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/export.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 #include <linux/slab.h>
  38
  39 #include <asm/pgalloc.h>
  40 #include <asm/prom.h>
  41 #include <asm/io.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/pgtable.h>
  44 #include <asm/mmu.h>
  45 #include <asm/smp.h>
  46 #include <asm/machdep.h>
  47 #include <asm/btext.h>
  48 #include <asm/tlb.h>
  49 #include <asm/sections.h>
  50 #include <asm/sparsemem.h>
  51 #include <asm/vdso.h>
  52 #include <asm/fixmap.h>
  53 #include <asm/swiotlb.h>
  54
  55 #include "mmu_decl.h"
  56
  57 #ifndef CPU_FTR_COHERENT_ICACHE
  58 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  59 #define CPU_FTR_NOEXECUTE       0
  60 #endif
  61
  62 int init_bootmem_done;
  63 int mem_init_done;
  64 phys_addr_t memory_limit;
  65
  66 #ifdef CONFIG_HIGHMEM
  67 pte_t *kmap_pte;
  68 pgprot_t kmap_prot;
  69
  70 EXPORT_SYMBOL(kmap_prot);
  71 EXPORT_SYMBOL(kmap_pte);
  72
  73 static inline pte_t *virt_to_kpte(unsigned long vaddr)
  74 {
  75         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  76                         vaddr), vaddr), vaddr);
  77 }
  78 #endif
  79
  80 int page_is_ram(unsigned long pfn)
  81 {
  82 #ifndef CONFIG_PPC64    /* XXX for now */
  83         return pfn < max_pfn;
  84 #else
  85         unsigned long paddr = (pfn << PAGE_SHIFT);
  86         struct memblock_region *reg;
  87
  88         for_each_memblock(memory, reg)
  89                 if (paddr >= reg->base && paddr < (reg->base + reg->size))
  90                         return 1;
  91         return 0;
  92 #endif
  93 }
  94
  95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  96                               unsigned long size, pgprot_t vma_prot)
  97 {
  98         if (ppc_md.phys_mem_access_prot)
  99                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 100
 101         if (!page_is_ram(pfn))
 102                 vma_prot = pgprot_noncached(vma_prot);
 103
 104         return vma_prot;
 105 }
 106 EXPORT_SYMBOL(phys_mem_access_prot);
 107
 108 #ifdef CONFIG_MEMORY_HOTPLUG
 109
 110 #ifdef CONFIG_NUMA
 111 int memory_add_physaddr_to_nid(u64 start)
 112 {
 113         return hot_add_scn_to_nid(start);
 114 }
 115 #endif
 116
 117 int arch_add_memory(int nid, u64 start, u64 size)
 118 {
 119         struct pglist_data *pgdata;
 120         struct zone *zone;
 121         unsigned long start_pfn = start >> PAGE_SHIFT;
 122         unsigned long nr_pages = size >> PAGE_SHIFT;
 123
 124         pgdata = NODE_DATA(nid);
 125
 126         start = (unsigned long)__va(start);
 127         if (create_section_mapping(start, start + size))
 128                 return -EINVAL;
 129
 130         /* this should work for most non-highmem platforms */
 131         zone = pgdata->node_zones;
 132
 133         return __add_pages(nid, zone, start_pfn, nr_pages);
 134 }
 135 #endif /* CONFIG_MEMORY_HOTPLUG */
 136
 137 /*
 138  * walk_memory_resource() needs to make sure there is no holes in a given
 139  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 140  * Instead it maintains it in memblock.memory structures.  Walk through the
 141  * memory regions, find holes and callback for contiguous regions.
 142  */
 143 int
 144 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 145                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 146 {
 147         struct memblock_region *reg;
 148         unsigned long end_pfn = start_pfn + nr_pages;
 149         unsigned long tstart, tend;
 150         int ret = -1;
 151
 152         for_each_memblock(memory, reg) {
 153                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 154                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 155                 if (tstart >= tend)
 156                         continue;
 157                 ret = (*func)(tstart, tend - tstart, arg);
 158                 if (ret)
 159                         break;
 160         }
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 164
 165 /*
 166  * Initialize the bootmem system and give it all the memory we
 167  * have available.  If we are using highmem, we only put the
 168  * lowmem into the bootmem system.
 169  */
 170 #ifndef CONFIG_NEED_MULTIPLE_NODES
 171 void __init do_init_bootmem(void)
 172 {
 173         unsigned long start, bootmap_pages;
 174         unsigned long total_pages;
 175         struct memblock_region *reg;
 176         int boot_mapsize;
 177
 178         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 179         total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 180 #ifdef CONFIG_HIGHMEM
 181         total_pages = total_lowmem >> PAGE_SHIFT;
 182         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 183 #endif
 184
 185         /*
 186          * Find an area to use for the bootmem bitmap.  Calculate the size of
 187          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 188          * Add 1 additional page in case the address isn't page-aligned.
 189          */
 190         bootmap_pages = bootmem_bootmap_pages(total_pages);
 191
 192         start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 193
 194         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 195         boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
 196
 197         /* Add active regions with valid PFNs */
 198         for_each_memblock(memory, reg) {
 199                 unsigned long start_pfn, end_pfn;
 200                 start_pfn = memblock_region_memory_base_pfn(reg);
 201                 end_pfn = memblock_region_memory_end_pfn(reg);
 202                 add_active_range(0, start_pfn, end_pfn);
 203         }
 204
 205         /* Add all physical memory to the bootmem map, mark each area
 206          * present.
 207          */
 208 #ifdef CONFIG_HIGHMEM
 209         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 210
 211         /* reserve the sections we're already using */
 212         for_each_memblock(reserved, reg) {
 213                 unsigned long top = reg->base + reg->size - 1;
 214                 if (top < lowmem_end_addr)
 215                         reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 216                 else if (reg->base < lowmem_end_addr) {
 217                         unsigned long trunc_size = lowmem_end_addr - reg->base;
 218                         reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
 219                 }
 220         }
 221 #else
 222         free_bootmem_with_active_regions(0, max_pfn);
 223
 224         /* reserve the sections we're already using */
 225         for_each_memblock(reserved, reg)
 226                 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 227 #endif
 228         /* XXX need to clip this if using highmem? */
 229         sparse_memory_present_with_active_regions(0);
 230
 231         init_bootmem_done = 1;
 232 }
 233
 234 /* mark pages that don't exist as nosave */
 235 static int __init mark_nonram_nosave(void)
 236 {
 237         struct memblock_region *reg, *prev = NULL;
 238
 239         for_each_memblock(memory, reg) {
 240                 if (prev &&
 241                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 242                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 243                                                memblock_region_memory_base_pfn(reg));
 244                 prev = reg;
 245         }
 246         return 0;
 247 }
 248
 249 /*
 250  * paging_init() sets up the page tables - in fact we've already done this.
 251  */
 252 void __init paging_init(void)
 253 {
 254         unsigned long long total_ram = memblock_phys_mem_size();
 255         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 256         unsigned long max_zone_pfns[MAX_NR_ZONES];
 257
 258 #ifdef CONFIG_PPC32
 259         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 260         unsigned long end = __fix_to_virt(FIX_HOLE);
 261
 262         for (; v < end; v += PAGE_SIZE)
 263                 map_page(v, 0, 0); /* XXX gross */
 264 #endif
 265
 266 #ifdef CONFIG_HIGHMEM
 267         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 268         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 269
 270         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 271         kmap_prot = PAGE_KERNEL;
 272 #endif /* CONFIG_HIGHMEM */
 273
 274         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 275                (unsigned long long)top_of_ram, total_ram);
 276         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 277                (long int)((top_of_ram - total_ram) >> 20));
 278         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 279 #ifdef CONFIG_HIGHMEM
 280         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 281         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 282 #else
 283         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 284 #endif
 285         free_area_init_nodes(max_zone_pfns);
 286
 287         mark_nonram_nosave();
 288 }
 289 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 290
 291 void __init mem_init(void)
 292 {
 293 #ifdef CONFIG_NEED_MULTIPLE_NODES
 294         int nid;
 295 #endif
 296         pg_data_t *pgdat;
 297         unsigned long i;
 298         struct page *page;
 299         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 300
 301 #ifdef CONFIG_SWIOTLB
 302         if (ppc_swiotlb_enable)
 303                 swiotlb_init(1);
 304 #endif
 305
 306         num_physpages = memblock_phys_mem_size() >> PAGE_SHIFT;
 307         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 308
 309 #ifdef CONFIG_NEED_MULTIPLE_NODES
 310         for_each_online_node(nid) {
 311                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 312                         printk("freeing bootmem node %d\n", nid);
 313                         totalram_pages +=
 314                                 free_all_bootmem_node(NODE_DATA(nid));
 315                 }
 316         }
 317 #else
 318         max_mapnr = max_pfn;
 319         totalram_pages += free_all_bootmem();
 320 #endif
 321         for_each_online_pgdat(pgdat) {
 322                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 323                         if (!pfn_valid(pgdat->node_start_pfn + i))
 324                                 continue;
 325                         page = pgdat_page_nr(pgdat, i);
 326                         if (PageReserved(page))
 327                                 reservedpages++;
 328                 }
 329         }
 330
 331         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 332         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 333         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 334         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 335
 336 #ifdef CONFIG_HIGHMEM
 337         {
 338                 unsigned long pfn, highmem_mapnr;
 339
 340                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 341                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 342                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 343                         struct page *page = pfn_to_page(pfn);
 344                         if (memblock_is_reserved(paddr))
 345                                 continue;
 346                         ClearPageReserved(page);
 347                         init_page_count(page);
 348                         __free_page(page);
 349                         totalhigh_pages++;
 350                         reservedpages--;
 351                 }
 352                 totalram_pages += totalhigh_pages;
 353                 printk(KERN_DEBUG "High memory: %luk\n",
 354                        totalhigh_pages << (PAGE_SHIFT-10));
 355         }
 356 #endif /* CONFIG_HIGHMEM */
 357
 358 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 359         /*
 360          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 361          * functions.... do it here for the non-smp case.
 362          */
 363         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 364                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 365 #endif
 366
 367         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 368                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 369                 nr_free_pages() << (PAGE_SHIFT-10),
 370                 num_physpages << (PAGE_SHIFT-10),
 371                 codesize >> 10,
 372                 reservedpages << (PAGE_SHIFT-10),
 373                 datasize >> 10,
 374                 bsssize >> 10,
 375                 initsize >> 10);
 376
 377 #ifdef CONFIG_PPC32
 378         pr_info("Kernel virtual memory layout:\n");
 379         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 380 #ifdef CONFIG_HIGHMEM
 381         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 382                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 383 #endif /* CONFIG_HIGHMEM */
 384 #ifdef CONFIG_NOT_COHERENT_CACHE
 385         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 386                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 387 #endif /* CONFIG_NOT_COHERENT_CACHE */
 388         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 389                 ioremap_bot, IOREMAP_TOP);
 390         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 391                 VMALLOC_START, VMALLOC_END);
 392 #endif /* CONFIG_PPC32 */
 393
 394         mem_init_done = 1;
 395 }
 396
 397 void free_initmem(void)
 398 {
 399         unsigned long addr;
 400
 401         ppc_md.progress = ppc_printk_progress;
 402
 403         addr = (unsigned long)__init_begin;
 404         for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
 405                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 406                 ClearPageReserved(virt_to_page(addr));
 407                 init_page_count(virt_to_page(addr));
 408                 free_page(addr);
 409                 totalram_pages++;
 410         }
 411         pr_info("Freeing unused kernel memory: %luk freed\n",
 412                 ((unsigned long)__init_end -
 413                 (unsigned long)__init_begin) >> 10);
 414 }
 415
 416 #ifdef CONFIG_BLK_DEV_INITRD
 417 void __init free_initrd_mem(unsigned long start, unsigned long end)
 418 {
 419         if (start >= end)
 420                 return;
 421
 422         start = _ALIGN_DOWN(start, PAGE_SIZE);
 423         end = _ALIGN_UP(end, PAGE_SIZE);
 424         pr_info("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 425
 426         for (; start < end; start += PAGE_SIZE) {
 427                 ClearPageReserved(virt_to_page(start));
 428                 init_page_count(virt_to_page(start));
 429                 free_page(start);
 430                 totalram_pages++;
 431         }
 432 }
 433 #endif
 434
 435 /*
 436  * This is called when a page has been modified by the kernel.
 437  * It just marks the page as not i-cache clean.  We do the i-cache
 438  * flush later when the page is given to a user process, if necessary.
 439  */
 440 void flush_dcache_page(struct page *page)
 441 {
 442         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 443                 return;
 444         /* avoid an atomic op if possible */
 445         if (test_bit(PG_arch_1, &page->flags))
 446                 clear_bit(PG_arch_1, &page->flags);
 447 }
 448 EXPORT_SYMBOL(flush_dcache_page);
 449
 450 void flush_dcache_icache_page(struct page *page)
 451 {
 452 #ifdef CONFIG_HUGETLB_PAGE
 453         if (PageCompound(page)) {
 454                 flush_dcache_icache_hugepage(page);
 455                 return;
 456         }
 457 #endif
 458 #ifdef CONFIG_BOOKE
 459         {
 460                 void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 461                 __flush_dcache_icache(start);
 462                 kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 463         }
 464 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 465         /* On 8xx there is no need to kmap since highmem is not supported */
 466         __flush_dcache_icache(page_address(page));
 467 #else
 468         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 469 #endif
 470 }
 471
 472 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 473 {
 474         clear_page(page);
 475
 476         /*
 477          * We shouldn't have to do this, but some versions of glibc
 478          * require it (ld.so assumes zero filled pages are icache clean)
 479          * - Anton
 480          */
 481         flush_dcache_page(pg);
 482 }
 483 EXPORT_SYMBOL(clear_user_page);
 484
 485 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 486                     struct page *pg)
 487 {
 488         copy_page(vto, vfrom);
 489
 490         /*
 491          * We should be able to use the following optimisation, however
 492          * there are two problems.
 493          * Firstly a bug in some versions of binutils meant PLT sections
 494          * were not marked executable.
 495          * Secondly the first word in the GOT section is blrl, used
 496          * to establish the GOT address. Until recently the GOT was
 497          * not marked executable.
 498          * - Anton
 499          */
 500 #if 0
 501         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 502                 return;
 503 #endif
 504
 505         flush_dcache_page(pg);
 506 }
 507
 508 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 509                              unsigned long addr, int len)
 510 {
 511         unsigned long maddr;
 512
 513         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 514         flush_icache_range(maddr, maddr + len);
 515         kunmap(page);
 516 }
 517 EXPORT_SYMBOL(flush_icache_user_range);
 518
 519 /*
 520  * This is called at the end of handling a user page fault, when the
 521  * fault has been handled by updating a PTE in the linux page tables.
 522  * We use it to preload an HPTE into the hash table corresponding to
 523  * the updated linux PTE.
 524  *
 525  * This must always be called with the pte lock held.
 526  */
 527 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 528                       pte_t *ptep)
 529 {
 530 #ifdef CONFIG_PPC_STD_MMU
 531         unsigned long access = 0, trap;
 532
 533         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 534         if (!pte_young(*ptep) || address >= TASK_SIZE)
 535                 return;
 536
 537         /* We try to figure out if we are coming from an instruction
 538          * access fault and pass that down to __hash_page so we avoid
 539          * double-faulting on execution of fresh text. We have to test
 540          * for regs NULL since init will get here first thing at boot
 541          *
 542          * We also avoid filling the hash if not coming from a fault
 543          */
 544         if (current->thread.regs == NULL)
 545                 return;
 546         trap = TRAP(current->thread.regs);
 547         if (trap == 0x400)
 548                 access |= _PAGE_EXEC;
 549         else if (trap != 0x300)
 550                 return;
 551         hash_preload(vma->vm_mm, address, access, trap);
 552 #endif /* CONFIG_PPC_STD_MMU */
 553 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
 554         && defined(CONFIG_HUGETLB_PAGE)
 555         if (is_vm_hugetlb_page(vma))
 556                 book3e_hugetlb_preload(vma->vm_mm, address, *ptep);
 557 #endif
 558 }
 559
 560 /*
 561  * System memory should not be in /proc/iomem but various tools expect it
 562  * (eg kdump).
 563  */
 564 static int add_system_ram_resources(void)
 565 {
 566         struct memblock_region *reg;
 567
 568         for_each_memblock(memory, reg) {
 569                 struct resource *res;
 570                 unsigned long base = reg->base;
 571                 unsigned long size = reg->size;
 572
 573                 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 574                 WARN_ON(!res);
 575
 576                 if (res) {
 577                         res->name = "System RAM";
 578                         res->start = base;
 579                         res->end = base + size - 1;
 580                         res->flags = IORESOURCE_MEM;
 581                         WARN_ON(request_resource(&iomem_resource, res) < 0);
 582                 }
 583         }
 584
 585         return 0;
 586 }
 587 subsys_initcall(add_system_ram_resources);