3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
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
9 * Derived from "arch/i386/mm/init.c"
10 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 * Dave Engebretsen <engebret@us.ibm.com>
13 * Rework for PPC64 port.
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.
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>
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>
47 #include <asm/pgalloc.h>
52 #include <asm/mmu_context.h>
53 #include <asm/pgtable.h>
55 #include <asm/uaccess.h>
57 #include <asm/machdep.h>
60 #include <asm/processor.h>
61 #include <asm/mmzone.h>
62 #include <asm/cputable.h>
63 #include <asm/sections.h>
64 #include <asm/iommu.h>
69 #ifdef CONFIG_PPC_STD_MMU_64
70 #if PGTABLE_RANGE > USER_VSID_RANGE
71 #warning Limited user VSID range means pagetable space is wasted
74 #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
75 #warning TASK_SIZE is smaller than it needs to be.
77 #endif /* CONFIG_PPC_STD_MMU_64 */
79 phys_addr_t memstart_addr = ~0;
80 EXPORT_SYMBOL_GPL(memstart_addr);
81 phys_addr_t kernstart_addr;
82 EXPORT_SYMBOL_GPL(kernstart_addr);
84 static void pgd_ctor(void *addr)
86 memset(addr, 0, PGD_TABLE_SIZE);
89 static void pmd_ctor(void *addr)
91 memset(addr, 0, PMD_TABLE_SIZE);
94 struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
97 * Create a kmem_cache() for pagetables. This is not used for PTE
98 * pages - they're linked to struct page, come from the normal free
99 * pages pool and have a different entry size (see real_pte_t) to
100 * everything else. Caches created by this function are used for all
101 * the higher level pagetables, and for hugepage pagetables.
103 void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
106 unsigned long table_size = sizeof(void *) << shift;
107 unsigned long align = table_size;
109 /* When batching pgtable pointers for RCU freeing, we store
110 * the index size in the low bits. Table alignment must be
111 * big enough to fit it.
113 * Likewise, hugeapge pagetable pointers contain a (different)
114 * shift value in the low bits. All tables must be aligned so
115 * as to leave enough 0 bits in the address to contain it. */
116 unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
117 HUGEPD_SHIFT_MASK + 1);
118 struct kmem_cache *new;
120 /* It would be nice if this was a BUILD_BUG_ON(), but at the
121 * moment, gcc doesn't seem to recognize is_power_of_2 as a
122 * constant expression, so so much for that. */
123 BUG_ON(!is_power_of_2(minalign));
124 BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
126 if (PGT_CACHE(shift))
127 return; /* Already have a cache of this size */
129 align = max_t(unsigned long, align, minalign);
130 name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
131 new = kmem_cache_create(name, table_size, align, 0, ctor);
132 pgtable_cache[shift - 1] = new;
133 pr_debug("Allocated pgtable cache for order %d\n", shift);
137 void pgtable_cache_init(void)
139 pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
140 pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
141 if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
142 panic("Couldn't allocate pgtable caches");
144 /* In all current configs, when the PUD index exists it's the
145 * same size as either the pgd or pmd index. Verify that the
146 * initialization above has also created a PUD cache. This
147 * will need re-examiniation if we add new possibilities for
148 * the pagetable layout. */
149 BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
152 #ifdef CONFIG_SPARSEMEM_VMEMMAP
154 * Given an address within the vmemmap, determine the pfn of the page that
155 * represents the start of the section it is within. Note that we have to
156 * do this by hand as the proffered address may not be correctly aligned.
157 * Subtraction of non-aligned pointers produces undefined results.
159 static unsigned long __meminit vmemmap_section_start(unsigned long page)
161 unsigned long offset = page - ((unsigned long)(vmemmap));
163 /* Return the pfn of the start of the section. */
164 return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
168 * Check if this vmemmap page is already initialised. If any section
169 * which overlaps this vmemmap page is initialised then this page is
170 * initialised already.
172 static int __meminit vmemmap_populated(unsigned long start, int page_size)
174 unsigned long end = start + page_size;
176 for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
177 if (pfn_valid(vmemmap_section_start(start)))
183 /* On hash-based CPUs, the vmemmap is bolted in the hash table.
185 * On Book3E CPUs, the vmemmap is currently mapped in the top half of
186 * the vmalloc space using normal page tables, though the size of
187 * pages encoded in the PTEs can be different
190 #ifdef CONFIG_PPC_BOOK3E
191 static void __meminit vmemmap_create_mapping(unsigned long start,
192 unsigned long page_size,
195 /* Create a PTE encoding without page size */
196 unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
199 /* PTEs only contain page size encodings up to 32M */
200 BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
202 /* Encode the size in the PTE */
203 flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
205 /* For each PTE for that area, map things. Note that we don't
206 * increment phys because all PTEs are of the large size and
207 * thus must have the low bits clear
209 for (i = 0; i < page_size; i += PAGE_SIZE)
210 BUG_ON(map_kernel_page(start + i, phys, flags));
212 #else /* CONFIG_PPC_BOOK3E */
213 static void __meminit vmemmap_create_mapping(unsigned long start,
214 unsigned long page_size,
217 int mapped = htab_bolt_mapping(start, start + page_size, phys,
218 PAGE_KERNEL, mmu_vmemmap_psize,
222 #endif /* CONFIG_PPC_BOOK3E */
224 struct vmemmap_backing *vmemmap_list;
226 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
228 static struct vmemmap_backing *next;
231 /* allocate a page when required and hand out chunks */
232 if (!next || !num_left) {
233 next = vmemmap_alloc_block(PAGE_SIZE, node);
234 if (unlikely(!next)) {
238 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
246 static __meminit void vmemmap_list_populate(unsigned long phys,
250 struct vmemmap_backing *vmem_back;
252 vmem_back = vmemmap_list_alloc(node);
253 if (unlikely(!vmem_back)) {
258 vmem_back->phys = phys;
259 vmem_back->virt_addr = start;
260 vmem_back->list = vmemmap_list;
262 vmemmap_list = vmem_back;
265 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
267 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
269 /* Align to the page size of the linear mapping. */
270 start = _ALIGN_DOWN(start, page_size);
272 pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node);
274 for (; start < end; start += page_size) {
277 if (vmemmap_populated(start, page_size))
280 p = vmemmap_alloc_block(page_size, node);
284 vmemmap_list_populate(__pa(p), start, node);
286 pr_debug(" * %016lx..%016lx allocated at %p\n",
287 start, start + page_size, p);
289 vmemmap_create_mapping(start, page_size, __pa(p));
295 void vmemmap_free(unsigned long start, unsigned long end)
299 #endif /* CONFIG_SPARSEMEM_VMEMMAP */