linux/arch/powerpc/mm/init_64.c
<<
>>
Prefs
   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
  86static void pgd_ctor(void *addr)
  87{
  88        memset(addr, 0, PGD_TABLE_SIZE);
  89}
  90
  91static void pmd_ctor(void *addr)
  92{
  93        memset(addr, 0, PMD_TABLE_SIZE);
  94}
  95
  96struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
  97
  98/*
  99 * Create a kmem_cache() for pagetables.  This is not used for PTE
 100 * pages - they're linked to struct page, come from the normal free
 101 * pages pool and have a different entry size (see real_pte_t) to
 102 * everything else.  Caches created by this function are used for all
 103 * the higher level pagetables, and for hugepage pagetables.
 104 */
 105void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
 106{
 107        char *name;
 108        unsigned long table_size = sizeof(void *) << shift;
 109        unsigned long align = table_size;
 110
 111        /* When batching pgtable pointers for RCU freeing, we store
 112         * the index size in the low bits.  Table alignment must be
 113         * big enough to fit it.
 114         *
 115         * Likewise, hugeapge pagetable pointers contain a (different)
 116         * shift value in the low bits.  All tables must be aligned so
 117         * as to leave enough 0 bits in the address to contain it. */
 118        unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
 119                                     HUGEPD_SHIFT_MASK + 1);
 120        struct kmem_cache *new;
 121
 122        /* It would be nice if this was a BUILD_BUG_ON(), but at the
 123         * moment, gcc doesn't seem to recognize is_power_of_2 as a
 124         * constant expression, so so much for that. */
 125        BUG_ON(!is_power_of_2(minalign));
 126        BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
 127
 128        if (PGT_CACHE(shift))
 129                return; /* Already have a cache of this size */
 130
 131        align = max_t(unsigned long, align, minalign);
 132        name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
 133        new = kmem_cache_create(name, table_size, align, 0, ctor);
 134        PGT_CACHE(shift) = new;
 135
 136        pr_debug("Allocated pgtable cache for order %d\n", shift);
 137}
 138
 139
 140void pgtable_cache_init(void)
 141{
 142        pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
 143        pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
 144        if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
 145                panic("Couldn't allocate pgtable caches");
 146
 147        /* In all current configs, when the PUD index exists it's the
 148         * same size as either the pgd or pmd index.  Verify that the
 149         * initialization above has also created a PUD cache.  This
 150         * will need re-examiniation if we add new possibilities for
 151         * the pagetable layout. */
 152        BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
 153}
 154
 155#ifdef CONFIG_SPARSEMEM_VMEMMAP
 156/*
 157 * Given an address within the vmemmap, determine the pfn of the page that
 158 * represents the start of the section it is within.  Note that we have to
 159 * do this by hand as the proffered address may not be correctly aligned.
 160 * Subtraction of non-aligned pointers produces undefined results.
 161 */
 162static unsigned long __meminit vmemmap_section_start(unsigned long page)
 163{
 164        unsigned long offset = page - ((unsigned long)(vmemmap));
 165
 166        /* Return the pfn of the start of the section. */
 167        return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
 168}
 169
 170/*
 171 * Check if this vmemmap page is already initialised.  If any section
 172 * which overlaps this vmemmap page is initialised then this page is
 173 * initialised already.
 174 */
 175static int __meminit vmemmap_populated(unsigned long start, int page_size)
 176{
 177        unsigned long end = start + page_size;
 178
 179        for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
 180                if (pfn_valid(vmemmap_section_start(start)))
 181                        return 1;
 182
 183        return 0;
 184}
 185
 186/* On hash-based CPUs, the vmemmap is bolted in the hash table.
 187 *
 188 * On Book3E CPUs, the vmemmap is currently mapped in the top half of
 189 * the vmalloc space using normal page tables, though the size of
 190 * pages encoded in the PTEs can be different
 191 */
 192
 193#ifdef CONFIG_PPC_BOOK3E
 194static void __meminit vmemmap_create_mapping(unsigned long start,
 195                                             unsigned long page_size,
 196                                             unsigned long phys)
 197{
 198        /* Create a PTE encoding without page size */
 199        unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
 200                _PAGE_KERNEL_RW;
 201
 202        /* PTEs only contain page size encodings up to 32M */
 203        BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
 204
 205        /* Encode the size in the PTE */
 206        flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
 207
 208        /* For each PTE for that area, map things. Note that we don't
 209         * increment phys because all PTEs are of the large size and
 210         * thus must have the low bits clear
 211         */
 212        for (i = 0; i < page_size; i += PAGE_SIZE)
 213                BUG_ON(map_kernel_page(start + i, phys, flags));
 214}
 215#else /* CONFIG_PPC_BOOK3E */
 216static void __meminit vmemmap_create_mapping(unsigned long start,
 217                                             unsigned long page_size,
 218                                             unsigned long phys)
 219{
 220        int  mapped = htab_bolt_mapping(start, start + page_size, phys,
 221                                        PAGE_KERNEL, mmu_vmemmap_psize,
 222                                        mmu_kernel_ssize);
 223        BUG_ON(mapped < 0);
 224}
 225#endif /* CONFIG_PPC_BOOK3E */
 226
 227struct vmemmap_backing *vmemmap_list;
 228
 229static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
 230{
 231        static struct vmemmap_backing *next;
 232        static int num_left;
 233
 234        /* allocate a page when required and hand out chunks */
 235        if (!next || !num_left) {
 236                next = vmemmap_alloc_block(PAGE_SIZE, node);
 237                if (unlikely(!next)) {
 238                        WARN_ON(1);
 239                        return NULL;
 240                }
 241                num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
 242        }
 243
 244        num_left--;
 245
 246        return next++;
 247}
 248
 249static __meminit void vmemmap_list_populate(unsigned long phys,
 250                                            unsigned long start,
 251                                            int node)
 252{
 253        struct vmemmap_backing *vmem_back;
 254
 255        vmem_back = vmemmap_list_alloc(node);
 256        if (unlikely(!vmem_back)) {
 257                WARN_ON(1);
 258                return;
 259        }
 260
 261        vmem_back->phys = phys;
 262        vmem_back->virt_addr = start;
 263        vmem_back->list = vmemmap_list;
 264
 265        vmemmap_list = vmem_back;
 266}
 267
 268int __meminit vmemmap_populate(struct page *start_page,
 269                               unsigned long nr_pages, int node)
 270{
 271        unsigned long start = (unsigned long)start_page;
 272        unsigned long end = (unsigned long)(start_page + nr_pages);
 273        unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 274
 275        /* Align to the page size of the linear mapping. */
 276        start = _ALIGN_DOWN(start, page_size);
 277
 278        pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
 279                 start_page, nr_pages, node);
 280        pr_debug(" -> map %lx..%lx\n", start, end);
 281
 282        for (; start < end; start += page_size) {
 283                void *p;
 284
 285                if (vmemmap_populated(start, page_size))
 286                        continue;
 287
 288                p = vmemmap_alloc_block(page_size, node);
 289                if (!p)
 290                        return -ENOMEM;
 291
 292                vmemmap_list_populate(__pa(p), start, node);
 293
 294                pr_debug("      * %016lx..%016lx allocated at %p\n",
 295                         start, start + page_size, p);
 296
 297                vmemmap_create_mapping(start, page_size, __pa(p));
 298        }
 299
 300        return 0;
 301}
 302#endif /* CONFIG_SPARSEMEM_VMEMMAP */
 303
 304