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/bootinfo.h>
  47 #include <asm/prom.h>
  48 #include <asm/lmb.h>
  49 #include <asm/sections.h>
  50 #ifdef CONFIG_PPC64
  51 #include <asm/vdso.h>
  52 #endif
  53
  54 #include "mmu_decl.h"
  55
  56 #ifndef CPU_FTR_COHERENT_ICACHE
  57 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  58 #define CPU_FTR_NOEXECUTE       0
  59 #endif
  60
  61 int init_bootmem_done;
  62 int mem_init_done;
  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 addr,
  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, addr, size, vma_prot);
  97
  98         if (!page_is_ram(addr >> PAGE_SHIFT))
  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 void show_mem(void)
 106 {
 107         unsigned long total = 0, reserved = 0;
 108         unsigned long shared = 0, cached = 0;
 109         unsigned long highmem = 0;
 110         struct page *page;
 111         pg_data_t *pgdat;
 112         unsigned long i;
 113
 114         printk("Mem-info:\n");
 115         show_free_areas();
 116         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 117         for_each_pgdat(pgdat) {
 118                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 119                         page = pgdat_page_nr(pgdat, i);
 120                         total++;
 121                         if (PageHighMem(page))
 122                                 highmem++;
 123                         if (PageReserved(page))
 124                                 reserved++;
 125                         else if (PageSwapCache(page))
 126                                 cached++;
 127                         else if (page_count(page))
 128                                 shared += page_count(page) - 1;
 129                 }
 130         }
 131         printk("%ld pages of RAM\n", total);
 132 #ifdef CONFIG_HIGHMEM
 133         printk("%ld pages of HIGHMEM\n", highmem);
 134 #endif
 135         printk("%ld reserved pages\n", reserved);
 136         printk("%ld pages shared\n", shared);
 137         printk("%ld pages swap cached\n", cached);
 138 }
 139
 140 /*
 141  * Initialize the bootmem system and give it all the memory we
 142  * have available.  If we are using highmem, we only put the
 143  * lowmem into the bootmem system.
 144  */
 145 #ifndef CONFIG_NEED_MULTIPLE_NODES
 146 void __init do_init_bootmem(void)
 147 {
 148         unsigned long i;
 149         unsigned long start, bootmap_pages;
 150         unsigned long total_pages;
 151         int boot_mapsize;
 152
 153         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 154 #ifdef CONFIG_HIGHMEM
 155         total_pages = total_lowmem >> PAGE_SHIFT;
 156 #endif
 157
 158         /*
 159          * Find an area to use for the bootmem bitmap.  Calculate the size of
 160          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 161          * Add 1 additional page in case the address isn't page-aligned.
 162          */
 163         bootmap_pages = bootmem_bootmap_pages(total_pages);
 164
 165         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 166         BUG_ON(!start);
 167
 168         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 169
 170         /* Add all physical memory to the bootmem map, mark each area
 171          * present.
 172          */
 173         for (i = 0; i < lmb.memory.cnt; i++) {
 174                 unsigned long base = lmb.memory.region[i].base;
 175                 unsigned long size = lmb_size_bytes(&lmb.memory, i);
 176 #ifdef CONFIG_HIGHMEM
 177                 if (base >= total_lowmem)
 178                         continue;
 179                 if (base + size > total_lowmem)
 180                         size = total_lowmem - base;
 181 #endif
 182                 free_bootmem(base, size);
 183         }
 184
 185         /* reserve the sections we're already using */
 186         for (i = 0; i < lmb.reserved.cnt; i++)
 187                 reserve_bootmem(lmb.reserved.region[i].base,
 188                                 lmb_size_bytes(&lmb.reserved, i));
 189
 190         /* XXX need to clip this if using highmem? */
 191         for (i = 0; i < lmb.memory.cnt; i++)
 192                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
 193                                lmb_end_pfn(&lmb.memory, i));
 194         init_bootmem_done = 1;
 195 }
 196
 197 /*
 198  * paging_init() sets up the page tables - in fact we've already done this.
 199  */
 200 void __init paging_init(void)
 201 {
 202         unsigned long zones_size[MAX_NR_ZONES];
 203         unsigned long zholes_size[MAX_NR_ZONES];
 204         unsigned long total_ram = lmb_phys_mem_size();
 205         unsigned long top_of_ram = lmb_end_of_DRAM();
 206
 207 #ifdef CONFIG_HIGHMEM
 208         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 209         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
 210                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 211         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 212         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
 213                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 214         kmap_prot = PAGE_KERNEL;
 215 #endif /* CONFIG_HIGHMEM */
 216
 217         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 218                top_of_ram, total_ram);
 219         printk(KERN_INFO "Memory hole size: %ldMB\n",
 220                (top_of_ram - total_ram) >> 20);
 221         /*
 222          * All pages are DMA-able so we put them all in the DMA zone.
 223          */
 224         memset(zones_size, 0, sizeof(zones_size));
 225         memset(zholes_size, 0, sizeof(zholes_size));
 226
 227         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 228         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 229
 230 #ifdef CONFIG_HIGHMEM
 231         zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 232         zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
 233         zholes_size[ZONE_HIGHMEM] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 234 #else
 235         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 236         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 237 #endif /* CONFIG_HIGHMEM */
 238
 239         free_area_init_node(0, NODE_DATA(0), zones_size,
 240                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
 241 }
 242 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 243
 244 void __init mem_init(void)
 245 {
 246 #ifdef CONFIG_NEED_MULTIPLE_NODES
 247         int nid;
 248 #endif
 249         pg_data_t *pgdat;
 250         unsigned long i;
 251         struct page *page;
 252         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 253
 254         num_physpages = max_pfn;        /* RAM is assumed contiguous */
 255         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 256
 257 #ifdef CONFIG_NEED_MULTIPLE_NODES
 258         for_each_online_node(nid) {
 259                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 260                         printk("freeing bootmem node %x\n", nid);
 261                         totalram_pages +=
 262                                 free_all_bootmem_node(NODE_DATA(nid));
 263                 }
 264         }
 265 #else
 266         max_mapnr = num_physpages;
 267         totalram_pages += free_all_bootmem();
 268 #endif
 269         for_each_pgdat(pgdat) {
 270                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 271                         page = pgdat_page_nr(pgdat, i);
 272                         if (PageReserved(page))
 273                                 reservedpages++;
 274                 }
 275         }
 276
 277         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 278         datasize = (unsigned long)&__init_begin - (unsigned long)&_sdata;
 279         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 280         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 281
 282 #ifdef CONFIG_HIGHMEM
 283         {
 284                 unsigned long pfn, highmem_mapnr;
 285
 286                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 287                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 288                         struct page *page = pfn_to_page(pfn);
 289
 290                         ClearPageReserved(page);
 291                         set_page_count(page, 1);
 292                         __free_page(page);
 293                         totalhigh_pages++;
 294                 }
 295                 totalram_pages += totalhigh_pages;
 296                 printk(KERN_INFO "High memory: %luk\n",
 297                        totalhigh_pages << (PAGE_SHIFT-10));
 298         }
 299 #endif /* CONFIG_HIGHMEM */
 300
 301         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 302                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 303                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 304                 num_physpages << (PAGE_SHIFT-10),
 305                 codesize >> 10,
 306                 reservedpages << (PAGE_SHIFT-10),
 307                 datasize >> 10,
 308                 bsssize >> 10,
 309                 initsize >> 10);
 310
 311         mem_init_done = 1;
 312
 313 #ifdef CONFIG_PPC64
 314         /* Initialize the vDSO */
 315         vdso_init();
 316 #endif
 317 }
 318
 319 /*
 320  * This is called when a page has been modified by the kernel.
 321  * It just marks the page as not i-cache clean.  We do the i-cache
 322  * flush later when the page is given to a user process, if necessary.
 323  */
 324 void flush_dcache_page(struct page *page)
 325 {
 326         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 327                 return;
 328         /* avoid an atomic op if possible */
 329         if (test_bit(PG_arch_1, &page->flags))
 330                 clear_bit(PG_arch_1, &page->flags);
 331 }
 332 EXPORT_SYMBOL(flush_dcache_page);
 333
 334 void flush_dcache_icache_page(struct page *page)
 335 {
 336 #ifdef CONFIG_BOOKE
 337         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 338         __flush_dcache_icache(start);
 339         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 340 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 341         /* On 8xx there is no need to kmap since highmem is not supported */
 342         __flush_dcache_icache(page_address(page));
 343 #else
 344         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 345 #endif
 346
 347 }
 348 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 349 {
 350         clear_page(page);
 351
 352         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 353                 return;
 354         /*
 355          * We shouldnt have to do this, but some versions of glibc
 356          * require it (ld.so assumes zero filled pages are icache clean)
 357          * - Anton
 358          */
 359
 360         /* avoid an atomic op if possible */
 361         if (test_bit(PG_arch_1, &pg->flags))
 362                 clear_bit(PG_arch_1, &pg->flags);
 363 }
 364 EXPORT_SYMBOL(clear_user_page);
 365
 366 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 367                     struct page *pg)
 368 {
 369         copy_page(vto, vfrom);
 370
 371         /*
 372          * We should be able to use the following optimisation, however
 373          * there are two problems.
 374          * Firstly a bug in some versions of binutils meant PLT sections
 375          * were not marked executable.
 376          * Secondly the first word in the GOT section is blrl, used
 377          * to establish the GOT address. Until recently the GOT was
 378          * not marked executable.
 379          * - Anton
 380          */
 381 #if 0
 382         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 383                 return;
 384 #endif
 385
 386         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 387                 return;
 388
 389         /* avoid an atomic op if possible */
 390         if (test_bit(PG_arch_1, &pg->flags))
 391                 clear_bit(PG_arch_1, &pg->flags);
 392 }
 393
 394 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 395                              unsigned long addr, int len)
 396 {
 397         unsigned long maddr;
 398
 399         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 400         flush_icache_range(maddr, maddr + len);
 401         kunmap(page);
 402 }
 403 EXPORT_SYMBOL(flush_icache_user_range);
 404
 405 /*
 406  * This is called at the end of handling a user page fault, when the
 407  * fault has been handled by updating a PTE in the linux page tables.
 408  * We use it to preload an HPTE into the hash table corresponding to
 409  * the updated linux PTE.
 410  *
 411  * This must always be called with the mm->page_table_lock held
 412  */
 413 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 414                       pte_t pte)
 415 {
 416         /* handle i-cache coherency */
 417         unsigned long pfn = pte_pfn(pte);
 418 #ifdef CONFIG_PPC32
 419         pmd_t *pmd;
 420 #else
 421         unsigned long vsid;
 422         void *pgdir;
 423         pte_t *ptep;
 424         int local = 0;
 425         cpumask_t tmp;
 426         unsigned long flags;
 427 #endif
 428
 429         /* handle i-cache coherency */
 430         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 431             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 432             pfn_valid(pfn)) {
 433                 struct page *page = pfn_to_page(pfn);
 434                 if (!PageReserved(page)
 435                     && !test_bit(PG_arch_1, &page->flags)) {
 436                         if (vma->vm_mm == current->active_mm) {
 437 #ifdef CONFIG_8xx
 438                         /* On 8xx, cache control instructions (particularly
 439                          * "dcbst" from flush_dcache_icache) fault as write
 440                          * operation if there is an unpopulated TLB entry
 441                          * for the address in question. To workaround that,
 442                          * we invalidate the TLB here, thus avoiding dcbst
 443                          * misbehaviour.
 444                          */
 445                                 _tlbie(address);
 446 #endif
 447                                 __flush_dcache_icache((void *) address);
 448                         } else
 449                                 flush_dcache_icache_page(page);
 450                         set_bit(PG_arch_1, &page->flags);
 451                 }
 452         }
 453
 454 #ifdef CONFIG_PPC_STD_MMU
 455         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 456         if (!pte_young(pte) || address >= TASK_SIZE)
 457                 return;
 458 #ifdef CONFIG_PPC32
 459         if (Hash == 0)
 460                 return;
 461         pmd = pmd_offset(pgd_offset(vma->vm_mm, address), address);
 462         if (!pmd_none(*pmd))
 463                 add_hash_page(vma->vm_mm->context, address, pmd_val(*pmd));
 464 #else
 465         pgdir = vma->vm_mm->pgd;
 466         if (pgdir == NULL)
 467                 return;
 468
 469         ptep = find_linux_pte(pgdir, address);
 470         if (!ptep)
 471                 return;
 472
 473         vsid = get_vsid(vma->vm_mm->context.id, address);
 474
 475         local_irq_save(flags);
 476         tmp = cpumask_of_cpu(smp_processor_id());
 477         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
 478                 local = 1;
 479
 480         __hash_page(address, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
 481                     0x300, local);
 482         local_irq_restore(flags);
 483 #endif
 484 #endif
 485 }