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