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/gfp.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 #include <linux/suspend.h>
  35 #include <linux/memblock.h>
  36 #include <linux/hugetlb.h>
  37
  38 #include <asm/pgalloc.h>
  39 #include <asm/prom.h>
  40 #include <asm/io.h>
  41 #include <asm/mmu_context.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/mmu.h>
  44 #include <asm/smp.h>
  45 #include <asm/machdep.h>
  46 #include <asm/btext.h>
  47 #include <asm/tlb.h>
  48 #include <asm/sections.h>
  49 #include <asm/sparsemem.h>
  50 #include <asm/vdso.h>
  51 #include <asm/fixmap.h>
  52 #include <asm/swiotlb.h>
  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 phys_addr_t memory_limit;
  64
  65 #ifdef CONFIG_HIGHMEM
  66 pte_t *kmap_pte;
  67 pgprot_t kmap_prot;
  68
  69 EXPORT_SYMBOL(kmap_prot);
  70 EXPORT_SYMBOL(kmap_pte);
  71
  72 static inline pte_t *virt_to_kpte(unsigned long vaddr)
  73 {
  74         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  75                         vaddr), vaddr), vaddr);
  76 }
  77 #endif
  78
  79 int page_is_ram(unsigned long pfn)
  80 {
  81 #ifndef CONFIG_PPC64    /* XXX for now */
  82         return pfn < max_pfn;
  83 #else
  84         unsigned long paddr = (pfn << PAGE_SHIFT);
  85         struct memblock_region *reg;
  86
  87         for_each_memblock(memory, reg)
  88                 if (paddr >= reg->base && paddr < (reg->base + reg->size))
  89                         return 1;
  90         return 0;
  91 #endif
  92 }
  93
  94 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  95                               unsigned long size, pgprot_t vma_prot)
  96 {
  97         if (ppc_md.phys_mem_access_prot)
  98                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  99
 100         if (!page_is_ram(pfn))
 101                 vma_prot = pgprot_noncached(vma_prot);
 102
 103         return vma_prot;
 104 }
 105 EXPORT_SYMBOL(phys_mem_access_prot);
 106
 107 #ifdef CONFIG_MEMORY_HOTPLUG
 108
 109 #ifdef CONFIG_NUMA
 110 int memory_add_physaddr_to_nid(u64 start)
 111 {
 112         return hot_add_scn_to_nid(start);
 113 }
 114 #endif
 115
 116 int arch_add_memory(int nid, u64 start, u64 size)
 117 {
 118         struct pglist_data *pgdata;
 119         struct zone *zone;
 120         unsigned long start_pfn = start >> PAGE_SHIFT;
 121         unsigned long nr_pages = size >> PAGE_SHIFT;
 122
 123         pgdata = NODE_DATA(nid);
 124
 125         start = (unsigned long)__va(start);
 126         if (create_section_mapping(start, start + size))
 127                 return -EINVAL;
 128
 129         /* this should work for most non-highmem platforms */
 130         zone = pgdata->node_zones;
 131
 132         return __add_pages(nid, zone, start_pfn, nr_pages);
 133 }
 134 #endif /* CONFIG_MEMORY_HOTPLUG */
 135
 136 /*
 137  * walk_memory_resource() needs to make sure there is no holes in a given
 138  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 139  * Instead it maintains it in memblock.memory structures.  Walk through the
 140  * memory regions, find holes and callback for contiguous regions.
 141  */
 142 int
 143 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 144                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 145 {
 146         struct memblock_region *reg;
 147         unsigned long end_pfn = start_pfn + nr_pages;
 148         unsigned long tstart, tend;
 149         int ret = -1;
 150
 151         for_each_memblock(memory, reg) {
 152                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 153                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 154                 if (tstart >= tend)
 155                         continue;
 156                 ret = (*func)(tstart, tend - tstart, arg);
 157                 if (ret)
 158                         break;
 159         }
 160         return ret;
 161 }
 162 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 163
 164 /*
 165  * Initialize the bootmem system and give it all the memory we
 166  * have available.  If we are using highmem, we only put the
 167  * lowmem into the bootmem system.
 168  */
 169 #ifndef CONFIG_NEED_MULTIPLE_NODES
 170 void __init do_init_bootmem(void)
 171 {
 172         unsigned long start, bootmap_pages;
 173         unsigned long total_pages;
 174         struct memblock_region *reg;
 175         int boot_mapsize;
 176
 177         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 178         total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 179 #ifdef CONFIG_HIGHMEM
 180         total_pages = total_lowmem >> PAGE_SHIFT;
 181         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 182 #endif
 183
 184         /*
 185          * Find an area to use for the bootmem bitmap.  Calculate the size of
 186          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 187          * Add 1 additional page in case the address isn't page-aligned.
 188          */
 189         bootmap_pages = bootmem_bootmap_pages(total_pages);
 190
 191         start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 192
 193         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 194         boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
 195
 196         /* Add active regions with valid PFNs */
 197         for_each_memblock(memory, reg) {
 198                 unsigned long start_pfn, end_pfn;
 199                 start_pfn = memblock_region_memory_base_pfn(reg);
 200                 end_pfn = memblock_region_memory_end_pfn(reg);
 201                 add_active_range(0, start_pfn, end_pfn);
 202         }
 203
 204         /* Add all physical memory to the bootmem map, mark each area
 205          * present.
 206          */
 207 #ifdef CONFIG_HIGHMEM
 208         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 209
 210         /* reserve the sections we're already using */
 211         for_each_memblock(reserved, reg) {
 212                 unsigned long top = reg->base + reg->size - 1;
 213                 if (top < lowmem_end_addr)
 214                         reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 215                 else if (reg->base < lowmem_end_addr) {
 216                         unsigned long trunc_size = lowmem_end_addr - reg->base;
 217                         reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
 218                 }
 219         }
 220 #else
 221         free_bootmem_with_active_regions(0, max_pfn);
 222
 223         /* reserve the sections we're already using */
 224         for_each_memblock(reserved, reg)
 225                 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 226 #endif
 227         /* XXX need to clip this if using highmem? */
 228         sparse_memory_present_with_active_regions(0);
 229
 230         init_bootmem_done = 1;
 231 }
 232
 233 /* mark pages that don't exist as nosave */
 234 static int __init mark_nonram_nosave(void)
 235 {
 236         struct memblock_region *reg, *prev = NULL;
 237
 238         for_each_memblock(memory, reg) {
 239                 if (prev &&
 240                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 241                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 242                                                memblock_region_memory_base_pfn(reg));
 243                 prev = reg;
 244         }
 245         return 0;
 246 }
 247
 248 /*
 249  * paging_init() sets up the page tables - in fact we've already done this.
 250  */
 251 void __init paging_init(void)
 252 {
 253         unsigned long long total_ram = memblock_phys_mem_size();
 254         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 255         unsigned long max_zone_pfns[MAX_NR_ZONES];
 256
 257 #ifdef CONFIG_PPC32
 258         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 259         unsigned long end = __fix_to_virt(FIX_HOLE);
 260
 261         for (; v < end; v += PAGE_SIZE)
 262                 map_page(v, 0, 0); /* XXX gross */
 263 #endif
 264
 265 #ifdef CONFIG_HIGHMEM
 266         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 267         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 268
 269         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 270         kmap_prot = PAGE_KERNEL;
 271 #endif /* CONFIG_HIGHMEM */
 272
 273         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 274                (unsigned long long)top_of_ram, total_ram);
 275         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 276                (long int)((top_of_ram - total_ram) >> 20));
 277         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 278 #ifdef CONFIG_HIGHMEM
 279         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 280         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 281 #else
 282         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 283 #endif
 284         free_area_init_nodes(max_zone_pfns);
 285
 286         mark_nonram_nosave();
 287 }
 288 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 289
 290 void __init mem_init(void)
 291 {
 292 #ifdef CONFIG_NEED_MULTIPLE_NODES
 293         int nid;
 294 #endif
 295         pg_data_t *pgdat;
 296         unsigned long i;
 297         struct page *page;
 298         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 299
 300 #ifdef CONFIG_SWIOTLB
 301         if (ppc_swiotlb_enable)
 302                 swiotlb_init(1);
 303 #endif
 304
 305         num_physpages = memblock_phys_mem_size() >> PAGE_SHIFT;
 306         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 307
 308 #ifdef CONFIG_NEED_MULTIPLE_NODES
 309         for_each_online_node(nid) {
 310                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 311                         printk("freeing bootmem node %d\n", nid);
 312                         totalram_pages +=
 313                                 free_all_bootmem_node(NODE_DATA(nid));
 314                 }
 315         }
 316 #else
 317         max_mapnr = max_pfn;
 318         totalram_pages += free_all_bootmem();
 319 #endif
 320         for_each_online_pgdat(pgdat) {
 321                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 322                         if (!pfn_valid(pgdat->node_start_pfn + i))
 323                                 continue;
 324                         page = pgdat_page_nr(pgdat, i);
 325                         if (PageReserved(page))
 326                                 reservedpages++;
 327                 }
 328         }
 329
 330         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 331         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 332         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 333         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 334
 335 #ifdef CONFIG_HIGHMEM
 336         {
 337                 unsigned long pfn, highmem_mapnr;
 338
 339                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 340                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 341                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 342                         struct page *page = pfn_to_page(pfn);
 343                         if (memblock_is_reserved(paddr))
 344                                 continue;
 345                         ClearPageReserved(page);
 346                         init_page_count(page);
 347                         __free_page(page);
 348                         totalhigh_pages++;
 349                         reservedpages--;
 350                 }
 351                 totalram_pages += totalhigh_pages;
 352                 printk(KERN_DEBUG "High memory: %luk\n",
 353                        totalhigh_pages << (PAGE_SHIFT-10));
 354         }
 355 #endif /* CONFIG_HIGHMEM */
 356
 357 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 358         /*
 359          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 360          * functions.... do it here for the non-smp case.
 361          */
 362         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 363                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 364 #endif
 365
 366         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 367                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 368                 nr_free_pages() << (PAGE_SHIFT-10),
 369                 num_physpages << (PAGE_SHIFT-10),
 370                 codesize >> 10,
 371                 reservedpages << (PAGE_SHIFT-10),
 372                 datasize >> 10,
 373                 bsssize >> 10,
 374                 initsize >> 10);
 375
 376 #ifdef CONFIG_PPC32
 377         pr_info("Kernel virtual memory layout:\n");
 378         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 379 #ifdef CONFIG_HIGHMEM
 380         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 381                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 382 #endif /* CONFIG_HIGHMEM */
 383 #ifdef CONFIG_NOT_COHERENT_CACHE
 384         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 385                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 386 #endif /* CONFIG_NOT_COHERENT_CACHE */
 387         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 388                 ioremap_bot, IOREMAP_TOP);
 389         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 390                 VMALLOC_START, VMALLOC_END);
 391 #endif /* CONFIG_PPC32 */
 392
 393         mem_init_done = 1;
 394 }
 395
 396 void free_initmem(void)
 397 {
 398         unsigned long addr;
 399
 400         ppc_md.progress = ppc_printk_progress;
 401
 402         addr = (unsigned long)__init_begin;
 403         for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
 404                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 405                 ClearPageReserved(virt_to_page(addr));
 406                 init_page_count(virt_to_page(addr));
 407                 free_page(addr);
 408                 totalram_pages++;
 409         }
 410         pr_info("Freeing unused kernel memory: %luk freed\n",
 411                 ((unsigned long)__init_end -
 412                 (unsigned long)__init_begin) >> 10);
 413 }
 414
 415 #ifdef CONFIG_BLK_DEV_INITRD
 416 void __init free_initrd_mem(unsigned long start, unsigned long end)
 417 {
 418         if (start >= end)
 419                 return;
 420
 421         start = _ALIGN_DOWN(start, PAGE_SIZE);
 422         end = _ALIGN_UP(end, PAGE_SIZE);
 423         pr_info("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 424
 425         for (; start < end; start += PAGE_SIZE) {
 426                 ClearPageReserved(virt_to_page(start));
 427                 init_page_count(virt_to_page(start));
 428                 free_page(start);
 429                 totalram_pages++;
 430         }
 431 }
 432 #endif
 433
 434 /*
 435  * This is called when a page has been modified by the kernel.
 436  * It just marks the page as not i-cache clean.  We do the i-cache
 437  * flush later when the page is given to a user process, if necessary.
 438  */
 439 void flush_dcache_page(struct page *page)
 440 {
 441         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 442                 return;
 443         /* avoid an atomic op if possible */
 444         if (test_bit(PG_arch_1, &page->flags))
 445                 clear_bit(PG_arch_1, &page->flags);
 446 }
 447 EXPORT_SYMBOL(flush_dcache_page);
 448
 449 void flush_dcache_icache_page(struct page *page)
 450 {
 451 #ifdef CONFIG_HUGETLB_PAGE
 452         if (PageCompound(page)) {
 453                 flush_dcache_icache_hugepage(page);
 454                 return;
 455         }
 456 #endif
 457 #ifdef CONFIG_BOOKE
 458         {
 459                 void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 460                 __flush_dcache_icache(start);
 461                 kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 462         }
 463 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 464         /* On 8xx there is no need to kmap since highmem is not supported */
 465         __flush_dcache_icache(page_address(page));
 466 #else
 467         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 468 #endif
 469 }
 470
 471 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 472 {
 473         clear_page(page);
 474
 475         /*
 476          * We shouldn't have to do this, but some versions of glibc
 477          * require it (ld.so assumes zero filled pages are icache clean)
 478          * - Anton
 479          */
 480         flush_dcache_page(pg);
 481 }
 482 EXPORT_SYMBOL(clear_user_page);
 483
 484 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 485                     struct page *pg)
 486 {
 487         copy_page(vto, vfrom);
 488
 489         /*
 490          * We should be able to use the following optimisation, however
 491          * there are two problems.
 492          * Firstly a bug in some versions of binutils meant PLT sections
 493          * were not marked executable.
 494          * Secondly the first word in the GOT section is blrl, used
 495          * to establish the GOT address. Until recently the GOT was
 496          * not marked executable.
 497          * - Anton
 498          */
 499 #if 0
 500         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 501                 return;
 502 #endif
 503
 504         flush_dcache_page(pg);
 505 }
 506
 507 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 508                              unsigned long addr, int len)
 509 {
 510         unsigned long maddr;
 511
 512         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 513         flush_icache_range(maddr, maddr + len);
 514         kunmap(page);
 515 }
 516 EXPORT_SYMBOL(flush_icache_user_range);
 517
 518 /*
 519  * This is called at the end of handling a user page fault, when the
 520  * fault has been handled by updating a PTE in the linux page tables.
 521  * We use it to preload an HPTE into the hash table corresponding to
 522  * the updated linux PTE.
 523  *
 524  * This must always be called with the pte lock held.
 525  */
 526 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 527                       pte_t *ptep)
 528 {
 529 #ifdef CONFIG_PPC_STD_MMU
 530         unsigned long access = 0, trap;
 531
 532         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 533         if (!pte_young(*ptep) || address >= TASK_SIZE)
 534                 return;
 535
 536         /* We try to figure out if we are coming from an instruction
 537          * access fault and pass that down to __hash_page so we avoid
 538          * double-faulting on execution of fresh text. We have to test
 539          * for regs NULL since init will get here first thing at boot
 540          *
 541          * We also avoid filling the hash if not coming from a fault
 542          */
 543         if (current->thread.regs == NULL)
 544                 return;
 545         trap = TRAP(current->thread.regs);
 546         if (trap == 0x400)
 547                 access |= _PAGE_EXEC;
 548         else if (trap != 0x300)
 549                 return;
 550         hash_preload(vma->vm_mm, address, access, trap);
 551 #endif /* CONFIG_PPC_STD_MMU */
 552 }