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 #include <linux/vmalloc.h>
  39
  40 #include <asm/pgalloc.h>
  41 #include <asm/prom.h>
  42 #include <asm/io.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/mmu.h>
  46 #include <asm/smp.h>
  47 #include <asm/machdep.h>
  48 #include <asm/btext.h>
  49 #include <asm/tlb.h>
  50 #include <asm/sections.h>
  51 #include <asm/sparsemem.h>
  52 #include <asm/vdso.h>
  53 #include <asm/fixmap.h>
  54 #include <asm/swiotlb.h>
  55 #include <asm/rtas.h>
  56
  57 #include "mmu_decl.h"
  58
  59 #ifndef CPU_FTR_COHERENT_ICACHE
  60 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  61 #define CPU_FTR_NOEXECUTE       0
  62 #endif
  63
  64 unsigned long long memory_limit;
  65
  66 #ifdef CONFIG_HIGHMEM
  67 pte_t *kmap_pte;
  68 EXPORT_SYMBOL(kmap_pte);
  69 pgprot_t kmap_prot;
  70 EXPORT_SYMBOL(kmap_prot);
  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                 zone_for_memory(nid, start, size, 0);
 132
 133         return __add_pages(nid, zone, start_pfn, nr_pages);
 134 }
 135
 136 #ifdef CONFIG_MEMORY_HOTREMOVE
 137 int arch_remove_memory(u64 start, u64 size)
 138 {
 139         unsigned long start_pfn = start >> PAGE_SHIFT;
 140         unsigned long nr_pages = size >> PAGE_SHIFT;
 141         struct zone *zone;
 142         int ret;
 143
 144         zone = page_zone(pfn_to_page(start_pfn));
 145         ret = __remove_pages(zone, start_pfn, nr_pages);
 146         if (ret)
 147                 return ret;
 148
 149         /* Remove htab bolted mappings for this section of memory */
 150         start = (unsigned long)__va(start);
 151         ret = remove_section_mapping(start, start + size);
 152
 153         /* Ensure all vmalloc mappings are flushed in case they also
 154          * hit that section of memory
 155          */
 156         vm_unmap_aliases();
 157
 158         return ret;
 159 }
 160 #endif
 161 #endif /* CONFIG_MEMORY_HOTPLUG */
 162
 163 /*
 164  * walk_memory_resource() needs to make sure there is no holes in a given
 165  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 166  * Instead it maintains it in memblock.memory structures.  Walk through the
 167  * memory regions, find holes and callback for contiguous regions.
 168  */
 169 int
 170 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 171                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 172 {
 173         struct memblock_region *reg;
 174         unsigned long end_pfn = start_pfn + nr_pages;
 175         unsigned long tstart, tend;
 176         int ret = -1;
 177
 178         for_each_memblock(memory, reg) {
 179                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 180                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 181                 if (tstart >= tend)
 182                         continue;
 183                 ret = (*func)(tstart, tend - tstart, arg);
 184                 if (ret)
 185                         break;
 186         }
 187         return ret;
 188 }
 189 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 190
 191 #ifndef CONFIG_NEED_MULTIPLE_NODES
 192 void __init initmem_init(void)
 193 {
 194         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 195         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 196 #ifdef CONFIG_HIGHMEM
 197         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 198 #endif
 199
 200         /* Place all memblock_regions in the same node and merge contiguous
 201          * memblock_regions
 202          */
 203         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 204
 205         /* XXX need to clip this if using highmem? */
 206         sparse_memory_present_with_active_regions(0);
 207         sparse_init();
 208 }
 209
 210 /* mark pages that don't exist as nosave */
 211 static int __init mark_nonram_nosave(void)
 212 {
 213         struct memblock_region *reg, *prev = NULL;
 214
 215         for_each_memblock(memory, reg) {
 216                 if (prev &&
 217                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 218                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 219                                                memblock_region_memory_base_pfn(reg));
 220                 prev = reg;
 221         }
 222         return 0;
 223 }
 224 #else /* CONFIG_NEED_MULTIPLE_NODES */
 225 static int __init mark_nonram_nosave(void)
 226 {
 227         return 0;
 228 }
 229 #endif
 230
 231 static bool zone_limits_final;
 232
 233 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
 234         [0 ... MAX_NR_ZONES - 1] = ~0UL
 235 };
 236
 237 /*
 238  * Restrict the specified zone and all more restrictive zones
 239  * to be below the specified pfn.  May not be called after
 240  * paging_init().
 241  */
 242 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
 243 {
 244         int i;
 245
 246         if (WARN_ON(zone_limits_final))
 247                 return;
 248
 249         for (i = zone; i >= 0; i--) {
 250                 if (max_zone_pfns[i] > pfn_limit)
 251                         max_zone_pfns[i] = pfn_limit;
 252         }
 253 }
 254
 255 /*
 256  * Find the least restrictive zone that is entirely below the
 257  * specified pfn limit.  Returns < 0 if no suitable zone is found.
 258  *
 259  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
 260  * systems -- the DMA limit can be higher than any possible real pfn.
 261  */
 262 int dma_pfn_limit_to_zone(u64 pfn_limit)
 263 {
 264         enum zone_type top_zone = ZONE_NORMAL;
 265         int i;
 266
 267 #ifdef CONFIG_HIGHMEM
 268         top_zone = ZONE_HIGHMEM;
 269 #endif
 270
 271         for (i = top_zone; i >= 0; i--) {
 272                 if (max_zone_pfns[i] <= pfn_limit)
 273                         return i;
 274         }
 275
 276         return -EPERM;
 277 }
 278
 279 /*
 280  * paging_init() sets up the page tables - in fact we've already done this.
 281  */
 282 void __init paging_init(void)
 283 {
 284         unsigned long long total_ram = memblock_phys_mem_size();
 285         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 286         enum zone_type top_zone;
 287
 288 #ifdef CONFIG_PPC32
 289         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 290         unsigned long end = __fix_to_virt(FIX_HOLE);
 291
 292         for (; v < end; v += PAGE_SIZE)
 293                 map_page(v, 0, 0); /* XXX gross */
 294 #endif
 295
 296 #ifdef CONFIG_HIGHMEM
 297         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 298         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 299
 300         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 301         kmap_prot = PAGE_KERNEL;
 302 #endif /* CONFIG_HIGHMEM */
 303
 304         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 305                (unsigned long long)top_of_ram, total_ram);
 306         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 307                (long int)((top_of_ram - total_ram) >> 20));
 308
 309 #ifdef CONFIG_HIGHMEM
 310         top_zone = ZONE_HIGHMEM;
 311         limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
 312 #else
 313         top_zone = ZONE_NORMAL;
 314 #endif
 315
 316         limit_zone_pfn(top_zone, top_of_ram >> PAGE_SHIFT);
 317         zone_limits_final = true;
 318         free_area_init_nodes(max_zone_pfns);
 319
 320         mark_nonram_nosave();
 321 }
 322
 323 void __init mem_init(void)
 324 {
 325         /*
 326          * book3s is limited to 16 page sizes due to encoding this in
 327          * a 4-bit field for slices.
 328          */
 329         BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
 330
 331 #ifdef CONFIG_SWIOTLB
 332         swiotlb_init(0);
 333 #endif
 334
 335         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 336         set_max_mapnr(max_pfn);
 337         free_all_bootmem();
 338
 339 #ifdef CONFIG_HIGHMEM
 340         {
 341                 unsigned long pfn, highmem_mapnr;
 342
 343                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 344                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 345                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 346                         struct page *page = pfn_to_page(pfn);
 347                         if (!memblock_is_reserved(paddr))
 348                                 free_highmem_page(page);
 349                 }
 350         }
 351 #endif /* CONFIG_HIGHMEM */
 352
 353 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 354         /*
 355          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 356          * functions.... do it here for the non-smp case.
 357          */
 358         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 359                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 360 #endif
 361
 362         mem_init_print_info(NULL);
 363 #ifdef CONFIG_PPC32
 364         pr_info("Kernel virtual memory layout:\n");
 365         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 366 #ifdef CONFIG_HIGHMEM
 367         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 368                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 369 #endif /* CONFIG_HIGHMEM */
 370 #ifdef CONFIG_NOT_COHERENT_CACHE
 371         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 372                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 373 #endif /* CONFIG_NOT_COHERENT_CACHE */
 374         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 375                 ioremap_bot, IOREMAP_TOP);
 376         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 377                 VMALLOC_START, VMALLOC_END);
 378 #endif /* CONFIG_PPC32 */
 379 }
 380
 381 void free_initmem(void)
 382 {
 383         ppc_md.progress = ppc_printk_progress;
 384         free_initmem_default(POISON_FREE_INITMEM);
 385 }
 386
 387 #ifdef CONFIG_BLK_DEV_INITRD
 388 void __init free_initrd_mem(unsigned long start, unsigned long end)
 389 {
 390         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 391 }
 392 #endif
 393
 394 /*
 395  * This is called when a page has been modified by the kernel.
 396  * It just marks the page as not i-cache clean.  We do the i-cache
 397  * flush later when the page is given to a user process, if necessary.
 398  */
 399 void flush_dcache_page(struct page *page)
 400 {
 401         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 402                 return;
 403         /* avoid an atomic op if possible */
 404         if (test_bit(PG_arch_1, &page->flags))
 405                 clear_bit(PG_arch_1, &page->flags);
 406 }
 407 EXPORT_SYMBOL(flush_dcache_page);
 408
 409 void flush_dcache_icache_page(struct page *page)
 410 {
 411 #ifdef CONFIG_HUGETLB_PAGE
 412         if (PageCompound(page)) {
 413                 flush_dcache_icache_hugepage(page);
 414                 return;
 415         }
 416 #endif
 417 #ifdef CONFIG_BOOKE
 418         {
 419                 void *start = kmap_atomic(page);
 420                 __flush_dcache_icache(start);
 421                 kunmap_atomic(start);
 422         }
 423 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 424         /* On 8xx there is no need to kmap since highmem is not supported */
 425         __flush_dcache_icache(page_address(page));
 426 #else
 427         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 428 #endif
 429 }
 430 EXPORT_SYMBOL(flush_dcache_icache_page);
 431
 432 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 433 {
 434         clear_page(page);
 435
 436         /*
 437          * We shouldn't 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 *ptep)
 489 {
 490 #ifdef CONFIG_PPC_STD_MMU
 491         /*
 492          * We don't need to worry about _PAGE_PRESENT here because we are
 493          * called with either mm->page_table_lock held or ptl lock held
 494          */
 495         unsigned long access = 0, trap;
 496
 497         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 498         if (!pte_young(*ptep) || address >= TASK_SIZE)
 499                 return;
 500
 501         /* We try to figure out if we are coming from an instruction
 502          * access fault and pass that down to __hash_page so we avoid
 503          * double-faulting on execution of fresh text. We have to test
 504          * for regs NULL since init will get here first thing at boot
 505          *
 506          * We also avoid filling the hash if not coming from a fault
 507          */
 508         if (current->thread.regs == NULL)
 509                 return;
 510         trap = TRAP(current->thread.regs);
 511         if (trap == 0x400)
 512                 access |= _PAGE_EXEC;
 513         else if (trap != 0x300)
 514                 return;
 515         hash_preload(vma->vm_mm, address, access, trap);
 516 #endif /* CONFIG_PPC_STD_MMU */
 517 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
 518         && defined(CONFIG_HUGETLB_PAGE)
 519         if (is_vm_hugetlb_page(vma))
 520                 book3e_hugetlb_preload(vma, address, *ptep);
 521 #endif
 522 }
 523
 524 /*
 525  * System memory should not be in /proc/iomem but various tools expect it
 526  * (eg kdump).
 527  */
 528 static int __init add_system_ram_resources(void)
 529 {
 530         struct memblock_region *reg;
 531
 532         for_each_memblock(memory, reg) {
 533                 struct resource *res;
 534                 unsigned long base = reg->base;
 535                 unsigned long size = reg->size;
 536
 537                 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 538                 WARN_ON(!res);
 539
 540                 if (res) {
 541                         res->name = "System RAM";
 542                         res->start = base;
 543                         res->end = base + size - 1;
 544                         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 545                         WARN_ON(request_resource(&iomem_resource, res) < 0);
 546                 }
 547         }
 548
 549         return 0;
 550 }
 551 subsys_initcall(add_system_ram_resources);
 552
 553 #ifdef CONFIG_STRICT_DEVMEM
 554 /*
 555  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
 556  * is valid. The argument is a physical page number.
 557  *
 558  * Access has to be given to non-kernel-ram areas as well, these contain the
 559  * PCI mmio resources as well as potential bios/acpi data regions.
 560  */
 561 int devmem_is_allowed(unsigned long pfn)
 562 {
 563         if (iomem_is_exclusive(pfn << PAGE_SHIFT))
 564                 return 0;
 565         if (!page_is_ram(pfn))
 566                 return 1;
 567         if (page_is_rtas_user_buf(pfn))
 568                 return 1;
 569         return 0;
 570 }
 571 #endif /* CONFIG_STRICT_DEVMEM */