arch/ppc64/mm/init.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  *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
   9  *
  10  *  Derived from "arch/i386/mm/init.c"
  11  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  12  *
  13  *  Dave Engebretsen <engebret@us.ibm.com>
  14  *      Rework for PPC64 port.
  15  *
  16  *  This program is free software; you can redistribute it and/or
  17  *  modify it under the terms of the GNU General Public License
  18  *  as published by the Free Software Foundation; either version
  19  *  2 of the License, or (at your option) any later version.
  20  *
  21  */
  22
  23 #include <linux/config.h>
  24 #include <linux/signal.h>
  25 #include <linux/sched.h>
  26 #include <linux/kernel.h>
  27 #include <linux/errno.h>
  28 #include <linux/string.h>
  29 #include <linux/types.h>
  30 #include <linux/mman.h>
  31 #include <linux/mm.h>
  32 #include <linux/swap.h>
  33 #include <linux/stddef.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/init.h>
  36 #include <linux/delay.h>
  37 #include <linux/bootmem.h>
  38 #include <linux/highmem.h>
  39 #include <linux/idr.h>
  40 #include <linux/nodemask.h>
  41 #include <linux/module.h>
  42
  43 #include <asm/pgalloc.h>
  44 #include <asm/page.h>
  45 #include <asm/prom.h>
  46 #include <asm/lmb.h>
  47 #include <asm/rtas.h>
  48 #include <asm/io.h>
  49 #include <asm/mmu_context.h>
  50 #include <asm/pgtable.h>
  51 #include <asm/mmu.h>
  52 #include <asm/uaccess.h>
  53 #include <asm/smp.h>
  54 #include <asm/machdep.h>
  55 #include <asm/tlb.h>
  56 #include <asm/eeh.h>
  57 #include <asm/processor.h>
  58 #include <asm/mmzone.h>
  59 #include <asm/cputable.h>
  60 #include <asm/ppcdebug.h>
  61 #include <asm/sections.h>
  62 #include <asm/system.h>
  63 #include <asm/iommu.h>
  64 #include <asm/abs_addr.h>
  65 #include <asm/vdso.h>
  66 #include <asm/imalloc.h>
  67
  68 #if PGTABLE_RANGE > USER_VSID_RANGE
  69 #warning Limited user VSID range means pagetable space is wasted
  70 #endif
  71
  72 #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
  73 #warning TASK_SIZE is smaller than it needs to be.
  74 #endif
  75
  76 int mem_init_done;
  77 unsigned long ioremap_bot = IMALLOC_BASE;
  78 static unsigned long phbs_io_bot = PHBS_IO_BASE;
  79
  80 extern pgd_t swapper_pg_dir[];
  81 extern struct task_struct *current_set[NR_CPUS];
  82
  83 unsigned long klimit = (unsigned long)_end;
  84
  85 unsigned long _SDR1=0;
  86 unsigned long _ASR=0;
  87
  88 /* max amount of RAM to use */
  89 unsigned long __max_memory;
  90
  91 /* info on what we think the IO hole is */
  92 unsigned long   io_hole_start;
  93 unsigned long   io_hole_size;
  94
  95 void show_mem(void)
  96 {
  97         unsigned long total = 0, reserved = 0;
  98         unsigned long shared = 0, cached = 0;
  99         struct page *page;
 100         pg_data_t *pgdat;
 101         unsigned long i;
 102
 103         printk("Mem-info:\n");
 104         show_free_areas();
 105         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 106         for_each_pgdat(pgdat) {
 107                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 108                         page = pgdat_page_nr(pgdat, i);
 109                         total++;
 110                         if (PageReserved(page))
 111                                 reserved++;
 112                         else if (PageSwapCache(page))
 113                                 cached++;
 114                         else if (page_count(page))
 115                                 shared += page_count(page) - 1;
 116                 }
 117         }
 118         printk("%ld pages of RAM\n", total);
 119         printk("%ld reserved pages\n", reserved);
 120         printk("%ld pages shared\n", shared);
 121         printk("%ld pages swap cached\n", cached);
 122 }
 123
 124 #ifdef CONFIG_PPC_ISERIES
 125
 126 void __iomem *ioremap(unsigned long addr, unsigned long size)
 127 {
 128         return (void __iomem *)addr;
 129 }
 130
 131 extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
 132                        unsigned long flags)
 133 {
 134         return (void __iomem *)addr;
 135 }
 136
 137 void iounmap(volatile void __iomem *addr)
 138 {
 139         return;
 140 }
 141
 142 #else
 143
 144 /*
 145  * map_io_page currently only called by __ioremap
 146  * map_io_page adds an entry to the ioremap page table
 147  * and adds an entry to the HPT, possibly bolting it
 148  */
 149 static int map_io_page(unsigned long ea, unsigned long pa, int flags)
 150 {
 151         pgd_t *pgdp;
 152         pud_t *pudp;
 153         pmd_t *pmdp;
 154         pte_t *ptep;
 155         unsigned long vsid;
 156
 157         if (mem_init_done) {
 158                 pgdp = pgd_offset_k(ea);
 159                 pudp = pud_alloc(&init_mm, pgdp, ea);
 160                 if (!pudp)
 161                         return -ENOMEM;
 162                 pmdp = pmd_alloc(&init_mm, pudp, ea);
 163                 if (!pmdp)
 164                         return -ENOMEM;
 165                 ptep = pte_alloc_kernel(pmdp, ea);
 166                 if (!ptep)
 167                         return -ENOMEM;
 168                 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
 169                                                           __pgprot(flags)));
 170         } else {
 171                 unsigned long va, vpn, hash, hpteg;
 172
 173                 /*
 174                  * If the mm subsystem is not fully up, we cannot create a
 175                  * linux page table entry for this mapping.  Simply bolt an
 176                  * entry in the hardware page table.
 177                  */
 178                 vsid = get_kernel_vsid(ea);
 179                 va = (vsid << 28) | (ea & 0xFFFFFFF);
 180                 vpn = va >> PAGE_SHIFT;
 181
 182                 hash = hpt_hash(vpn, 0);
 183
 184                 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
 185
 186                 /* Panic if a pte grpup is full */
 187                 if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT,
 188                                        HPTE_V_BOLTED,
 189                                        _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX)
 190                     == -1) {
 191                         panic("map_io_page: could not insert mapping");
 192                 }
 193         }
 194         return 0;
 195 }
 196
 197
 198 static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
 199                             unsigned long ea, unsigned long size,
 200                             unsigned long flags)
 201 {
 202         unsigned long i;
 203
 204         if ((flags & _PAGE_PRESENT) == 0)
 205                 flags |= pgprot_val(PAGE_KERNEL);
 206
 207         for (i = 0; i < size; i += PAGE_SIZE)
 208                 if (map_io_page(ea+i, pa+i, flags))
 209                         return NULL;
 210
 211         return (void __iomem *) (ea + (addr & ~PAGE_MASK));
 212 }
 213
 214
 215 void __iomem *
 216 ioremap(unsigned long addr, unsigned long size)
 217 {
 218         return __ioremap(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED);
 219 }
 220
 221 void __iomem * __ioremap(unsigned long addr, unsigned long size,
 222                          unsigned long flags)
 223 {
 224         unsigned long pa, ea;
 225         void __iomem *ret;
 226
 227         /*
 228          * Choose an address to map it to.
 229          * Once the imalloc system is running, we use it.
 230          * Before that, we map using addresses going
 231          * up from ioremap_bot.  imalloc will use
 232          * the addresses from ioremap_bot through
 233          * IMALLOC_END
 234          *
 235          */
 236         pa = addr & PAGE_MASK;
 237         size = PAGE_ALIGN(addr + size) - pa;
 238
 239         if (size == 0)
 240                 return NULL;
 241
 242         if (mem_init_done) {
 243                 struct vm_struct *area;
 244                 area = im_get_free_area(size);
 245                 if (area == NULL)
 246                         return NULL;
 247                 ea = (unsigned long)(area->addr);
 248                 ret = __ioremap_com(addr, pa, ea, size, flags);
 249                 if (!ret)
 250                         im_free(area->addr);
 251         } else {
 252                 ea = ioremap_bot;
 253                 ret = __ioremap_com(addr, pa, ea, size, flags);
 254                 if (ret)
 255                         ioremap_bot += size;
 256         }
 257         return ret;
 258 }
 259
 260 #define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
 261
 262 int __ioremap_explicit(unsigned long pa, unsigned long ea,
 263                        unsigned long size, unsigned long flags)
 264 {
 265         struct vm_struct *area;
 266         void __iomem *ret;
 267
 268         /* For now, require page-aligned values for pa, ea, and size */
 269         if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
 270             !IS_PAGE_ALIGNED(size)) {
 271                 printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
 272                 return 1;
 273         }
 274
 275         if (!mem_init_done) {
 276                 /* Two things to consider in this case:
 277                  * 1) No records will be kept (imalloc, etc) that the region
 278                  *    has been remapped
 279                  * 2) It won't be easy to iounmap() the region later (because
 280                  *    of 1)
 281                  */
 282                 ;
 283         } else {
 284                 area = im_get_area(ea, size,
 285                         IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
 286                 if (area == NULL) {
 287                         /* Expected when PHB-dlpar is in play */
 288                         return 1;
 289                 }
 290                 if (ea != (unsigned long) area->addr) {
 291                         printk(KERN_ERR "unexpected addr return from "
 292                                "im_get_area\n");
 293                         return 1;
 294                 }
 295         }
 296
 297         ret = __ioremap_com(pa, pa, ea, size, flags);
 298         if (ret == NULL) {
 299                 printk(KERN_ERR "ioremap_explicit() allocation failure !\n");
 300                 return 1;
 301         }
 302         if (ret != (void *) ea) {
 303                 printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
 304                 return 1;
 305         }
 306
 307         return 0;
 308 }
 309
 310 /*
 311  * Unmap an IO region and remove it from imalloc'd list.
 312  * Access to IO memory should be serialized by driver.
 313  * This code is modeled after vmalloc code - unmap_vm_area()
 314  *
 315  * XXX  what about calls before mem_init_done (ie python_countermeasures())
 316  */
 317 void iounmap(volatile void __iomem *token)
 318 {
 319         void *addr;
 320
 321         if (!mem_init_done)
 322                 return;
 323
 324         addr = (void *) ((unsigned long __force) token & PAGE_MASK);
 325
 326         im_free(addr);
 327 }
 328
 329 static int iounmap_subset_regions(unsigned long addr, unsigned long size)
 330 {
 331         struct vm_struct *area;
 332
 333         /* Check whether subsets of this region exist */
 334         area = im_get_area(addr, size, IM_REGION_SUPERSET);
 335         if (area == NULL)
 336                 return 1;
 337
 338         while (area) {
 339                 iounmap((void __iomem *) area->addr);
 340                 area = im_get_area(addr, size,
 341                                 IM_REGION_SUPERSET);
 342         }
 343
 344         return 0;
 345 }
 346
 347 int iounmap_explicit(volatile void __iomem *start, unsigned long size)
 348 {
 349         struct vm_struct *area;
 350         unsigned long addr;
 351         int rc;
 352
 353         addr = (unsigned long __force) start & PAGE_MASK;
 354
 355         /* Verify that the region either exists or is a subset of an existing
 356          * region.  In the latter case, split the parent region to create
 357          * the exact region
 358          */
 359         area = im_get_area(addr, size,
 360                             IM_REGION_EXISTS | IM_REGION_SUBSET);
 361         if (area == NULL) {
 362                 /* Determine whether subset regions exist.  If so, unmap */
 363                 rc = iounmap_subset_regions(addr, size);
 364                 if (rc) {
 365                         printk(KERN_ERR
 366                                "%s() cannot unmap nonexistent range 0x%lx\n",
 367                                 __FUNCTION__, addr);
 368                         return 1;
 369                 }
 370         } else {
 371                 iounmap((void __iomem *) area->addr);
 372         }
 373         /*
 374          * FIXME! This can't be right:
 375         iounmap(area->addr);
 376          * Maybe it should be "iounmap(area);"
 377          */
 378         return 0;
 379 }
 380
 381 #endif
 382
 383 EXPORT_SYMBOL(ioremap);
 384 EXPORT_SYMBOL(__ioremap);
 385 EXPORT_SYMBOL(iounmap);
 386
 387 void free_initmem(void)
 388 {
 389         unsigned long addr;
 390
 391         addr = (unsigned long)__init_begin;
 392         for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
 393                 memset((void *)addr, 0xcc, PAGE_SIZE);
 394                 ClearPageReserved(virt_to_page(addr));
 395                 set_page_count(virt_to_page(addr), 1);
 396                 free_page(addr);
 397                 totalram_pages++;
 398         }
 399         printk ("Freeing unused kernel memory: %luk freed\n",
 400                 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
 401 }
 402
 403 #ifdef CONFIG_BLK_DEV_INITRD
 404 void free_initrd_mem(unsigned long start, unsigned long end)
 405 {
 406         if (start < end)
 407                 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 408         for (; start < end; start += PAGE_SIZE) {
 409                 ClearPageReserved(virt_to_page(start));
 410                 set_page_count(virt_to_page(start), 1);
 411                 free_page(start);
 412                 totalram_pages++;
 413         }
 414 }
 415 #endif
 416
 417 static DEFINE_SPINLOCK(mmu_context_lock);
 418 static DEFINE_IDR(mmu_context_idr);
 419
 420 int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
 421 {
 422         int index;
 423         int err;
 424
 425 again:
 426         if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
 427                 return -ENOMEM;
 428
 429         spin_lock(&mmu_context_lock);
 430         err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
 431         spin_unlock(&mmu_context_lock);
 432
 433         if (err == -EAGAIN)
 434                 goto again;
 435         else if (err)
 436                 return err;
 437
 438         if (index > MAX_CONTEXT) {
 439                 idr_remove(&mmu_context_idr, index);
 440                 return -ENOMEM;
 441         }
 442
 443         mm->context.id = index;
 444
 445         return 0;
 446 }
 447
 448 void destroy_context(struct mm_struct *mm)
 449 {
 450         spin_lock(&mmu_context_lock);
 451         idr_remove(&mmu_context_idr, mm->context.id);
 452         spin_unlock(&mmu_context_lock);
 453
 454         mm->context.id = NO_CONTEXT;
 455 }
 456
 457 /*
 458  * Do very early mm setup.
 459  */
 460 void __init mm_init_ppc64(void)
 461 {
 462 #ifndef CONFIG_PPC_ISERIES
 463         unsigned long i;
 464 #endif
 465
 466         ppc64_boot_msg(0x100, "MM Init");
 467
 468         /* This is the story of the IO hole... please, keep seated,
 469          * unfortunately, we are out of oxygen masks at the moment.
 470          * So we need some rough way to tell where your big IO hole
 471          * is. On pmac, it's between 2G and 4G, on POWER3, it's around
 472          * that area as well, on POWER4 we don't have one, etc...
 473          * We need that as a "hint" when sizing the TCE table on POWER3
 474          * So far, the simplest way that seem work well enough for us it
 475          * to just assume that the first discontinuity in our physical
 476          * RAM layout is the IO hole. That may not be correct in the future
 477          * (and isn't on iSeries but then we don't care ;)
 478          */
 479
 480 #ifndef CONFIG_PPC_ISERIES
 481         for (i = 1; i < lmb.memory.cnt; i++) {
 482                 unsigned long base, prevbase, prevsize;
 483
 484                 prevbase = lmb.memory.region[i-1].base;
 485                 prevsize = lmb.memory.region[i-1].size;
 486                 base = lmb.memory.region[i].base;
 487                 if (base > (prevbase + prevsize)) {
 488                         io_hole_start = prevbase + prevsize;
 489                         io_hole_size = base  - (prevbase + prevsize);
 490                         break;
 491                 }
 492         }
 493 #endif /* CONFIG_PPC_ISERIES */
 494         if (io_hole_start)
 495                 printk("IO Hole assumed to be %lx -> %lx\n",
 496                        io_hole_start, io_hole_start + io_hole_size - 1);
 497
 498         ppc64_boot_msg(0x100, "MM Init Done");
 499 }
 500
 501 /*
 502  * This is called by /dev/mem to know if a given address has to
 503  * be mapped non-cacheable or not
 504  */
 505 int page_is_ram(unsigned long pfn)
 506 {
 507         int i;
 508         unsigned long paddr = (pfn << PAGE_SHIFT);
 509
 510         for (i=0; i < lmb.memory.cnt; i++) {
 511                 unsigned long base;
 512
 513                 base = lmb.memory.region[i].base;
 514
 515                 if ((paddr >= base) &&
 516                         (paddr < (base + lmb.memory.region[i].size))) {
 517                         return 1;
 518                 }
 519         }
 520
 521         return 0;
 522 }
 523 EXPORT_SYMBOL(page_is_ram);
 524
 525 /*
 526  * Initialize the bootmem system and give it all the memory we
 527  * have available.
 528  */
 529 #ifndef CONFIG_NEED_MULTIPLE_NODES
 530 void __init do_init_bootmem(void)
 531 {
 532         unsigned long i;
 533         unsigned long start, bootmap_pages;
 534         unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 535         int boot_mapsize;
 536
 537         /*
 538          * Find an area to use for the bootmem bitmap.  Calculate the size of
 539          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 540          * Add 1 additional page in case the address isn't page-aligned.
 541          */
 542         bootmap_pages = bootmem_bootmap_pages(total_pages);
 543
 544         start = lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
 545         BUG_ON(!start);
 546
 547         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 548
 549         max_pfn = max_low_pfn;
 550
 551         /* Add all physical memory to the bootmem map, mark each area
 552          * present.
 553          */
 554         for (i=0; i < lmb.memory.cnt; i++)
 555                 free_bootmem(lmb.memory.region[i].base,
 556                              lmb_size_bytes(&lmb.memory, i));
 557
 558         /* reserve the sections we're already using */
 559         for (i=0; i < lmb.reserved.cnt; i++)
 560                 reserve_bootmem(lmb.reserved.region[i].base,
 561                                 lmb_size_bytes(&lmb.reserved, i));
 562
 563         for (i=0; i < lmb.memory.cnt; i++)
 564                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
 565                                lmb_end_pfn(&lmb.memory, i));
 566 }
 567
 568 /*
 569  * paging_init() sets up the page tables - in fact we've already done this.
 570  */
 571 void __init paging_init(void)
 572 {
 573         unsigned long zones_size[MAX_NR_ZONES];
 574         unsigned long zholes_size[MAX_NR_ZONES];
 575         unsigned long total_ram = lmb_phys_mem_size();
 576         unsigned long top_of_ram = lmb_end_of_DRAM();
 577
 578         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 579                top_of_ram, total_ram);
 580         printk(KERN_INFO "Memory hole size: %ldMB\n",
 581                (top_of_ram - total_ram) >> 20);
 582         /*
 583          * All pages are DMA-able so we put them all in the DMA zone.
 584          */
 585         memset(zones_size, 0, sizeof(zones_size));
 586         memset(zholes_size, 0, sizeof(zholes_size));
 587
 588         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 589         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 590
 591         free_area_init_node(0, NODE_DATA(0), zones_size,
 592                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
 593 }
 594 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
 595
 596 static struct kcore_list kcore_vmem;
 597
 598 static int __init setup_kcore(void)
 599 {
 600         int i;
 601
 602         for (i=0; i < lmb.memory.cnt; i++) {
 603                 unsigned long base, size;
 604                 struct kcore_list *kcore_mem;
 605
 606                 base = lmb.memory.region[i].base;
 607                 size = lmb.memory.region[i].size;
 608
 609                 /* GFP_ATOMIC to avoid might_sleep warnings during boot */
 610                 kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
 611                 if (!kcore_mem)
 612                         panic("mem_init: kmalloc failed\n");
 613
 614                 kclist_add(kcore_mem, __va(base), size);
 615         }
 616
 617         kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
 618
 619         return 0;
 620 }
 621 module_init(setup_kcore);
 622
 623 void __init mem_init(void)
 624 {
 625 #ifdef CONFIG_NEED_MULTIPLE_NODES
 626         int nid;
 627 #endif
 628         pg_data_t *pgdat;
 629         unsigned long i;
 630         struct page *page;
 631         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 632
 633         num_physpages = max_low_pfn;    /* RAM is assumed contiguous */
 634         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 635
 636 #ifdef CONFIG_NEED_MULTIPLE_NODES
 637         for_each_online_node(nid) {
 638                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 639                         printk("freeing bootmem node %x\n", nid);
 640                         totalram_pages +=
 641                                 free_all_bootmem_node(NODE_DATA(nid));
 642                 }
 643         }
 644 #else
 645         max_mapnr = num_physpages;
 646         totalram_pages += free_all_bootmem();
 647 #endif
 648
 649         for_each_pgdat(pgdat) {
 650                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 651                         page = pgdat_page_nr(pgdat, i);
 652                         if (PageReserved(page))
 653                                 reservedpages++;
 654                 }
 655         }
 656
 657         codesize = (unsigned long)&_etext - (unsigned long)&_stext;
 658         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 659         datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
 660         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 661
 662         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 663                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 664                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 665                 num_physpages << (PAGE_SHIFT-10),
 666                 codesize >> 10,
 667                 reservedpages << (PAGE_SHIFT-10),
 668                 datasize >> 10,
 669                 bsssize >> 10,
 670                 initsize >> 10);
 671
 672         mem_init_done = 1;
 673
 674         /* Initialize the vDSO */
 675         vdso_init();
 676 }
 677
 678 /*
 679  * This is called when a page has been modified by the kernel.
 680  * It just marks the page as not i-cache clean.  We do the i-cache
 681  * flush later when the page is given to a user process, if necessary.
 682  */
 683 void flush_dcache_page(struct page *page)
 684 {
 685         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 686                 return;
 687         /* avoid an atomic op if possible */
 688         if (test_bit(PG_arch_1, &page->flags))
 689                 clear_bit(PG_arch_1, &page->flags);
 690 }
 691 EXPORT_SYMBOL(flush_dcache_page);
 692
 693 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 694 {
 695         clear_page(page);
 696
 697         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 698                 return;
 699         /*
 700          * We shouldnt have to do this, but some versions of glibc
 701          * require it (ld.so assumes zero filled pages are icache clean)
 702          * - Anton
 703          */
 704
 705         /* avoid an atomic op if possible */
 706         if (test_bit(PG_arch_1, &pg->flags))
 707                 clear_bit(PG_arch_1, &pg->flags);
 708 }
 709 EXPORT_SYMBOL(clear_user_page);
 710
 711 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 712                     struct page *pg)
 713 {
 714         copy_page(vto, vfrom);
 715
 716         /*
 717          * We should be able to use the following optimisation, however
 718          * there are two problems.
 719          * Firstly a bug in some versions of binutils meant PLT sections
 720          * were not marked executable.
 721          * Secondly the first word in the GOT section is blrl, used
 722          * to establish the GOT address. Until recently the GOT was
 723          * not marked executable.
 724          * - Anton
 725          */
 726 #if 0
 727         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 728                 return;
 729 #endif
 730
 731         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 732                 return;
 733
 734         /* avoid an atomic op if possible */
 735         if (test_bit(PG_arch_1, &pg->flags))
 736                 clear_bit(PG_arch_1, &pg->flags);
 737 }
 738
 739 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 740                              unsigned long addr, int len)
 741 {
 742         unsigned long maddr;
 743
 744         maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
 745         flush_icache_range(maddr, maddr + len);
 746 }
 747 EXPORT_SYMBOL(flush_icache_user_range);
 748
 749 /*
 750  * This is called at the end of handling a user page fault, when the
 751  * fault has been handled by updating a PTE in the linux page tables.
 752  * We use it to preload an HPTE into the hash table corresponding to
 753  * the updated linux PTE.
 754  *
 755  * This must always be called with the mm->page_table_lock held
 756  */
 757 void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
 758                       pte_t pte)
 759 {
 760         unsigned long vsid;
 761         void *pgdir;
 762         pte_t *ptep;
 763         int local = 0;
 764         cpumask_t tmp;
 765         unsigned long flags;
 766
 767         /* handle i-cache coherency */
 768         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 769             !cpu_has_feature(CPU_FTR_NOEXECUTE)) {
 770                 unsigned long pfn = pte_pfn(pte);
 771                 if (pfn_valid(pfn)) {
 772                         struct page *page = pfn_to_page(pfn);
 773                         if (!PageReserved(page)
 774                             && !test_bit(PG_arch_1, &page->flags)) {
 775                                 __flush_dcache_icache(page_address(page));
 776                                 set_bit(PG_arch_1, &page->flags);
 777                         }
 778                 }
 779         }
 780
 781         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 782         if (!pte_young(pte))
 783                 return;
 784
 785         pgdir = vma->vm_mm->pgd;
 786         if (pgdir == NULL)
 787                 return;
 788
 789         ptep = find_linux_pte(pgdir, ea);
 790         if (!ptep)
 791                 return;
 792
 793         vsid = get_vsid(vma->vm_mm->context.id, ea);
 794
 795         local_irq_save(flags);
 796         tmp = cpumask_of_cpu(smp_processor_id());
 797         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
 798                 local = 1;
 799
 800         __hash_page(ea, 0, vsid, ptep, 0x300, local);
 801         local_irq_restore(flags);
 802 }
 803
 804 void __iomem * reserve_phb_iospace(unsigned long size)
 805 {
 806         void __iomem *virt_addr;
 807
 808         if (phbs_io_bot >= IMALLOC_BASE)
 809                 panic("reserve_phb_iospace(): phb io space overflow\n");
 810
 811         virt_addr = (void __iomem *) phbs_io_bot;
 812         phbs_io_bot += size;
 813
 814         return virt_addr;
 815 }
 816
 817 static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
 818 {
 819         memset(addr, 0, kmem_cache_size(cache));
 820 }
 821
 822 static const int pgtable_cache_size[2] = {
 823         PTE_TABLE_SIZE, PMD_TABLE_SIZE
 824 };
 825 static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = {
 826         "pgd_pte_cache", "pud_pmd_cache",
 827 };
 828
 829 kmem_cache_t *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)];
 830
 831 void pgtable_cache_init(void)
 832 {
 833         int i;
 834
 835         BUILD_BUG_ON(PTE_TABLE_SIZE != pgtable_cache_size[PTE_CACHE_NUM]);
 836         BUILD_BUG_ON(PMD_TABLE_SIZE != pgtable_cache_size[PMD_CACHE_NUM]);
 837         BUILD_BUG_ON(PUD_TABLE_SIZE != pgtable_cache_size[PUD_CACHE_NUM]);
 838         BUILD_BUG_ON(PGD_TABLE_SIZE != pgtable_cache_size[PGD_CACHE_NUM]);
 839
 840         for (i = 0; i < ARRAY_SIZE(pgtable_cache_size); i++) {
 841                 int size = pgtable_cache_size[i];
 842                 const char *name = pgtable_cache_name[i];
 843
 844                 pgtable_cache[i] = kmem_cache_create(name,
 845                                                      size, size,
 846                                                      SLAB_HWCACHE_ALIGN
 847                                                      | SLAB_MUST_HWCACHE_ALIGN,
 848                                                      zero_ctor,
 849                                                      NULL);
 850                 if (! pgtable_cache[i])
 851                         panic("pgtable_cache_init(): could not create %s!\n",
 852                               name);
 853         }
 854 }
 855
 856 pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
 857                               unsigned long size, pgprot_t vma_prot)
 858 {
 859         if (ppc_md.phys_mem_access_prot)
 860                 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
 861
 862         if (!page_is_ram(addr >> PAGE_SHIFT))
 863                 vma_prot = __pgprot(pgprot_val(vma_prot)
 864                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 865         return vma_prot;
 866 }
 867 EXPORT_SYMBOL(phys_mem_access_prot);