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/module.h>
  21 #include <linux/sched.h>
  22 #include <linux/kernel.h>
  23 #include <linux/errno.h>
  24 #include <linux/string.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/stddef.h>
  28 #include <linux/init.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/highmem.h>
  31 #include <linux/initrd.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/suspend.h>
  34 #include <linux/lmb.h>
  35
  36 #include <asm/pgalloc.h>
  37 #include <asm/prom.h>
  38 #include <asm/io.h>
  39 #include <asm/mmu_context.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/mmu.h>
  42 #include <asm/smp.h>
  43 #include <asm/machdep.h>
  44 #include <asm/btext.h>
  45 #include <asm/tlb.h>
  46 #include <asm/sections.h>
  47 #include <asm/vdso.h>
  48
  49 #include "mmu_decl.h"
  50
  51 #ifndef CPU_FTR_COHERENT_ICACHE
  52 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
  53 #define CPU_FTR_NOEXECUTE       0
  54 #endif
  55
  56 int init_bootmem_done;
  57 int mem_init_done;
  58 unsigned long memory_limit;
  59
  60 int page_is_ram(unsigned long pfn)
  61 {
  62         unsigned long paddr = (pfn << PAGE_SHIFT);
  63
  64 #ifndef CONFIG_PPC64    /* XXX for now */
  65         return paddr < __pa(high_memory);
  66 #else
  67         int i;
  68         for (i=0; i < lmb.memory.cnt; i++) {
  69                 unsigned long base;
  70
  71                 base = lmb.memory.region[i].base;
  72
  73                 if ((paddr >= base) &&
  74                         (paddr < (base + lmb.memory.region[i].size))) {
  75                         return 1;
  76                 }
  77         }
  78
  79         return 0;
  80 #endif
  81 }
  82
  83 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
  84                               unsigned long size, pgprot_t vma_prot)
  85 {
  86         if (ppc_md.phys_mem_access_prot)
  87                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
  88
  89         if (!page_is_ram(pfn))
  90                 vma_prot = __pgprot(pgprot_val(vma_prot)
  91                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
  92         return vma_prot;
  93 }
  94 EXPORT_SYMBOL(phys_mem_access_prot);
  95
  96 #ifdef CONFIG_MEMORY_HOTPLUG
  97
  98 void online_page(struct page *page)
  99 {
 100         ClearPageReserved(page);
 101         init_page_count(page);
 102         __free_page(page);
 103         totalram_pages++;
 104         num_physpages++;
 105 }
 106
 107 #ifdef CONFIG_NUMA
 108 int memory_add_physaddr_to_nid(u64 start)
 109 {
 110         return hot_add_scn_to_nid(start);
 111 }
 112 #endif
 113
 114 int arch_add_memory(int nid, u64 start, u64 size)
 115 {
 116         struct pglist_data *pgdata;
 117         struct zone *zone;
 118         unsigned long start_pfn = start >> PAGE_SHIFT;
 119         unsigned long nr_pages = size >> PAGE_SHIFT;
 120
 121         pgdata = NODE_DATA(nid);
 122
 123         start = (unsigned long)__va(start);
 124         create_section_mapping(start, start + size);
 125
 126         /* this should work for most non-highmem platforms */
 127         zone = pgdata->node_zones;
 128
 129         return __add_pages(zone, start_pfn, nr_pages);
 130 }
 131
 132 #ifdef CONFIG_MEMORY_HOTREMOVE
 133 int remove_memory(u64 start, u64 size)
 134 {
 135         unsigned long start_pfn, end_pfn;
 136         int ret;
 137
 138         start_pfn = start >> PAGE_SHIFT;
 139         end_pfn = start_pfn + (size >> PAGE_SHIFT);
 140         ret = offline_pages(start_pfn, end_pfn, 120 * HZ);
 141         if (ret)
 142                 goto out;
 143         /* Arch-specific calls go here - next patch */
 144 out:
 145         return ret;
 146 }
 147 #endif /* CONFIG_MEMORY_HOTREMOVE */
 148
 149 /*
 150  * walk_memory_resource() needs to make sure there is no holes in a given
 151  * memory range. On PPC64, since this range comes from /sysfs, the range
 152  * is guaranteed to be valid, non-overlapping and can not contain any
 153  * holes. By the time we get here (memory add or remove), /proc/device-tree
 154  * is updated and correct. Only reason we need to check against device-tree
 155  * would be if we allow user-land to specify a memory range through a
 156  * system call/ioctl etc. instead of doing offline/online through /sysfs.
 157  */
 158 int
 159 walk_memory_resource(unsigned long start_pfn, unsigned long nr_pages, void *arg,
 160                         int (*func)(unsigned long, unsigned long, void *))
 161 {
 162         return  (*func)(start_pfn, nr_pages, arg);
 163 }
 164
 165 #endif /* CONFIG_MEMORY_HOTPLUG */
 166
 167 void show_mem(void)
 168 {
 169         unsigned long total = 0, reserved = 0;
 170         unsigned long shared = 0, cached = 0;
 171         unsigned long highmem = 0;
 172         struct page *page;
 173         pg_data_t *pgdat;
 174         unsigned long i;
 175
 176         printk("Mem-info:\n");
 177         show_free_areas();
 178         for_each_online_pgdat(pgdat) {
 179                 unsigned long flags;
 180                 pgdat_resize_lock(pgdat, &flags);
 181                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 182                         if (!pfn_valid(pgdat->node_start_pfn + i))
 183                                 continue;
 184                         page = pgdat_page_nr(pgdat, i);
 185                         total++;
 186                         if (PageHighMem(page))
 187                                 highmem++;
 188                         if (PageReserved(page))
 189                                 reserved++;
 190                         else if (PageSwapCache(page))
 191                                 cached++;
 192                         else if (page_count(page))
 193                                 shared += page_count(page) - 1;
 194                 }
 195                 pgdat_resize_unlock(pgdat, &flags);
 196         }
 197         printk("%ld pages of RAM\n", total);
 198 #ifdef CONFIG_HIGHMEM
 199         printk("%ld pages of HIGHMEM\n", highmem);
 200 #endif
 201         printk("%ld reserved pages\n", reserved);
 202         printk("%ld pages shared\n", shared);
 203         printk("%ld pages swap cached\n", cached);
 204 }
 205
 206 /*
 207  * Initialize the bootmem system and give it all the memory we
 208  * have available.  If we are using highmem, we only put the
 209  * lowmem into the bootmem system.
 210  */
 211 #ifndef CONFIG_NEED_MULTIPLE_NODES
 212 void __init do_init_bootmem(void)
 213 {
 214         unsigned long i;
 215         unsigned long start, bootmap_pages;
 216         unsigned long total_pages;
 217         int boot_mapsize;
 218
 219         max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
 220         total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
 221 #ifdef CONFIG_HIGHMEM
 222         total_pages = total_lowmem >> PAGE_SHIFT;
 223         max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
 224 #endif
 225
 226         /*
 227          * Find an area to use for the bootmem bitmap.  Calculate the size of
 228          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 229          * Add 1 additional page in case the address isn't page-aligned.
 230          */
 231         bootmap_pages = bootmem_bootmap_pages(total_pages);
 232
 233         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
 234
 235         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 236
 237         /* Add active regions with valid PFNs */
 238         for (i = 0; i < lmb.memory.cnt; i++) {
 239                 unsigned long start_pfn, end_pfn;
 240                 start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
 241                 end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
 242                 add_active_range(0, start_pfn, end_pfn);
 243         }
 244
 245         /* Add all physical memory to the bootmem map, mark each area
 246          * present.
 247          */
 248 #ifdef CONFIG_HIGHMEM
 249         free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
 250
 251         /* reserve the sections we're already using */
 252         for (i = 0; i < lmb.reserved.cnt; i++) {
 253                 unsigned long addr = lmb.reserved.region[i].base +
 254                                      lmb_size_bytes(&lmb.reserved, i) - 1;
 255                 if (addr < lowmem_end_addr)
 256                         reserve_bootmem(lmb.reserved.region[i].base,
 257                                         lmb_size_bytes(&lmb.reserved, i),
 258                                         BOOTMEM_DEFAULT);
 259                 else if (lmb.reserved.region[i].base < lowmem_end_addr) {
 260                         unsigned long adjusted_size = lowmem_end_addr -
 261                                       lmb.reserved.region[i].base;
 262                         reserve_bootmem(lmb.reserved.region[i].base,
 263                                         adjusted_size, BOOTMEM_DEFAULT);
 264                 }
 265         }
 266 #else
 267         free_bootmem_with_active_regions(0, max_pfn);
 268
 269         /* reserve the sections we're already using */
 270         for (i = 0; i < lmb.reserved.cnt; i++)
 271                 reserve_bootmem(lmb.reserved.region[i].base,
 272                                 lmb_size_bytes(&lmb.reserved, i),
 273                                 BOOTMEM_DEFAULT);
 274
 275 #endif
 276         /* XXX need to clip this if using highmem? */
 277         sparse_memory_present_with_active_regions(0);
 278
 279         init_bootmem_done = 1;
 280 }
 281
 282 /* mark pages that don't exist as nosave */
 283 static int __init mark_nonram_nosave(void)
 284 {
 285         unsigned long lmb_next_region_start_pfn,
 286                       lmb_region_max_pfn;
 287         int i;
 288
 289         for (i = 0; i < lmb.memory.cnt - 1; i++) {
 290                 lmb_region_max_pfn =
 291                         (lmb.memory.region[i].base >> PAGE_SHIFT) +
 292                         (lmb.memory.region[i].size >> PAGE_SHIFT);
 293                 lmb_next_region_start_pfn =
 294                         lmb.memory.region[i+1].base >> PAGE_SHIFT;
 295
 296                 if (lmb_region_max_pfn < lmb_next_region_start_pfn)
 297                         register_nosave_region(lmb_region_max_pfn,
 298                                                lmb_next_region_start_pfn);
 299         }
 300
 301         return 0;
 302 }
 303
 304 /*
 305  * paging_init() sets up the page tables - in fact we've already done this.
 306  */
 307 void __init paging_init(void)
 308 {
 309         unsigned long total_ram = lmb_phys_mem_size();
 310         unsigned long top_of_ram = lmb_end_of_DRAM();
 311         unsigned long max_zone_pfns[MAX_NR_ZONES];
 312
 313 #ifdef CONFIG_HIGHMEM
 314         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
 315         pkmap_page_table = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
 316                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 317         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
 318         kmap_pte = pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k
 319                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN),
 320                          KMAP_FIX_BEGIN);
 321         kmap_prot = PAGE_KERNEL;
 322 #endif /* CONFIG_HIGHMEM */
 323
 324         printk(KERN_DEBUG "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 325                top_of_ram, total_ram);
 326         printk(KERN_DEBUG "Memory hole size: %ldMB\n",
 327                (top_of_ram - total_ram) >> 20);
 328         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 329 #ifdef CONFIG_HIGHMEM
 330         max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
 331         max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
 332 #else
 333         max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 334 #endif
 335         free_area_init_nodes(max_zone_pfns);
 336
 337         mark_nonram_nosave();
 338 }
 339 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 340
 341 void __init mem_init(void)
 342 {
 343 #ifdef CONFIG_NEED_MULTIPLE_NODES
 344         int nid;
 345 #endif
 346         pg_data_t *pgdat;
 347         unsigned long i;
 348         struct page *page;
 349         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 350
 351         num_physpages = lmb.memory.size >> PAGE_SHIFT;
 352         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 353
 354 #ifdef CONFIG_NEED_MULTIPLE_NODES
 355         for_each_online_node(nid) {
 356                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 357                         printk("freeing bootmem node %d\n", nid);
 358                         totalram_pages +=
 359                                 free_all_bootmem_node(NODE_DATA(nid));
 360                 }
 361         }
 362 #else
 363         max_mapnr = max_pfn;
 364         totalram_pages += free_all_bootmem();
 365 #endif
 366         for_each_online_pgdat(pgdat) {
 367                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 368                         if (!pfn_valid(pgdat->node_start_pfn + i))
 369                                 continue;
 370                         page = pgdat_page_nr(pgdat, i);
 371                         if (PageReserved(page))
 372                                 reservedpages++;
 373                 }
 374         }
 375
 376         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
 377         datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
 378         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 379         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 380
 381 #ifdef CONFIG_HIGHMEM
 382         {
 383                 unsigned long pfn, highmem_mapnr;
 384
 385                 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
 386                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
 387                         struct page *page = pfn_to_page(pfn);
 388                         if (lmb_is_reserved(pfn << PAGE_SHIFT))
 389                                 continue;
 390                         ClearPageReserved(page);
 391                         init_page_count(page);
 392                         __free_page(page);
 393                         totalhigh_pages++;
 394                         reservedpages--;
 395                 }
 396                 totalram_pages += totalhigh_pages;
 397                 printk(KERN_DEBUG "High memory: %luk\n",
 398                        totalhigh_pages << (PAGE_SHIFT-10));
 399         }
 400 #endif /* CONFIG_HIGHMEM */
 401
 402         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 403                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 404                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 405                 num_physpages << (PAGE_SHIFT-10),
 406                 codesize >> 10,
 407                 reservedpages << (PAGE_SHIFT-10),
 408                 datasize >> 10,
 409                 bsssize >> 10,
 410                 initsize >> 10);
 411
 412         mem_init_done = 1;
 413 }
 414
 415 /*
 416  * This is called when a page has been modified by the kernel.
 417  * It just marks the page as not i-cache clean.  We do the i-cache
 418  * flush later when the page is given to a user process, if necessary.
 419  */
 420 void flush_dcache_page(struct page *page)
 421 {
 422         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 423                 return;
 424         /* avoid an atomic op if possible */
 425         if (test_bit(PG_arch_1, &page->flags))
 426                 clear_bit(PG_arch_1, &page->flags);
 427 }
 428 EXPORT_SYMBOL(flush_dcache_page);
 429
 430 void flush_dcache_icache_page(struct page *page)
 431 {
 432 #ifdef CONFIG_BOOKE
 433         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
 434         __flush_dcache_icache(start);
 435         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
 436 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
 437         /* On 8xx there is no need to kmap since highmem is not supported */
 438         __flush_dcache_icache(page_address(page));
 439 #else
 440         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
 441 #endif
 442
 443 }
 444 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 445 {
 446         clear_page(page);
 447
 448         /*
 449          * We shouldnt have to do this, but some versions of glibc
 450          * require it (ld.so assumes zero filled pages are icache clean)
 451          * - Anton
 452          */
 453         flush_dcache_page(pg);
 454 }
 455 EXPORT_SYMBOL(clear_user_page);
 456
 457 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 458                     struct page *pg)
 459 {
 460         copy_page(vto, vfrom);
 461
 462         /*
 463          * We should be able to use the following optimisation, however
 464          * there are two problems.
 465          * Firstly a bug in some versions of binutils meant PLT sections
 466          * were not marked executable.
 467          * Secondly the first word in the GOT section is blrl, used
 468          * to establish the GOT address. Until recently the GOT was
 469          * not marked executable.
 470          * - Anton
 471          */
 472 #if 0
 473         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 474                 return;
 475 #endif
 476
 477         flush_dcache_page(pg);
 478 }
 479
 480 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 481                              unsigned long addr, int len)
 482 {
 483         unsigned long maddr;
 484
 485         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
 486         flush_icache_range(maddr, maddr + len);
 487         kunmap(page);
 488 }
 489 EXPORT_SYMBOL(flush_icache_user_range);
 490
 491 /*
 492  * This is called at the end of handling a user page fault, when the
 493  * fault has been handled by updating a PTE in the linux page tables.
 494  * We use it to preload an HPTE into the hash table corresponding to
 495  * the updated linux PTE.
 496  *
 497  * This must always be called with the pte lock held.
 498  */
 499 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 500                       pte_t pte)
 501 {
 502 #ifdef CONFIG_PPC_STD_MMU
 503         unsigned long access = 0, trap;
 504 #endif
 505         unsigned long pfn = pte_pfn(pte);
 506
 507         /* handle i-cache coherency */
 508         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 509             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
 510             pfn_valid(pfn)) {
 511                 struct page *page = pfn_to_page(pfn);
 512 #ifdef CONFIG_8xx
 513                 /* On 8xx, cache control instructions (particularly
 514                  * "dcbst" from flush_dcache_icache) fault as write
 515                  * operation if there is an unpopulated TLB entry
 516                  * for the address in question. To workaround that,
 517                  * we invalidate the TLB here, thus avoiding dcbst
 518                  * misbehaviour.
 519                  */
 520                 _tlbie(address, 0 /* 8xx doesn't care about PID */);
 521 #endif
 522                 /* The _PAGE_USER test should really be _PAGE_EXEC, but
 523                  * older glibc versions execute some code from no-exec
 524                  * pages, which for now we are supporting.  If exec-only
 525                  * pages are ever implemented, this will have to change.
 526                  */
 527                 if (!PageReserved(page) && (pte_val(pte) & _PAGE_USER)
 528                     && !test_bit(PG_arch_1, &page->flags)) {
 529                         if (vma->vm_mm == current->active_mm) {
 530                                 __flush_dcache_icache((void *) address);
 531                         } else
 532                                 flush_dcache_icache_page(page);
 533                         set_bit(PG_arch_1, &page->flags);
 534                 }
 535         }
 536
 537 #ifdef CONFIG_PPC_STD_MMU
 538         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 539         if (!pte_young(pte) || address >= TASK_SIZE)
 540                 return;
 541
 542         /* We try to figure out if we are coming from an instruction
 543          * access fault and pass that down to __hash_page so we avoid
 544          * double-faulting on execution of fresh text. We have to test
 545          * for regs NULL since init will get here first thing at boot
 546          *
 547          * We also avoid filling the hash if not coming from a fault
 548          */
 549         if (current->thread.regs == NULL)
 550                 return;
 551         trap = TRAP(current->thread.regs);
 552         if (trap == 0x400)
 553                 access |= _PAGE_EXEC;
 554         else if (trap != 0x300)
 555                 return;
 556         hash_preload(vma->vm_mm, address, access, trap);
 557 #endif /* CONFIG_PPC_STD_MMU */
 558 }