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/config.h>
  22 #include <linux/module.h>
  23 #include <linux/sched.h>
  24 #include <linux/kernel.h>
  25 #include <linux/errno.h>
  26 #include <linux/string.h>
  27 #include <linux/types.h>
  28 #include <linux/mm.h>
  29 #include <linux/stddef.h>
  30 #include <linux/init.h>
  31 #include <linux/bootmem.h>
  32 #include <linux/highmem.h>
  33 #include <linux/initrd.h>
  34 #include <linux/pagemap.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/prom.h>
  47 #include <asm/lmb.h>
  48 #include <asm/sections.h>
  49 #ifdef CONFIG_PPC64
  50 #include <asm/vdso.h>
  51 #endif
  52
  53 #include "mmu_decl.h"
  54
  55 #ifndef CPU_FTR_COHERENT_ICACHE
  56 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  57 #define CPU_FTR_NOEXECUTE       0
  58 #endif
  59
  60 int init_bootmem_done;
  61 int mem_init_done;
  62 unsigned long memory_limit;
  63
  64 extern void hash_preload(struct mm_struct *mm, unsigned long ea,
  65                          unsigned long access, unsigned long trap);
  66
  67 /*
  68  * This is called by /dev/mem to know if a given address has to
  69  * be mapped non-cacheable or not
  70  */
  71 int page_is_ram(unsigned long pfn)
  72 {
  73         unsigned long paddr = (pfn << PAGE_SHIFT);
  74
  75 #ifndef CONFIG_PPC64    /* XXX for now */
  76         return paddr < __pa(high_memory);
  77 #else
  78         int i;
  79         for (i=0; i < lmb.memory.cnt; i++) {
  80                 unsigned long base;
  81
  82                 base = lmb.memory.region[i].base;
  83
  84                 if ((paddr >= base) &&
  85                         (paddr < (base + lmb.memory.region[i].size))) {
  86                         return 1;
  87                 }
  88         }
  89
  90         return 0;
  91 #endif
  92 }
  93 EXPORT_SYMBOL(page_is_ram);
  94
  95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  96                               unsigned long size, pgprot_t vma_prot)
  97 {
  98         if (ppc_md.phys_mem_access_prot)
  99                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 100
 101         if (!page_is_ram(pfn))
 102                 vma_prot = __pgprot(pgprot_val(vma_prot)
 103                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 104         return vma_prot;
 105 }
 106 EXPORT_SYMBOL(phys_mem_access_prot);
 107
 108 #ifdef CONFIG_MEMORY_HOTPLUG
 109
 110 void online_page(struct page *page)
 111 {
 112         ClearPageReserved(page);
 113         free_cold_page(page);
 114         totalram_pages++;
 115         num_physpages++;
 116 }
 117
 118 /*
 119  * This works only for the non-NUMA case.  Later, we'll need a lookup
 120  * to convert from real physical addresses to nid, that doesn't use
 121  * pfn_to_nid().
 122  */
 123 int __devinit add_memory(u64 start, u64 size)
 124 {
 125         struct pglist_data *pgdata = NODE_DATA(0);
 126         struct zone *zone;
 127         unsigned long start_pfn = start >> PAGE_SHIFT;
 128         unsigned long nr_pages = size >> PAGE_SHIFT;
 129
 130         /* this should work for most non-highmem platforms */
 131         zone = pgdata->node_zones;
 132
 133         return __add_pages(zone, start_pfn, nr_pages);
 134
 135         return 0;
 136 }
 137
 138 /*
 139  * First pass at this code will check to determine if the remove
 140  * request is within the RMO.  Do not allow removal within the RMO.
 141  */
 142 int __devinit remove_memory(u64 start, u64 size)
 143 {
 144         struct zone *zone;
 145         unsigned long start_pfn, end_pfn, nr_pages;
 146
 147         start_pfn = start >> PAGE_SHIFT;
 148         nr_pages = size >> PAGE_SHIFT;
 149         end_pfn = start_pfn + nr_pages;
 150
 151         printk("%s(): Attempting to remove memoy in range "
 152                         "%lx to %lx\n", __func__, start, start+size);
 153         /*
 154          * check for range within RMO
 155          */
 156         zone = page_zone(pfn_to_page(start_pfn));
 157
 158         printk("%s(): memory will be removed from "
 159                         "the %s zone\n", __func__, zone->name);
 160
 161         /*
 162          * not handling removing memory ranges that
 163          * overlap multiple zones yet
 164          */
 165         if (end_pfn > (zone->zone_start_pfn + zone->spanned_pages))
 166                 goto overlap;
 167
 168         /* make sure it is NOT in RMO */
 169         if ((start < lmb.rmo_size) || ((start+size) < lmb.rmo_size)) {
 170                 printk("%s(): range to be removed must NOT be in RMO!\n",
 171                         __func__);
 172                 goto in_rmo;
 173         }
 174
 175         return __remove_pages(zone, start_pfn, nr_pages);
 176
 177 overlap:
 178         printk("%s(): memory range to be removed overlaps "
 179                 "multiple zones!!!\n", __func__);
 180 in_rmo:
 181         return -1;
 182 }
 183 #endif /* CONFIG_MEMORY_HOTPLUG */
 184
 185 void show_mem(void)
 186 {
 187         unsigned long total = 0, reserved = 0;
 188         unsigned long shared = 0, cached = 0;
 189         unsigned long highmem = 0;
 190         struct page *page;
 191         pg_data_t *pgdat;
 192         unsigned long i;
 193
 194         printk("Mem-info:\n");
 195         show_free_areas();
 196         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 197         for_each_pgdat(pgdat) {
 198                 unsigned long flags;
 199                 pgdat_resize_lock(pgdat, &flags);
 200                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 201                         page = pgdat_page_nr(pgdat, i);
 202                         total++;
 203                         if (PageHighMem(page))
 204                                 highmem++;
 205                         if (PageReserved(page))
 206                                 reserved++;
 207                         else if (PageSwapCache(page))
 208                                 cached++;
 209                         else if (page_count(page))
 210                                 shared += page_count(page) - 1;
 211                 }
 212                 pgdat_resize_unlock(pgdat, &flags);
 213         }
 214         printk("%ld pages of RAM\n", total);
 215 #ifdef CONFIG_HIGHMEM
 216         printk("%ld pages of HIGHMEM\n", highmem);
 217 #endif
 218         printk("%ld reserved pages\n", reserved);
 219         printk("%ld pages shared\n", shared);
 220         printk("%ld pages swap cached\n", cached);
 221 }
 222
 223 /*
 224  * Initialize the bootmem system and give it all the memory we
 225  * have available.  If we are using highmem, we only put the
 226  * lowmem into the bootmem system.
 227  */
 228 #ifndef CONFIG_NEED_MULTIPLE_NODES
 229 void __init do_init_bootmem(void)
 230 {
 231         unsigned long i;
 232         unsigned long start, bootmap_pages;
 233         unsigned long total_pages;
 234         int boot_mapsize;
 235
 236         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 237 #ifdef CONFIG_HIGHMEM
 238         total_pages = total_lowmem >> PAGE_SHIFT;
 239 #endif
 240
 241         /*
 242          * Find an area to use for the bootmem bitmap.  Calculate the size of
 243          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 244          * Add 1 additional page in case the address isn't page-aligned.
 245          */
 246         bootmap_pages = bootmem_bootmap_pages(total_pages);
 247
 248         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 249         BUG_ON(!start);
 250
 251         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 252
 253         /* Add all physical memory to the bootmem map, mark each area
 254          * present.
 255          */
 256         for (i = 0; i < lmb.memory.cnt; i++) {
 257                 unsigned long base = lmb.memory.region[i].base;
 258                 unsigned long size = lmb_size_bytes(&lmb.memory, i);
 259 #ifdef CONFIG_HIGHMEM
 260                 if (base >= total_lowmem)
 261                         continue;
 262                 if (base + size > total_lowmem)
 263                         size = total_lowmem - base;
 264 #endif
 265                 free_bootmem(base, size);
 266         }
 267
 268         /* reserve the sections we're already using */
 269         for (i = 0; i < lmb.reserved.cnt; i++)
 270                 reserve_bootmem(lmb.reserved.region[i].base,
 271                                 lmb_size_bytes(&lmb.reserved, i));
 272
 273         /* XXX need to clip this if using highmem? */
 274         for (i = 0; i < lmb.memory.cnt; i++)
 275                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
 276                                lmb_end_pfn(&lmb.memory, i));
 277         init_bootmem_done = 1;
 278 }
 279
 280 /*
 281  * paging_init() sets up the page tables - in fact we've already done this.
 282  */
 283 void __init paging_init(void)
 284 {
 285         unsigned long zones_size[MAX_NR_ZONES];
 286         unsigned long zholes_size[MAX_NR_ZONES];
 287         unsigned long total_ram = lmb_phys_mem_size();
 288         unsigned long top_of_ram = lmb_end_of_DRAM();
 289
 290 #ifdef CONFIG_HIGHMEM
 291         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 292         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
 293                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 294         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 295         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
 296                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 297         kmap_prot = PAGE_KERNEL;
 298 #endif /* CONFIG_HIGHMEM */
 299
 300         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 301                top_of_ram, total_ram);
 302         printk(KERN_INFO "Memory hole size: %ldMB\n",
 303                (top_of_ram - total_ram) >> 20);
 304         /*
 305          * All pages are DMA-able so we put them all in the DMA zone.
 306          */
 307         memset(zones_size, 0, sizeof(zones_size));
 308         memset(zholes_size, 0, sizeof(zholes_size));
 309
 310         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 311         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 312
 313 #ifdef CONFIG_HIGHMEM
 314         zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 315         zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
 316         zholes_size[ZONE_HIGHMEM] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 317 #else
 318         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 319         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 320 #endif /* CONFIG_HIGHMEM */
 321
 322         free_area_init_node(0, NODE_DATA(0), zones_size,
 323                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
 324 }
 325 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 326
 327 void __init mem_init(void)
 328 {
 329 #ifdef CONFIG_NEED_MULTIPLE_NODES
 330         int nid;
 331 #endif
 332         pg_data_t *pgdat;
 333         unsigned long i;
 334         struct page *page;
 335         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 336
 337         num_physpages = max_pfn;        /* RAM is assumed contiguous */
 338         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 339
 340 #ifdef CONFIG_NEED_MULTIPLE_NODES
 341         for_each_online_node(nid) {
 342                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 343                         printk("freeing bootmem node %x\n", nid);
 344                         totalram_pages +=
 345                                 free_all_bootmem_node(NODE_DATA(nid));
 346                 }
 347         }
 348 #else
 349         max_mapnr = num_physpages;
 350         totalram_pages += free_all_bootmem();
 351 #endif
 352         for_each_pgdat(pgdat) {
 353                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 354                         page = pgdat_page_nr(pgdat, i);
 355                         if (PageReserved(page))
 356                                 reservedpages++;
 357                 }
 358         }
 359
 360         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 361         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 362         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 363         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 364
 365 #ifdef CONFIG_HIGHMEM
 366         {
 367                 unsigned long pfn, highmem_mapnr;
 368
 369                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 370                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 371                         struct page *page = pfn_to_page(pfn);
 372
 373                         ClearPageReserved(page);
 374                         set_page_count(page, 1);
 375                         __free_page(page);
 376                         totalhigh_pages++;
 377                 }
 378                 totalram_pages += totalhigh_pages;
 379                 printk(KERN_INFO "High memory: %luk\n",
 380                        totalhigh_pages << (PAGE_SHIFT-10));
 381         }
 382 #endif /* CONFIG_HIGHMEM */
 383
 384         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 385                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 386                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 387                 num_physpages << (PAGE_SHIFT-10),
 388                 codesize >> 10,
 389                 reservedpages << (PAGE_SHIFT-10),
 390                 datasize >> 10,
 391                 bsssize >> 10,
 392                 initsize >> 10);
 393
 394         mem_init_done = 1;
 395
 396 #ifdef CONFIG_PPC64
 397         /* Initialize the vDSO */
 398         vdso_init();
 399 #endif
 400 }
 401
 402 /*
 403  * This is called when a page has been modified by the kernel.
 404  * It just marks the page as not i-cache clean.  We do the i-cache
 405  * flush later when the page is given to a user process, if necessary.
 406  */
 407 void flush_dcache_page(struct page *page)
 408 {
 409         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 410                 return;
 411         /* avoid an atomic op if possible */
 412         if (test_bit(PG_arch_1, &page->flags))
 413                 clear_bit(PG_arch_1, &page->flags);
 414 }
 415 EXPORT_SYMBOL(flush_dcache_page);
 416
 417 void flush_dcache_icache_page(struct page *page)
 418 {
 419 #ifdef CONFIG_BOOKE
 420         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 421         __flush_dcache_icache(start);
 422         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 423 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 424         /* On 8xx there is no need to kmap since highmem is not supported */
 425         __flush_dcache_icache(page_address(page));
 426 #else
 427         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 428 #endif
 429
 430 }
 431 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 432 {
 433         clear_page(page);
 434
 435         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 436                 return;
 437         /*
 438          * We shouldnt have to do this, but some versions of glibc
 439          * require it (ld.so assumes zero filled pages are icache clean)
 440          * - Anton
 441          */
 442
 443         /* avoid an atomic op if possible */
 444         if (test_bit(PG_arch_1, &pg->flags))
 445                 clear_bit(PG_arch_1, &pg->flags);
 446 }
 447 EXPORT_SYMBOL(clear_user_page);
 448
 449 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 450                     struct page *pg)
 451 {
 452         copy_page(vto, vfrom);
 453
 454         /*
 455          * We should be able to use the following optimisation, however
 456          * there are two problems.
 457          * Firstly a bug in some versions of binutils meant PLT sections
 458          * were not marked executable.
 459          * Secondly the first word in the GOT section is blrl, used
 460          * to establish the GOT address. Until recently the GOT was
 461          * not marked executable.
 462          * - Anton
 463          */
 464 #if 0
 465         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 466                 return;
 467 #endif
 468
 469         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 470                 return;
 471
 472         /* avoid an atomic op if possible */
 473         if (test_bit(PG_arch_1, &pg->flags))
 474                 clear_bit(PG_arch_1, &pg->flags);
 475 }
 476
 477 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 478                              unsigned long addr, int len)
 479 {
 480         unsigned long maddr;
 481
 482         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 483         flush_icache_range(maddr, maddr + len);
 484         kunmap(page);
 485 }
 486 EXPORT_SYMBOL(flush_icache_user_range);
 487
 488 /*
 489  * This is called at the end of handling a user page fault, when the
 490  * fault has been handled by updating a PTE in the linux page tables.
 491  * We use it to preload an HPTE into the hash table corresponding to
 492  * the updated linux PTE.
 493  *
 494  * This must always be called with the mm->page_table_lock held
 495  */
 496 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 497                       pte_t pte)
 498 {
 499 #ifdef CONFIG_PPC_STD_MMU
 500         unsigned long access = 0, trap;
 501 #endif
 502         unsigned long pfn = pte_pfn(pte);
 503
 504         /* handle i-cache coherency */
 505         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 506             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 507             pfn_valid(pfn)) {
 508                 struct page *page = pfn_to_page(pfn);
 509                 if (!PageReserved(page)
 510                     && !test_bit(PG_arch_1, &page->flags)) {
 511                         if (vma->vm_mm == current->active_mm) {
 512 #ifdef CONFIG_8xx
 513                         /* On 8xx, cache control instructions (particularly
 514                          * "dcbst" from flush_dcache_icache) fault as write
 515                          * operation if there is an unpopulated TLB entry
 516                          * for the address in question. To workaround that,
 517                          * we invalidate the TLB here, thus avoiding dcbst
 518                          * misbehaviour.
 519                          */
 520                                 _tlbie(address);
 521 #endif
 522                                 __flush_dcache_icache((void *) address);
 523                         } else
 524                                 flush_dcache_icache_page(page);
 525                         set_bit(PG_arch_1, &page->flags);
 526                 }
 527         }
 528
 529 #ifdef CONFIG_PPC_STD_MMU
 530         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 531         if (!pte_young(pte) || address >= TASK_SIZE)
 532                 return;
 533
 534         /* We try to figure out if we are coming from an instruction
 535          * access fault and pass that down to __hash_page so we avoid
 536          * double-faulting on execution of fresh text. We have to test
 537          * for regs NULL since init will get here first thing at boot
 538          *
 539          * We also avoid filling the hash if not coming from a fault
 540          */
 541         if (current->thread.regs == NULL)
 542                 return;
 543         trap = TRAP(current->thread.regs);
 544         if (trap == 0x400)
 545                 access |= _PAGE_EXEC;
 546         else if (trap != 0x300)
 547                 return;
 548         hash_preload(vma->vm_mm, address, access, trap);
 549 #endif /* CONFIG_PPC_STD_MMU */
 550 }