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 | ||
91 | min_low_pfn = min(min_low_pfn, bdp->node_boot_start>>PAGE_SHIFT); | |
92 | max_low_pfn = max(max_low_pfn, bdp->node_low_pfn); | |
93 | ||
94 | return 0; | |
95 | } | |
96 | ||
97 | /** | |
564601a5 | 98 | * early_nr_cpus_node - return number of cpus on a given node |
1da177e4 LT |
99 | * @node: node to check |
100 | * | |
564601a5 | 101 | * Count the number of cpus on @node. We can't use nr_cpus_node() yet because |
1da177e4 | 102 | * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been |
564601a5 | 103 | * called yet. Note that node 0 will also count all non-existent cpus. |
1da177e4 | 104 | */ |
dd0932d9 | 105 | static int __meminit early_nr_cpus_node(int node) |
1da177e4 LT |
106 | { |
107 | int cpu, n = 0; | |
108 | ||
109 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
110 | if (node == node_cpuid[cpu].nid) | |
564601a5 | 111 | n++; |
1da177e4 LT |
112 | |
113 | return n; | |
114 | } | |
115 | ||
564601a5 | 116 | /** |
117 | * compute_pernodesize - compute size of pernode data | |
118 | * @node: the node id. | |
119 | */ | |
dd0932d9 | 120 | static unsigned long __meminit compute_pernodesize(int node) |
564601a5 | 121 | { |
122 | unsigned long pernodesize = 0, cpus; | |
123 | ||
124 | cpus = early_nr_cpus_node(node); | |
125 | pernodesize += PERCPU_PAGE_SIZE * cpus; | |
126 | pernodesize += node * L1_CACHE_BYTES; | |
127 | pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t)); | |
128 | pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
129 | pernodesize = PAGE_ALIGN(pernodesize); | |
130 | return pernodesize; | |
131 | } | |
1da177e4 | 132 | |
8d7e3517 TL |
133 | /** |
134 | * per_cpu_node_setup - setup per-cpu areas on each node | |
135 | * @cpu_data: per-cpu area on this node | |
136 | * @node: node to setup | |
137 | * | |
138 | * Copy the static per-cpu data into the region we just set aside and then | |
139 | * setup __per_cpu_offset for each CPU on this node. Return a pointer to | |
140 | * the end of the area. | |
141 | */ | |
142 | static void *per_cpu_node_setup(void *cpu_data, int node) | |
143 | { | |
144 | #ifdef CONFIG_SMP | |
145 | int cpu; | |
146 | ||
147 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
148 | if (node == node_cpuid[cpu].nid) { | |
149 | memcpy(__va(cpu_data), __phys_per_cpu_start, | |
150 | __per_cpu_end - __per_cpu_start); | |
151 | __per_cpu_offset[cpu] = (char*)__va(cpu_data) - | |
152 | __per_cpu_start; | |
153 | cpu_data += PERCPU_PAGE_SIZE; | |
154 | } | |
155 | } | |
156 | #endif | |
157 | return cpu_data; | |
158 | } | |
159 | ||
1da177e4 | 160 | /** |
564601a5 | 161 | * fill_pernode - initialize pernode data. |
162 | * @node: the node id. | |
163 | * @pernode: physical address of pernode data | |
164 | * @pernodesize: size of the pernode data | |
1da177e4 | 165 | */ |
564601a5 | 166 | static void __init fill_pernode(int node, unsigned long pernode, |
167 | unsigned long pernodesize) | |
1da177e4 | 168 | { |
564601a5 | 169 | void *cpu_data; |
8d7e3517 | 170 | int cpus = early_nr_cpus_node(node); |
564601a5 | 171 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
1da177e4 | 172 | |
564601a5 | 173 | mem_data[node].pernode_addr = pernode; |
174 | mem_data[node].pernode_size = pernodesize; | |
175 | memset(__va(pernode), 0, pernodesize); | |
1da177e4 | 176 | |
564601a5 | 177 | cpu_data = (void *)pernode; |
178 | pernode += PERCPU_PAGE_SIZE * cpus; | |
179 | pernode += node * L1_CACHE_BYTES; | |
180 | ||
ae5a2c1c | 181 | pgdat_list[node] = __va(pernode); |
564601a5 | 182 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
183 | ||
184 | mem_data[node].node_data = __va(pernode); | |
185 | pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data)); | |
186 | ||
ae5a2c1c | 187 | pgdat_list[node]->bdata = bdp; |
564601a5 | 188 | pernode += L1_CACHE_ALIGN(sizeof(pg_data_t)); |
189 | ||
8d7e3517 | 190 | cpu_data = per_cpu_node_setup(cpu_data, node); |
1da177e4 | 191 | |
564601a5 | 192 | return; |
193 | } | |
8d7e3517 | 194 | |
1da177e4 LT |
195 | /** |
196 | * find_pernode_space - allocate memory for memory map and per-node structures | |
197 | * @start: physical start of range | |
198 | * @len: length of range | |
199 | * @node: node where this range resides | |
200 | * | |
201 | * This routine reserves space for the per-cpu data struct, the list of | |
202 | * pg_data_ts and the per-node data struct. Each node will have something like | |
203 | * the following in the first chunk of addr. space large enough to hold it. | |
204 | * | |
205 | * ________________________ | |
206 | * | | | |
207 | * |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first | |
208 | * | PERCPU_PAGE_SIZE * | start and length big enough | |
209 | * | cpus_on_this_node | Node 0 will also have entries for all non-existent cpus. | |
210 | * |------------------------| | |
211 | * | local pg_data_t * | | |
212 | * |------------------------| | |
213 | * | local ia64_node_data | | |
214 | * |------------------------| | |
215 | * | ??? | | |
216 | * |________________________| | |
217 | * | |
218 | * Once this space has been set aside, the bootmem maps are initialized. We | |
219 | * could probably move the allocation of the per-cpu and ia64_node_data space | |
220 | * outside of this function and use alloc_bootmem_node(), but doing it here | |
221 | * is straightforward and we get the alignments we want so... | |
222 | */ | |
223 | static int __init find_pernode_space(unsigned long start, unsigned long len, | |
224 | int node) | |
225 | { | |
564601a5 | 226 | unsigned long epfn; |
1da177e4 | 227 | unsigned long pernodesize = 0, pernode, pages, mapsize; |
1da177e4 LT |
228 | struct bootmem_data *bdp = &mem_data[node].bootmem_data; |
229 | ||
230 | epfn = (start + len) >> PAGE_SHIFT; | |
231 | ||
232 | pages = bdp->node_low_pfn - (bdp->node_boot_start >> PAGE_SHIFT); | |
233 | mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
234 | ||
235 | /* | |
236 | * Make sure this memory falls within this node's usable memory | |
237 | * since we may have thrown some away in build_maps(). | |
238 | */ | |
239 | if (start < bdp->node_boot_start || epfn > bdp->node_low_pfn) | |
240 | return 0; | |
241 | ||
242 | /* Don't setup this node's local space twice... */ | |
243 | if (mem_data[node].pernode_addr) | |
244 | return 0; | |
245 | ||
246 | /* | |
247 | * Calculate total size needed, incl. what's necessary | |
248 | * for good alignment and alias prevention. | |
249 | */ | |
564601a5 | 250 | pernodesize = compute_pernodesize(node); |
1da177e4 LT |
251 | pernode = NODEDATA_ALIGN(start, node); |
252 | ||
253 | /* Is this range big enough for what we want to store here? */ | |
564601a5 | 254 | if (start + len > (pernode + pernodesize + mapsize)) |
255 | fill_pernode(node, pernode, pernodesize); | |
1da177e4 LT |
256 | |
257 | return 0; | |
258 | } | |
259 | ||
260 | /** | |
261 | * free_node_bootmem - free bootmem allocator memory for use | |
262 | * @start: physical start of range | |
263 | * @len: length of range | |
264 | * @node: node where this range resides | |
265 | * | |
266 | * Simply calls the bootmem allocator to free the specified ranged from | |
267 | * the given pg_data_t's bdata struct. After this function has been called | |
268 | * for all the entries in the EFI memory map, the bootmem allocator will | |
269 | * be ready to service allocation requests. | |
270 | */ | |
271 | static int __init free_node_bootmem(unsigned long start, unsigned long len, | |
272 | int node) | |
273 | { | |
ae5a2c1c | 274 | free_bootmem_node(pgdat_list[node], start, len); |
1da177e4 LT |
275 | |
276 | return 0; | |
277 | } | |
278 | ||
279 | /** | |
280 | * reserve_pernode_space - reserve memory for per-node space | |
281 | * | |
282 | * Reserve the space used by the bootmem maps & per-node space in the boot | |
283 | * allocator so that when we actually create the real mem maps we don't | |
284 | * use their memory. | |
285 | */ | |
286 | static void __init reserve_pernode_space(void) | |
287 | { | |
288 | unsigned long base, size, pages; | |
289 | struct bootmem_data *bdp; | |
290 | int node; | |
291 | ||
292 | for_each_online_node(node) { | |
ae5a2c1c | 293 | pg_data_t *pdp = pgdat_list[node]; |
1da177e4 | 294 | |
564601a5 | 295 | if (node_isset(node, memory_less_mask)) |
296 | continue; | |
297 | ||
1da177e4 LT |
298 | bdp = pdp->bdata; |
299 | ||
300 | /* First the bootmem_map itself */ | |
301 | pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT); | |
302 | size = bootmem_bootmap_pages(pages) << PAGE_SHIFT; | |
303 | base = __pa(bdp->node_bootmem_map); | |
304 | reserve_bootmem_node(pdp, base, size); | |
305 | ||
306 | /* Now the per-node space */ | |
307 | size = mem_data[node].pernode_size; | |
308 | base = __pa(mem_data[node].pernode_addr); | |
309 | reserve_bootmem_node(pdp, base, size); | |
310 | } | |
311 | } | |
312 | ||
7049027c YG |
313 | static void __meminit scatter_node_data(void) |
314 | { | |
315 | pg_data_t **dst; | |
316 | int node; | |
317 | ||
dd8041f1 YG |
318 | /* |
319 | * for_each_online_node() can't be used at here. | |
320 | * node_online_map is not set for hot-added nodes at this time, | |
321 | * because we are halfway through initialization of the new node's | |
322 | * structures. If for_each_online_node() is used, a new node's | |
323 | * pg_data_ptrs will be not initialized. Insted of using it, | |
324 | * pgdat_list[] is checked. | |
325 | */ | |
326 | for_each_node(node) { | |
327 | if (pgdat_list[node]) { | |
328 | dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs; | |
329 | memcpy(dst, pgdat_list, sizeof(pgdat_list)); | |
330 | } | |
7049027c YG |
331 | } |
332 | } | |
333 | ||
1da177e4 LT |
334 | /** |
335 | * initialize_pernode_data - fixup per-cpu & per-node pointers | |
336 | * | |
337 | * Each node's per-node area has a copy of the global pg_data_t list, so | |
338 | * we copy that to each node here, as well as setting the per-cpu pointer | |
339 | * to the local node data structure. The active_cpus field of the per-node | |
340 | * structure gets setup by the platform_cpu_init() function later. | |
341 | */ | |
342 | static void __init initialize_pernode_data(void) | |
343 | { | |
8d7e3517 | 344 | int cpu, node; |
1da177e4 | 345 | |
7049027c YG |
346 | scatter_node_data(); |
347 | ||
8d7e3517 | 348 | #ifdef CONFIG_SMP |
1da177e4 LT |
349 | /* Set the node_data pointer for each per-cpu struct */ |
350 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
351 | node = node_cpuid[cpu].nid; | |
352 | per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data; | |
353 | } | |
8d7e3517 TL |
354 | #else |
355 | { | |
356 | struct cpuinfo_ia64 *cpu0_cpu_info; | |
357 | cpu = 0; | |
358 | node = node_cpuid[cpu].nid; | |
359 | cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start + | |
360 | ((char *)&per_cpu__cpu_info - __per_cpu_start)); | |
361 | cpu0_cpu_info->node_data = mem_data[node].node_data; | |
362 | } | |
363 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
364 | } |
365 | ||
564601a5 | 366 | /** |
367 | * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit | |
368 | * node but fall back to any other node when __alloc_bootmem_node fails | |
369 | * for best. | |
370 | * @nid: node id | |
371 | * @pernodesize: size of this node's pernode data | |
564601a5 | 372 | */ |
97835245 | 373 | static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize) |
564601a5 | 374 | { |
375 | void *ptr = NULL; | |
376 | u8 best = 0xff; | |
97835245 | 377 | int bestnode = -1, node, anynode = 0; |
564601a5 | 378 | |
379 | for_each_online_node(node) { | |
380 | if (node_isset(node, memory_less_mask)) | |
381 | continue; | |
382 | else if (node_distance(nid, node) < best) { | |
383 | best = node_distance(nid, node); | |
384 | bestnode = node; | |
385 | } | |
97835245 | 386 | anynode = node; |
564601a5 | 387 | } |
388 | ||
97835245 BP |
389 | if (bestnode == -1) |
390 | bestnode = anynode; | |
391 | ||
ae5a2c1c | 392 | ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize, |
97835245 | 393 | PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); |
564601a5 | 394 | |
564601a5 | 395 | return ptr; |
396 | } | |
397 | ||
564601a5 | 398 | /** |
399 | * memory_less_nodes - allocate and initialize CPU only nodes pernode | |
400 | * information. | |
401 | */ | |
402 | static void __init memory_less_nodes(void) | |
403 | { | |
404 | unsigned long pernodesize; | |
405 | void *pernode; | |
406 | int node; | |
407 | ||
408 | for_each_node_mask(node, memory_less_mask) { | |
409 | pernodesize = compute_pernodesize(node); | |
97835245 | 410 | pernode = memory_less_node_alloc(node, pernodesize); |
564601a5 | 411 | fill_pernode(node, __pa(pernode), pernodesize); |
412 | } | |
413 | ||
414 | return; | |
415 | } | |
416 | ||
1da177e4 LT |
417 | /** |
418 | * find_memory - walk the EFI memory map and setup the bootmem allocator | |
419 | * | |
420 | * Called early in boot to setup the bootmem allocator, and to | |
421 | * allocate the per-cpu and per-node structures. | |
422 | */ | |
423 | void __init find_memory(void) | |
424 | { | |
425 | int node; | |
426 | ||
427 | reserve_memory(); | |
428 | ||
429 | if (num_online_nodes() == 0) { | |
430 | printk(KERN_ERR "node info missing!\n"); | |
431 | node_set_online(0); | |
432 | } | |
433 | ||
564601a5 | 434 | nodes_or(memory_less_mask, memory_less_mask, node_online_map); |
1da177e4 LT |
435 | min_low_pfn = -1; |
436 | max_low_pfn = 0; | |
437 | ||
1da177e4 LT |
438 | /* These actually end up getting called by call_pernode_memory() */ |
439 | efi_memmap_walk(filter_rsvd_memory, build_node_maps); | |
440 | efi_memmap_walk(filter_rsvd_memory, find_pernode_space); | |
441 | ||
564601a5 | 442 | for_each_online_node(node) |
443 | if (mem_data[node].bootmem_data.node_low_pfn) { | |
444 | node_clear(node, memory_less_mask); | |
445 | mem_data[node].min_pfn = ~0UL; | |
446 | } | |
139b8304 BP |
447 | |
448 | efi_memmap_walk(register_active_ranges, NULL); | |
449 | ||
1da177e4 LT |
450 | /* |
451 | * Initialize the boot memory maps in reverse order since that's | |
452 | * what the bootmem allocator expects | |
453 | */ | |
454 | for (node = MAX_NUMNODES - 1; node >= 0; node--) { | |
455 | unsigned long pernode, pernodesize, map; | |
456 | struct bootmem_data *bdp; | |
457 | ||
458 | if (!node_online(node)) | |
459 | continue; | |
564601a5 | 460 | else if (node_isset(node, memory_less_mask)) |
461 | continue; | |
1da177e4 LT |
462 | |
463 | bdp = &mem_data[node].bootmem_data; | |
464 | pernode = mem_data[node].pernode_addr; | |
465 | pernodesize = mem_data[node].pernode_size; | |
466 | map = pernode + pernodesize; | |
467 | ||
ae5a2c1c | 468 | init_bootmem_node(pgdat_list[node], |
1da177e4 LT |
469 | map>>PAGE_SHIFT, |
470 | bdp->node_boot_start>>PAGE_SHIFT, | |
471 | bdp->node_low_pfn); | |
472 | } | |
473 | ||
474 | efi_memmap_walk(filter_rsvd_memory, free_node_bootmem); | |
475 | ||
476 | reserve_pernode_space(); | |
564601a5 | 477 | memory_less_nodes(); |
1da177e4 LT |
478 | initialize_pernode_data(); |
479 | ||
480 | max_pfn = max_low_pfn; | |
481 | ||
482 | find_initrd(); | |
483 | } | |
484 | ||
8d7e3517 | 485 | #ifdef CONFIG_SMP |
1da177e4 LT |
486 | /** |
487 | * per_cpu_init - setup per-cpu variables | |
488 | * | |
489 | * find_pernode_space() does most of this already, we just need to set | |
490 | * local_per_cpu_offset | |
491 | */ | |
244fd545 | 492 | void __cpuinit *per_cpu_init(void) |
1da177e4 LT |
493 | { |
494 | int cpu; | |
ff741906 AR |
495 | static int first_time = 1; |
496 | ||
1da177e4 | 497 | |
8d7e3517 TL |
498 | if (smp_processor_id() != 0) |
499 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
500 | ||
ff741906 AR |
501 | if (first_time) { |
502 | first_time = 0; | |
503 | for (cpu = 0; cpu < NR_CPUS; cpu++) | |
504 | per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu]; | |
505 | } | |
1da177e4 LT |
506 | |
507 | return __per_cpu_start + __per_cpu_offset[smp_processor_id()]; | |
508 | } | |
8d7e3517 | 509 | #endif /* CONFIG_SMP */ |
1da177e4 LT |
510 | |
511 | /** | |
512 | * show_mem - give short summary of memory stats | |
513 | * | |
514 | * Shows a simple page count of reserved and used pages in the system. | |
515 | * For discontig machines, it does this on a per-pgdat basis. | |
516 | */ | |
517 | void show_mem(void) | |
518 | { | |
519 | int i, total_reserved = 0; | |
520 | int total_shared = 0, total_cached = 0; | |
521 | unsigned long total_present = 0; | |
522 | pg_data_t *pgdat; | |
523 | ||
709a6c1c | 524 | printk(KERN_INFO "Mem-info:\n"); |
1da177e4 | 525 | show_free_areas(); |
709a6c1c JS |
526 | printk(KERN_INFO "Free swap: %6ldkB\n", |
527 | nr_swap_pages<<(PAGE_SHIFT-10)); | |
816add4e | 528 | printk(KERN_INFO "Node memory in pages:\n"); |
ec936fc5 | 529 | for_each_online_pgdat(pgdat) { |
208d54e5 DH |
530 | unsigned long present; |
531 | unsigned long flags; | |
1da177e4 | 532 | int shared = 0, cached = 0, reserved = 0; |
208d54e5 | 533 | |
208d54e5 DH |
534 | pgdat_resize_lock(pgdat, &flags); |
535 | present = pgdat->node_present_pages; | |
1da177e4 | 536 | for(i = 0; i < pgdat->node_spanned_pages; i++) { |
2d4b1fa2 BP |
537 | struct page *page; |
538 | if (pfn_valid(pgdat->node_start_pfn + i)) | |
539 | page = pfn_to_page(pgdat->node_start_pfn + i); | |
ace1d816 | 540 | else { |
e44e41d0 BP |
541 | i = vmemmap_find_next_valid_pfn(pgdat->node_id, |
542 | i) - 1; | |
1da177e4 | 543 | continue; |
ace1d816 | 544 | } |
408fde81 | 545 | if (PageReserved(page)) |
1da177e4 | 546 | reserved++; |
408fde81 | 547 | else if (PageSwapCache(page)) |
1da177e4 | 548 | cached++; |
408fde81 DH |
549 | else if (page_count(page)) |
550 | shared += page_count(page)-1; | |
1da177e4 | 551 | } |
208d54e5 | 552 | pgdat_resize_unlock(pgdat, &flags); |
1da177e4 LT |
553 | total_present += present; |
554 | total_reserved += reserved; | |
555 | total_cached += cached; | |
556 | total_shared += shared; | |
816add4e JS |
557 | printk(KERN_INFO "Node %4d: RAM: %11ld, rsvd: %8d, " |
558 | "shrd: %10d, swpd: %10d\n", pgdat->node_id, | |
559 | present, reserved, shared, cached); | |
1da177e4 | 560 | } |
709a6c1c JS |
561 | printk(KERN_INFO "%ld pages of RAM\n", total_present); |
562 | printk(KERN_INFO "%d reserved pages\n", total_reserved); | |
563 | printk(KERN_INFO "%d pages shared\n", total_shared); | |
564 | printk(KERN_INFO "%d pages swap cached\n", total_cached); | |
565 | printk(KERN_INFO "Total of %ld pages in page table cache\n", | |
566 | pgtable_quicklist_total_size()); | |
567 | printk(KERN_INFO "%d free buffer pages\n", nr_free_buffer_pages()); | |
1da177e4 LT |
568 | } |
569 | ||
570 | /** | |
571 | * call_pernode_memory - use SRAT to call callback functions with node info | |
572 | * @start: physical start of range | |
573 | * @len: length of range | |
574 | * @arg: function to call for each range | |
575 | * | |
576 | * efi_memmap_walk() knows nothing about layout of memory across nodes. Find | |
577 | * out to which node a block of memory belongs. Ignore memory that we cannot | |
578 | * identify, and split blocks that run across multiple nodes. | |
579 | * | |
580 | * Take this opportunity to round the start address up and the end address | |
581 | * down to page boundaries. | |
582 | */ | |
583 | void call_pernode_memory(unsigned long start, unsigned long len, void *arg) | |
584 | { | |
585 | unsigned long rs, re, end = start + len; | |
586 | void (*func)(unsigned long, unsigned long, int); | |
587 | int i; | |
588 | ||
589 | start = PAGE_ALIGN(start); | |
590 | end &= PAGE_MASK; | |
591 | if (start >= end) | |
592 | return; | |
593 | ||
594 | func = arg; | |
595 | ||
596 | if (!num_node_memblks) { | |
597 | /* No SRAT table, so assume one node (node 0) */ | |
598 | if (start < end) | |
599 | (*func)(start, end - start, 0); | |
600 | return; | |
601 | } | |
602 | ||
603 | for (i = 0; i < num_node_memblks; i++) { | |
604 | rs = max(start, node_memblk[i].start_paddr); | |
605 | re = min(end, node_memblk[i].start_paddr + | |
606 | node_memblk[i].size); | |
607 | ||
608 | if (rs < re) | |
609 | (*func)(rs, re - rs, node_memblk[i].nid); | |
610 | ||
611 | if (re == end) | |
612 | break; | |
613 | } | |
614 | } | |
615 | ||
616 | /** | |
617 | * count_node_pages - callback to build per-node memory info structures | |
618 | * @start: physical start of range | |
619 | * @len: length of range | |
620 | * @node: node where this range resides | |
621 | * | |
622 | * Each node has it's own number of physical pages, DMAable pages, start, and | |
623 | * end page frame number. This routine will be called by call_pernode_memory() | |
624 | * for each piece of usable memory and will setup these values for each node. | |
625 | * Very similar to build_maps(). | |
626 | */ | |
627 | static __init int count_node_pages(unsigned long start, unsigned long len, int node) | |
628 | { | |
629 | unsigned long end = start + len; | |
630 | ||
631 | mem_data[node].num_physpages += len >> PAGE_SHIFT; | |
09ae1f58 | 632 | #ifdef CONFIG_ZONE_DMA |
1da177e4 LT |
633 | if (start <= __pa(MAX_DMA_ADDRESS)) |
634 | mem_data[node].num_dma_physpages += | |
635 | (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT; | |
09ae1f58 | 636 | #endif |
1da177e4 LT |
637 | start = GRANULEROUNDDOWN(start); |
638 | start = ORDERROUNDDOWN(start); | |
639 | end = GRANULEROUNDUP(end); | |
640 | mem_data[node].max_pfn = max(mem_data[node].max_pfn, | |
641 | end >> PAGE_SHIFT); | |
642 | mem_data[node].min_pfn = min(mem_data[node].min_pfn, | |
643 | start >> PAGE_SHIFT); | |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
648 | /** | |
649 | * paging_init - setup page tables | |
650 | * | |
651 | * paging_init() sets up the page tables for each node of the system and frees | |
652 | * the bootmem allocator memory for general use. | |
653 | */ | |
654 | void __init paging_init(void) | |
655 | { | |
656 | unsigned long max_dma; | |
1da177e4 | 657 | unsigned long pfn_offset = 0; |
05e0caad | 658 | unsigned long max_pfn = 0; |
1da177e4 | 659 | int node; |
05e0caad | 660 | unsigned long max_zone_pfns[MAX_NR_ZONES]; |
1da177e4 LT |
661 | |
662 | max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
663 | ||
1da177e4 LT |
664 | efi_memmap_walk(filter_rsvd_memory, count_node_pages); |
665 | ||
524fd988 BP |
666 | sparse_memory_present_with_active_regions(MAX_NUMNODES); |
667 | sparse_init(); | |
668 | ||
2d4b1fa2 | 669 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
921eea1c BP |
670 | vmalloc_end -= PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) * |
671 | sizeof(struct page)); | |
564601a5 | 672 | vmem_map = (struct page *) vmalloc_end; |
673 | efi_memmap_walk(create_mem_map_page_table, NULL); | |
674 | printk("Virtual mem_map starts at 0x%p\n", vmem_map); | |
2d4b1fa2 | 675 | #endif |
564601a5 | 676 | |
1da177e4 | 677 | for_each_online_node(node) { |
1da177e4 | 678 | num_physpages += mem_data[node].num_physpages; |
1da177e4 LT |
679 | pfn_offset = mem_data[node].min_pfn; |
680 | ||
2d4b1fa2 | 681 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
1da177e4 | 682 | NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset; |
2d4b1fa2 | 683 | #endif |
05e0caad MG |
684 | if (mem_data[node].max_pfn > max_pfn) |
685 | max_pfn = mem_data[node].max_pfn; | |
1da177e4 LT |
686 | } |
687 | ||
6391af17 | 688 | memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); |
09ae1f58 | 689 | #ifdef CONFIG_ZONE_DMA |
05e0caad | 690 | max_zone_pfns[ZONE_DMA] = max_dma; |
09ae1f58 | 691 | #endif |
05e0caad MG |
692 | max_zone_pfns[ZONE_NORMAL] = max_pfn; |
693 | free_area_init_nodes(max_zone_pfns); | |
694 | ||
1da177e4 LT |
695 | zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page)); |
696 | } | |
7049027c | 697 | |
dd0932d9 YG |
698 | pg_data_t *arch_alloc_nodedata(int nid) |
699 | { | |
700 | unsigned long size = compute_pernodesize(nid); | |
701 | ||
702 | return kzalloc(size, GFP_KERNEL); | |
703 | } | |
704 | ||
705 | void arch_free_nodedata(pg_data_t *pgdat) | |
706 | { | |
707 | kfree(pgdat); | |
708 | } | |
709 | ||
7049027c YG |
710 | void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat) |
711 | { | |
712 | pgdat_list[update_node] = update_pgdat; | |
713 | scatter_node_data(); | |
714 | } |