Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved. | |
3 | * Copyright (c) 2001 Intel Corp. | |
4 | * Copyright (c) 2001 Tony Luck <tony.luck@intel.com> | |
5 | * Copyright (c) 2002 NEC Corp. | |
6 | * Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com> | |
7 | * Copyright (c) 2004 Silicon Graphics, Inc | |
8 | * Russ Anderson <rja@sgi.com> | |
9 | * Jesse Barnes <jbarnes@sgi.com> | |
10 | * Jack Steiner <steiner@sgi.com> | |
11 | */ | |
12 | ||
13 | /* | |
14 | * Platform initialization for Discontig Memory | |
15 | */ | |
16 | ||
17 | #include <linux/kernel.h> | |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/bootmem.h> | |
21 | #include <linux/acpi.h> | |
22 | #include <linux/efi.h> | |
23 | #include <linux/nodemask.h> | |
24 | #include <asm/pgalloc.h> | |
25 | #include <asm/tlb.h> | |
26 | #include <asm/meminit.h> | |
27 | #include <asm/numa.h> | |
28 | #include <asm/sections.h> | |
29 | ||
30 | /* | |
31 | * Track per-node information needed to setup the boot memory allocator, the | |
32 | * per-node areas, and the real VM. | |
33 | */ | |
34 | struct early_node_data { | |
35 | struct ia64_node_data *node_data; | |
1da177e4 LT |
36 | unsigned long pernode_addr; |
37 | unsigned long pernode_size; | |
38 | struct bootmem_data bootmem_data; | |
39 | unsigned long num_physpages; | |
09ae1f58 | 40 | #ifdef CONFIG_ZONE_DMA |
1da177e4 | 41 | unsigned long num_dma_physpages; |
09ae1f58 | 42 | #endif |
1da177e4 LT |
43 | unsigned long min_pfn; |
44 | unsigned long max_pfn; | |
45 | }; | |
46 | ||
47 | static struct early_node_data mem_data[MAX_NUMNODES] __initdata; | |
564601a5 | 48 | static nodemask_t memory_less_mask __initdata; |
1da177e4 | 49 | |
ae5a2c1c YG |
50 | static pg_data_t *pgdat_list[MAX_NUMNODES]; |
51 | ||
1da177e4 LT |
52 | /* |
53 | * To prevent cache aliasing effects, align per-node structures so that they | |
54 | * start at addresses that are strided by node number. | |
55 | */ | |
acb7f672 | 56 | #define MAX_NODE_ALIGN_OFFSET (32 * 1024 * 1024) |
1da177e4 | 57 | #define NODEDATA_ALIGN(addr, node) \ |
acb7f672 JS |
58 | ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + \ |
59 | (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1))) | |
1da177e4 LT |
60 | |
61 | /** | |
62 | * build_node_maps - callback to setup bootmem structs for each node | |
63 | * @start: physical start of range | |
64 | * @len: length of range | |
65 | * @node: node where this range resides | |
66 | * | |
67 | * We allocate a struct bootmem_data for each piece of memory that we wish to | |
68 | * treat as a virtually contiguous block (i.e. each node). Each such block | |
69 | * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down | |
70 | * if necessary. Any non-existent pages will simply be part of the virtual | |
71 | * memmap. We also update min_low_pfn and max_low_pfn here as we receive | |
72 | * memory ranges from the caller. | |
73 | */ | |
74 | static int __init build_node_maps(unsigned long start, unsigned long len, | |
75 | int node) | |
76 | { | |
77 | unsigned long cstart, epfn, end = start + len; | |
78 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; | |
79 | ||
80 | epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT; | |
81 | cstart = GRANULEROUNDDOWN(start); | |
82 | ||
83 | if (!bdp->node_low_pfn) { | |
84 | bdp->node_boot_start = cstart; | |
85 | bdp->node_low_pfn = epfn; | |
86 | } else { | |
87 | bdp->node_boot_start = min(cstart, bdp->node_boot_start); | |
88 | bdp->node_low_pfn = max(epfn, bdp->node_low_pfn); | |
89 | } | |
90 | ||
1da177e4 LT |
91 | return 0; |
92 | } | |
93 | ||
94 | /** | |
564601a5 | 95 | * early_nr_cpus_node - return number of cpus on a given node |
1da177e4 LT |
96 | * @node: node to check |
97 | * | |
564601a5 | 98 | * Count the number of cpus on @node. We can't use nr_cpus_node() yet because |
1da177e4 | 99 | * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been |
564601a5 | 100 | * called yet. Note that node 0 will also count all non-existent cpus. |
1da177e4 | 101 | */ |
dd0932d9 | 102 | static int __meminit early_nr_cpus_node(int node) |
1da177e4 LT |
103 | { |
104 | int cpu, n = 0; | |
105 | ||
106 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
107 | if (node == node_cpuid[cpu].nid) | |
564601a5 | 108 | n++; |
1da177e4 LT |
109 | |
110 | return n; | |
111 | } | |
112 | ||
564601a5 | 113 | /** |
114 | * compute_pernodesize - compute size of pernode data | |
115 | * @node: the node id. | |
116 | */ | |
dd0932d9 | 117 | static unsigned long __meminit compute_pernodesize(int node) |
564601a5 | 118 | { |
119 | unsigned long pernodesize = 0, cpus; | |
120 | ||
121 | cpus = early_nr_cpus_node(node); | |
122 | pernodesize += PERCPU_PAGE_SIZE * cpus; | |
123 | pernodesize += node * L1_CACHE_BYTES; | |
124 | pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t)); | |
125 | pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
126 | pernodesize = PAGE_ALIGN(pernodesize); | |
127 | return pernodesize; | |
128 | } | |
1da177e4 | 129 | |
8d7e3517 TL |
130 | /** |
131 | * per_cpu_node_setup - setup per-cpu areas on each node | |
132 | * @cpu_data: per-cpu area on this node | |
133 | * @node: node to setup | |
134 | * | |
135 | * Copy the static per-cpu data into the region we just set aside and then | |
136 | * setup __per_cpu_offset for each CPU on this node. Return a pointer to | |
137 | * the end of the area. | |
138 | */ | |
139 | static void *per_cpu_node_setup(void *cpu_data, int node) | |
140 | { | |
141 | #ifdef CONFIG_SMP | |
142 | int cpu; | |
143 | ||
144 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
145 | if (node == node_cpuid[cpu].nid) { | |
146 | memcpy(__va(cpu_data), __phys_per_cpu_start, | |
147 | __per_cpu_end - __per_cpu_start); | |
148 | __per_cpu_offset[cpu] = (char*)__va(cpu_data) - | |
149 | __per_cpu_start; | |
150 | cpu_data += PERCPU_PAGE_SIZE; | |
151 | } | |
152 | } | |
153 | #endif | |
154 | return cpu_data; | |
155 | } | |
156 | ||
1da177e4 | 157 | /** |
564601a5 | 158 | * fill_pernode - initialize pernode data. |
159 | * @node: the node id. | |
160 | * @pernode: physical address of pernode data | |
161 | * @pernodesize: size of the pernode data | |
1da177e4 | 162 | */ |
564601a5 | 163 | static void __init fill_pernode(int node, unsigned long pernode, |
164 | unsigned long pernodesize) | |
1da177e4 | 165 | { |
564601a5 | 166 | void *cpu_data; |
8d7e3517 | 167 | int cpus = early_nr_cpus_node(node); |
564601a5 | 168 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
1da177e4 | 169 | |
564601a5 | 170 | mem_data[node].pernode_addr = pernode; |
171 | mem_data[node].pernode_size = pernodesize; | |
172 | memset(__va(pernode), 0, pernodesize); | |
1da177e4 | 173 | |
564601a5 | 174 | cpu_data = (void *)pernode; |
175 | pernode += PERCPU_PAGE_SIZE * cpus; | |
176 | pernode += node * L1_CACHE_BYTES; | |
177 | ||
ae5a2c1c | 178 | pgdat_list[node] = __va(pernode); |
564601a5 | 179 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
180 | ||
181 | mem_data[node].node_data = __va(pernode); | |
182 | pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
183 | ||
ae5a2c1c | 184 | pgdat_list[node]->bdata = bdp; |
564601a5 | 185 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
186 | ||
8d7e3517 | 187 | cpu_data = per_cpu_node_setup(cpu_data, node); |
1da177e4 | 188 | |
564601a5 | 189 | return; |
190 | } | |
8d7e3517 | 191 | |
1da177e4 LT |
192 | /** |
193 | * find_pernode_space - allocate memory for memory map and per-node structures | |
194 | * @start: physical start of range | |
195 | * @len: length of range | |
196 | * @node: node where this range resides | |
197 | * | |
198 | * This routine reserves space for the per-cpu data struct, the list of | |
199 | * pg_data_ts and the per-node data struct. Each node will have something like | |
200 | * the following in the first chunk of addr. space large enough to hold it. | |
201 | * | |
202 | * ________________________ | |
203 | * | | | |
204 | * |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first | |
205 | * | PERCPU_PAGE_SIZE * | start and length big enough | |
206 | * | cpus_on_this_node | Node 0 will also have entries for all non-existent cpus. | |
207 | * |------------------------| | |
208 | * | local pg_data_t * | | |
209 | * |------------------------| | |
210 | * | local ia64_node_data | | |
211 | * |------------------------| | |
212 | * | ??? | | |
213 | * |________________________| | |
214 | * | |
215 | * Once this space has been set aside, the bootmem maps are initialized. We | |
216 | * could probably move the allocation of the per-cpu and ia64_node_data space | |
217 | * outside of this function and use alloc_bootmem_node(), but doing it here | |
218 | * is straightforward and we get the alignments we want so... | |
219 | */ | |
220 | static int __init find_pernode_space(unsigned long start, unsigned long len, | |
221 | int node) | |
222 | { | |
564601a5 | 223 | unsigned long epfn; |
1da177e4 | 224 | unsigned long pernodesize = 0, pernode, pages, mapsize; |
1da177e4 LT |
225 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
226 | ||
227 | epfn = (start + len) >> PAGE_SHIFT; | |
228 | ||
229 | pages = bdp->node_low_pfn - (bdp->node_boot_start >> PAGE_SHIFT); | |
230 | mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
231 | ||
232 | /* | |
233 | * Make sure this memory falls within this node's usable memory | |
234 | * since we may have thrown some away in build_maps(). | |
235 | */ | |
236 | if (start < bdp->node_boot_start || epfn > bdp->node_low_pfn) | |
237 | return 0; | |
238 | ||
239 | /* Don't setup this node's local space twice... */ | |
240 | if (mem_data[node].pernode_addr) | |
241 | return 0; | |
242 | ||
243 | /* | |
244 | * Calculate total size needed, incl. what's necessary | |
245 | * for good alignment and alias prevention. | |
246 | */ | |
564601a5 | 247 | pernodesize = compute_pernodesize(node); |
1da177e4 LT |
248 | pernode = NODEDATA_ALIGN(start, node); |
249 | ||
250 | /* Is this range big enough for what we want to store here? */ | |
564601a5 | 251 | if (start + len > (pernode + pernodesize + mapsize)) |
252 | fill_pernode(node, pernode, pernodesize); | |
1da177e4 LT |
253 | |
254 | return 0; | |
255 | } | |
256 | ||
257 | /** | |
258 | * free_node_bootmem - free bootmem allocator memory for use | |
259 | * @start: physical start of range | |
260 | * @len: length of range | |
261 | * @node: node where this range resides | |
262 | * | |
263 | * Simply calls the bootmem allocator to free the specified ranged from | |
264 | * the given pg_data_t's bdata struct. After this function has been called | |
265 | * for all the entries in the EFI memory map, the bootmem allocator will | |
266 | * be ready to service allocation requests. | |
267 | */ | |
268 | static int __init free_node_bootmem(unsigned long start, unsigned long len, | |
269 | int node) | |
270 | { | |
ae5a2c1c | 271 | free_bootmem_node(pgdat_list[node], start, len); |
1da177e4 LT |
272 | |
273 | return 0; | |
274 | } | |
275 | ||
276 | /** | |
277 | * reserve_pernode_space - reserve memory for per-node space | |
278 | * | |
279 | * Reserve the space used by the bootmem maps & per-node space in the boot | |
280 | * allocator so that when we actually create the real mem maps we don't | |
281 | * use their memory. | |
282 | */ | |
283 | static void __init reserve_pernode_space(void) | |
284 | { | |
285 | unsigned long base, size, pages; | |
286 | struct bootmem_data *bdp; | |
287 | int node; | |
288 | ||
289 | for_each_online_node(node) { | |
ae5a2c1c | 290 | pg_data_t *pdp = pgdat_list[node]; |
1da177e4 | 291 | |
564601a5 | 292 | if (node_isset(node, memory_less_mask)) |
293 | continue; | |
294 | ||
1da177e4 LT |
295 | bdp = pdp->bdata; |
296 | ||
297 | /* First the bootmem_map itself */ | |
298 | pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT); | |
299 | size = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
300 | base = __pa(bdp->node_bootmem_map); | |
301 | reserve_bootmem_node(pdp, base, size); | |
302 | ||
303 | /* Now the per-node space */ | |
304 | size = mem_data[node].pernode_size; | |
305 | base = __pa(mem_data[node].pernode_addr); | |
306 | reserve_bootmem_node(pdp, base, size); | |
307 | } | |
308 | } | |
309 | ||
7049027c YG |
310 | static void __meminit scatter_node_data(void) |
311 | { | |
312 | pg_data_t **dst; | |
313 | int node; | |
314 | ||
dd8041f1 YG |
315 | /* |
316 | * for_each_online_node() can't be used at here. | |
317 | * node_online_map is not set for hot-added nodes at this time, | |
318 | * because we are halfway through initialization of the new node's | |
319 | * structures. If for_each_online_node() is used, a new node's | |
320 | * pg_data_ptrs will be not initialized. Insted of using it, | |
321 | * pgdat_list[] is checked. | |
322 | */ | |
323 | for_each_node(node) { | |
324 | if (pgdat_list[node]) { | |
325 | dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs; | |
326 | memcpy(dst, pgdat_list, sizeof(pgdat_list)); | |
327 | } | |
7049027c YG |
328 | } |
329 | } | |
330 | ||
1da177e4 LT |
331 | /** |
332 | * initialize_pernode_data - fixup per-cpu & per-node pointers | |
333 | * | |
334 | * Each node's per-node area has a copy of the global pg_data_t list, so | |
335 | * we copy that to each node here, as well as setting the per-cpu pointer | |
336 | * to the local node data structure. The active_cpus field of the per-node | |
337 | * structure gets setup by the platform_cpu_init() function later. | |
338 | */ | |
339 | static void __init initialize_pernode_data(void) | |
340 | { | |
8d7e3517 | 341 | int cpu, node; |
1da177e4 | 342 | |
7049027c YG |
343 | scatter_node_data(); |
344 | ||
8d7e3517 | 345 | #ifdef CONFIG_SMP |
1da177e4 LT |
346 | /* Set the node_data pointer for each per-cpu struct */ |
347 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
348 | node = node_cpuid[cpu].nid; | |
349 | per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data; | |
350 | } | |
8d7e3517 TL |
351 | #else |
352 | { | |
353 | struct cpuinfo_ia64 *cpu0_cpu_info; | |
354 | cpu = 0; | |
355 | node = node_cpuid[cpu].nid; | |
356 | cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start + | |
357 | ((char *)&per_cpu__cpu_info - __per_cpu_start)); | |
358 | cpu0_cpu_info->node_data = mem_data[node].node_data; | |
359 | } | |
360 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
361 | } |
362 | ||
564601a5 | 363 | /** |
364 | * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit | |
365 | * node but fall back to any other node when __alloc_bootmem_node fails | |
366 | * for best. | |
367 | * @nid: node id | |
368 | * @pernodesize: size of this node's pernode data | |
564601a5 | 369 | */ |
97835245 | 370 | static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize) |
564601a5 | 371 | { |
372 | void *ptr = NULL; | |
373 | u8 best = 0xff; | |
97835245 | 374 | int bestnode = -1, node, anynode = 0; |
564601a5 | 375 | |
376 | for_each_online_node(node) { | |
377 | if (node_isset(node, memory_less_mask)) | |
378 | continue; | |
379 | else if (node_distance(nid, node) < best) { | |
380 | best = node_distance(nid, node); | |
381 | bestnode = node; | |
382 | } | |
97835245 | 383 | anynode = node; |
564601a5 | 384 | } |
385 | ||
97835245 BP |
386 | if (bestnode == -1) |
387 | bestnode = anynode; | |
388 | ||
ae5a2c1c | 389 | ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize, |
97835245 | 390 | PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); |
564601a5 | 391 | |
564601a5 | 392 | return ptr; |
393 | } | |
394 | ||
564601a5 | 395 | /** |
396 | * memory_less_nodes - allocate and initialize CPU only nodes pernode | |
397 | * information. | |
398 | */ | |
399 | static void __init memory_less_nodes(void) | |
400 | { | |
401 | unsigned long pernodesize; | |
402 | void *pernode; | |
403 | int node; | |
404 | ||
405 | for_each_node_mask(node, memory_less_mask) { | |
406 | pernodesize = compute_pernodesize(node); | |
97835245 | 407 | pernode = memory_less_node_alloc(node, pernodesize); |
564601a5 | 408 | fill_pernode(node, __pa(pernode), pernodesize); |
409 | } | |
410 | ||
411 | return; | |
412 | } | |
413 | ||
1da177e4 LT |
414 | /** |
415 | * find_memory - walk the EFI memory map and setup the bootmem allocator | |
416 | * | |
417 | * Called early in boot to setup the bootmem allocator, and to | |
418 | * allocate the per-cpu and per-node structures. | |
419 | */ | |
420 | void __init find_memory(void) | |
421 | { | |
422 | int node; | |
423 | ||
424 | reserve_memory(); | |
425 | ||
426 | if (num_online_nodes() == 0) { | |
427 | printk(KERN_ERR "node info missing!\n"); | |
428 | node_set_online(0); | |
429 | } | |
430 | ||
564601a5 | 431 | nodes_or(memory_less_mask, memory_less_mask, node_online_map); |
1da177e4 LT |
432 | min_low_pfn = -1; |
433 | max_low_pfn = 0; | |
434 | ||
1da177e4 LT |
435 | /* These actually end up getting called by call_pernode_memory() */ |
436 | efi_memmap_walk(filter_rsvd_memory, build_node_maps); | |
437 | efi_memmap_walk(filter_rsvd_memory, find_pernode_space); | |
a3f5c338 | 438 | efi_memmap_walk(find_max_min_low_pfn, NULL); |
1da177e4 | 439 | |
564601a5 | 440 | for_each_online_node(node) |
441 | if (mem_data[node].bootmem_data.node_low_pfn) { | |
442 | node_clear(node, memory_less_mask); | |
443 | mem_data[node].min_pfn = ~0UL; | |
444 | } | |
139b8304 BP |
445 | |
446 | efi_memmap_walk(register_active_ranges, NULL); | |
447 | ||
1da177e4 LT |
448 | /* |
449 | * Initialize the boot memory maps in reverse order since that's | |
450 | * what the bootmem allocator expects | |
451 | */ | |
452 | for (node = MAX_NUMNODES - 1; node >= 0; node--) { | |
453 | unsigned long pernode, pernodesize, map; | |
454 | struct bootmem_data *bdp; | |
455 | ||
456 | if (!node_online(node)) | |
457 | continue; | |
564601a5 | 458 | else if (node_isset(node, memory_less_mask)) |
459 | continue; | |
1da177e4 LT |
460 | |
461 | bdp = &mem_data[node].bootmem_data; | |
462 | pernode = mem_data[node].pernode_addr; | |
463 | pernodesize = mem_data[node].pernode_size; | |
464 | map = pernode + pernodesize; | |
465 | ||
ae5a2c1c | 466 | init_bootmem_node(pgdat_list[node], |
1da177e4 LT |
467 | map>>PAGE_SHIFT, |
468 | bdp->node_boot_start>>PAGE_SHIFT, | |
469 | bdp->node_low_pfn); | |
470 | } | |
471 | ||
472 | efi_memmap_walk(filter_rsvd_memory, free_node_bootmem); | |
473 | ||
474 | reserve_pernode_space(); | |
564601a5 | 475 | memory_less_nodes(); |
1da177e4 LT |
476 | initialize_pernode_data(); |
477 | ||
478 | max_pfn = max_low_pfn; | |
479 | ||
480 | find_initrd(); | |
481 | } | |
482 | ||
8d7e3517 | 483 | #ifdef CONFIG_SMP |
1da177e4 LT |
484 | /** |
485 | * per_cpu_init - setup per-cpu variables | |
486 | * | |
487 | * find_pernode_space() does most of this already, we just need to set | |
488 | * local_per_cpu_offset | |
489 | */ | |
244fd545 | 490 | void __cpuinit *per_cpu_init(void) |
1da177e4 LT |
491 | { |
492 | int cpu; | |
ff741906 AR |
493 | static int first_time = 1; |
494 | ||
1da177e4 | 495 | |
8d7e3517 TL |
496 | if (smp_processor_id() != 0) |
497 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
498 | ||
ff741906 AR |
499 | if (first_time) { |
500 | first_time = 0; | |
501 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
502 | per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu]; | |
503 | } | |
1da177e4 LT |
504 | |
505 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
506 | } | |
8d7e3517 | 507 | #endif /* CONFIG_SMP */ |
1da177e4 LT |
508 | |
509 | /** | |
510 | * show_mem - give short summary of memory stats | |
511 | * | |
512 | * Shows a simple page count of reserved and used pages in the system. | |
513 | * For discontig machines, it does this on a per-pgdat basis. | |
514 | */ | |
515 | void show_mem(void) | |
516 | { | |
517 | int i, total_reserved = 0; | |
518 | int total_shared = 0, total_cached = 0; | |
519 | unsigned long total_present = 0; | |
520 | pg_data_t *pgdat; | |
521 | ||
709a6c1c | 522 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 523 | show_free_areas(); |
709a6c1c JS |
524 | printk(KERN_INFO "Free swap: %6ldkB\n", |
525 | nr_swap_pages<<(PAGE_SHIFT-10)); | |
816add4e | 526 | printk(KERN_INFO "Node memory in pages:\n"); |
ec936fc5 | 527 | for_each_online_pgdat(pgdat) { |
208d54e5 DH |
528 | unsigned long present; |
529 | unsigned long flags; | |
1da177e4 | 530 | int shared = 0, cached = 0, reserved = 0; |
208d54e5 | 531 | |
208d54e5 DH |
532 | pgdat_resize_lock(pgdat, &flags); |
533 | present = pgdat->node_present_pages; | |
1da177e4 | 534 | for(i = 0; i < pgdat->node_spanned_pages; i++) { |
2d4b1fa2 BP |
535 | struct page *page; |
536 | if (pfn_valid(pgdat->node_start_pfn + i)) | |
537 | page = pfn_to_page(pgdat->node_start_pfn + i); | |
ace1d816 | 538 | else { |
e44e41d0 BP |
539 | i = vmemmap_find_next_valid_pfn(pgdat->node_id, |
540 | i) - 1; | |
1da177e4 | 541 | continue; |
ace1d816 | 542 | } |
408fde81 | 543 | if (PageReserved(page)) |
1da177e4 | 544 | reserved++; |
408fde81 | 545 | else if (PageSwapCache(page)) |
1da177e4 | 546 | cached++; |
408fde81 DH |
547 | else if (page_count(page)) |
548 | shared += page_count(page)-1; | |
1da177e4 | 549 | } |
208d54e5 | 550 | pgdat_resize_unlock(pgdat, &flags); |
1da177e4 LT |
551 | total_present += present; |
552 | total_reserved += reserved; | |
553 | total_cached += cached; | |
554 | total_shared += shared; | |
816add4e JS |
555 | printk(KERN_INFO "Node %4d: RAM: %11ld, rsvd: %8d, " |
556 | "shrd: %10d, swpd: %10d\n", pgdat->node_id, | |
557 | present, reserved, shared, cached); | |
1da177e4 | 558 | } |
709a6c1c JS |
559 | printk(KERN_INFO "%ld pages of RAM\n", total_present); |
560 | printk(KERN_INFO "%d reserved pages\n", total_reserved); | |
561 | printk(KERN_INFO "%d pages shared\n", total_shared); | |
562 | printk(KERN_INFO "%d pages swap cached\n", total_cached); | |
563 | printk(KERN_INFO "Total of %ld pages in page table cache\n", | |
564 | pgtable_quicklist_total_size()); | |
565 | printk(KERN_INFO "%d free buffer pages\n", nr_free_buffer_pages()); | |
1da177e4 LT |
566 | } |
567 | ||
568 | /** | |
569 | * call_pernode_memory - use SRAT to call callback functions with node info | |
570 | * @start: physical start of range | |
571 | * @len: length of range | |
572 | * @arg: function to call for each range | |
573 | * | |
574 | * efi_memmap_walk() knows nothing about layout of memory across nodes. Find | |
575 | * out to which node a block of memory belongs. Ignore memory that we cannot | |
576 | * identify, and split blocks that run across multiple nodes. | |
577 | * | |
578 | * Take this opportunity to round the start address up and the end address | |
579 | * down to page boundaries. | |
580 | */ | |
581 | void call_pernode_memory(unsigned long start, unsigned long len, void *arg) | |
582 | { | |
583 | unsigned long rs, re, end = start + len; | |
584 | void (*func)(unsigned long, unsigned long, int); | |
585 | int i; | |
586 | ||
587 | start = PAGE_ALIGN(start); | |
588 | end &= PAGE_MASK; | |
589 | if (start >= end) | |
590 | return; | |
591 | ||
592 | func = arg; | |
593 | ||
594 | if (!num_node_memblks) { | |
595 | /* No SRAT table, so assume one node (node 0) */ | |
596 | if (start < end) | |
597 | (*func)(start, end - start, 0); | |
598 | return; | |
599 | } | |
600 | ||
601 | for (i = 0; i < num_node_memblks; i++) { | |
602 | rs = max(start, node_memblk[i].start_paddr); | |
603 | re = min(end, node_memblk[i].start_paddr + | |
604 | node_memblk[i].size); | |
605 | ||
606 | if (rs < re) | |
607 | (*func)(rs, re - rs, node_memblk[i].nid); | |
608 | ||
609 | if (re == end) | |
610 | break; | |
611 | } | |
612 | } | |
613 | ||
614 | /** | |
615 | * count_node_pages - callback to build per-node memory info structures | |
616 | * @start: physical start of range | |
617 | * @len: length of range | |
618 | * @node: node where this range resides | |
619 | * | |
620 | * Each node has it's own number of physical pages, DMAable pages, start, and | |
621 | * end page frame number. This routine will be called by call_pernode_memory() | |
622 | * for each piece of usable memory and will setup these values for each node. | |
623 | * Very similar to build_maps(). | |
624 | */ | |
625 | static __init int count_node_pages(unsigned long start, unsigned long len, int node) | |
626 | { | |
627 | unsigned long end = start + len; | |
628 | ||
629 | mem_data[node].num_physpages += len >> PAGE_SHIFT; | |
09ae1f58 | 630 | #ifdef CONFIG_ZONE_DMA |
1da177e4 LT |
631 | if (start <= __pa(MAX_DMA_ADDRESS)) |
632 | mem_data[node].num_dma_physpages += | |
633 | (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT; | |
09ae1f58 | 634 | #endif |
1da177e4 LT |
635 | start = GRANULEROUNDDOWN(start); |
636 | start = ORDERROUNDDOWN(start); | |
637 | end = GRANULEROUNDUP(end); | |
638 | mem_data[node].max_pfn = max(mem_data[node].max_pfn, | |
639 | end >> PAGE_SHIFT); | |
640 | mem_data[node].min_pfn = min(mem_data[node].min_pfn, | |
641 | start >> PAGE_SHIFT); | |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | /** | |
647 | * paging_init - setup page tables | |
648 | * | |
649 | * paging_init() sets up the page tables for each node of the system and frees | |
650 | * the bootmem allocator memory for general use. | |
651 | */ | |
652 | void __init paging_init(void) | |
653 | { | |
654 | unsigned long max_dma; | |
1da177e4 | 655 | unsigned long pfn_offset = 0; |
05e0caad | 656 | unsigned long max_pfn = 0; |
1da177e4 | 657 | int node; |
05e0caad | 658 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1da177e4 LT |
659 | |
660 | max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
661 | ||
1da177e4 LT |
662 | efi_memmap_walk(filter_rsvd_memory, count_node_pages); |
663 | ||
524fd988 BP |
664 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
665 | sparse_init(); | |
666 | ||
2d4b1fa2 | 667 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
921eea1c BP |
668 | vmalloc_end -= PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) * |
669 | sizeof(struct page)); | |
564601a5 | 670 | vmem_map = (struct page *) vmalloc_end; |
671 | efi_memmap_walk(create_mem_map_page_table, NULL); | |
672 | printk("Virtual mem_map starts at 0x%p\n", vmem_map); | |
2d4b1fa2 | 673 | #endif |
564601a5 | 674 | |
1da177e4 | 675 | for_each_online_node(node) { |
1da177e4 | 676 | num_physpages += mem_data[node].num_physpages; |
1da177e4 LT |
677 | pfn_offset = mem_data[node].min_pfn; |
678 | ||
2d4b1fa2 | 679 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
1da177e4 | 680 | NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset; |
2d4b1fa2 | 681 | #endif |
05e0caad MG |
682 | if (mem_data[node].max_pfn > max_pfn) |
683 | max_pfn = mem_data[node].max_pfn; | |
1da177e4 LT |
684 | } |
685 | ||
6391af17 | 686 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
09ae1f58 | 687 | #ifdef CONFIG_ZONE_DMA |
05e0caad | 688 | max_zone_pfns[ZONE_DMA] = max_dma; |
09ae1f58 | 689 | #endif |
05e0caad MG |
690 | max_zone_pfns[ZONE_NORMAL] = max_pfn; |
691 | free_area_init_nodes(max_zone_pfns); | |
692 | ||
1da177e4 LT |
693 | zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page)); |
694 | } | |
7049027c | 695 | |
dd0932d9 YG |
696 | pg_data_t *arch_alloc_nodedata(int nid) |
697 | { | |
698 | unsigned long size = compute_pernodesize(nid); | |
699 | ||
700 | return kzalloc(size, GFP_KERNEL); | |
701 | } | |
702 | ||
703 | void arch_free_nodedata(pg_data_t *pgdat) | |
704 | { | |
705 | kfree(pgdat); | |
706 | } | |
707 | ||
7049027c YG |
708 | void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat) |
709 | { | |
710 | pgdat_list[update_node] = update_pgdat; | |
711 | scatter_node_data(); | |
712 | } |