arch/ppc/mm/init.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/bootinfo.h>
  46
  47 #include "mem_pieces.h"
  48 #include "mmu_decl.h"
  49
  50 #if defined(CONFIG_KERNEL_START_BOOL) || defined(CONFIG_LOWMEM_SIZE_BOOL)
  51 /* The amount of lowmem must be within 0xF0000000 - KERNELBASE. */
  52 #if (CONFIG_LOWMEM_SIZE > (0xF0000000 - KERNELBASE))
  53 #error "You must adjust CONFIG_LOWMEM_SIZE or CONFIG_START_KERNEL"
  54 #endif
  55 #endif
  56 #define MAX_LOW_MEM     CONFIG_LOWMEM_SIZE
  57
  58 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  59
  60 unsigned long total_memory;
  61 unsigned long total_lowmem;
  62
  63 unsigned long ppc_memstart;
  64 unsigned long ppc_memoffset = PAGE_OFFSET;
  65
  66 int mem_init_done;
  67 int init_bootmem_done;
  68 int boot_mapsize;
  69
  70 extern char _end[];
  71 extern char etext[], _stext[];
  72 extern char __init_begin, __init_end;
  73
  74 #ifdef CONFIG_HIGHMEM
  75 pte_t *kmap_pte;
  76 pgprot_t kmap_prot;
  77
  78 EXPORT_SYMBOL(kmap_prot);
  79 EXPORT_SYMBOL(kmap_pte);
  80 #endif
  81
  82 void MMU_init(void);
  83 void set_phys_avail(unsigned long total_ram);
  84
  85 /* XXX should be in current.h  -- paulus */
  86 extern struct task_struct *current_set[NR_CPUS];
  87
  88 char *klimit = _end;
  89 struct mem_pieces phys_avail;
  90
  91 /*
  92  * this tells the system to map all of ram with the segregs
  93  * (i.e. page tables) instead of the bats.
  94  * -- Cort
  95  */
  96 int __map_without_bats;
  97 int __map_without_ltlbs;
  98
  99 /* max amount of RAM to use */
 100 unsigned long __max_memory;
 101 /* max amount of low RAM to map in */
 102 unsigned long __max_low_memory = MAX_LOW_MEM;
 103
 104 void show_mem(void)
 105 {
 106         int i,free = 0,total = 0,reserved = 0;
 107         int shared = 0, cached = 0;
 108         int highmem = 0;
 109
 110         printk("Mem-info:\n");
 111         show_free_areas();
 112         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 113         i = max_mapnr;
 114         while (i-- > 0) {
 115                 total++;
 116                 if (PageHighMem(mem_map+i))
 117                         highmem++;
 118                 if (PageReserved(mem_map+i))
 119                         reserved++;
 120                 else if (PageSwapCache(mem_map+i))
 121                         cached++;
 122                 else if (!page_count(mem_map+i))
 123                         free++;
 124                 else
 125                         shared += page_count(mem_map+i) - 1;
 126         }
 127         printk("%d pages of RAM\n",total);
 128         printk("%d pages of HIGHMEM\n", highmem);
 129         printk("%d free pages\n",free);
 130         printk("%d reserved pages\n",reserved);
 131         printk("%d pages shared\n",shared);
 132         printk("%d pages swap cached\n",cached);
 133 }
 134
 135 /* Free up now-unused memory */
 136 static void free_sec(unsigned long start, unsigned long end, const char *name)
 137 {
 138         unsigned long cnt = 0;
 139
 140         while (start < end) {
 141                 ClearPageReserved(virt_to_page(start));
 142                 init_page_count(virt_to_page(start));
 143                 free_page(start);
 144                 cnt++;
 145                 start += PAGE_SIZE;
 146         }
 147         if (cnt) {
 148                 printk(" %ldk %s", cnt << (PAGE_SHIFT - 10), name);
 149                 totalram_pages += cnt;
 150         }
 151 }
 152
 153 void free_initmem(void)
 154 {
 155 #define FREESEC(TYPE) \
 156         free_sec((unsigned long)(&__ ## TYPE ## _begin), \
 157                  (unsigned long)(&__ ## TYPE ## _end), \
 158                  #TYPE);
 159
 160         printk ("Freeing unused kernel memory:");
 161         FREESEC(init);
 162         printk("\n");
 163         ppc_md.progress = NULL;
 164 #undef FREESEC
 165 }
 166
 167 #ifdef CONFIG_BLK_DEV_INITRD
 168 void free_initrd_mem(unsigned long start, unsigned long end)
 169 {
 170         printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 171
 172         for (; start < end; start += PAGE_SIZE) {
 173                 ClearPageReserved(virt_to_page(start));
 174                 init_page_count(virt_to_page(start));
 175                 free_page(start);
 176                 totalram_pages++;
 177         }
 178 }
 179 #endif
 180
 181 /*
 182  * Check for command-line options that affect what MMU_init will do.
 183  */
 184 void MMU_setup(void)
 185 {
 186         /* Check for nobats option (used in mapin_ram). */
 187         if (strstr(cmd_line, "nobats")) {
 188                 __map_without_bats = 1;
 189         }
 190
 191         if (strstr(cmd_line, "noltlbs")) {
 192                 __map_without_ltlbs = 1;
 193         }
 194
 195         /* Look for mem= option on command line */
 196         if (strstr(cmd_line, "mem=")) {
 197                 char *p, *q;
 198                 unsigned long maxmem = 0;
 199
 200                 for (q = cmd_line; (p = strstr(q, "mem=")) != 0; ) {
 201                         q = p + 4;
 202                         if (p > cmd_line && p[-1] != ' ')
 203                                 continue;
 204                         maxmem = simple_strtoul(q, &q, 0);
 205                         if (*q == 'k' || *q == 'K') {
 206                                 maxmem <<= 10;
 207                                 ++q;
 208                         } else if (*q == 'm' || *q == 'M') {
 209                                 maxmem <<= 20;
 210                                 ++q;
 211                         }
 212                 }
 213                 __max_memory = maxmem;
 214         }
 215 }
 216
 217 /*
 218  * MMU_init sets up the basic memory mappings for the kernel,
 219  * including both RAM and possibly some I/O regions,
 220  * and sets up the page tables and the MMU hardware ready to go.
 221  */
 222 void __init MMU_init(void)
 223 {
 224         if (ppc_md.progress)
 225                 ppc_md.progress("MMU:enter", 0x111);
 226
 227         /* parse args from command line */
 228         MMU_setup();
 229
 230         /*
 231          * Figure out how much memory we have, how much
 232          * is lowmem, and how much is highmem.  If we were
 233          * passed the total memory size from the bootloader,
 234          * just use it.
 235          */
 236         if (boot_mem_size)
 237                 total_memory = boot_mem_size;
 238         else
 239                 total_memory = ppc_md.find_end_of_memory();
 240
 241         if (__max_memory && total_memory > __max_memory)
 242                 total_memory = __max_memory;
 243         total_lowmem = total_memory;
 244         if (total_lowmem > __max_low_memory) {
 245                 total_lowmem = __max_low_memory;
 246 #ifndef CONFIG_HIGHMEM
 247                 total_memory = total_lowmem;
 248 #endif /* CONFIG_HIGHMEM */
 249         }
 250         set_phys_avail(total_lowmem);
 251
 252         /* Initialize the MMU hardware */
 253         if (ppc_md.progress)
 254                 ppc_md.progress("MMU:hw init", 0x300);
 255         MMU_init_hw();
 256
 257         /* Map in all of RAM starting at KERNELBASE */
 258         if (ppc_md.progress)
 259                 ppc_md.progress("MMU:mapin", 0x301);
 260         mapin_ram();
 261
 262 #ifdef CONFIG_HIGHMEM
 263         ioremap_base = PKMAP_BASE;
 264 #else
 265         ioremap_base = 0xfe000000UL;    /* for now, could be 0xfffff000 */
 266 #endif /* CONFIG_HIGHMEM */
 267         ioremap_bot = ioremap_base;
 268
 269         /* Map in I/O resources */
 270         if (ppc_md.progress)
 271                 ppc_md.progress("MMU:setio", 0x302);
 272         if (ppc_md.setup_io_mappings)
 273                 ppc_md.setup_io_mappings();
 274
 275         /* Initialize the context management stuff */
 276         mmu_context_init();
 277
 278         if (ppc_md.progress)
 279                 ppc_md.progress("MMU:exit", 0x211);
 280
 281 #ifdef CONFIG_BOOTX_TEXT
 282         /* By default, we are no longer mapped */
 283         boot_text_mapped = 0;
 284         /* Must be done last, or ppc_md.progress will die. */
 285         map_boot_text();
 286 #endif
 287 }
 288
 289 /* This is only called until mem_init is done. */
 290 void __init *early_get_page(void)
 291 {
 292         void *p;
 293
 294         if (init_bootmem_done) {
 295                 p = alloc_bootmem_pages(PAGE_SIZE);
 296         } else {
 297                 p = mem_pieces_find(PAGE_SIZE, PAGE_SIZE);
 298         }
 299         return p;
 300 }
 301
 302 /*
 303  * Initialize the bootmem system and give it all the memory we
 304  * have available.
 305  */
 306 void __init do_init_bootmem(void)
 307 {
 308         unsigned long start, size;
 309         int i;
 310
 311         /*
 312          * Find an area to use for the bootmem bitmap.
 313          * We look for the first area which is at least
 314          * 128kB in length (128kB is enough for a bitmap
 315          * for 4GB of memory, using 4kB pages), plus 1 page
 316          * (in case the address isn't page-aligned).
 317          */
 318         start = 0;
 319         size = 0;
 320         for (i = 0; i < phys_avail.n_regions; ++i) {
 321                 unsigned long a = phys_avail.regions[i].address;
 322                 unsigned long s = phys_avail.regions[i].size;
 323                 if (s <= size)
 324                         continue;
 325                 start = a;
 326                 size = s;
 327                 if (s >= 33 * PAGE_SIZE)
 328                         break;
 329         }
 330         start = PAGE_ALIGN(start);
 331
 332         min_low_pfn = start >> PAGE_SHIFT;
 333         max_low_pfn = (PPC_MEMSTART + total_lowmem) >> PAGE_SHIFT;
 334         max_pfn = (PPC_MEMSTART + total_memory) >> PAGE_SHIFT;
 335         boot_mapsize = init_bootmem_node(&contig_page_data, min_low_pfn,
 336                                          PPC_MEMSTART >> PAGE_SHIFT,
 337                                          max_low_pfn);
 338
 339         /* remove the bootmem bitmap from the available memory */
 340         mem_pieces_remove(&phys_avail, start, boot_mapsize, 1);
 341
 342         /* add everything in phys_avail into the bootmem map */
 343         for (i = 0; i < phys_avail.n_regions; ++i)
 344                 free_bootmem(phys_avail.regions[i].address,
 345                              phys_avail.regions[i].size);
 346
 347         init_bootmem_done = 1;
 348 }
 349
 350 /*
 351  * paging_init() sets up the page tables - in fact we've already done this.
 352  */
 353 void __init paging_init(void)
 354 {
 355         unsigned long start_pfn, end_pfn;
 356         unsigned long max_zone_pfns[MAX_NR_ZONES];
 357 #ifdef CONFIG_HIGHMEM
 358         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 359         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
 360                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 361         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 362         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
 363                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 364         kmap_prot = PAGE_KERNEL;
 365 #endif /* CONFIG_HIGHMEM */
 366         /* All pages are DMA-able so we put them all in the DMA zone. */
 367         start_pfn = __pa(PAGE_OFFSET) >> PAGE_SHIFT;
 368         end_pfn = start_pfn + (total_memory >> PAGE_SHIFT);
 369         add_active_range(0, start_pfn, end_pfn);
 370
 371         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 372 #ifdef CONFIG_HIGHMEM
 373         max_zone_pfns[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 374         max_zone_pfns[ZONE_HIGHMEM] = total_memory >> PAGE_SHIFT;
 375 #else
 376         max_zone_pfns[ZONE_DMA] = total_memory >> PAGE_SHIFT;
 377 #endif /* CONFIG_HIGHMEM */
 378         free_area_init_nodes(max_zone_pfns);
 379 }
 380
 381 void __init mem_init(void)
 382 {
 383         unsigned long addr;
 384         int codepages = 0;
 385         int datapages = 0;
 386         int initpages = 0;
 387 #ifdef CONFIG_HIGHMEM
 388         unsigned long highmem_mapnr;
 389
 390         highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 391 #endif /* CONFIG_HIGHMEM */
 392         max_mapnr = total_memory >> PAGE_SHIFT;
 393
 394         high_memory = (void *) __va(PPC_MEMSTART + total_lowmem);
 395         num_physpages = max_mapnr;      /* RAM is assumed contiguous */
 396
 397         totalram_pages += free_all_bootmem();
 398
 399 #ifdef CONFIG_BLK_DEV_INITRD
 400         /* if we are booted from BootX with an initial ramdisk,
 401            make sure the ramdisk pages aren't reserved. */
 402         if (initrd_start) {
 403                 for (addr = initrd_start; addr < initrd_end; addr += PAGE_SIZE)
 404                         ClearPageReserved(virt_to_page(addr));
 405         }
 406 #endif /* CONFIG_BLK_DEV_INITRD */
 407
 408         for (addr = PAGE_OFFSET; addr < (unsigned long)high_memory;
 409              addr += PAGE_SIZE) {
 410                 if (!PageReserved(virt_to_page(addr)))
 411                         continue;
 412                 if (addr < (ulong) etext)
 413                         codepages++;
 414                 else if (addr >= (unsigned long)&__init_begin
 415                          && addr < (unsigned long)&__init_end)
 416                         initpages++;
 417                 else if (addr < (ulong) klimit)
 418                         datapages++;
 419         }
 420
 421 #ifdef CONFIG_HIGHMEM
 422         {
 423                 unsigned long pfn;
 424
 425                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 426                         struct page *page = mem_map + pfn;
 427
 428                         ClearPageReserved(page);
 429                         init_page_count(page);
 430                         __free_page(page);
 431                         totalhigh_pages++;
 432                 }
 433                 totalram_pages += totalhigh_pages;
 434         }
 435 #endif /* CONFIG_HIGHMEM */
 436
 437         printk("Memory: %luk available (%dk kernel code, %dk data, %dk init, %ldk highmem)\n",
 438                (unsigned long)nr_free_pages()<< (PAGE_SHIFT-10),
 439                codepages<< (PAGE_SHIFT-10), datapages<< (PAGE_SHIFT-10),
 440                initpages<< (PAGE_SHIFT-10),
 441                (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
 442
 443         mem_init_done = 1;
 444 }
 445
 446 /*
 447  * Set phys_avail to the amount of physical memory,
 448  * less the kernel text/data/bss.
 449  */
 450 void __init
 451 set_phys_avail(unsigned long total_memory)
 452 {
 453         unsigned long kstart, ksize;
 454
 455         /*
 456          * Initially, available physical memory is equivalent to all
 457          * physical memory.
 458          */
 459
 460         phys_avail.regions[0].address = PPC_MEMSTART;
 461         phys_avail.regions[0].size = total_memory;
 462         phys_avail.n_regions = 1;
 463
 464         /*
 465          * Map out the kernel text/data/bss from the available physical
 466          * memory.
 467          */
 468
 469         kstart = __pa(_stext);  /* should be 0 */
 470         ksize = PAGE_ALIGN(klimit - _stext);
 471
 472         mem_pieces_remove(&phys_avail, kstart, ksize, 0);
 473         mem_pieces_remove(&phys_avail, 0, 0x4000, 0);
 474
 475 #if defined(CONFIG_BLK_DEV_INITRD)
 476         /* Remove the init RAM disk from the available memory. */
 477         if (initrd_start) {
 478                 mem_pieces_remove(&phys_avail, __pa(initrd_start),
 479                                   initrd_end - initrd_start, 1);
 480         }
 481 #endif /* CONFIG_BLK_DEV_INITRD */
 482 }
 483
 484 /* Mark some memory as reserved by removing it from phys_avail. */
 485 void __init reserve_phys_mem(unsigned long start, unsigned long size)
 486 {
 487         mem_pieces_remove(&phys_avail, start, size, 1);
 488 }
 489
 490 /*
 491  * This is called when a page has been modified by the kernel.
 492  * It just marks the page as not i-cache clean.  We do the i-cache
 493  * flush later when the page is given to a user process, if necessary.
 494  */
 495 void flush_dcache_page(struct page *page)
 496 {
 497         clear_bit(PG_arch_1, &page->flags);
 498 }
 499
 500 void flush_dcache_icache_page(struct page *page)
 501 {
 502 #ifdef CONFIG_BOOKE
 503         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 504         __flush_dcache_icache(start);
 505         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 506 #elif defined(CONFIG_8xx)
 507         /* On 8xx there is no need to kmap since highmem is not supported */
 508         __flush_dcache_icache(page_address(page));
 509 #else
 510         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 511 #endif
 512
 513 }
 514 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 515 {
 516         clear_page(page);
 517         clear_bit(PG_arch_1, &pg->flags);
 518 }
 519
 520 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 521                     struct page *pg)
 522 {
 523         copy_page(vto, vfrom);
 524         clear_bit(PG_arch_1, &pg->flags);
 525 }
 526
 527 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 528                              unsigned long addr, int len)
 529 {
 530         unsigned long maddr;
 531
 532         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 533         flush_icache_range(maddr, maddr + len);
 534         kunmap(page);
 535 }
 536
 537 /*
 538  * This is called at the end of handling a user page fault, when the
 539  * fault has been handled by updating a PTE in the linux page tables.
 540  * We use it to preload an HPTE into the hash table corresponding to
 541  * the updated linux PTE.
 542  */
 543 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 544                       pte_t pte)
 545 {
 546         /* handle i-cache coherency */
 547         unsigned long pfn = pte_pfn(pte);
 548
 549         if (pfn_valid(pfn)) {
 550                 struct page *page = pfn_to_page(pfn);
 551 #ifdef CONFIG_8xx
 552                 /* On 8xx, the TLB handlers work in 2 stages:
 553                  * First, a zeroed entry is loaded by TLBMiss handler,
 554                  * which causes the TLBError handler to be triggered.
 555                  * That means the zeroed TLB has to be invalidated
 556                  * whenever a page miss occurs.
 557                  */
 558                 _tlbie(address, 0 /* 8xx doesn't care about PID */);
 559 #endif
 560                 if (!PageReserved(page)
 561                     && !test_bit(PG_arch_1, &page->flags)) {
 562                         if (vma->vm_mm == current->active_mm)
 563                                 __flush_dcache_icache((void *) address);
 564                         else
 565                                 flush_dcache_icache_page(page);
 566                         set_bit(PG_arch_1, &page->flags);
 567                 }
 568         }
 569
 570 #ifdef CONFIG_PPC_STD_MMU
 571         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 572         if (Hash != 0 && pte_young(pte)) {
 573                 struct mm_struct *mm;
 574                 pmd_t *pmd;
 575
 576                 mm = (address < TASK_SIZE)? vma->vm_mm: &init_mm;
 577                 pmd = pmd_offset(pgd_offset(mm, address), address);
 578                 if (!pmd_none(*pmd))
 579                         add_hash_page(mm->context.id, address, pmd_val(*pmd));
 580         }
 581 #endif
 582 }
 583
 584 /*
 585  * This is called by /dev/mem to know if a given address has to
 586  * be mapped non-cacheable or not
 587  */
 588 int page_is_ram(unsigned long pfn)
 589 {
 590         return pfn < max_pfn;
 591 }
 592
 593 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 594                               unsigned long size, pgprot_t vma_prot)
 595 {
 596         if (ppc_md.phys_mem_access_prot)
 597                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 598
 599         if (!page_is_ram(pfn))
 600                 vma_prot = __pgprot(pgprot_val(vma_prot)
 601                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 602         return vma_prot;
 603 }
 604 EXPORT_SYMBOL(phys_mem_access_prot);