blob: 5a4398a6006bcca5a0bce02e53f06601f399a09e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/arch/x86_64/mm/init.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
7 */
8
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/signal.h>
10#include <linux/sched.h>
11#include <linux/kernel.h>
12#include <linux/errno.h>
13#include <linux/string.h>
14#include <linux/types.h>
15#include <linux/ptrace.h>
16#include <linux/mman.h>
17#include <linux/mm.h>
18#include <linux/swap.h>
19#include <linux/smp.h>
20#include <linux/init.h>
Thomas Gleixner11034d52008-05-12 15:43:36 +020021#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/pagemap.h>
23#include <linux/bootmem.h>
24#include <linux/proc_fs.h>
Andi Kleen59170892005-11-05 17:25:53 +010025#include <linux/pci.h>
Jan Beulich6fb14752007-05-02 19:27:10 +020026#include <linux/pfn.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070027#include <linux/poison.h>
Muli Ben-Yehuda17a941d2006-01-11 22:44:42 +010028#include <linux/dma-mapping.h>
Matt Tolentino44df75e2006-01-17 07:03:41 +010029#include <linux/module.h>
30#include <linux/memory_hotplug.h>
Konrad Rzeszutekae32b122007-05-02 19:27:11 +020031#include <linux/nmi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33#include <asm/processor.h>
Ingo Molnar46eaa672008-10-12 15:06:29 +020034#include <asm/bios_ebda.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/system.h>
36#include <asm/uaccess.h>
37#include <asm/pgtable.h>
38#include <asm/pgalloc.h>
39#include <asm/dma.h>
40#include <asm/fixmap.h>
41#include <asm/e820.h>
42#include <asm/apic.h>
43#include <asm/tlb.h>
44#include <asm/mmu_context.h>
45#include <asm/proto.h>
46#include <asm/smp.h>
Andi Kleen2bc04142005-11-05 17:25:53 +010047#include <asm/sections.h>
Thomas Gleixner718fc132008-01-30 13:30:17 +010048#include <asm/kdebug.h>
Thomas Gleixneraaa64e02008-01-30 13:30:17 +010049#include <asm/numa.h>
Harvey Harrison7bfeab92008-02-12 12:12:01 -080050#include <asm/cacheflush.h>
Pekka Enberg4fcb2082009-03-05 14:55:08 +020051#include <asm/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Andi Kleene18c6872005-11-05 17:25:53 +010053static unsigned long dma_reserve __initdata;
54
Ingo Molnar00d1c5e2008-04-17 17:40:45 +020055static int __init parse_direct_gbpages_off(char *arg)
56{
57 direct_gbpages = 0;
58 return 0;
59}
60early_param("nogbpages", parse_direct_gbpages_off);
61
62static int __init parse_direct_gbpages_on(char *arg)
63{
64 direct_gbpages = 1;
65 return 0;
66}
67early_param("gbpages", parse_direct_gbpages_on);
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069/*
70 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
71 * physical space so we can cache the place of the first one and move
72 * around without checking the pgd every time.
73 */
74
Jeremy Fitzhardingebe43d722008-09-07 15:21:13 -070075pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
Yinghai Lubd220a22008-09-05 00:58:28 -070076EXPORT_SYMBOL_GPL(__supported_pte_mask);
77
Yinghai Lubd220a22008-09-05 00:58:28 -070078int force_personality32;
79
Ingo Molnardeed05b2008-09-05 10:23:26 +020080/*
81 * noexec32=on|off
82 * Control non executable heap for 32bit processes.
83 * To control the stack too use noexec=off
84 *
85 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
86 * off PROT_READ implies PROT_EXEC
87 */
Yinghai Lubd220a22008-09-05 00:58:28 -070088static int __init nonx32_setup(char *str)
89{
90 if (!strcmp(str, "on"))
91 force_personality32 &= ~READ_IMPLIES_EXEC;
92 else if (!strcmp(str, "off"))
93 force_personality32 |= READ_IMPLIES_EXEC;
94 return 1;
95}
96__setup("noexec32=", nonx32_setup);
97
Marcin Slusarz8d6ea9672008-08-15 18:32:24 +020098/*
99 * NOTE: This function is marked __ref because it calls __init function
100 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
101 */
102static __ref void *spp_getpage(void)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100103{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 void *ptr;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 if (after_bootmem)
Vegard Nossum9e730232009-02-22 11:28:25 +0100107 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 else
109 ptr = alloc_bootmem_pages(PAGE_SIZE);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100110
111 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
112 panic("set_pte_phys: cannot allocate page data %s\n",
113 after_bootmem ? "after bootmem" : "");
114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100116 pr_debug("spp_getpage %p\n", ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100118 return ptr;
119}
120
Jeremy Fitzhardingef254f392009-03-03 12:02:57 -0800121static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
Tejun Heo458a3e62009-02-24 11:57:21 +0900122{
123 if (pgd_none(*pgd)) {
124 pud_t *pud = (pud_t *)spp_getpage();
125 pgd_populate(&init_mm, pgd, pud);
126 if (pud != pud_offset(pgd, 0))
127 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
128 pud, pud_offset(pgd, 0));
129 }
130 return pud_offset(pgd, vaddr);
131}
132
Jeremy Fitzhardingef254f392009-03-03 12:02:57 -0800133static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
Tejun Heo458a3e62009-02-24 11:57:21 +0900134{
135 if (pud_none(*pud)) {
136 pmd_t *pmd = (pmd_t *) spp_getpage();
137 pud_populate(&init_mm, pud, pmd);
138 if (pmd != pmd_offset(pud, 0))
139 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
140 pmd, pmd_offset(pud, 0));
141 }
142 return pmd_offset(pud, vaddr);
143}
144
Jeremy Fitzhardingef254f392009-03-03 12:02:57 -0800145static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
Tejun Heo458a3e62009-02-24 11:57:21 +0900146{
147 if (pmd_none(*pmd)) {
148 pte_t *pte = (pte_t *) spp_getpage();
149 pmd_populate_kernel(&init_mm, pmd, pte);
150 if (pte != pte_offset_kernel(pmd, 0))
151 printk(KERN_ERR "PAGETABLE BUG #02!\n");
152 }
153 return pte_offset_kernel(pmd, vaddr);
154}
155
156void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 pud_t *pud;
159 pmd_t *pmd;
Jeremy Fitzhardinged494a962008-06-17 11:41:59 -0700160 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Eduardo Habkost0814e0b2008-06-25 00:19:22 -0400162 pud = pud_page + pud_index(vaddr);
Tejun Heo458a3e62009-02-24 11:57:21 +0900163 pmd = fill_pmd(pud, vaddr);
164 pte = fill_pte(pmd, vaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 set_pte(pte, new_pte);
167
168 /*
169 * It's enough to flush this one mapping.
170 * (PGE mappings get flushed as well)
171 */
172 __flush_tlb_one(vaddr);
173}
174
Tejun Heo458a3e62009-02-24 11:57:21 +0900175void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
Eduardo Habkost0814e0b2008-06-25 00:19:22 -0400176{
177 pgd_t *pgd;
178 pud_t *pud_page;
179
180 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
181
182 pgd = pgd_offset_k(vaddr);
183 if (pgd_none(*pgd)) {
184 printk(KERN_ERR
185 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
186 return;
187 }
188 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
189 set_pte_vaddr_pud(pud_page, vaddr, pteval);
190}
191
Tejun Heo458a3e62009-02-24 11:57:21 +0900192pmd_t * __init populate_extra_pmd(unsigned long vaddr)
Tejun Heo11124412009-02-20 16:29:09 +0900193{
194 pgd_t *pgd;
195 pud_t *pud;
196
197 pgd = pgd_offset_k(vaddr);
Tejun Heo458a3e62009-02-24 11:57:21 +0900198 pud = fill_pud(pgd, vaddr);
199 return fill_pmd(pud, vaddr);
200}
Tejun Heo11124412009-02-20 16:29:09 +0900201
Tejun Heo458a3e62009-02-24 11:57:21 +0900202pte_t * __init populate_extra_pte(unsigned long vaddr)
203{
204 pmd_t *pmd;
205
206 pmd = populate_extra_pmd(vaddr);
207 return fill_pte(pmd, vaddr);
Tejun Heo11124412009-02-20 16:29:09 +0900208}
209
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100210/*
Jack Steiner3a9e1892008-07-01 14:45:32 -0500211 * Create large page table mappings for a range of physical addresses.
212 */
213static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
214 pgprot_t prot)
215{
216 pgd_t *pgd;
217 pud_t *pud;
218 pmd_t *pmd;
219
220 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
221 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
222 pgd = pgd_offset_k((unsigned long)__va(phys));
223 if (pgd_none(*pgd)) {
224 pud = (pud_t *) spp_getpage();
225 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
226 _PAGE_USER));
227 }
228 pud = pud_offset(pgd, (unsigned long)__va(phys));
229 if (pud_none(*pud)) {
230 pmd = (pmd_t *) spp_getpage();
231 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
232 _PAGE_USER));
233 }
234 pmd = pmd_offset(pud, phys);
235 BUG_ON(!pmd_none(*pmd));
236 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
237 }
238}
239
240void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
241{
242 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
243}
244
245void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
246{
247 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
248}
249
250/*
Ingo Molnar88f3aec2008-02-21 11:04:11 +0100251 * The head.S code sets up the kernel high mapping:
252 *
253 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100254 *
255 * phys_addr holds the negative offset to the kernel, which is added
256 * to the compile time generated pmds. This results in invalid pmds up
257 * to the point where we hit the physaddr 0 mapping.
258 *
259 * We limit the mappings to the region from _text to _end. _end is
260 * rounded up to the 2MB boundary. This catches the invalid pmds as
261 * well, as they are located before _text:
262 */
263void __init cleanup_highmap(void)
264{
265 unsigned long vaddr = __START_KERNEL_map;
Joerg Roedeld86bb0d2008-07-25 16:48:57 +0200266 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100267 pmd_t *pmd = level2_kernel_pgt;
268 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
269
270 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
Hugh Dickins2884f112008-05-28 19:36:07 +0100271 if (pmd_none(*pmd))
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100272 continue;
273 if (vaddr < (unsigned long) _text || vaddr > end)
274 set_pmd(pmd, __pmd(0));
275 }
276}
277
Jan Beulich9482ac62008-08-21 14:28:42 +0100278static __ref void *alloc_low_page(unsigned long *phys)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100279{
Pekka Enberg298af9d2009-03-05 14:55:06 +0200280 unsigned long pfn = e820_table_end++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 void *adr;
282
Matt Tolentino44df75e2006-01-17 07:03:41 +0100283 if (after_bootmem) {
Vegard Nossum9e730232009-02-22 11:28:25 +0100284 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100285 *phys = __pa(adr);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100286
Matt Tolentino44df75e2006-01-17 07:03:41 +0100287 return adr;
288 }
289
Pekka Enberg298af9d2009-03-05 14:55:06 +0200290 if (pfn >= e820_table_top)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100291 panic("alloc_low_page: ran out of memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Jeremy Fitzhardinge14941772008-09-07 15:21:15 -0700293 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200294 memset(adr, 0, PAGE_SIZE);
295 *phys = pfn * PAGE_SIZE;
296 return adr;
297}
298
Jan Beulich9482ac62008-08-21 14:28:42 +0100299static __ref void unmap_low_page(void *adr)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100300{
Matt Tolentino44df75e2006-01-17 07:03:41 +0100301 if (after_bootmem)
302 return;
303
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200304 early_iounmap(adr, PAGE_SIZE);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100305}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700307static unsigned long __meminit
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700308phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
309 pgprot_t prot)
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400310{
311 unsigned pages = 0;
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700312 unsigned long last_map_addr = end;
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400313 int i;
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700314
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400315 pte_t *pte = pte_page + pte_index(addr);
316
317 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
318
319 if (addr >= end) {
320 if (!after_bootmem) {
321 for(; i < PTRS_PER_PTE; i++, pte++)
322 set_pte(pte, __pte(0));
323 }
324 break;
325 }
326
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700327 /*
328 * We will re-use the existing mapping.
329 * Xen for example has some special requirements, like mapping
330 * pagetable pages as RO. So assume someone who pre-setup
331 * these mappings are more intelligent.
332 */
Yinghai Lu3afa3942008-10-25 22:58:21 -0700333 if (pte_val(*pte)) {
334 pages++;
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400335 continue;
Yinghai Lu3afa3942008-10-25 22:58:21 -0700336 }
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400337
338 if (0)
339 printk(" pte=%p addr=%lx pte=%016lx\n",
340 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400341 pages++;
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700342 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400343 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400344 }
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700345
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400346 update_page_count(PG_LEVEL_4K, pages);
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700347
348 return last_map_addr;
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400349}
350
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700351static unsigned long __meminit
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700352phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
353 pgprot_t prot)
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400354{
355 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
356
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700357 return phys_pte_init(pte, address, end, prot);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400358}
359
Andi Kleencc615032008-03-12 03:53:28 +0100360static unsigned long __meminit
Yinghai Lub50efd22008-07-08 01:41:05 -0700361phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700362 unsigned long page_size_mask, pgprot_t prot)
Matt Tolentino44df75e2006-01-17 07:03:41 +0100363{
Andi Kleence0c0e52008-05-02 11:46:49 +0200364 unsigned long pages = 0;
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700365 unsigned long last_map_addr = end;
Andi Kleence0c0e52008-05-02 11:46:49 +0200366
Keith Mannthey6ad91652006-09-26 10:52:36 +0200367 int i = pmd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Keith Mannthey6ad91652006-09-26 10:52:36 +0200369 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400370 unsigned long pte_phys;
Keith Mannthey6ad91652006-09-26 10:52:36 +0200371 pmd_t *pmd = pmd_page + pmd_index(address);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400372 pte_t *pte;
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700373 pgprot_t new_prot = prot;
Matt Tolentino44df75e2006-01-17 07:03:41 +0100374
Jan Beulich5f51e132006-06-26 13:59:02 +0200375 if (address >= end) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100376 if (!after_bootmem) {
Jan Beulich5f51e132006-06-26 13:59:02 +0200377 for (; i < PTRS_PER_PMD; i++, pmd++)
378 set_pmd(pmd, __pmd(0));
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100379 }
Matt Tolentino44df75e2006-01-17 07:03:41 +0100380 break;
381 }
Keith Mannthey6ad91652006-09-26 10:52:36 +0200382
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400383 if (pmd_val(*pmd)) {
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100384 if (!pmd_large(*pmd)) {
385 spin_lock(&init_mm.page_table_lock);
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700386 last_map_addr = phys_pte_update(pmd, address,
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700387 end, prot);
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100388 spin_unlock(&init_mm.page_table_lock);
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700389 continue;
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100390 }
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700391 /*
392 * If we are ok with PG_LEVEL_2M mapping, then we will
393 * use the existing mapping,
394 *
395 * Otherwise, we will split the large page mapping but
396 * use the same existing protection bits except for
397 * large page, so that we don't violate Intel's TLB
398 * Application note (317080) which says, while changing
399 * the page sizes, new and old translations should
400 * not differ with respect to page frame and
401 * attributes.
402 */
Yinghai Lu3afa3942008-10-25 22:58:21 -0700403 if (page_size_mask & (1 << PG_LEVEL_2M)) {
404 pages++;
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700405 continue;
Yinghai Lu3afa3942008-10-25 22:58:21 -0700406 }
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700407 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400408 }
Keith Mannthey6ad91652006-09-26 10:52:36 +0200409
Yinghai Lub50efd22008-07-08 01:41:05 -0700410 if (page_size_mask & (1<<PG_LEVEL_2M)) {
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400411 pages++;
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100412 spin_lock(&init_mm.page_table_lock);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400413 set_pte((pte_t *)pmd,
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700414 pfn_pte(address >> PAGE_SHIFT,
415 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100416 spin_unlock(&init_mm.page_table_lock);
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700417 last_map_addr = (address & PMD_MASK) + PMD_SIZE;
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400418 continue;
419 }
420
421 pte = alloc_low_page(&pte_phys);
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700422 last_map_addr = phys_pte_init(pte, address, end, new_prot);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400423 unmap_low_page(pte);
424
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100425 spin_lock(&init_mm.page_table_lock);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400426 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100427 spin_unlock(&init_mm.page_table_lock);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100428 }
Andi Kleence0c0e52008-05-02 11:46:49 +0200429 update_page_count(PG_LEVEL_2M, pages);
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700430 return last_map_addr;
Matt Tolentino44df75e2006-01-17 07:03:41 +0100431}
432
Andi Kleencc615032008-03-12 03:53:28 +0100433static unsigned long __meminit
Yinghai Lub50efd22008-07-08 01:41:05 -0700434phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700435 unsigned long page_size_mask, pgprot_t prot)
Matt Tolentino44df75e2006-01-17 07:03:41 +0100436{
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100437 pmd_t *pmd = pmd_offset(pud, 0);
Andi Kleencc615032008-03-12 03:53:28 +0100438 unsigned long last_map_addr;
439
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700440 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
Keith Mannthey6ad91652006-09-26 10:52:36 +0200441 __flush_tlb_all();
Andi Kleencc615032008-03-12 03:53:28 +0100442 return last_map_addr;
Matt Tolentino44df75e2006-01-17 07:03:41 +0100443}
444
Andi Kleencc615032008-03-12 03:53:28 +0100445static unsigned long __meminit
Yinghai Lub50efd22008-07-08 01:41:05 -0700446phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
447 unsigned long page_size_mask)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100448{
Andi Kleence0c0e52008-05-02 11:46:49 +0200449 unsigned long pages = 0;
Andi Kleencc615032008-03-12 03:53:28 +0100450 unsigned long last_map_addr = end;
Keith Mannthey6ad91652006-09-26 10:52:36 +0200451 int i = pud_index(addr);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100452
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100453 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
Keith Mannthey6ad91652006-09-26 10:52:36 +0200454 unsigned long pmd_phys;
455 pud_t *pud = pud_page + pud_index(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 pmd_t *pmd;
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700457 pgprot_t prot = PAGE_KERNEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Keith Mannthey6ad91652006-09-26 10:52:36 +0200459 if (addr >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100462 if (!after_bootmem &&
463 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
464 set_pud(pud, __pud(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 continue;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Keith Mannthey6ad91652006-09-26 10:52:36 +0200468 if (pud_val(*pud)) {
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700469 if (!pud_large(*pud)) {
Yinghai Lub50efd22008-07-08 01:41:05 -0700470 last_map_addr = phys_pmd_update(pud, addr, end,
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700471 page_size_mask, prot);
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700472 continue;
473 }
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700474 /*
475 * If we are ok with PG_LEVEL_1G mapping, then we will
476 * use the existing mapping.
477 *
478 * Otherwise, we will split the gbpage mapping but use
479 * the same existing protection bits except for large
480 * page, so that we don't violate Intel's TLB
481 * Application note (317080) which says, while changing
482 * the page sizes, new and old translations should
483 * not differ with respect to page frame and
484 * attributes.
485 */
Yinghai Lu3afa3942008-10-25 22:58:21 -0700486 if (page_size_mask & (1 << PG_LEVEL_1G)) {
487 pages++;
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700488 continue;
Yinghai Lu3afa3942008-10-25 22:58:21 -0700489 }
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700490 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
Andi Kleenef925762008-04-17 17:40:45 +0200491 }
492
Yinghai Lub50efd22008-07-08 01:41:05 -0700493 if (page_size_mask & (1<<PG_LEVEL_1G)) {
Andi Kleence0c0e52008-05-02 11:46:49 +0200494 pages++;
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100495 spin_lock(&init_mm.page_table_lock);
Andi Kleenef925762008-04-17 17:40:45 +0200496 set_pte((pte_t *)pud,
497 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100498 spin_unlock(&init_mm.page_table_lock);
Andi Kleencc615032008-03-12 03:53:28 +0100499 last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
Keith Mannthey6ad91652006-09-26 10:52:36 +0200500 continue;
501 }
502
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200503 pmd = alloc_low_page(&pmd_phys);
Suresh Siddhab27a43c2008-10-07 13:58:46 -0700504 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
505 prot);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400506 unmap_low_page(pmd);
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100507
508 spin_lock(&init_mm.page_table_lock);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400509 pud_populate(&init_mm, pud, __va(pmd_phys));
Matt Tolentino44df75e2006-01-17 07:03:41 +0100510 spin_unlock(&init_mm.page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
Andi Kleen1a2b4412008-01-30 13:33:54 +0100512 __flush_tlb_all();
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700513
Andi Kleence0c0e52008-05-02 11:46:49 +0200514 update_page_count(PG_LEVEL_1G, pages);
Andi Kleencc615032008-03-12 03:53:28 +0100515
Yinghai Lu1a0db382008-06-24 14:56:20 -0700516 return last_map_addr;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100517}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400519static unsigned long __meminit
Yinghai Lub50efd22008-07-08 01:41:05 -0700520phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
521 unsigned long page_size_mask)
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400522{
523 pud_t *pud;
524
525 pud = (pud_t *)pgd_page_vaddr(*pgd);
526
Yinghai Lub50efd22008-07-08 01:41:05 -0700527 return phys_pud_init(pud, addr, end, page_size_mask);
Jeremy Fitzhardinge4f9c11d2008-06-25 00:19:19 -0400528}
529
Shaohua Li41d840e2009-06-12 12:57:52 +0800530unsigned long __meminit
Pekka Enbergf7650902009-03-05 14:55:05 +0200531kernel_physical_mapping_init(unsigned long start,
532 unsigned long end,
533 unsigned long page_size_mask)
Yinghai Lub50efd22008-07-08 01:41:05 -0700534{
535
536 unsigned long next, last_map_addr = end;
537
538 start = (unsigned long)__va(start);
539 end = (unsigned long)__va(end);
540
541 for (; start < end; start = next) {
542 pgd_t *pgd = pgd_offset_k(start);
543 unsigned long pud_phys;
544 pud_t *pud;
545
Jack Steinere22146e2008-07-16 11:11:59 -0500546 next = (start + PGDIR_SIZE) & PGDIR_MASK;
Yinghai Lub50efd22008-07-08 01:41:05 -0700547 if (next > end)
548 next = end;
549
550 if (pgd_val(*pgd)) {
551 last_map_addr = phys_pud_update(pgd, __pa(start),
552 __pa(end), page_size_mask);
553 continue;
554 }
555
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100556 pud = alloc_low_page(&pud_phys);
Yinghai Lub50efd22008-07-08 01:41:05 -0700557 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
558 page_size_mask);
559 unmap_low_page(pud);
Jan Beulich8ae3a5a2008-08-21 14:27:22 +0100560
561 spin_lock(&init_mm.page_table_lock);
562 pgd_populate(&init_mm, pgd, __va(pud_phys));
563 spin_unlock(&init_mm.page_table_lock);
Yinghai Lub50efd22008-07-08 01:41:05 -0700564 }
Suresh Siddhaa2699e42008-09-23 14:00:38 -0700565 __flush_tlb_all();
Yinghai Lub50efd22008-07-08 01:41:05 -0700566
567 return last_map_addr;
568}
Yinghai Lu7b16eb82008-07-09 20:15:02 -0700569
Matt Tolentino2b976902005-06-23 00:08:06 -0700570#ifndef CONFIG_NUMA
Yinghai Lu1f75d7e2008-06-22 02:44:49 -0700571void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
572{
573 unsigned long bootmap_size, bootmap;
574
575 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
576 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
577 PAGE_SIZE);
578 if (bootmap == -1L)
579 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
Yinghai Lu346cafe2008-06-23 03:06:14 -0700580 /* don't touch min_low_pfn */
581 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
582 0, end_pfn);
Yinghai Lu1f75d7e2008-06-22 02:44:49 -0700583 e820_register_active_regions(0, start_pfn, end_pfn);
584 free_bootmem_with_active_regions(0, end_pfn);
585 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
586 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
587}
Pekka Enberg3551f882009-05-07 15:35:41 +0300588#endif
Yinghai Lu1f75d7e2008-06-22 02:44:49 -0700589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590void __init paging_init(void)
591{
Mel Gorman6391af12006-10-11 01:20:39 -0700592 unsigned long max_zone_pfns[MAX_NR_ZONES];
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100593
Mel Gorman6391af12006-10-11 01:20:39 -0700594 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
595 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
596 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
Yinghai Luc987d122008-06-24 22:14:09 -0700597 max_zone_pfns[ZONE_NORMAL] = max_pfn;
Mel Gorman6391af12006-10-11 01:20:39 -0700598
Pekka Enberg3551f882009-05-07 15:35:41 +0300599 sparse_memory_present_with_active_regions(MAX_NUMNODES);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100600 sparse_init();
Yinghai Lu44b57282009-07-08 09:50:19 -0700601
602 /*
603 * clear the default setting with node 0
604 * note: don't use nodes_clear here, that is really clearing when
605 * numa support is not compiled in, and later node_set_state
606 * will not set it back.
607 */
608 node_clear_state(0, N_NORMAL_MEMORY);
609
Mel Gorman5cb248a2006-09-27 01:49:52 -0700610 free_area_init_nodes(max_zone_pfns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100613/*
Matt Tolentino44df75e2006-01-17 07:03:41 +0100614 * Memory hotplug specific functions
Matt Tolentino44df75e2006-01-17 07:03:41 +0100615 */
Yasunori Gotobc02af92006-06-27 02:53:30 -0700616#ifdef CONFIG_MEMORY_HOTPLUG
617/*
Yasunori Gotobc02af92006-06-27 02:53:30 -0700618 * Memory is added always to NORMAL zone. This means you will never get
619 * additional DMA/DMA32 memory.
620 */
621int arch_add_memory(int nid, u64 start, u64 size)
622{
623 struct pglist_data *pgdat = NODE_DATA(nid);
Christoph Lameter776ed982006-09-25 23:31:09 -0700624 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
Andi Kleencc615032008-03-12 03:53:28 +0100625 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700626 unsigned long nr_pages = size >> PAGE_SHIFT;
627 int ret;
628
Shaohua Li60817c92008-10-27 13:03:18 -0700629 last_mapped_pfn = init_memory_mapping(start, start + size);
Andi Kleencc615032008-03-12 03:53:28 +0100630 if (last_mapped_pfn > max_pfn_mapped)
631 max_pfn_mapped = last_mapped_pfn;
Keith Mannthey45e0b782006-09-30 23:27:09 -0700632
Gary Hadec04fc582009-01-06 14:39:14 -0800633 ret = __add_pages(nid, zone, start_pfn, nr_pages);
Gary Hadefe8b8682008-10-28 16:43:14 -0700634 WARN_ON_ONCE(ret);
Yasunori Gotobc02af92006-06-27 02:53:30 -0700635
Yasunori Gotobc02af92006-06-27 02:53:30 -0700636 return ret;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700637}
638EXPORT_SYMBOL_GPL(arch_add_memory);
639
Yasunori Goto82432292006-11-18 22:19:40 -0800640#if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
Keith Mannthey4942e992006-09-30 23:27:06 -0700641int memory_add_physaddr_to_nid(u64 start)
642{
643 return 0;
644}
Keith Mannthey8c2676a2006-09-30 23:27:07 -0700645EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
Keith Mannthey4942e992006-09-30 23:27:06 -0700646#endif
647
Keith Mannthey45e0b782006-09-30 23:27:09 -0700648#endif /* CONFIG_MEMORY_HOTPLUG */
649
KAMEZAWA Hiroyuki81ac3ad2009-09-22 16:45:49 -0700650static struct kcore_list kcore_vsyscall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
652void __init mem_init(void)
653{
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200654 long codesize, reservedpages, datasize, initsize;
Yinghai Lu11a6b0c2008-10-14 18:59:18 -0700655 unsigned long absent_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Jon Mason0dc243a2006-06-26 13:58:11 +0200657 pci_iommu_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Yinghai Lu48ddb152008-01-30 13:32:36 +0100659 /* clear_bss() already clear the empty_zero_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 reservedpages = 0;
662
663 /* this will put all low memory onto the freelists */
Matt Tolentino2b976902005-06-23 00:08:06 -0700664#ifdef CONFIG_NUMA
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200665 totalram_pages = numa_free_all_bootmem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666#else
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200667 totalram_pages = free_all_bootmem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668#endif
Yinghai Lu11a6b0c2008-10-14 18:59:18 -0700669
670 absent_pages = absent_pages_in_range(0, max_pfn);
671 reservedpages = max_pfn - totalram_pages - absent_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 after_bootmem = 1;
673
674 codesize = (unsigned long) &_etext - (unsigned long) &_text;
675 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
676 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
677
678 /* Register memory areas for /proc/kcore */
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100679 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
KAMEZAWA Hiroyukic30bb2a2009-09-22 16:45:43 -0700680 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100682 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
Yinghai Lu11a6b0c2008-10-14 18:59:18 -0700683 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
Geert Uytterhoevencc013a82009-09-21 17:02:36 -0700684 nr_free_pages() << (PAGE_SHIFT-10),
Yinghai Luc987d122008-06-24 22:14:09 -0700685 max_pfn << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 codesize >> 10,
Yinghai Lu11a6b0c2008-10-14 18:59:18 -0700687 absent_pages << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 reservedpages << (PAGE_SHIFT-10),
689 datasize >> 10,
690 initsize >> 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
Arjan van de Ven67df1972006-01-06 00:12:04 -0800693#ifdef CONFIG_DEBUG_RODATA
Arjan van de Venedeed302008-01-30 13:34:08 +0100694const int rodata_test_data = 0xC3;
695EXPORT_SYMBOL_GPL(rodata_test_data);
Arjan van de Ven67df1972006-01-06 00:12:04 -0800696
Steven Rostedt16239632009-02-17 17:57:30 -0500697static int kernel_set_to_readonly;
698
699void set_kernel_text_rw(void)
700{
701 unsigned long start = PFN_ALIGN(_stext);
702 unsigned long end = PFN_ALIGN(__start_rodata);
703
704 if (!kernel_set_to_readonly)
705 return;
706
707 pr_debug("Set kernel text: %lx - %lx for read write\n",
708 start, end);
709
710 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
711}
712
713void set_kernel_text_ro(void)
714{
715 unsigned long start = PFN_ALIGN(_stext);
716 unsigned long end = PFN_ALIGN(__start_rodata);
717
718 if (!kernel_set_to_readonly)
719 return;
720
721 pr_debug("Set kernel text: %lx - %lx for read only\n",
722 start, end);
723
724 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
725}
726
Arjan van de Ven67df1972006-01-06 00:12:04 -0800727void mark_rodata_ro(void)
728{
Mathieu Desnoyers4e4eee02008-02-02 15:42:20 -0500729 unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
Steven Rostedt8f0f9962008-05-12 21:20:56 +0200730 unsigned long rodata_start =
731 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
732
Jan Beulich6fb14752007-05-02 19:27:10 +0200733 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700734 (end - start) >> 10);
Arjan van de Ven984bb802008-02-06 22:39:45 +0100735 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
736
Steven Rostedt16239632009-02-17 17:57:30 -0500737 kernel_set_to_readonly = 1;
738
Arjan van de Ven984bb802008-02-06 22:39:45 +0100739 /*
740 * The rodata section (but not the kernel text!) should also be
741 * not-executable.
742 */
Pekka Paalanen72b59d62008-05-12 21:21:01 +0200743 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
Arjan van de Ven67df1972006-01-06 00:12:04 -0800744
Arjan van de Ven1a487252008-01-30 13:34:09 +0100745 rodata_test();
746
Andi Kleen0c42f392008-01-30 13:33:42 +0100747#ifdef CONFIG_CPA_DEBUG
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100748 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
Arjan van de Ven6d238cc2008-01-30 13:34:06 +0100749 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
Andi Kleen0c42f392008-01-30 13:33:42 +0100750
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100751 printk(KERN_INFO "Testing CPA: again\n");
Arjan van de Ven6d238cc2008-01-30 13:34:06 +0100752 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
Andi Kleen0c42f392008-01-30 13:33:42 +0100753#endif
Arjan van de Ven67df1972006-01-06 00:12:04 -0800754}
Mathieu Desnoyers4e4eee02008-02-02 15:42:20 -0500755
Arjan van de Ven67df1972006-01-06 00:12:04 -0800756#endif
757
Yinghai Lud2dbf342008-06-13 02:00:56 -0700758int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
759 int flags)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100760{
Matt Tolentino2b976902005-06-23 00:08:06 -0700761#ifdef CONFIG_NUMA
Yinghai Lu8b3cd092008-03-18 12:50:21 -0700762 int nid, next_nid;
Yinghai Lu6a07a0e2008-06-23 14:02:36 -0700763 int ret;
Andi Kleen5e58a022006-11-14 16:57:46 +0100764#endif
765 unsigned long pfn = phys >> PAGE_SHIFT;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100766
Yinghai Luc987d122008-06-24 22:14:09 -0700767 if (pfn >= max_pfn) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100768 /*
769 * This can happen with kdump kernels when accessing
770 * firmware tables:
771 */
Thomas Gleixner67794292008-03-21 21:27:10 +0100772 if (pfn < max_pfn_mapped)
Bernhard Walle8b2ef1d2008-06-08 15:46:30 +0200773 return -EFAULT;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100774
Yinghai Lu6a07a0e2008-06-23 14:02:36 -0700775 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
Andi Kleen5e58a022006-11-14 16:57:46 +0100776 phys, len);
Bernhard Walle8b2ef1d2008-06-08 15:46:30 +0200777 return -EFAULT;
Andi Kleen5e58a022006-11-14 16:57:46 +0100778 }
779
780 /* Should check here against the e820 map to avoid double free */
781#ifdef CONFIG_NUMA
Yinghai Lu8b3cd092008-03-18 12:50:21 -0700782 nid = phys_to_nid(phys);
783 next_nid = phys_to_nid(phys + len - 1);
784 if (nid == next_nid)
Bernhard Walle8b2ef1d2008-06-08 15:46:30 +0200785 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
Yinghai Lu8b3cd092008-03-18 12:50:21 -0700786 else
Bernhard Walle8b2ef1d2008-06-08 15:46:30 +0200787 ret = reserve_bootmem(phys, len, flags);
788
789 if (ret != 0)
790 return ret;
791
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100792#else
Amerigo Wanga6a06f72009-08-21 04:34:45 -0400793 reserve_bootmem(phys, len, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794#endif
Yinghai Lu8b3cd092008-03-18 12:50:21 -0700795
Mel Gorman0e0b8642006-09-27 01:49:56 -0700796 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
Andi Kleene18c6872005-11-05 17:25:53 +0100797 dma_reserve += len / PAGE_SIZE;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700798 set_dma_reserve(dma_reserve);
799 }
Bernhard Walle8b2ef1d2008-06-08 15:46:30 +0200800
801 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100804int kern_addr_valid(unsigned long addr)
805{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100807 pgd_t *pgd;
808 pud_t *pud;
809 pmd_t *pmd;
810 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812 if (above != 0 && above != -1UL)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100813 return 0;
814
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 pgd = pgd_offset_k(addr);
816 if (pgd_none(*pgd))
817 return 0;
818
819 pud = pud_offset(pgd, addr);
820 if (pud_none(*pud))
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100821 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 pmd = pmd_offset(pud, addr);
824 if (pmd_none(*pmd))
825 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100826
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 if (pmd_large(*pmd))
828 return pfn_valid(pmd_pfn(*pmd));
829
830 pte = pte_offset_kernel(pmd, addr);
831 if (pte_none(*pte))
832 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 return pfn_valid(pte_pfn(*pte));
835}
836
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100837/*
838 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
839 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
840 * not need special handling anymore:
841 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842static struct vm_area_struct gate_vma = {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100843 .vm_start = VSYSCALL_START,
844 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
845 .vm_page_prot = PAGE_READONLY_EXEC,
846 .vm_flags = VM_READ | VM_EXEC
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847};
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
850{
851#ifdef CONFIG_IA32_EMULATION
Andi Kleen1e014412005-04-16 15:24:55 -0700852 if (test_tsk_thread_flag(tsk, TIF_IA32))
853 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854#endif
855 return &gate_vma;
856}
857
858int in_gate_area(struct task_struct *task, unsigned long addr)
859{
860 struct vm_area_struct *vma = get_gate_vma(task);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100861
Andi Kleen1e014412005-04-16 15:24:55 -0700862 if (!vma)
863 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 return (addr >= vma->vm_start) && (addr < vma->vm_end);
866}
867
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100868/*
869 * Use this when you have no reliable task/vma, typically from interrupt
870 * context. It is less reliable than using the task's vma and may give
871 * false positives:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
873int in_gate_area_no_task(unsigned long addr)
874{
Andi Kleen1e014412005-04-16 15:24:55 -0700875 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
Zou Nan hai2e1c49d2007-06-01 00:46:28 -0700877
Andi Kleen2aae9502007-07-21 17:10:01 +0200878const char *arch_vma_name(struct vm_area_struct *vma)
879{
880 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
881 return "[vdso]";
882 if (vma == &gate_vma)
883 return "[vsyscall]";
884 return NULL;
885}
Christoph Lameter0889eba2007-10-16 01:24:15 -0700886
887#ifdef CONFIG_SPARSEMEM_VMEMMAP
888/*
889 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
890 */
Yinghai Luc2b91e22008-04-12 01:19:24 -0700891static long __meminitdata addr_start, addr_end;
892static void __meminitdata *p_start, *p_end;
893static int __meminitdata node_start;
894
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100895int __meminit
896vmemmap_populate(struct page *start_page, unsigned long size, int node)
Christoph Lameter0889eba2007-10-16 01:24:15 -0700897{
898 unsigned long addr = (unsigned long)start_page;
899 unsigned long end = (unsigned long)(start_page + size);
900 unsigned long next;
901 pgd_t *pgd;
902 pud_t *pud;
903 pmd_t *pmd;
904
905 for (; addr < end; addr = next) {
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400906 void *p = NULL;
Christoph Lameter0889eba2007-10-16 01:24:15 -0700907
908 pgd = vmemmap_pgd_populate(addr, node);
909 if (!pgd)
910 return -ENOMEM;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100911
Christoph Lameter0889eba2007-10-16 01:24:15 -0700912 pud = vmemmap_pud_populate(pgd, addr, node);
913 if (!pud)
914 return -ENOMEM;
915
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400916 if (!cpu_has_pse) {
917 next = (addr + PAGE_SIZE) & PAGE_MASK;
918 pmd = vmemmap_pmd_populate(pud, addr, node);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100919
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400920 if (!pmd)
921 return -ENOMEM;
922
923 p = vmemmap_pte_populate(pmd, addr, node);
924
Christoph Lameter0889eba2007-10-16 01:24:15 -0700925 if (!p)
926 return -ENOMEM;
927
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400928 addr_end = addr + PAGE_SIZE;
929 p_end = p + PAGE_SIZE;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100930 } else {
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400931 next = pmd_addr_end(addr, end);
932
933 pmd = pmd_offset(pud, addr);
934 if (pmd_none(*pmd)) {
935 pte_t entry;
936
937 p = vmemmap_alloc_block(PMD_SIZE, node);
938 if (!p)
939 return -ENOMEM;
940
941 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
942 PAGE_KERNEL_LARGE);
943 set_pmd(pmd, __pmd(pte_val(entry)));
944
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400945 /* check to see if we have contiguous blocks */
946 if (p_end != p || node_start != node) {
947 if (p_start)
948 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
949 addr_start, addr_end-1, p_start, p_end-1, node_start);
950 addr_start = addr;
951 node_start = node;
952 p_start = p;
953 }
Yinghai Lu49c980d2008-07-03 12:29:34 -0700954
955 addr_end = addr + PMD_SIZE;
956 p_end = p + PMD_SIZE;
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400957 } else
958 vmemmap_verify((pte_t *)pmd, node, addr, next);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100959 }
Jeremy Fitzhardinge7c934d32008-06-25 00:19:20 -0400960
Christoph Lameter0889eba2007-10-16 01:24:15 -0700961 }
Christoph Lameter0889eba2007-10-16 01:24:15 -0700962 return 0;
963}
Yinghai Luc2b91e22008-04-12 01:19:24 -0700964
965void __meminit vmemmap_populate_print_last(void)
966{
967 if (p_start) {
968 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
969 addr_start, addr_end-1, p_start, p_end-1, node_start);
970 p_start = NULL;
971 p_end = NULL;
972 node_start = 0;
973 }
974}
Christoph Lameter0889eba2007-10-16 01:24:15 -0700975#endif