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; | |
36 | pg_data_t *pgdat; | |
37 | unsigned long pernode_addr; | |
38 | unsigned long pernode_size; | |
39 | struct bootmem_data bootmem_data; | |
40 | unsigned long num_physpages; | |
41 | unsigned long num_dma_physpages; | |
42 | unsigned long min_pfn; | |
43 | unsigned long max_pfn; | |
44 | }; | |
45 | ||
46 | static struct early_node_data mem_data[MAX_NUMNODES] __initdata; | |
564601a5 | 47 | static nodemask_t memory_less_mask __initdata; |
1da177e4 LT |
48 | |
49 | /* | |
50 | * To prevent cache aliasing effects, align per-node structures so that they | |
51 | * start at addresses that are strided by node number. | |
52 | */ | |
acb7f672 | 53 | #define MAX_NODE_ALIGN_OFFSET (32 * 1024 * 1024) |
1da177e4 | 54 | #define NODEDATA_ALIGN(addr, node) \ |
acb7f672 JS |
55 | ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + \ |
56 | (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1))) | |
1da177e4 LT |
57 | |
58 | /** | |
59 | * build_node_maps - callback to setup bootmem structs for each node | |
60 | * @start: physical start of range | |
61 | * @len: length of range | |
62 | * @node: node where this range resides | |
63 | * | |
64 | * We allocate a struct bootmem_data for each piece of memory that we wish to | |
65 | * treat as a virtually contiguous block (i.e. each node). Each such block | |
66 | * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down | |
67 | * if necessary. Any non-existent pages will simply be part of the virtual | |
68 | * memmap. We also update min_low_pfn and max_low_pfn here as we receive | |
69 | * memory ranges from the caller. | |
70 | */ | |
71 | static int __init build_node_maps(unsigned long start, unsigned long len, | |
72 | int node) | |
73 | { | |
74 | unsigned long cstart, epfn, end = start + len; | |
75 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; | |
76 | ||
77 | epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT; | |
78 | cstart = GRANULEROUNDDOWN(start); | |
79 | ||
80 | if (!bdp->node_low_pfn) { | |
81 | bdp->node_boot_start = cstart; | |
82 | bdp->node_low_pfn = epfn; | |
83 | } else { | |
84 | bdp->node_boot_start = min(cstart, bdp->node_boot_start); | |
85 | bdp->node_low_pfn = max(epfn, bdp->node_low_pfn); | |
86 | } | |
87 | ||
88 | min_low_pfn = min(min_low_pfn, bdp->node_boot_start>>PAGE_SHIFT); | |
89 | max_low_pfn = max(max_low_pfn, bdp->node_low_pfn); | |
90 | ||
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 | */ |
564601a5 | 102 | static int __init 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 | */ | |
117 | static unsigned long __init compute_pernodesize(int node) | |
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 | ||
178 | mem_data[node].pgdat = __va(pernode); | |
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 | ||
184 | mem_data[node].pgdat->bdata = bdp; | |
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 | { | |
271 | free_bootmem_node(mem_data[node].pgdat, start, len); | |
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) { | |
290 | pg_data_t *pdp = mem_data[node].pgdat; | |
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 | ||
310 | /** | |
311 | * initialize_pernode_data - fixup per-cpu & per-node pointers | |
312 | * | |
313 | * Each node's per-node area has a copy of the global pg_data_t list, so | |
314 | * we copy that to each node here, as well as setting the per-cpu pointer | |
315 | * to the local node data structure. The active_cpus field of the per-node | |
316 | * structure gets setup by the platform_cpu_init() function later. | |
317 | */ | |
318 | static void __init initialize_pernode_data(void) | |
319 | { | |
1da177e4 | 320 | pg_data_t *pgdat_list[MAX_NUMNODES]; |
8d7e3517 | 321 | int cpu, node; |
1da177e4 LT |
322 | |
323 | for_each_online_node(node) | |
324 | pgdat_list[node] = mem_data[node].pgdat; | |
325 | ||
326 | /* Copy the pg_data_t list to each node and init the node field */ | |
327 | for_each_online_node(node) { | |
328 | memcpy(mem_data[node].node_data->pg_data_ptrs, pgdat_list, | |
329 | sizeof(pgdat_list)); | |
330 | } | |
8d7e3517 | 331 | #ifdef CONFIG_SMP |
1da177e4 LT |
332 | /* Set the node_data pointer for each per-cpu struct */ |
333 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
334 | node = node_cpuid[cpu].nid; | |
335 | per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data; | |
336 | } | |
8d7e3517 TL |
337 | #else |
338 | { | |
339 | struct cpuinfo_ia64 *cpu0_cpu_info; | |
340 | cpu = 0; | |
341 | node = node_cpuid[cpu].nid; | |
342 | cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start + | |
343 | ((char *)&per_cpu__cpu_info - __per_cpu_start)); | |
344 | cpu0_cpu_info->node_data = mem_data[node].node_data; | |
345 | } | |
346 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
347 | } |
348 | ||
564601a5 | 349 | /** |
350 | * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit | |
351 | * node but fall back to any other node when __alloc_bootmem_node fails | |
352 | * for best. | |
353 | * @nid: node id | |
354 | * @pernodesize: size of this node's pernode data | |
564601a5 | 355 | */ |
97835245 | 356 | static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize) |
564601a5 | 357 | { |
358 | void *ptr = NULL; | |
359 | u8 best = 0xff; | |
97835245 | 360 | int bestnode = -1, node, anynode = 0; |
564601a5 | 361 | |
362 | for_each_online_node(node) { | |
363 | if (node_isset(node, memory_less_mask)) | |
364 | continue; | |
365 | else if (node_distance(nid, node) < best) { | |
366 | best = node_distance(nid, node); | |
367 | bestnode = node; | |
368 | } | |
97835245 | 369 | anynode = node; |
564601a5 | 370 | } |
371 | ||
97835245 BP |
372 | if (bestnode == -1) |
373 | bestnode = anynode; | |
374 | ||
375 | ptr = __alloc_bootmem_node(mem_data[bestnode].pgdat, pernodesize, | |
376 | PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); | |
564601a5 | 377 | |
564601a5 | 378 | return ptr; |
379 | } | |
380 | ||
381 | /** | |
382 | * pgdat_insert - insert the pgdat into global pgdat_list | |
383 | * @pgdat: the pgdat for a node. | |
384 | */ | |
385 | static void __init pgdat_insert(pg_data_t *pgdat) | |
386 | { | |
387 | pg_data_t *prev = NULL, *next; | |
388 | ||
389 | for_each_pgdat(next) | |
390 | if (pgdat->node_id < next->node_id) | |
391 | break; | |
392 | else | |
393 | prev = next; | |
394 | ||
395 | if (prev) { | |
396 | prev->pgdat_next = pgdat; | |
397 | pgdat->pgdat_next = next; | |
398 | } else { | |
399 | pgdat->pgdat_next = pgdat_list; | |
400 | pgdat_list = pgdat; | |
401 | } | |
402 | ||
403 | return; | |
404 | } | |
405 | ||
406 | /** | |
407 | * memory_less_nodes - allocate and initialize CPU only nodes pernode | |
408 | * information. | |
409 | */ | |
410 | static void __init memory_less_nodes(void) | |
411 | { | |
412 | unsigned long pernodesize; | |
413 | void *pernode; | |
414 | int node; | |
415 | ||
416 | for_each_node_mask(node, memory_less_mask) { | |
417 | pernodesize = compute_pernodesize(node); | |
97835245 | 418 | pernode = memory_less_node_alloc(node, pernodesize); |
564601a5 | 419 | fill_pernode(node, __pa(pernode), pernodesize); |
420 | } | |
421 | ||
422 | return; | |
423 | } | |
424 | ||
2d4b1fa2 BP |
425 | #ifdef CONFIG_SPARSEMEM |
426 | /** | |
427 | * register_sparse_mem - notify SPARSEMEM that this memory range exists. | |
428 | * @start: physical start of range | |
429 | * @end: physical end of range | |
430 | * @arg: unused | |
431 | * | |
432 | * Simply calls SPARSEMEM to register memory section(s). | |
433 | */ | |
434 | static int __init register_sparse_mem(unsigned long start, unsigned long end, | |
435 | void *arg) | |
436 | { | |
437 | int nid; | |
438 | ||
439 | start = __pa(start) >> PAGE_SHIFT; | |
440 | end = __pa(end) >> PAGE_SHIFT; | |
441 | nid = early_pfn_to_nid(start); | |
442 | memory_present(nid, start, end); | |
443 | ||
444 | return 0; | |
445 | } | |
446 | ||
447 | static void __init arch_sparse_init(void) | |
448 | { | |
449 | efi_memmap_walk(register_sparse_mem, NULL); | |
450 | sparse_init(); | |
451 | } | |
452 | #else | |
453 | #define arch_sparse_init() do {} while (0) | |
454 | #endif | |
455 | ||
1da177e4 LT |
456 | /** |
457 | * find_memory - walk the EFI memory map and setup the bootmem allocator | |
458 | * | |
459 | * Called early in boot to setup the bootmem allocator, and to | |
460 | * allocate the per-cpu and per-node structures. | |
461 | */ | |
462 | void __init find_memory(void) | |
463 | { | |
464 | int node; | |
465 | ||
466 | reserve_memory(); | |
467 | ||
468 | if (num_online_nodes() == 0) { | |
469 | printk(KERN_ERR "node info missing!\n"); | |
470 | node_set_online(0); | |
471 | } | |
472 | ||
564601a5 | 473 | nodes_or(memory_less_mask, memory_less_mask, node_online_map); |
1da177e4 LT |
474 | min_low_pfn = -1; |
475 | max_low_pfn = 0; | |
476 | ||
1da177e4 LT |
477 | /* These actually end up getting called by call_pernode_memory() */ |
478 | efi_memmap_walk(filter_rsvd_memory, build_node_maps); | |
479 | efi_memmap_walk(filter_rsvd_memory, find_pernode_space); | |
480 | ||
564601a5 | 481 | for_each_online_node(node) |
482 | if (mem_data[node].bootmem_data.node_low_pfn) { | |
483 | node_clear(node, memory_less_mask); | |
484 | mem_data[node].min_pfn = ~0UL; | |
485 | } | |
1da177e4 LT |
486 | /* |
487 | * Initialize the boot memory maps in reverse order since that's | |
488 | * what the bootmem allocator expects | |
489 | */ | |
490 | for (node = MAX_NUMNODES - 1; node >= 0; node--) { | |
491 | unsigned long pernode, pernodesize, map; | |
492 | struct bootmem_data *bdp; | |
493 | ||
494 | if (!node_online(node)) | |
495 | continue; | |
564601a5 | 496 | else if (node_isset(node, memory_less_mask)) |
497 | continue; | |
1da177e4 LT |
498 | |
499 | bdp = &mem_data[node].bootmem_data; | |
500 | pernode = mem_data[node].pernode_addr; | |
501 | pernodesize = mem_data[node].pernode_size; | |
502 | map = pernode + pernodesize; | |
503 | ||
1da177e4 LT |
504 | init_bootmem_node(mem_data[node].pgdat, |
505 | map>>PAGE_SHIFT, | |
506 | bdp->node_boot_start>>PAGE_SHIFT, | |
507 | bdp->node_low_pfn); | |
508 | } | |
509 | ||
510 | efi_memmap_walk(filter_rsvd_memory, free_node_bootmem); | |
511 | ||
512 | reserve_pernode_space(); | |
564601a5 | 513 | memory_less_nodes(); |
1da177e4 LT |
514 | initialize_pernode_data(); |
515 | ||
516 | max_pfn = max_low_pfn; | |
517 | ||
518 | find_initrd(); | |
519 | } | |
520 | ||
8d7e3517 | 521 | #ifdef CONFIG_SMP |
1da177e4 LT |
522 | /** |
523 | * per_cpu_init - setup per-cpu variables | |
524 | * | |
525 | * find_pernode_space() does most of this already, we just need to set | |
526 | * local_per_cpu_offset | |
527 | */ | |
244fd545 | 528 | void __cpuinit *per_cpu_init(void) |
1da177e4 LT |
529 | { |
530 | int cpu; | |
ff741906 AR |
531 | static int first_time = 1; |
532 | ||
1da177e4 | 533 | |
8d7e3517 TL |
534 | if (smp_processor_id() != 0) |
535 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
536 | ||
ff741906 AR |
537 | if (first_time) { |
538 | first_time = 0; | |
539 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
540 | per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu]; | |
541 | } | |
1da177e4 LT |
542 | |
543 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
544 | } | |
8d7e3517 | 545 | #endif /* CONFIG_SMP */ |
1da177e4 LT |
546 | |
547 | /** | |
548 | * show_mem - give short summary of memory stats | |
549 | * | |
550 | * Shows a simple page count of reserved and used pages in the system. | |
551 | * For discontig machines, it does this on a per-pgdat basis. | |
552 | */ | |
553 | void show_mem(void) | |
554 | { | |
555 | int i, total_reserved = 0; | |
556 | int total_shared = 0, total_cached = 0; | |
557 | unsigned long total_present = 0; | |
558 | pg_data_t *pgdat; | |
559 | ||
560 | printk("Mem-info:\n"); | |
561 | show_free_areas(); | |
562 | printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); | |
563 | for_each_pgdat(pgdat) { | |
208d54e5 DH |
564 | unsigned long present; |
565 | unsigned long flags; | |
1da177e4 | 566 | int shared = 0, cached = 0, reserved = 0; |
208d54e5 | 567 | |
1da177e4 | 568 | printk("Node ID: %d\n", pgdat->node_id); |
208d54e5 DH |
569 | pgdat_resize_lock(pgdat, &flags); |
570 | present = pgdat->node_present_pages; | |
1da177e4 | 571 | for(i = 0; i < pgdat->node_spanned_pages; i++) { |
2d4b1fa2 BP |
572 | struct page *page; |
573 | if (pfn_valid(pgdat->node_start_pfn + i)) | |
574 | page = pfn_to_page(pgdat->node_start_pfn + i); | |
575 | else | |
1da177e4 | 576 | continue; |
408fde81 | 577 | if (PageReserved(page)) |
1da177e4 | 578 | reserved++; |
408fde81 | 579 | else if (PageSwapCache(page)) |
1da177e4 | 580 | cached++; |
408fde81 DH |
581 | else if (page_count(page)) |
582 | shared += page_count(page)-1; | |
1da177e4 | 583 | } |
208d54e5 | 584 | pgdat_resize_unlock(pgdat, &flags); |
1da177e4 LT |
585 | total_present += present; |
586 | total_reserved += reserved; | |
587 | total_cached += cached; | |
588 | total_shared += shared; | |
589 | printk("\t%ld pages of RAM\n", present); | |
590 | printk("\t%d reserved pages\n", reserved); | |
591 | printk("\t%d pages shared\n", shared); | |
592 | printk("\t%d pages swap cached\n", cached); | |
593 | } | |
594 | printk("%ld pages of RAM\n", total_present); | |
595 | printk("%d reserved pages\n", total_reserved); | |
596 | printk("%d pages shared\n", total_shared); | |
597 | printk("%d pages swap cached\n", total_cached); | |
fde740e4 RH |
598 | printk("Total of %ld pages in page table cache\n", |
599 | pgtable_quicklist_total_size()); | |
1da177e4 LT |
600 | printk("%d free buffer pages\n", nr_free_buffer_pages()); |
601 | } | |
602 | ||
603 | /** | |
604 | * call_pernode_memory - use SRAT to call callback functions with node info | |
605 | * @start: physical start of range | |
606 | * @len: length of range | |
607 | * @arg: function to call for each range | |
608 | * | |
609 | * efi_memmap_walk() knows nothing about layout of memory across nodes. Find | |
610 | * out to which node a block of memory belongs. Ignore memory that we cannot | |
611 | * identify, and split blocks that run across multiple nodes. | |
612 | * | |
613 | * Take this opportunity to round the start address up and the end address | |
614 | * down to page boundaries. | |
615 | */ | |
616 | void call_pernode_memory(unsigned long start, unsigned long len, void *arg) | |
617 | { | |
618 | unsigned long rs, re, end = start + len; | |
619 | void (*func)(unsigned long, unsigned long, int); | |
620 | int i; | |
621 | ||
622 | start = PAGE_ALIGN(start); | |
623 | end &= PAGE_MASK; | |
624 | if (start >= end) | |
625 | return; | |
626 | ||
627 | func = arg; | |
628 | ||
629 | if (!num_node_memblks) { | |
630 | /* No SRAT table, so assume one node (node 0) */ | |
631 | if (start < end) | |
632 | (*func)(start, end - start, 0); | |
633 | return; | |
634 | } | |
635 | ||
636 | for (i = 0; i < num_node_memblks; i++) { | |
637 | rs = max(start, node_memblk[i].start_paddr); | |
638 | re = min(end, node_memblk[i].start_paddr + | |
639 | node_memblk[i].size); | |
640 | ||
641 | if (rs < re) | |
642 | (*func)(rs, re - rs, node_memblk[i].nid); | |
643 | ||
644 | if (re == end) | |
645 | break; | |
646 | } | |
647 | } | |
648 | ||
649 | /** | |
650 | * count_node_pages - callback to build per-node memory info structures | |
651 | * @start: physical start of range | |
652 | * @len: length of range | |
653 | * @node: node where this range resides | |
654 | * | |
655 | * Each node has it's own number of physical pages, DMAable pages, start, and | |
656 | * end page frame number. This routine will be called by call_pernode_memory() | |
657 | * for each piece of usable memory and will setup these values for each node. | |
658 | * Very similar to build_maps(). | |
659 | */ | |
660 | static __init int count_node_pages(unsigned long start, unsigned long len, int node) | |
661 | { | |
662 | unsigned long end = start + len; | |
663 | ||
664 | mem_data[node].num_physpages += len >> PAGE_SHIFT; | |
665 | if (start <= __pa(MAX_DMA_ADDRESS)) | |
666 | mem_data[node].num_dma_physpages += | |
667 | (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT; | |
668 | start = GRANULEROUNDDOWN(start); | |
669 | start = ORDERROUNDDOWN(start); | |
670 | end = GRANULEROUNDUP(end); | |
671 | mem_data[node].max_pfn = max(mem_data[node].max_pfn, | |
672 | end >> PAGE_SHIFT); | |
673 | mem_data[node].min_pfn = min(mem_data[node].min_pfn, | |
674 | start >> PAGE_SHIFT); | |
675 | ||
676 | return 0; | |
677 | } | |
678 | ||
679 | /** | |
680 | * paging_init - setup page tables | |
681 | * | |
682 | * paging_init() sets up the page tables for each node of the system and frees | |
683 | * the bootmem allocator memory for general use. | |
684 | */ | |
685 | void __init paging_init(void) | |
686 | { | |
687 | unsigned long max_dma; | |
688 | unsigned long zones_size[MAX_NR_ZONES]; | |
689 | unsigned long zholes_size[MAX_NR_ZONES]; | |
690 | unsigned long pfn_offset = 0; | |
691 | int node; | |
692 | ||
693 | max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
694 | ||
2d4b1fa2 BP |
695 | arch_sparse_init(); |
696 | ||
1da177e4 LT |
697 | efi_memmap_walk(filter_rsvd_memory, count_node_pages); |
698 | ||
2d4b1fa2 | 699 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
564601a5 | 700 | vmalloc_end -= PAGE_ALIGN(max_low_pfn * sizeof(struct page)); |
701 | vmem_map = (struct page *) vmalloc_end; | |
702 | efi_memmap_walk(create_mem_map_page_table, NULL); | |
703 | printk("Virtual mem_map starts at 0x%p\n", vmem_map); | |
2d4b1fa2 | 704 | #endif |
564601a5 | 705 | |
1da177e4 LT |
706 | for_each_online_node(node) { |
707 | memset(zones_size, 0, sizeof(zones_size)); | |
708 | memset(zholes_size, 0, sizeof(zholes_size)); | |
709 | ||
710 | num_physpages += mem_data[node].num_physpages; | |
711 | ||
712 | if (mem_data[node].min_pfn >= max_dma) { | |
713 | /* All of this node's memory is above ZONE_DMA */ | |
714 | zones_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
715 | mem_data[node].min_pfn; | |
716 | zholes_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
717 | mem_data[node].min_pfn - | |
718 | mem_data[node].num_physpages; | |
719 | } else if (mem_data[node].max_pfn < max_dma) { | |
720 | /* All of this node's memory is in ZONE_DMA */ | |
721 | zones_size[ZONE_DMA] = mem_data[node].max_pfn - | |
722 | mem_data[node].min_pfn; | |
723 | zholes_size[ZONE_DMA] = mem_data[node].max_pfn - | |
724 | mem_data[node].min_pfn - | |
725 | mem_data[node].num_dma_physpages; | |
726 | } else { | |
727 | /* This node has memory in both zones */ | |
728 | zones_size[ZONE_DMA] = max_dma - | |
729 | mem_data[node].min_pfn; | |
730 | zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - | |
731 | mem_data[node].num_dma_physpages; | |
732 | zones_size[ZONE_NORMAL] = mem_data[node].max_pfn - | |
733 | max_dma; | |
734 | zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - | |
735 | (mem_data[node].num_physpages - | |
736 | mem_data[node].num_dma_physpages); | |
737 | } | |
738 | ||
1da177e4 LT |
739 | pfn_offset = mem_data[node].min_pfn; |
740 | ||
2d4b1fa2 | 741 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
1da177e4 | 742 | NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset; |
2d4b1fa2 | 743 | #endif |
1da177e4 LT |
744 | free_area_init_node(node, NODE_DATA(node), zones_size, |
745 | pfn_offset, zholes_size); | |
746 | } | |
747 | ||
564601a5 | 748 | /* |
749 | * Make memory less nodes become a member of the known nodes. | |
750 | */ | |
751 | for_each_node_mask(node, memory_less_mask) | |
752 | pgdat_insert(mem_data[node].pgdat); | |
753 | ||
1da177e4 LT |
754 | zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page)); |
755 | } |