1 /* 2 * linux/arch/arm/mm/init.c 3 * 4 * Copyright (C) 1995-2002 Russell King 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 */ 10 #include <linux/config.h> 11 #include <linux/kernel.h> 12 #include <linux/errno.h> 13 #include <linux/ptrace.h> 14 #include <linux/swap.h> 15 #include <linux/init.h> 16 #include <linux/bootmem.h> 17 #include <linux/mman.h> 18 #include <linux/nodemask.h> 19 #include <linux/initrd.h> 20 21 #include <asm/mach-types.h> 22 #include <asm/hardware.h> 23 #include <asm/setup.h> 24 #include <asm/tlb.h> 25 26 #include <asm/mach/arch.h> 27 #include <asm/mach/map.h> 28 29 #define TABLE_SIZE (2 * PTRS_PER_PTE * sizeof(pte_t)) 30 31 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); 32 33 extern pgd_t swapper_pg_dir[PTRS_PER_PGD]; 34 extern void _stext, _text, _etext, __data_start, _end, __init_begin, __init_end; 35 extern unsigned long phys_initrd_start; 36 extern unsigned long phys_initrd_size; 37 38 /* 39 * The sole use of this is to pass memory configuration 40 * data from paging_init to mem_init. 41 */ 42 static struct meminfo meminfo __initdata = { 0, }; 43 44 /* 45 * empty_zero_page is a special page that is used for 46 * zero-initialized data and COW. 47 */ 48 struct page *empty_zero_page; 49 50 void show_mem(void) 51 { 52 int free = 0, total = 0, reserved = 0; 53 int shared = 0, cached = 0, slab = 0, node; 54 55 printk("Mem-info:\n"); 56 show_free_areas(); 57 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); 58 59 for_each_online_node(node) { 60 struct page *page, *end; 61 62 page = NODE_MEM_MAP(node); 63 end = page + NODE_DATA(node)->node_spanned_pages; 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 } 80 81 printk("%d pages of RAM\n", total); 82 printk("%d free pages\n", free); 83 printk("%d reserved pages\n", reserved); 84 printk("%d slab pages\n", slab); 85 printk("%d pages shared\n", shared); 86 printk("%d pages swap cached\n", cached); 87 } 88 89 struct node_info { 90 unsigned int start; 91 unsigned int end; 92 int bootmap_pages; 93 }; 94 95 #define O_PFN_DOWN(x) ((x) >> PAGE_SHIFT) 96 #define V_PFN_DOWN(x) O_PFN_DOWN(__pa(x)) 97 98 #define O_PFN_UP(x) (PAGE_ALIGN(x) >> PAGE_SHIFT) 99 #define V_PFN_UP(x) O_PFN_UP(__pa(x)) 100 101 #define PFN_SIZE(x) ((x) >> PAGE_SHIFT) 102 #define PFN_RANGE(s,e) PFN_SIZE(PAGE_ALIGN((unsigned long)(e)) - \ 103 (((unsigned long)(s)) & PAGE_MASK)) 104 105 /* 106 * FIXME: We really want to avoid allocating the bootmap bitmap 107 * over the top of the initrd. Hopefully, this is located towards 108 * the start of a bank, so if we allocate the bootmap bitmap at 109 * the end, we won't clash. 110 */ 111 static unsigned int __init 112 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages) 113 { 114 unsigned int start_pfn, bank, bootmap_pfn; 115 116 start_pfn = V_PFN_UP(&_end); 117 bootmap_pfn = 0; 118 119 for (bank = 0; bank < mi->nr_banks; bank ++) { 120 unsigned int start, end; 121 122 if (mi->bank[bank].node != node) 123 continue; 124 125 start = O_PFN_UP(mi->bank[bank].start); 126 end = O_PFN_DOWN(mi->bank[bank].size + 127 mi->bank[bank].start); 128 129 if (end < start_pfn) 130 continue; 131 132 if (start < start_pfn) 133 start = start_pfn; 134 135 if (end <= start) 136 continue; 137 138 if (end - start >= bootmap_pages) { 139 bootmap_pfn = start; 140 break; 141 } 142 } 143 144 if (bootmap_pfn == 0) 145 BUG(); 146 147 return bootmap_pfn; 148 } 149 150 /* 151 * Scan the memory info structure and pull out: 152 * - the end of memory 153 * - the number of nodes 154 * - the pfn range of each node 155 * - the number of bootmem bitmap pages 156 */ 157 static unsigned int __init 158 find_memend_and_nodes(struct meminfo *mi, struct node_info *np) 159 { 160 unsigned int i, bootmem_pages = 0, memend_pfn = 0; 161 162 for (i = 0; i < MAX_NUMNODES; i++) { 163 np[i].start = -1U; 164 np[i].end = 0; 165 np[i].bootmap_pages = 0; 166 } 167 168 for (i = 0; i < mi->nr_banks; i++) { 169 unsigned long start, end; 170 int node; 171 172 if (mi->bank[i].size == 0) { 173 /* 174 * Mark this bank with an invalid node number 175 */ 176 mi->bank[i].node = -1; 177 continue; 178 } 179 180 node = mi->bank[i].node; 181 182 /* 183 * Make sure we haven't exceeded the maximum number of nodes 184 * that we have in this configuration. If we have, we're in 185 * trouble. (maybe we ought to limit, instead of bugging?) 186 */ 187 if (node >= MAX_NUMNODES) 188 BUG(); 189 node_set_online(node); 190 191 /* 192 * Get the start and end pfns for this bank 193 */ 194 start = O_PFN_UP(mi->bank[i].start); 195 end = O_PFN_DOWN(mi->bank[i].start + mi->bank[i].size); 196 197 if (np[node].start > start) 198 np[node].start = start; 199 200 if (np[node].end < end) 201 np[node].end = end; 202 203 if (memend_pfn < end) 204 memend_pfn = end; 205 } 206 207 /* 208 * Calculate the number of pages we require to 209 * store the bootmem bitmaps. 210 */ 211 for_each_online_node(i) { 212 if (np[i].end == 0) 213 continue; 214 215 np[i].bootmap_pages = bootmem_bootmap_pages(np[i].end - 216 np[i].start); 217 bootmem_pages += np[i].bootmap_pages; 218 } 219 220 high_memory = __va(memend_pfn << PAGE_SHIFT); 221 222 /* 223 * This doesn't seem to be used by the Linux memory 224 * manager any more. If we can get rid of it, we 225 * also get rid of some of the stuff above as well. 226 * 227 * Note: max_low_pfn and max_pfn reflect the number 228 * of _pages_ in the system, not the maximum PFN. 229 */ 230 max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET); 231 max_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET); 232 233 return bootmem_pages; 234 } 235 236 static int __init check_initrd(struct meminfo *mi) 237 { 238 int initrd_node = -2; 239 #ifdef CONFIG_BLK_DEV_INITRD 240 unsigned long end = phys_initrd_start + phys_initrd_size; 241 242 /* 243 * Make sure that the initrd is within a valid area of 244 * memory. 245 */ 246 if (phys_initrd_size) { 247 unsigned int i; 248 249 initrd_node = -1; 250 251 for (i = 0; i < mi->nr_banks; i++) { 252 unsigned long bank_end; 253 254 bank_end = mi->bank[i].start + mi->bank[i].size; 255 256 if (mi->bank[i].start <= phys_initrd_start && 257 end <= bank_end) 258 initrd_node = mi->bank[i].node; 259 } 260 } 261 262 if (initrd_node == -1) { 263 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond " 264 "physical memory - disabling initrd\n", 265 phys_initrd_start, end); 266 phys_initrd_start = phys_initrd_size = 0; 267 } 268 #endif 269 270 return initrd_node; 271 } 272 273 /* 274 * Reserve the various regions of node 0 275 */ 276 static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int bootmap_pages) 277 { 278 pg_data_t *pgdat = NODE_DATA(0); 279 unsigned long res_size = 0; 280 281 /* 282 * Register the kernel text and data with bootmem. 283 * Note that this can only be in node 0. 284 */ 285 #ifdef CONFIG_XIP_KERNEL 286 reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start); 287 #else 288 reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext); 289 #endif 290 291 /* 292 * Reserve the page tables. These are already in use, 293 * and can only be in node 0. 294 */ 295 reserve_bootmem_node(pgdat, __pa(swapper_pg_dir), 296 PTRS_PER_PGD * sizeof(pgd_t)); 297 298 /* 299 * And don't forget to reserve the allocator bitmap, 300 * which will be freed later. 301 */ 302 reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT, 303 bootmap_pages << PAGE_SHIFT); 304 305 /* 306 * Hmm... This should go elsewhere, but we really really need to 307 * stop things allocating the low memory; ideally we need a better 308 * implementation of GFP_DMA which does not assume that DMA-able 309 * memory starts at zero. 310 */ 311 if (machine_is_integrator() || machine_is_cintegrator()) 312 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET; 313 314 /* 315 * These should likewise go elsewhere. They pre-reserve the 316 * screen memory region at the start of main system memory. 317 */ 318 if (machine_is_edb7211()) 319 res_size = 0x00020000; 320 if (machine_is_p720t()) 321 res_size = 0x00014000; 322 323 #ifdef CONFIG_SA1111 324 /* 325 * Because of the SA1111 DMA bug, we want to preserve our 326 * precious DMA-able memory... 327 */ 328 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET; 329 #endif 330 if (res_size) 331 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size); 332 } 333 334 /* 335 * Register all available RAM in this node with the bootmem allocator. 336 */ 337 static inline void free_bootmem_node_bank(int node, struct meminfo *mi) 338 { 339 pg_data_t *pgdat = NODE_DATA(node); 340 int bank; 341 342 for (bank = 0; bank < mi->nr_banks; bank++) 343 if (mi->bank[bank].node == node) 344 free_bootmem_node(pgdat, mi->bank[bank].start, 345 mi->bank[bank].size); 346 } 347 348 /* 349 * Initialise the bootmem allocator for all nodes. This is called 350 * early during the architecture specific initialisation. 351 */ 352 static void __init bootmem_init(struct meminfo *mi) 353 { 354 struct node_info node_info[MAX_NUMNODES], *np = node_info; 355 unsigned int bootmap_pages, bootmap_pfn, map_pg; 356 int node, initrd_node; 357 358 bootmap_pages = find_memend_and_nodes(mi, np); 359 bootmap_pfn = find_bootmap_pfn(0, mi, bootmap_pages); 360 initrd_node = check_initrd(mi); 361 362 map_pg = bootmap_pfn; 363 364 /* 365 * Initialise the bootmem nodes. 366 * 367 * What we really want to do is: 368 * 369 * unmap_all_regions_except_kernel(); 370 * for_each_node_in_reverse_order(node) { 371 * map_node(node); 372 * allocate_bootmem_map(node); 373 * init_bootmem_node(node); 374 * free_bootmem_node(node); 375 * } 376 * 377 * but this is a 2.5-type change. For now, we just set 378 * the nodes up in reverse order. 379 * 380 * (we could also do with rolling bootmem_init and paging_init 381 * into one generic "memory_init" type function). 382 */ 383 np += num_online_nodes() - 1; 384 for (node = num_online_nodes() - 1; node >= 0; node--, np--) { 385 /* 386 * If there are no pages in this node, ignore it. 387 * Note that node 0 must always have some pages. 388 */ 389 if (np->end == 0 || !node_online(node)) { 390 if (node == 0) 391 BUG(); 392 continue; 393 } 394 395 /* 396 * Initialise the bootmem allocator. 397 */ 398 init_bootmem_node(NODE_DATA(node), map_pg, np->start, np->end); 399 free_bootmem_node_bank(node, mi); 400 map_pg += np->bootmap_pages; 401 402 /* 403 * If this is node 0, we need to reserve some areas ASAP - 404 * we may use bootmem on node 0 to setup the other nodes. 405 */ 406 if (node == 0) 407 reserve_node_zero(bootmap_pfn, bootmap_pages); 408 } 409 410 411 #ifdef CONFIG_BLK_DEV_INITRD 412 if (phys_initrd_size && initrd_node >= 0) { 413 reserve_bootmem_node(NODE_DATA(initrd_node), phys_initrd_start, 414 phys_initrd_size); 415 initrd_start = __phys_to_virt(phys_initrd_start); 416 initrd_end = initrd_start + phys_initrd_size; 417 } 418 #endif 419 420 BUG_ON(map_pg != bootmap_pfn + bootmap_pages); 421 } 422 423 /* 424 * paging_init() sets up the page tables, initialises the zone memory 425 * maps, and sets up the zero page, bad page and bad page tables. 426 */ 427 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc) 428 { 429 void *zero_page; 430 int node; 431 432 bootmem_init(mi); 433 434 memcpy(&meminfo, mi, sizeof(meminfo)); 435 436 /* 437 * allocate the zero page. Note that we count on this going ok. 438 */ 439 zero_page = alloc_bootmem_low_pages(PAGE_SIZE); 440 441 /* 442 * initialise the page tables. 443 */ 444 memtable_init(mi); 445 if (mdesc->map_io) 446 mdesc->map_io(); 447 flush_tlb_all(); 448 449 /* 450 * initialise the zones within each node 451 */ 452 for_each_online_node(node) { 453 unsigned long zone_size[MAX_NR_ZONES]; 454 unsigned long zhole_size[MAX_NR_ZONES]; 455 struct bootmem_data *bdata; 456 pg_data_t *pgdat; 457 int i; 458 459 /* 460 * Initialise the zone size information. 461 */ 462 for (i = 0; i < MAX_NR_ZONES; i++) { 463 zone_size[i] = 0; 464 zhole_size[i] = 0; 465 } 466 467 pgdat = NODE_DATA(node); 468 bdata = pgdat->bdata; 469 470 /* 471 * The size of this node has already been determined. 472 * If we need to do anything fancy with the allocation 473 * of this memory to the zones, now is the time to do 474 * it. 475 */ 476 zone_size[0] = bdata->node_low_pfn - 477 (bdata->node_boot_start >> PAGE_SHIFT); 478 479 /* 480 * If this zone has zero size, skip it. 481 */ 482 if (!zone_size[0]) 483 continue; 484 485 /* 486 * For each bank in this node, calculate the size of the 487 * holes. holes = node_size - sum(bank_sizes_in_node) 488 */ 489 zhole_size[0] = zone_size[0]; 490 for (i = 0; i < mi->nr_banks; i++) { 491 if (mi->bank[i].node != node) 492 continue; 493 494 zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT; 495 } 496 497 /* 498 * Adjust the sizes according to any special 499 * requirements for this machine type. 500 */ 501 arch_adjust_zones(node, zone_size, zhole_size); 502 503 free_area_init_node(node, pgdat, zone_size, 504 bdata->node_boot_start >> PAGE_SHIFT, zhole_size); 505 } 506 507 /* 508 * finish off the bad pages once 509 * the mem_map is initialised 510 */ 511 memzero(zero_page, PAGE_SIZE); 512 empty_zero_page = virt_to_page(zero_page); 513 flush_dcache_page(empty_zero_page); 514 } 515 516 static inline void free_area(unsigned long addr, unsigned long end, char *s) 517 { 518 unsigned int size = (end - addr) >> 10; 519 520 for (; addr < end; addr += PAGE_SIZE) { 521 struct page *page = virt_to_page(addr); 522 ClearPageReserved(page); 523 set_page_count(page, 1); 524 free_page(addr); 525 totalram_pages++; 526 } 527 528 if (size && s) 529 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size); 530 } 531 532 /* 533 * mem_init() marks the free areas in the mem_map and tells us how much 534 * memory is free. This is done after various parts of the system have 535 * claimed their memory after the kernel image. 536 */ 537 void __init mem_init(void) 538 { 539 unsigned int codepages, datapages, initpages; 540 int i, node; 541 542 codepages = &_etext - &_text; 543 datapages = &_end - &__data_start; 544 initpages = &__init_end - &__init_begin; 545 546 #ifndef CONFIG_DISCONTIGMEM 547 max_mapnr = virt_to_page(high_memory) - mem_map; 548 #endif 549 550 /* 551 * We may have non-contiguous memory. 552 */ 553 if (meminfo.nr_banks != 1) 554 create_memmap_holes(&meminfo); 555 556 /* this will put all unused low memory onto the freelists */ 557 for_each_online_node(node) { 558 pg_data_t *pgdat = NODE_DATA(node); 559 560 if (pgdat->node_spanned_pages != 0) 561 totalram_pages += free_all_bootmem_node(pgdat); 562 } 563 564 #ifdef CONFIG_SA1111 565 /* now that our DMA memory is actually so designated, we can free it */ 566 free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL); 567 #endif 568 569 /* 570 * Since our memory may not be contiguous, calculate the 571 * real number of pages we have in this system 572 */ 573 printk(KERN_INFO "Memory:"); 574 575 num_physpages = 0; 576 for (i = 0; i < meminfo.nr_banks; i++) { 577 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT; 578 printk(" %ldMB", meminfo.bank[i].size >> 20); 579 } 580 581 printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT)); 582 printk(KERN_NOTICE "Memory: %luKB available (%dK code, " 583 "%dK data, %dK init)\n", 584 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), 585 codepages >> 10, datapages >> 10, initpages >> 10); 586 587 if (PAGE_SIZE >= 16384 && num_physpages <= 128) { 588 extern int sysctl_overcommit_memory; 589 /* 590 * On a machine this small we won't get 591 * anywhere without overcommit, so turn 592 * it on by default. 593 */ 594 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS; 595 } 596 } 597 598 void free_initmem(void) 599 { 600 if (!machine_is_integrator() && !machine_is_cintegrator()) { 601 free_area((unsigned long)(&__init_begin), 602 (unsigned long)(&__init_end), 603 "init"); 604 } 605 } 606 607 #ifdef CONFIG_BLK_DEV_INITRD 608 609 static int keep_initrd; 610 611 void free_initrd_mem(unsigned long start, unsigned long end) 612 { 613 if (!keep_initrd) 614 free_area(start, end, "initrd"); 615 } 616 617 static int __init keepinitrd_setup(char *__unused) 618 { 619 keep_initrd = 1; 620 return 1; 621 } 622 623 __setup("keepinitrd", keepinitrd_setup); 624 #endif 625