2 * Written by: Patricia Gaughen <gone@us.ibm.com>, IBM Corporation
3 * August 2002: added remote node KVA remap - Martin J. Bligh
5 * Copyright (C) 2002, IBM Corp.
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
17 * NON INFRINGEMENT. See the GNU General Public License for more
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/bootmem.h>
27 #include <linux/mmzone.h>
28 #include <linux/highmem.h>
29 #include <linux/initrd.h>
30 #include <linux/nodemask.h>
31 #include <linux/module.h>
32 #include <linux/kexec.h>
33 #include <linux/pfn.h>
34 #include <linux/swap.h>
35 #include <linux/acpi.h>
38 #include <asm/setup.h>
39 #include <asm/mmzone.h>
40 #include <asm/bios_ebda.h>
41 #include <asm/proto.h>
43 struct pglist_data
*node_data
[MAX_NUMNODES
] __read_mostly
;
44 EXPORT_SYMBOL(node_data
);
47 * numa interface - we expect the numa architecture specific code to have
48 * populated the following initialisation.
50 * 1) node_online_map - the map of all nodes configured (online) in the system
51 * 2) node_start_pfn - the starting page frame number for a node
52 * 3) node_end_pfn - the ending page fram number for a node
54 unsigned long node_start_pfn
[MAX_NUMNODES
] __read_mostly
;
55 unsigned long node_end_pfn
[MAX_NUMNODES
] __read_mostly
;
58 #ifdef CONFIG_DISCONTIGMEM
60 * 4) physnode_map - the mapping between a pfn and owning node
61 * physnode_map keeps track of the physical memory layout of a generic
62 * numa node on a 64Mb break (each element of the array will
63 * represent 64Mb of memory and will be marked by the node id. so,
64 * if the first gig is on node 0, and the second gig is on node 1
65 * physnode_map will contain:
67 * physnode_map[0-15] = 0;
68 * physnode_map[16-31] = 1;
69 * physnode_map[32- ] = -1;
71 s8 physnode_map
[MAX_ELEMENTS
] __read_mostly
= { [0 ... (MAX_ELEMENTS
- 1)] = -1};
72 EXPORT_SYMBOL(physnode_map
);
74 void memory_present(int nid
, unsigned long start
, unsigned long end
)
78 printk(KERN_INFO
"Node: %d, start_pfn: %lx, end_pfn: %lx\n",
80 printk(KERN_DEBUG
" Setting physnode_map array to node %d for pfns:\n", nid
);
81 printk(KERN_DEBUG
" ");
82 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_ELEMENT
) {
83 physnode_map
[pfn
/ PAGES_PER_ELEMENT
] = nid
;
84 printk(KERN_CONT
"%lx ", pfn
);
86 printk(KERN_CONT
"\n");
89 unsigned long node_memmap_size_bytes(int nid
, unsigned long start_pfn
,
90 unsigned long end_pfn
)
92 unsigned long nr_pages
= end_pfn
- start_pfn
;
97 return (nr_pages
+ 1) * sizeof(struct page
);
101 extern unsigned long find_max_low_pfn(void);
102 extern unsigned long highend_pfn
, highstart_pfn
;
104 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
106 unsigned long node_remap_size
[MAX_NUMNODES
];
107 static void *node_remap_start_vaddr
[MAX_NUMNODES
];
108 void set_pmd_pfn(unsigned long vaddr
, unsigned long pfn
, pgprot_t flags
);
110 static unsigned long kva_start_pfn
;
111 static unsigned long kva_pages
;
113 * FLAT - support for basic PC memory model with discontig enabled, essentially
114 * a single node with all available processors in it with a flat
117 int __init
get_memcfg_numa_flat(void)
119 printk(KERN_DEBUG
"NUMA - single node, flat memory mode\n");
121 node_start_pfn
[0] = 0;
122 node_end_pfn
[0] = max_pfn
;
123 e820_register_active_regions(0, 0, max_pfn
);
124 memory_present(0, 0, max_pfn
);
125 node_remap_size
[0] = node_memmap_size_bytes(0, 0, max_pfn
);
127 /* Indicate there is one node available. */
128 nodes_clear(node_online_map
);
134 * Find the highest page frame number we have available for the node
136 static void __init
propagate_e820_map_node(int nid
)
138 if (node_end_pfn
[nid
] > max_pfn
)
139 node_end_pfn
[nid
] = max_pfn
;
141 * if a user has given mem=XXXX, then we need to make sure
142 * that the node _starts_ before that, too, not just ends
144 if (node_start_pfn
[nid
] > max_pfn
)
145 node_start_pfn
[nid
] = max_pfn
;
146 BUG_ON(node_start_pfn
[nid
] > node_end_pfn
[nid
]);
150 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
151 * method. For node zero take this from the bottom of memory, for
152 * subsequent nodes place them at node_remap_start_vaddr which contains
153 * node local data in physically node local memory. See setup_memory()
156 static void __init
allocate_pgdat(int nid
)
160 if (node_has_online_mem(nid
) && node_remap_start_vaddr
[nid
])
161 NODE_DATA(nid
) = (pg_data_t
*)node_remap_start_vaddr
[nid
];
163 unsigned long pgdat_phys
;
164 pgdat_phys
= find_e820_area(min_low_pfn
<<PAGE_SHIFT
,
165 max_pfn_mapped
<<PAGE_SHIFT
,
168 NODE_DATA(nid
) = (pg_data_t
*)(pfn_to_kaddr(pgdat_phys
>>PAGE_SHIFT
));
169 memset(buf
, 0, sizeof(buf
));
170 sprintf(buf
, "NODE_DATA %d", nid
);
171 reserve_early(pgdat_phys
, pgdat_phys
+ sizeof(pg_data_t
), buf
);
173 printk(KERN_DEBUG
"allocate_pgdat: node %d NODE_DATA %08lx\n",
174 nid
, (unsigned long)NODE_DATA(nid
));
178 * In the DISCONTIGMEM and SPARSEMEM memory model, a portion of the kernel
179 * virtual address space (KVA) is reserved and portions of nodes are mapped
180 * using it. This is to allow node-local memory to be allocated for
181 * structures that would normally require ZONE_NORMAL. The memory is
182 * allocated with alloc_remap() and callers should be prepared to allocate
183 * from the bootmem allocator instead.
185 static unsigned long node_remap_start_pfn
[MAX_NUMNODES
];
186 static void *node_remap_end_vaddr
[MAX_NUMNODES
];
187 static void *node_remap_alloc_vaddr
[MAX_NUMNODES
];
188 static unsigned long node_remap_offset
[MAX_NUMNODES
];
190 void *alloc_remap(int nid
, unsigned long size
)
192 void *allocation
= node_remap_alloc_vaddr
[nid
];
194 size
= ALIGN(size
, L1_CACHE_BYTES
);
196 if (!allocation
|| (allocation
+ size
) >= node_remap_end_vaddr
[nid
])
199 node_remap_alloc_vaddr
[nid
] += size
;
200 memset(allocation
, 0, size
);
205 static void __init
remap_numa_kva(void)
211 for_each_online_node(node
) {
212 printk(KERN_DEBUG
"remap_numa_kva: node %d\n", node
);
213 for (pfn
=0; pfn
< node_remap_size
[node
]; pfn
+= PTRS_PER_PTE
) {
214 vaddr
= node_remap_start_vaddr
[node
]+(pfn
<<PAGE_SHIFT
);
215 printk(KERN_DEBUG
"remap_numa_kva: %08lx to pfn %08lx\n",
216 (unsigned long)vaddr
,
217 node_remap_start_pfn
[node
] + pfn
);
218 set_pmd_pfn((ulong
) vaddr
,
219 node_remap_start_pfn
[node
] + pfn
,
225 static unsigned long calculate_numa_remap_pages(void)
228 unsigned long size
, reserve_pages
= 0;
230 for_each_online_node(nid
) {
235 * The acpi/srat node info can show hot-add memroy zones
236 * where memory could be added but not currently present.
238 printk(KERN_DEBUG
"node %d pfn: [%lx - %lx]\n",
239 nid
, node_start_pfn
[nid
], node_end_pfn
[nid
]);
240 if (node_start_pfn
[nid
] > max_pfn
)
242 if (!node_end_pfn
[nid
])
244 if (node_end_pfn
[nid
] > max_pfn
)
245 node_end_pfn
[nid
] = max_pfn
;
247 /* ensure the remap includes space for the pgdat. */
248 size
= node_remap_size
[nid
] + sizeof(pg_data_t
);
250 /* convert size to large (pmd size) pages, rounding up */
251 size
= (size
+ LARGE_PAGE_BYTES
- 1) / LARGE_PAGE_BYTES
;
252 /* now the roundup is correct, convert to PAGE_SIZE pages */
253 size
= size
* PTRS_PER_PTE
;
255 node_kva_target
= round_down(node_end_pfn
[nid
] - size
,
257 node_kva_target
<<= PAGE_SHIFT
;
259 node_kva_final
= find_e820_area(node_kva_target
,
260 ((u64
)node_end_pfn
[nid
])<<PAGE_SHIFT
,
261 ((u64
)size
)<<PAGE_SHIFT
,
263 node_kva_target
-= LARGE_PAGE_BYTES
;
264 } while (node_kva_final
== -1ULL &&
265 (node_kva_target
>>PAGE_SHIFT
) > (node_start_pfn
[nid
]));
267 if (node_kva_final
== -1ULL)
268 panic("Can not get kva ram\n");
270 node_remap_size
[nid
] = size
;
271 node_remap_offset
[nid
] = reserve_pages
;
272 reserve_pages
+= size
;
273 printk(KERN_DEBUG
"Reserving %ld pages of KVA for lmem_map of"
274 " node %d at %llx\n",
275 size
, nid
, node_kva_final
>>PAGE_SHIFT
);
278 * prevent kva address below max_low_pfn want it on system
279 * with less memory later.
280 * layout will be: KVA address , KVA RAM
282 * we are supposed to only record the one less then max_low_pfn
283 * but we could have some hole in high memory, and it will only
284 * check page_is_ram(pfn) && !page_is_reserved_early(pfn) to decide
286 * So reserve_early here, hope we don't run out of that array
288 reserve_early(node_kva_final
,
289 node_kva_final
+(((u64
)size
)<<PAGE_SHIFT
),
292 node_remap_start_pfn
[nid
] = node_kva_final
>>PAGE_SHIFT
;
293 remove_active_range(nid
, node_remap_start_pfn
[nid
],
294 node_remap_start_pfn
[nid
] + size
);
296 printk(KERN_INFO
"Reserving total of %lx pages for numa KVA remap\n",
298 return reserve_pages
;
301 static void init_remap_allocator(int nid
)
303 node_remap_start_vaddr
[nid
] = pfn_to_kaddr(
304 kva_start_pfn
+ node_remap_offset
[nid
]);
305 node_remap_end_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
306 (node_remap_size
[nid
] * PAGE_SIZE
);
307 node_remap_alloc_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
308 ALIGN(sizeof(pg_data_t
), PAGE_SIZE
);
310 printk(KERN_DEBUG
"node %d will remap to vaddr %08lx - %08lx\n", nid
,
311 (ulong
) node_remap_start_vaddr
[nid
],
312 (ulong
) node_remap_end_vaddr
[nid
]);
315 void __init
initmem_init(unsigned long start_pfn
,
316 unsigned long end_pfn
)
322 * When mapping a NUMA machine we allocate the node_mem_map arrays
323 * from node local memory. They are then mapped directly into KVA
324 * between zone normal and vmalloc space. Calculate the size of
325 * this space and use it to adjust the boundary between ZONE_NORMAL
331 kva_pages
= roundup(calculate_numa_remap_pages(), PTRS_PER_PTE
);
333 kva_target_pfn
= round_down(max_low_pfn
- kva_pages
, PTRS_PER_PTE
);
335 kva_start_pfn
= find_e820_area(kva_target_pfn
<<PAGE_SHIFT
,
336 max_low_pfn
<<PAGE_SHIFT
,
337 kva_pages
<<PAGE_SHIFT
,
338 PTRS_PER_PTE
<<PAGE_SHIFT
) >> PAGE_SHIFT
;
339 kva_target_pfn
-= PTRS_PER_PTE
;
340 } while (kva_start_pfn
== -1UL && kva_target_pfn
> min_low_pfn
);
342 if (kva_start_pfn
== -1UL)
343 panic("Can not get kva space\n");
345 printk(KERN_INFO
"kva_start_pfn ~ %lx max_low_pfn ~ %lx\n",
346 kva_start_pfn
, max_low_pfn
);
347 printk(KERN_INFO
"max_pfn = %lx\n", max_pfn
);
349 /* avoid clash with initrd */
350 reserve_early(kva_start_pfn
<<PAGE_SHIFT
,
351 (kva_start_pfn
+ kva_pages
)<<PAGE_SHIFT
,
353 #ifdef CONFIG_HIGHMEM
354 highstart_pfn
= highend_pfn
= max_pfn
;
355 if (max_pfn
> max_low_pfn
)
356 highstart_pfn
= max_low_pfn
;
357 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
358 pages_to_mb(highend_pfn
- highstart_pfn
));
359 num_physpages
= highend_pfn
;
360 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
362 num_physpages
= max_low_pfn
;
363 high_memory
= (void *) __va(max_low_pfn
* PAGE_SIZE
- 1) + 1;
365 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
366 pages_to_mb(max_low_pfn
));
367 printk(KERN_DEBUG
"max_low_pfn = %lx, highstart_pfn = %lx\n",
368 max_low_pfn
, highstart_pfn
);
370 printk(KERN_DEBUG
"Low memory ends at vaddr %08lx\n",
371 (ulong
) pfn_to_kaddr(max_low_pfn
));
372 for_each_online_node(nid
) {
373 init_remap_allocator(nid
);
379 printk(KERN_DEBUG
"High memory starts at vaddr %08lx\n",
380 (ulong
) pfn_to_kaddr(highstart_pfn
));
381 for_each_online_node(nid
)
382 propagate_e820_map_node(nid
);
384 for_each_online_node(nid
)
385 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
387 NODE_DATA(0)->bdata
= &bootmem_node_data
[0];
388 setup_bootmem_allocator();
391 void __init
set_highmem_pages_init(void)
393 #ifdef CONFIG_HIGHMEM
397 for_each_zone(zone
) {
398 unsigned long zone_start_pfn
, zone_end_pfn
;
400 if (!is_highmem(zone
))
403 zone_start_pfn
= zone
->zone_start_pfn
;
404 zone_end_pfn
= zone_start_pfn
+ zone
->spanned_pages
;
406 nid
= zone_to_nid(zone
);
407 printk(KERN_INFO
"Initializing %s for node %d (%08lx:%08lx)\n",
408 zone
->name
, nid
, zone_start_pfn
, zone_end_pfn
);
410 add_highpages_with_active_regions(nid
, zone_start_pfn
,
413 totalram_pages
+= totalhigh_pages
;
417 #ifdef CONFIG_MEMORY_HOTPLUG
418 static int paddr_to_nid(u64 addr
)
421 unsigned long pfn
= PFN_DOWN(addr
);
424 if (node_start_pfn
[nid
] <= pfn
&&
425 pfn
< node_end_pfn
[nid
])
432 * This function is used to ask node id BEFORE memmap and mem_section's
433 * initialization (pfn_to_nid() can't be used yet).
434 * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
436 int memory_add_physaddr_to_nid(u64 addr
)
438 int nid
= paddr_to_nid(addr
);
439 return (nid
>= 0) ? nid
: 0;
442 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);