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; | |
40 | unsigned long num_dma_physpages; | |
41 | unsigned long min_pfn; | |
42 | unsigned long max_pfn; | |
43 | }; | |
44 | ||
45 | static struct early_node_data mem_data[MAX_NUMNODES] __initdata; | |
564601a5 | 46 | static nodemask_t memory_less_mask __initdata; |
1da177e4 | 47 | |
ae5a2c1c YG |
48 | static pg_data_t *pgdat_list[MAX_NUMNODES]; |
49 | ||
1da177e4 LT |
50 | /* |
51 | * To prevent cache aliasing effects, align per-node structures so that they | |
52 | * start at addresses that are strided by node number. | |
53 | */ | |
acb7f672 | 54 | #define MAX_NODE_ALIGN_OFFSET (32 * 1024 * 1024) |
1da177e4 | 55 | #define NODEDATA_ALIGN(addr, node) \ |
acb7f672 JS |
56 | ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + \ |
57 | (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1))) | |
1da177e4 LT |
58 | |
59 | /** | |
60 | * build_node_maps - callback to setup bootmem structs for each node | |
61 | * @start: physical start of range | |
62 | * @len: length of range | |
63 | * @node: node where this range resides | |
64 | * | |
65 | * We allocate a struct bootmem_data for each piece of memory that we wish to | |
66 | * treat as a virtually contiguous block (i.e. each node). Each such block | |
67 | * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down | |
68 | * if necessary. Any non-existent pages will simply be part of the virtual | |
69 | * memmap. We also update min_low_pfn and max_low_pfn here as we receive | |
70 | * memory ranges from the caller. | |
71 | */ | |
72 | static int __init build_node_maps(unsigned long start, unsigned long len, | |
73 | int node) | |
74 | { | |
75 | unsigned long cstart, epfn, end = start + len; | |
76 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; | |
77 | ||
78 | epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT; | |
79 | cstart = GRANULEROUNDDOWN(start); | |
80 | ||
81 | if (!bdp->node_low_pfn) { | |
82 | bdp->node_boot_start = cstart; | |
83 | bdp->node_low_pfn = epfn; | |
84 | } else { | |
85 | bdp->node_boot_start = min(cstart, bdp->node_boot_start); | |
86 | bdp->node_low_pfn = max(epfn, bdp->node_low_pfn); | |
87 | } | |
88 | ||
89 | min_low_pfn = min(min_low_pfn, bdp->node_boot_start>>PAGE_SHIFT); | |
90 | max_low_pfn = max(max_low_pfn, bdp->node_low_pfn); | |
91 | ||
92 | return 0; | |
93 | } | |
94 | ||
95 | /** | |
564601a5 | 96 | * early_nr_cpus_node - return number of cpus on a given node |
1da177e4 LT |
97 | * @node: node to check |
98 | * | |
564601a5 | 99 | * Count the number of cpus on @node. We can't use nr_cpus_node() yet because |
1da177e4 | 100 | * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been |
564601a5 | 101 | * called yet. Note that node 0 will also count all non-existent cpus. |
1da177e4 | 102 | */ |
dd0932d9 | 103 | static int __meminit early_nr_cpus_node(int node) |
1da177e4 LT |
104 | { |
105 | int cpu, n = 0; | |
106 | ||
107 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
108 | if (node == node_cpuid[cpu].nid) | |
564601a5 | 109 | n++; |
1da177e4 LT |
110 | |
111 | return n; | |
112 | } | |
113 | ||
564601a5 | 114 | /** |
115 | * compute_pernodesize - compute size of pernode data | |
116 | * @node: the node id. | |
117 | */ | |
dd0932d9 | 118 | static unsigned long __meminit compute_pernodesize(int node) |
564601a5 | 119 | { |
120 | unsigned long pernodesize = 0, cpus; | |
121 | ||
122 | cpus = early_nr_cpus_node(node); | |
123 | pernodesize += PERCPU_PAGE_SIZE * cpus; | |
124 | pernodesize += node * L1_CACHE_BYTES; | |
125 | pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t)); | |
126 | pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
127 | pernodesize = PAGE_ALIGN(pernodesize); | |
128 | return pernodesize; | |
129 | } | |
1da177e4 | 130 | |
8d7e3517 TL |
131 | /** |
132 | * per_cpu_node_setup - setup per-cpu areas on each node | |
133 | * @cpu_data: per-cpu area on this node | |
134 | * @node: node to setup | |
135 | * | |
136 | * Copy the static per-cpu data into the region we just set aside and then | |
137 | * setup __per_cpu_offset for each CPU on this node. Return a pointer to | |
138 | * the end of the area. | |
139 | */ | |
140 | static void *per_cpu_node_setup(void *cpu_data, int node) | |
141 | { | |
142 | #ifdef CONFIG_SMP | |
143 | int cpu; | |
144 | ||
145 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
146 | if (node == node_cpuid[cpu].nid) { | |
147 | memcpy(__va(cpu_data), __phys_per_cpu_start, | |
148 | __per_cpu_end - __per_cpu_start); | |
149 | __per_cpu_offset[cpu] = (char*)__va(cpu_data) - | |
150 | __per_cpu_start; | |
151 | cpu_data += PERCPU_PAGE_SIZE; | |
152 | } | |
153 | } | |
154 | #endif | |
155 | return cpu_data; | |
156 | } | |
157 | ||
1da177e4 | 158 | /** |
564601a5 | 159 | * fill_pernode - initialize pernode data. |
160 | * @node: the node id. | |
161 | * @pernode: physical address of pernode data | |
162 | * @pernodesize: size of the pernode data | |
1da177e4 | 163 | */ |
564601a5 | 164 | static void __init fill_pernode(int node, unsigned long pernode, |
165 | unsigned long pernodesize) | |
1da177e4 | 166 | { |
564601a5 | 167 | void *cpu_data; |
8d7e3517 | 168 | int cpus = early_nr_cpus_node(node); |
564601a5 | 169 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
1da177e4 | 170 | |
564601a5 | 171 | mem_data[node].pernode_addr = pernode; |
172 | mem_data[node].pernode_size = pernodesize; | |
173 | memset(__va(pernode), 0, pernodesize); | |
1da177e4 | 174 | |
564601a5 | 175 | cpu_data = (void *)pernode; |
176 | pernode += PERCPU_PAGE_SIZE * cpus; | |
177 | pernode += node * L1_CACHE_BYTES; | |
178 | ||
ae5a2c1c | 179 | pgdat_list[node] = __va(pernode); |
564601a5 | 180 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
181 | ||
182 | mem_data[node].node_data = __va(pernode); | |
183 | pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
184 | ||
ae5a2c1c | 185 | pgdat_list[node]->bdata = bdp; |
564601a5 | 186 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
187 | ||
8d7e3517 | 188 | cpu_data = per_cpu_node_setup(cpu_data, node); |
1da177e4 | 189 | |
564601a5 | 190 | return; |
191 | } | |
8d7e3517 | 192 | |
1da177e4 LT |
193 | /** |
194 | * find_pernode_space - allocate memory for memory map and per-node structures | |
195 | * @start: physical start of range | |
196 | * @len: length of range | |
197 | * @node: node where this range resides | |
198 | * | |
199 | * This routine reserves space for the per-cpu data struct, the list of | |
200 | * pg_data_ts and the per-node data struct. Each node will have something like | |
201 | * the following in the first chunk of addr. space large enough to hold it. | |
202 | * | |
203 | * ________________________ | |
204 | * | | | |
205 | * |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first | |
206 | * | PERCPU_PAGE_SIZE * | start and length big enough | |
207 | * | cpus_on_this_node | Node 0 will also have entries for all non-existent cpus. | |
208 | * |------------------------| | |
209 | * | local pg_data_t * | | |
210 | * |------------------------| | |
211 | * | local ia64_node_data | | |
212 | * |------------------------| | |
213 | * | ??? | | |
214 | * |________________________| | |
215 | * | |
216 | * Once this space has been set aside, the bootmem maps are initialized. We | |
217 | * could probably move the allocation of the per-cpu and ia64_node_data space | |
218 | * outside of this function and use alloc_bootmem_node(), but doing it here | |
219 | * is straightforward and we get the alignments we want so... | |
220 | */ | |
221 | static int __init find_pernode_space(unsigned long start, unsigned long len, | |
222 | int node) | |
223 | { | |
564601a5 | 224 | unsigned long epfn; |
1da177e4 | 225 | unsigned long pernodesize = 0, pernode, pages, mapsize; |
1da177e4 LT |
226 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
227 | ||
228 | epfn = (start + len) >> PAGE_SHIFT; | |
229 | ||
230 | pages = bdp->node_low_pfn - (bdp->node_boot_start >> PAGE_SHIFT); | |
231 | mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
232 | ||
233 | /* | |
234 | * Make sure this memory falls within this node's usable memory | |
235 | * since we may have thrown some away in build_maps(). | |
236 | */ | |
237 | if (start < bdp->node_boot_start || epfn > bdp->node_low_pfn) | |
238 | return 0; | |
239 | ||
240 | /* Don't setup this node's local space twice... */ | |
241 | if (mem_data[node].pernode_addr) | |
242 | return 0; | |
243 | ||
244 | /* | |
245 | * Calculate total size needed, incl. what's necessary | |
246 | * for good alignment and alias prevention. | |
247 | */ | |
564601a5 | 248 | pernodesize = compute_pernodesize(node); |
1da177e4 LT |
249 | pernode = NODEDATA_ALIGN(start, node); |
250 | ||
251 | /* Is this range big enough for what we want to store here? */ | |
564601a5 | 252 | if (start + len > (pernode + pernodesize + mapsize)) |
253 | fill_pernode(node, pernode, pernodesize); | |
1da177e4 LT |
254 | |
255 | return 0; | |
256 | } | |
257 | ||
258 | /** | |
259 | * free_node_bootmem - free bootmem allocator memory for use | |
260 | * @start: physical start of range | |
261 | * @len: length of range | |
262 | * @node: node where this range resides | |
263 | * | |
264 | * Simply calls the bootmem allocator to free the specified ranged from | |
265 | * the given pg_data_t's bdata struct. After this function has been called | |
266 | * for all the entries in the EFI memory map, the bootmem allocator will | |
267 | * be ready to service allocation requests. | |
268 | */ | |
269 | static int __init free_node_bootmem(unsigned long start, unsigned long len, | |
270 | int node) | |
271 | { | |
ae5a2c1c | 272 | free_bootmem_node(pgdat_list[node], start, len); |
1da177e4 LT |
273 | |
274 | return 0; | |
275 | } | |
276 | ||
277 | /** | |
278 | * reserve_pernode_space - reserve memory for per-node space | |
279 | * | |
280 | * Reserve the space used by the bootmem maps & per-node space in the boot | |
281 | * allocator so that when we actually create the real mem maps we don't | |
282 | * use their memory. | |
283 | */ | |
284 | static void __init reserve_pernode_space(void) | |
285 | { | |
286 | unsigned long base, size, pages; | |
287 | struct bootmem_data *bdp; | |
288 | int node; | |
289 | ||
290 | for_each_online_node(node) { | |
ae5a2c1c | 291 | pg_data_t *pdp = pgdat_list[node]; |
1da177e4 | 292 | |
564601a5 | 293 | if (node_isset(node, memory_less_mask)) |
294 | continue; | |
295 | ||
1da177e4 LT |
296 | bdp = pdp->bdata; |
297 | ||
298 | /* First the bootmem_map itself */ | |
299 | pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT); | |
300 | size = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
301 | base = __pa(bdp->node_bootmem_map); | |
302 | reserve_bootmem_node(pdp, base, size); | |
303 | ||
304 | /* Now the per-node space */ | |
305 | size = mem_data[node].pernode_size; | |
306 | base = __pa(mem_data[node].pernode_addr); | |
307 | reserve_bootmem_node(pdp, base, size); | |
308 | } | |
309 | } | |
310 | ||
7049027c YG |
311 | static void __meminit scatter_node_data(void) |
312 | { | |
313 | pg_data_t **dst; | |
314 | int node; | |
315 | ||
dd8041f1 YG |
316 | /* |
317 | * for_each_online_node() can't be used at here. | |
318 | * node_online_map is not set for hot-added nodes at this time, | |
319 | * because we are halfway through initialization of the new node's | |
320 | * structures. If for_each_online_node() is used, a new node's | |
321 | * pg_data_ptrs will be not initialized. Insted of using it, | |
322 | * pgdat_list[] is checked. | |
323 | */ | |
324 | for_each_node(node) { | |
325 | if (pgdat_list[node]) { | |
326 | dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs; | |
327 | memcpy(dst, pgdat_list, sizeof(pgdat_list)); | |
328 | } | |
7049027c YG |
329 | } |
330 | } | |
331 | ||
1da177e4 LT |
332 | /** |
333 | * initialize_pernode_data - fixup per-cpu & per-node pointers | |
334 | * | |
335 | * Each node's per-node area has a copy of the global pg_data_t list, so | |
336 | * we copy that to each node here, as well as setting the per-cpu pointer | |
337 | * to the local node data structure. The active_cpus field of the per-node | |
338 | * structure gets setup by the platform_cpu_init() function later. | |
339 | */ | |
340 | static void __init initialize_pernode_data(void) | |
341 | { | |
8d7e3517 | 342 | int cpu, node; |
1da177e4 | 343 | |
7049027c YG |
344 | scatter_node_data(); |
345 | ||
8d7e3517 | 346 | #ifdef CONFIG_SMP |
1da177e4 LT |
347 | /* Set the node_data pointer for each per-cpu struct */ |
348 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
349 | node = node_cpuid[cpu].nid; | |
350 | per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data; | |
351 | } | |
8d7e3517 TL |
352 | #else |
353 | { | |
354 | struct cpuinfo_ia64 *cpu0_cpu_info; | |
355 | cpu = 0; | |
356 | node = node_cpuid[cpu].nid; | |
357 | cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start + | |
358 | ((char *)&per_cpu__cpu_info - __per_cpu_start)); | |
359 | cpu0_cpu_info->node_data = mem_data[node].node_data; | |
360 | } | |
361 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
362 | } |
363 | ||
564601a5 | 364 | /** |
365 | * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit | |
366 | * node but fall back to any other node when __alloc_bootmem_node fails | |
367 | * for best. | |
368 | * @nid: node id | |
369 | * @pernodesize: size of this node's pernode data | |
564601a5 | 370 | */ |
97835245 | 371 | static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize) |
564601a5 | 372 | { |
373 | void *ptr = NULL; | |
374 | u8 best = 0xff; | |
97835245 | 375 | int bestnode = -1, node, anynode = 0; |
564601a5 | 376 | |
377 | for_each_online_node(node) { | |
378 | if (node_isset(node, memory_less_mask)) | |
379 | continue; | |
380 | else if (node_distance(nid, node) < best) { | |
381 | best = node_distance(nid, node); | |
382 | bestnode = node; | |
383 | } | |
97835245 | 384 | anynode = node; |
564601a5 | 385 | } |
386 | ||
97835245 BP |
387 | if (bestnode == -1) |
388 | bestnode = anynode; | |
389 | ||
ae5a2c1c | 390 | ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize, |
97835245 | 391 | PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); |
564601a5 | 392 | |
564601a5 | 393 | return ptr; |
394 | } | |
395 | ||
564601a5 | 396 | /** |
397 | * memory_less_nodes - allocate and initialize CPU only nodes pernode | |
398 | * information. | |
399 | */ | |
400 | static void __init memory_less_nodes(void) | |
401 | { | |
402 | unsigned long pernodesize; | |
403 | void *pernode; | |
404 | int node; | |
405 | ||
406 | for_each_node_mask(node, memory_less_mask) { | |
407 | pernodesize = compute_pernodesize(node); | |
97835245 | 408 | pernode = memory_less_node_alloc(node, pernodesize); |
564601a5 | 409 | fill_pernode(node, __pa(pernode), pernodesize); |
410 | } | |
411 | ||
412 | return; | |
413 | } | |
414 | ||
2d4b1fa2 BP |
415 | #ifdef CONFIG_SPARSEMEM |
416 | /** | |
417 | * register_sparse_mem - notify SPARSEMEM that this memory range exists. | |
418 | * @start: physical start of range | |
419 | * @end: physical end of range | |
420 | * @arg: unused | |
421 | * | |
422 | * Simply calls SPARSEMEM to register memory section(s). | |
423 | */ | |
424 | static int __init register_sparse_mem(unsigned long start, unsigned long end, | |
425 | void *arg) | |
426 | { | |
427 | int nid; | |
428 | ||
429 | start = __pa(start) >> PAGE_SHIFT; | |
430 | end = __pa(end) >> PAGE_SHIFT; | |
431 | nid = early_pfn_to_nid(start); | |
432 | memory_present(nid, start, end); | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
437 | static void __init arch_sparse_init(void) | |
438 | { | |
439 | efi_memmap_walk(register_sparse_mem, NULL); | |
440 | sparse_init(); | |
441 | } | |
442 | #else | |
443 | #define arch_sparse_init() do {} while (0) | |
444 | #endif | |
445 | ||
1da177e4 LT |
446 | /** |
447 | * find_memory - walk the EFI memory map and setup the bootmem allocator | |
448 | * | |
449 | * Called early in boot to setup the bootmem allocator, and to | |
450 | * allocate the per-cpu and per-node structures. | |
451 | */ | |
452 | void __init find_memory(void) | |
453 | { | |
454 | int node; | |
455 | ||
456 | reserve_memory(); | |
457 | ||
458 | if (num_online_nodes() == 0) { | |
459 | printk(KERN_ERR "node info missing!\n"); | |
460 | node_set_online(0); | |
461 | } | |
462 | ||
564601a5 | 463 | nodes_or(memory_less_mask, memory_less_mask, node_online_map); |
1da177e4 LT |
464 | min_low_pfn = -1; |
465 | max_low_pfn = 0; | |
466 | ||
1da177e4 LT |
467 | /* These actually end up getting called by call_pernode_memory() */ |
468 | efi_memmap_walk(filter_rsvd_memory, build_node_maps); | |
469 | efi_memmap_walk(filter_rsvd_memory, find_pernode_space); | |
470 | ||
564601a5 | 471 | for_each_online_node(node) |
472 | if (mem_data[node].bootmem_data.node_low_pfn) { | |
473 | node_clear(node, memory_less_mask); | |
474 | mem_data[node].min_pfn = ~0UL; | |
475 | } | |
1da177e4 LT |
476 | /* |
477 | * Initialize the boot memory maps in reverse order since that's | |
478 | * what the bootmem allocator expects | |
479 | */ | |
480 | for (node = MAX_NUMNODES - 1; node >= 0; node--) { | |
481 | unsigned long pernode, pernodesize, map; | |
482 | struct bootmem_data *bdp; | |
483 | ||
484 | if (!node_online(node)) | |
485 | continue; | |
564601a5 | 486 | else if (node_isset(node, memory_less_mask)) |
487 | continue; | |
1da177e4 LT |
488 | |
489 | bdp = &mem_data[node].bootmem_data; | |
490 | pernode = mem_data[node].pernode_addr; | |
491 | pernodesize = mem_data[node].pernode_size; | |
492 | map = pernode + pernodesize; | |
493 | ||
ae5a2c1c | 494 | init_bootmem_node(pgdat_list[node], |
1da177e4 LT |
495 | map>>PAGE_SHIFT, |
496 | bdp->node_boot_start>>PAGE_SHIFT, | |
497 | bdp->node_low_pfn); | |
498 | } | |
499 | ||
500 | efi_memmap_walk(filter_rsvd_memory, free_node_bootmem); | |
501 | ||
502 | reserve_pernode_space(); | |
564601a5 | 503 | memory_less_nodes(); |
1da177e4 LT |
504 | initialize_pernode_data(); |
505 | ||
506 | max_pfn = max_low_pfn; | |
507 | ||
508 | find_initrd(); | |
509 | } | |
510 | ||
8d7e3517 | 511 | #ifdef CONFIG_SMP |
1da177e4 LT |
512 | /** |
513 | * per_cpu_init - setup per-cpu variables | |
514 | * | |
515 | * find_pernode_space() does most of this already, we just need to set | |
516 | * local_per_cpu_offset | |
517 | */ | |
244fd545 | 518 | void __cpuinit *per_cpu_init(void) |
1da177e4 LT |
519 | { |
520 | int cpu; | |
ff741906 AR |
521 | static int first_time = 1; |
522 | ||
1da177e4 | 523 | |
8d7e3517 TL |
524 | if (smp_processor_id() != 0) |
525 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
526 | ||
ff741906 AR |
527 | if (first_time) { |
528 | first_time = 0; | |
529 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
530 | per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu]; | |
531 | } | |
1da177e4 LT |
532 | |
533 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
534 | } | |
8d7e3517 | 535 | #endif /* CONFIG_SMP */ |
1da177e4 | 536 | |
ace1d816 RH |
537 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
538 | static inline int find_next_valid_pfn_for_pgdat(pg_data_t *pgdat, int i) | |
539 | { | |
540 | unsigned long end_address, hole_next_pfn; | |
541 | unsigned long stop_address; | |
542 | ||
543 | end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i]; | |
544 | end_address = PAGE_ALIGN(end_address); | |
545 | ||
546 | stop_address = (unsigned long) &vmem_map[ | |
547 | pgdat->node_start_pfn + pgdat->node_spanned_pages]; | |
548 | ||
549 | do { | |
550 | pgd_t *pgd; | |
551 | pud_t *pud; | |
552 | pmd_t *pmd; | |
553 | pte_t *pte; | |
554 | ||
555 | pgd = pgd_offset_k(end_address); | |
556 | if (pgd_none(*pgd)) { | |
557 | end_address += PGDIR_SIZE; | |
558 | continue; | |
559 | } | |
560 | ||
561 | pud = pud_offset(pgd, end_address); | |
562 | if (pud_none(*pud)) { | |
563 | end_address += PUD_SIZE; | |
564 | continue; | |
565 | } | |
566 | ||
567 | pmd = pmd_offset(pud, end_address); | |
568 | if (pmd_none(*pmd)) { | |
569 | end_address += PMD_SIZE; | |
570 | continue; | |
571 | } | |
572 | ||
573 | pte = pte_offset_kernel(pmd, end_address); | |
574 | retry_pte: | |
575 | if (pte_none(*pte)) { | |
576 | end_address += PAGE_SIZE; | |
577 | pte++; | |
578 | if ((end_address < stop_address) && | |
579 | (end_address != ALIGN(end_address, 1UL << PMD_SHIFT))) | |
580 | goto retry_pte; | |
581 | continue; | |
582 | } | |
583 | /* Found next valid vmem_map page */ | |
584 | break; | |
585 | } while (end_address < stop_address); | |
586 | ||
587 | end_address = min(end_address, stop_address); | |
588 | end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1; | |
589 | hole_next_pfn = end_address / sizeof(struct page); | |
590 | return hole_next_pfn - pgdat->node_start_pfn; | |
591 | } | |
592 | #else | |
593 | static inline int find_next_valid_pfn_for_pgdat(pg_data_t *pgdat, int i) | |
594 | { | |
595 | return i + 1; | |
596 | } | |
597 | #endif | |
598 | ||
1da177e4 LT |
599 | /** |
600 | * show_mem - give short summary of memory stats | |
601 | * | |
602 | * Shows a simple page count of reserved and used pages in the system. | |
603 | * For discontig machines, it does this on a per-pgdat basis. | |
604 | */ | |
605 | void show_mem(void) | |
606 | { | |
607 | int i, total_reserved = 0; | |
608 | int total_shared = 0, total_cached = 0; | |
609 | unsigned long total_present = 0; | |
610 | pg_data_t *pgdat; | |
611 | ||
612 | printk("Mem-info:\n"); | |
613 | show_free_areas(); | |
614 | printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); | |
ec936fc5 | 615 | for_each_online_pgdat(pgdat) { |
208d54e5 DH |
616 | unsigned long present; |
617 | unsigned long flags; | |
1da177e4 | 618 | int shared = 0, cached = 0, reserved = 0; |
208d54e5 | 619 | |
1da177e4 | 620 | printk("Node ID: %d\n", pgdat->node_id); |
208d54e5 DH |
621 | pgdat_resize_lock(pgdat, &flags); |
622 | present = pgdat->node_present_pages; | |
1da177e4 | 623 | for(i = 0; i < pgdat->node_spanned_pages; i++) { |
2d4b1fa2 BP |
624 | struct page *page; |
625 | if (pfn_valid(pgdat->node_start_pfn + i)) | |
626 | page = pfn_to_page(pgdat->node_start_pfn + i); | |
ace1d816 RH |
627 | else { |
628 | i = find_next_valid_pfn_for_pgdat(pgdat, i) - 1; | |
1da177e4 | 629 | continue; |
ace1d816 | 630 | } |
408fde81 | 631 | if (PageReserved(page)) |
1da177e4 | 632 | reserved++; |
408fde81 | 633 | else if (PageSwapCache(page)) |
1da177e4 | 634 | cached++; |
408fde81 DH |
635 | else if (page_count(page)) |
636 | shared += page_count(page)-1; | |
1da177e4 | 637 | } |
208d54e5 | 638 | pgdat_resize_unlock(pgdat, &flags); |
1da177e4 LT |
639 | total_present += present; |
640 | total_reserved += reserved; | |
641 | total_cached += cached; | |
642 | total_shared += shared; | |
643 | printk("\t%ld pages of RAM\n", present); | |
644 | printk("\t%d reserved pages\n", reserved); | |
645 | printk("\t%d pages shared\n", shared); | |
646 | printk("\t%d pages swap cached\n", cached); | |
647 | } | |
648 | printk("%ld pages of RAM\n", total_present); | |
649 | printk("%d reserved pages\n", total_reserved); | |
650 | printk("%d pages shared\n", total_shared); | |
651 | printk("%d pages swap cached\n", total_cached); | |
fde740e4 RH |
652 | printk("Total of %ld pages in page table cache\n", |
653 | pgtable_quicklist_total_size()); | |
1da177e4 LT |
654 | printk("%d free buffer pages\n", nr_free_buffer_pages()); |
655 | } | |
656 | ||
657 | /** | |
658 | * call_pernode_memory - use SRAT to call callback functions with node info | |
659 | * @start: physical start of range | |
660 | * @len: length of range | |
661 | * @arg: function to call for each range | |
662 | * | |
663 | * efi_memmap_walk() knows nothing about layout of memory across nodes. Find | |
664 | * out to which node a block of memory belongs. Ignore memory that we cannot | |
665 | * identify, and split blocks that run across multiple nodes. | |
666 | * | |
667 | * Take this opportunity to round the start address up and the end address | |
668 | * down to page boundaries. | |
669 | */ | |
670 | void call_pernode_memory(unsigned long start, unsigned long len, void *arg) | |
671 | { | |
672 | unsigned long rs, re, end = start + len; | |
673 | void (*func)(unsigned long, unsigned long, int); | |
674 | int i; | |
675 | ||
676 | start = PAGE_ALIGN(start); | |
677 | end &= PAGE_MASK; | |
678 | if (start >= end) | |
679 | return; | |
680 | ||
681 | func = arg; | |
682 | ||
683 | if (!num_node_memblks) { | |
684 | /* No SRAT table, so assume one node (node 0) */ | |
685 | if (start < end) | |
686 | (*func)(start, end - start, 0); | |
687 | return; | |
688 | } | |
689 | ||
690 | for (i = 0; i < num_node_memblks; i++) { | |
691 | rs = max(start, node_memblk[i].start_paddr); | |
692 | re = min(end, node_memblk[i].start_paddr + | |
693 | node_memblk[i].size); | |
694 | ||
695 | if (rs < re) | |
696 | (*func)(rs, re - rs, node_memblk[i].nid); | |
697 | ||
698 | if (re == end) | |
699 | break; | |
700 | } | |
701 | } | |
702 | ||
703 | /** | |
704 | * count_node_pages - callback to build per-node memory info structures | |
705 | * @start: physical start of range | |
706 | * @len: length of range | |
707 | * @node: node where this range resides | |
708 | * | |
709 | * Each node has it's own number of physical pages, DMAable pages, start, and | |
710 | * end page frame number. This routine will be called by call_pernode_memory() | |
711 | * for each piece of usable memory and will setup these values for each node. | |
712 | * Very similar to build_maps(). | |
713 | */ | |
714 | static __init int count_node_pages(unsigned long start, unsigned long len, int node) | |
715 | { | |
716 | unsigned long end = start + len; | |
717 | ||
718 | mem_data[node].num_physpages += len >> PAGE_SHIFT; | |
719 | if (start <= __pa(MAX_DMA_ADDRESS)) | |
720 | mem_data[node].num_dma_physpages += | |
721 | (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT; | |
722 | start = GRANULEROUNDDOWN(start); | |
723 | start = ORDERROUNDDOWN(start); | |
724 | end = GRANULEROUNDUP(end); | |
725 | mem_data[node].max_pfn = max(mem_data[node].max_pfn, | |
726 | end >> PAGE_SHIFT); | |
727 | mem_data[node].min_pfn = min(mem_data[node].min_pfn, | |
728 | start >> PAGE_SHIFT); | |
729 | ||
730 | return 0; | |
731 | } | |
732 | ||
733 | /** | |
734 | * paging_init - setup page tables | |
735 | * | |
736 | * paging_init() sets up the page tables for each node of the system and frees | |
737 | * the bootmem allocator memory for general use. | |
738 | */ | |
739 | void __init paging_init(void) | |
740 | { | |
741 | unsigned long max_dma; | |
742 | unsigned long zones_size[MAX_NR_ZONES]; | |
743 | unsigned long zholes_size[MAX_NR_ZONES]; | |
744 | unsigned long pfn_offset = 0; | |
745 | int node; | |
746 | ||
747 | max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
748 | ||
2d4b1fa2 BP |
749 | arch_sparse_init(); |
750 | ||
1da177e4 LT |
751 | efi_memmap_walk(filter_rsvd_memory, count_node_pages); |
752 | ||
2d4b1fa2 | 753 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
564601a5 | 754 | vmalloc_end -= PAGE_ALIGN(max_low_pfn * sizeof(struct page)); |
755 | vmem_map = (struct page *) vmalloc_end; | |
756 | efi_memmap_walk(create_mem_map_page_table, NULL); | |
757 | printk("Virtual mem_map starts at 0x%p\n", vmem_map); | |
2d4b1fa2 | 758 | #endif |
564601a5 | 759 | |
1da177e4 LT |
760 | for_each_online_node(node) { |
761 | memset(zones_size, 0, sizeof(zones_size)); | |
762 | memset(zholes_size, 0, sizeof(zholes_size)); | |
763 | ||
764 | num_physpages += mem_data[node].num_physpages; | |
765 | ||
766 | if (mem_data[node].min_pfn >= max_dma) { | |
767 | /* All of this node's memory is above ZONE_DMA */ | |
768 | zones_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
769 | mem_data[node].min_pfn; | |
770 | zholes_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
771 | mem_data[node].min_pfn - | |
772 | mem_data[node].num_physpages; | |
773 | } else if (mem_data[node].max_pfn < max_dma) { | |
774 | /* All of this node's memory is in ZONE_DMA */ | |
775 | zones_size[ZONE_DMA] = mem_data[node].max_pfn - | |
776 | mem_data[node].min_pfn; | |
777 | zholes_size[ZONE_DMA] = mem_data[node].max_pfn - | |
778 | mem_data[node].min_pfn - | |
779 | mem_data[node].num_dma_physpages; | |
780 | } else { | |
781 | /* This node has memory in both zones */ | |
782 | zones_size[ZONE_DMA] = max_dma - | |
783 | mem_data[node].min_pfn; | |
784 | zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - | |
785 | mem_data[node].num_dma_physpages; | |
786 | zones_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
787 | max_dma; | |
788 | zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - | |
789 | (mem_data[node].num_physpages - | |
790 | mem_data[node].num_dma_physpages); | |
791 | } | |
792 | ||
1da177e4 LT |
793 | pfn_offset = mem_data[node].min_pfn; |
794 | ||
2d4b1fa2 | 795 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
1da177e4 | 796 | NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset; |
2d4b1fa2 | 797 | #endif |
1da177e4 LT |
798 | free_area_init_node(node, NODE_DATA(node), zones_size, |
799 | pfn_offset, zholes_size); | |
800 | } | |
801 | ||
802 | zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page)); | |
803 | } | |
7049027c | 804 | |
dd0932d9 YG |
805 | pg_data_t *arch_alloc_nodedata(int nid) |
806 | { | |
807 | unsigned long size = compute_pernodesize(nid); | |
808 | ||
809 | return kzalloc(size, GFP_KERNEL); | |
810 | } | |
811 | ||
812 | void arch_free_nodedata(pg_data_t *pgdat) | |
813 | { | |
814 | kfree(pgdat); | |
815 | } | |
816 | ||
7049027c YG |
817 | void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat) |
818 | { | |
819 | pgdat_list[update_node] = update_pgdat; | |
820 | scatter_node_data(); | |
821 | } |