Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/arm/mm/init.c | |
3 | * | |
90072059 | 4 | * Copyright (C) 1995-2005 Russell King |
1da177e4 LT |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 as | |
8 | * published by the Free Software Foundation. | |
9 | */ | |
1da177e4 LT |
10 | #include <linux/kernel.h> |
11 | #include <linux/errno.h> | |
1da177e4 LT |
12 | #include <linux/swap.h> |
13 | #include <linux/init.h> | |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/mman.h> | |
16 | #include <linux/nodemask.h> | |
17 | #include <linux/initrd.h> | |
18 | ||
19 | #include <asm/mach-types.h> | |
1da177e4 | 20 | #include <asm/setup.h> |
74d02fb9 | 21 | #include <asm/sizes.h> |
1da177e4 LT |
22 | #include <asm/tlb.h> |
23 | ||
24 | #include <asm/mach/arch.h> | |
25 | #include <asm/mach/map.h> | |
26 | ||
1b2e2b73 RK |
27 | #include "mm.h" |
28 | ||
d111e8f9 | 29 | extern void _text, _etext, __data_start, _end, __init_begin, __init_end; |
1da177e4 LT |
30 | extern unsigned long phys_initrd_start; |
31 | extern unsigned long phys_initrd_size; | |
32 | ||
33 | /* | |
5e709827 RL |
34 | * This is used to pass memory configuration data from paging_init |
35 | * to mem_init, and by show_mem() to skip holes in the memory map. | |
1da177e4 | 36 | */ |
5e709827 RL |
37 | static struct meminfo meminfo = { 0, }; |
38 | ||
39 | #define for_each_nodebank(iter,mi,no) \ | |
40 | for (iter = 0; iter < mi->nr_banks; iter++) \ | |
41 | if (mi->bank[iter].node == no) | |
1da177e4 | 42 | |
1da177e4 LT |
43 | void show_mem(void) |
44 | { | |
45 | int free = 0, total = 0, reserved = 0; | |
5e709827 RL |
46 | int shared = 0, cached = 0, slab = 0, node, i; |
47 | struct meminfo * mi = &meminfo; | |
1da177e4 LT |
48 | |
49 | printk("Mem-info:\n"); | |
50 | show_free_areas(); | |
1da177e4 | 51 | for_each_online_node(node) { |
204ecae4 RK |
52 | pg_data_t *n = NODE_DATA(node); |
53 | struct page *map = n->node_mem_map - n->node_start_pfn; | |
54 | ||
5e709827 RL |
55 | for_each_nodebank (i,mi,node) { |
56 | unsigned int pfn1, pfn2; | |
57 | struct page *page, *end; | |
58 | ||
204ecae4 RK |
59 | pfn1 = __phys_to_pfn(mi->bank[i].start); |
60 | pfn2 = __phys_to_pfn(mi->bank[i].size + mi->bank[i].start); | |
5e709827 | 61 | |
204ecae4 RK |
62 | page = map + pfn1; |
63 | end = map + pfn2; | |
5e709827 RL |
64 | |
65 | do { | |
66 | total++; | |
67 | if (PageReserved(page)) | |
68 | reserved++; | |
69 | else if (PageSwapCache(page)) | |
70 | cached++; | |
71 | else if (PageSlab(page)) | |
72 | slab++; | |
73 | else if (!page_count(page)) | |
74 | free++; | |
75 | else | |
76 | shared += page_count(page) - 1; | |
77 | page++; | |
78 | } while (page < end); | |
79 | } | |
1da177e4 LT |
80 | } |
81 | ||
82 | printk("%d pages of RAM\n", total); | |
83 | printk("%d free pages\n", free); | |
84 | printk("%d reserved pages\n", reserved); | |
85 | printk("%d slab pages\n", slab); | |
86 | printk("%d pages shared\n", shared); | |
87 | printk("%d pages swap cached\n", cached); | |
88 | } | |
89 | ||
1da177e4 LT |
90 | /* |
91 | * FIXME: We really want to avoid allocating the bootmap bitmap | |
92 | * over the top of the initrd. Hopefully, this is located towards | |
93 | * the start of a bank, so if we allocate the bootmap bitmap at | |
94 | * the end, we won't clash. | |
95 | */ | |
96 | static unsigned int __init | |
97 | find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages) | |
98 | { | |
99 | unsigned int start_pfn, bank, bootmap_pfn; | |
100 | ||
90072059 | 101 | start_pfn = PAGE_ALIGN(__pa(&_end)) >> PAGE_SHIFT; |
1da177e4 LT |
102 | bootmap_pfn = 0; |
103 | ||
90072059 | 104 | for_each_nodebank(bank, mi, node) { |
1da177e4 LT |
105 | unsigned int start, end; |
106 | ||
92a8cbed RK |
107 | start = mi->bank[bank].start >> PAGE_SHIFT; |
108 | end = (mi->bank[bank].size + | |
109 | mi->bank[bank].start) >> PAGE_SHIFT; | |
1da177e4 LT |
110 | |
111 | if (end < start_pfn) | |
112 | continue; | |
113 | ||
114 | if (start < start_pfn) | |
115 | start = start_pfn; | |
116 | ||
117 | if (end <= start) | |
118 | continue; | |
119 | ||
120 | if (end - start >= bootmap_pages) { | |
121 | bootmap_pfn = start; | |
122 | break; | |
123 | } | |
124 | } | |
125 | ||
126 | if (bootmap_pfn == 0) | |
127 | BUG(); | |
128 | ||
129 | return bootmap_pfn; | |
130 | } | |
131 | ||
1da177e4 LT |
132 | static int __init check_initrd(struct meminfo *mi) |
133 | { | |
134 | int initrd_node = -2; | |
135 | #ifdef CONFIG_BLK_DEV_INITRD | |
136 | unsigned long end = phys_initrd_start + phys_initrd_size; | |
137 | ||
138 | /* | |
139 | * Make sure that the initrd is within a valid area of | |
140 | * memory. | |
141 | */ | |
142 | if (phys_initrd_size) { | |
143 | unsigned int i; | |
144 | ||
145 | initrd_node = -1; | |
146 | ||
147 | for (i = 0; i < mi->nr_banks; i++) { | |
148 | unsigned long bank_end; | |
149 | ||
150 | bank_end = mi->bank[i].start + mi->bank[i].size; | |
151 | ||
152 | if (mi->bank[i].start <= phys_initrd_start && | |
153 | end <= bank_end) | |
154 | initrd_node = mi->bank[i].node; | |
155 | } | |
156 | } | |
157 | ||
158 | if (initrd_node == -1) { | |
159 | printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond " | |
160 | "physical memory - disabling initrd\n", | |
161 | phys_initrd_start, end); | |
162 | phys_initrd_start = phys_initrd_size = 0; | |
163 | } | |
164 | #endif | |
165 | ||
166 | return initrd_node; | |
167 | } | |
168 | ||
456335e2 RK |
169 | static inline void map_memory_bank(struct membank *bank) |
170 | { | |
d111e8f9 | 171 | #ifdef CONFIG_MMU |
456335e2 RK |
172 | struct map_desc map; |
173 | ||
174 | map.pfn = __phys_to_pfn(bank->start); | |
175 | map.virtual = __phys_to_virt(bank->start); | |
176 | map.length = bank->size; | |
177 | map.type = MT_MEMORY; | |
178 | ||
179 | create_mapping(&map); | |
d111e8f9 | 180 | #endif |
456335e2 RK |
181 | } |
182 | ||
90072059 RK |
183 | static unsigned long __init |
184 | bootmem_init_node(int node, int initrd_node, struct meminfo *mi) | |
1da177e4 | 185 | { |
90072059 RK |
186 | unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES]; |
187 | unsigned long start_pfn, end_pfn, boot_pfn; | |
188 | unsigned int boot_pages; | |
189 | pg_data_t *pgdat; | |
190 | int i; | |
1da177e4 | 191 | |
90072059 RK |
192 | start_pfn = -1UL; |
193 | end_pfn = 0; | |
1da177e4 | 194 | |
90072059 RK |
195 | /* |
196 | * Calculate the pfn range, and map the memory banks for this node. | |
197 | */ | |
198 | for_each_nodebank(i, mi, node) { | |
456335e2 | 199 | struct membank *bank = &mi->bank[i]; |
90072059 | 200 | unsigned long start, end; |
1da177e4 | 201 | |
456335e2 RK |
202 | start = bank->start >> PAGE_SHIFT; |
203 | end = (bank->start + bank->size) >> PAGE_SHIFT; | |
1da177e4 | 204 | |
90072059 RK |
205 | if (start_pfn > start) |
206 | start_pfn = start; | |
207 | if (end_pfn < end) | |
208 | end_pfn = end; | |
209 | ||
456335e2 | 210 | map_memory_bank(bank); |
90072059 | 211 | } |
1da177e4 LT |
212 | |
213 | /* | |
90072059 | 214 | * If there is no memory in this node, ignore it. |
1da177e4 | 215 | */ |
90072059 RK |
216 | if (end_pfn == 0) |
217 | return end_pfn; | |
1da177e4 | 218 | |
90072059 RK |
219 | /* |
220 | * Allocate the bootmem bitmap page. | |
221 | */ | |
222 | boot_pages = bootmem_bootmap_pages(end_pfn - start_pfn); | |
223 | boot_pfn = find_bootmap_pfn(node, mi, boot_pages); | |
1da177e4 | 224 | |
90072059 RK |
225 | /* |
226 | * Initialise the bootmem allocator for this node, handing the | |
227 | * memory banks over to bootmem. | |
228 | */ | |
229 | node_set_online(node); | |
230 | pgdat = NODE_DATA(node); | |
231 | init_bootmem_node(pgdat, boot_pfn, start_pfn, end_pfn); | |
1da177e4 | 232 | |
90072059 RK |
233 | for_each_nodebank(i, mi, node) |
234 | free_bootmem_node(pgdat, mi->bank[i].start, mi->bank[i].size); | |
235 | ||
236 | /* | |
237 | * Reserve the bootmem bitmap for this node. | |
238 | */ | |
239 | reserve_bootmem_node(pgdat, boot_pfn << PAGE_SHIFT, | |
72a7fe39 | 240 | boot_pages << PAGE_SHIFT, BOOTMEM_DEFAULT); |
1da177e4 LT |
241 | |
242 | #ifdef CONFIG_BLK_DEV_INITRD | |
90072059 RK |
243 | /* |
244 | * If the initrd is in this node, reserve its memory. | |
245 | */ | |
246 | if (node == initrd_node) { | |
247 | reserve_bootmem_node(pgdat, phys_initrd_start, | |
72a7fe39 | 248 | phys_initrd_size, BOOTMEM_DEFAULT); |
1da177e4 LT |
249 | initrd_start = __phys_to_virt(phys_initrd_start); |
250 | initrd_end = initrd_start + phys_initrd_size; | |
251 | } | |
252 | #endif | |
253 | ||
90072059 RK |
254 | /* |
255 | * Finally, reserve any node zero regions. | |
256 | */ | |
257 | if (node == 0) | |
258 | reserve_node_zero(pgdat); | |
259 | ||
260 | /* | |
261 | * initialise the zones within this node. | |
262 | */ | |
263 | memset(zone_size, 0, sizeof(zone_size)); | |
264 | memset(zhole_size, 0, sizeof(zhole_size)); | |
265 | ||
266 | /* | |
267 | * The size of this node has already been determined. If we need | |
268 | * to do anything fancy with the allocation of this memory to the | |
269 | * zones, now is the time to do it. | |
270 | */ | |
271 | zone_size[0] = end_pfn - start_pfn; | |
272 | ||
273 | /* | |
274 | * For each bank in this node, calculate the size of the holes. | |
275 | * holes = node_size - sum(bank_sizes_in_node) | |
276 | */ | |
277 | zhole_size[0] = zone_size[0]; | |
278 | for_each_nodebank(i, mi, node) | |
279 | zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT; | |
280 | ||
281 | /* | |
282 | * Adjust the sizes according to any special requirements for | |
283 | * this machine type. | |
284 | */ | |
285 | arch_adjust_zones(node, zone_size, zhole_size); | |
286 | ||
9109fb7b | 287 | free_area_init_node(node, zone_size, start_pfn, zhole_size); |
90072059 RK |
288 | |
289 | return end_pfn; | |
1da177e4 LT |
290 | } |
291 | ||
d111e8f9 | 292 | void __init bootmem_init(struct meminfo *mi) |
1da177e4 | 293 | { |
456335e2 | 294 | unsigned long memend_pfn = 0; |
90072059 | 295 | int node, initrd_node, i; |
1da177e4 | 296 | |
90072059 RK |
297 | /* |
298 | * Invalidate the node number for empty or invalid memory banks | |
299 | */ | |
300 | for (i = 0; i < mi->nr_banks; i++) | |
301 | if (mi->bank[i].size == 0 || mi->bank[i].node >= MAX_NUMNODES) | |
302 | mi->bank[i].node = -1; | |
1da177e4 LT |
303 | |
304 | memcpy(&meminfo, mi, sizeof(meminfo)); | |
305 | ||
1da177e4 | 306 | /* |
90072059 | 307 | * Locate which node contains the ramdisk image, if any. |
1da177e4 | 308 | */ |
90072059 | 309 | initrd_node = check_initrd(mi); |
1da177e4 | 310 | |
90072059 RK |
311 | /* |
312 | * Run through each node initialising the bootmem allocator. | |
313 | */ | |
314 | for_each_node(node) { | |
315 | unsigned long end_pfn; | |
1da177e4 | 316 | |
90072059 | 317 | end_pfn = bootmem_init_node(node, initrd_node, mi); |
1da177e4 LT |
318 | |
319 | /* | |
90072059 | 320 | * Remember the highest memory PFN. |
1da177e4 | 321 | */ |
90072059 RK |
322 | if (end_pfn > memend_pfn) |
323 | memend_pfn = end_pfn; | |
324 | } | |
1da177e4 | 325 | |
90072059 | 326 | high_memory = __va(memend_pfn << PAGE_SHIFT); |
1da177e4 | 327 | |
90072059 RK |
328 | /* |
329 | * This doesn't seem to be used by the Linux memory manager any | |
330 | * more, but is used by ll_rw_block. If we can get rid of it, we | |
331 | * also get rid of some of the stuff above as well. | |
332 | * | |
333 | * Note: max_low_pfn and max_pfn reflect the number of _pages_ in | |
334 | * the system, not the maximum PFN. | |
335 | */ | |
336 | max_pfn = max_low_pfn = memend_pfn - PHYS_PFN_OFFSET; | |
337 | } | |
1da177e4 | 338 | |
1da177e4 LT |
339 | static inline void free_area(unsigned long addr, unsigned long end, char *s) |
340 | { | |
341 | unsigned int size = (end - addr) >> 10; | |
342 | ||
343 | for (; addr < end; addr += PAGE_SIZE) { | |
344 | struct page *page = virt_to_page(addr); | |
345 | ClearPageReserved(page); | |
7835e98b | 346 | init_page_count(page); |
1da177e4 LT |
347 | free_page(addr); |
348 | totalram_pages++; | |
349 | } | |
350 | ||
351 | if (size && s) | |
352 | printk(KERN_INFO "Freeing %s memory: %dK\n", s, size); | |
353 | } | |
354 | ||
a013053d RK |
355 | static inline void |
356 | free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn) | |
357 | { | |
358 | struct page *start_pg, *end_pg; | |
359 | unsigned long pg, pgend; | |
360 | ||
361 | /* | |
362 | * Convert start_pfn/end_pfn to a struct page pointer. | |
363 | */ | |
364 | start_pg = pfn_to_page(start_pfn); | |
365 | end_pg = pfn_to_page(end_pfn); | |
366 | ||
367 | /* | |
368 | * Convert to physical addresses, and | |
369 | * round start upwards and end downwards. | |
370 | */ | |
371 | pg = PAGE_ALIGN(__pa(start_pg)); | |
372 | pgend = __pa(end_pg) & PAGE_MASK; | |
373 | ||
374 | /* | |
375 | * If there are free pages between these, | |
376 | * free the section of the memmap array. | |
377 | */ | |
378 | if (pg < pgend) | |
379 | free_bootmem_node(NODE_DATA(node), pg, pgend - pg); | |
380 | } | |
381 | ||
382 | /* | |
383 | * The mem_map array can get very big. Free the unused area of the memory map. | |
384 | */ | |
385 | static void __init free_unused_memmap_node(int node, struct meminfo *mi) | |
386 | { | |
387 | unsigned long bank_start, prev_bank_end = 0; | |
388 | unsigned int i; | |
389 | ||
390 | /* | |
391 | * [FIXME] This relies on each bank being in address order. This | |
392 | * may not be the case, especially if the user has provided the | |
393 | * information on the command line. | |
394 | */ | |
90072059 | 395 | for_each_nodebank(i, mi, node) { |
a013053d RK |
396 | bank_start = mi->bank[i].start >> PAGE_SHIFT; |
397 | if (bank_start < prev_bank_end) { | |
398 | printk(KERN_ERR "MEM: unordered memory banks. " | |
399 | "Not freeing memmap.\n"); | |
400 | break; | |
401 | } | |
402 | ||
403 | /* | |
404 | * If we had a previous bank, and there is a space | |
405 | * between the current bank and the previous, free it. | |
406 | */ | |
407 | if (prev_bank_end && prev_bank_end != bank_start) | |
408 | free_memmap(node, prev_bank_end, bank_start); | |
409 | ||
410 | prev_bank_end = (mi->bank[i].start + | |
411 | mi->bank[i].size) >> PAGE_SHIFT; | |
412 | } | |
413 | } | |
414 | ||
1da177e4 LT |
415 | /* |
416 | * mem_init() marks the free areas in the mem_map and tells us how much | |
417 | * memory is free. This is done after various parts of the system have | |
418 | * claimed their memory after the kernel image. | |
419 | */ | |
420 | void __init mem_init(void) | |
421 | { | |
422 | unsigned int codepages, datapages, initpages; | |
423 | int i, node; | |
424 | ||
425 | codepages = &_etext - &_text; | |
426 | datapages = &_end - &__data_start; | |
427 | initpages = &__init_end - &__init_begin; | |
428 | ||
429 | #ifndef CONFIG_DISCONTIGMEM | |
430 | max_mapnr = virt_to_page(high_memory) - mem_map; | |
431 | #endif | |
432 | ||
1da177e4 LT |
433 | /* this will put all unused low memory onto the freelists */ |
434 | for_each_online_node(node) { | |
435 | pg_data_t *pgdat = NODE_DATA(node); | |
436 | ||
a013053d RK |
437 | free_unused_memmap_node(node, &meminfo); |
438 | ||
1da177e4 LT |
439 | if (pgdat->node_spanned_pages != 0) |
440 | totalram_pages += free_all_bootmem_node(pgdat); | |
441 | } | |
442 | ||
443 | #ifdef CONFIG_SA1111 | |
444 | /* now that our DMA memory is actually so designated, we can free it */ | |
445 | free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL); | |
446 | #endif | |
447 | ||
448 | /* | |
449 | * Since our memory may not be contiguous, calculate the | |
450 | * real number of pages we have in this system | |
451 | */ | |
452 | printk(KERN_INFO "Memory:"); | |
453 | ||
454 | num_physpages = 0; | |
455 | for (i = 0; i < meminfo.nr_banks; i++) { | |
456 | num_physpages += meminfo.bank[i].size >> PAGE_SHIFT; | |
457 | printk(" %ldMB", meminfo.bank[i].size >> 20); | |
458 | } | |
459 | ||
460 | printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT)); | |
461 | printk(KERN_NOTICE "Memory: %luKB available (%dK code, " | |
462 | "%dK data, %dK init)\n", | |
463 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), | |
464 | codepages >> 10, datapages >> 10, initpages >> 10); | |
465 | ||
466 | if (PAGE_SIZE >= 16384 && num_physpages <= 128) { | |
467 | extern int sysctl_overcommit_memory; | |
468 | /* | |
469 | * On a machine this small we won't get | |
470 | * anywhere without overcommit, so turn | |
471 | * it on by default. | |
472 | */ | |
473 | sysctl_overcommit_memory = OVERCOMMIT_ALWAYS; | |
474 | } | |
475 | } | |
476 | ||
477 | void free_initmem(void) | |
478 | { | |
479 | if (!machine_is_integrator() && !machine_is_cintegrator()) { | |
480 | free_area((unsigned long)(&__init_begin), | |
481 | (unsigned long)(&__init_end), | |
482 | "init"); | |
483 | } | |
484 | } | |
485 | ||
486 | #ifdef CONFIG_BLK_DEV_INITRD | |
487 | ||
488 | static int keep_initrd; | |
489 | ||
490 | void free_initrd_mem(unsigned long start, unsigned long end) | |
491 | { | |
492 | if (!keep_initrd) | |
493 | free_area(start, end, "initrd"); | |
494 | } | |
495 | ||
496 | static int __init keepinitrd_setup(char *__unused) | |
497 | { | |
498 | keep_initrd = 1; | |
499 | return 1; | |
500 | } | |
501 | ||
502 | __setup("keepinitrd", keepinitrd_setup); | |
503 | #endif |