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