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.
25 #include <linux/config.h>
27 #include <linux/bootmem.h>
28 #include <linux/mmzone.h>
29 #include <linux/highmem.h>
30 #include <linux/initrd.h>
31 #include <linux/nodemask.h>
32 #include <linux/module.h>
34 #include <asm/setup.h>
35 #include <asm/mmzone.h>
36 #include <bios_ebda.h>
38 struct pglist_data
*node_data
[MAX_NUMNODES
];
39 EXPORT_SYMBOL(node_data
);
40 bootmem_data_t node0_bdata
;
43 * numa interface - we expect the numa architecture specfic code to have
44 * populated the following initialisation.
46 * 1) node_online_map - the map of all nodes configured (online) in the system
47 * 2) node_start_pfn - the starting page frame number for a node
48 * 3) node_end_pfn - the ending page fram number for a node
50 unsigned long node_start_pfn
[MAX_NUMNODES
];
51 unsigned long node_end_pfn
[MAX_NUMNODES
];
54 #ifdef CONFIG_DISCONTIGMEM
56 * 4) physnode_map - the mapping between a pfn and owning node
57 * physnode_map keeps track of the physical memory layout of a generic
58 * numa node on a 256Mb break (each element of the array will
59 * represent 256Mb of memory and will be marked by the node id. so,
60 * if the first gig is on node 0, and the second gig is on node 1
61 * physnode_map will contain:
63 * physnode_map[0-3] = 0;
64 * physnode_map[4-7] = 1;
65 * physnode_map[8- ] = -1;
67 s8 physnode_map
[MAX_ELEMENTS
] = { [0 ... (MAX_ELEMENTS
- 1)] = -1};
68 EXPORT_SYMBOL(physnode_map
);
70 void memory_present(int nid
, unsigned long start
, unsigned long end
)
74 printk(KERN_INFO
"Node: %d, start_pfn: %ld, end_pfn: %ld\n",
76 printk(KERN_DEBUG
" Setting physnode_map array to node %d for pfns:\n", nid
);
77 printk(KERN_DEBUG
" ");
78 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_ELEMENT
) {
79 physnode_map
[pfn
/ PAGES_PER_ELEMENT
] = nid
;
85 unsigned long node_memmap_size_bytes(int nid
, unsigned long start_pfn
,
86 unsigned long end_pfn
)
88 unsigned long nr_pages
= end_pfn
- start_pfn
;
93 return (nr_pages
+ 1) * sizeof(struct page
);
97 extern unsigned long find_max_low_pfn(void);
98 extern void find_max_pfn(void);
99 extern void one_highpage_init(struct page
*, int, int);
101 extern struct e820map e820
;
102 extern unsigned long init_pg_tables_end
;
103 extern unsigned long highend_pfn
, highstart_pfn
;
104 extern unsigned long max_low_pfn
;
105 extern unsigned long totalram_pages
;
106 extern unsigned long totalhigh_pages
;
108 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
110 unsigned long node_remap_start_pfn
[MAX_NUMNODES
];
111 unsigned long node_remap_size
[MAX_NUMNODES
];
112 unsigned long node_remap_offset
[MAX_NUMNODES
];
113 void *node_remap_start_vaddr
[MAX_NUMNODES
];
114 void set_pmd_pfn(unsigned long vaddr
, unsigned long pfn
, pgprot_t flags
);
116 void *node_remap_end_vaddr
[MAX_NUMNODES
];
117 void *node_remap_alloc_vaddr
[MAX_NUMNODES
];
120 * FLAT - support for basic PC memory model with discontig enabled, essentially
121 * a single node with all available processors in it with a flat
124 int __init
get_memcfg_numa_flat(void)
126 printk("NUMA - single node, flat memory mode\n");
128 /* Run the memory configuration and find the top of memory. */
130 node_start_pfn
[0] = 0;
131 node_end_pfn
[0] = max_pfn
;
132 memory_present(0, 0, max_pfn
);
134 /* Indicate there is one node available. */
135 nodes_clear(node_online_map
);
141 * Find the highest page frame number we have available for the node
143 static void __init
find_max_pfn_node(int nid
)
145 if (node_end_pfn
[nid
] > max_pfn
)
146 node_end_pfn
[nid
] = max_pfn
;
148 * if a user has given mem=XXXX, then we need to make sure
149 * that the node _starts_ before that, too, not just ends
151 if (node_start_pfn
[nid
] > max_pfn
)
152 node_start_pfn
[nid
] = max_pfn
;
153 if (node_start_pfn
[nid
] > node_end_pfn
[nid
])
157 /* Find the owning node for a pfn. */
158 int early_pfn_to_nid(unsigned long pfn
)
163 if (node_end_pfn
[nid
] == 0)
165 if (node_start_pfn
[nid
] <= pfn
&& node_end_pfn
[nid
] >= pfn
)
173 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
174 * method. For node zero take this from the bottom of memory, for
175 * subsequent nodes place them at node_remap_start_vaddr which contains
176 * node local data in physically node local memory. See setup_memory()
179 static void __init
allocate_pgdat(int nid
)
181 if (nid
&& node_has_online_mem(nid
))
182 NODE_DATA(nid
) = (pg_data_t
*)node_remap_start_vaddr
[nid
];
184 NODE_DATA(nid
) = (pg_data_t
*)(__va(min_low_pfn
<< PAGE_SHIFT
));
185 min_low_pfn
+= PFN_UP(sizeof(pg_data_t
));
189 void *alloc_remap(int nid
, unsigned long size
)
191 void *allocation
= node_remap_alloc_vaddr
[nid
];
193 size
= ALIGN(size
, L1_CACHE_BYTES
);
195 if (!allocation
|| (allocation
+ size
) >= node_remap_end_vaddr
[nid
])
198 node_remap_alloc_vaddr
[nid
] += size
;
199 memset(allocation
, 0, size
);
204 void __init
remap_numa_kva(void)
210 for_each_online_node(node
) {
211 for (pfn
=0; pfn
< node_remap_size
[node
]; pfn
+= PTRS_PER_PTE
) {
212 vaddr
= node_remap_start_vaddr
[node
]+(pfn
<<PAGE_SHIFT
);
213 set_pmd_pfn((ulong
) vaddr
,
214 node_remap_start_pfn
[node
] + pfn
,
220 static unsigned long calculate_numa_remap_pages(void)
223 unsigned long size
, reserve_pages
= 0;
226 for_each_online_node(nid
) {
228 * The acpi/srat node info can show hot-add memroy zones
229 * where memory could be added but not currently present.
231 if (node_start_pfn
[nid
] > max_pfn
)
233 if (node_end_pfn
[nid
] > max_pfn
)
234 node_end_pfn
[nid
] = max_pfn
;
236 /* ensure the remap includes space for the pgdat. */
237 size
= node_remap_size
[nid
] + sizeof(pg_data_t
);
239 /* convert size to large (pmd size) pages, rounding up */
240 size
= (size
+ LARGE_PAGE_BYTES
- 1) / LARGE_PAGE_BYTES
;
241 /* now the roundup is correct, convert to PAGE_SIZE pages */
242 size
= size
* PTRS_PER_PTE
;
245 * Validate the region we are allocating only contains valid
248 for (pfn
= node_end_pfn
[nid
] - size
;
249 pfn
< node_end_pfn
[nid
]; pfn
++)
250 if (!page_is_ram(pfn
))
253 if (pfn
!= node_end_pfn
[nid
])
256 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
258 node_remap_size
[nid
] = size
;
259 node_remap_offset
[nid
] = reserve_pages
;
260 reserve_pages
+= size
;
261 printk("Shrinking node %d from %ld pages to %ld pages\n",
262 nid
, node_end_pfn
[nid
], node_end_pfn
[nid
] - size
);
263 node_end_pfn
[nid
] -= size
;
264 node_remap_start_pfn
[nid
] = node_end_pfn
[nid
];
266 printk("Reserving total of %ld pages for numa KVA remap\n",
268 return reserve_pages
;
271 extern void setup_bootmem_allocator(void);
272 unsigned long __init
setup_memory(void)
275 unsigned long system_start_pfn
, system_max_low_pfn
;
276 unsigned long reserve_pages
;
279 * When mapping a NUMA machine we allocate the node_mem_map arrays
280 * from node local memory. They are then mapped directly into KVA
281 * between zone normal and vmalloc space. Calculate the size of
282 * this space and use it to adjust the boundry between ZONE_NORMAL
288 reserve_pages
= calculate_numa_remap_pages();
290 /* partially used pages are not usable - thus round upwards */
291 system_start_pfn
= min_low_pfn
= PFN_UP(init_pg_tables_end
);
293 system_max_low_pfn
= max_low_pfn
= find_max_low_pfn() - reserve_pages
;
294 printk("reserve_pages = %ld find_max_low_pfn() ~ %ld\n",
295 reserve_pages
, max_low_pfn
+ reserve_pages
);
296 printk("max_pfn = %ld\n", max_pfn
);
297 #ifdef CONFIG_HIGHMEM
298 highstart_pfn
= highend_pfn
= max_pfn
;
299 if (max_pfn
> system_max_low_pfn
)
300 highstart_pfn
= system_max_low_pfn
;
301 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
302 pages_to_mb(highend_pfn
- highstart_pfn
));
304 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
305 pages_to_mb(system_max_low_pfn
));
306 printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
307 min_low_pfn
, max_low_pfn
, highstart_pfn
);
309 printk("Low memory ends at vaddr %08lx\n",
310 (ulong
) pfn_to_kaddr(max_low_pfn
));
311 for_each_online_node(nid
) {
312 node_remap_start_vaddr
[nid
] = pfn_to_kaddr(
313 highstart_pfn
+ node_remap_offset
[nid
]);
314 /* Init the node remap allocator */
315 node_remap_end_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
316 (node_remap_size
[nid
] * PAGE_SIZE
);
317 node_remap_alloc_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
318 ALIGN(sizeof(pg_data_t
), PAGE_SIZE
);
321 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid
,
322 (ulong
) node_remap_start_vaddr
[nid
],
323 (ulong
) pfn_to_kaddr(highstart_pfn
324 + node_remap_offset
[nid
] + node_remap_size
[nid
]));
326 printk("High memory starts at vaddr %08lx\n",
327 (ulong
) pfn_to_kaddr(highstart_pfn
));
328 vmalloc_earlyreserve
= reserve_pages
* PAGE_SIZE
;
329 for_each_online_node(nid
)
330 find_max_pfn_node(nid
);
332 memset(NODE_DATA(0), 0, sizeof(struct pglist_data
));
333 NODE_DATA(0)->bdata
= &node0_bdata
;
334 setup_bootmem_allocator();
338 void __init
zone_sizes_init(void)
343 * Insert nodes into pgdat_list backward so they appear in order.
344 * Clobber node 0's links and NULL out pgdat_list before starting.
347 for (nid
= MAX_NUMNODES
- 1; nid
>= 0; nid
--) {
348 if (!node_online(nid
))
350 NODE_DATA(nid
)->pgdat_next
= pgdat_list
;
351 pgdat_list
= NODE_DATA(nid
);
354 for_each_online_node(nid
) {
355 unsigned long zones_size
[MAX_NR_ZONES
] = {0, 0, 0};
356 unsigned long *zholes_size
;
357 unsigned int max_dma
;
359 unsigned long low
= max_low_pfn
;
360 unsigned long start
= node_start_pfn
[nid
];
361 unsigned long high
= node_end_pfn
[nid
];
363 max_dma
= virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
365 if (node_has_online_mem(nid
)){
367 #ifdef CONFIG_HIGHMEM
368 BUG_ON(start
> high
);
369 zones_size
[ZONE_HIGHMEM
] = high
- start
;
373 zones_size
[ZONE_DMA
] = low
;
375 BUG_ON(max_dma
> low
);
377 zones_size
[ZONE_DMA
] = max_dma
;
378 zones_size
[ZONE_NORMAL
] = low
- max_dma
;
379 #ifdef CONFIG_HIGHMEM
380 zones_size
[ZONE_HIGHMEM
] = high
- low
;
386 zholes_size
= get_zholes_size(nid
);
388 free_area_init_node(nid
, NODE_DATA(nid
), zones_size
, start
,
394 void __init
set_highmem_pages_init(int bad_ppro
)
396 #ifdef CONFIG_HIGHMEM
400 for_each_zone(zone
) {
401 unsigned long node_pfn
, zone_start_pfn
, zone_end_pfn
;
403 if (!is_highmem(zone
))
406 zone_start_pfn
= zone
->zone_start_pfn
;
407 zone_end_pfn
= zone_start_pfn
+ zone
->spanned_pages
;
409 printk("Initializing %s for node %d (%08lx:%08lx)\n",
410 zone
->name
, zone
->zone_pgdat
->node_id
,
411 zone_start_pfn
, zone_end_pfn
);
413 for (node_pfn
= zone_start_pfn
; node_pfn
< zone_end_pfn
; node_pfn
++) {
414 if (!pfn_valid(node_pfn
))
416 page
= pfn_to_page(node_pfn
);
417 one_highpage_init(page
, node_pfn
, bad_ppro
);
420 totalram_pages
+= totalhigh_pages
;