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 #include <linux/lmb.h>
  35
  36 #include <asm/pgalloc.h>
  37 #include <asm/prom.h>
  38 #include <asm/io.h>
  39 #include <asm/mmu_context.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/mmu.h>
  42 #include <asm/smp.h>
  43 #include <asm/machdep.h>
  44 #include <asm/btext.h>
  45 #include <asm/tlb.h>
  46 #include <asm/sections.h>
  47 #include <asm/vdso.h>
  48 #include <asm/fixmap.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 #ifdef CONFIG_HIGHMEM
  62 pte_t *kmap_pte;
  63 pgprot_t kmap_prot;
  64
  65 EXPORT_SYMBOL(kmap_prot);
  66 EXPORT_SYMBOL(kmap_pte);
  67
  68 static inline pte_t *virt_to_kpte(unsigned long vaddr)
  69 {
  70         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  71                         vaddr), vaddr), vaddr);
  72 }
  73 #endif
  74
  75 int page_is_ram(unsigned long pfn)
  76 {
  77         unsigned long paddr = (pfn << PAGE_SHIFT);
  78
  79 #ifndef CONFIG_PPC64    /* XXX for now */
  80         return paddr < __pa(high_memory);
  81 #else
  82         int i;
  83         for (i=0; i < lmb.memory.cnt; i++) {
  84                 unsigned long base;
  85
  86                 base = lmb.memory.region[i].base;
  87
  88                 if ((paddr >= base) &&
  89                         (paddr < (base + lmb.memory.region[i].size))) {
  90                         return 1;
  91                 }
  92         }
  93
  94         return 0;
  95 #endif
  96 }
  97
  98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  99                               unsigned long size, pgprot_t vma_prot)
 100 {
 101         if (ppc_md.phys_mem_access_prot)
 102                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 103
 104         if (!page_is_ram(pfn))
 105                 vma_prot = __pgprot(pgprot_val(vma_prot)
 106                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 107         return vma_prot;
 108 }
 109 EXPORT_SYMBOL(phys_mem_access_prot);
 110
 111 #ifdef CONFIG_MEMORY_HOTPLUG
 112
 113 #ifdef CONFIG_NUMA
 114 int memory_add_physaddr_to_nid(u64 start)
 115 {
 116         return hot_add_scn_to_nid(start);
 117 }
 118 #endif
 119
 120 int arch_add_memory(int nid, u64 start, u64 size)
 121 {
 122         struct pglist_data *pgdata;
 123         struct zone *zone;
 124         unsigned long start_pfn = start >> PAGE_SHIFT;
 125         unsigned long nr_pages = size >> PAGE_SHIFT;
 126
 127         pgdata = NODE_DATA(nid);
 128
 129         start = (unsigned long)__va(start);
 130         create_section_mapping(start, start + size);
 131
 132         /* this should work for most non-highmem platforms */
 133         zone = pgdata->node_zones;
 134
 135         return __add_pages(zone, start_pfn, nr_pages);
 136 }
 137
 138 #ifdef CONFIG_MEMORY_HOTREMOVE
 139 int remove_memory(u64 start, u64 size)
 140 {
 141         unsigned long start_pfn, end_pfn;
 142         int ret;
 143
 144         start_pfn = start >> PAGE_SHIFT;
 145         end_pfn = start_pfn + (size >> PAGE_SHIFT);
 146         ret = offline_pages(start_pfn, end_pfn, 120 * HZ);
 147         if (ret)
 148                 goto out;
 149         /* Arch-specific calls go here - next patch */
 150 out:
 151         return ret;
 152 }
 153 #endif /* CONFIG_MEMORY_HOTREMOVE */
 154
 155 /*
 156  * walk_memory_resource() needs to make sure there is no holes in a given
 157  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 158  * Instead it maintains it in lmb.memory structures.  Walk through the
 159  * memory regions, find holes and callback for contiguous regions.
 160  */
 161 int
 162 walk_memory_resource(unsigned long start_pfn, unsigned long nr_pages, void *arg,
 163                         int (*func)(unsigned long, unsigned long, void *))
 164 {
 165         struct lmb_property res;
 166         unsigned long pfn, len;
 167         u64 end;
 168         int ret = -1;
 169
 170         res.base = (u64) start_pfn << PAGE_SHIFT;
 171         res.size = (u64) nr_pages << PAGE_SHIFT;
 172
 173         end = res.base + res.size - 1;
 174         while ((res.base < end) && (lmb_find(&res) >= 0)) {
 175                 pfn = (unsigned long)(res.base >> PAGE_SHIFT);
 176                 len = (unsigned long)(res.size >> PAGE_SHIFT);
 177                 ret = (*func)(pfn, len, arg);
 178                 if (ret)
 179                         break;
 180                 res.base += (res.size + 1);
 181                 res.size = (end - res.base + 1);
 182         }
 183         return ret;
 184 }
 185 EXPORT_SYMBOL_GPL(walk_memory_resource);
 186
 187 #endif /* CONFIG_MEMORY_HOTPLUG */
 188
 189 void show_mem(void)
 190 {
 191         unsigned long total = 0, reserved = 0;
 192         unsigned long shared = 0, cached = 0;
 193         unsigned long highmem = 0;
 194         struct page *page;
 195         pg_data_t *pgdat;
 196         unsigned long i;
 197
 198         printk("Mem-info:\n");
 199         show_free_areas();
 200         for_each_online_pgdat(pgdat) {
 201                 unsigned long flags;
 202                 pgdat_resize_lock(pgdat, &flags);
 203                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 204                         if (!pfn_valid(pgdat->node_start_pfn + i))
 205                                 continue;
 206                         page = pgdat_page_nr(pgdat, i);
 207                         total++;
 208                         if (PageHighMem(page))
 209                                 highmem++;
 210                         if (PageReserved(page))
 211                                 reserved++;
 212                         else if (PageSwapCache(page))
 213                                 cached++;
 214                         else if (page_count(page))
 215                                 shared += page_count(page) - 1;
 216                 }
 217                 pgdat_resize_unlock(pgdat, &flags);
 218         }
 219         printk("%ld pages of RAM\n", total);
 220 #ifdef CONFIG_HIGHMEM
 221         printk("%ld pages of HIGHMEM\n", highmem);
 222 #endif
 223         printk("%ld reserved pages\n", reserved);
 224         printk("%ld pages shared\n", shared);
 225         printk("%ld pages swap cached\n", cached);
 226 }
 227
 228 /*
 229  * Initialize the bootmem system and give it all the memory we
 230  * have available.  If we are using highmem, we only put the
 231  * lowmem into the bootmem system.
 232  */
 233 #ifndef CONFIG_NEED_MULTIPLE_NODES
 234 void __init do_init_bootmem(void)
 235 {
 236         unsigned long i;
 237         unsigned long start, bootmap_pages;
 238         unsigned long total_pages;
 239         int boot_mapsize;
 240
 241         max_low_pfn = max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
 242         total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 243 #ifdef CONFIG_HIGHMEM
 244         total_pages = total_lowmem >> PAGE_SHIFT;
 245         max_low_pfn = lowmem_end_addr >> 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         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 258         boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
 259
 260         /* Add active regions with valid PFNs */
 261         for (i = 0; i < lmb.memory.cnt; i++) {
 262                 unsigned long start_pfn, end_pfn;
 263                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
 264                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
 265                 add_active_range(0, start_pfn, end_pfn);
 266         }
 267
 268         /* Add all physical memory to the bootmem map, mark each area
 269          * present.
 270          */
 271 #ifdef CONFIG_HIGHMEM
 272         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 273
 274         /* reserve the sections we're already using */
 275         for (i = 0; i < lmb.reserved.cnt; i++) {
 276                 unsigned long addr = lmb.reserved.region[i].base +
 277                                      lmb_size_bytes(&lmb.reserved, i) - 1;
 278                 if (addr < lowmem_end_addr)
 279                         reserve_bootmem(lmb.reserved.region[i].base,
 280                                         lmb_size_bytes(&lmb.reserved, i),
 281                                         BOOTMEM_DEFAULT);
 282                 else if (lmb.reserved.region[i].base < lowmem_end_addr) {
 283                         unsigned long adjusted_size = lowmem_end_addr -
 284                                       lmb.reserved.region[i].base;
 285                         reserve_bootmem(lmb.reserved.region[i].base,
 286                                         adjusted_size, BOOTMEM_DEFAULT);
 287                 }
 288         }
 289 #else
 290         free_bootmem_with_active_regions(0, max_pfn);
 291
 292         /* reserve the sections we're already using */
 293         for (i = 0; i < lmb.reserved.cnt; i++)
 294                 reserve_bootmem(lmb.reserved.region[i].base,
 295                                 lmb_size_bytes(&lmb.reserved, i),
 296                                 BOOTMEM_DEFAULT);
 297
 298 #endif
 299         /* XXX need to clip this if using highmem? */
 300         sparse_memory_present_with_active_regions(0);
 301
 302         init_bootmem_done = 1;
 303 }
 304
 305 /* mark pages that don't exist as nosave */
 306 static int __init mark_nonram_nosave(void)
 307 {
 308         unsigned long lmb_next_region_start_pfn,
 309                       lmb_region_max_pfn;
 310         int i;
 311
 312         for (i = 0; i < lmb.memory.cnt - 1; i++) {
 313                 lmb_region_max_pfn =
 314                         (lmb.memory.region[i].base >> PAGE_SHIFT) +
 315                         (lmb.memory.region[i].size >> PAGE_SHIFT);
 316                 lmb_next_region_start_pfn =
 317                         lmb.memory.region[i+1].base >> PAGE_SHIFT;
 318
 319                 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
 320                         register_nosave_region(lmb_region_max_pfn,
 321                                                lmb_next_region_start_pfn);
 322         }
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * paging_init() sets up the page tables - in fact we've already done this.
 329  */
 330 void __init paging_init(void)
 331 {
 332         unsigned long total_ram = lmb_phys_mem_size();
 333         unsigned long top_of_ram = lmb_end_of_DRAM();
 334         unsigned long max_zone_pfns[MAX_NR_ZONES];
 335
 336 #ifdef CONFIG_PPC32
 337         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 338         unsigned long end = __fix_to_virt(FIX_HOLE);
 339
 340         for (; v < end; v += PAGE_SIZE)
 341                 map_page(v, 0, 0); /* XXX gross */
 342 #endif
 343
 344 #ifdef CONFIG_HIGHMEM
 345         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 346         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 347
 348         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 349         kmap_prot = PAGE_KERNEL;
 350 #endif /* CONFIG_HIGHMEM */
 351
 352         printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 353                top_of_ram, total_ram);
 354         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 355                (top_of_ram - total_ram) >> 20);
 356         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 357 #ifdef CONFIG_HIGHMEM
 358         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 359         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 360 #else
 361         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 362 #endif
 363         free_area_init_nodes(max_zone_pfns);
 364
 365         mark_nonram_nosave();
 366 }
 367 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 368
 369 void __init mem_init(void)
 370 {
 371 #ifdef CONFIG_NEED_MULTIPLE_NODES
 372         int nid;
 373 #endif
 374         pg_data_t *pgdat;
 375         unsigned long i;
 376         struct page *page;
 377         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 378
 379         num_physpages = lmb.memory.size >> PAGE_SHIFT;
 380         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 381
 382 #ifdef CONFIG_NEED_MULTIPLE_NODES
 383         for_each_online_node(nid) {
 384                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 385                         printk("freeing bootmem node %d\n", nid);
 386                         totalram_pages +=
 387                                 free_all_bootmem_node(NODE_DATA(nid));
 388                 }
 389         }
 390 #else
 391         max_mapnr = max_pfn;
 392         totalram_pages += free_all_bootmem();
 393 #endif
 394         for_each_online_pgdat(pgdat) {
 395                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 396                         if (!pfn_valid(pgdat->node_start_pfn + i))
 397                                 continue;
 398                         page = pgdat_page_nr(pgdat, i);
 399                         if (PageReserved(page))
 400                                 reservedpages++;
 401                 }
 402         }
 403
 404         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 405         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 406         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 407         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 408
 409 #ifdef CONFIG_HIGHMEM
 410         {
 411                 unsigned long pfn, highmem_mapnr;
 412
 413                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 414                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 415                         struct page *page = pfn_to_page(pfn);
 416                         if (lmb_is_reserved(pfn << PAGE_SHIFT))
 417                                 continue;
 418                         ClearPageReserved(page);
 419                         init_page_count(page);
 420                         __free_page(page);
 421                         totalhigh_pages++;
 422                         reservedpages--;
 423                 }
 424                 totalram_pages += totalhigh_pages;
 425                 printk(KERN_DEBUG "High memory: %luk\n",
 426                        totalhigh_pages << (PAGE_SHIFT-10));
 427         }
 428 #endif /* CONFIG_HIGHMEM */
 429
 430         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 431                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 432                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 433                 num_physpages << (PAGE_SHIFT-10),
 434                 codesize >> 10,
 435                 reservedpages << (PAGE_SHIFT-10),
 436                 datasize >> 10,
 437                 bsssize >> 10,
 438                 initsize >> 10);
 439
 440         mem_init_done = 1;
 441 }
 442
 443 /*
 444  * This is called when a page has been modified by the kernel.
 445  * It just marks the page as not i-cache clean.  We do the i-cache
 446  * flush later when the page is given to a user process, if necessary.
 447  */
 448 void flush_dcache_page(struct page *page)
 449 {
 450         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 451                 return;
 452         /* avoid an atomic op if possible */
 453         if (test_bit(PG_arch_1, &page->flags))
 454                 clear_bit(PG_arch_1, &page->flags);
 455 }
 456 EXPORT_SYMBOL(flush_dcache_page);
 457
 458 void flush_dcache_icache_page(struct page *page)
 459 {
 460 #ifdef CONFIG_BOOKE
 461         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 462         __flush_dcache_icache(start);
 463         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 464 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 465         /* On 8xx there is no need to kmap since highmem is not supported */
 466         __flush_dcache_icache(page_address(page));
 467 #else
 468         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 469 #endif
 470
 471 }
 472 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 473 {
 474         clear_page(page);
 475
 476         /*
 477          * We shouldnt have to do this, but some versions of glibc
 478          * require it (ld.so assumes zero filled pages are icache clean)
 479          * - Anton
 480          */
 481         flush_dcache_page(pg);
 482 }
 483 EXPORT_SYMBOL(clear_user_page);
 484
 485 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 486                     struct page *pg)
 487 {
 488         copy_page(vto, vfrom);
 489
 490         /*
 491          * We should be able to use the following optimisation, however
 492          * there are two problems.
 493          * Firstly a bug in some versions of binutils meant PLT sections
 494          * were not marked executable.
 495          * Secondly the first word in the GOT section is blrl, used
 496          * to establish the GOT address. Until recently the GOT was
 497          * not marked executable.
 498          * - Anton
 499          */
 500 #if 0
 501         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 502                 return;
 503 #endif
 504
 505         flush_dcache_page(pg);
 506 }
 507
 508 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 509                              unsigned long addr, int len)
 510 {
 511         unsigned long maddr;
 512
 513         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 514         flush_icache_range(maddr, maddr + len);
 515         kunmap(page);
 516 }
 517 EXPORT_SYMBOL(flush_icache_user_range);
 518
 519 /*
 520  * This is called at the end of handling a user page fault, when the
 521  * fault has been handled by updating a PTE in the linux page tables.
 522  * We use it to preload an HPTE into the hash table corresponding to
 523  * the updated linux PTE.
 524  *
 525  * This must always be called with the pte lock held.
 526  */
 527 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 528                       pte_t pte)
 529 {
 530 #ifdef CONFIG_PPC_STD_MMU
 531         unsigned long access = 0, trap;
 532 #endif
 533         unsigned long pfn = pte_pfn(pte);
 534
 535         /* handle i-cache coherency */
 536         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 537             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 538             pfn_valid(pfn)) {
 539                 struct page *page = pfn_to_page(pfn);
 540 #ifdef CONFIG_8xx
 541                 /* On 8xx, cache control instructions (particularly
 542                  * "dcbst" from flush_dcache_icache) fault as write
 543                  * operation if there is an unpopulated TLB entry
 544                  * for the address in question. To workaround that,
 545                  * we invalidate the TLB here, thus avoiding dcbst
 546                  * misbehaviour.
 547                  */
 548                 _tlbie(address, 0 /* 8xx doesn't care about PID */);
 549 #endif
 550                 /* The _PAGE_USER test should really be _PAGE_EXEC, but
 551                  * older glibc versions execute some code from no-exec
 552                  * pages, which for now we are supporting.  If exec-only
 553                  * pages are ever implemented, this will have to change.
 554                  */
 555                 if (!PageReserved(page) && (pte_val(pte) & _PAGE_USER)
 556                     && !test_bit(PG_arch_1, &page->flags)) {
 557                         if (vma->vm_mm == current->active_mm) {
 558                                 __flush_dcache_icache((void *) address);
 559                         } else
 560                                 flush_dcache_icache_page(page);
 561                         set_bit(PG_arch_1, &page->flags);
 562                 }
 563         }
 564
 565 #ifdef CONFIG_PPC_STD_MMU
 566         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 567         if (!pte_young(pte) || address >= TASK_SIZE)
 568                 return;
 569
 570         /* We try to figure out if we are coming from an instruction
 571          * access fault and pass that down to __hash_page so we avoid
 572          * double-faulting on execution of fresh text. We have to test
 573          * for regs NULL since init will get here first thing at boot
 574          *
 575          * We also avoid filling the hash if not coming from a fault
 576          */
 577         if (current->thread.regs == NULL)
 578                 return;
 579         trap = TRAP(current->thread.regs);
 580         if (trap == 0x400)
 581                 access |= _PAGE_EXEC;
 582         else if (trap != 0x300)
 583                 return;
 584         hash_preload(vma->vm_mm, address, access, trap);
 585 #endif /* CONFIG_PPC_STD_MMU */
 586 }