arch/powerpc/mm/init_64.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  *
   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
  81 phys_addr_t memstart_addr = ~0;
  82 EXPORT_SYMBOL_GPL(memstart_addr);
  83 phys_addr_t kernstart_addr;
  84 EXPORT_SYMBOL_GPL(kernstart_addr);
  85
  86 void 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 static void pgd_ctor(void *addr)
 103 {
 104         memset(addr, 0, PGD_TABLE_SIZE);
 105 }
 106
 107 static void pmd_ctor(void *addr)
 108 {
 109         memset(addr, 0, PMD_TABLE_SIZE);
 110 }
 111
 112 struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
 113
 114 /*
 115  * Create a kmem_cache() for pagetables.  This is not used for PTE
 116  * pages - they're linked to struct page, come from the normal free
 117  * pages pool and have a different entry size (see real_pte_t) to
 118  * everything else.  Caches created by this function are used for all
 119  * the higher level pagetables, and for hugepage pagetables.
 120  */
 121 void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
 122 {
 123         char *name;
 124         unsigned long table_size = sizeof(void *) << shift;
 125         unsigned long align = table_size;
 126
 127         /* When batching pgtable pointers for RCU freeing, we store
 128          * the index size in the low bits.  Table alignment must be
 129          * big enough to fit it.
 130          *
 131          * Likewise, hugeapge pagetable pointers contain a (different)
 132          * shift value in the low bits.  All tables must be aligned so
 133          * as to leave enough 0 bits in the address to contain it. */
 134         unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
 135                                      HUGEPD_SHIFT_MASK + 1);
 136         struct kmem_cache *new;
 137
 138         /* It would be nice if this was a BUILD_BUG_ON(), but at the
 139          * moment, gcc doesn't seem to recognize is_power_of_2 as a
 140          * constant expression, so so much for that. */
 141         BUG_ON(!is_power_of_2(minalign));
 142         BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
 143
 144         if (PGT_CACHE(shift))
 145                 return; /* Already have a cache of this size */
 146
 147         align = max_t(unsigned long, align, minalign);
 148         name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
 149         new = kmem_cache_create(name, table_size, align, 0, ctor);
 150         PGT_CACHE(shift) = new;
 151
 152         pr_debug("Allocated pgtable cache for order %d\n", shift);
 153 }
 154
 155
 156 void pgtable_cache_init(void)
 157 {
 158         pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
 159         pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
 160         if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
 161                 panic("Couldn't allocate pgtable caches");
 162
 163         /* In all current configs, when the PUD index exists it's the
 164          * same size as either the pgd or pmd index.  Verify that the
 165          * initialization above has also created a PUD cache.  This
 166          * will need re-examiniation if we add new possibilities for
 167          * the pagetable layout. */
 168         BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
 169 }
 170
 171 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 172 /*
 173  * Given an address within the vmemmap, determine the pfn of the page that
 174  * represents the start of the section it is within.  Note that we have to
 175  * do this by hand as the proffered address may not be correctly aligned.
 176  * Subtraction of non-aligned pointers produces undefined results.
 177  */
 178 static unsigned long __meminit vmemmap_section_start(unsigned long page)
 179 {
 180         unsigned long offset = page - ((unsigned long)(vmemmap));
 181
 182         /* Return the pfn of the start of the section. */
 183         return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
 184 }
 185
 186 /*
 187  * Check if this vmemmap page is already initialised.  If any section
 188  * which overlaps this vmemmap page is initialised then this page is
 189  * initialised already.
 190  */
 191 static int __meminit vmemmap_populated(unsigned long start, int page_size)
 192 {
 193         unsigned long end = start + page_size;
 194
 195         for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
 196                 if (pfn_valid(vmemmap_section_start(start)))
 197                         return 1;
 198
 199         return 0;
 200 }
 201
 202 /* On hash-based CPUs, the vmemmap is bolted in the hash table.
 203  *
 204  * On Book3E CPUs, the vmemmap is currently mapped in the top half of
 205  * the vmalloc space using normal page tables, though the size of
 206  * pages encoded in the PTEs can be different
 207  */
 208
 209 #ifdef CONFIG_PPC_BOOK3E
 210 static void __meminit vmemmap_create_mapping(unsigned long start,
 211                                              unsigned long page_size,
 212                                              unsigned long phys)
 213 {
 214         /* Create a PTE encoding without page size */
 215         unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
 216                 _PAGE_KERNEL_RW;
 217
 218         /* PTEs only contain page size encodings up to 32M */
 219         BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
 220
 221         /* Encode the size in the PTE */
 222         flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
 223
 224         /* For each PTE for that area, map things. Note that we don't
 225          * increment phys because all PTEs are of the large size and
 226          * thus must have the low bits clear
 227          */
 228         for (i = 0; i < page_size; i += PAGE_SIZE)
 229                 BUG_ON(map_kernel_page(start + i, phys, flags));
 230 }
 231 #else /* CONFIG_PPC_BOOK3E */
 232 static void __meminit vmemmap_create_mapping(unsigned long start,
 233                                              unsigned long page_size,
 234                                              unsigned long phys)
 235 {
 236         int  mapped = htab_bolt_mapping(start, start + page_size, phys,
 237                                         PAGE_KERNEL, mmu_vmemmap_psize,
 238                                         mmu_kernel_ssize);
 239         BUG_ON(mapped < 0);
 240 }
 241 #endif /* CONFIG_PPC_BOOK3E */
 242
 243 struct vmemmap_backing *vmemmap_list;
 244
 245 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
 246 {
 247         static struct vmemmap_backing *next;
 248         static int num_left;
 249
 250         /* allocate a page when required and hand out chunks */
 251         if (!next || !num_left) {
 252                 next = vmemmap_alloc_block(PAGE_SIZE, node);
 253                 if (unlikely(!next)) {
 254                         WARN_ON(1);
 255                         return NULL;
 256                 }
 257                 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
 258         }
 259
 260         num_left--;
 261
 262         return next++;
 263 }
 264
 265 static __meminit void vmemmap_list_populate(unsigned long phys,
 266                                             unsigned long start,
 267                                             int node)
 268 {
 269         struct vmemmap_backing *vmem_back;
 270
 271         vmem_back = vmemmap_list_alloc(node);
 272         if (unlikely(!vmem_back)) {
 273                 WARN_ON(1);
 274                 return;
 275         }
 276
 277         vmem_back->phys = phys;
 278         vmem_back->virt_addr = start;
 279         vmem_back->list = vmemmap_list;
 280
 281         vmemmap_list = vmem_back;
 282 }
 283
 284 int __meminit vmemmap_populate(struct page *start_page,
 285                                unsigned long nr_pages, int node)
 286 {
 287         unsigned long start = (unsigned long)start_page;
 288         unsigned long end = (unsigned long)(start_page + nr_pages);
 289         unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 290
 291         /* Align to the page size of the linear mapping. */
 292         start = _ALIGN_DOWN(start, page_size);
 293
 294         pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
 295                  start_page, nr_pages, node);
 296         pr_debug(" -> map %lx..%lx\n", start, end);
 297
 298         for (; start < end; start += page_size) {
 299                 void *p;
 300
 301                 if (vmemmap_populated(start, page_size))
 302                         continue;
 303
 304                 p = vmemmap_alloc_block(page_size, node);
 305                 if (!p)
 306                         return -ENOMEM;
 307
 308                 vmemmap_list_populate(__pa(p), start, node);
 309
 310                 pr_debug("      * %016lx..%016lx allocated at %p\n",
 311                          start, start + page_size, p);
 312
 313                 vmemmap_create_mapping(start, page_size, __pa(p));
 314         }
 315
 316         return 0;
 317 }
 318 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
 319