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  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
   9  *
  10  *  Derived from "arch/i386/mm/init.c"
  11  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  12  *
  13  *  This program is free software; you can redistribute it and/or
  14  *  modify it under the terms of the GNU General Public License
  15  *  as published by the Free Software Foundation; either version
  16  *  2 of the License, or (at your option) any later version.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/sched.h>
  22 #include <linux/kernel.h>
  23 #include <linux/errno.h>
  24 #include <linux/string.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/stddef.h>
  28 #include <linux/init.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/highmem.h>
  31 #include <linux/initrd.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/suspend.h>
  34 #include <linux/lmb.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/sections.h>
  47 #include <asm/sparsemem.h>
  48 #include <asm/vdso.h>
  49 #include <asm/fixmap.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 phys_addr_t memory_limit;
  61
  62 #ifdef CONFIG_HIGHMEM
  63 pte_t *kmap_pte;
  64 pgprot_t kmap_prot;
  65
  66 EXPORT_SYMBOL(kmap_prot);
  67 EXPORT_SYMBOL(kmap_pte);
  68
  69 static inline pte_t *virt_to_kpte(unsigned long vaddr)
  70 {
  71         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  72                         vaddr), vaddr), vaddr);
  73 }
  74 #endif
  75
  76 int page_is_ram(unsigned long pfn)
  77 {
  78 #ifndef CONFIG_PPC64    /* XXX for now */
  79         return pfn < max_pfn;
  80 #else
  81         unsigned long paddr = (pfn << PAGE_SHIFT);
  82         int i;
  83         for (i=0; i < lmb.memory.cnt; i++) {
  84                 unsigned long base;
  85
  86                 base = lmb.memory.region[i].base;
  87
  88                 if ((paddr >= base) &&
  89                         (paddr < (base + lmb.memory.region[i].size))) {
  90                         return 1;
  91                 }
  92         }
  93
  94         return 0;
  95 #endif
  96 }
  97
  98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  99                               unsigned long size, pgprot_t vma_prot)
 100 {
 101         if (ppc_md.phys_mem_access_prot)
 102                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 103
 104         if (!page_is_ram(pfn))
 105                 vma_prot = pgprot_noncached(vma_prot);
 106
 107         return vma_prot;
 108 }
 109 EXPORT_SYMBOL(phys_mem_access_prot);
 110
 111 #ifdef CONFIG_MEMORY_HOTPLUG
 112
 113 #ifdef CONFIG_NUMA
 114 int memory_add_physaddr_to_nid(u64 start)
 115 {
 116         return hot_add_scn_to_nid(start);
 117 }
 118 #endif
 119
 120 int arch_add_memory(int nid, u64 start, u64 size)
 121 {
 122         struct pglist_data *pgdata;
 123         struct zone *zone;
 124         unsigned long start_pfn = start >> PAGE_SHIFT;
 125         unsigned long nr_pages = size >> PAGE_SHIFT;
 126
 127         pgdata = NODE_DATA(nid);
 128
 129         start = (unsigned long)__va(start);
 130         create_section_mapping(start, start + size);
 131
 132         /* this should work for most non-highmem platforms */
 133         zone = pgdata->node_zones;
 134
 135         return __add_pages(nid, zone, start_pfn, nr_pages);
 136 }
 137 #endif /* CONFIG_MEMORY_HOTPLUG */
 138
 139 /*
 140  * walk_memory_resource() needs to make sure there is no holes in a given
 141  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 142  * Instead it maintains it in lmb.memory structures.  Walk through the
 143  * memory regions, find holes and callback for contiguous regions.
 144  */
 145 int
 146 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 147                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 148 {
 149         struct lmb_property res;
 150         unsigned long pfn, len;
 151         u64 end;
 152         int ret = -1;
 153
 154         res.base = (u64) start_pfn << PAGE_SHIFT;
 155         res.size = (u64) nr_pages << PAGE_SHIFT;
 156
 157         end = res.base + res.size - 1;
 158         while ((res.base < end) && (lmb_find(&res) >= 0)) {
 159                 pfn = (unsigned long)(res.base >> PAGE_SHIFT);
 160                 len = (unsigned long)(res.size >> PAGE_SHIFT);
 161                 ret = (*func)(pfn, len, arg);
 162                 if (ret)
 163                         break;
 164                 res.base += (res.size + 1);
 165                 res.size = (end - res.base + 1);
 166         }
 167         return ret;
 168 }
 169 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 170
 171 /*
 172  * Initialize the bootmem system and give it all the memory we
 173  * have available.  If we are using highmem, we only put the
 174  * lowmem into the bootmem system.
 175  */
 176 #ifndef CONFIG_NEED_MULTIPLE_NODES
 177 void __init do_init_bootmem(void)
 178 {
 179         unsigned long i;
 180         unsigned long start, bootmap_pages;
 181         unsigned long total_pages;
 182         int boot_mapsize;
 183
 184         max_low_pfn = max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
 185         total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 186 #ifdef CONFIG_HIGHMEM
 187         total_pages = total_lowmem >> PAGE_SHIFT;
 188         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 189 #endif
 190
 191         /*
 192          * Find an area to use for the bootmem bitmap.  Calculate the size of
 193          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 194          * Add 1 additional page in case the address isn't page-aligned.
 195          */
 196         bootmap_pages = bootmem_bootmap_pages(total_pages);
 197
 198         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 199
 200         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 201         boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
 202
 203         /* Add active regions with valid PFNs */
 204         for (i = 0; i < lmb.memory.cnt; i++) {
 205                 unsigned long start_pfn, end_pfn;
 206                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
 207                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
 208                 add_active_range(0, start_pfn, end_pfn);
 209         }
 210
 211         /* Add all physical memory to the bootmem map, mark each area
 212          * present.
 213          */
 214 #ifdef CONFIG_HIGHMEM
 215         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 216
 217         /* reserve the sections we're already using */
 218         for (i = 0; i < lmb.reserved.cnt; i++) {
 219                 unsigned long addr = lmb.reserved.region[i].base +
 220                                      lmb_size_bytes(&lmb.reserved, i) - 1;
 221                 if (addr < lowmem_end_addr)
 222                         reserve_bootmem(lmb.reserved.region[i].base,
 223                                         lmb_size_bytes(&lmb.reserved, i),
 224                                         BOOTMEM_DEFAULT);
 225                 else if (lmb.reserved.region[i].base < lowmem_end_addr) {
 226                         unsigned long adjusted_size = lowmem_end_addr -
 227                                       lmb.reserved.region[i].base;
 228                         reserve_bootmem(lmb.reserved.region[i].base,
 229                                         adjusted_size, BOOTMEM_DEFAULT);
 230                 }
 231         }
 232 #else
 233         free_bootmem_with_active_regions(0, max_pfn);
 234
 235         /* reserve the sections we're already using */
 236         for (i = 0; i < lmb.reserved.cnt; i++)
 237                 reserve_bootmem(lmb.reserved.region[i].base,
 238                                 lmb_size_bytes(&lmb.reserved, i),
 239                                 BOOTMEM_DEFAULT);
 240
 241 #endif
 242         /* XXX need to clip this if using highmem? */
 243         sparse_memory_present_with_active_regions(0);
 244
 245         init_bootmem_done = 1;
 246 }
 247
 248 /* mark pages that don't exist as nosave */
 249 static int __init mark_nonram_nosave(void)
 250 {
 251         unsigned long lmb_next_region_start_pfn,
 252                       lmb_region_max_pfn;
 253         int i;
 254
 255         for (i = 0; i < lmb.memory.cnt - 1; i++) {
 256                 lmb_region_max_pfn =
 257                         (lmb.memory.region[i].base >> PAGE_SHIFT) +
 258                         (lmb.memory.region[i].size >> PAGE_SHIFT);
 259                 lmb_next_region_start_pfn =
 260                         lmb.memory.region[i+1].base >> PAGE_SHIFT;
 261
 262                 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
 263                         register_nosave_region(lmb_region_max_pfn,
 264                                                lmb_next_region_start_pfn);
 265         }
 266
 267         return 0;
 268 }
 269
 270 /*
 271  * paging_init() sets up the page tables - in fact we've already done this.
 272  */
 273 void __init paging_init(void)
 274 {
 275         unsigned long total_ram = lmb_phys_mem_size();
 276         phys_addr_t top_of_ram = lmb_end_of_DRAM();
 277         unsigned long max_zone_pfns[MAX_NR_ZONES];
 278
 279 #ifdef CONFIG_PPC32
 280         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 281         unsigned long end = __fix_to_virt(FIX_HOLE);
 282
 283         for (; v < end; v += PAGE_SIZE)
 284                 map_page(v, 0, 0); /* XXX gross */
 285 #endif
 286
 287 #ifdef CONFIG_HIGHMEM
 288         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 289         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 290
 291         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 292         kmap_prot = PAGE_KERNEL;
 293 #endif /* CONFIG_HIGHMEM */
 294
 295         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
 296                (unsigned long long)top_of_ram, total_ram);
 297         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 298                (long int)((top_of_ram - total_ram) >> 20));
 299         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 300 #ifdef CONFIG_HIGHMEM
 301         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 302         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 303 #else
 304         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 305 #endif
 306         free_area_init_nodes(max_zone_pfns);
 307
 308         mark_nonram_nosave();
 309 }
 310 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 311
 312 void __init mem_init(void)
 313 {
 314 #ifdef CONFIG_NEED_MULTIPLE_NODES
 315         int nid;
 316 #endif
 317         pg_data_t *pgdat;
 318         unsigned long i;
 319         struct page *page;
 320         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 321
 322         num_physpages = lmb.memory.size >> PAGE_SHIFT;
 323         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 324
 325 #ifdef CONFIG_NEED_MULTIPLE_NODES
 326         for_each_online_node(nid) {
 327                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 328                         printk("freeing bootmem node %d\n", nid);
 329                         totalram_pages +=
 330                                 free_all_bootmem_node(NODE_DATA(nid));
 331                 }
 332         }
 333 #else
 334         max_mapnr = max_pfn;
 335         totalram_pages += free_all_bootmem();
 336 #endif
 337         for_each_online_pgdat(pgdat) {
 338                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 339                         if (!pfn_valid(pgdat->node_start_pfn + i))
 340                                 continue;
 341                         page = pgdat_page_nr(pgdat, i);
 342                         if (PageReserved(page))
 343                                 reservedpages++;
 344                 }
 345         }
 346
 347         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 348         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 349         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 350         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 351
 352 #ifdef CONFIG_HIGHMEM
 353         {
 354                 unsigned long pfn, highmem_mapnr;
 355
 356                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 357                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 358                         struct page *page = pfn_to_page(pfn);
 359                         if (lmb_is_reserved(pfn << PAGE_SHIFT))
 360                                 continue;
 361                         ClearPageReserved(page);
 362                         init_page_count(page);
 363                         __free_page(page);
 364                         totalhigh_pages++;
 365                         reservedpages--;
 366                 }
 367                 totalram_pages += totalhigh_pages;
 368                 printk(KERN_DEBUG "High memory: %luk\n",
 369                        totalhigh_pages << (PAGE_SHIFT-10));
 370         }
 371 #endif /* CONFIG_HIGHMEM */
 372
 373         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 374                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 375                 nr_free_pages() << (PAGE_SHIFT-10),
 376                 num_physpages << (PAGE_SHIFT-10),
 377                 codesize >> 10,
 378                 reservedpages << (PAGE_SHIFT-10),
 379                 datasize >> 10,
 380                 bsssize >> 10,
 381                 initsize >> 10);
 382
 383 #ifdef CONFIG_PPC32
 384         pr_info("Kernel virtual memory layout:\n");
 385         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 386 #ifdef CONFIG_HIGHMEM
 387         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 388                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 389 #endif /* CONFIG_HIGHMEM */
 390 #ifdef CONFIG_NOT_COHERENT_CACHE
 391         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 392                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 393 #endif /* CONFIG_NOT_COHERENT_CACHE */
 394         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 395                 ioremap_bot, IOREMAP_TOP);
 396         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 397                 VMALLOC_START, VMALLOC_END);
 398 #endif /* CONFIG_PPC32 */
 399
 400         mem_init_done = 1;
 401 }
 402
 403 /*
 404  * This is called when a page has been modified by the kernel.
 405  * It just marks the page as not i-cache clean.  We do the i-cache
 406  * flush later when the page is given to a user process, if necessary.
 407  */
 408 void flush_dcache_page(struct page *page)
 409 {
 410         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 411                 return;
 412         /* avoid an atomic op if possible */
 413         if (test_bit(PG_arch_1, &page->flags))
 414                 clear_bit(PG_arch_1, &page->flags);
 415 }
 416 EXPORT_SYMBOL(flush_dcache_page);
 417
 418 void flush_dcache_icache_page(struct page *page)
 419 {
 420 #ifdef CONFIG_BOOKE
 421         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 422         __flush_dcache_icache(start);
 423         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 424 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 425         /* On 8xx there is no need to kmap since highmem is not supported */
 426         __flush_dcache_icache(page_address(page));
 427 #else
 428         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 429 #endif
 430
 431 }
 432 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 433 {
 434         clear_page(page);
 435
 436         /*
 437          * We shouldnt have to do this, but some versions of glibc
 438          * require it (ld.so assumes zero filled pages are icache clean)
 439          * - Anton
 440          */
 441         flush_dcache_page(pg);
 442 }
 443 EXPORT_SYMBOL(clear_user_page);
 444
 445 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 446                     struct page *pg)
 447 {
 448         copy_page(vto, vfrom);
 449
 450         /*
 451          * We should be able to use the following optimisation, however
 452          * there are two problems.
 453          * Firstly a bug in some versions of binutils meant PLT sections
 454          * were not marked executable.
 455          * Secondly the first word in the GOT section is blrl, used
 456          * to establish the GOT address. Until recently the GOT was
 457          * not marked executable.
 458          * - Anton
 459          */
 460 #if 0
 461         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 462                 return;
 463 #endif
 464
 465         flush_dcache_page(pg);
 466 }
 467
 468 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 469                              unsigned long addr, int len)
 470 {
 471         unsigned long maddr;
 472
 473         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 474         flush_icache_range(maddr, maddr + len);
 475         kunmap(page);
 476 }
 477 EXPORT_SYMBOL(flush_icache_user_range);
 478
 479 /*
 480  * This is called at the end of handling a user page fault, when the
 481  * fault has been handled by updating a PTE in the linux page tables.
 482  * We use it to preload an HPTE into the hash table corresponding to
 483  * the updated linux PTE.
 484  *
 485  * This must always be called with the pte lock held.
 486  */
 487 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 488                       pte_t pte)
 489 {
 490 #ifdef CONFIG_PPC_STD_MMU
 491         unsigned long access = 0, trap;
 492
 493         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 494         if (!pte_young(pte) || address >= TASK_SIZE)
 495                 return;
 496
 497         /* We try to figure out if we are coming from an instruction
 498          * access fault and pass that down to __hash_page so we avoid
 499          * double-faulting on execution of fresh text. We have to test
 500          * for regs NULL since init will get here first thing at boot
 501          *
 502          * We also avoid filling the hash if not coming from a fault
 503          */
 504         if (current->thread.regs == NULL)
 505                 return;
 506         trap = TRAP(current->thread.regs);
 507         if (trap == 0x400)
 508                 access |= _PAGE_EXEC;
 509         else if (trap != 0x300)
 510                 return;
 511         hash_preload(vma->vm_mm, address, access, trap);
 512 #endif /* CONFIG_PPC_STD_MMU */
 513 }