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
);
296 * Sanity check to catch more bad NUMA configurations (they are amazingly
297 * common). Make sure the nodes cover all memory.
299 static int __init
nodes_cover_memory(const struct bootnode
*nodes
)
301 unsigned long numaram
, e820ram
;
305 for_each_node_mask(i
, mem_nodes_parsed
) {
306 unsigned long s
= nodes
[i
].start
>> PAGE_SHIFT
;
307 unsigned long e
= nodes
[i
].end
>> PAGE_SHIFT
;
309 numaram
-= __absent_pages_in_range(i
, s
, e
);
310 if ((long)numaram
< 0)
315 (memblock_x86_hole_size(0, max_pfn
<<PAGE_SHIFT
) >> PAGE_SHIFT
);
316 /* We seem to lose 3 pages somewhere. Allow 1M of slack. */
317 if ((long)(e820ram
- numaram
) >= (1<<(20 - PAGE_SHIFT
))) {
318 printk(KERN_ERR
"NUMA: nodes only cover %luMB of your %luMB e820 RAM. Not used.\n",
319 (numaram
<< PAGE_SHIFT
) >> 20,
320 (e820ram
<< PAGE_SHIFT
) >> 20);
326 static int __init
numa_register_memblks(void)
328 struct numa_meminfo
*mi
= &numa_meminfo
;
332 * Join together blocks on the same node, holes between
333 * which don't overlap with memory on other nodes.
335 for (i
= 0; i
< mi
->nr_blks
; ++i
) {
336 struct numa_memblk
*bi
= &mi
->blk
[i
];
339 for (j
= i
+ 1; j
< mi
->nr_blks
; ++j
) {
340 struct numa_memblk
*bj
= &mi
->blk
[j
];
341 unsigned long start
, end
;
343 if (bi
->nid
!= bj
->nid
)
345 start
= min(bi
->end
, bj
->end
);
346 end
= max(bi
->start
, bj
->start
);
347 for (k
= 0; k
< mi
->nr_blks
; ++k
) {
348 struct numa_memblk
*bk
= &mi
->blk
[k
];
350 if (bi
->nid
== bk
->nid
)
352 if (start
< bk
->end
&& end
> bk
->start
)
357 start
= min(bi
->start
, bj
->start
);
358 end
= max(bi
->end
, bj
->end
);
359 printk(KERN_INFO
"NUMA: Node %d [%Lx,%Lx) + [%Lx,%Lx) -> [%lx,%lx)\n",
360 bi
->nid
, bi
->start
, bi
->end
, bj
->start
, bj
->end
,
364 k
= --mi
->nr_blks
- j
;
365 memmove(mi
->blk
+ j
, mi
->blk
+ j
+ 1,
366 k
* sizeof(mi
->blk
[0]));
371 memnode_shift
= compute_hash_shift(mi
);
372 if (memnode_shift
< 0) {
373 printk(KERN_ERR
"NUMA: No NUMA node hash function found. Contact maintainer\n");
377 for (i
= 0; i
< mi
->nr_blks
; i
++)
378 memblock_x86_register_active_regions(mi
->blk
[i
].nid
,
379 mi
->blk
[i
].start
>> PAGE_SHIFT
,
380 mi
->blk
[i
].end
>> PAGE_SHIFT
);
382 /* for out of order entries */
384 if (!nodes_cover_memory(numa_nodes
))
387 init_memory_mapping_high();
389 /* Finally register nodes. */
390 for_each_node_mask(i
, node_possible_map
)
391 setup_node_bootmem(i
, numa_nodes
[i
].start
, numa_nodes
[i
].end
);
394 * Try again in case setup_node_bootmem missed one due to missing
397 for_each_node_mask(i
, node_possible_map
)
399 setup_node_bootmem(i
, numa_nodes
[i
].start
,
405 #ifdef CONFIG_NUMA_EMU
407 static struct bootnode nodes
[MAX_NUMNODES
] __initdata
;
408 static struct bootnode physnodes
[MAX_NUMNODES
] __cpuinitdata
;
409 static char *cmdline __initdata
;
411 void __init
numa_emu_cmdline(char *str
)
416 static int __init
setup_physnodes(unsigned long start
, unsigned long end
)
421 memset(physnodes
, 0, sizeof(physnodes
));
423 for_each_node_mask(i
, mem_nodes_parsed
) {
424 physnodes
[i
].start
= numa_nodes
[i
].start
;
425 physnodes
[i
].end
= numa_nodes
[i
].end
;
429 * Basic sanity checking on the physical node map: there may be errors
430 * if the SRAT or AMD code incorrectly reported the topology or the mem=
431 * kernel parameter is used.
433 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
434 if (physnodes
[i
].start
== physnodes
[i
].end
)
436 if (physnodes
[i
].start
> end
) {
437 physnodes
[i
].end
= physnodes
[i
].start
;
440 if (physnodes
[i
].end
< start
) {
441 physnodes
[i
].start
= physnodes
[i
].end
;
444 if (physnodes
[i
].start
< start
)
445 physnodes
[i
].start
= start
;
446 if (physnodes
[i
].end
> end
)
447 physnodes
[i
].end
= end
;
452 * If no physical topology was detected, a single node is faked to cover
453 * the entire address space.
456 physnodes
[ret
].start
= start
;
457 physnodes
[ret
].end
= end
;
463 static void __init
fake_physnodes(int acpi
, int amd
, int nr_nodes
)
468 #ifdef CONFIG_ACPI_NUMA
470 acpi_fake_nodes(nodes
, nr_nodes
);
472 #ifdef CONFIG_AMD_NUMA
474 amd_fake_nodes(nodes
, nr_nodes
);
477 for (i
= 0; i
< nr_cpu_ids
; i
++)
482 * Setups up nid to range from addr to addr + size. If the end
483 * boundary is greater than max_addr, then max_addr is used instead.
484 * The return value is 0 if there is additional memory left for
485 * allocation past addr and -1 otherwise. addr is adjusted to be at
486 * the end of the node.
488 static int __init
setup_node_range(int nid
, u64
*addr
, u64 size
, u64 max_addr
)
491 nodes
[nid
].start
= *addr
;
493 if (*addr
>= max_addr
) {
497 nodes
[nid
].end
= *addr
;
498 node_set(nid
, node_possible_map
);
499 printk(KERN_INFO
"Faking node %d at %016Lx-%016Lx (%LuMB)\n", nid
,
500 nodes
[nid
].start
, nodes
[nid
].end
,
501 (nodes
[nid
].end
- nodes
[nid
].start
) >> 20);
506 * Sets up nr_nodes fake nodes interleaved over physical nodes ranging from addr
507 * to max_addr. The return value is the number of nodes allocated.
509 static int __init
split_nodes_interleave(u64 addr
, u64 max_addr
, int nr_nodes
)
511 nodemask_t physnode_mask
= NODE_MASK_NONE
;
519 if (nr_nodes
> MAX_NUMNODES
) {
520 pr_info("numa=fake=%d too large, reducing to %d\n",
521 nr_nodes
, MAX_NUMNODES
);
522 nr_nodes
= MAX_NUMNODES
;
525 size
= (max_addr
- addr
- memblock_x86_hole_size(addr
, max_addr
)) / nr_nodes
;
527 * Calculate the number of big nodes that can be allocated as a result
528 * of consolidating the remainder.
530 big
= ((size
& ~FAKE_NODE_MIN_HASH_MASK
) * nr_nodes
) /
533 size
&= FAKE_NODE_MIN_HASH_MASK
;
535 pr_err("Not enough memory for each node. "
536 "NUMA emulation disabled.\n");
540 for (i
= 0; i
< MAX_NUMNODES
; i
++)
541 if (physnodes
[i
].start
!= physnodes
[i
].end
)
542 node_set(i
, physnode_mask
);
545 * Continue to fill physical nodes with fake nodes until there is no
546 * memory left on any of them.
548 while (nodes_weight(physnode_mask
)) {
549 for_each_node_mask(i
, physnode_mask
) {
550 u64 end
= physnodes
[i
].start
+ size
;
551 u64 dma32_end
= PFN_PHYS(MAX_DMA32_PFN
);
554 end
+= FAKE_NODE_MIN_SIZE
;
557 * Continue to add memory to this fake node if its
558 * non-reserved memory is less than the per-node size.
560 while (end
- physnodes
[i
].start
-
561 memblock_x86_hole_size(physnodes
[i
].start
, end
) < size
) {
562 end
+= FAKE_NODE_MIN_SIZE
;
563 if (end
> physnodes
[i
].end
) {
564 end
= physnodes
[i
].end
;
570 * If there won't be at least FAKE_NODE_MIN_SIZE of
571 * non-reserved memory in ZONE_DMA32 for the next node,
572 * this one must extend to the boundary.
574 if (end
< dma32_end
&& dma32_end
- end
-
575 memblock_x86_hole_size(end
, dma32_end
) < FAKE_NODE_MIN_SIZE
)
579 * If there won't be enough non-reserved memory for the
580 * next node, this one must extend to the end of the
583 if (physnodes
[i
].end
- end
-
584 memblock_x86_hole_size(end
, physnodes
[i
].end
) < size
)
585 end
= physnodes
[i
].end
;
588 * Avoid allocating more nodes than requested, which can
589 * happen as a result of rounding down each node's size
590 * to FAKE_NODE_MIN_SIZE.
592 if (nodes_weight(physnode_mask
) + ret
>= nr_nodes
)
593 end
= physnodes
[i
].end
;
595 if (setup_node_range(ret
++, &physnodes
[i
].start
,
596 end
- physnodes
[i
].start
,
597 physnodes
[i
].end
) < 0)
598 node_clear(i
, physnode_mask
);
605 * Returns the end address of a node so that there is at least `size' amount of
606 * non-reserved memory or `max_addr' is reached.
608 static u64 __init
find_end_of_node(u64 start
, u64 max_addr
, u64 size
)
610 u64 end
= start
+ size
;
612 while (end
- start
- memblock_x86_hole_size(start
, end
) < size
) {
613 end
+= FAKE_NODE_MIN_SIZE
;
614 if (end
> max_addr
) {
623 * Sets up fake nodes of `size' interleaved over physical nodes ranging from
624 * `addr' to `max_addr'. The return value is the number of nodes allocated.
626 static int __init
split_nodes_size_interleave(u64 addr
, u64 max_addr
, u64 size
)
628 nodemask_t physnode_mask
= NODE_MASK_NONE
;
636 * The limit on emulated nodes is MAX_NUMNODES, so the size per node is
637 * increased accordingly if the requested size is too small. This
638 * creates a uniform distribution of node sizes across the entire
639 * machine (but not necessarily over physical nodes).
641 min_size
= (max_addr
- addr
- memblock_x86_hole_size(addr
, max_addr
)) /
643 min_size
= max(min_size
, FAKE_NODE_MIN_SIZE
);
644 if ((min_size
& FAKE_NODE_MIN_HASH_MASK
) < min_size
)
645 min_size
= (min_size
+ FAKE_NODE_MIN_SIZE
) &
646 FAKE_NODE_MIN_HASH_MASK
;
647 if (size
< min_size
) {
648 pr_err("Fake node size %LuMB too small, increasing to %LuMB\n",
649 size
>> 20, min_size
>> 20);
652 size
&= FAKE_NODE_MIN_HASH_MASK
;
654 for (i
= 0; i
< MAX_NUMNODES
; i
++)
655 if (physnodes
[i
].start
!= physnodes
[i
].end
)
656 node_set(i
, physnode_mask
);
658 * Fill physical nodes with fake nodes of size until there is no memory
659 * left on any of them.
661 while (nodes_weight(physnode_mask
)) {
662 for_each_node_mask(i
, physnode_mask
) {
663 u64 dma32_end
= MAX_DMA32_PFN
<< PAGE_SHIFT
;
666 end
= find_end_of_node(physnodes
[i
].start
,
667 physnodes
[i
].end
, size
);
669 * If there won't be at least FAKE_NODE_MIN_SIZE of
670 * non-reserved memory in ZONE_DMA32 for the next node,
671 * this one must extend to the boundary.
673 if (end
< dma32_end
&& dma32_end
- end
-
674 memblock_x86_hole_size(end
, dma32_end
) < FAKE_NODE_MIN_SIZE
)
678 * If there won't be enough non-reserved memory for the
679 * next node, this one must extend to the end of the
682 if (physnodes
[i
].end
- end
-
683 memblock_x86_hole_size(end
, physnodes
[i
].end
) < size
)
684 end
= physnodes
[i
].end
;
687 * Setup the fake node that will be allocated as bootmem
688 * later. If setup_node_range() returns non-zero, there
689 * is no more memory available on this physical node.
691 if (setup_node_range(ret
++, &physnodes
[i
].start
,
692 end
- physnodes
[i
].start
,
693 physnodes
[i
].end
) < 0)
694 node_clear(i
, physnode_mask
);
701 * Sets up the system RAM area from start_pfn to last_pfn according to the
702 * numa=fake command-line option.
704 static int __init
numa_emulation(unsigned long start_pfn
,
705 unsigned long last_pfn
, int acpi
, int amd
)
707 static struct numa_meminfo ei __initdata
;
708 u64 addr
= start_pfn
<< PAGE_SHIFT
;
709 u64 max_addr
= last_pfn
<< PAGE_SHIFT
;
714 * If the numa=fake command-line contains a 'M' or 'G', it represents
715 * the fixed node size. Otherwise, if it is just a single number N,
716 * split the system RAM into N fake nodes.
718 if (strchr(cmdline
, 'M') || strchr(cmdline
, 'G')) {
721 size
= memparse(cmdline
, &cmdline
);
722 num_nodes
= split_nodes_size_interleave(addr
, max_addr
, size
);
726 n
= simple_strtoul(cmdline
, NULL
, 0);
727 num_nodes
= split_nodes_interleave(addr
, max_addr
, n
);
733 ei
.nr_blks
= num_nodes
;
734 for (i
= 0; i
< ei
.nr_blks
; i
++) {
735 ei
.blk
[i
].start
= nodes
[i
].start
;
736 ei
.blk
[i
].end
= nodes
[i
].end
;
740 memnode_shift
= compute_hash_shift(&ei
);
741 if (memnode_shift
< 0) {
743 printk(KERN_ERR
"No NUMA hash function found. NUMA emulation "
749 * We need to vacate all active ranges that may have been registered for
750 * the e820 memory map.
752 remove_all_active_ranges();
753 for_each_node_mask(i
, node_possible_map
)
754 memblock_x86_register_active_regions(i
, nodes
[i
].start
>> PAGE_SHIFT
,
755 nodes
[i
].end
>> PAGE_SHIFT
);
756 init_memory_mapping_high();
757 for_each_node_mask(i
, node_possible_map
)
758 setup_node_bootmem(i
, nodes
[i
].start
, nodes
[i
].end
);
759 setup_physnodes(addr
, max_addr
);
760 fake_physnodes(acpi
, amd
, num_nodes
);
764 #endif /* CONFIG_NUMA_EMU */
766 static int dummy_numa_init(void)
768 printk(KERN_INFO
"%s\n",
769 numa_off
? "NUMA turned off" : "No NUMA configuration found");
770 printk(KERN_INFO
"Faking a node at %016lx-%016lx\n",
771 0LU, max_pfn
<< PAGE_SHIFT
);
773 node_set(0, cpu_nodes_parsed
);
774 node_set(0, mem_nodes_parsed
);
775 numa_add_memblk(0, 0, (u64
)max_pfn
<< PAGE_SHIFT
);
776 numa_nodes
[0].start
= 0;
777 numa_nodes
[0].end
= (u64
)max_pfn
<< PAGE_SHIFT
;
782 void __init
initmem_init(void)
784 int (*numa_init
[])(void) = { [2] = dummy_numa_init
};
788 #ifdef CONFIG_ACPI_NUMA
789 numa_init
[0] = x86_acpi_numa_init
;
791 #ifdef CONFIG_AMD_NUMA
792 numa_init
[1] = amd_numa_init
;
796 for (i
= 0; i
< ARRAY_SIZE(numa_init
); i
++) {
800 for (j
= 0; j
< MAX_LOCAL_APIC
; j
++)
801 set_apicid_to_node(j
, NUMA_NO_NODE
);
803 nodes_clear(cpu_nodes_parsed
);
804 nodes_clear(mem_nodes_parsed
);
805 nodes_clear(node_possible_map
);
806 nodes_clear(node_online_map
);
807 memset(&numa_meminfo
, 0, sizeof(numa_meminfo
));
808 memset(numa_nodes
, 0, sizeof(numa_nodes
));
809 remove_all_active_ranges();
811 if (numa_init
[i
]() < 0)
814 /* clean up the node list */
815 for (j
= 0; j
< MAX_NUMNODES
; j
++)
816 cutoff_node(j
, 0, max_pfn
<< PAGE_SHIFT
);
818 #ifdef CONFIG_NUMA_EMU
819 setup_physnodes(0, max_pfn
<< PAGE_SHIFT
);
820 if (cmdline
&& !numa_emulation(0, max_pfn
, i
== 0, i
== 1))
822 setup_physnodes(0, max_pfn
<< PAGE_SHIFT
);
823 nodes_clear(node_possible_map
);
824 nodes_clear(node_online_map
);
826 /* Account for nodes with cpus and no memory */
827 nodes_or(node_possible_map
, mem_nodes_parsed
, cpu_nodes_parsed
);
828 if (WARN_ON(nodes_empty(node_possible_map
)))
831 if (numa_register_memblks() < 0)
834 for (j
= 0; j
< nr_cpu_ids
; j
++) {
835 int nid
= early_cpu_to_node(j
);
837 if (nid
== NUMA_NO_NODE
)
839 if (!node_online(nid
))
848 unsigned long __init
numa_free_all_bootmem(void)
850 unsigned long pages
= 0;
853 for_each_online_node(i
)
854 pages
+= free_all_bootmem_node(NODE_DATA(i
));
856 pages
+= free_all_memory_core_early(MAX_NUMNODES
);
861 int __cpuinit
numa_cpu_node(int cpu
)
863 int apicid
= early_per_cpu(x86_cpu_to_apicid
, cpu
);
865 if (apicid
!= BAD_APICID
)
866 return __apicid_to_node
[apicid
];
871 * UGLINESS AHEAD: Currently, CONFIG_NUMA_EMU is 64bit only and makes use
872 * of 64bit specific data structures. The distinction is artificial and
873 * should be removed. numa_{add|remove}_cpu() are implemented in numa.c
874 * for both 32 and 64bit when CONFIG_NUMA_EMU is disabled but here when
877 * NUMA emulation is planned to be made generic and the following and other
878 * related code should be moved to numa.c.
880 #ifdef CONFIG_NUMA_EMU
881 # ifndef CONFIG_DEBUG_PER_CPU_MAPS
882 void __cpuinit
numa_add_cpu(int cpu
)
887 nid
= numa_cpu_node(cpu
);
888 if (nid
== NUMA_NO_NODE
)
889 nid
= early_cpu_to_node(cpu
);
890 BUG_ON(nid
== NUMA_NO_NODE
|| !node_online(nid
));
893 * Use the starting address of the emulated node to find which physical
894 * node it is allocated on.
896 addr
= node_start_pfn(nid
) << PAGE_SHIFT
;
897 for (physnid
= 0; physnid
< MAX_NUMNODES
; physnid
++)
898 if (addr
>= physnodes
[physnid
].start
&&
899 addr
< physnodes
[physnid
].end
)
903 * Map the cpu to each emulated node that is allocated on the physical
904 * node of the cpu's apic id.
906 for_each_online_node(nid
) {
907 addr
= node_start_pfn(nid
) << PAGE_SHIFT
;
908 if (addr
>= physnodes
[physnid
].start
&&
909 addr
< physnodes
[physnid
].end
)
910 cpumask_set_cpu(cpu
, node_to_cpumask_map
[nid
]);
914 void __cpuinit
numa_remove_cpu(int cpu
)
918 for_each_online_node(i
)
919 cpumask_clear_cpu(cpu
, node_to_cpumask_map
[i
]);
921 # else /* !CONFIG_DEBUG_PER_CPU_MAPS */
922 static void __cpuinit
numa_set_cpumask(int cpu
, int enable
)
924 int node
= early_cpu_to_node(cpu
);
925 struct cpumask
*mask
;
928 if (node
== NUMA_NO_NODE
) {
929 /* early_cpu_to_node() already emits a warning and trace */
932 for_each_online_node(i
) {
935 addr
= node_start_pfn(i
) << PAGE_SHIFT
;
936 if (addr
< physnodes
[node
].start
||
937 addr
>= physnodes
[node
].end
)
939 mask
= debug_cpumask_set_cpu(cpu
, enable
);
944 cpumask_set_cpu(cpu
, mask
);
946 cpumask_clear_cpu(cpu
, mask
);
950 void __cpuinit
numa_add_cpu(int cpu
)
952 numa_set_cpumask(cpu
, 1);
955 void __cpuinit
numa_remove_cpu(int cpu
)
957 numa_set_cpumask(cpu
, 0);
959 # endif /* !CONFIG_DEBUG_PER_CPU_MAPS */
960 #endif /* CONFIG_NUMA_EMU */