xref: /openbmc/linux/arch/powerpc/mm/init_64.c (revision 7dd65feb)
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/lmb.h>
44 #include <linux/hugetlb.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 <asm/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/system.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 phys_addr_t kernstart_addr;
82 
83 void free_initmem(void)
84 {
85 	unsigned long addr;
86 
87 	addr = (unsigned long)__init_begin;
88 	for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
89 		memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
90 		ClearPageReserved(virt_to_page(addr));
91 		init_page_count(virt_to_page(addr));
92 		free_page(addr);
93 		totalram_pages++;
94 	}
95 	printk ("Freeing unused kernel memory: %luk freed\n",
96 		((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
97 }
98 
99 #ifdef CONFIG_BLK_DEV_INITRD
100 void free_initrd_mem(unsigned long start, unsigned long end)
101 {
102 	if (start < end)
103 		printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
104 	for (; start < end; start += PAGE_SIZE) {
105 		ClearPageReserved(virt_to_page(start));
106 		init_page_count(virt_to_page(start));
107 		free_page(start);
108 		totalram_pages++;
109 	}
110 }
111 #endif
112 
113 static void pgd_ctor(void *addr)
114 {
115 	memset(addr, 0, PGD_TABLE_SIZE);
116 }
117 
118 static void pmd_ctor(void *addr)
119 {
120 	memset(addr, 0, PMD_TABLE_SIZE);
121 }
122 
123 struct kmem_cache *pgtable_cache[MAX_PGTABLE_INDEX_SIZE];
124 
125 /*
126  * Create a kmem_cache() for pagetables.  This is not used for PTE
127  * pages - they're linked to struct page, come from the normal free
128  * pages pool and have a different entry size (see real_pte_t) to
129  * everything else.  Caches created by this function are used for all
130  * the higher level pagetables, and for hugepage pagetables.
131  */
132 void pgtable_cache_add(unsigned shift, void (*ctor)(void *))
133 {
134 	char *name;
135 	unsigned long table_size = sizeof(void *) << shift;
136 	unsigned long align = table_size;
137 
138 	/* When batching pgtable pointers for RCU freeing, we store
139 	 * the index size in the low bits.  Table alignment must be
140 	 * big enough to fit it.
141 	 *
142 	 * Likewise, hugeapge pagetable pointers contain a (different)
143 	 * shift value in the low bits.  All tables must be aligned so
144 	 * as to leave enough 0 bits in the address to contain it. */
145 	unsigned long minalign = max(MAX_PGTABLE_INDEX_SIZE + 1,
146 				     HUGEPD_SHIFT_MASK + 1);
147 	struct kmem_cache *new;
148 
149 	/* It would be nice if this was a BUILD_BUG_ON(), but at the
150 	 * moment, gcc doesn't seem to recognize is_power_of_2 as a
151 	 * constant expression, so so much for that. */
152 	BUG_ON(!is_power_of_2(minalign));
153 	BUG_ON((shift < 1) || (shift > MAX_PGTABLE_INDEX_SIZE));
154 
155 	if (PGT_CACHE(shift))
156 		return; /* Already have a cache of this size */
157 
158 	align = max_t(unsigned long, align, minalign);
159 	name = kasprintf(GFP_KERNEL, "pgtable-2^%d", shift);
160 	new = kmem_cache_create(name, table_size, align, 0, ctor);
161 	PGT_CACHE(shift) = new;
162 
163 	pr_debug("Allocated pgtable cache for order %d\n", shift);
164 }
165 
166 
167 void pgtable_cache_init(void)
168 {
169 	pgtable_cache_add(PGD_INDEX_SIZE, pgd_ctor);
170 	pgtable_cache_add(PMD_INDEX_SIZE, pmd_ctor);
171 	if (!PGT_CACHE(PGD_INDEX_SIZE) || !PGT_CACHE(PMD_INDEX_SIZE))
172 		panic("Couldn't allocate pgtable caches");
173 
174 	/* In all current configs, when the PUD index exists it's the
175 	 * same size as either the pgd or pmd index.  Verify that the
176 	 * initialization above has also created a PUD cache.  This
177 	 * will need re-examiniation if we add new possibilities for
178 	 * the pagetable layout. */
179 	BUG_ON(PUD_INDEX_SIZE && !PGT_CACHE(PUD_INDEX_SIZE));
180 }
181 
182 #ifdef CONFIG_SPARSEMEM_VMEMMAP
183 /*
184  * Given an address within the vmemmap, determine the pfn of the page that
185  * represents the start of the section it is within.  Note that we have to
186  * do this by hand as the proffered address may not be correctly aligned.
187  * Subtraction of non-aligned pointers produces undefined results.
188  */
189 static unsigned long __meminit vmemmap_section_start(unsigned long page)
190 {
191 	unsigned long offset = page - ((unsigned long)(vmemmap));
192 
193 	/* Return the pfn of the start of the section. */
194 	return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
195 }
196 
197 /*
198  * Check if this vmemmap page is already initialised.  If any section
199  * which overlaps this vmemmap page is initialised then this page is
200  * initialised already.
201  */
202 static int __meminit vmemmap_populated(unsigned long start, int page_size)
203 {
204 	unsigned long end = start + page_size;
205 
206 	for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
207 		if (pfn_valid(vmemmap_section_start(start)))
208 			return 1;
209 
210 	return 0;
211 }
212 
213 /* On hash-based CPUs, the vmemmap is bolted in the hash table.
214  *
215  * On Book3E CPUs, the vmemmap is currently mapped in the top half of
216  * the vmalloc space using normal page tables, though the size of
217  * pages encoded in the PTEs can be different
218  */
219 
220 #ifdef CONFIG_PPC_BOOK3E
221 static void __meminit vmemmap_create_mapping(unsigned long start,
222 					     unsigned long page_size,
223 					     unsigned long phys)
224 {
225 	/* Create a PTE encoding without page size */
226 	unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
227 		_PAGE_KERNEL_RW;
228 
229 	/* PTEs only contain page size encodings up to 32M */
230 	BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].enc > 0xf);
231 
232 	/* Encode the size in the PTE */
233 	flags |= mmu_psize_defs[mmu_vmemmap_psize].enc << 8;
234 
235 	/* For each PTE for that area, map things. Note that we don't
236 	 * increment phys because all PTEs are of the large size and
237 	 * thus must have the low bits clear
238 	 */
239 	for (i = 0; i < page_size; i += PAGE_SIZE)
240 		BUG_ON(map_kernel_page(start + i, phys, flags));
241 }
242 #else /* CONFIG_PPC_BOOK3E */
243 static void __meminit vmemmap_create_mapping(unsigned long start,
244 					     unsigned long page_size,
245 					     unsigned long phys)
246 {
247 	int  mapped = htab_bolt_mapping(start, start + page_size, phys,
248 					PAGE_KERNEL, mmu_vmemmap_psize,
249 					mmu_kernel_ssize);
250 	BUG_ON(mapped < 0);
251 }
252 #endif /* CONFIG_PPC_BOOK3E */
253 
254 int __meminit vmemmap_populate(struct page *start_page,
255 			       unsigned long nr_pages, int node)
256 {
257 	unsigned long start = (unsigned long)start_page;
258 	unsigned long end = (unsigned long)(start_page + nr_pages);
259 	unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
260 
261 	/* Align to the page size of the linear mapping. */
262 	start = _ALIGN_DOWN(start, page_size);
263 
264 	pr_debug("vmemmap_populate page %p, %ld pages, node %d\n",
265 		 start_page, nr_pages, node);
266 	pr_debug(" -> map %lx..%lx\n", start, end);
267 
268 	for (; start < end; start += page_size) {
269 		void *p;
270 
271 		if (vmemmap_populated(start, page_size))
272 			continue;
273 
274 		p = vmemmap_alloc_block(page_size, node);
275 		if (!p)
276 			return -ENOMEM;
277 
278 		pr_debug("      * %016lx..%016lx allocated at %p\n",
279 			 start, start + page_size, p);
280 
281 		vmemmap_create_mapping(start, page_size, __pa(p));
282 	}
283 
284 	return 0;
285 }
286 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
287