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/highmem.h>
  38 #include <linux/idr.h>
  39 #include <linux/nodemask.h>
  40 #include <linux/module.h>
  41 #include <linux/poison.h>
  42 #include <linux/memblock.h>
  43 #include <linux/hugetlb.h>
  44 #include <linux/slab.h>
  45
  46 #include <asm/pgalloc.h>
  47 #include <asm/page.h>
  48 #include <asm/prom.h>
  49 #include <asm/rtas.h>
  50 #include <asm/io.h>
  51 #include <asm/mmu_context.h>
  52 #include <asm/pgtable.h>
  53 #include <asm/mmu.h>
  54 #include <linux/uaccess.h>
  55 #include <asm/smp.h>
  56 #include <asm/machdep.h>
  57 #include <asm/tlb.h>
  58 #include <asm/eeh.h>
  59 #include <asm/processor.h>
  60 #include <asm/mmzone.h>
  61 #include <asm/cputable.h>
  62 #include <asm/sections.h>
  63 #include <asm/iommu.h>
  64 #include <asm/vdso.h>
  65
  66 #include "mmu_decl.h"
  67
  68 #ifdef CONFIG_PPC_STD_MMU_64
  69 #if H_PGTABLE_RANGE > USER_VSID_RANGE
  70 #warning Limited user VSID range means pagetable space is wasted
  71 #endif
  72
  73 #if (TASK_SIZE_USER64 < H_PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
  74 #warning TASK_SIZE is smaller than it needs to be.
  75 #endif
  76 #endif /* CONFIG_PPC_STD_MMU_64 */
  77
  78 phys_addr_t memstart_addr = ~0;
  79 EXPORT_SYMBOL_GPL(memstart_addr);
  80 phys_addr_t kernstart_addr;
  81 EXPORT_SYMBOL_GPL(kernstart_addr);
  82
  83 #ifdef CONFIG_SPARSEMEM_VMEMMAP
  84 /*
  85  * Given an address within the vmemmap, determine the pfn of the page that
  86  * represents the start of the section it is within.  Note that we have to
  87  * do this by hand as the proffered address may not be correctly aligned.
  88  * Subtraction of non-aligned pointers produces undefined results.
  89  */
  90 static unsigned long __meminit vmemmap_section_start(unsigned long page)
  91 {
  92         unsigned long offset = page - ((unsigned long)(vmemmap));
  93
  94         /* Return the pfn of the start of the section. */
  95         return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
  96 }
  97
  98 /*
  99  * Check if this vmemmap page is already initialised.  If any section
 100  * which overlaps this vmemmap page is initialised then this page is
 101  * initialised already.
 102  */
 103 static int __meminit vmemmap_populated(unsigned long start, int page_size)
 104 {
 105         unsigned long end = start + page_size;
 106         start = (unsigned long)(pfn_to_page(vmemmap_section_start(start)));
 107
 108         for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
 109                 if (pfn_valid(page_to_pfn((struct page *)start)))
 110                         return 1;
 111
 112         return 0;
 113 }
 114
 115 struct vmemmap_backing *vmemmap_list;
 116 static struct vmemmap_backing *next;
 117 static int num_left;
 118 static int num_freed;
 119
 120 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
 121 {
 122         struct vmemmap_backing *vmem_back;
 123         /* get from freed entries first */
 124         if (num_freed) {
 125                 num_freed--;
 126                 vmem_back = next;
 127                 next = next->list;
 128
 129                 return vmem_back;
 130         }
 131
 132         /* allocate a page when required and hand out chunks */
 133         if (!num_left) {
 134                 next = vmemmap_alloc_block(PAGE_SIZE, node);
 135                 if (unlikely(!next)) {
 136                         WARN_ON(1);
 137                         return NULL;
 138                 }
 139                 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
 140         }
 141
 142         num_left--;
 143
 144         return next++;
 145 }
 146
 147 static __meminit void vmemmap_list_populate(unsigned long phys,
 148                                             unsigned long start,
 149                                             int node)
 150 {
 151         struct vmemmap_backing *vmem_back;
 152
 153         vmem_back = vmemmap_list_alloc(node);
 154         if (unlikely(!vmem_back)) {
 155                 WARN_ON(1);
 156                 return;
 157         }
 158
 159         vmem_back->phys = phys;
 160         vmem_back->virt_addr = start;
 161         vmem_back->list = vmemmap_list;
 162
 163         vmemmap_list = vmem_back;
 164 }
 165
 166 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
 167 {
 168         unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 169
 170         /* Align to the page size of the linear mapping. */
 171         start = _ALIGN_DOWN(start, page_size);
 172
 173         pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node);
 174
 175         for (; start < end; start += page_size) {
 176                 void *p;
 177                 int rc;
 178
 179                 if (vmemmap_populated(start, page_size))
 180                         continue;
 181
 182                 p = vmemmap_alloc_block(page_size, node);
 183                 if (!p)
 184                         return -ENOMEM;
 185
 186                 vmemmap_list_populate(__pa(p), start, node);
 187
 188                 pr_debug("      * %016lx..%016lx allocated at %p\n",
 189                          start, start + page_size, p);
 190
 191                 rc = vmemmap_create_mapping(start, page_size, __pa(p));
 192                 if (rc < 0) {
 193                         pr_warning(
 194                                 "vmemmap_populate: Unable to create vmemmap mapping: %d\n",
 195                                 rc);
 196                         return -EFAULT;
 197                 }
 198         }
 199
 200         return 0;
 201 }
 202
 203 #ifdef CONFIG_MEMORY_HOTPLUG
 204 static unsigned long vmemmap_list_free(unsigned long start)
 205 {
 206         struct vmemmap_backing *vmem_back, *vmem_back_prev;
 207
 208         vmem_back_prev = vmem_back = vmemmap_list;
 209
 210         /* look for it with prev pointer recorded */
 211         for (; vmem_back; vmem_back = vmem_back->list) {
 212                 if (vmem_back->virt_addr == start)
 213                         break;
 214                 vmem_back_prev = vmem_back;
 215         }
 216
 217         if (unlikely(!vmem_back)) {
 218                 WARN_ON(1);
 219                 return 0;
 220         }
 221
 222         /* remove it from vmemmap_list */
 223         if (vmem_back == vmemmap_list) /* remove head */
 224                 vmemmap_list = vmem_back->list;
 225         else
 226                 vmem_back_prev->list = vmem_back->list;
 227
 228         /* next point to this freed entry */
 229         vmem_back->list = next;
 230         next = vmem_back;
 231         num_freed++;
 232
 233         return vmem_back->phys;
 234 }
 235
 236 void __ref vmemmap_free(unsigned long start, unsigned long end)
 237 {
 238         unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 239
 240         start = _ALIGN_DOWN(start, page_size);
 241
 242         pr_debug("vmemmap_free %lx...%lx\n", start, end);
 243
 244         for (; start < end; start += page_size) {
 245                 unsigned long addr;
 246
 247                 /*
 248                  * the section has already be marked as invalid, so
 249                  * vmemmap_populated() true means some other sections still
 250                  * in this page, so skip it.
 251                  */
 252                 if (vmemmap_populated(start, page_size))
 253                         continue;
 254
 255                 addr = vmemmap_list_free(start);
 256                 if (addr) {
 257                         struct page *page = pfn_to_page(addr >> PAGE_SHIFT);
 258
 259                         if (PageReserved(page)) {
 260                                 /* allocated from bootmem */
 261                                 if (page_size < PAGE_SIZE) {
 262                                         /*
 263                                          * this shouldn't happen, but if it is
 264                                          * the case, leave the memory there
 265                                          */
 266                                         WARN_ON_ONCE(1);
 267                                 } else {
 268                                         unsigned int nr_pages =
 269                                                 1 << get_order(page_size);
 270                                         while (nr_pages--)
 271                                                 free_reserved_page(page++);
 272                                 }
 273                         } else
 274                                 free_pages((unsigned long)(__va(addr)),
 275                                                         get_order(page_size));
 276
 277                         vmemmap_remove_mapping(start, page_size);
 278                 }
 279         }
 280 }
 281 #endif
 282 void register_page_bootmem_memmap(unsigned long section_nr,
 283                                   struct page *start_page, unsigned long size)
 284 {
 285 }
 286
 287 /*
 288  * We do not have access to the sparsemem vmemmap, so we fallback to
 289  * walking the list of sparsemem blocks which we already maintain for
 290  * the sake of crashdump. In the long run, we might want to maintain
 291  * a tree if performance of that linear walk becomes a problem.
 292  *
 293  * realmode_pfn_to_page functions can fail due to:
 294  * 1) As real sparsemem blocks do not lay in RAM continously (they
 295  * are in virtual address space which is not available in the real mode),
 296  * the requested page struct can be split between blocks so get_page/put_page
 297  * may fail.
 298  * 2) When huge pages are used, the get_page/put_page API will fail
 299  * in real mode as the linked addresses in the page struct are virtual
 300  * too.
 301  */
 302 struct page *realmode_pfn_to_page(unsigned long pfn)
 303 {
 304         struct vmemmap_backing *vmem_back;
 305         struct page *page;
 306         unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
 307         unsigned long pg_va = (unsigned long) pfn_to_page(pfn);
 308
 309         for (vmem_back = vmemmap_list; vmem_back; vmem_back = vmem_back->list) {
 310                 if (pg_va < vmem_back->virt_addr)
 311                         continue;
 312
 313                 /* After vmemmap_list entry free is possible, need check all */
 314                 if ((pg_va + sizeof(struct page)) <=
 315                                 (vmem_back->virt_addr + page_size)) {
 316                         page = (struct page *) (vmem_back->phys + pg_va -
 317                                 vmem_back->virt_addr);
 318                         return page;
 319                 }
 320         }
 321
 322         /* Probably that page struct is split between real pages */
 323         return NULL;
 324 }
 325 EXPORT_SYMBOL_GPL(realmode_pfn_to_page);
 326
 327 #elif defined(CONFIG_FLATMEM)
 328
 329 struct page *realmode_pfn_to_page(unsigned long pfn)
 330 {
 331         struct page *page = pfn_to_page(pfn);
 332         return page;
 333 }
 334 EXPORT_SYMBOL_GPL(realmode_pfn_to_page);
 335
 336 #endif /* CONFIG_SPARSEMEM_VMEMMAP/CONFIG_FLATMEM */
 337
 338 #ifdef CONFIG_PPC_STD_MMU_64
 339 static bool disable_radix;
 340 static int __init parse_disable_radix(char *p)
 341 {
 342         disable_radix = true;
 343         return 0;
 344 }
 345 early_param("disable_radix", parse_disable_radix);
 346
 347 void __init mmu_early_init_devtree(void)
 348 {
 349         /* Disable radix mode based on kernel command line. */
 350         if (disable_radix)
 351                 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
 352
 353         if (early_radix_enabled())
 354                 radix__early_init_devtree();
 355         else
 356                 hash__early_init_devtree();
 357 }
 358 #endif /* CONFIG_PPC_STD_MMU_64 */