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 #include <linux/memremap.h>
  40
  41 #include <asm/pgalloc.h>
  42 #include <asm/prom.h>
  43 #include <asm/io.h>
  44 #include <asm/mmu_context.h>
  45 #include <asm/pgtable.h>
  46 #include <asm/mmu.h>
  47 #include <asm/smp.h>
  48 #include <asm/machdep.h>
  49 #include <asm/btext.h>
  50 #include <asm/tlb.h>
  51 #include <asm/sections.h>
  52 #include <asm/sparsemem.h>
  53 #include <asm/vdso.h>
  54 #include <asm/fixmap.h>
  55 #include <asm/swiotlb.h>
  56 #include <asm/rtas.h>
  57
  58 #include "mmu_decl.h"
  59
  60 #ifndef CPU_FTR_COHERENT_ICACHE
  61 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  62 #define CPU_FTR_NOEXECUTE       0
  63 #endif
  64
  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 #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, struct vmem_altmap *altmap,
 131                 bool want_memblock)
 132 {
 133         unsigned long start_pfn = start >> PAGE_SHIFT;
 134         unsigned long nr_pages = size >> PAGE_SHIFT;
 135         int rc;
 136
 137         resize_hpt_for_hotplug(memblock_phys_mem_size());
 138
 139         start = (unsigned long)__va(start);
 140         rc = create_section_mapping(start, start + size);
 141         if (rc) {
 142                 pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
 143                         start, start + size, rc);
 144                 return -EFAULT;
 145         }
 146
 147         return __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
 148 }
 149
 150 #ifdef CONFIG_MEMORY_HOTREMOVE
 151 int arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
 152 {
 153         unsigned long start_pfn = start >> PAGE_SHIFT;
 154         unsigned long nr_pages = size >> PAGE_SHIFT;
 155         struct page *page;
 156         int ret;
 157
 158         /*
 159          * If we have an altmap then we need to skip over any reserved PFNs
 160          * when querying the zone.
 161          */
 162         page = pfn_to_page(start_pfn);
 163         if (altmap)
 164                 page += vmem_altmap_offset(altmap);
 165
 166         ret = __remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
 167         if (ret)
 168                 return ret;
 169
 170         /* Remove htab bolted mappings for this section of memory */
 171         start = (unsigned long)__va(start);
 172         ret = remove_section_mapping(start, start + size);
 173
 174         /* Ensure all vmalloc mappings are flushed in case they also
 175          * hit that section of memory
 176          */
 177         vm_unmap_aliases();
 178
 179         resize_hpt_for_hotplug(memblock_phys_mem_size());
 180
 181         return ret;
 182 }
 183 #endif
 184 #endif /* CONFIG_MEMORY_HOTPLUG */
 185
 186 /*
 187  * walk_memory_resource() needs to make sure there is no holes in a given
 188  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
 189  * Instead it maintains it in memblock.memory structures.  Walk through the
 190  * memory regions, find holes and callback for contiguous regions.
 191  */
 192 int
 193 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
 194                 void *arg, int (*func)(unsigned long, unsigned long, void *))
 195 {
 196         struct memblock_region *reg;
 197         unsigned long end_pfn = start_pfn + nr_pages;
 198         unsigned long tstart, tend;
 199         int ret = -1;
 200
 201         for_each_memblock(memory, reg) {
 202                 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
 203                 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
 204                 if (tstart >= tend)
 205                         continue;
 206                 ret = (*func)(tstart, tend - tstart, arg);
 207                 if (ret)
 208                         break;
 209         }
 210         return ret;
 211 }
 212 EXPORT_SYMBOL_GPL(walk_system_ram_range);
 213
 214 #ifndef CONFIG_NEED_MULTIPLE_NODES
 215 void __init initmem_init(void)
 216 {
 217         max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
 218         min_low_pfn = MEMORY_START >> PAGE_SHIFT;
 219 #ifdef CONFIG_HIGHMEM
 220         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 221 #endif
 222
 223         /* Place all memblock_regions in the same node and merge contiguous
 224          * memblock_regions
 225          */
 226         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 227
 228         /* XXX need to clip this if using highmem? */
 229         sparse_memory_present_with_active_regions(0);
 230         sparse_init();
 231 }
 232
 233 /* mark pages that don't exist as nosave */
 234 static int __init mark_nonram_nosave(void)
 235 {
 236         struct memblock_region *reg, *prev = NULL;
 237
 238         for_each_memblock(memory, reg) {
 239                 if (prev &&
 240                     memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
 241                         register_nosave_region(memblock_region_memory_end_pfn(prev),
 242                                                memblock_region_memory_base_pfn(reg));
 243                 prev = reg;
 244         }
 245         return 0;
 246 }
 247 #else /* CONFIG_NEED_MULTIPLE_NODES */
 248 static int __init mark_nonram_nosave(void)
 249 {
 250         return 0;
 251 }
 252 #endif
 253
 254 static bool zone_limits_final;
 255
 256 /*
 257  * The memory zones past TOP_ZONE are managed by generic mm code.
 258  * These should be set to zero since that's what every other
 259  * architecture does.
 260  */
 261 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
 262         [0            ... TOP_ZONE        ] = ~0UL,
 263         [TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
 264 };
 265
 266 /*
 267  * Restrict the specified zone and all more restrictive zones
 268  * to be below the specified pfn.  May not be called after
 269  * paging_init().
 270  */
 271 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
 272 {
 273         int i;
 274
 275         if (WARN_ON(zone_limits_final))
 276                 return;
 277
 278         for (i = zone; i >= 0; i--) {
 279                 if (max_zone_pfns[i] > pfn_limit)
 280                         max_zone_pfns[i] = pfn_limit;
 281         }
 282 }
 283
 284 /*
 285  * Find the least restrictive zone that is entirely below the
 286  * specified pfn limit.  Returns < 0 if no suitable zone is found.
 287  *
 288  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
 289  * systems -- the DMA limit can be higher than any possible real pfn.
 290  */
 291 int dma_pfn_limit_to_zone(u64 pfn_limit)
 292 {
 293         int i;
 294
 295         for (i = TOP_ZONE; i >= 0; i--) {
 296                 if (max_zone_pfns[i] <= pfn_limit)
 297                         return i;
 298         }
 299
 300         return -EPERM;
 301 }
 302
 303 /*
 304  * paging_init() sets up the page tables - in fact we've already done this.
 305  */
 306 void __init paging_init(void)
 307 {
 308         unsigned long long total_ram = memblock_phys_mem_size();
 309         phys_addr_t top_of_ram = memblock_end_of_DRAM();
 310
 311 #ifdef CONFIG_PPC32
 312         unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
 313         unsigned long end = __fix_to_virt(FIX_HOLE);
 314
 315         for (; v < end; v += PAGE_SIZE)
 316                 map_kernel_page(v, 0, 0); /* XXX gross */
 317 #endif
 318
 319 #ifdef CONFIG_HIGHMEM
 320         map_kernel_page(PKMAP_BASE, 0, 0);      /* XXX gross */
 321         pkmap_page_table = virt_to_kpte(PKMAP_BASE);
 322
 323         kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
 324         kmap_prot = PAGE_KERNEL;
 325 #endif /* CONFIG_HIGHMEM */
 326
 327         printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
 328                (unsigned long long)top_of_ram, total_ram);
 329         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 330                (long int)((top_of_ram - total_ram) >> 20));
 331
 332 #ifdef CONFIG_HIGHMEM
 333         limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
 334 #endif
 335         limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
 336         zone_limits_final = true;
 337         free_area_init_nodes(max_zone_pfns);
 338
 339         mark_nonram_nosave();
 340 }
 341
 342 void __init mem_init(void)
 343 {
 344         /*
 345          * book3s is limited to 16 page sizes due to encoding this in
 346          * a 4-bit field for slices.
 347          */
 348         BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
 349
 350 #ifdef CONFIG_SWIOTLB
 351         swiotlb_init(0);
 352 #endif
 353
 354         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 355         set_max_mapnr(max_pfn);
 356         free_all_bootmem();
 357
 358 #ifdef CONFIG_HIGHMEM
 359         {
 360                 unsigned long pfn, highmem_mapnr;
 361
 362                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 363                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 364                         phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
 365                         struct page *page = pfn_to_page(pfn);
 366                         if (!memblock_is_reserved(paddr))
 367                                 free_highmem_page(page);
 368                 }
 369         }
 370 #endif /* CONFIG_HIGHMEM */
 371
 372 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
 373         /*
 374          * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
 375          * functions.... do it here for the non-smp case.
 376          */
 377         per_cpu(next_tlbcam_idx, smp_processor_id()) =
 378                 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
 379 #endif
 380
 381         mem_init_print_info(NULL);
 382 #ifdef CONFIG_PPC32
 383         pr_info("Kernel virtual memory layout:\n");
 384         pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
 385 #ifdef CONFIG_HIGHMEM
 386         pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
 387                 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
 388 #endif /* CONFIG_HIGHMEM */
 389 #ifdef CONFIG_NOT_COHERENT_CACHE
 390         pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 391                 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
 392 #endif /* CONFIG_NOT_COHERENT_CACHE */
 393         pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 394                 ioremap_bot, IOREMAP_TOP);
 395         pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 396                 VMALLOC_START, VMALLOC_END);
 397 #endif /* CONFIG_PPC32 */
 398 }
 399
 400 void free_initmem(void)
 401 {
 402         ppc_md.progress = ppc_printk_progress;
 403         mark_initmem_nx();
 404         free_initmem_default(POISON_FREE_INITMEM);
 405 }
 406
 407 #ifdef CONFIG_BLK_DEV_INITRD
 408 void __init free_initrd_mem(unsigned long start, unsigned long end)
 409 {
 410         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 411 }
 412 #endif
 413
 414 /*
 415  * This is called when a page has been modified by the kernel.
 416  * It just marks the page as not i-cache clean.  We do the i-cache
 417  * flush later when the page is given to a user process, if necessary.
 418  */
 419 void flush_dcache_page(struct page *page)
 420 {
 421         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 422                 return;
 423         /* avoid an atomic op if possible */
 424         if (test_bit(PG_arch_1, &page->flags))
 425                 clear_bit(PG_arch_1, &page->flags);
 426 }
 427 EXPORT_SYMBOL(flush_dcache_page);
 428
 429 void flush_dcache_icache_page(struct page *page)
 430 {
 431 #ifdef CONFIG_HUGETLB_PAGE
 432         if (PageCompound(page)) {
 433                 flush_dcache_icache_hugepage(page);
 434                 return;
 435         }
 436 #endif
 437 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
 438         /* On 8xx there is no need to kmap since highmem is not supported */
 439         __flush_dcache_icache(page_address(page));
 440 #else
 441         if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
 442                 void *start = kmap_atomic(page);
 443                 __flush_dcache_icache(start);
 444                 kunmap_atomic(start);
 445         } else {
 446                 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 447         }
 448 #endif
 449 }
 450 EXPORT_SYMBOL(flush_dcache_icache_page);
 451
 452 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 453 {
 454         clear_page(page);
 455
 456         /*
 457          * We shouldn't have to do this, but some versions of glibc
 458          * require it (ld.so assumes zero filled pages are icache clean)
 459          * - Anton
 460          */
 461         flush_dcache_page(pg);
 462 }
 463 EXPORT_SYMBOL(clear_user_page);
 464
 465 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 466                     struct page *pg)
 467 {
 468         copy_page(vto, vfrom);
 469
 470         /*
 471          * We should be able to use the following optimisation, however
 472          * there are two problems.
 473          * Firstly a bug in some versions of binutils meant PLT sections
 474          * were not marked executable.
 475          * Secondly the first word in the GOT section is blrl, used
 476          * to establish the GOT address. Until recently the GOT was
 477          * not marked executable.
 478          * - Anton
 479          */
 480 #if 0
 481         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 482                 return;
 483 #endif
 484
 485         flush_dcache_page(pg);
 486 }
 487
 488 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 489                              unsigned long addr, int len)
 490 {
 491         unsigned long maddr;
 492
 493         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 494         flush_icache_range(maddr, maddr + len);
 495         kunmap(page);
 496 }
 497 EXPORT_SYMBOL(flush_icache_user_range);
 498
 499 /*
 500  * This is called at the end of handling a user page fault, when the
 501  * fault has been handled by updating a PTE in the linux page tables.
 502  * We use it to preload an HPTE into the hash table corresponding to
 503  * the updated linux PTE.
 504  *
 505  * This must always be called with the pte lock held.
 506  */
 507 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 508                       pte_t *ptep)
 509 {
 510 #ifdef CONFIG_PPC_STD_MMU
 511         /*
 512          * We don't need to worry about _PAGE_PRESENT here because we are
 513          * called with either mm->page_table_lock held or ptl lock held
 514          */
 515         unsigned long access, trap;
 516
 517         if (radix_enabled())
 518                 return;
 519
 520         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 521         if (!pte_young(*ptep) || address >= TASK_SIZE)
 522                 return;
 523
 524         /* We try to figure out if we are coming from an instruction
 525          * access fault and pass that down to __hash_page so we avoid
 526          * double-faulting on execution of fresh text. We have to test
 527          * for regs NULL since init will get here first thing at boot
 528          *
 529          * We also avoid filling the hash if not coming from a fault
 530          */
 531
 532         trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
 533         switch (trap) {
 534         case 0x300:
 535                 access = 0UL;
 536                 break;
 537         case 0x400:
 538                 access = _PAGE_EXEC;
 539                 break;
 540         default:
 541                 return;
 542         }
 543
 544         hash_preload(vma->vm_mm, address, access, trap);
 545 #endif /* CONFIG_PPC_STD_MMU */
 546 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
 547         && defined(CONFIG_HUGETLB_PAGE)
 548         if (is_vm_hugetlb_page(vma))
 549                 book3e_hugetlb_preload(vma, address, *ptep);
 550 #endif
 551 }
 552
 553 /*
 554  * System memory should not be in /proc/iomem but various tools expect it
 555  * (eg kdump).
 556  */
 557 static int __init add_system_ram_resources(void)
 558 {
 559         struct memblock_region *reg;
 560
 561         for_each_memblock(memory, reg) {
 562                 struct resource *res;
 563                 unsigned long base = reg->base;
 564                 unsigned long size = reg->size;
 565
 566                 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 567                 WARN_ON(!res);
 568
 569                 if (res) {
 570                         res->name = "System RAM";
 571                         res->start = base;
 572                         res->end = base + size - 1;
 573                         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 574                         WARN_ON(request_resource(&iomem_resource, res) < 0);
 575                 }
 576         }
 577
 578         return 0;
 579 }
 580 subsys_initcall(add_system_ram_resources);
 581
 582 #ifdef CONFIG_STRICT_DEVMEM
 583 /*
 584  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
 585  * is valid. The argument is a physical page number.
 586  *
 587  * Access has to be given to non-kernel-ram areas as well, these contain the
 588  * PCI mmio resources as well as potential bios/acpi data regions.
 589  */
 590 int devmem_is_allowed(unsigned long pfn)
 591 {
 592         if (page_is_rtas_user_buf(pfn))
 593                 return 1;
 594         if (iomem_is_exclusive(PFN_PHYS(pfn)))
 595                 return 0;
 596         if (!page_is_ram(pfn))
 597                 return 1;
 598         return 0;
 599 }
 600 #endif /* CONFIG_STRICT_DEVMEM */