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 #include <asm/rtas.h>
  55
  56 #include "mmu_decl.h"
  57
  58 #ifndef CPU_FTR_COHERENT_ICACHE
  59 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  60 #define CPU_FTR_NOEXECUTE       0
  61 #endif
  62
  63 int init_bootmem_done;
  64 int mem_init_done;
  65 unsigned long long memory_limit;
  66
  67 #ifdef CONFIG_HIGHMEM
  68 pte_t *kmap_pte;
  69 EXPORT_SYMBOL(kmap_pte);
  70 pgprot_t kmap_prot;
  71 EXPORT_SYMBOL(kmap_prot);
  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                 zone_for_memory(nid, start, size, 0);
 133
 134         return __add_pages(nid, zone, start_pfn, nr_pages);
 135 }
 136
 137 #ifdef CONFIG_MEMORY_HOTREMOVE
 138 int arch_remove_memory(u64 start, u64 size)
 139 {
 140         unsigned long start_pfn = start >> PAGE_SHIFT;
 141         unsigned long nr_pages = size >> PAGE_SHIFT;
 142         struct zone *zone;
 143         int ret;
 144
 145         zone = page_zone(pfn_to_page(start_pfn));
 146         ret = __remove_pages(zone, start_pfn, nr_pages);
 147         if (!ret && (ppc_md.remove_memory))
 148                 ret = ppc_md.remove_memory(start, size);
 149
 150         return ret;
 151 }
 152 #endif
 153 #endif /* CONFIG_MEMORY_HOTPLUG */
 154
 155 /*
 156  * walk_memory_resource() needs to make sure there is no holes in a given
 157  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 158  * Instead it maintains it in memblock.memory structures.  Walk through the
 159  * memory regions, find holes and callback for contiguous regions.
 160  */
 161 int
 162 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 163                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 164 {
 165         struct memblock_region *reg;
 166         unsigned long end_pfn = start_pfn + nr_pages;
 167         unsigned long tstart, tend;
 168         int ret = -1;
 169
 170         for_each_memblock(memory, reg) {
 171                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 172                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 173                 if (tstart >= tend)
 174                         continue;
 175                 ret = (*func)(tstart, tend - tstart, arg);
 176                 if (ret)
 177                         break;
 178         }
 179         return ret;
 180 }
 181 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 182
 183 /*
 184  * Initialize the bootmem system and give it all the memory we
 185  * have available.  If we are using highmem, we only put the
 186  * lowmem into the bootmem system.
 187  */
 188 #ifndef CONFIG_NEED_MULTIPLE_NODES
 189 void __init do_init_bootmem(void)
 190 {
 191         unsigned long start, bootmap_pages;
 192         unsigned long total_pages;
 193         struct memblock_region *reg;
 194         int boot_mapsize;
 195
 196         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 197         total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 198 #ifdef CONFIG_HIGHMEM
 199         total_pages = total_lowmem >> PAGE_SHIFT;
 200         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 201 #endif
 202
 203         /*
 204          * Find an area to use for the bootmem bitmap.  Calculate the size of
 205          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 206          * Add 1 additional page in case the address isn't page-aligned.
 207          */
 208         bootmap_pages = bootmem_bootmap_pages(total_pages);
 209
 210         start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 211
 212         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 213         boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
 214
 215         /* Place all memblock_regions in the same node and merge contiguous
 216          * memblock_regions
 217          */
 218         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 219
 220         /* Add all physical memory to the bootmem map, mark each area
 221          * present.
 222          */
 223 #ifdef CONFIG_HIGHMEM
 224         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 225
 226         /* reserve the sections we're already using */
 227         for_each_memblock(reserved, reg) {
 228                 unsigned long top = reg->base + reg->size - 1;
 229                 if (top < lowmem_end_addr)
 230                         reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 231                 else if (reg->base < lowmem_end_addr) {
 232                         unsigned long trunc_size = lowmem_end_addr - reg->base;
 233                         reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
 234                 }
 235         }
 236 #else
 237         free_bootmem_with_active_regions(0, max_pfn);
 238
 239         /* reserve the sections we're already using */
 240         for_each_memblock(reserved, reg)
 241                 reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
 242 #endif
 243         /* XXX need to clip this if using highmem? */
 244         sparse_memory_present_with_active_regions(0);
 245
 246         init_bootmem_done = 1;
 247 }
 248
 249 /* mark pages that don't exist as nosave */
 250 static int __init mark_nonram_nosave(void)
 251 {
 252         struct memblock_region *reg, *prev = NULL;
 253
 254         for_each_memblock(memory, reg) {
 255                 if (prev &&
 256                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 257                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 258                                                memblock_region_memory_base_pfn(reg));
 259                 prev = reg;
 260         }
 261         return 0;
 262 }
 263
 264 /*
 265  * paging_init() sets up the page tables - in fact we've already done this.
 266  */
 267 void __init paging_init(void)
 268 {
 269         unsigned long long total_ram = memblock_phys_mem_size();
 270         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 271         unsigned long max_zone_pfns[MAX_NR_ZONES];
 272
 273 #ifdef CONFIG_PPC32
 274         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 275         unsigned long end = __fix_to_virt(FIX_HOLE);
 276
 277         for (; v < end; v += PAGE_SIZE)
 278                 map_page(v, 0, 0); /* XXX gross */
 279 #endif
 280
 281 #ifdef CONFIG_HIGHMEM
 282         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 283         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 284
 285         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 286         kmap_prot = PAGE_KERNEL;
 287 #endif /* CONFIG_HIGHMEM */
 288
 289         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 290                (unsigned long long)top_of_ram, total_ram);
 291         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 292                (long int)((top_of_ram - total_ram) >> 20));
 293         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 294 #ifdef CONFIG_HIGHMEM
 295         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 296         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 297 #else
 298         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 299 #endif
 300         free_area_init_nodes(max_zone_pfns);
 301
 302         mark_nonram_nosave();
 303 }
 304 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 305
 306 static void __init register_page_bootmem_info(void)
 307 {
 308         int i;
 309
 310         for_each_online_node(i)
 311                 register_page_bootmem_info_node(NODE_DATA(i));
 312 }
 313
 314 void __init mem_init(void)
 315 {
 316         /*
 317          * book3s is limited to 16 page sizes due to encoding this in
 318          * a 4-bit field for slices.
 319          */
 320         BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
 321
 322 #ifdef CONFIG_SWIOTLB
 323         swiotlb_init(0);
 324 #endif
 325
 326         register_page_bootmem_info();
 327         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 328         set_max_mapnr(max_pfn);
 329         free_all_bootmem();
 330
 331 #ifdef CONFIG_HIGHMEM
 332         {
 333                 unsigned long pfn, highmem_mapnr;
 334
 335                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 336                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 337                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 338                         struct page *page = pfn_to_page(pfn);
 339                         if (!memblock_is_reserved(paddr))
 340                                 free_highmem_page(page);
 341                 }
 342         }
 343 #endif /* CONFIG_HIGHMEM */
 344
 345 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 346         /*
 347          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 348          * functions.... do it here for the non-smp case.
 349          */
 350         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 351                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 352 #endif
 353
 354         mem_init_print_info(NULL);
 355 #ifdef CONFIG_PPC32
 356         pr_info("Kernel virtual memory layout:\n");
 357         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 358 #ifdef CONFIG_HIGHMEM
 359         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 360                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 361 #endif /* CONFIG_HIGHMEM */
 362 #ifdef CONFIG_NOT_COHERENT_CACHE
 363         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 364                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 365 #endif /* CONFIG_NOT_COHERENT_CACHE */
 366         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 367                 ioremap_bot, IOREMAP_TOP);
 368         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 369                 VMALLOC_START, VMALLOC_END);
 370 #endif /* CONFIG_PPC32 */
 371
 372         mem_init_done = 1;
 373 }
 374
 375 void free_initmem(void)
 376 {
 377         ppc_md.progress = ppc_printk_progress;
 378         free_initmem_default(POISON_FREE_INITMEM);
 379 }
 380
 381 #ifdef CONFIG_BLK_DEV_INITRD
 382 void __init free_initrd_mem(unsigned long start, unsigned long end)
 383 {
 384         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 385 }
 386 #endif
 387
 388 /*
 389  * This is called when a page has been modified by the kernel.
 390  * It just marks the page as not i-cache clean.  We do the i-cache
 391  * flush later when the page is given to a user process, if necessary.
 392  */
 393 void flush_dcache_page(struct page *page)
 394 {
 395         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 396                 return;
 397         /* avoid an atomic op if possible */
 398         if (test_bit(PG_arch_1, &page->flags))
 399                 clear_bit(PG_arch_1, &page->flags);
 400 }
 401 EXPORT_SYMBOL(flush_dcache_page);
 402
 403 void flush_dcache_icache_page(struct page *page)
 404 {
 405 #ifdef CONFIG_HUGETLB_PAGE
 406         if (PageCompound(page)) {
 407                 flush_dcache_icache_hugepage(page);
 408                 return;
 409         }
 410 #endif
 411 #ifdef CONFIG_BOOKE
 412         {
 413                 void *start = kmap_atomic(page);
 414                 __flush_dcache_icache(start);
 415                 kunmap_atomic(start);
 416         }
 417 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 418         /* On 8xx there is no need to kmap since highmem is not supported */
 419         __flush_dcache_icache(page_address(page));
 420 #else
 421         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 422 #endif
 423 }
 424 EXPORT_SYMBOL(flush_dcache_icache_page);
 425
 426 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 427 {
 428         clear_page(page);
 429
 430         /*
 431          * We shouldn't have to do this, but some versions of glibc
 432          * require it (ld.so assumes zero filled pages are icache clean)
 433          * - Anton
 434          */
 435         flush_dcache_page(pg);
 436 }
 437 EXPORT_SYMBOL(clear_user_page);
 438
 439 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 440                     struct page *pg)
 441 {
 442         copy_page(vto, vfrom);
 443
 444         /*
 445          * We should be able to use the following optimisation, however
 446          * there are two problems.
 447          * Firstly a bug in some versions of binutils meant PLT sections
 448          * were not marked executable.
 449          * Secondly the first word in the GOT section is blrl, used
 450          * to establish the GOT address. Until recently the GOT was
 451          * not marked executable.
 452          * - Anton
 453          */
 454 #if 0
 455         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 456                 return;
 457 #endif
 458
 459         flush_dcache_page(pg);
 460 }
 461
 462 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 463                              unsigned long addr, int len)
 464 {
 465         unsigned long maddr;
 466
 467         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 468         flush_icache_range(maddr, maddr + len);
 469         kunmap(page);
 470 }
 471 EXPORT_SYMBOL(flush_icache_user_range);
 472
 473 /*
 474  * This is called at the end of handling a user page fault, when the
 475  * fault has been handled by updating a PTE in the linux page tables.
 476  * We use it to preload an HPTE into the hash table corresponding to
 477  * the updated linux PTE.
 478  *
 479  * This must always be called with the pte lock held.
 480  */
 481 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 482                       pte_t *ptep)
 483 {
 484 #ifdef CONFIG_PPC_STD_MMU
 485         /*
 486          * We don't need to worry about _PAGE_PRESENT here because we are
 487          * called with either mm->page_table_lock held or ptl lock held
 488          */
 489         unsigned long access = 0, trap;
 490
 491         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 492         if (!pte_young(*ptep) || address >= TASK_SIZE)
 493                 return;
 494
 495         /* We try to figure out if we are coming from an instruction
 496          * access fault and pass that down to __hash_page so we avoid
 497          * double-faulting on execution of fresh text. We have to test
 498          * for regs NULL since init will get here first thing at boot
 499          *
 500          * We also avoid filling the hash if not coming from a fault
 501          */
 502         if (current->thread.regs == NULL)
 503                 return;
 504         trap = TRAP(current->thread.regs);
 505         if (trap == 0x400)
 506                 access |= _PAGE_EXEC;
 507         else if (trap != 0x300)
 508                 return;
 509         hash_preload(vma->vm_mm, address, access, trap);
 510 #endif /* CONFIG_PPC_STD_MMU */
 511 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
 512         && defined(CONFIG_HUGETLB_PAGE)
 513         if (is_vm_hugetlb_page(vma))
 514                 book3e_hugetlb_preload(vma, address, *ptep);
 515 #endif
 516 }
 517
 518 /*
 519  * System memory should not be in /proc/iomem but various tools expect it
 520  * (eg kdump).
 521  */
 522 static int __init add_system_ram_resources(void)
 523 {
 524         struct memblock_region *reg;
 525
 526         for_each_memblock(memory, reg) {
 527                 struct resource *res;
 528                 unsigned long base = reg->base;
 529                 unsigned long size = reg->size;
 530
 531                 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 532                 WARN_ON(!res);
 533
 534                 if (res) {
 535                         res->name = "System RAM";
 536                         res->start = base;
 537                         res->end = base + size - 1;
 538                         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 539                         WARN_ON(request_resource(&iomem_resource, res) < 0);
 540                 }
 541         }
 542
 543         return 0;
 544 }
 545 subsys_initcall(add_system_ram_resources);
 546
 547 #ifdef CONFIG_STRICT_DEVMEM
 548 /*
 549  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
 550  * is valid. The argument is a physical page number.
 551  *
 552  * Access has to be given to non-kernel-ram areas as well, these contain the
 553  * PCI mmio resources as well as potential bios/acpi data regions.
 554  */
 555 int devmem_is_allowed(unsigned long pfn)
 556 {
 557         if (iomem_is_exclusive(pfn << PAGE_SHIFT))
 558                 return 0;
 559         if (!page_is_ram(pfn))
 560                 return 1;
 561         if (page_is_rtas_user_buf(pfn))
 562                 return 1;
 563         return 0;
 564 }
 565 #endif /* CONFIG_STRICT_DEVMEM */