xref: /openbmc/linux/arch/powerpc/mm/init_64.c (revision 93dc544c)
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 
45 #include <asm/pgalloc.h>
46 #include <asm/page.h>
47 #include <asm/prom.h>
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>
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>
66 
67 #include "mmu_decl.h"
68 
69 #if PGTABLE_RANGE > USER_VSID_RANGE
70 #warning Limited user VSID range means pagetable space is wasted
71 #endif
72 
73 #if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
74 #warning TASK_SIZE is smaller than it needs to be.
75 #endif
76 
77 phys_addr_t memstart_addr = ~0;
78 phys_addr_t kernstart_addr;
79 
80 void free_initmem(void)
81 {
82 	unsigned long addr;
83 
84 	addr = (unsigned long)__init_begin;
85 	for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
86 		memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
87 		ClearPageReserved(virt_to_page(addr));
88 		init_page_count(virt_to_page(addr));
89 		free_page(addr);
90 		totalram_pages++;
91 	}
92 	printk ("Freeing unused kernel memory: %luk freed\n",
93 		((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
94 }
95 
96 #ifdef CONFIG_BLK_DEV_INITRD
97 void free_initrd_mem(unsigned long start, unsigned long end)
98 {
99 	if (start < end)
100 		printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
101 	for (; start < end; start += PAGE_SIZE) {
102 		ClearPageReserved(virt_to_page(start));
103 		init_page_count(virt_to_page(start));
104 		free_page(start);
105 		totalram_pages++;
106 	}
107 }
108 #endif
109 
110 #ifdef CONFIG_PROC_KCORE
111 static struct kcore_list kcore_vmem;
112 
113 static int __init setup_kcore(void)
114 {
115 	int i;
116 
117 	for (i=0; i < lmb.memory.cnt; i++) {
118 		unsigned long base, size;
119 		struct kcore_list *kcore_mem;
120 
121 		base = lmb.memory.region[i].base;
122 		size = lmb.memory.region[i].size;
123 
124 		/* GFP_ATOMIC to avoid might_sleep warnings during boot */
125 		kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
126 		if (!kcore_mem)
127 			panic("%s: kmalloc failed\n", __func__);
128 
129 		kclist_add(kcore_mem, __va(base), size);
130 	}
131 
132 	kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
133 
134 	return 0;
135 }
136 module_init(setup_kcore);
137 #endif
138 
139 static void pgd_ctor(void *addr)
140 {
141 	memset(addr, 0, PGD_TABLE_SIZE);
142 }
143 
144 static void pmd_ctor(void *addr)
145 {
146 	memset(addr, 0, PMD_TABLE_SIZE);
147 }
148 
149 static const unsigned int pgtable_cache_size[2] = {
150 	PGD_TABLE_SIZE, PMD_TABLE_SIZE
151 };
152 static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = {
153 #ifdef CONFIG_PPC_64K_PAGES
154 	"pgd_cache", "pmd_cache",
155 #else
156 	"pgd_cache", "pud_pmd_cache",
157 #endif /* CONFIG_PPC_64K_PAGES */
158 };
159 
160 #ifdef CONFIG_HUGETLB_PAGE
161 /* Hugepages need an extra cache per hugepagesize, initialized in
162  * hugetlbpage.c.  We can't put into the tables above, because HPAGE_SHIFT
163  * is not compile time constant. */
164 struct kmem_cache *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)+MMU_PAGE_COUNT];
165 #else
166 struct kmem_cache *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)];
167 #endif
168 
169 void pgtable_cache_init(void)
170 {
171 	pgtable_cache[0] = kmem_cache_create(pgtable_cache_name[0], PGD_TABLE_SIZE, PGD_TABLE_SIZE, SLAB_PANIC, pgd_ctor);
172 	pgtable_cache[1] = kmem_cache_create(pgtable_cache_name[1], PMD_TABLE_SIZE, PMD_TABLE_SIZE, SLAB_PANIC, pmd_ctor);
173 }
174 
175 #ifdef CONFIG_SPARSEMEM_VMEMMAP
176 /*
177  * Given an address within the vmemmap, determine the pfn of the page that
178  * represents the start of the section it is within.  Note that we have to
179  * do this by hand as the proffered address may not be correctly aligned.
180  * Subtraction of non-aligned pointers produces undefined results.
181  */
182 static unsigned long __meminit vmemmap_section_start(unsigned long page)
183 {
184 	unsigned long offset = page - ((unsigned long)(vmemmap));
185 
186 	/* Return the pfn of the start of the section. */
187 	return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
188 }
189 
190 /*
191  * Check if this vmemmap page is already initialised.  If any section
192  * which overlaps this vmemmap page is initialised then this page is
193  * initialised already.
194  */
195 static int __meminit vmemmap_populated(unsigned long start, int page_size)
196 {
197 	unsigned long end = start + page_size;
198 
199 	for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
200 		if (pfn_valid(vmemmap_section_start(start)))
201 			return 1;
202 
203 	return 0;
204 }
205 
206 int __meminit vmemmap_populate(struct page *start_page,
207 			       unsigned long nr_pages, int node)
208 {
209 	unsigned long mode_rw;
210 	unsigned long start = (unsigned long)start_page;
211 	unsigned long end = (unsigned long)(start_page + nr_pages);
212 	unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
213 
214 	mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
215 
216 	/* Align to the page size of the linear mapping. */
217 	start = _ALIGN_DOWN(start, page_size);
218 
219 	for (; start < end; start += page_size) {
220 		int mapped;
221 		void *p;
222 
223 		if (vmemmap_populated(start, page_size))
224 			continue;
225 
226 		p = vmemmap_alloc_block(page_size, node);
227 		if (!p)
228 			return -ENOMEM;
229 
230 		pr_debug("vmemmap %08lx allocated at %p, physical %08lx.\n",
231 			start, p, __pa(p));
232 
233 		mapped = htab_bolt_mapping(start, start + page_size,
234 					__pa(p), mode_rw, mmu_vmemmap_psize,
235 					mmu_kernel_ssize);
236 		BUG_ON(mapped < 0);
237 	}
238 
239 	return 0;
240 }
241 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
242