powerpc/mm: Use page fragments for allocation page table at PMD level
[linux-2.6-block.git] / arch / powerpc / mm / pgtable_64.c
CommitLineData
14cf11af
PM
1/*
2 * This file contains ioremap and related functions for 64-bit machines.
3 *
4 * Derived from arch/ppc64/mm/init.c
5 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6 *
7 * Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
8 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
9 * Copyright (C) 1996 Paul Mackerras
14cf11af
PM
10 *
11 * Derived from "arch/i386/mm/init.c"
12 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
13 *
14 * Dave Engebretsen <engebret@us.ibm.com>
15 * Rework for PPC64 port.
16 *
17 * This program is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU General Public License
19 * as published by the Free Software Foundation; either version
20 * 2 of the License, or (at your option) any later version.
21 *
22 */
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>
66b15db6 29#include <linux/export.h>
14cf11af
PM
30#include <linux/types.h>
31#include <linux/mman.h>
32#include <linux/mm.h>
33#include <linux/swap.h>
34#include <linux/stddef.h>
35#include <linux/vmalloc.h>
5a0e3ad6 36#include <linux/slab.h>
06743521 37#include <linux/hugetlb.h>
14cf11af
PM
38
39#include <asm/pgalloc.h>
40#include <asm/page.h>
41#include <asm/prom.h>
14cf11af
PM
42#include <asm/io.h>
43#include <asm/mmu_context.h>
44#include <asm/pgtable.h>
45#include <asm/mmu.h>
14cf11af
PM
46#include <asm/smp.h>
47#include <asm/machdep.h>
48#include <asm/tlb.h>
14cf11af 49#include <asm/processor.h>
14cf11af 50#include <asm/cputable.h>
14cf11af 51#include <asm/sections.h>
5e203d68 52#include <asm/firmware.h>
68cf0d64 53#include <asm/dma.h>
800fc3ee
DG
54
55#include "mmu_decl.h"
14cf11af 56
14cf11af 57
50de596d
AK
58#ifdef CONFIG_PPC_BOOK3S_64
59/*
60 * partition table and process table for ISA 3.0
61 */
62struct prtb_entry *process_tb;
63struct patb_entry *partition_tb;
dd1842a2
AK
64/*
65 * page table size
66 */
67unsigned long __pte_index_size;
68EXPORT_SYMBOL(__pte_index_size);
69unsigned long __pmd_index_size;
70EXPORT_SYMBOL(__pmd_index_size);
71unsigned long __pud_index_size;
72EXPORT_SYMBOL(__pud_index_size);
73unsigned long __pgd_index_size;
74EXPORT_SYMBOL(__pgd_index_size);
fae22116
AK
75unsigned long __pud_cache_index;
76EXPORT_SYMBOL(__pud_cache_index);
dd1842a2
AK
77unsigned long __pte_table_size;
78EXPORT_SYMBOL(__pte_table_size);
79unsigned long __pmd_table_size;
80EXPORT_SYMBOL(__pmd_table_size);
81unsigned long __pud_table_size;
82EXPORT_SYMBOL(__pud_table_size);
83unsigned long __pgd_table_size;
84EXPORT_SYMBOL(__pgd_table_size);
a2f41eb9
AK
85unsigned long __pmd_val_bits;
86EXPORT_SYMBOL(__pmd_val_bits);
87unsigned long __pud_val_bits;
88EXPORT_SYMBOL(__pud_val_bits);
89unsigned long __pgd_val_bits;
90EXPORT_SYMBOL(__pgd_val_bits);
d6a9996e
AK
91unsigned long __kernel_virt_start;
92EXPORT_SYMBOL(__kernel_virt_start);
93unsigned long __kernel_virt_size;
94EXPORT_SYMBOL(__kernel_virt_size);
95unsigned long __vmalloc_start;
96EXPORT_SYMBOL(__vmalloc_start);
97unsigned long __vmalloc_end;
98EXPORT_SYMBOL(__vmalloc_end);
63ee9b2f
ME
99unsigned long __kernel_io_start;
100EXPORT_SYMBOL(__kernel_io_start);
d6a9996e
AK
101struct page *vmemmap;
102EXPORT_SYMBOL(vmemmap);
5ed7ecd0
AK
103unsigned long __pte_frag_nr;
104EXPORT_SYMBOL(__pte_frag_nr);
105unsigned long __pte_frag_size_shift;
106EXPORT_SYMBOL(__pte_frag_size_shift);
d6a9996e
AK
107unsigned long ioremap_bot;
108#else /* !CONFIG_PPC_BOOK3S_64 */
78f1dbde 109unsigned long ioremap_bot = IOREMAP_BASE;
d6a9996e 110#endif
a245067e 111
3d5134ee
BH
112/**
113 * __ioremap_at - Low level function to establish the page tables
114 * for an IO mapping
115 */
116void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
14cf11af
PM
117 unsigned long flags)
118{
119 unsigned long i;
120
a1f242ff 121 /* Make sure we have the base flags */
14cf11af
PM
122 if ((flags & _PAGE_PRESENT) == 0)
123 flags |= pgprot_val(PAGE_KERNEL);
124
a1f242ff 125 /* We don't support the 4K PFN hack with ioremap */
945537df 126 if (flags & H_PAGE_4K_PFN)
a1f242ff
BH
127 return NULL;
128
3d5134ee
BH
129 WARN_ON(pa & ~PAGE_MASK);
130 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
131 WARN_ON(size & ~PAGE_MASK);
132
14cf11af 133 for (i = 0; i < size; i += PAGE_SIZE)
a245067e 134 if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
14cf11af
PM
135 return NULL;
136
3d5134ee
BH
137 return (void __iomem *)ea;
138}
139
140/**
141 * __iounmap_from - Low level function to tear down the page tables
142 * for an IO mapping. This is used for mappings that
143 * are manipulated manually, like partial unmapping of
144 * PCI IOs or ISA space.
145 */
146void __iounmap_at(void *ea, unsigned long size)
147{
148 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
149 WARN_ON(size & ~PAGE_MASK);
150
151 unmap_kernel_range((unsigned long)ea, size);
14cf11af
PM
152}
153
1cdab55d
BH
154void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
155 unsigned long flags, void *caller)
14cf11af 156{
3d5134ee 157 phys_addr_t paligned;
14cf11af
PM
158 void __iomem *ret;
159
160 /*
161 * Choose an address to map it to.
162 * Once the imalloc system is running, we use it.
163 * Before that, we map using addresses going
164 * up from ioremap_bot. imalloc will use
165 * the addresses from ioremap_bot through
166 * IMALLOC_END
167 *
168 */
3d5134ee
BH
169 paligned = addr & PAGE_MASK;
170 size = PAGE_ALIGN(addr + size) - paligned;
14cf11af 171
3d5134ee 172 if ((size == 0) || (paligned == 0))
14cf11af
PM
173 return NULL;
174
f691fa10 175 if (slab_is_available()) {
14cf11af 176 struct vm_struct *area;
3d5134ee 177
1cdab55d
BH
178 area = __get_vm_area_caller(size, VM_IOREMAP,
179 ioremap_bot, IOREMAP_END,
180 caller);
14cf11af
PM
181 if (area == NULL)
182 return NULL;
7a9d1256
ME
183
184 area->phys_addr = paligned;
3d5134ee 185 ret = __ioremap_at(paligned, area->addr, size, flags);
14cf11af 186 if (!ret)
3d5134ee 187 vunmap(area->addr);
14cf11af 188 } else {
3d5134ee 189 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
14cf11af
PM
190 if (ret)
191 ioremap_bot += size;
192 }
3d5134ee
BH
193
194 if (ret)
195 ret += addr & ~PAGE_MASK;
14cf11af
PM
196 return ret;
197}
198
1cdab55d
BH
199void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
200 unsigned long flags)
201{
202 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
203}
4cb3cee0 204
68a64357 205void __iomem * ioremap(phys_addr_t addr, unsigned long size)
4cb3cee0 206{
72176dd0 207 unsigned long flags = pgprot_val(pgprot_noncached(__pgprot(0)));
1cdab55d 208 void *caller = __builtin_return_address(0);
4cb3cee0
BH
209
210 if (ppc_md.ioremap)
1cdab55d
BH
211 return ppc_md.ioremap(addr, size, flags, caller);
212 return __ioremap_caller(addr, size, flags, caller);
4cb3cee0
BH
213}
214
be135f40
AB
215void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
216{
72176dd0 217 unsigned long flags = pgprot_val(pgprot_noncached_wc(__pgprot(0)));
be135f40
AB
218 void *caller = __builtin_return_address(0);
219
220 if (ppc_md.ioremap)
221 return ppc_md.ioremap(addr, size, flags, caller);
222 return __ioremap_caller(addr, size, flags, caller);
223}
224
40f1ce7f 225void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
4cb3cee0
BH
226 unsigned long flags)
227{
1cdab55d
BH
228 void *caller = __builtin_return_address(0);
229
a1f242ff 230 /* writeable implies dirty for kernel addresses */
c7d54842 231 if (flags & _PAGE_WRITE)
a1f242ff
BH
232 flags |= _PAGE_DIRTY;
233
ac29c640
AK
234 /* we don't want to let _PAGE_EXEC leak out */
235 flags &= ~_PAGE_EXEC;
236 /*
237 * Force kernel mapping.
238 */
ac29c640 239 flags &= ~_PAGE_USER;
812fadcb 240 flags |= _PAGE_PRIVILEGED;
55052eec 241
4cb3cee0 242 if (ppc_md.ioremap)
1cdab55d
BH
243 return ppc_md.ioremap(addr, size, flags, caller);
244 return __ioremap_caller(addr, size, flags, caller);
4cb3cee0
BH
245}
246
247
14cf11af
PM
248/*
249 * Unmap an IO region and remove it from imalloc'd list.
250 * Access to IO memory should be serialized by driver.
14cf11af 251 */
68a64357 252void __iounmap(volatile void __iomem *token)
14cf11af
PM
253{
254 void *addr;
255
f691fa10 256 if (!slab_is_available())
14cf11af
PM
257 return;
258
3d5134ee
BH
259 addr = (void *) ((unsigned long __force)
260 PCI_FIX_ADDR(token) & PAGE_MASK);
261 if ((unsigned long)addr < ioremap_bot) {
262 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
263 " at 0x%p\n", addr);
264 return;
265 }
266 vunmap(addr);
14cf11af
PM
267}
268
68a64357 269void iounmap(volatile void __iomem *token)
4cb3cee0
BH
270{
271 if (ppc_md.iounmap)
272 ppc_md.iounmap(token);
273 else
274 __iounmap(token);
275}
276
14cf11af 277EXPORT_SYMBOL(ioremap);
be135f40 278EXPORT_SYMBOL(ioremap_wc);
40f1ce7f 279EXPORT_SYMBOL(ioremap_prot);
14cf11af 280EXPORT_SYMBOL(__ioremap);
a302cb9d 281EXPORT_SYMBOL(__ioremap_at);
14cf11af 282EXPORT_SYMBOL(iounmap);
4cb3cee0 283EXPORT_SYMBOL(__iounmap);
a302cb9d 284EXPORT_SYMBOL(__iounmap_at);
5c1f6ee9 285
06743521
AK
286#ifndef __PAGETABLE_PUD_FOLDED
287/* 4 level page table */
288struct page *pgd_page(pgd_t pgd)
289{
290 if (pgd_huge(pgd))
291 return pte_page(pgd_pte(pgd));
292 return virt_to_page(pgd_page_vaddr(pgd));
293}
294#endif
295
296struct page *pud_page(pud_t pud)
297{
298 if (pud_huge(pud))
299 return pte_page(pud_pte(pud));
300 return virt_to_page(pud_page_vaddr(pud));
301}
302
074c2eae
AK
303/*
304 * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
305 * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
306 */
307struct page *pmd_page(pmd_t pmd)
308{
ebd31197 309 if (pmd_trans_huge(pmd) || pmd_huge(pmd) || pmd_devmap(pmd))
e34aa03c 310 return pte_page(pmd_pte(pmd));
074c2eae
AK
311 return virt_to_page(pmd_page_vaddr(pmd));
312}
313
cd65d697
BS
314#ifdef CONFIG_STRICT_KERNEL_RWX
315void mark_rodata_ro(void)
316{
317 if (!mmu_has_feature(MMU_FTR_KERNEL_RO)) {
318 pr_warn("Warning: Unable to mark rodata read only on this CPU.\n");
319 return;
320 }
321
7614ff32
BS
322 if (radix_enabled())
323 radix__mark_rodata_ro();
324 else
cd65d697
BS
325 hash__mark_rodata_ro();
326}
029d9252
ME
327
328void mark_initmem_nx(void)
329{
330 if (radix_enabled())
331 radix__mark_initmem_nx();
332 else
333 hash__mark_initmem_nx();
334}
cd65d697 335#endif