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/iommu.h>
65 #include <asm/abs_addr.h>
66 #include <asm/vdso.h>
67
68 #include "mmu_decl.h"
69
70 #ifdef CONFIG_PPC_STD_MMU_64
71 #if PGTABLE_RANGE > USER_VSID_RANGE
72 #warning Limited user VSID range means pagetable space is wasted
73 #endif
74
75 #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
76 #warning TASK_SIZE is smaller than it needs to be.
77 #endif
78 #endif /* CONFIG_PPC_STD_MMU_64 */
79
80 phys_addr_t memstart_addr = ~0;
81 EXPORT_SYMBOL_GPL(memstart_addr);
82 phys_addr_t kernstart_addr;
83 EXPORT_SYMBOL_GPL(kernstart_addr);
84
pgd_ctor(void * addr)85 static void pgd_ctor(void *addr)
86 {
87 memset(addr, 0, PGD_TABLE_SIZE);
88 }
89
pmd_ctor(void * addr)90 static void pmd_ctor(void *addr)
91 {
92 memset(addr, 0, PMD_TABLE_SIZE);
93 }
94
95 struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
96
97 /*
98 * Create a kmem_cache() for pagetables. This is not used for PTE
99 * pages - they're linked to struct page, come from the normal free
100 * pages pool and have a different entry size (see real_pte_t) to
101 * everything else. Caches created by this function are used for all
102 * the higher level pagetables, and for hugepage pagetables.
103 */
pgtable_cache_add(unsigned shift,void (* ctor)(void *))104 void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
105 {
106 char *name;
107 unsigned long table_size = sizeof(void *) << shift;
108 unsigned long align = table_size;
109
110 /* When batching pgtable pointers for RCU freeing, we store
111 * the index size in the low bits. Table alignment must be
112 * big enough to fit it.
113 *
114 * Likewise, hugeapge pagetable pointers contain a (different)
115 * shift value in the low bits. All tables must be aligned so
116 * as to leave enough 0 bits in the address to contain it. */
117 unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
118 HUGEPD_SHIFT_MASK + 1);
119 struct kmem_cache *new;
120
121 /* It would be nice if this was a BUILD_BUG_ON(), but at the
122 * moment, gcc doesn't seem to recognize is_power_of_2 as a
123 * constant expression, so so much for that. */
124 BUG_ON(!is_power_of_2(minalign));
125 BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
126
127 if (PGT_CACHE(shift))
128 return; /* Already have a cache of this size */
129
130 align = max_t(unsigned long, align, minalign);
131 name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
132 new = kmem_cache_create(name, table_size, align, 0, ctor);
133 PGT_CACHE(shift) = new;
134
135 pr_debug("Allocated pgtable cache for order %d\n", shift);
136 }
137
138
pgtable_cache_init(void)139 void pgtable_cache_init(void)
140 {
141 pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
142 pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
143 if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
144 panic("Couldn't allocate pgtable caches");
145
146 /* In all current configs, when the PUD index exists it's the
147 * same size as either the pgd or pmd index. Verify that the
148 * initialization above has also created a PUD cache. This
149 * will need re-examiniation if we add new possibilities for
150 * the pagetable layout. */
151 BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
152 }
153
154 #ifdef CONFIG_SPARSEMEM_VMEMMAP
155 /*
156 * Given an address within the vmemmap, determine the pfn of the page that
157 * represents the start of the section it is within. Note that we have to
158 * do this by hand as the proffered address may not be correctly aligned.
159 * Subtraction of non-aligned pointers produces undefined results.
160 */
vmemmap_section_start(unsigned long page)161 static unsigned long __meminit vmemmap_section_start(unsigned long page)
162 {
163 unsigned long offset = page - ((unsigned long)(vmemmap));
164
165 /* Return the pfn of the start of the section. */
166 return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
167 }
168
169 /*
170 * Check if this vmemmap page is already initialised. If any section
171 * which overlaps this vmemmap page is initialised then this page is
172 * initialised already.
173 */
vmemmap_populated(unsigned long start,int page_size)174 static int __meminit vmemmap_populated(unsigned long start, int page_size)
175 {
176 unsigned long end = start + page_size;
177
178 for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
179 if (pfn_valid(vmemmap_section_start(start)))
180 return 1;
181
182 return 0;
183 }
184
185 /* On hash-based CPUs, the vmemmap is bolted in the hash table.
186 *
187 * On Book3E CPUs, the vmemmap is currently mapped in the top half of
188 * the vmalloc space using normal page tables, though the size of
189 * pages encoded in the PTEs can be different
190 */
191
192 #ifdef CONFIG_PPC_BOOK3E
vmemmap_create_mapping(unsigned long start,unsigned long page_size,unsigned long phys)193 static void __meminit vmemmap_create_mapping(unsigned long start,
194 unsigned long page_size,
195 unsigned long phys)
196 {
197 /* Create a PTE encoding without page size */
198 unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
199 _PAGE_KERNEL_RW;
200
201 /* PTEs only contain page size encodings up to 32M */
202 BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
203
204 /* Encode the size in the PTE */
205 flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
206
207 /* For each PTE for that area, map things. Note that we don't
208 * increment phys because all PTEs are of the large size and
209 * thus must have the low bits clear
210 */
211 for (i = 0; i < page_size; i += PAGE_SIZE)
212 BUG_ON(map_kernel_page(start + i, phys, flags));
213 }
214 #else /* CONFIG_PPC_BOOK3E */
vmemmap_create_mapping(unsigned long start,unsigned long page_size,unsigned long phys)215 static void __meminit vmemmap_create_mapping(unsigned long start,
216 unsigned long page_size,
217 unsigned long phys)
218 {
219 int mapped = htab_bolt_mapping(start, start + page_size, phys,
220 PAGE_KERNEL, mmu_vmemmap_psize,
221 mmu_kernel_ssize);
222 BUG_ON(mapped < 0);
223 }
224 #endif /* CONFIG_PPC_BOOK3E */
225
226 struct vmemmap_backing *vmemmap_list;
227
vmemmap_list_alloc(int node)228 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
229 {
230 static struct vmemmap_backing *next;
231 static int num_left;
232
233 /* allocate a page when required and hand out chunks */
234 if (!next || !num_left) {
235 next = vmemmap_alloc_block(PAGE_SIZE, node);
236 if (unlikely(!next)) {
237 WARN_ON(1);
238 return NULL;
239 }
240 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
241 }
242
243 num_left--;
244
245 return next++;
246 }
247
vmemmap_list_populate(unsigned long phys,unsigned long start,int node)248 static __meminit void vmemmap_list_populate(unsigned long phys,
249 unsigned long start,
250 int node)
251 {
252 struct vmemmap_backing *vmem_back;
253
254 vmem_back = vmemmap_list_alloc(node);
255 if (unlikely(!vmem_back)) {
256 WARN_ON(1);
257 return;
258 }
259
260 vmem_back->phys = phys;
261 vmem_back->virt_addr = start;
262 vmem_back->list = vmemmap_list;
263
264 vmemmap_list = vmem_back;
265 }
266
vmemmap_populate(struct page * start_page,unsigned long nr_pages,int node)267 int __meminit vmemmap_populate(struct page *start_page,
268 unsigned long nr_pages, int node)
269 {
270 unsigned long start = (unsigned long)start_page;
271 unsigned long end = (unsigned long)(start_page + nr_pages);
272 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
273
274 /* Align to the page size of the linear mapping. */
275 start = _ALIGN_DOWN(start, page_size);
276
277 pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
278 start_page, nr_pages, node);
279 pr_debug(" -> map %lx..%lx\n", start, end);
280
281 for (; start < end; start += page_size) {
282 void *p;
283
284 if (vmemmap_populated(start, page_size))
285 continue;
286
287 p = vmemmap_alloc_block(page_size, node);
288 if (!p)
289 return -ENOMEM;
290
291 vmemmap_list_populate(__pa(p), start, node);
292
293 pr_debug(" * %016lx..%016lx allocated at %p\n",
294 start, start + page_size, p);
295
296 vmemmap_create_mapping(start, page_size, __pa(p));
297 }
298
299 return 0;
300 }
301 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
302
303