xref: /openbmc/linux/arch/powerpc/mm/init_64.c (revision 05bcf503)
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/vdso.h>
66 
67 #include "mmu_decl.h"
68 
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
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
77 #endif /* CONFIG_PPC_STD_MMU_64 */
78 
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);
83 
84 static void pgd_ctor(void *addr)
85 {
86 	memset(addr, 0, PGD_TABLE_SIZE);
87 }
88 
89 static void pmd_ctor(void *addr)
90 {
91 	memset(addr, 0, PMD_TABLE_SIZE);
92 }
93 
94 struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
95 
96 /*
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.
102  */
103 void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
104 {
105 	char *name;
106 	unsigned long table_size = sizeof(void *) << shift;
107 	unsigned long align = table_size;
108 
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.
112 	 *
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;
119 
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));
125 
126 	if (PGT_CACHE(shift))
127 		return; /* Already have a cache of this size */
128 
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 	PGT_CACHE(shift) = new;
133 
134 	pr_debug("Allocated pgtable cache for order %d\n", shift);
135 }
136 
137 
138 void pgtable_cache_init(void)
139 {
140 	pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
141 	pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
142 	if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
143 		panic("Couldn't allocate pgtable caches");
144 
145 	/* In all current configs, when the PUD index exists it's the
146 	 * same size as either the pgd or pmd index.  Verify that the
147 	 * initialization above has also created a PUD cache.  This
148 	 * will need re-examiniation if we add new possibilities for
149 	 * the pagetable layout. */
150 	BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
151 }
152 
153 #ifdef CONFIG_SPARSEMEM_VMEMMAP
154 /*
155  * Given an address within the vmemmap, determine the pfn of the page that
156  * represents the start of the section it is within.  Note that we have to
157  * do this by hand as the proffered address may not be correctly aligned.
158  * Subtraction of non-aligned pointers produces undefined results.
159  */
160 static unsigned long __meminit vmemmap_section_start(unsigned long page)
161 {
162 	unsigned long offset = page - ((unsigned long)(vmemmap));
163 
164 	/* Return the pfn of the start of the section. */
165 	return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
166 }
167 
168 /*
169  * Check if this vmemmap page is already initialised.  If any section
170  * which overlaps this vmemmap page is initialised then this page is
171  * initialised already.
172  */
173 static int __meminit vmemmap_populated(unsigned long start, int page_size)
174 {
175 	unsigned long end = start + page_size;
176 
177 	for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
178 		if (pfn_valid(vmemmap_section_start(start)))
179 			return 1;
180 
181 	return 0;
182 }
183 
184 /* On hash-based CPUs, the vmemmap is bolted in the hash table.
185  *
186  * On Book3E CPUs, the vmemmap is currently mapped in the top half of
187  * the vmalloc space using normal page tables, though the size of
188  * pages encoded in the PTEs can be different
189  */
190 
191 #ifdef CONFIG_PPC_BOOK3E
192 static void __meminit vmemmap_create_mapping(unsigned long start,
193 					     unsigned long page_size,
194 					     unsigned long phys)
195 {
196 	/* Create a PTE encoding without page size */
197 	unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
198 		_PAGE_KERNEL_RW;
199 
200 	/* PTEs only contain page size encodings up to 32M */
201 	BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
202 
203 	/* Encode the size in the PTE */
204 	flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
205 
206 	/* For each PTE for that area, map things. Note that we don't
207 	 * increment phys because all PTEs are of the large size and
208 	 * thus must have the low bits clear
209 	 */
210 	for (i = 0; i < page_size; i += PAGE_SIZE)
211 		BUG_ON(map_kernel_page(start + i, phys, flags));
212 }
213 #else /* CONFIG_PPC_BOOK3E */
214 static void __meminit vmemmap_create_mapping(unsigned long start,
215 					     unsigned long page_size,
216 					     unsigned long phys)
217 {
218 	int  mapped = htab_bolt_mapping(start, start + page_size, phys,
219 					PAGE_KERNEL, mmu_vmemmap_psize,
220 					mmu_kernel_ssize);
221 	BUG_ON(mapped < 0);
222 }
223 #endif /* CONFIG_PPC_BOOK3E */
224 
225 struct vmemmap_backing *vmemmap_list;
226 
227 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
228 {
229 	static struct vmemmap_backing *next;
230 	static int num_left;
231 
232 	/* allocate a page when required and hand out chunks */
233 	if (!next || !num_left) {
234 		next = vmemmap_alloc_block(PAGE_SIZE, node);
235 		if (unlikely(!next)) {
236 			WARN_ON(1);
237 			return NULL;
238 		}
239 		num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
240 	}
241 
242 	num_left--;
243 
244 	return next++;
245 }
246 
247 static __meminit void vmemmap_list_populate(unsigned long phys,
248 					    unsigned long start,
249 					    int node)
250 {
251 	struct vmemmap_backing *vmem_back;
252 
253 	vmem_back = vmemmap_list_alloc(node);
254 	if (unlikely(!vmem_back)) {
255 		WARN_ON(1);
256 		return;
257 	}
258 
259 	vmem_back->phys = phys;
260 	vmem_back->virt_addr = start;
261 	vmem_back->list = vmemmap_list;
262 
263 	vmemmap_list = vmem_back;
264 }
265 
266 int __meminit vmemmap_populate(struct page *start_page,
267 			       unsigned long nr_pages, int node)
268 {
269 	unsigned long start = (unsigned long)start_page;
270 	unsigned long end = (unsigned long)(start_page + nr_pages);
271 	unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
272 
273 	/* Align to the page size of the linear mapping. */
274 	start = _ALIGN_DOWN(start, page_size);
275 
276 	pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
277 		 start_page, nr_pages, node);
278 	pr_debug(" -> map %lx..%lx\n", start, end);
279 
280 	for (; start < end; start += page_size) {
281 		void *p;
282 
283 		if (vmemmap_populated(start, page_size))
284 			continue;
285 
286 		p = vmemmap_alloc_block(page_size, node);
287 		if (!p)
288 			return -ENOMEM;
289 
290 		vmemmap_list_populate(__pa(p), start, node);
291 
292 		pr_debug("      * %016lx..%016lx allocated at %p\n",
293 			 start, start + page_size, p);
294 
295 		vmemmap_create_mapping(start, page_size, __pa(p));
296 	}
297 
298 	return 0;
299 }
300 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
301 
302