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>
37 #include <asm/setup.h>
38 #include <asm/mmzone.h>
39 #include <bios_ebda.h>
41 struct pglist_data
*node_data
[MAX_NUMNODES
] __read_mostly
;
42 EXPORT_SYMBOL(node_data
);
43 static bootmem_data_t node0_bdata
;
46 * numa interface - we expect the numa architecture specific code to have
47 * populated the following initialisation.
49 * 1) node_online_map - the map of all nodes configured (online) in the system
50 * 2) node_start_pfn - the starting page frame number for a node
51 * 3) node_end_pfn - the ending page fram number for a node
53 unsigned long node_start_pfn
[MAX_NUMNODES
] __read_mostly
;
54 unsigned long node_end_pfn
[MAX_NUMNODES
] __read_mostly
;
57 #ifdef CONFIG_DISCONTIGMEM
59 * 4) physnode_map - the mapping between a pfn and owning node
60 * physnode_map keeps track of the physical memory layout of a generic
61 * numa node on a 256Mb break (each element of the array will
62 * represent 256Mb of memory and will be marked by the node id. so,
63 * if the first gig is on node 0, and the second gig is on node 1
64 * physnode_map will contain:
66 * physnode_map[0-3] = 0;
67 * physnode_map[4-7] = 1;
68 * physnode_map[8- ] = -1;
70 s8 physnode_map
[MAX_ELEMENTS
] __read_mostly
= { [0 ... (MAX_ELEMENTS
- 1)] = -1};
71 EXPORT_SYMBOL(physnode_map
);
73 void memory_present(int nid
, unsigned long start
, unsigned long end
)
77 printk(KERN_INFO
"Node: %d, start_pfn: %ld, end_pfn: %ld\n",
79 printk(KERN_DEBUG
" Setting physnode_map array to node %d for pfns:\n", nid
);
80 printk(KERN_DEBUG
" ");
81 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_ELEMENT
) {
82 physnode_map
[pfn
/ PAGES_PER_ELEMENT
] = nid
;
88 unsigned long node_memmap_size_bytes(int nid
, unsigned long start_pfn
,
89 unsigned long end_pfn
)
91 unsigned long nr_pages
= end_pfn
- start_pfn
;
96 return (nr_pages
+ 1) * sizeof(struct page
);
100 extern unsigned long find_max_low_pfn(void);
101 extern void add_one_highpage_init(struct page
*, int, int);
102 extern unsigned long highend_pfn
, highstart_pfn
;
104 #define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
106 static unsigned long node_remap_start_pfn
[MAX_NUMNODES
];
107 unsigned long node_remap_size
[MAX_NUMNODES
];
108 static unsigned long node_remap_offset
[MAX_NUMNODES
];
109 static void *node_remap_start_vaddr
[MAX_NUMNODES
];
110 void set_pmd_pfn(unsigned long vaddr
, unsigned long pfn
, pgprot_t flags
);
112 static void *node_remap_end_vaddr
[MAX_NUMNODES
];
113 static void *node_remap_alloc_vaddr
[MAX_NUMNODES
];
114 static unsigned long kva_start_pfn
;
115 static unsigned long kva_pages
;
117 * FLAT - support for basic PC memory model with discontig enabled, essentially
118 * a single node with all available processors in it with a flat
121 int __init
get_memcfg_numa_flat(void)
123 printk("NUMA - single node, flat memory mode\n");
125 /* Run the memory configuration and find the top of memory. */
127 node_start_pfn
[0] = 0;
128 node_end_pfn
[0] = max_pfn
;
129 memory_present(0, 0, max_pfn
);
131 /* Indicate there is one node available. */
132 nodes_clear(node_online_map
);
138 * Find the highest page frame number we have available for the node
140 static void __init
find_max_pfn_node(int nid
)
142 if (node_end_pfn
[nid
] > max_pfn
)
143 node_end_pfn
[nid
] = max_pfn
;
145 * if a user has given mem=XXXX, then we need to make sure
146 * that the node _starts_ before that, too, not just ends
148 if (node_start_pfn
[nid
] > max_pfn
)
149 node_start_pfn
[nid
] = max_pfn
;
150 BUG_ON(node_start_pfn
[nid
] > node_end_pfn
[nid
]);
154 * Allocate memory for the pg_data_t for this node via a crude pre-bootmem
155 * method. For node zero take this from the bottom of memory, for
156 * subsequent nodes place them at node_remap_start_vaddr which contains
157 * node local data in physically node local memory. See setup_memory()
160 static void __init
allocate_pgdat(int nid
)
162 if (nid
&& node_has_online_mem(nid
))
163 NODE_DATA(nid
) = (pg_data_t
*)node_remap_start_vaddr
[nid
];
165 NODE_DATA(nid
) = (pg_data_t
*)(pfn_to_kaddr(min_low_pfn
));
166 min_low_pfn
+= PFN_UP(sizeof(pg_data_t
));
170 void *alloc_remap(int nid
, unsigned long size
)
172 void *allocation
= node_remap_alloc_vaddr
[nid
];
174 size
= ALIGN(size
, L1_CACHE_BYTES
);
176 if (!allocation
|| (allocation
+ size
) >= node_remap_end_vaddr
[nid
])
179 node_remap_alloc_vaddr
[nid
] += size
;
180 memset(allocation
, 0, size
);
185 void __init
remap_numa_kva(void)
191 for_each_online_node(node
) {
192 for (pfn
=0; pfn
< node_remap_size
[node
]; pfn
+= PTRS_PER_PTE
) {
193 vaddr
= node_remap_start_vaddr
[node
]+(pfn
<<PAGE_SHIFT
);
194 set_pmd_pfn((ulong
) vaddr
,
195 node_remap_start_pfn
[node
] + pfn
,
201 static unsigned long calculate_numa_remap_pages(void)
204 unsigned long size
, reserve_pages
= 0;
207 for_each_online_node(nid
) {
208 unsigned old_end_pfn
= node_end_pfn
[nid
];
211 * The acpi/srat node info can show hot-add memroy zones
212 * where memory could be added but not currently present.
214 if (node_start_pfn
[nid
] > max_pfn
)
216 if (node_end_pfn
[nid
] > max_pfn
)
217 node_end_pfn
[nid
] = max_pfn
;
219 /* ensure the remap includes space for the pgdat. */
220 size
= node_remap_size
[nid
] + sizeof(pg_data_t
);
222 /* convert size to large (pmd size) pages, rounding up */
223 size
= (size
+ LARGE_PAGE_BYTES
- 1) / LARGE_PAGE_BYTES
;
224 /* now the roundup is correct, convert to PAGE_SIZE pages */
225 size
= size
* PTRS_PER_PTE
;
228 * Validate the region we are allocating only contains valid
231 for (pfn
= node_end_pfn
[nid
] - size
;
232 pfn
< node_end_pfn
[nid
]; pfn
++)
233 if (!page_is_ram(pfn
))
236 if (pfn
!= node_end_pfn
[nid
])
239 printk("Reserving %ld pages of KVA for lmem_map of node %d\n",
241 node_remap_size
[nid
] = size
;
242 node_remap_offset
[nid
] = reserve_pages
;
243 reserve_pages
+= size
;
244 printk("Shrinking node %d from %ld pages to %ld pages\n",
245 nid
, node_end_pfn
[nid
], node_end_pfn
[nid
] - size
);
247 if (node_end_pfn
[nid
] & (PTRS_PER_PTE
-1)) {
249 * Align node_end_pfn[] and node_remap_start_pfn[] to
250 * pmd boundary. remap_numa_kva will barf otherwise.
252 printk("Shrinking node %d further by %ld pages for proper alignment\n",
253 nid
, node_end_pfn
[nid
] & (PTRS_PER_PTE
-1));
254 size
+= node_end_pfn
[nid
] & (PTRS_PER_PTE
-1);
257 node_end_pfn
[nid
] -= size
;
258 node_remap_start_pfn
[nid
] = node_end_pfn
[nid
];
259 shrink_active_range(nid
, old_end_pfn
, node_end_pfn
[nid
]);
261 printk("Reserving total of %ld pages for numa KVA remap\n",
263 return reserve_pages
;
266 extern void setup_bootmem_allocator(void);
267 unsigned long __init
setup_memory(void)
270 unsigned long system_start_pfn
, system_max_low_pfn
;
273 * When mapping a NUMA machine we allocate the node_mem_map arrays
274 * from node local memory. They are then mapped directly into KVA
275 * between zone normal and vmalloc space. Calculate the size of
276 * this space and use it to adjust the boundary between ZONE_NORMAL
282 kva_pages
= calculate_numa_remap_pages();
284 /* partially used pages are not usable - thus round upwards */
285 system_start_pfn
= min_low_pfn
= PFN_UP(init_pg_tables_end
);
287 kva_start_pfn
= find_max_low_pfn() - kva_pages
;
289 #ifdef CONFIG_BLK_DEV_INITRD
290 /* Numa kva area is below the initrd */
291 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
)
292 kva_start_pfn
= PFN_DOWN(boot_params
.hdr
.ramdisk_image
)
295 kva_start_pfn
-= kva_start_pfn
& (PTRS_PER_PTE
-1);
297 system_max_low_pfn
= max_low_pfn
= find_max_low_pfn();
298 printk("kva_start_pfn ~ %ld find_max_low_pfn() ~ %ld\n",
299 kva_start_pfn
, max_low_pfn
);
300 printk("max_pfn = %ld\n", max_pfn
);
301 #ifdef CONFIG_HIGHMEM
302 highstart_pfn
= highend_pfn
= max_pfn
;
303 if (max_pfn
> system_max_low_pfn
)
304 highstart_pfn
= system_max_low_pfn
;
305 printk(KERN_NOTICE
"%ldMB HIGHMEM available.\n",
306 pages_to_mb(highend_pfn
- highstart_pfn
));
307 num_physpages
= highend_pfn
;
308 high_memory
= (void *) __va(highstart_pfn
* PAGE_SIZE
- 1) + 1;
310 num_physpages
= system_max_low_pfn
;
311 high_memory
= (void *) __va(system_max_low_pfn
* PAGE_SIZE
- 1) + 1;
313 printk(KERN_NOTICE
"%ldMB LOWMEM available.\n",
314 pages_to_mb(system_max_low_pfn
));
315 printk("min_low_pfn = %ld, max_low_pfn = %ld, highstart_pfn = %ld\n",
316 min_low_pfn
, max_low_pfn
, highstart_pfn
);
318 printk("Low memory ends at vaddr %08lx\n",
319 (ulong
) pfn_to_kaddr(max_low_pfn
));
320 for_each_online_node(nid
) {
321 node_remap_start_vaddr
[nid
] = pfn_to_kaddr(
322 kva_start_pfn
+ node_remap_offset
[nid
]);
323 /* Init the node remap allocator */
324 node_remap_end_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
325 (node_remap_size
[nid
] * PAGE_SIZE
);
326 node_remap_alloc_vaddr
[nid
] = node_remap_start_vaddr
[nid
] +
327 ALIGN(sizeof(pg_data_t
), PAGE_SIZE
);
330 printk ("node %d will remap to vaddr %08lx - %08lx\n", nid
,
331 (ulong
) node_remap_start_vaddr
[nid
],
332 (ulong
) pfn_to_kaddr(highstart_pfn
333 + node_remap_offset
[nid
] + node_remap_size
[nid
]));
335 printk("High memory starts at vaddr %08lx\n",
336 (ulong
) pfn_to_kaddr(highstart_pfn
));
337 for_each_online_node(nid
)
338 find_max_pfn_node(nid
);
340 memset(NODE_DATA(0), 0, sizeof(struct pglist_data
));
341 NODE_DATA(0)->bdata
= &node0_bdata
;
342 setup_bootmem_allocator();
346 void __init
numa_kva_reserve(void)
348 reserve_bootmem(PFN_PHYS(kva_start_pfn
),PFN_PHYS(kva_pages
));
351 void __init
zone_sizes_init(void)
354 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
355 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
356 max_zone_pfns
[ZONE_DMA
] =
357 virt_to_phys((char *)MAX_DMA_ADDRESS
) >> PAGE_SHIFT
;
358 max_zone_pfns
[ZONE_NORMAL
] = max_low_pfn
;
359 #ifdef CONFIG_HIGHMEM
360 max_zone_pfns
[ZONE_HIGHMEM
] = highend_pfn
;
363 /* If SRAT has not registered memory, register it now */
364 if (find_max_pfn_with_active_regions() == 0) {
365 for_each_online_node(nid
) {
366 if (node_has_online_mem(nid
))
367 add_active_range(nid
, node_start_pfn
[nid
],
372 free_area_init_nodes(max_zone_pfns
);
376 void __init
set_highmem_pages_init(int bad_ppro
)
378 #ifdef CONFIG_HIGHMEM
382 for_each_zone(zone
) {
383 unsigned long node_pfn
, zone_start_pfn
, zone_end_pfn
;
385 if (!is_highmem(zone
))
388 zone_start_pfn
= zone
->zone_start_pfn
;
389 zone_end_pfn
= zone_start_pfn
+ zone
->spanned_pages
;
391 printk("Initializing %s for node %d (%08lx:%08lx)\n",
392 zone
->name
, zone_to_nid(zone
),
393 zone_start_pfn
, zone_end_pfn
);
395 for (node_pfn
= zone_start_pfn
; node_pfn
< zone_end_pfn
; node_pfn
++) {
396 if (!pfn_valid(node_pfn
))
398 page
= pfn_to_page(node_pfn
);
399 add_one_highpage_init(page
, node_pfn
, bad_ppro
);
402 totalram_pages
+= totalhigh_pages
;
406 #ifdef CONFIG_MEMORY_HOTPLUG
407 static int paddr_to_nid(u64 addr
)
410 unsigned long pfn
= PFN_DOWN(addr
);
413 if (node_start_pfn
[nid
] <= pfn
&&
414 pfn
< node_end_pfn
[nid
])
421 * This function is used to ask node id BEFORE memmap and mem_section's
422 * initialization (pfn_to_nid() can't be used yet).
423 * If _PXM is not defined on ACPI's DSDT, node id must be found by this.
425 int memory_add_physaddr_to_nid(u64 addr
)
427 int nid
= paddr_to_nid(addr
);
428 return (nid
>= 0) ? nid
: 0;
431 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid
);