arch/powerpc/mm/mem.c

   1 /*
   2  *  PowerPC version
   3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   4  *
   5  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
   6  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
   7  *    Copyright (C) 1996 Paul Mackerras
   8  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
   9  *
  10  *  Derived from "arch/i386/mm/init.c"
  11  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  12  *
  13  *  This program is free software; you can redistribute it and/or
  14  *  modify it under the terms of the GNU General Public License
  15  *  as published by the Free Software Foundation; either version
  16  *  2 of the License, or (at your option) any later version.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/sched.h>
  22 #include <linux/kernel.h>
  23 #include <linux/errno.h>
  24 #include <linux/string.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/stddef.h>
  28 #include <linux/init.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/highmem.h>
  31 #include <linux/initrd.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/suspend.h>
  34
  35 #include <asm/pgalloc.h>
  36 #include <asm/prom.h>
  37 #include <asm/io.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/mmu.h>
  41 #include <asm/smp.h>
  42 #include <asm/machdep.h>
  43 #include <asm/btext.h>
  44 #include <asm/tlb.h>
  45 #include <asm/lmb.h>
  46 #include <asm/sections.h>
  47 #include <asm/vdso.h>
  48
  49 #include "mmu_decl.h"
  50
  51 #ifndef CPU_FTR_COHERENT_ICACHE
  52 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  53 #define CPU_FTR_NOEXECUTE       0
  54 #endif
  55
  56 int init_bootmem_done;
  57 int mem_init_done;
  58 unsigned long memory_limit;
  59
  60 int page_is_ram(unsigned long pfn)
  61 {
  62         unsigned long paddr = (pfn << PAGE_SHIFT);
  63
  64 #ifndef CONFIG_PPC64    /* XXX for now */
  65         return paddr < __pa(high_memory);
  66 #else
  67         int i;
  68         for (i=0; i < lmb.memory.cnt; i++) {
  69                 unsigned long base;
  70
  71                 base = lmb.memory.region[i].base;
  72
  73                 if ((paddr >= base) &&
  74                         (paddr < (base + lmb.memory.region[i].size))) {
  75                         return 1;
  76                 }
  77         }
  78
  79         return 0;
  80 #endif
  81 }
  82
  83 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  84                               unsigned long size, pgprot_t vma_prot)
  85 {
  86         if (ppc_md.phys_mem_access_prot)
  87                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  88
  89         if (!page_is_ram(pfn))
  90                 vma_prot = __pgprot(pgprot_val(vma_prot)
  91                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
  92         return vma_prot;
  93 }
  94 EXPORT_SYMBOL(phys_mem_access_prot);
  95
  96 #ifdef CONFIG_MEMORY_HOTPLUG
  97
  98 void online_page(struct page *page)
  99 {
 100         ClearPageReserved(page);
 101         init_page_count(page);
 102         __free_page(page);
 103         totalram_pages++;
 104         num_physpages++;
 105 }
 106
 107 #ifdef CONFIG_NUMA
 108 int memory_add_physaddr_to_nid(u64 start)
 109 {
 110         return hot_add_scn_to_nid(start);
 111 }
 112 #endif
 113
 114 int __devinit arch_add_memory(int nid, u64 start, u64 size)
 115 {
 116         struct pglist_data *pgdata;
 117         struct zone *zone;
 118         unsigned long start_pfn = start >> PAGE_SHIFT;
 119         unsigned long nr_pages = size >> PAGE_SHIFT;
 120
 121         pgdata = NODE_DATA(nid);
 122
 123         start = (unsigned long)__va(start);
 124         create_section_mapping(start, start + size);
 125
 126         /* this should work for most non-highmem platforms */
 127         zone = pgdata->node_zones;
 128
 129         return __add_pages(zone, start_pfn, nr_pages);
 130 }
 131
 132 #endif /* CONFIG_MEMORY_HOTPLUG */
 133
 134 void show_mem(void)
 135 {
 136         unsigned long total = 0, reserved = 0;
 137         unsigned long shared = 0, cached = 0;
 138         unsigned long highmem = 0;
 139         struct page *page;
 140         pg_data_t *pgdat;
 141         unsigned long i;
 142
 143         printk("Mem-info:\n");
 144         show_free_areas();
 145         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 146         for_each_online_pgdat(pgdat) {
 147                 unsigned long flags;
 148                 pgdat_resize_lock(pgdat, &flags);
 149                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 150                         if (!pfn_valid(pgdat->node_start_pfn + i))
 151                                 continue;
 152                         page = pgdat_page_nr(pgdat, i);
 153                         total++;
 154                         if (PageHighMem(page))
 155                                 highmem++;
 156                         if (PageReserved(page))
 157                                 reserved++;
 158                         else if (PageSwapCache(page))
 159                                 cached++;
 160                         else if (page_count(page))
 161                                 shared += page_count(page) - 1;
 162                 }
 163                 pgdat_resize_unlock(pgdat, &flags);
 164         }
 165         printk("%ld pages of RAM\n", total);
 166 #ifdef CONFIG_HIGHMEM
 167         printk("%ld pages of HIGHMEM\n", highmem);
 168 #endif
 169         printk("%ld reserved pages\n", reserved);
 170         printk("%ld pages shared\n", shared);
 171         printk("%ld pages swap cached\n", cached);
 172 }
 173
 174 /*
 175  * Initialize the bootmem system and give it all the memory we
 176  * have available.  If we are using highmem, we only put the
 177  * lowmem into the bootmem system.
 178  */
 179 #ifndef CONFIG_NEED_MULTIPLE_NODES
 180 void __init do_init_bootmem(void)
 181 {
 182         unsigned long i;
 183         unsigned long start, bootmap_pages;
 184         unsigned long total_pages;
 185         int boot_mapsize;
 186
 187         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 188 #ifdef CONFIG_HIGHMEM
 189         total_pages = total_lowmem >> PAGE_SHIFT;
 190 #endif
 191
 192         /*
 193          * Find an area to use for the bootmem bitmap.  Calculate the size of
 194          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 195          * Add 1 additional page in case the address isn't page-aligned.
 196          */
 197         bootmap_pages = bootmem_bootmap_pages(total_pages);
 198
 199         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 200
 201         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 202
 203         /* Add active regions with valid PFNs */
 204         for (i = 0; i < lmb.memory.cnt; i++) {
 205                 unsigned long start_pfn, end_pfn;
 206                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
 207                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
 208                 add_active_range(0, start_pfn, end_pfn);
 209         }
 210
 211         /* Add all physical memory to the bootmem map, mark each area
 212          * present.
 213          */
 214 #ifdef CONFIG_HIGHMEM
 215         free_bootmem_with_active_regions(0, total_lowmem >> PAGE_SHIFT);
 216
 217         /* reserve the sections we're already using */
 218         for (i = 0; i < lmb.reserved.cnt; i++) {
 219                 unsigned long addr = lmb.reserved.region[i].base +
 220                                      lmb_size_bytes(&lmb.reserved, i) - 1;
 221                 if (addr < total_lowmem)
 222                         reserve_bootmem(lmb.reserved.region[i].base,
 223                                         lmb_size_bytes(&lmb.reserved, i));
 224                 else if (lmb.reserved.region[i].base < total_lowmem) {
 225                         unsigned long adjusted_size = total_lowmem -
 226                                       lmb.reserved.region[i].base;
 227                         reserve_bootmem(lmb.reserved.region[i].base,
 228                                         adjusted_size);
 229                 }
 230         }
 231 #else
 232         free_bootmem_with_active_regions(0, max_pfn);
 233
 234         /* reserve the sections we're already using */
 235         for (i = 0; i < lmb.reserved.cnt; i++)
 236                 reserve_bootmem(lmb.reserved.region[i].base,
 237                                 lmb_size_bytes(&lmb.reserved, i));
 238
 239 #endif
 240         /* XXX need to clip this if using highmem? */
 241         sparse_memory_present_with_active_regions(0);
 242
 243         init_bootmem_done = 1;
 244 }
 245
 246 /* mark pages that don't exist as nosave */
 247 static int __init mark_nonram_nosave(void)
 248 {
 249         unsigned long lmb_next_region_start_pfn,
 250                       lmb_region_max_pfn;
 251         int i;
 252
 253         for (i = 0; i < lmb.memory.cnt - 1; i++) {
 254                 lmb_region_max_pfn =
 255                         (lmb.memory.region[i].base >> PAGE_SHIFT) +
 256                         (lmb.memory.region[i].size >> PAGE_SHIFT);
 257                 lmb_next_region_start_pfn =
 258                         lmb.memory.region[i+1].base >> PAGE_SHIFT;
 259
 260                 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
 261                         register_nosave_region(lmb_region_max_pfn,
 262                                                lmb_next_region_start_pfn);
 263         }
 264
 265         return 0;
 266 }
 267
 268 /*
 269  * paging_init() sets up the page tables - in fact we've already done this.
 270  */
 271 void __init paging_init(void)
 272 {
 273         unsigned long total_ram = lmb_phys_mem_size();
 274         unsigned long top_of_ram = lmb_end_of_DRAM();
 275         unsigned long max_zone_pfns[MAX_NR_ZONES];
 276
 277 #ifdef CONFIG_HIGHMEM
 278         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 279         pkmap_page_table = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
 280                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 281         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 282         kmap_pte = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
 283                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN),
 284                          KMAP_FIX_BEGIN);
 285         kmap_prot = PAGE_KERNEL;
 286 #endif /* CONFIG_HIGHMEM */
 287
 288         printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 289                top_of_ram, total_ram);
 290         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 291                (top_of_ram - total_ram) >> 20);
 292         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 293 #ifdef CONFIG_HIGHMEM
 294         max_zone_pfns[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 295         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 296 #else
 297         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 298 #endif
 299         free_area_init_nodes(max_zone_pfns);
 300
 301         mark_nonram_nosave();
 302 }
 303 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 304
 305 void __init mem_init(void)
 306 {
 307 #ifdef CONFIG_NEED_MULTIPLE_NODES
 308         int nid;
 309 #endif
 310         pg_data_t *pgdat;
 311         unsigned long i;
 312         struct page *page;
 313         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 314
 315         num_physpages = lmb.memory.size >> PAGE_SHIFT;
 316         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 317
 318 #ifdef CONFIG_NEED_MULTIPLE_NODES
 319         for_each_online_node(nid) {
 320                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 321                         printk("freeing bootmem node %d\n", nid);
 322                         totalram_pages +=
 323                                 free_all_bootmem_node(NODE_DATA(nid));
 324                 }
 325         }
 326 #else
 327         max_mapnr = max_pfn;
 328         totalram_pages += free_all_bootmem();
 329 #endif
 330         for_each_online_pgdat(pgdat) {
 331                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 332                         if (!pfn_valid(pgdat->node_start_pfn + i))
 333                                 continue;
 334                         page = pgdat_page_nr(pgdat, i);
 335                         if (PageReserved(page))
 336                                 reservedpages++;
 337                 }
 338         }
 339
 340         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 341         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 342         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 343         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 344
 345 #ifdef CONFIG_HIGHMEM
 346         {
 347                 unsigned long pfn, highmem_mapnr;
 348
 349                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 350                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 351                         struct page *page = pfn_to_page(pfn);
 352                         if (lmb_is_reserved(pfn << PAGE_SHIFT))
 353                                 continue;
 354                         ClearPageReserved(page);
 355                         init_page_count(page);
 356                         __free_page(page);
 357                         totalhigh_pages++;
 358                         reservedpages--;
 359                 }
 360                 totalram_pages += totalhigh_pages;
 361                 printk(KERN_DEBUG "High memory: %luk\n",
 362                        totalhigh_pages << (PAGE_SHIFT-10));
 363         }
 364 #endif /* CONFIG_HIGHMEM */
 365
 366         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 367                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 368                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 369                 num_physpages << (PAGE_SHIFT-10),
 370                 codesize >> 10,
 371                 reservedpages << (PAGE_SHIFT-10),
 372                 datasize >> 10,
 373                 bsssize >> 10,
 374                 initsize >> 10);
 375
 376         mem_init_done = 1;
 377 }
 378
 379 /*
 380  * This is called when a page has been modified by the kernel.
 381  * It just marks the page as not i-cache clean.  We do the i-cache
 382  * flush later when the page is given to a user process, if necessary.
 383  */
 384 void flush_dcache_page(struct page *page)
 385 {
 386         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 387                 return;
 388         /* avoid an atomic op if possible */
 389         if (test_bit(PG_arch_1, &page->flags))
 390                 clear_bit(PG_arch_1, &page->flags);
 391 }
 392 EXPORT_SYMBOL(flush_dcache_page);
 393
 394 void flush_dcache_icache_page(struct page *page)
 395 {
 396 #ifdef CONFIG_BOOKE
 397         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 398         __flush_dcache_icache(start);
 399         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 400 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 401         /* On 8xx there is no need to kmap since highmem is not supported */
 402         __flush_dcache_icache(page_address(page));
 403 #else
 404         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 405 #endif
 406
 407 }
 408 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 409 {
 410         clear_page(page);
 411
 412         /*
 413          * We shouldnt have to do this, but some versions of glibc
 414          * require it (ld.so assumes zero filled pages are icache clean)
 415          * - Anton
 416          */
 417         flush_dcache_page(pg);
 418 }
 419 EXPORT_SYMBOL(clear_user_page);
 420
 421 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 422                     struct page *pg)
 423 {
 424         copy_page(vto, vfrom);
 425
 426         /*
 427          * We should be able to use the following optimisation, however
 428          * there are two problems.
 429          * Firstly a bug in some versions of binutils meant PLT sections
 430          * were not marked executable.
 431          * Secondly the first word in the GOT section is blrl, used
 432          * to establish the GOT address. Until recently the GOT was
 433          * not marked executable.
 434          * - Anton
 435          */
 436 #if 0
 437         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 438                 return;
 439 #endif
 440
 441         flush_dcache_page(pg);
 442 }
 443
 444 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 445                              unsigned long addr, int len)
 446 {
 447         unsigned long maddr;
 448
 449         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 450         flush_icache_range(maddr, maddr + len);
 451         kunmap(page);
 452 }
 453 EXPORT_SYMBOL(flush_icache_user_range);
 454
 455 /*
 456  * This is called at the end of handling a user page fault, when the
 457  * fault has been handled by updating a PTE in the linux page tables.
 458  * We use it to preload an HPTE into the hash table corresponding to
 459  * the updated linux PTE.
 460  *
 461  * This must always be called with the pte lock held.
 462  */
 463 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 464                       pte_t pte)
 465 {
 466 #ifdef CONFIG_PPC_STD_MMU
 467         unsigned long access = 0, trap;
 468 #endif
 469         unsigned long pfn = pte_pfn(pte);
 470
 471         /* handle i-cache coherency */
 472         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 473             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 474             pfn_valid(pfn)) {
 475                 struct page *page = pfn_to_page(pfn);
 476 #ifdef CONFIG_8xx
 477                 /* On 8xx, cache control instructions (particularly
 478                  * "dcbst" from flush_dcache_icache) fault as write
 479                  * operation if there is an unpopulated TLB entry
 480                  * for the address in question. To workaround that,
 481                  * we invalidate the TLB here, thus avoiding dcbst
 482                  * misbehaviour.
 483                  */
 484                 _tlbie(address, 0 /* 8xx doesn't care about PID */);
 485 #endif
 486                 if (!PageReserved(page)
 487                     && !test_bit(PG_arch_1, &page->flags)) {
 488                         if (vma->vm_mm == current->active_mm) {
 489                                 __flush_dcache_icache((void *) address);
 490                         } else
 491                                 flush_dcache_icache_page(page);
 492                         set_bit(PG_arch_1, &page->flags);
 493                 }
 494         }
 495
 496 #ifdef CONFIG_PPC_STD_MMU
 497         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 498         if (!pte_young(pte) || address >= TASK_SIZE)
 499                 return;
 500
 501         /* We try to figure out if we are coming from an instruction
 502          * access fault and pass that down to __hash_page so we avoid
 503          * double-faulting on execution of fresh text. We have to test
 504          * for regs NULL since init will get here first thing at boot
 505          *
 506          * We also avoid filling the hash if not coming from a fault
 507          */
 508         if (current->thread.regs == NULL)
 509                 return;
 510         trap = TRAP(current->thread.regs);
 511         if (trap == 0x400)
 512                 access |= _PAGE_EXEC;
 513         else if (trap != 0x300)
 514                 return;
 515         hash_preload(vma->vm_mm, address, access, trap);
 516 #endif /* CONFIG_PPC_STD_MMU */
 517 }