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
  51 #include "mmu_decl.h"
  52
  53 #ifndef CPU_FTR_COHERENT_ICACHE
  54 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  55 #define CPU_FTR_NOEXECUTE       0
  56 #endif
  57
  58 int init_bootmem_done;
  59 int mem_init_done;
  60
  61 /*
  62  * This is called by /dev/mem to know if a given address has to
  63  * be mapped non-cacheable or not
  64  */
  65 int page_is_ram(unsigned long pfn)
  66 {
  67         unsigned long paddr = (pfn << PAGE_SHIFT);
  68
  69 #ifndef CONFIG_PPC64    /* XXX for now */
  70         return paddr < __pa(high_memory);
  71 #else
  72         int i;
  73         for (i=0; i < lmb.memory.cnt; i++) {
  74                 unsigned long base;
  75
  76                 base = lmb.memory.region[i].base;
  77
  78                 if ((paddr >= base) &&
  79                         (paddr < (base + lmb.memory.region[i].size))) {
  80                         return 1;
  81                 }
  82         }
  83
  84         return 0;
  85 #endif
  86 }
  87 EXPORT_SYMBOL(page_is_ram);
  88
  89 pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
  90                               unsigned long size, pgprot_t vma_prot)
  91 {
  92         if (ppc_md.phys_mem_access_prot)
  93                 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
  94
  95         if (!page_is_ram(addr >> PAGE_SHIFT))
  96                 vma_prot = __pgprot(pgprot_val(vma_prot)
  97                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
  98         return vma_prot;
  99 }
 100 EXPORT_SYMBOL(phys_mem_access_prot);
 101
 102 void show_mem(void)
 103 {
 104         unsigned long total = 0, reserved = 0;
 105         unsigned long shared = 0, cached = 0;
 106         unsigned long highmem = 0;
 107         struct page *page;
 108         pg_data_t *pgdat;
 109         unsigned long i;
 110
 111         printk("Mem-info:\n");
 112         show_free_areas();
 113         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 114         for_each_pgdat(pgdat) {
 115                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 116                         page = pgdat_page_nr(pgdat, i);
 117                         total++;
 118                         if (PageHighMem(page))
 119                                 highmem++;
 120                         if (PageReserved(page))
 121                                 reserved++;
 122                         else if (PageSwapCache(page))
 123                                 cached++;
 124                         else if (page_count(page))
 125                                 shared += page_count(page) - 1;
 126                 }
 127         }
 128         printk("%ld pages of RAM\n", total);
 129 #ifdef CONFIG_HIGHMEM
 130         printk("%ld pages of HIGHMEM\n", highmem);
 131 #endif
 132         printk("%ld reserved pages\n", reserved);
 133         printk("%ld pages shared\n", shared);
 134         printk("%ld pages swap cached\n", cached);
 135 }
 136
 137 /*
 138  * Initialize the bootmem system and give it all the memory we
 139  * have available.  If we are using highmem, we only put the
 140  * lowmem into the bootmem system.
 141  */
 142 #ifndef CONFIG_NEED_MULTIPLE_NODES
 143 void __init do_init_bootmem(void)
 144 {
 145         unsigned long i;
 146         unsigned long start, bootmap_pages;
 147         unsigned long total_pages;
 148         int boot_mapsize;
 149
 150         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 151 #ifdef CONFIG_HIGHMEM
 152         total_pages = total_lowmem >> PAGE_SHIFT;
 153 #endif
 154
 155         /*
 156          * Find an area to use for the bootmem bitmap.  Calculate the size of
 157          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 158          * Add 1 additional page in case the address isn't page-aligned.
 159          */
 160         bootmap_pages = bootmem_bootmap_pages(total_pages);
 161
 162         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 163         BUG_ON(!start);
 164
 165         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 166
 167         /* Add all physical memory to the bootmem map, mark each area
 168          * present.
 169          */
 170         for (i = 0; i < lmb.memory.cnt; i++) {
 171                 unsigned long base = lmb.memory.region[i].base;
 172                 unsigned long size = lmb_size_bytes(&lmb.memory, i);
 173 #ifdef CONFIG_HIGHMEM
 174                 if (base >= total_lowmem)
 175                         continue;
 176                 if (base + size > total_lowmem)
 177                         size = total_lowmem - base;
 178 #endif
 179                 free_bootmem(base, size);
 180         }
 181
 182         /* reserve the sections we're already using */
 183         for (i = 0; i < lmb.reserved.cnt; i++)
 184                 reserve_bootmem(lmb.reserved.region[i].base,
 185                                 lmb_size_bytes(&lmb.reserved, i));
 186
 187         /* XXX need to clip this if using highmem? */
 188         for (i = 0; i < lmb.memory.cnt; i++)
 189                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
 190                                lmb_end_pfn(&lmb.memory, i));
 191         init_bootmem_done = 1;
 192 }
 193
 194 /*
 195  * paging_init() sets up the page tables - in fact we've already done this.
 196  */
 197 void __init paging_init(void)
 198 {
 199         unsigned long zones_size[MAX_NR_ZONES];
 200         unsigned long zholes_size[MAX_NR_ZONES];
 201         unsigned long total_ram = lmb_phys_mem_size();
 202         unsigned long top_of_ram = lmb_end_of_DRAM();
 203
 204 #ifdef CONFIG_HIGHMEM
 205         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 206         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
 207                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 208         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 209         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
 210                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 211         kmap_prot = PAGE_KERNEL;
 212 #endif /* CONFIG_HIGHMEM */
 213
 214         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 215                top_of_ram, total_ram);
 216         printk(KERN_INFO "Memory hole size: %ldMB\n",
 217                (top_of_ram - total_ram) >> 20);
 218         /*
 219          * All pages are DMA-able so we put them all in the DMA zone.
 220          */
 221         memset(zones_size, 0, sizeof(zones_size));
 222         memset(zholes_size, 0, sizeof(zholes_size));
 223
 224         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 225         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 226
 227 #ifdef CONFIG_HIGHMEM
 228         zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
 229         zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
 230         zholes_size[ZONE_HIGHMEM] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 231 #else
 232         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 233         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 234 #endif /* CONFIG_HIGHMEM */
 235
 236         free_area_init_node(0, NODE_DATA(0), zones_size,
 237                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
 238 }
 239 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 240
 241 void __init mem_init(void)
 242 {
 243 #ifdef CONFIG_NEED_MULTIPLE_NODES
 244         int nid;
 245 #endif
 246         pg_data_t *pgdat;
 247         unsigned long i;
 248         struct page *page;
 249         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 250
 251         num_physpages = max_pfn;        /* RAM is assumed contiguous */
 252         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 253
 254 #ifdef CONFIG_NEED_MULTIPLE_NODES
 255         for_each_online_node(nid) {
 256                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 257                         printk("freeing bootmem node %x\n", nid);
 258                         totalram_pages +=
 259                                 free_all_bootmem_node(NODE_DATA(nid));
 260                 }
 261         }
 262 #else
 263         max_mapnr = num_physpages;
 264         totalram_pages += free_all_bootmem();
 265 #endif
 266         for_each_pgdat(pgdat) {
 267                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 268                         page = pgdat_page_nr(pgdat, i);
 269                         if (PageReserved(page))
 270                                 reservedpages++;
 271                 }
 272         }
 273
 274         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 275         datasize = (unsigned long)&__init_begin - (unsigned long)&_sdata;
 276         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 277         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 278
 279 #ifdef CONFIG_HIGHMEM
 280         {
 281                 unsigned long pfn, highmem_mapnr;
 282
 283                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
 284                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 285                         struct page *page = pfn_to_page(pfn);
 286
 287                         ClearPageReserved(page);
 288                         set_page_count(page, 1);
 289                         __free_page(page);
 290                         totalhigh_pages++;
 291                 }
 292                 totalram_pages += totalhigh_pages;
 293                 printk(KERN_INFO "High memory: %luk\n",
 294                        totalhigh_pages << (PAGE_SHIFT-10));
 295         }
 296 #endif /* CONFIG_HIGHMEM */
 297
 298         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 299                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 300                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 301                 num_physpages << (PAGE_SHIFT-10),
 302                 codesize >> 10,
 303                 reservedpages << (PAGE_SHIFT-10),
 304                 datasize >> 10,
 305                 bsssize >> 10,
 306                 initsize >> 10);
 307
 308         mem_init_done = 1;
 309
 310 #ifdef CONFIG_PPC64
 311         /* Initialize the vDSO */
 312         vdso_init();
 313 #endif
 314 }
 315
 316 /*
 317  * This is called when a page has been modified by the kernel.
 318  * It just marks the page as not i-cache clean.  We do the i-cache
 319  * flush later when the page is given to a user process, if necessary.
 320  */
 321 void flush_dcache_page(struct page *page)
 322 {
 323         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 324                 return;
 325         /* avoid an atomic op if possible */
 326         if (test_bit(PG_arch_1, &page->flags))
 327                 clear_bit(PG_arch_1, &page->flags);
 328 }
 329 EXPORT_SYMBOL(flush_dcache_page);
 330
 331 void flush_dcache_icache_page(struct page *page)
 332 {
 333 #ifdef CONFIG_BOOKE
 334         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 335         __flush_dcache_icache(start);
 336         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 337 #elif defined(CONFIG_8xx)
 338         /* On 8xx there is no need to kmap since highmem is not supported */
 339         __flush_dcache_icache(page_address(page));
 340 #else
 341         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 342 #endif
 343
 344 }
 345 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 346 {
 347         clear_page(page);
 348
 349         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 350                 return;
 351         /*
 352          * We shouldnt have to do this, but some versions of glibc
 353          * require it (ld.so assumes zero filled pages are icache clean)
 354          * - Anton
 355          */
 356
 357         /* avoid an atomic op if possible */
 358         if (test_bit(PG_arch_1, &pg->flags))
 359                 clear_bit(PG_arch_1, &pg->flags);
 360 }
 361 EXPORT_SYMBOL(clear_user_page);
 362
 363 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 364                     struct page *pg)
 365 {
 366         copy_page(vto, vfrom);
 367
 368         /*
 369          * We should be able to use the following optimisation, however
 370          * there are two problems.
 371          * Firstly a bug in some versions of binutils meant PLT sections
 372          * were not marked executable.
 373          * Secondly the first word in the GOT section is blrl, used
 374          * to establish the GOT address. Until recently the GOT was
 375          * not marked executable.
 376          * - Anton
 377          */
 378 #if 0
 379         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 380                 return;
 381 #endif
 382
 383         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 384                 return;
 385
 386         /* avoid an atomic op if possible */
 387         if (test_bit(PG_arch_1, &pg->flags))
 388                 clear_bit(PG_arch_1, &pg->flags);
 389 }
 390
 391 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 392                              unsigned long addr, int len)
 393 {
 394         unsigned long maddr;
 395
 396         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 397         flush_icache_range(maddr, maddr + len);
 398         kunmap(page);
 399 }
 400 EXPORT_SYMBOL(flush_icache_user_range);
 401
 402 /*
 403  * This is called at the end of handling a user page fault, when the
 404  * fault has been handled by updating a PTE in the linux page tables.
 405  * We use it to preload an HPTE into the hash table corresponding to
 406  * the updated linux PTE.
 407  *
 408  * This must always be called with the mm->page_table_lock held
 409  */
 410 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 411                       pte_t pte)
 412 {
 413         /* handle i-cache coherency */
 414         unsigned long pfn = pte_pfn(pte);
 415 #ifdef CONFIG_PPC32
 416         pmd_t *pmd;
 417 #else
 418         unsigned long vsid;
 419         void *pgdir;
 420         pte_t *ptep;
 421         int local = 0;
 422         cpumask_t tmp;
 423         unsigned long flags;
 424 #endif
 425
 426         /* handle i-cache coherency */
 427         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 428             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 429             pfn_valid(pfn)) {
 430                 struct page *page = pfn_to_page(pfn);
 431                 if (!PageReserved(page)
 432                     && !test_bit(PG_arch_1, &page->flags)) {
 433                         if (vma->vm_mm == current->active_mm) {
 434 #ifdef CONFIG_8xx
 435                         /* On 8xx, cache control instructions (particularly
 436                          * "dcbst" from flush_dcache_icache) fault as write
 437                          * operation if there is an unpopulated TLB entry
 438                          * for the address in question. To workaround that,
 439                          * we invalidate the TLB here, thus avoiding dcbst
 440                          * misbehaviour.
 441                          */
 442                                 _tlbie(address);
 443 #endif
 444                                 __flush_dcache_icache((void *) address);
 445                         } else
 446                                 flush_dcache_icache_page(page);
 447                         set_bit(PG_arch_1, &page->flags);
 448                 }
 449         }
 450
 451 #ifdef CONFIG_PPC_STD_MMU
 452         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 453         if (!pte_young(pte) || address >= TASK_SIZE)
 454                 return;
 455 #ifdef CONFIG_PPC32
 456         if (Hash == 0)
 457                 return;
 458         pmd = pmd_offset(pgd_offset(vma->vm_mm, address), address);
 459         if (!pmd_none(*pmd))
 460                 add_hash_page(vma->vm_mm->context, address, pmd_val(*pmd));
 461 #else
 462         pgdir = vma->vm_mm->pgd;
 463         if (pgdir == NULL)
 464                 return;
 465
 466         ptep = find_linux_pte(pgdir, ea);
 467         if (!ptep)
 468                 return;
 469
 470         vsid = get_vsid(vma->vm_mm->context.id, ea);
 471
 472         local_irq_save(flags);
 473         tmp = cpumask_of_cpu(smp_processor_id());
 474         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
 475                 local = 1;
 476
 477         __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
 478                     0x300, local);
 479         local_irq_restore(flags);
 480 #endif
 481 #endif
 482 }