
848 if (!node_online(nid
))
857 unsigned long __init
numa_free_all_bootmem(void)
859 unsigned long pages
= 0;
862 for_each_online_node(i
)
863 pages
+= free_all_bootmem_node(NODE_DATA(i
));
865 pages
+= free_all_memory_core_early(MAX_NUMNODES
);
870 int __cpuinit
numa_cpu_node(int cpu
)
872 int apicid
= early_per_cpu(x86_cpu_to_apicid
, cpu
);
874 if (apicid
!= BAD_APICID
)
875 return __apicid_to_node
[apicid
];
880 * UGLINESS AHEAD: Currently, CONFIG_NUMA_EMU is 64bit only and makes use
881 * of 64bit specific data structures. The distinction is artificial and
882 * should be removed. numa_{add|remove}_cpu() are implemented in numa.c
883 * for both 32 and 64bit when CONFIG_NUMA_EMU is disabled but here when
886 * NUMA emulation is planned to be made generic and the following and other
887 * related code should be moved to numa.c.
889 #ifdef CONFIG_NUMA_EMU
890 # ifndef CONFIG_DEBUG_PER_CPU_MAPS
891 void __cpuinit
numa_add_cpu(int cpu
)
896 nid
= numa_cpu_node(cpu
);
897 if (nid
== NUMA_NO_NODE
)
898 nid
= early_cpu_to_node(cpu
);
899 BUG_ON(nid
== NUMA_NO_NODE
|| !node_online(nid
));
902 * Use the starting address of the emulated node to find which physical
903 * node it is allocated on.
905 addr
= node_start_pfn(nid
) << PAGE_SHIFT
;
906 for (physnid
= 0; physnid
< MAX_NUMNODES
; physnid
++)
907 if (addr
>= physnodes
[physnid
].start
&&
908 addr
< physnodes
[physnid
].end
)
912 * Map the cpu to each emulated node that is allocated on the physical
913 * node of the cpu's apic id.
915 for_each_online_node(nid
) {
916 addr
= node_start_pfn(nid
) << PAGE_SHIFT
;
917 if (addr
>= physnodes
[physnid
].start
&&
918 addr
< physnodes
[physnid
].end
)
919 cpumask_set_cpu(cpu
, node_to_cpumask_map
[nid
]);
923 void __cpuinit
numa_remove_cpu(int cpu
)
927 for_each_online_node(i
)
928 cpumask_clear_cpu(cpu
, node_to_cpumask_map
[i
]);
930 # else /* !CONFIG_DEBUG_PER_CPU_MAPS */
931 static void __cpuinit
numa_set_cpumask(int cpu
, int enable
)
933 int node
= early_cpu_to_node(cpu
);
934 struct cpumask
*mask
;
937 if (node
== NUMA_NO_NODE
) {
938 /* early_cpu_to_node() already emits a warning and trace */
941 for_each_online_node(i
) {
944 addr
= node_start_pfn(i
) << PAGE_SHIFT
;
945 if (addr
< physnodes
[node
].start
||
946 addr
>= physnodes
[node
].end
)
948 mask
= debug_cpumask_set_cpu(cpu
, enable
);
953 cpumask_set_cpu(cpu
, mask
);
955 cpumask_clear_cpu(cpu
, mask
);
959 void __cpuinit
numa_add_cpu(int cpu
)
961 numa_set_cpumask(cpu
, 1);
964 void __cpuinit
numa_remove_cpu(int cpu
)
966 numa_set_cpumask(cpu
, 0);
968 # endif /* !CONFIG_DEBUG_PER_CPU_MAPS */
969 #endif /* CONFIG_NUMA_EMU */