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/abs_addr.h>
  46 #include <asm/prom.h>
  47 #include <asm/lmb.h>
  48 #include <asm/rtas.h>
  49 #include <asm/io.h>
  50 #include <asm/mmu_context.h>
  51 #include <asm/pgtable.h>
  52 #include <asm/mmu.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/smp.h>
  55 #include <asm/machdep.h>
  56 #include <asm/tlb.h>
  57 #include <asm/eeh.h>
  58 #include <asm/processor.h>
  59 #include <asm/mmzone.h>
  60 #include <asm/cputable.h>
  61 #include <asm/ppcdebug.h>
  62 #include <asm/sections.h>
  63 #include <asm/system.h>
  64 #include <asm/iommu.h>
  65 #include <asm/abs_addr.h>
  66 #include <asm/vdso.h>
  67
  68 int mem_init_done;
  69 unsigned long ioremap_bot = IMALLOC_BASE;
  70 static unsigned long phbs_io_bot = PHBS_IO_BASE;
  71
  72 extern pgd_t swapper_pg_dir[];
  73 extern struct task_struct *current_set[NR_CPUS];
  74
  75 extern pgd_t ioremap_dir[];
  76 pgd_t * ioremap_pgd = (pgd_t *)&ioremap_dir;
  77
  78 unsigned long klimit = (unsigned long)_end;
  79
  80 unsigned long _SDR1=0;
  81 unsigned long _ASR=0;
  82
  83 /* max amount of RAM to use */
  84 unsigned long __max_memory;
  85
  86 /* info on what we think the IO hole is */
  87 unsigned long   io_hole_start;
  88 unsigned long   io_hole_size;
  89
  90 void show_mem(void)
  91 {
  92         unsigned long total = 0, reserved = 0;
  93         unsigned long shared = 0, cached = 0;
  94         struct page *page;
  95         pg_data_t *pgdat;
  96         unsigned long i;
  97
  98         printk("Mem-info:\n");
  99         show_free_areas();
 100         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
 101         for_each_pgdat(pgdat) {
 102                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 103                         page = pgdat->node_mem_map + i;
 104                         total++;
 105                         if (PageReserved(page))
 106                                 reserved++;
 107                         else if (PageSwapCache(page))
 108                                 cached++;
 109                         else if (page_count(page))
 110                                 shared += page_count(page) - 1;
 111                 }
 112         }
 113         printk("%ld pages of RAM\n", total);
 114         printk("%ld reserved pages\n", reserved);
 115         printk("%ld pages shared\n", shared);
 116         printk("%ld pages swap cached\n", cached);
 117 }
 118
 119 #ifdef CONFIG_PPC_ISERIES
 120
 121 void __iomem *ioremap(unsigned long addr, unsigned long size)
 122 {
 123         return (void __iomem *)addr;
 124 }
 125
 126 extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
 127                        unsigned long flags)
 128 {
 129         return (void __iomem *)addr;
 130 }
 131
 132 void iounmap(volatile void __iomem *addr)
 133 {
 134         return;
 135 }
 136
 137 #else
 138
 139 /*
 140  * map_io_page currently only called by __ioremap
 141  * map_io_page adds an entry to the ioremap page table
 142  * and adds an entry to the HPT, possibly bolting it
 143  */
 144 static void map_io_page(unsigned long ea, unsigned long pa, int flags)
 145 {
 146         pgd_t *pgdp;
 147         pmd_t *pmdp;
 148         pte_t *ptep;
 149         unsigned long vsid;
 150
 151         if (mem_init_done) {
 152                 spin_lock(&ioremap_mm.page_table_lock);
 153                 pgdp = pgd_offset_i(ea);
 154                 pmdp = pmd_alloc(&ioremap_mm, pgdp, ea);
 155                 ptep = pte_alloc_kernel(&ioremap_mm, pmdp, ea);
 156
 157                 pa = abs_to_phys(pa);
 158                 set_pte_at(&ioremap_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags)));
 159                 spin_unlock(&ioremap_mm.page_table_lock);
 160         } else {
 161                 unsigned long va, vpn, hash, hpteg;
 162
 163                 /*
 164                  * If the mm subsystem is not fully up, we cannot create a
 165                  * linux page table entry for this mapping.  Simply bolt an
 166                  * entry in the hardware page table.
 167                  */
 168                 vsid = get_kernel_vsid(ea);
 169                 va = (vsid << 28) | (ea & 0xFFFFFFF);
 170                 vpn = va >> PAGE_SHIFT;
 171
 172                 hash = hpt_hash(vpn, 0);
 173
 174                 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
 175
 176                 /* Panic if a pte grpup is full */
 177                 if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT, 0,
 178                                        _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX,
 179                                        1, 0) == -1) {
 180                         panic("map_io_page: could not insert mapping");
 181                 }
 182         }
 183 }
 184
 185
 186 static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
 187                             unsigned long ea, unsigned long size,
 188                             unsigned long flags)
 189 {
 190         unsigned long i;
 191
 192         if ((flags & _PAGE_PRESENT) == 0)
 193                 flags |= pgprot_val(PAGE_KERNEL);
 194         if (flags & (_PAGE_NO_CACHE | _PAGE_WRITETHRU))
 195                 flags |= _PAGE_GUARDED;
 196
 197         for (i = 0; i < size; i += PAGE_SIZE) {
 198                 map_io_page(ea+i, pa+i, flags);
 199         }
 200
 201         return (void __iomem *) (ea + (addr & ~PAGE_MASK));
 202 }
 203
 204
 205 void __iomem *
 206 ioremap(unsigned long addr, unsigned long size)
 207 {
 208         return __ioremap(addr, size, _PAGE_NO_CACHE);
 209 }
 210
 211 void __iomem *
 212 __ioremap(unsigned long addr, unsigned long size, unsigned long flags)
 213 {
 214         unsigned long pa, ea;
 215
 216         /*
 217          * Choose an address to map it to.
 218          * Once the imalloc system is running, we use it.
 219          * Before that, we map using addresses going
 220          * up from ioremap_bot.  imalloc will use
 221          * the addresses from ioremap_bot through
 222          * IMALLOC_END (0xE000001fffffffff)
 223          *
 224          */
 225         pa = addr & PAGE_MASK;
 226         size = PAGE_ALIGN(addr + size) - pa;
 227
 228         if (size == 0)
 229                 return NULL;
 230
 231         if (mem_init_done) {
 232                 struct vm_struct *area;
 233                 area = im_get_free_area(size);
 234                 if (area == NULL)
 235                         return NULL;
 236                 ea = (unsigned long)(area->addr);
 237         } else {
 238                 ea = ioremap_bot;
 239                 ioremap_bot += size;
 240         }
 241
 242         return __ioremap_com(addr, pa, ea, size, flags);
 243 }
 244
 245 #define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
 246
 247 int __ioremap_explicit(unsigned long pa, unsigned long ea,
 248                        unsigned long size, unsigned long flags)
 249 {
 250         struct vm_struct *area;
 251
 252         /* For now, require page-aligned values for pa, ea, and size */
 253         if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
 254             !IS_PAGE_ALIGNED(size)) {
 255                 printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
 256                 return 1;
 257         }
 258
 259         if (!mem_init_done) {
 260                 /* Two things to consider in this case:
 261                  * 1) No records will be kept (imalloc, etc) that the region
 262                  *    has been remapped
 263                  * 2) It won't be easy to iounmap() the region later (because
 264                  *    of 1)
 265                  */
 266                 ;
 267         } else {
 268                 area = im_get_area(ea, size,
 269                         IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
 270                 if (area == NULL) {
 271                         /* Expected when PHB-dlpar is in play */
 272                         return 1;
 273                 }
 274                 if (ea != (unsigned long) area->addr) {
 275                         printk(KERN_ERR "unexpected addr return from im_get_area\n");
 276                         return 1;
 277                 }
 278         }
 279
 280         if (__ioremap_com(pa, pa, ea, size, flags) != (void *) ea) {
 281                 printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
 282                 return 1;
 283         }
 284
 285         return 0;
 286 }
 287
 288 static void unmap_im_area_pte(pmd_t *pmd, unsigned long address,
 289                                   unsigned long size)
 290 {
 291         unsigned long base, end;
 292         pte_t *pte;
 293
 294         if (pmd_none(*pmd))
 295                 return;
 296         if (pmd_bad(*pmd)) {
 297                 pmd_ERROR(*pmd);
 298                 pmd_clear(pmd);
 299                 return;
 300         }
 301
 302         pte = pte_offset_kernel(pmd, address);
 303         base = address & PMD_MASK;
 304         address &= ~PMD_MASK;
 305         end = address + size;
 306         if (end > PMD_SIZE)
 307                 end = PMD_SIZE;
 308
 309         do {
 310                 pte_t page;
 311                 page = ptep_get_and_clear(&ioremap_mm, base + address, pte);
 312                 address += PAGE_SIZE;
 313                 pte++;
 314                 if (pte_none(page))
 315                         continue;
 316                 if (pte_present(page))
 317                         continue;
 318                 printk(KERN_CRIT "Whee.. Swapped out page in kernel page table\n");
 319         } while (address < end);
 320 }
 321
 322 static void unmap_im_area_pmd(pgd_t *dir, unsigned long address,
 323                                   unsigned long size)
 324 {
 325         unsigned long base, end;
 326         pmd_t *pmd;
 327
 328         if (pgd_none(*dir))
 329                 return;
 330         if (pgd_bad(*dir)) {
 331                 pgd_ERROR(*dir);
 332                 pgd_clear(dir);
 333                 return;
 334         }
 335
 336         pmd = pmd_offset(dir, address);
 337         base = address & PGDIR_MASK;
 338         address &= ~PGDIR_MASK;
 339         end = address + size;
 340         if (end > PGDIR_SIZE)
 341                 end = PGDIR_SIZE;
 342
 343         do {
 344                 unmap_im_area_pte(pmd, base + address, end - address);
 345                 address = (address + PMD_SIZE) & PMD_MASK;
 346                 pmd++;
 347         } while (address < end);
 348 }
 349
 350 /*
 351  * Unmap an IO region and remove it from imalloc'd list.
 352  * Access to IO memory should be serialized by driver.
 353  * This code is modeled after vmalloc code - unmap_vm_area()
 354  *
 355  * XXX  what about calls before mem_init_done (ie python_countermeasures())
 356  */
 357 void iounmap(volatile void __iomem *token)
 358 {
 359         unsigned long address, start, end, size;
 360         struct mm_struct *mm;
 361         pgd_t *dir;
 362         void *addr;
 363
 364         if (!mem_init_done) {
 365                 return;
 366         }
 367
 368         addr = (void *) ((unsigned long __force) token & PAGE_MASK);
 369
 370         if ((size = im_free(addr)) == 0) {
 371                 return;
 372         }
 373
 374         address = (unsigned long)addr;
 375         start = address;
 376         end = address + size;
 377
 378         mm = &ioremap_mm;
 379         spin_lock(&mm->page_table_lock);
 380
 381         dir = pgd_offset_i(address);
 382         flush_cache_vunmap(address, end);
 383         do {
 384                 unmap_im_area_pmd(dir, address, end - address);
 385                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 386                 dir++;
 387         } while (address && (address < end));
 388         flush_tlb_kernel_range(start, end);
 389
 390         spin_unlock(&mm->page_table_lock);
 391         return;
 392 }
 393
 394 static int iounmap_subset_regions(unsigned long addr, unsigned long size)
 395 {
 396         struct vm_struct *area;
 397
 398         /* Check whether subsets of this region exist */
 399         area = im_get_area(addr, size, IM_REGION_SUPERSET);
 400         if (area == NULL)
 401                 return 1;
 402
 403         while (area) {
 404                 iounmap((void __iomem *) area->addr);
 405                 area = im_get_area(addr, size,
 406                                 IM_REGION_SUPERSET);
 407         }
 408
 409         return 0;
 410 }
 411
 412 int iounmap_explicit(volatile void __iomem *start, unsigned long size)
 413 {
 414         struct vm_struct *area;
 415         unsigned long addr;
 416         int rc;
 417
 418         addr = (unsigned long __force) start & PAGE_MASK;
 419
 420         /* Verify that the region either exists or is a subset of an existing
 421          * region.  In the latter case, split the parent region to create
 422          * the exact region
 423          */
 424         area = im_get_area(addr, size,
 425                             IM_REGION_EXISTS | IM_REGION_SUBSET);
 426         if (area == NULL) {
 427                 /* Determine whether subset regions exist.  If so, unmap */
 428                 rc = iounmap_subset_regions(addr, size);
 429                 if (rc) {
 430                         printk(KERN_ERR
 431                                "%s() cannot unmap nonexistent range 0x%lx\n",
 432                                 __FUNCTION__, addr);
 433                         return 1;
 434                 }
 435         } else {
 436                 iounmap((void __iomem *) area->addr);
 437         }
 438         /*
 439          * FIXME! This can't be right:
 440         iounmap(area->addr);
 441          * Maybe it should be "iounmap(area);"
 442          */
 443         return 0;
 444 }
 445
 446 #endif
 447
 448 EXPORT_SYMBOL(ioremap);
 449 EXPORT_SYMBOL(__ioremap);
 450 EXPORT_SYMBOL(iounmap);
 451
 452 void free_initmem(void)
 453 {
 454         unsigned long addr;
 455
 456         addr = (unsigned long)__init_begin;
 457         for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
 458                 ClearPageReserved(virt_to_page(addr));
 459                 set_page_count(virt_to_page(addr), 1);
 460                 free_page(addr);
 461                 totalram_pages++;
 462         }
 463         printk ("Freeing unused kernel memory: %luk freed\n",
 464                 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
 465 }
 466
 467 #ifdef CONFIG_BLK_DEV_INITRD
 468 void free_initrd_mem(unsigned long start, unsigned long end)
 469 {
 470         if (start < end)
 471                 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 472         for (; start < end; start += PAGE_SIZE) {
 473                 ClearPageReserved(virt_to_page(start));
 474                 set_page_count(virt_to_page(start), 1);
 475                 free_page(start);
 476                 totalram_pages++;
 477         }
 478 }
 479 #endif
 480
 481 static DEFINE_SPINLOCK(mmu_context_lock);
 482 static DEFINE_IDR(mmu_context_idr);
 483
 484 int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
 485 {
 486         int index;
 487         int err;
 488
 489 #ifdef CONFIG_HUGETLB_PAGE
 490         /* We leave htlb_segs as it was, but for a fork, we need to
 491          * clear the huge_pgdir. */
 492         mm->context.huge_pgdir = NULL;
 493 #endif
 494
 495 again:
 496         if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
 497                 return -ENOMEM;
 498
 499         spin_lock(&mmu_context_lock);
 500         err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
 501         spin_unlock(&mmu_context_lock);
 502
 503         if (err == -EAGAIN)
 504                 goto again;
 505         else if (err)
 506                 return err;
 507
 508         if (index > MAX_CONTEXT) {
 509                 idr_remove(&mmu_context_idr, index);
 510                 return -ENOMEM;
 511         }
 512
 513         mm->context.id = index;
 514
 515         return 0;
 516 }
 517
 518 void destroy_context(struct mm_struct *mm)
 519 {
 520         spin_lock(&mmu_context_lock);
 521         idr_remove(&mmu_context_idr, mm->context.id);
 522         spin_unlock(&mmu_context_lock);
 523
 524         mm->context.id = NO_CONTEXT;
 525
 526         hugetlb_mm_free_pgd(mm);
 527 }
 528
 529 /*
 530  * Do very early mm setup.
 531  */
 532 void __init mm_init_ppc64(void)
 533 {
 534 #ifndef CONFIG_PPC_ISERIES
 535         unsigned long i;
 536 #endif
 537
 538         ppc64_boot_msg(0x100, "MM Init");
 539
 540         /* This is the story of the IO hole... please, keep seated,
 541          * unfortunately, we are out of oxygen masks at the moment.
 542          * So we need some rough way to tell where your big IO hole
 543          * is. On pmac, it's between 2G and 4G, on POWER3, it's around
 544          * that area as well, on POWER4 we don't have one, etc...
 545          * We need that as a "hint" when sizing the TCE table on POWER3
 546          * So far, the simplest way that seem work well enough for us it
 547          * to just assume that the first discontinuity in our physical
 548          * RAM layout is the IO hole. That may not be correct in the future
 549          * (and isn't on iSeries but then we don't care ;)
 550          */
 551
 552 #ifndef CONFIG_PPC_ISERIES
 553         for (i = 1; i < lmb.memory.cnt; i++) {
 554                 unsigned long base, prevbase, prevsize;
 555
 556                 prevbase = lmb.memory.region[i-1].physbase;
 557                 prevsize = lmb.memory.region[i-1].size;
 558                 base = lmb.memory.region[i].physbase;
 559                 if (base > (prevbase + prevsize)) {
 560                         io_hole_start = prevbase + prevsize;
 561                         io_hole_size = base  - (prevbase + prevsize);
 562                         break;
 563                 }
 564         }
 565 #endif /* CONFIG_PPC_ISERIES */
 566         if (io_hole_start)
 567                 printk("IO Hole assumed to be %lx -> %lx\n",
 568                        io_hole_start, io_hole_start + io_hole_size - 1);
 569
 570         ppc64_boot_msg(0x100, "MM Init Done");
 571 }
 572
 573 /*
 574  * This is called by /dev/mem to know if a given address has to
 575  * be mapped non-cacheable or not
 576  */
 577 int page_is_ram(unsigned long pfn)
 578 {
 579         int i;
 580         unsigned long paddr = (pfn << PAGE_SHIFT);
 581
 582         for (i=0; i < lmb.memory.cnt; i++) {
 583                 unsigned long base;
 584
 585 #ifdef CONFIG_MSCHUNKS
 586                 base = lmb.memory.region[i].physbase;
 587 #else
 588                 base = lmb.memory.region[i].base;
 589 #endif
 590                 if ((paddr >= base) &&
 591                         (paddr < (base + lmb.memory.region[i].size))) {
 592                         return 1;
 593                 }
 594         }
 595
 596         return 0;
 597 }
 598 EXPORT_SYMBOL(page_is_ram);
 599
 600 /*
 601  * Initialize the bootmem system and give it all the memory we
 602  * have available.
 603  */
 604 #ifndef CONFIG_DISCONTIGMEM
 605 void __init do_init_bootmem(void)
 606 {
 607         unsigned long i;
 608         unsigned long start, bootmap_pages;
 609         unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
 610         int boot_mapsize;
 611
 612         /*
 613          * Find an area to use for the bootmem bitmap.  Calculate the size of
 614          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
 615          * Add 1 additional page in case the address isn't page-aligned.
 616          */
 617         bootmap_pages = bootmem_bootmap_pages(total_pages);
 618
 619         start = abs_to_phys(lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE));
 620         BUG_ON(!start);
 621
 622         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
 623
 624         max_pfn = max_low_pfn;
 625
 626         /* add all physical memory to the bootmem map. Also find the first */
 627         for (i=0; i < lmb.memory.cnt; i++) {
 628                 unsigned long physbase, size;
 629
 630                 physbase = lmb.memory.region[i].physbase;
 631                 size = lmb.memory.region[i].size;
 632                 free_bootmem(physbase, size);
 633         }
 634
 635         /* reserve the sections we're already using */
 636         for (i=0; i < lmb.reserved.cnt; i++) {
 637                 unsigned long physbase = lmb.reserved.region[i].physbase;
 638                 unsigned long size = lmb.reserved.region[i].size;
 639
 640                 reserve_bootmem(physbase, size);
 641         }
 642 }
 643
 644 /*
 645  * paging_init() sets up the page tables - in fact we've already done this.
 646  */
 647 void __init paging_init(void)
 648 {
 649         unsigned long zones_size[MAX_NR_ZONES];
 650         unsigned long zholes_size[MAX_NR_ZONES];
 651         unsigned long total_ram = lmb_phys_mem_size();
 652         unsigned long top_of_ram = lmb_end_of_DRAM();
 653
 654         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
 655                top_of_ram, total_ram);
 656         printk(KERN_INFO "Memory hole size: %ldMB\n",
 657                (top_of_ram - total_ram) >> 20);
 658         /*
 659          * All pages are DMA-able so we put them all in the DMA zone.
 660          */
 661         memset(zones_size, 0, sizeof(zones_size));
 662         memset(zholes_size, 0, sizeof(zholes_size));
 663
 664         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
 665         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
 666
 667         free_area_init_node(0, &contig_page_data, zones_size,
 668                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
 669 }
 670 #endif /* CONFIG_DISCONTIGMEM */
 671
 672 static struct kcore_list kcore_vmem;
 673
 674 static int __init setup_kcore(void)
 675 {
 676         int i;
 677
 678         for (i=0; i < lmb.memory.cnt; i++) {
 679                 unsigned long physbase, size;
 680                 struct kcore_list *kcore_mem;
 681
 682                 physbase = lmb.memory.region[i].physbase;
 683                 size = lmb.memory.region[i].size;
 684
 685                 /* GFP_ATOMIC to avoid might_sleep warnings during boot */
 686                 kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
 687                 if (!kcore_mem)
 688                         panic("mem_init: kmalloc failed\n");
 689
 690                 kclist_add(kcore_mem, __va(physbase), size);
 691         }
 692
 693         kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
 694
 695         return 0;
 696 }
 697 module_init(setup_kcore);
 698
 699 void __init mem_init(void)
 700 {
 701 #ifdef CONFIG_DISCONTIGMEM
 702         int nid;
 703 #endif
 704         pg_data_t *pgdat;
 705         unsigned long i;
 706         struct page *page;
 707         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
 708
 709         num_physpages = max_low_pfn;    /* RAM is assumed contiguous */
 710         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 711
 712 #ifdef CONFIG_DISCONTIGMEM
 713         for_each_online_node(nid) {
 714                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
 715                         printk("freeing bootmem node %x\n", nid);
 716                         totalram_pages +=
 717                                 free_all_bootmem_node(NODE_DATA(nid));
 718                 }
 719         }
 720 #else
 721         max_mapnr = num_physpages;
 722         totalram_pages += free_all_bootmem();
 723 #endif
 724
 725         for_each_pgdat(pgdat) {
 726                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
 727                         page = pgdat->node_mem_map + i;
 728                         if (PageReserved(page))
 729                                 reservedpages++;
 730                 }
 731         }
 732
 733         codesize = (unsigned long)&_etext - (unsigned long)&_stext;
 734         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
 735         datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
 736         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
 737
 738         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
 739                "%luk reserved, %luk data, %luk bss, %luk init)\n",
 740                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
 741                 num_physpages << (PAGE_SHIFT-10),
 742                 codesize >> 10,
 743                 reservedpages << (PAGE_SHIFT-10),
 744                 datasize >> 10,
 745                 bsssize >> 10,
 746                 initsize >> 10);
 747
 748         mem_init_done = 1;
 749
 750 #ifdef CONFIG_PPC_ISERIES
 751         iommu_vio_init();
 752 #endif
 753         /* Initialize the vDSO */
 754         vdso_init();
 755 }
 756
 757 /*
 758  * This is called when a page has been modified by the kernel.
 759  * It just marks the page as not i-cache clean.  We do the i-cache
 760  * flush later when the page is given to a user process, if necessary.
 761  */
 762 void flush_dcache_page(struct page *page)
 763 {
 764         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 765                 return;
 766         /* avoid an atomic op if possible */
 767         if (test_bit(PG_arch_1, &page->flags))
 768                 clear_bit(PG_arch_1, &page->flags);
 769 }
 770 EXPORT_SYMBOL(flush_dcache_page);
 771
 772 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 773 {
 774         clear_page(page);
 775
 776         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 777                 return;
 778         /*
 779          * We shouldnt have to do this, but some versions of glibc
 780          * require it (ld.so assumes zero filled pages are icache clean)
 781          * - Anton
 782          */
 783
 784         /* avoid an atomic op if possible */
 785         if (test_bit(PG_arch_1, &pg->flags))
 786                 clear_bit(PG_arch_1, &pg->flags);
 787 }
 788 EXPORT_SYMBOL(clear_user_page);
 789
 790 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 791                     struct page *pg)
 792 {
 793         copy_page(vto, vfrom);
 794
 795         /*
 796          * We should be able to use the following optimisation, however
 797          * there are two problems.
 798          * Firstly a bug in some versions of binutils meant PLT sections
 799          * were not marked executable.
 800          * Secondly the first word in the GOT section is blrl, used
 801          * to establish the GOT address. Until recently the GOT was
 802          * not marked executable.
 803          * - Anton
 804          */
 805 #if 0
 806         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
 807                 return;
 808 #endif
 809
 810         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
 811                 return;
 812
 813         /* avoid an atomic op if possible */
 814         if (test_bit(PG_arch_1, &pg->flags))
 815                 clear_bit(PG_arch_1, &pg->flags);
 816 }
 817
 818 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 819                              unsigned long addr, int len)
 820 {
 821         unsigned long maddr;
 822
 823         maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
 824         flush_icache_range(maddr, maddr + len);
 825 }
 826 EXPORT_SYMBOL(flush_icache_user_range);
 827
 828 /*
 829  * This is called at the end of handling a user page fault, when the
 830  * fault has been handled by updating a PTE in the linux page tables.
 831  * We use it to preload an HPTE into the hash table corresponding to
 832  * the updated linux PTE.
 833  *
 834  * This must always be called with the mm->page_table_lock held
 835  */
 836 void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
 837                       pte_t pte)
 838 {
 839         unsigned long vsid;
 840         void *pgdir;
 841         pte_t *ptep;
 842         int local = 0;
 843         cpumask_t tmp;
 844         unsigned long flags;
 845
 846         /* handle i-cache coherency */
 847         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
 848             !cpu_has_feature(CPU_FTR_NOEXECUTE)) {
 849                 unsigned long pfn = pte_pfn(pte);
 850                 if (pfn_valid(pfn)) {
 851                         struct page *page = pfn_to_page(pfn);
 852                         if (!PageReserved(page)
 853                             && !test_bit(PG_arch_1, &page->flags)) {
 854                                 __flush_dcache_icache(page_address(page));
 855                                 set_bit(PG_arch_1, &page->flags);
 856                         }
 857                 }
 858         }
 859
 860         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
 861         if (!pte_young(pte))
 862                 return;
 863
 864         pgdir = vma->vm_mm->pgd;
 865         if (pgdir == NULL)
 866                 return;
 867
 868         ptep = find_linux_pte(pgdir, ea);
 869         if (!ptep)
 870                 return;
 871
 872         vsid = get_vsid(vma->vm_mm->context.id, ea);
 873
 874         local_irq_save(flags);
 875         tmp = cpumask_of_cpu(smp_processor_id());
 876         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
 877                 local = 1;
 878
 879         __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
 880                     0x300, local);
 881         local_irq_restore(flags);
 882 }
 883
 884 void __iomem * reserve_phb_iospace(unsigned long size)
 885 {
 886         void __iomem *virt_addr;
 887
 888         if (phbs_io_bot >= IMALLOC_BASE)
 889                 panic("reserve_phb_iospace(): phb io space overflow\n");
 890
 891         virt_addr = (void __iomem *) phbs_io_bot;
 892         phbs_io_bot += size;
 893
 894         return virt_addr;
 895 }
 896
 897 kmem_cache_t *zero_cache;
 898
 899 static void zero_ctor(void *pte, kmem_cache_t *cache, unsigned long flags)
 900 {
 901         memset(pte, 0, PAGE_SIZE);
 902 }
 903
 904 void pgtable_cache_init(void)
 905 {
 906         zero_cache = kmem_cache_create("zero",
 907                                 PAGE_SIZE,
 908                                 0,
 909                                 SLAB_HWCACHE_ALIGN | SLAB_MUST_HWCACHE_ALIGN,
 910                                 zero_ctor,
 911                                 NULL);
 912         if (!zero_cache)
 913                 panic("pgtable_cache_init(): could not create zero_cache!\n");
 914 }
 915
 916 pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
 917                               unsigned long size, pgprot_t vma_prot)
 918 {
 919         if (ppc_md.phys_mem_access_prot)
 920                 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
 921
 922         if (!page_is_ram(addr >> PAGE_SHIFT))
 923                 vma_prot = __pgprot(pgprot_val(vma_prot)
 924                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
 925         return vma_prot;
 926 }
 927 EXPORT_SYMBOL(phys_mem_access_prot);