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 bool init_mem_is_free;
  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 #define TOP_ZONE ZONE_HIGHMEM
  73
  74 static inline pte_t *virt_to_kpte(unsigned long vaddr)
  75 {
  76         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
  77                         vaddr), vaddr), vaddr);
  78 }
  79 #else
  80 #define TOP_ZONE ZONE_NORMAL
  81 #endif
  82
  83 int page_is_ram(unsigned long pfn)
  84 {
  85 #ifndef CONFIG_PPC64    /* XXX for now */
  86         return pfn < max_pfn;
  87 #else
  88         unsigned long paddr = (pfn << PAGE_SHIFT);
  89         struct memblock_region *reg;
  90
  91         for_each_memblock(memory, reg)
  92                 if (paddr >= reg->base && paddr < (reg->base + reg->size))
  93                         return 1;
  94         return 0;
  95 #endif
  96 }
  97
  98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  99                               unsigned long size, pgprot_t vma_prot)
 100 {
 101         if (ppc_md.phys_mem_access_prot)
 102                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
 103
 104         if (!page_is_ram(pfn))
 105                 vma_prot = pgprot_noncached(vma_prot);
 106
 107         return vma_prot;
 108 }
 109 EXPORT_SYMBOL(phys_mem_access_prot);
 110
 111 #ifdef CONFIG_MEMORY_HOTPLUG
 112
 113 #ifdef CONFIG_NUMA
 114 int memory_add_physaddr_to_nid(u64 start)
 115 {
 116         return hot_add_scn_to_nid(start);
 117 }
 118 #endif
 119
 120 int __weak create_section_mapping(unsigned long start, unsigned long end)
 121 {
 122         return -ENODEV;
 123 }
 124
 125 int __weak remove_section_mapping(unsigned long start, unsigned long end)
 126 {
 127         return -ENODEV;
 128 }
 129
 130 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
 131 {
 132         struct pglist_data *pgdata;
 133         struct zone *zone;
 134         unsigned long start_pfn = start >> PAGE_SHIFT;
 135         unsigned long nr_pages = size >> PAGE_SHIFT;
 136         int rc;
 137
 138         pgdata = NODE_DATA(nid);
 139
 140         start = (unsigned long)__va(start);
 141         rc = create_section_mapping(start, start + size);
 142         if (rc) {
 143                 pr_warning(
 144                         "Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
 145                         start, start + size, rc);
 146                 return -EFAULT;
 147         }
 148
 149         /* this should work for most non-highmem platforms */
 150         zone = pgdata->node_zones +
 151                 zone_for_memory(nid, start, size, 0, for_device);
 152
 153         return __add_pages(nid, zone, start_pfn, nr_pages);
 154 }
 155
 156 #ifdef CONFIG_MEMORY_HOTREMOVE
 157 int arch_remove_memory(u64 start, u64 size)
 158 {
 159         unsigned long start_pfn = start >> PAGE_SHIFT;
 160         unsigned long nr_pages = size >> PAGE_SHIFT;
 161         struct zone *zone;
 162         int ret;
 163
 164         zone = page_zone(pfn_to_page(start_pfn));
 165         ret = __remove_pages(zone, start_pfn, nr_pages);
 166         if (ret)
 167                 return ret;
 168
 169         /* Remove htab bolted mappings for this section of memory */
 170         start = (unsigned long)__va(start);
 171         ret = remove_section_mapping(start, start + size);
 172
 173         /* Ensure all vmalloc mappings are flushed in case they also
 174          * hit that section of memory
 175          */
 176         vm_unmap_aliases();
 177
 178         return ret;
 179 }
 180 #endif
 181 #endif /* CONFIG_MEMORY_HOTPLUG */
 182
 183 /*
 184  * walk_memory_resource() needs to make sure there is no holes in a given
 185  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 186  * Instead it maintains it in memblock.memory structures.  Walk through the
 187  * memory regions, find holes and callback for contiguous regions.
 188  */
 189 int
 190 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 191                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 192 {
 193         struct memblock_region *reg;
 194         unsigned long end_pfn = start_pfn + nr_pages;
 195         unsigned long tstart, tend;
 196         int ret = -1;
 197
 198         for_each_memblock(memory, reg) {
 199                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 200                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 201                 if (tstart >= tend)
 202                         continue;
 203                 ret = (*func)(tstart, tend - tstart, arg);
 204                 if (ret)
 205                         break;
 206         }
 207         return ret;
 208 }
 209 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 210
 211 #ifndef CONFIG_NEED_MULTIPLE_NODES
 212 void __init initmem_init(void)
 213 {
 214         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 215         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 216 #ifdef CONFIG_HIGHMEM
 217         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 218 #endif
 219
 220         /* Place all memblock_regions in the same node and merge contiguous
 221          * memblock_regions
 222          */
 223         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 224
 225         /* XXX need to clip this if using highmem? */
 226         sparse_memory_present_with_active_regions(0);
 227         sparse_init();
 228 }
 229
 230 /* mark pages that don't exist as nosave */
 231 static int __init mark_nonram_nosave(void)
 232 {
 233         struct memblock_region *reg, *prev = NULL;
 234
 235         for_each_memblock(memory, reg) {
 236                 if (prev &&
 237                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 238                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 239                                                memblock_region_memory_base_pfn(reg));
 240                 prev = reg;
 241         }
 242         return 0;
 243 }
 244 #else /* CONFIG_NEED_MULTIPLE_NODES */
 245 static int __init mark_nonram_nosave(void)
 246 {
 247         return 0;
 248 }
 249 #endif
 250
 251 static bool zone_limits_final;
 252
 253 /*
 254  * The memory zones past TOP_ZONE are managed by generic mm code.
 255  * These should be set to zero since that's what every other
 256  * architecture does.
 257  */
 258 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
 259         [0            ... TOP_ZONE        ] = ~0UL,
 260         [TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
 261 };
 262
 263 /*
 264  * Restrict the specified zone and all more restrictive zones
 265  * to be below the specified pfn.  May not be called after
 266  * paging_init().
 267  */
 268 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
 269 {
 270         int i;
 271
 272         if (WARN_ON(zone_limits_final))
 273                 return;
 274
 275         for (i = zone; i >= 0; i--) {
 276                 if (max_zone_pfns[i] > pfn_limit)
 277                         max_zone_pfns[i] = pfn_limit;
 278         }
 279 }
 280
 281 /*
 282  * Find the least restrictive zone that is entirely below the
 283  * specified pfn limit.  Returns < 0 if no suitable zone is found.
 284  *
 285  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
 286  * systems -- the DMA limit can be higher than any possible real pfn.
 287  */
 288 int dma_pfn_limit_to_zone(u64 pfn_limit)
 289 {
 290         int i;
 291
 292         for (i = TOP_ZONE; i >= 0; i--) {
 293                 if (max_zone_pfns[i] <= pfn_limit)
 294                         return i;
 295         }
 296
 297         return -EPERM;
 298 }
 299
 300 /*
 301  * paging_init() sets up the page tables - in fact we've already done this.
 302  */
 303 void __init paging_init(void)
 304 {
 305         unsigned long long total_ram = memblock_phys_mem_size();
 306         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 307
 308 #ifdef CONFIG_PPC32
 309         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 310         unsigned long end = __fix_to_virt(FIX_HOLE);
 311
 312         for (; v < end; v += PAGE_SIZE)
 313                 map_page(v, 0, 0); /* XXX gross */
 314 #endif
 315
 316 #ifdef CONFIG_HIGHMEM
 317         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 318         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 319
 320         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 321         kmap_prot = PAGE_KERNEL;
 322 #endif /* CONFIG_HIGHMEM */
 323
 324         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 325                (unsigned long long)top_of_ram, total_ram);
 326         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 327                (long int)((top_of_ram - total_ram) >> 20));
 328
 329 #ifdef CONFIG_HIGHMEM
 330         limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
 331 #endif
 332         limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
 333         zone_limits_final = true;
 334         free_area_init_nodes(max_zone_pfns);
 335
 336         mark_nonram_nosave();
 337 }
 338
 339 void __init mem_init(void)
 340 {
 341         /*
 342          * book3s is limited to 16 page sizes due to encoding this in
 343          * a 4-bit field for slices.
 344          */
 345         BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
 346
 347 #ifdef CONFIG_SWIOTLB
 348         swiotlb_init(0);
 349 #endif
 350
 351         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 352         set_max_mapnr(max_pfn);
 353         free_all_bootmem();
 354
 355 #ifdef CONFIG_HIGHMEM
 356         {
 357                 unsigned long pfn, highmem_mapnr;
 358
 359                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 360                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 361                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 362                         struct page *page = pfn_to_page(pfn);
 363                         if (!memblock_is_reserved(paddr))
 364                                 free_highmem_page(page);
 365                 }
 366         }
 367 #endif /* CONFIG_HIGHMEM */
 368
 369 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 370         /*
 371          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 372          * functions.... do it here for the non-smp case.
 373          */
 374         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 375                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 376 #endif
 377
 378         mem_init_print_info(NULL);
 379 #ifdef CONFIG_PPC32
 380         pr_info("Kernel virtual memory layout:\n");
 381         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 382 #ifdef CONFIG_HIGHMEM
 383         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 384                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 385 #endif /* CONFIG_HIGHMEM */
 386 #ifdef CONFIG_NOT_COHERENT_CACHE
 387         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 388                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 389 #endif /* CONFIG_NOT_COHERENT_CACHE */
 390         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 391                 ioremap_bot, IOREMAP_TOP);
 392         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 393                 VMALLOC_START, VMALLOC_END);
 394 #endif /* CONFIG_PPC32 */
 395 }
 396
 397 void free_initmem(void)
 398 {
 399         ppc_md.progress = ppc_printk_progress;
 400         init_mem_is_free = true;
 401         free_initmem_default(POISON_FREE_INITMEM);
 402 }
 403
 404 #ifdef CONFIG_BLK_DEV_INITRD
 405 void __init free_initrd_mem(unsigned long start, unsigned long end)
 406 {
 407         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 408 }
 409 #endif
 410
 411 /*
 412  * This is called when a page has been modified by the kernel.
 413  * It just marks the page as not i-cache clean.  We do the i-cache
 414  * flush later when the page is given to a user process, if necessary.
 415  */
 416 void flush_dcache_page(struct page *page)
 417 {
 418         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 419                 return;
 420         /* avoid an atomic op if possible */
 421         if (test_bit(PG_arch_1, &page->flags))
 422                 clear_bit(PG_arch_1, &page->flags);
 423 }
 424 EXPORT_SYMBOL(flush_dcache_page);
 425
 426 void flush_dcache_icache_page(struct page *page)
 427 {
 428 #ifdef CONFIG_HUGETLB_PAGE
 429         if (PageCompound(page)) {
 430                 flush_dcache_icache_hugepage(page);
 431                 return;
 432         }
 433 #endif
 434 #if defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 435         /* On 8xx there is no need to kmap since highmem is not supported */
 436         __flush_dcache_icache(page_address(page));
 437 #else
 438         if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
 439                 void *start = kmap_atomic(page);
 440                 __flush_dcache_icache(start);
 441                 kunmap_atomic(start);
 442         } else {
 443                 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 444         }
 445 #endif
 446 }
 447 EXPORT_SYMBOL(flush_dcache_icache_page);
 448
 449 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 450 {
 451         clear_page(page);
 452
 453         /*
 454          * We shouldn't have to do this, but some versions of glibc
 455          * require it (ld.so assumes zero filled pages are icache clean)
 456          * - Anton
 457          */
 458         flush_dcache_page(pg);
 459 }
 460 EXPORT_SYMBOL(clear_user_page);
 461
 462 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 463                     struct page *pg)
 464 {
 465         copy_page(vto, vfrom);
 466
 467         /*
 468          * We should be able to use the following optimisation, however
 469          * there are two problems.
 470          * Firstly a bug in some versions of binutils meant PLT sections
 471          * were not marked executable.
 472          * Secondly the first word in the GOT section is blrl, used
 473          * to establish the GOT address. Until recently the GOT was
 474          * not marked executable.
 475          * - Anton
 476          */
 477 #if 0
 478         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 479                 return;
 480 #endif
 481
 482         flush_dcache_page(pg);
 483 }
 484
 485 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 486                              unsigned long addr, int len)
 487 {
 488         unsigned long maddr;
 489
 490         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 491         flush_icache_range(maddr, maddr + len);
 492         kunmap(page);
 493 }
 494 EXPORT_SYMBOL(flush_icache_user_range);
 495
 496 /*
 497  * This is called at the end of handling a user page fault, when the
 498  * fault has been handled by updating a PTE in the linux page tables.
 499  * We use it to preload an HPTE into the hash table corresponding to
 500  * the updated linux PTE.
 501  *
 502  * This must always be called with the pte lock held.
 503  */
 504 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 505                       pte_t *ptep)
 506 {
 507 #ifdef CONFIG_PPC_STD_MMU
 508         /*
 509          * We don't need to worry about _PAGE_PRESENT here because we are
 510          * called with either mm->page_table_lock held or ptl lock held
 511          */
 512         unsigned long access, trap;
 513
 514         if (radix_enabled())
 515                 return;
 516
 517         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 518         if (!pte_young(*ptep) || address >= TASK_SIZE)
 519                 return;
 520
 521         /* We try to figure out if we are coming from an instruction
 522          * access fault and pass that down to __hash_page so we avoid
 523          * double-faulting on execution of fresh text. We have to test
 524          * for regs NULL since init will get here first thing at boot
 525          *
 526          * We also avoid filling the hash if not coming from a fault
 527          */
 528
 529         trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
 530         switch (trap) {
 531         case 0x300:
 532                 access = 0UL;
 533                 break;
 534         case 0x400:
 535                 access = _PAGE_EXEC;
 536                 break;
 537         default:
 538                 return;
 539         }
 540
 541         hash_preload(vma->vm_mm, address, access, trap);
 542 #endif /* CONFIG_PPC_STD_MMU */
 543 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
 544         && defined(CONFIG_HUGETLB_PAGE)
 545         if (is_vm_hugetlb_page(vma))
 546                 book3e_hugetlb_preload(vma, address, *ptep);
 547 #endif
 548 }
 549
 550 /*
 551  * System memory should not be in /proc/iomem but various tools expect it
 552  * (eg kdump).
 553  */
 554 static int __init add_system_ram_resources(void)
 555 {
 556         struct memblock_region *reg;
 557
 558         for_each_memblock(memory, reg) {
 559                 struct resource *res;
 560                 unsigned long base = reg->base;
 561                 unsigned long size = reg->size;
 562
 563                 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 564                 WARN_ON(!res);
 565
 566                 if (res) {
 567                         res->name = "System RAM";
 568                         res->start = base;
 569                         res->end = base + size - 1;
 570                         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 571                         WARN_ON(request_resource(&iomem_resource, res) < 0);
 572                 }
 573         }
 574
 575         return 0;
 576 }
 577 subsys_initcall(add_system_ram_resources);
 578
 579 #ifdef CONFIG_STRICT_DEVMEM
 580 /*
 581  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
 582  * is valid. The argument is a physical page number.
 583  *
 584  * Access has to be given to non-kernel-ram areas as well, these contain the
 585  * PCI mmio resources as well as potential bios/acpi data regions.
 586  */
 587 int devmem_is_allowed(unsigned long pfn)
 588 {
 589         if (page_is_rtas_user_buf(pfn))
 590                 return 1;
 591         if (iomem_is_exclusive(PFN_PHYS(pfn)))
 592                 return 0;
 593         if (!page_is_ram(pfn))
 594                 return 1;
 595         return 0;
 596 }
 597 #endif /* CONFIG_STRICT_DEVMEM */