]>
Commit | Line | Data |
---|---|---|
14cf11af PM |
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 | |
14cf11af PM |
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 | ||
cec08e7a BH |
22 | #undef DEBUG |
23 | ||
14cf11af PM |
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> | |
c9cf5528 | 42 | #include <linux/poison.h> |
d9b2b2a2 | 43 | #include <linux/lmb.h> |
14cf11af PM |
44 | |
45 | #include <asm/pgalloc.h> | |
46 | #include <asm/page.h> | |
47 | #include <asm/prom.h> | |
14cf11af PM |
48 | #include <asm/rtas.h> |
49 | #include <asm/io.h> | |
50 | #include <asm/mmu_context.h> | |
51 | #include <asm/pgtable.h> | |
52 | #include <asm/mmu.h> | |
53 | #include <asm/uaccess.h> | |
54 | #include <asm/smp.h> | |
55 | #include <asm/machdep.h> | |
56 | #include <asm/tlb.h> | |
57 | #include <asm/eeh.h> | |
58 | #include <asm/processor.h> | |
59 | #include <asm/mmzone.h> | |
60 | #include <asm/cputable.h> | |
14cf11af PM |
61 | #include <asm/sections.h> |
62 | #include <asm/system.h> | |
63 | #include <asm/iommu.h> | |
64 | #include <asm/abs_addr.h> | |
65 | #include <asm/vdso.h> | |
800fc3ee DG |
66 | |
67 | #include "mmu_decl.h" | |
14cf11af | 68 | |
94491685 | 69 | #ifdef CONFIG_PPC_STD_MMU_64 |
14cf11af PM |
70 | #if PGTABLE_RANGE > USER_VSID_RANGE |
71 | #warning Limited user VSID range means pagetable space is wasted | |
72 | #endif | |
73 | ||
74 | #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE) | |
75 | #warning TASK_SIZE is smaller than it needs to be. | |
76 | #endif | |
94491685 | 77 | #endif /* CONFIG_PPC_STD_MMU_64 */ |
14cf11af | 78 | |
37dd2bad KG |
79 | phys_addr_t memstart_addr = ~0; |
80 | phys_addr_t kernstart_addr; | |
d7917ba7 | 81 | |
14cf11af PM |
82 | void free_initmem(void) |
83 | { | |
84 | unsigned long addr; | |
85 | ||
86 | addr = (unsigned long)__init_begin; | |
87 | for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) { | |
c9cf5528 | 88 | memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE); |
14cf11af | 89 | ClearPageReserved(virt_to_page(addr)); |
7835e98b | 90 | init_page_count(virt_to_page(addr)); |
14cf11af PM |
91 | free_page(addr); |
92 | totalram_pages++; | |
93 | } | |
94 | printk ("Freeing unused kernel memory: %luk freed\n", | |
95 | ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10); | |
96 | } | |
97 | ||
98 | #ifdef CONFIG_BLK_DEV_INITRD | |
99 | void free_initrd_mem(unsigned long start, unsigned long end) | |
100 | { | |
101 | if (start < end) | |
102 | printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10); | |
103 | for (; start < end; start += PAGE_SIZE) { | |
104 | ClearPageReserved(virt_to_page(start)); | |
7835e98b | 105 | init_page_count(virt_to_page(start)); |
14cf11af PM |
106 | free_page(start); |
107 | totalram_pages++; | |
108 | } | |
109 | } | |
110 | #endif | |
111 | ||
51cc5068 | 112 | static void pgd_ctor(void *addr) |
14cf11af | 113 | { |
51cc5068 AD |
114 | memset(addr, 0, PGD_TABLE_SIZE); |
115 | } | |
116 | ||
117 | static void pmd_ctor(void *addr) | |
118 | { | |
119 | memset(addr, 0, PMD_TABLE_SIZE); | |
14cf11af PM |
120 | } |
121 | ||
a0668cdc DG |
122 | struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE]; |
123 | ||
124 | /* | |
125 | * Create a kmem_cache() for pagetables. This is not used for PTE | |
126 | * pages - they're linked to struct page, come from the normal free | |
127 | * pages pool and have a different entry size (see real_pte_t) to | |
128 | * everything else. Caches created by this function are used for all | |
129 | * the higher level pagetables, and for hugepage pagetables. | |
130 | */ | |
131 | void pgtable_cache_add(unsigned shift, void (*ctor)(void *)) | |
132 | { | |
133 | char *name; | |
134 | unsigned long table_size = sizeof(void *) << shift; | |
135 | unsigned long align = table_size; | |
136 | ||
137 | /* When batching pgtable pointers for RCU freeing, we store | |
138 | * the index size in the low bits. Table alignment must be | |
139 | * big enough to fit it */ | |
140 | unsigned long minalign = MAX_PGTABLE_INDEX_SIZE + 1; | |
141 | struct kmem_cache *new; | |
142 | ||
143 | /* It would be nice if this was a BUILD_BUG_ON(), but at the | |
144 | * moment, gcc doesn't seem to recognize is_power_of_2 as a | |
145 | * constant expression, so so much for that. */ | |
146 | BUG_ON(!is_power_of_2(minalign)); | |
147 | BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE)); | |
148 | ||
149 | if (PGT_CACHE(shift)) | |
150 | return; /* Already have a cache of this size */ | |
151 | ||
152 | align = max_t(unsigned long, align, minalign); | |
153 | name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift); | |
154 | new = kmem_cache_create(name, table_size, align, 0, ctor); | |
155 | PGT_CACHE(shift) = new; | |
156 | ||
157 | pr_debug("Allocated pgtable cache for order %d\n", shift); | |
158 | } | |
159 | ||
14cf11af PM |
160 | |
161 | void pgtable_cache_init(void) | |
162 | { | |
a0668cdc DG |
163 | pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor); |
164 | pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor); | |
165 | if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE)) | |
166 | panic("Couldn't allocate pgtable caches"); | |
167 | ||
168 | /* In all current configs, when the PUD index exists it's the | |
169 | * same size as either the pgd or pmd index. Verify that the | |
170 | * initialization above has also created a PUD cache. This | |
171 | * will need re-examiniation if we add new possibilities for | |
172 | * the pagetable layout. */ | |
173 | BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE)); | |
14cf11af | 174 | } |
d29eff7b AW |
175 | |
176 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
177 | /* | |
178 | * Given an address within the vmemmap, determine the pfn of the page that | |
179 | * represents the start of the section it is within. Note that we have to | |
180 | * do this by hand as the proffered address may not be correctly aligned. | |
181 | * Subtraction of non-aligned pointers produces undefined results. | |
182 | */ | |
09de9ff8 | 183 | static unsigned long __meminit vmemmap_section_start(unsigned long page) |
d29eff7b AW |
184 | { |
185 | unsigned long offset = page - ((unsigned long)(vmemmap)); | |
186 | ||
187 | /* Return the pfn of the start of the section. */ | |
188 | return (offset / sizeof(struct page)) & PAGE_SECTION_MASK; | |
189 | } | |
190 | ||
191 | /* | |
192 | * Check if this vmemmap page is already initialised. If any section | |
193 | * which overlaps this vmemmap page is initialised then this page is | |
194 | * initialised already. | |
195 | */ | |
09de9ff8 | 196 | static int __meminit vmemmap_populated(unsigned long start, int page_size) |
d29eff7b AW |
197 | { |
198 | unsigned long end = start + page_size; | |
199 | ||
200 | for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page))) | |
201 | if (pfn_valid(vmemmap_section_start(start))) | |
202 | return 1; | |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
32a74949 BH |
207 | /* On hash-based CPUs, the vmemmap is bolted in the hash table. |
208 | * | |
209 | * On Book3E CPUs, the vmemmap is currently mapped in the top half of | |
210 | * the vmalloc space using normal page tables, though the size of | |
211 | * pages encoded in the PTEs can be different | |
212 | */ | |
213 | ||
214 | #ifdef CONFIG_PPC_BOOK3E | |
215 | static void __meminit vmemmap_create_mapping(unsigned long start, | |
216 | unsigned long page_size, | |
217 | unsigned long phys) | |
218 | { | |
219 | /* Create a PTE encoding without page size */ | |
220 | unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED | | |
221 | _PAGE_KERNEL_RW; | |
222 | ||
223 | /* PTEs only contain page size encodings up to 32M */ | |
224 | BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf); | |
225 | ||
226 | /* Encode the size in the PTE */ | |
227 | flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8; | |
228 | ||
229 | /* For each PTE for that area, map things. Note that we don't | |
230 | * increment phys because all PTEs are of the large size and | |
231 | * thus must have the low bits clear | |
232 | */ | |
233 | for (i = 0; i < page_size; i += PAGE_SIZE) | |
234 | BUG_ON(map_kernel_page(start + i, phys, flags)); | |
235 | } | |
236 | #else /* CONFIG_PPC_BOOK3E */ | |
237 | static void __meminit vmemmap_create_mapping(unsigned long start, | |
238 | unsigned long page_size, | |
239 | unsigned long phys) | |
240 | { | |
241 | int mapped = htab_bolt_mapping(start, start + page_size, phys, | |
242 | PAGE_KERNEL, mmu_vmemmap_psize, | |
243 | mmu_kernel_ssize); | |
244 | BUG_ON(mapped < 0); | |
245 | } | |
246 | #endif /* CONFIG_PPC_BOOK3E */ | |
247 | ||
d29eff7b | 248 | int __meminit vmemmap_populate(struct page *start_page, |
cec08e7a | 249 | unsigned long nr_pages, int node) |
d29eff7b | 250 | { |
d29eff7b AW |
251 | unsigned long start = (unsigned long)start_page; |
252 | unsigned long end = (unsigned long)(start_page + nr_pages); | |
cec08e7a | 253 | unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; |
d29eff7b | 254 | |
d29eff7b AW |
255 | /* Align to the page size of the linear mapping. */ |
256 | start = _ALIGN_DOWN(start, page_size); | |
257 | ||
32a74949 BH |
258 | pr_debug("vmemmap_populate page %p, %ld pages, node %d\n", |
259 | start_page, nr_pages, node); | |
260 | pr_debug(" -> map %lx..%lx\n", start, end); | |
261 | ||
d29eff7b | 262 | for (; start < end; start += page_size) { |
d29eff7b AW |
263 | void *p; |
264 | ||
265 | if (vmemmap_populated(start, page_size)) | |
266 | continue; | |
267 | ||
268 | p = vmemmap_alloc_block(page_size, node); | |
269 | if (!p) | |
270 | return -ENOMEM; | |
271 | ||
32a74949 BH |
272 | pr_debug(" * %016lx..%016lx allocated at %p\n", |
273 | start, start + page_size, p); | |
d29eff7b | 274 | |
32a74949 | 275 | vmemmap_create_mapping(start, page_size, __pa(p)); |
d29eff7b AW |
276 | } |
277 | ||
278 | return 0; | |
279 | } | |
cec08e7a | 280 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |