Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/alpha/mm/init.c |
| 3 | * |
| 4 | * Copyright (C) 1995 Linus Torvalds |
| 5 | */ |
| 6 | |
| 7 | /* 2.3.x zone allocator, 1999 Andrea Arcangeli <andrea@suse.de> */ |
| 8 | |
| 9 | #include <linux/config.h> |
akpm@osdl.org | df2e71f | 2006-01-09 20:51:37 -0800 | [diff] [blame^] | 10 | #include <linux/pagemap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | #include <linux/signal.h> |
| 12 | #include <linux/sched.h> |
| 13 | #include <linux/kernel.h> |
| 14 | #include <linux/errno.h> |
| 15 | #include <linux/string.h> |
| 16 | #include <linux/types.h> |
| 17 | #include <linux/ptrace.h> |
| 18 | #include <linux/mman.h> |
| 19 | #include <linux/mm.h> |
| 20 | #include <linux/swap.h> |
| 21 | #include <linux/init.h> |
| 22 | #include <linux/bootmem.h> /* max_low_pfn */ |
| 23 | #include <linux/vmalloc.h> |
| 24 | |
| 25 | #include <asm/system.h> |
| 26 | #include <asm/uaccess.h> |
| 27 | #include <asm/pgtable.h> |
| 28 | #include <asm/pgalloc.h> |
| 29 | #include <asm/hwrpb.h> |
| 30 | #include <asm/dma.h> |
| 31 | #include <asm/mmu_context.h> |
| 32 | #include <asm/console.h> |
| 33 | #include <asm/tlb.h> |
| 34 | |
| 35 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); |
| 36 | |
| 37 | extern void die_if_kernel(char *,struct pt_regs *,long); |
| 38 | |
| 39 | static struct pcb_struct original_pcb; |
| 40 | |
| 41 | pgd_t * |
| 42 | pgd_alloc(struct mm_struct *mm) |
| 43 | { |
| 44 | pgd_t *ret, *init; |
| 45 | |
| 46 | ret = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); |
| 47 | init = pgd_offset(&init_mm, 0UL); |
| 48 | if (ret) { |
| 49 | #ifdef CONFIG_ALPHA_LARGE_VMALLOC |
| 50 | memcpy (ret + USER_PTRS_PER_PGD, init + USER_PTRS_PER_PGD, |
| 51 | (PTRS_PER_PGD - USER_PTRS_PER_PGD - 1)*sizeof(pgd_t)); |
| 52 | #else |
| 53 | pgd_val(ret[PTRS_PER_PGD-2]) = pgd_val(init[PTRS_PER_PGD-2]); |
| 54 | #endif |
| 55 | |
| 56 | /* The last PGD entry is the VPTB self-map. */ |
| 57 | pgd_val(ret[PTRS_PER_PGD-1]) |
| 58 | = pte_val(mk_pte(virt_to_page(ret), PAGE_KERNEL)); |
| 59 | } |
| 60 | return ret; |
| 61 | } |
| 62 | |
| 63 | pte_t * |
| 64 | pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address) |
| 65 | { |
| 66 | pte_t *pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO); |
| 67 | return pte; |
| 68 | } |
| 69 | |
| 70 | |
| 71 | /* |
| 72 | * BAD_PAGE is the page that is used for page faults when linux |
| 73 | * is out-of-memory. Older versions of linux just did a |
| 74 | * do_exit(), but using this instead means there is less risk |
| 75 | * for a process dying in kernel mode, possibly leaving an inode |
| 76 | * unused etc.. |
| 77 | * |
| 78 | * BAD_PAGETABLE is the accompanying page-table: it is initialized |
| 79 | * to point to BAD_PAGE entries. |
| 80 | * |
| 81 | * ZERO_PAGE is a special page that is used for zero-initialized |
| 82 | * data and COW. |
| 83 | */ |
| 84 | pmd_t * |
| 85 | __bad_pagetable(void) |
| 86 | { |
| 87 | memset((void *) EMPTY_PGT, 0, PAGE_SIZE); |
| 88 | return (pmd_t *) EMPTY_PGT; |
| 89 | } |
| 90 | |
| 91 | pte_t |
| 92 | __bad_page(void) |
| 93 | { |
| 94 | memset((void *) EMPTY_PGE, 0, PAGE_SIZE); |
| 95 | return pte_mkdirty(mk_pte(virt_to_page(EMPTY_PGE), PAGE_SHARED)); |
| 96 | } |
| 97 | |
| 98 | #ifndef CONFIG_DISCONTIGMEM |
| 99 | void |
| 100 | show_mem(void) |
| 101 | { |
| 102 | long i,free = 0,total = 0,reserved = 0; |
| 103 | long shared = 0, cached = 0; |
| 104 | |
| 105 | printk("\nMem-info:\n"); |
| 106 | show_free_areas(); |
| 107 | printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); |
| 108 | i = max_mapnr; |
| 109 | while (i-- > 0) { |
| 110 | total++; |
| 111 | if (PageReserved(mem_map+i)) |
| 112 | reserved++; |
| 113 | else if (PageSwapCache(mem_map+i)) |
| 114 | cached++; |
| 115 | else if (!page_count(mem_map+i)) |
| 116 | free++; |
| 117 | else |
| 118 | shared += page_count(mem_map + i) - 1; |
| 119 | } |
| 120 | printk("%ld pages of RAM\n",total); |
| 121 | printk("%ld free pages\n",free); |
| 122 | printk("%ld reserved pages\n",reserved); |
| 123 | printk("%ld pages shared\n",shared); |
| 124 | printk("%ld pages swap cached\n",cached); |
| 125 | } |
| 126 | #endif |
| 127 | |
| 128 | static inline unsigned long |
| 129 | load_PCB(struct pcb_struct *pcb) |
| 130 | { |
| 131 | register unsigned long sp __asm__("$30"); |
| 132 | pcb->ksp = sp; |
| 133 | return __reload_thread(pcb); |
| 134 | } |
| 135 | |
| 136 | /* Set up initial PCB, VPTB, and other such nicities. */ |
| 137 | |
| 138 | static inline void |
| 139 | switch_to_system_map(void) |
| 140 | { |
| 141 | unsigned long newptbr; |
| 142 | unsigned long original_pcb_ptr; |
| 143 | |
| 144 | /* Initialize the kernel's page tables. Linux puts the vptb in |
| 145 | the last slot of the L1 page table. */ |
| 146 | memset(swapper_pg_dir, 0, PAGE_SIZE); |
| 147 | newptbr = ((unsigned long) swapper_pg_dir - PAGE_OFFSET) >> PAGE_SHIFT; |
| 148 | pgd_val(swapper_pg_dir[1023]) = |
| 149 | (newptbr << 32) | pgprot_val(PAGE_KERNEL); |
| 150 | |
| 151 | /* Set the vptb. This is often done by the bootloader, but |
| 152 | shouldn't be required. */ |
| 153 | if (hwrpb->vptb != 0xfffffffe00000000UL) { |
| 154 | wrvptptr(0xfffffffe00000000UL); |
| 155 | hwrpb->vptb = 0xfffffffe00000000UL; |
| 156 | hwrpb_update_checksum(hwrpb); |
| 157 | } |
| 158 | |
| 159 | /* Also set up the real kernel PCB while we're at it. */ |
| 160 | init_thread_info.pcb.ptbr = newptbr; |
| 161 | init_thread_info.pcb.flags = 1; /* set FEN, clear everything else */ |
| 162 | original_pcb_ptr = load_PCB(&init_thread_info.pcb); |
| 163 | tbia(); |
| 164 | |
| 165 | /* Save off the contents of the original PCB so that we can |
| 166 | restore the original console's page tables for a clean reboot. |
| 167 | |
| 168 | Note that the PCB is supposed to be a physical address, but |
| 169 | since KSEG values also happen to work, folks get confused. |
| 170 | Check this here. */ |
| 171 | |
| 172 | if (original_pcb_ptr < PAGE_OFFSET) { |
| 173 | original_pcb_ptr = (unsigned long) |
| 174 | phys_to_virt(original_pcb_ptr); |
| 175 | } |
| 176 | original_pcb = *(struct pcb_struct *) original_pcb_ptr; |
| 177 | } |
| 178 | |
| 179 | int callback_init_done; |
| 180 | |
| 181 | void * __init |
| 182 | callback_init(void * kernel_end) |
| 183 | { |
| 184 | struct crb_struct * crb; |
| 185 | pgd_t *pgd; |
| 186 | pmd_t *pmd; |
| 187 | void *two_pages; |
| 188 | |
| 189 | /* Starting at the HWRPB, locate the CRB. */ |
| 190 | crb = (struct crb_struct *)((char *)hwrpb + hwrpb->crb_offset); |
| 191 | |
| 192 | if (alpha_using_srm) { |
| 193 | /* Tell the console whither it is to be remapped. */ |
| 194 | if (srm_fixup(VMALLOC_START, (unsigned long)hwrpb)) |
| 195 | __halt(); /* "We're boned." --Bender */ |
| 196 | |
| 197 | /* Edit the procedure descriptors for DISPATCH and FIXUP. */ |
| 198 | crb->dispatch_va = (struct procdesc_struct *) |
| 199 | (VMALLOC_START + (unsigned long)crb->dispatch_va |
| 200 | - crb->map[0].va); |
| 201 | crb->fixup_va = (struct procdesc_struct *) |
| 202 | (VMALLOC_START + (unsigned long)crb->fixup_va |
| 203 | - crb->map[0].va); |
| 204 | } |
| 205 | |
| 206 | switch_to_system_map(); |
| 207 | |
| 208 | /* Allocate one PGD and one PMD. In the case of SRM, we'll need |
| 209 | these to actually remap the console. There is an assumption |
| 210 | here that only one of each is needed, and this allows for 8MB. |
| 211 | On systems with larger consoles, additional pages will be |
| 212 | allocated as needed during the mapping process. |
| 213 | |
| 214 | In the case of not SRM, but not CONFIG_ALPHA_LARGE_VMALLOC, |
| 215 | we need to allocate the PGD we use for vmalloc before we start |
| 216 | forking other tasks. */ |
| 217 | |
| 218 | two_pages = (void *) |
| 219 | (((unsigned long)kernel_end + ~PAGE_MASK) & PAGE_MASK); |
| 220 | kernel_end = two_pages + 2*PAGE_SIZE; |
| 221 | memset(two_pages, 0, 2*PAGE_SIZE); |
| 222 | |
| 223 | pgd = pgd_offset_k(VMALLOC_START); |
| 224 | pgd_set(pgd, (pmd_t *)two_pages); |
| 225 | pmd = pmd_offset(pgd, VMALLOC_START); |
| 226 | pmd_set(pmd, (pte_t *)(two_pages + PAGE_SIZE)); |
| 227 | |
| 228 | if (alpha_using_srm) { |
| 229 | static struct vm_struct console_remap_vm; |
| 230 | unsigned long vaddr = VMALLOC_START; |
| 231 | unsigned long i, j; |
| 232 | |
| 233 | /* Set up the third level PTEs and update the virtual |
| 234 | addresses of the CRB entries. */ |
| 235 | for (i = 0; i < crb->map_entries; ++i) { |
| 236 | unsigned long pfn = crb->map[i].pa >> PAGE_SHIFT; |
| 237 | crb->map[i].va = vaddr; |
| 238 | for (j = 0; j < crb->map[i].count; ++j) { |
| 239 | /* Newer console's (especially on larger |
| 240 | systems) may require more pages of |
| 241 | PTEs. Grab additional pages as needed. */ |
| 242 | if (pmd != pmd_offset(pgd, vaddr)) { |
| 243 | memset(kernel_end, 0, PAGE_SIZE); |
| 244 | pmd = pmd_offset(pgd, vaddr); |
| 245 | pmd_set(pmd, (pte_t *)kernel_end); |
| 246 | kernel_end += PAGE_SIZE; |
| 247 | } |
| 248 | set_pte(pte_offset_kernel(pmd, vaddr), |
| 249 | pfn_pte(pfn, PAGE_KERNEL)); |
| 250 | pfn++; |
| 251 | vaddr += PAGE_SIZE; |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | /* Let vmalloc know that we've allocated some space. */ |
| 256 | console_remap_vm.flags = VM_ALLOC; |
| 257 | console_remap_vm.addr = (void *) VMALLOC_START; |
| 258 | console_remap_vm.size = vaddr - VMALLOC_START; |
| 259 | vmlist = &console_remap_vm; |
| 260 | } |
| 261 | |
| 262 | callback_init_done = 1; |
| 263 | return kernel_end; |
| 264 | } |
| 265 | |
| 266 | |
| 267 | #ifndef CONFIG_DISCONTIGMEM |
| 268 | /* |
| 269 | * paging_init() sets up the memory map. |
| 270 | */ |
| 271 | void |
| 272 | paging_init(void) |
| 273 | { |
| 274 | unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0}; |
| 275 | unsigned long dma_pfn, high_pfn; |
| 276 | |
| 277 | dma_pfn = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; |
| 278 | high_pfn = max_pfn = max_low_pfn; |
| 279 | |
| 280 | if (dma_pfn >= high_pfn) |
| 281 | zones_size[ZONE_DMA] = high_pfn; |
| 282 | else { |
| 283 | zones_size[ZONE_DMA] = dma_pfn; |
| 284 | zones_size[ZONE_NORMAL] = high_pfn - dma_pfn; |
| 285 | } |
| 286 | |
| 287 | /* Initialize mem_map[]. */ |
| 288 | free_area_init(zones_size); |
| 289 | |
| 290 | /* Initialize the kernel's ZERO_PGE. */ |
| 291 | memset((void *)ZERO_PGE, 0, PAGE_SIZE); |
| 292 | } |
| 293 | #endif /* CONFIG_DISCONTIGMEM */ |
| 294 | |
| 295 | #if defined(CONFIG_ALPHA_GENERIC) || defined(CONFIG_ALPHA_SRM) |
| 296 | void |
| 297 | srm_paging_stop (void) |
| 298 | { |
| 299 | /* Move the vptb back to where the SRM console expects it. */ |
| 300 | swapper_pg_dir[1] = swapper_pg_dir[1023]; |
| 301 | tbia(); |
| 302 | wrvptptr(0x200000000UL); |
| 303 | hwrpb->vptb = 0x200000000UL; |
| 304 | hwrpb_update_checksum(hwrpb); |
| 305 | |
| 306 | /* Reload the page tables that the console had in use. */ |
| 307 | load_PCB(&original_pcb); |
| 308 | tbia(); |
| 309 | } |
| 310 | #endif |
| 311 | |
| 312 | #ifndef CONFIG_DISCONTIGMEM |
| 313 | static void __init |
| 314 | printk_memory_info(void) |
| 315 | { |
| 316 | unsigned long codesize, reservedpages, datasize, initsize, tmp; |
| 317 | extern int page_is_ram(unsigned long) __init; |
| 318 | extern char _text, _etext, _data, _edata; |
| 319 | extern char __init_begin, __init_end; |
| 320 | |
| 321 | /* printk all informations */ |
| 322 | reservedpages = 0; |
| 323 | for (tmp = 0; tmp < max_low_pfn; tmp++) |
| 324 | /* |
| 325 | * Only count reserved RAM pages |
| 326 | */ |
| 327 | if (page_is_ram(tmp) && PageReserved(mem_map+tmp)) |
| 328 | reservedpages++; |
| 329 | |
| 330 | codesize = (unsigned long) &_etext - (unsigned long) &_text; |
| 331 | datasize = (unsigned long) &_edata - (unsigned long) &_data; |
| 332 | initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; |
| 333 | |
| 334 | printk("Memory: %luk/%luk available (%luk kernel code, %luk reserved, %luk data, %luk init)\n", |
| 335 | (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), |
| 336 | max_mapnr << (PAGE_SHIFT-10), |
| 337 | codesize >> 10, |
| 338 | reservedpages << (PAGE_SHIFT-10), |
| 339 | datasize >> 10, |
| 340 | initsize >> 10); |
| 341 | } |
| 342 | |
| 343 | void __init |
| 344 | mem_init(void) |
| 345 | { |
| 346 | max_mapnr = num_physpages = max_low_pfn; |
| 347 | totalram_pages += free_all_bootmem(); |
| 348 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE); |
| 349 | |
| 350 | printk_memory_info(); |
| 351 | } |
| 352 | #endif /* CONFIG_DISCONTIGMEM */ |
| 353 | |
| 354 | void |
| 355 | free_reserved_mem(void *start, void *end) |
| 356 | { |
| 357 | void *__start = start; |
| 358 | for (; __start < end; __start += PAGE_SIZE) { |
| 359 | ClearPageReserved(virt_to_page(__start)); |
| 360 | set_page_count(virt_to_page(__start), 1); |
| 361 | free_page((long)__start); |
| 362 | totalram_pages++; |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | void |
| 367 | free_initmem(void) |
| 368 | { |
| 369 | extern char __init_begin, __init_end; |
| 370 | |
| 371 | free_reserved_mem(&__init_begin, &__init_end); |
| 372 | printk ("Freeing unused kernel memory: %ldk freed\n", |
| 373 | (&__init_end - &__init_begin) >> 10); |
| 374 | } |
| 375 | |
| 376 | #ifdef CONFIG_BLK_DEV_INITRD |
| 377 | void |
| 378 | free_initrd_mem(unsigned long start, unsigned long end) |
| 379 | { |
| 380 | free_reserved_mem((void *)start, (void *)end); |
| 381 | printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10); |
| 382 | } |
| 383 | #endif |