LXR linux/arch/powerpc/mm/init

   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 *
   9 *  Derived from "arch/i386/mm/init.c"
  10 *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  11 *
  12 *  Dave Engebretsen <engebret@us.ibm.com>
  13 *      Rework for PPC64 port.
  14 *
  15 *  This program is free software; you can redistribute it and/or
  16 *  modify it under the terms of the GNU General Public License
  17 *  as published by the Free Software Foundation; either version
  18 *  2 of the License, or (at your option) any later version.
  19 *
  20 */
  21
  22#undef DEBUG
  23
  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#include <linux/poison.h>
  43#include <linux/memblock.h>
  44#include <linux/hugetlb.h>
  45#include <linux/slab.h>
  46
  47#include <asm/pgalloc.h>
  48#include <asm/page.h>
  49#include <asm/prom.h>
  50#include <asm/rtas.h>
  51#include <asm/io.h>
  52#include <asm/mmu_context.h>
  53#include <asm/pgtable.h>
  54#include <asm/mmu.h>
  55#include <asm/uaccess.h>
  56#include <asm/smp.h>
  57#include <asm/machdep.h>
  58#include <asm/tlb.h>
  59#include <asm/eeh.h>
  60#include <asm/processor.h>
  61#include <asm/mmzone.h>
  62#include <asm/cputable.h>
  63#include <asm/sections.h>
  64#include <asm/system.h>
  65#include <asm/iommu.h>
  66#include <asm/abs_addr.h>
  67#include <asm/vdso.h>
  68
  69#include "mmu_decl.h"
  70
  71#ifdef CONFIG_PPC_STD_MMU_64
  72#if PGTABLE_RANGE > USER_VSID_RANGE
  73#warning Limited user VSID range means pagetable space is wasted
  74#endif
  75
  76#if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
  77#warning TASK_SIZE is smaller than it needs to be.
  78#endif
  79#endif /* CONFIG_PPC_STD_MMU_64 */
  80
  81phys_addr_t memstart_addr = ~0;
  82EXPORT_SYMBOL_GPL(memstart_addr);
  83phys_addr_t kernstart_addr;
  84EXPORT_SYMBOL_GPL(kernstart_addr);
  85
  86void free_initmem(void)
  87{
  88        unsigned long addr;
  89
  90        addr = (unsigned long)__init_begin;
  91        for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
  92                memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
  93                ClearPageReserved(virt_to_page(addr));
  94                init_page_count(virt_to_page(addr));
  95                free_page(addr);
  96                totalram_pages++;
  97        }
  98        printk ("Freeing unused kernel memory: %luk freed\n",
  99                ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
 100}
 101
 102#ifdef CONFIG_BLK_DEV_INITRD
 103void free_initrd_mem(unsigned long start, unsigned long end)
 104{
 105        if (start < end)
 106                printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 107        for (; start < end; start += PAGE_SIZE) {
 108                ClearPageReserved(virt_to_page(start));
 109                init_page_count(virt_to_page(start));
 110                free_page(start);
 111                totalram_pages++;
 112        }
 113}
 114#endif
 115
 116static void pgd_ctor(void *addr)
 117{
 118        memset(addr, 0, PGD_TABLE_SIZE);
 119}
 120
 121static void pmd_ctor(void *addr)
 122{
 123        memset(addr, 0, PMD_TABLE_SIZE);
 124}
 125
 126struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
 127
 128/*
 129 * Create a kmem_cache() for pagetables.  This is not used for PTE
 130 * pages - they're linked to struct page, come from the normal free
 131 * pages pool and have a different entry size (see real_pte_t) to
 132 * everything else.  Caches created by this function are used for all
 133 * the higher level pagetables, and for hugepage pagetables.
 134 */
 135void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
 136{
 137        char *name;
 138        unsigned long table_size = sizeof(void *) << shift;
 139        unsigned long align = table_size;
 140
 141        /* When batching pgtable pointers for RCU freeing, we store
 142         * the index size in the low bits.  Table alignment must be
 143         * big enough to fit it.
 144         *
 145         * Likewise, hugeapge pagetable pointers contain a (different)
 146         * shift value in the low bits.  All tables must be aligned so
 147         * as to leave enough 0 bits in the address to contain it. */
 148        unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
 149                                     HUGEPD_SHIFT_MASK + 1);
 150        struct kmem_cache *new;
 151
 152        /* It would be nice if this was a BUILD_BUG_ON(), but at the
 153         * moment, gcc doesn't seem to recognize is_power_of_2 as a
 154         * constant expression, so so much for that. */
 155        BUG_ON(!is_power_of_2(minalign));
 156        BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
 157
 158        if (PGT_CACHE(shift))
 159                return; /* Already have a cache of this size */
 160
 161        align = max_t(unsigned long, align, minalign);
 162        name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
 163        new = kmem_cache_create(name, table_size, align, 0, ctor);
 164        PGT_CACHE(shift) = new;
 165
 166        pr_debug("Allocated pgtable cache for order %d\n", shift);
 167}
 168
 169
 170void pgtable_cache_init(void)
 171{
 172        pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
 173        pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
 174        if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
 175                panic("Couldn't allocate pgtable caches");
 176
 177        /* In all current configs, when the PUD index exists it's the
 178         * same size as either the pgd or pmd index.  Verify that the
 179         * initialization above has also created a PUD cache.  This
 180         * will need re-examiniation if we add new possibilities for
 181         * the pagetable layout. */
 182        BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
 183}
 184
 185#ifdef CONFIG_SPARSEMEM_VMEMMAP
 186/*
 187 * Given an address within the vmemmap, determine the pfn of the page that
 188 * represents the start of the section it is within.  Note that we have to
 189 * do this by hand as the proffered address may not be correctly aligned.
 190 * Subtraction of non-aligned pointers produces undefined results.
 191 */
 192static unsigned long __meminit vmemmap_section_start(unsigned long page)
 193{
 194        unsigned long offset = page - ((unsigned long)(vmemmap));
 195
 196        /* Return the pfn of the start of the section. */
 197        return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
 198}
 199
 200/*
 201 * Check if this vmemmap page is already initialised.  If any section
 202 * which overlaps this vmemmap page is initialised then this page is
 203 * initialised already.
 204 */
 205static int __meminit vmemmap_populated(unsigned long start, int page_size)
 206{
 207        unsigned long end = start + page_size;
 208
 209        for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
 210                if (pfn_valid(vmemmap_section_start(start)))
 211                        return 1;
 212
 213        return 0;
 214}
 215
 216/* On hash-based CPUs, the vmemmap is bolted in the hash table.
 217 *
 218 * On Book3E CPUs, the vmemmap is currently mapped in the top half of
 219 * the vmalloc space using normal page tables, though the size of
 220 * pages encoded in the PTEs can be different
 221 */
 222
 223#ifdef CONFIG_PPC_BOOK3E
 224static void __meminit vmemmap_create_mapping(unsigned long start,
 225                                             unsigned long page_size,
 226                                             unsigned long phys)
 227{
 228        /* Create a PTE encoding without page size */
 229        unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
 230                _PAGE_KERNEL_RW;
 231
 232        /* PTEs only contain page size encodings up to 32M */
 233        BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
 234
 235        /* Encode the size in the PTE */
 236        flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
 237
 238        /* For each PTE for that area, map things. Note that we don't
 239         * increment phys because all PTEs are of the large size and
 240         * thus must have the low bits clear
 241         */
 242        for (i = 0; i < page_size; i += PAGE_SIZE)
 243                BUG_ON(map_kernel_page(start + i, phys, flags));
 244}
 245#else /* CONFIG_PPC_BOOK3E */
 246static void __meminit vmemmap_create_mapping(unsigned long start,
 247                                             unsigned long page_size,
 248                                             unsigned long phys)
 249{
 250        int  mapped = htab_bolt_mapping(start, start + page_size, phys,
 251                                        PAGE_KERNEL, mmu_vmemmap_psize,
 252                                        mmu_kernel_ssize);
 253        BUG_ON(mapped < 0);
 254}
 255#endif /* CONFIG_PPC_BOOK3E */
 256
 257struct vmemmap_backing *vmemmap_list;
 258
 259static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
 260{
 261        static struct vmemmap_backing *next;
 262        static int num_left;
 263
 264        /* allocate a page when required and hand out chunks */
 265        if (!next || !num_left) {
 266                next = vmemmap_alloc_block(PAGE_SIZE, node);
 267                if (unlikely(!next)) {
 268                        WARN_ON(1);
 269                        return NULL;
 270                }
 271                num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
 272        }
 273
 274        num_left--;
 275
 276        return next++;
 277}
 278
 279static __meminit void vmemmap_list_populate(unsigned long phys,
 280                                            unsigned long start,
 281                                            int node)
 282{
 283        struct vmemmap_backing *vmem_back;
 284
 285        vmem_back = vmemmap_list_alloc(node);
 286        if (unlikely(!vmem_back)) {
 287                WARN_ON(1);
 288                return;
 289        }
 290
 291        vmem_back->phys = phys;
 292        vmem_back->virt_addr = start;
 293        vmem_back->list = vmemmap_list;
 294
 295        vmemmap_list = vmem_back;
 296}
 297
 298int __meminit vmemmap_populate(struct page *start_page,
 299                               unsigned long nr_pages, int node)
 300{
 301        unsigned long start = (unsigned long)start_page;
 302        unsigned long end = (unsigned long)(start_page + nr_pages);
 303        unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 304
 305        /* Align to the page size of the linear mapping. */
 306        start = _ALIGN_DOWN(start, page_size);
 307
 308        pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
 309                 start_page, nr_pages, node);
 310        pr_debug(" -> map %lx..%lx\n", start, end);
 311
 312        for (; start < end; start += page_size) {
 313                void *p;
 314
 315                if (vmemmap_populated(start, page_size))
 316                        continue;
 317
 318                p = vmemmap_alloc_block(page_size, node);
 319                if (!p)
 320                        return -ENOMEM;
 321
 322                vmemmap_list_populate(__pa(p), start, node);
 323
 324                pr_debug("      * %016lx..%016lx allocated at %p\n",
 325                         start, start + page_size, p);
 326
 327                vmemmap_create_mapping(start, page_size, __pa(p));
 328        }
 329
 330        return 0;
 331}
 332#endif /* CONFIG_SPARSEMEM_VMEMMAP */
 333
 334