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  *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
   9  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
  10  *
  11  *  Derived from "arch/i386/mm/init.c"
  12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  13  *
  14  *  This program is free software; you can redistribute it and/or
  15  *  modify it under the terms of the GNU General Public License
  16  *  as published by the Free Software Foundation; either version
  17  *  2 of the License, or (at your option) any later version.
  18  *
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/sched.h>
  23 #include <linux/kernel.h>
  24 #include <linux/errno.h>
  25 #include <linux/string.h>
  26 #include <linux/types.h>
  27 #include <linux/mm.h>
  28 #include <linux/stddef.h>
  29 #include <linux/init.h>
  30 #include <linux/bootmem.h>
  31 #include <linux/highmem.h>
  32 #include <linux/initrd.h>
  33 #include <linux/pagemap.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/prom.h>
  46 #include <asm/lmb.h>
  47 #include <asm/sections.h>
  48 #include <asm/vdso.h>
  49
  50 #include "mmu_decl.h"
  51
  52 #ifndef CPU_FTR_COHERENT_ICACHE
  53 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  54 #define CPU_FTR_NOEXECUTE       0
  55 #endif
  56
  57 int init_bootmem_done;
  58 int mem_init_done;
  59 unsigned long memory_limit;
  60
  61 extern void hash_preload(struct mm_struct *mm, unsigned long ea,
  62                          unsigned long access, unsigned long trap);
  63
  64 /*
  65  * This is called by /dev/mem to know if a given address has to
  66  * be mapped non-cacheable or not
  67  */
  68 int page_is_ram(unsigned long pfn)
  69 {
  70         unsigned long paddr = (pfn << PAGE_SHIFT);
  71
  72 #ifndef CONFIG_PPC64    /* XXX for now */
  73         return paddr < __pa(high_memory);
  74 #else
  75         int i;
  76         for (i=0; i < lmb.memory.cnt; i++) {
  77                 unsigned long base;
  78
  79                 base = lmb.memory.region[i].base;
  80
  81                 if ((paddr >= base) &&
  82                         (paddr < (base + lmb.memory.region[i].size))) {
  83                         return 1;
  84                 }
  85         }
  86
  87         return 0;
  88 #endif
  89 }
  90 EXPORT_SYMBOL(page_is_ram);
  91
  92 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  93                               unsigned long size, pgprot_t vma_prot)
  94 {
  95         if (ppc_md.phys_mem_access_prot)
  96                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  97
  98         if (!page_is_ram(pfn))
  99                 vma_prot = __pgprot(pgprot_val(vma_prot)
 100                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 101         return vma_prot;
 102 }
 103 EXPORT_SYMBOL(phys_mem_access_prot);
 104
 105 #ifdef CONFIG_MEMORY_HOTPLUG
 106
 107 void online_page(struct page *page)
 108 {
 109         ClearPageReserved(page);
 110         init_page_count(page);
 111         __free_page(page);
 112         totalram_pages++;
 113         num_physpages++;
 114 }
 115
 116 #ifdef CONFIG_NUMA
 117 int memory_add_physaddr_to_nid(u64 start)
 118 {
 119         return hot_add_scn_to_nid(start);
 120 }
 121 #endif
 122
 123 int __devinit arch_add_memory(int nid, u64 start, u64 size)
 124 {
 125         struct pglist_data *pgdata;
 126         struct zone *zone;
 127         unsigned long start_pfn = start >> PAGE_SHIFT;
 128         unsigned long nr_pages = size >> PAGE_SHIFT;
 129
 130         pgdata = NODE_DATA(nid);
 131
 132         start = (unsigned long)__va(start);
 133         create_section_mapping(start, start + size);
 134
 135         /* this should work for most non-highmem platforms */
 136         zone = pgdata->node_zones;
 137
 138         return __add_pages(zone, start_pfn, nr_pages);
 139
 140         return 0;
 141 }
 142
 143 /*
 144  * First pass at this code will check to determine if the remove
 145  * request is within the RMO.  Do not allow removal within the RMO.
 146  */
 147 int __devinit remove_memory(u64 start, u64 size)
 148 {
 149         struct zone *zone;
 150         unsigned long start_pfn, end_pfn, nr_pages;
 151
 152         start_pfn = start >> PAGE_SHIFT;
 153         nr_pages = size >> PAGE_SHIFT;
 154         end_pfn = start_pfn + nr_pages;
 155
 156         printk("%s(): Attempting to remove memoy in range "
 157                         "%lx to %lx\n", __func__, start, start+size);
 158         /*
 159          * check for range within RMO
 160          */
 161         zone = page_zone(pfn_to_page(start_pfn));
 162
 163         printk("%s(): memory will be removed from "
 164                         "the %s zone\n", __func__, zone->name);
 165
 166         /*
 167          * not handling removing memory ranges that
 168          * overlap multiple zones yet
 169          */
 170         if (end_pfn > (zone->zone_start_pfn + zone->spanned_pages))
 171                 goto overlap;
 172
 173         /* make sure it is NOT in RMO */
 174         if ((start < lmb.rmo_size) || ((start+size) < lmb.rmo_size)) {
 175                 printk("%s(): range to be removed must NOT be in RMO!\n",
 176                         __func__);
 177                 goto in_rmo;
 178         }
 179
 180         return __remove_pages(zone, start_pfn, nr_pages);
 181
 182 overlap:
 183         printk("%s(): memory range to be removed overlaps "
 184                 "multiple zones!!!\n", __func__);
 185 in_rmo:
 186         return -1;
 187 }
 188 #endif /* CONFIG_MEMORY_HOTPLUG */
 189
 190 void show_mem(void)
 191 {
 192         unsigned long total = 0, reserved = 0;
 193         unsigned long shared = 0, cached = 0;
 194         unsigned long highmem = 0;
 195         struct page *page;
 196         pg_data_t *pgdat;
 197         unsigned long i;
 198
 199         printk("Mem-info:\n");
 200         show_free_areas();
 201         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 202         for_each_online_pgdat(pgdat) {
 203                 unsigned long flags;
 204                 pgdat_resize_lock(pgdat, &flags);
 205                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 206                         if (!pfn_valid(pgdat->node_start_pfn + i))
 207                                 continue;
 208                         page = pgdat_page_nr(pgdat, i);
 209                         total++;
 210                         if (PageHighMem(page))
 211                                 highmem++;
 212                         if (PageReserved(page))
 213                                 reserved++;
 214                         else if (PageSwapCache(page))
 215                                 cached++;
 216                         else if (page_count(page))
 217                                 shared += page_count(page) - 1;
 218                 }
 219                 pgdat_resize_unlock(pgdat, &flags);
 220         }
 221         printk("%ld pages of RAM\n", total);
 222 #ifdef CONFIG_HIGHMEM
 223         printk("%ld pages of HIGHMEM\n", highmem);
 224 #endif
 225         printk("%ld reserved pages\n", reserved);
 226         printk("%ld pages shared\n", shared);
 227         printk("%ld pages swap cached\n", cached);
 228 }
 229
 230 /*
 231  * Initialize the bootmem system and give it all the memory we
 232  * have available.  If we are using highmem, we only put the
 233  * lowmem into the bootmem system.
 234  */
 235 #ifndef CONFIG_NEED_MULTIPLE_NODES
 236 void __init do_init_bootmem(void)
 237 {
 238         unsigned long i;
 239         unsigned long start, bootmap_pages;
 240         unsigned long total_pages;
 241         int boot_mapsize;
 242
 243         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 244 #ifdef CONFIG_HIGHMEM
 245         total_pages = total_lowmem >> PAGE_SHIFT;
 246 #endif
 247
 248         /*
 249          * Find an area to use for the bootmem bitmap.  Calculate the size of
 250          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 251          * Add 1 additional page in case the address isn't page-aligned.
 252          */
 253         bootmap_pages = bootmem_bootmap_pages(total_pages);
 254
 255         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 256
 257         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 258
 259         /* Add active regions with valid PFNs */
 260         for (i = 0; i < lmb.memory.cnt; i++) {
 261                 unsigned long start_pfn, end_pfn;
 262                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
 263                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
 264                 add_active_range(0, start_pfn, end_pfn);
 265         }
 266
 267         /* Add all physical memory to the bootmem map, mark each area
 268          * present.
 269          */
 270 #ifdef CONFIG_HIGHMEM
 271         free_bootmem_with_active_regions(0, total_lowmem >> PAGE_SHIFT);
 272 #else
 273         free_bootmem_with_active_regions(0, max_pfn);
 274 #endif
 275
 276         /* reserve the sections we're already using */
 277         for (i = 0; i < lmb.reserved.cnt; i++)
 278                 reserve_bootmem(lmb.reserved.region[i].base,
 279                                 lmb_size_bytes(&lmb.reserved, i));
 280
 281         /* XXX need to clip this if using highmem? */
 282         sparse_memory_present_with_active_regions(0);
 283
 284         init_bootmem_done = 1;
 285 }
 286
 287 /*
 288  * paging_init() sets up the page tables - in fact we've already done this.
 289  */
 290 void __init paging_init(void)
 291 {
 292         unsigned long total_ram = lmb_phys_mem_size();
 293         unsigned long top_of_ram = lmb_end_of_DRAM();
 294         unsigned long max_zone_pfns[MAX_NR_ZONES];
 295
 296 #ifdef CONFIG_HIGHMEM
 297         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 298         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
 299                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 300         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 301         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
 302                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 303         kmap_prot = PAGE_KERNEL;
 304 #endif /* CONFIG_HIGHMEM */
 305
 306         printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 307                top_of_ram, total_ram);
 308         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 309                (top_of_ram - total_ram) >> 20);
 310         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 311 #ifdef CONFIG_HIGHMEM
 312         max_zone_pfns[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 313         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 314 #else
 315         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 316 #endif
 317         free_area_init_nodes(max_zone_pfns);
 318 }
 319 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 320
 321 void __init mem_init(void)
 322 {
 323 #ifdef CONFIG_NEED_MULTIPLE_NODES
 324         int nid;
 325 #endif
 326         pg_data_t *pgdat;
 327         unsigned long i;
 328         struct page *page;
 329         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 330
 331         num_physpages = lmb.memory.size >> PAGE_SHIFT;
 332         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 333
 334 #ifdef CONFIG_NEED_MULTIPLE_NODES
 335         for_each_online_node(nid) {
 336                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 337                         printk("freeing bootmem node %d\n", nid);
 338                         totalram_pages +=
 339                                 free_all_bootmem_node(NODE_DATA(nid));
 340                 }
 341         }
 342 #else
 343         max_mapnr = max_pfn;
 344         totalram_pages += free_all_bootmem();
 345 #endif
 346         for_each_online_pgdat(pgdat) {
 347                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 348                         if (!pfn_valid(pgdat->node_start_pfn + i))
 349                                 continue;
 350                         page = pgdat_page_nr(pgdat, i);
 351                         if (PageReserved(page))
 352                                 reservedpages++;
 353                 }
 354         }
 355
 356         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 357         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 358         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 359         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 360
 361 #ifdef CONFIG_HIGHMEM
 362         {
 363                 unsigned long pfn, highmem_mapnr;
 364
 365                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 366                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 367                         struct page *page = pfn_to_page(pfn);
 368
 369                         ClearPageReserved(page);
 370                         init_page_count(page);
 371                         __free_page(page);
 372                         totalhigh_pages++;
 373                 }
 374                 totalram_pages += totalhigh_pages;
 375                 printk(KERN_DEBUG "High memory: %luk\n",
 376                        totalhigh_pages << (PAGE_SHIFT-10));
 377         }
 378 #endif /* CONFIG_HIGHMEM */
 379
 380         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 381                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 382                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 383                 num_physpages << (PAGE_SHIFT-10),
 384                 codesize >> 10,
 385                 reservedpages << (PAGE_SHIFT-10),
 386                 datasize >> 10,
 387                 bsssize >> 10,
 388                 initsize >> 10);
 389
 390         mem_init_done = 1;
 391
 392         /* Initialize the vDSO */
 393         vdso_init();
 394 }
 395
 396 /*
 397  * This is called when a page has been modified by the kernel.
 398  * It just marks the page as not i-cache clean.  We do the i-cache
 399  * flush later when the page is given to a user process, if necessary.
 400  */
 401 void flush_dcache_page(struct page *page)
 402 {
 403         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 404                 return;
 405         /* avoid an atomic op if possible */
 406         if (test_bit(PG_arch_1, &page->flags))
 407                 clear_bit(PG_arch_1, &page->flags);
 408 }
 409 EXPORT_SYMBOL(flush_dcache_page);
 410
 411 void flush_dcache_icache_page(struct page *page)
 412 {
 413 #ifdef CONFIG_BOOKE
 414         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 415         __flush_dcache_icache(start);
 416         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 417 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 418         /* On 8xx there is no need to kmap since highmem is not supported */
 419         __flush_dcache_icache(page_address(page));
 420 #else
 421         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 422 #endif
 423
 424 }
 425 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 426 {
 427         clear_page(page);
 428
 429         /*
 430          * We shouldnt have to do this, but some versions of glibc
 431          * require it (ld.so assumes zero filled pages are icache clean)
 432          * - Anton
 433          */
 434         flush_dcache_page(pg);
 435 }
 436 EXPORT_SYMBOL(clear_user_page);
 437
 438 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 439                     struct page *pg)
 440 {
 441         copy_page(vto, vfrom);
 442
 443         /*
 444          * We should be able to use the following optimisation, however
 445          * there are two problems.
 446          * Firstly a bug in some versions of binutils meant PLT sections
 447          * were not marked executable.
 448          * Secondly the first word in the GOT section is blrl, used
 449          * to establish the GOT address. Until recently the GOT was
 450          * not marked executable.
 451          * - Anton
 452          */
 453 #if 0
 454         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 455                 return;
 456 #endif
 457
 458         flush_dcache_page(pg);
 459 }
 460
 461 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 462                              unsigned long addr, int len)
 463 {
 464         unsigned long maddr;
 465
 466         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 467         flush_icache_range(maddr, maddr + len);
 468         kunmap(page);
 469 }
 470 EXPORT_SYMBOL(flush_icache_user_range);
 471
 472 /*
 473  * This is called at the end of handling a user page fault, when the
 474  * fault has been handled by updating a PTE in the linux page tables.
 475  * We use it to preload an HPTE into the hash table corresponding to
 476  * the updated linux PTE.
 477  *
 478  * This must always be called with the pte lock held.
 479  */
 480 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 481                       pte_t pte)
 482 {
 483 #ifdef CONFIG_PPC_STD_MMU
 484         unsigned long access = 0, trap;
 485 #endif
 486         unsigned long pfn = pte_pfn(pte);
 487
 488         /* handle i-cache coherency */
 489         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 490             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 491             pfn_valid(pfn)) {
 492                 struct page *page = pfn_to_page(pfn);
 493                 if (!PageReserved(page)
 494                     && !test_bit(PG_arch_1, &page->flags)) {
 495                         if (vma->vm_mm == current->active_mm) {
 496 #ifdef CONFIG_8xx
 497                         /* On 8xx, cache control instructions (particularly
 498                          * "dcbst" from flush_dcache_icache) fault as write
 499                          * operation if there is an unpopulated TLB entry
 500                          * for the address in question. To workaround that,
 501                          * we invalidate the TLB here, thus avoiding dcbst
 502                          * misbehaviour.
 503                          */
 504                                 _tlbie(address);
 505 #endif
 506                                 __flush_dcache_icache((void *) address);
 507                         } else
 508                                 flush_dcache_icache_page(page);
 509                         set_bit(PG_arch_1, &page->flags);
 510                 }
 511         }
 512
 513 #ifdef CONFIG_PPC_STD_MMU
 514         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 515         if (!pte_young(pte) || address >= TASK_SIZE)
 516                 return;
 517
 518         /* We try to figure out if we are coming from an instruction
 519          * access fault and pass that down to __hash_page so we avoid
 520          * double-faulting on execution of fresh text. We have to test
 521          * for regs NULL since init will get here first thing at boot
 522          *
 523          * We also avoid filling the hash if not coming from a fault
 524          */
 525         if (current->thread.regs == NULL)
 526                 return;
 527         trap = TRAP(current->thread.regs);
 528         if (trap == 0x400)
 529                 access |= _PAGE_EXEC;
 530         else if (trap != 0x300)
 531                 return;
 532         hash_preload(vma->vm_mm, address, access, trap);
 533 #endif /* CONFIG_PPC_STD_MMU */
 534 }