blob: ae66d8d4c58d5a2df98feedd3a1e18853bc91389 [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>
21#include <linux/pagemap.h>
22#include <linux/bootmem.h>
23#include <linux/proc_fs.h>
Andi Kleen59170892005-11-05 17:25:53 +010024#include <linux/pci.h>
Jan Beulich6fb14752007-05-02 19:27:10 +020025#include <linux/pfn.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070026#include <linux/poison.h>
Muli Ben-Yehuda17a941d2006-01-11 22:44:42 +010027#include <linux/dma-mapping.h>
Matt Tolentino44df75e2006-01-17 07:03:41 +010028#include <linux/module.h>
29#include <linux/memory_hotplug.h>
Konrad Rzeszutekae32b122007-05-02 19:27:11 +020030#include <linux/nmi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/processor.h>
33#include <asm/system.h>
34#include <asm/uaccess.h>
35#include <asm/pgtable.h>
36#include <asm/pgalloc.h>
37#include <asm/dma.h>
38#include <asm/fixmap.h>
39#include <asm/e820.h>
40#include <asm/apic.h>
41#include <asm/tlb.h>
42#include <asm/mmu_context.h>
43#include <asm/proto.h>
44#include <asm/smp.h>
Andi Kleen2bc04142005-11-05 17:25:53 +010045#include <asm/sections.h>
Thomas Gleixner718fc132008-01-30 13:30:17 +010046#include <asm/kdebug.h>
Thomas Gleixneraaa64e02008-01-30 13:30:17 +010047#include <asm/numa.h>
Harvey Harrison7bfeab92008-02-12 12:12:01 -080048#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Thomas Gleixner14a62c32008-01-30 13:34:10 +010050const struct dma_mapping_ops *dma_ops;
Muli Ben-Yehuda17a941d2006-01-11 22:44:42 +010051EXPORT_SYMBOL(dma_ops);
52
Andi Kleene18c6872005-11-05 17:25:53 +010053static unsigned long dma_reserve __initdata;
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
56
Ingo Molnar00d1c5e2008-04-17 17:40:45 +020057int direct_gbpages __meminitdata
58#ifdef CONFIG_DIRECT_GBPAGES
59 = 1
60#endif
61;
62
63static int __init parse_direct_gbpages_off(char *arg)
64{
65 direct_gbpages = 0;
66 return 0;
67}
68early_param("nogbpages", parse_direct_gbpages_off);
69
70static int __init parse_direct_gbpages_on(char *arg)
71{
72 direct_gbpages = 1;
73 return 0;
74}
75early_param("gbpages", parse_direct_gbpages_on);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
78 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
79 * physical space so we can cache the place of the first one and move
80 * around without checking the pgd every time.
81 */
82
83void show_mem(void)
84{
Andi Kleene92343c2005-09-12 18:49:24 +020085 long i, total = 0, reserved = 0;
86 long shared = 0, cached = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 struct page *page;
Thomas Gleixner14a62c32008-01-30 13:34:10 +010088 pg_data_t *pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Andi Kleene92343c2005-09-12 18:49:24 +020090 printk(KERN_INFO "Mem-info:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 show_free_areas();
Thomas Gleixner14a62c32008-01-30 13:34:10 +010092 printk(KERN_INFO "Free swap: %6ldkB\n",
93 nr_swap_pages << (PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -080095 for_each_online_pgdat(pgdat) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +010096 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
97 /*
98 * This loop can take a while with 256 GB and
99 * 4k pages so defer the NMI watchdog:
100 */
101 if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
Konrad Rzeszutekae32b122007-05-02 19:27:11 +0200102 touch_nmi_watchdog();
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100103
Bob Picco12710a52007-06-08 13:47:00 -0700104 if (!pfn_valid(pgdat->node_start_pfn + i))
105 continue;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 page = pfn_to_page(pgdat->node_start_pfn + i);
108 total++;
Andi Kleene92343c2005-09-12 18:49:24 +0200109 if (PageReserved(page))
110 reserved++;
111 else if (PageSwapCache(page))
112 cached++;
113 else if (page_count(page))
114 shared += page_count(page) - 1;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 }
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100117 printk(KERN_INFO "%lu pages of RAM\n", total);
118 printk(KERN_INFO "%lu reserved pages\n", reserved);
119 printk(KERN_INFO "%lu pages shared\n", shared);
120 printk(KERN_INFO "%lu pages swap cached\n", cached);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123int after_bootmem;
124
Andi Kleen5f44a662006-03-25 16:30:25 +0100125static __init void *spp_getpage(void)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100126{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 void *ptr;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 if (after_bootmem)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100130 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 else
132 ptr = alloc_bootmem_pages(PAGE_SIZE);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100133
134 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
135 panic("set_pte_phys: cannot allocate page data %s\n",
136 after_bootmem ? "after bootmem" : "");
137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100139 pr_debug("spp_getpage %p\n", ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100141 return ptr;
142}
143
144static __init void
145set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 pgd_t *pgd;
148 pud_t *pud;
149 pmd_t *pmd;
150 pte_t *pte, new_pte;
151
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100152 pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 pgd = pgd_offset_k(vaddr);
155 if (pgd_none(*pgd)) {
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100156 printk(KERN_ERR
157 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 return;
159 }
160 pud = pud_offset(pgd, vaddr);
161 if (pud_none(*pud)) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100162 pmd = (pmd_t *) spp_getpage();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
164 if (pmd != pmd_offset(pud, 0)) {
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100165 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100166 pmd, pmd_offset(pud, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 return;
168 }
169 }
170 pmd = pmd_offset(pud, vaddr);
171 if (pmd_none(*pmd)) {
172 pte = (pte_t *) spp_getpage();
173 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
174 if (pte != pte_offset_kernel(pmd, 0)) {
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100175 printk(KERN_ERR "PAGETABLE BUG #02!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return;
177 }
178 }
179 new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
180
181 pte = pte_offset_kernel(pmd, vaddr);
182 if (!pte_none(*pte) &&
183 pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
184 pte_ERROR(*pte);
185 set_pte(pte, new_pte);
186
187 /*
188 * It's enough to flush this one mapping.
189 * (PGE mappings get flushed as well)
190 */
191 __flush_tlb_one(vaddr);
192}
193
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100194/*
Ingo Molnar88f3aec2008-02-21 11:04:11 +0100195 * The head.S code sets up the kernel high mapping:
196 *
197 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
Thomas Gleixner31eedd82008-02-15 17:29:12 +0100198 *
199 * phys_addr holds the negative offset to the kernel, which is added
200 * to the compile time generated pmds. This results in invalid pmds up
201 * to the point where we hit the physaddr 0 mapping.
202 *
203 * We limit the mappings to the region from _text to _end. _end is
204 * rounded up to the 2MB boundary. This catches the invalid pmds as
205 * well, as they are located before _text:
206 */
207void __init cleanup_highmap(void)
208{
209 unsigned long vaddr = __START_KERNEL_map;
210 unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
211 pmd_t *pmd = level2_kernel_pgt;
212 pmd_t *last_pmd = pmd + PTRS_PER_PMD;
213
214 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
215 if (!pmd_present(*pmd))
216 continue;
217 if (vaddr < (unsigned long) _text || vaddr > end)
218 set_pmd(pmd, __pmd(0));
219 }
220}
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222/* NOTE: this is meant to be run only at boot */
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100223void __init
224__set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 unsigned long address = __fix_to_virt(idx);
227
228 if (idx >= __end_of_fixed_addresses) {
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100229 printk(KERN_ERR "Invalid __set_fixmap\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return;
231 }
232 set_pte_phys(address, phys, prot);
233}
234
Andi Kleen75175272008-01-30 13:33:17 +0100235static unsigned long __initdata table_start;
236static unsigned long __meminitdata table_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200238static __meminit void *alloc_low_page(unsigned long *phys)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100239{
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200240 unsigned long pfn = table_end++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 void *adr;
242
Matt Tolentino44df75e2006-01-17 07:03:41 +0100243 if (after_bootmem) {
244 adr = (void *)get_zeroed_page(GFP_ATOMIC);
245 *phys = __pa(adr);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100246
Matt Tolentino44df75e2006-01-17 07:03:41 +0100247 return adr;
248 }
249
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100250 if (pfn >= end_pfn)
251 panic("alloc_low_page: ran out of memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200253 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
254 memset(adr, 0, PAGE_SIZE);
255 *phys = pfn * PAGE_SIZE;
256 return adr;
257}
258
259static __meminit void unmap_low_page(void *adr)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100260{
Matt Tolentino44df75e2006-01-17 07:03:41 +0100261 if (after_bootmem)
262 return;
263
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200264 early_iounmap(adr, PAGE_SIZE);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100267/* Must run before zap_low_mappings */
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700268__meminit void *early_ioremap(unsigned long addr, unsigned long size)
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100269{
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200270 pmd_t *pmd, *last_pmd;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100271 unsigned long vaddr;
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200272 int i, pmds;
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100273
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200274 pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
275 vaddr = __START_KERNEL_map;
276 pmd = level2_kernel_pgt;
277 last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100278
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200279 for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
280 for (i = 0; i < pmds; i++) {
281 if (pmd_present(pmd[i]))
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100282 goto continue_outer_loop;
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200283 }
284 vaddr += addr & ~PMD_MASK;
285 addr &= PMD_MASK;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100286
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200287 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
Joerg Roedel929fd582008-01-30 13:31:08 +0100288 set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
Andi Kleen1a2b4412008-01-30 13:33:54 +0100289 __flush_tlb_all();
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100290
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200291 return (void *)vaddr;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100292continue_outer_loop:
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200293 ;
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100294 }
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100295 printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100296
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200297 return NULL;
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100298}
299
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100300/*
301 * To avoid virtual aliases later:
302 */
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700303__meminit void early_iounmap(void *addr, unsigned long size)
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100304{
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200305 unsigned long vaddr;
306 pmd_t *pmd;
307 int i, pmds;
308
309 vaddr = (unsigned long)addr;
310 pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
311 pmd = level2_kernel_pgt + pmd_index(vaddr);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100312
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200313 for (i = 0; i < pmds; i++)
314 pmd_clear(pmd + i);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100315
Andi Kleen1a2b4412008-01-30 13:33:54 +0100316 __flush_tlb_all();
Andi Kleenf2d3efe2006-03-25 16:30:22 +0100317}
318
Matt Tolentino44df75e2006-01-17 07:03:41 +0100319static void __meminit
Keith Mannthey6ad91652006-09-26 10:52:36 +0200320phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
Matt Tolentino44df75e2006-01-17 07:03:41 +0100321{
Keith Mannthey6ad91652006-09-26 10:52:36 +0200322 int i = pmd_index(address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Keith Mannthey6ad91652006-09-26 10:52:36 +0200324 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
Keith Mannthey6ad91652006-09-26 10:52:36 +0200325 pmd_t *pmd = pmd_page + pmd_index(address);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100326
Jan Beulich5f51e132006-06-26 13:59:02 +0200327 if (address >= end) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100328 if (!after_bootmem) {
Jan Beulich5f51e132006-06-26 13:59:02 +0200329 for (; i < PTRS_PER_PMD; i++, pmd++)
330 set_pmd(pmd, __pmd(0));
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100331 }
Matt Tolentino44df75e2006-01-17 07:03:41 +0100332 break;
333 }
Keith Mannthey6ad91652006-09-26 10:52:36 +0200334
335 if (pmd_val(*pmd))
336 continue;
337
Andi Kleend4f71f72008-02-04 16:48:09 +0100338 set_pte((pte_t *)pmd,
339 pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
Matt Tolentino44df75e2006-01-17 07:03:41 +0100340 }
341}
342
343static void __meminit
344phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
345{
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100346 pmd_t *pmd = pmd_offset(pud, 0);
Keith Mannthey6ad91652006-09-26 10:52:36 +0200347 spin_lock(&init_mm.page_table_lock);
348 phys_pmd_init(pmd, address, end);
349 spin_unlock(&init_mm.page_table_lock);
350 __flush_tlb_all();
Matt Tolentino44df75e2006-01-17 07:03:41 +0100351}
352
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100353static void __meminit
354phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
355{
Keith Mannthey6ad91652006-09-26 10:52:36 +0200356 int i = pud_index(addr);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100357
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100358 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
Keith Mannthey6ad91652006-09-26 10:52:36 +0200359 unsigned long pmd_phys;
360 pud_t *pud = pud_page + pud_index(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 pmd_t *pmd;
362
Keith Mannthey6ad91652006-09-26 10:52:36 +0200363 if (addr >= end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100366 if (!after_bootmem &&
367 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
368 set_pud(pud, __pud(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 continue;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Keith Mannthey6ad91652006-09-26 10:52:36 +0200372 if (pud_val(*pud)) {
Andi Kleenef925762008-04-17 17:40:45 +0200373 if (!pud_large(*pud))
374 phys_pmd_update(pud, addr, end);
375 continue;
376 }
377
378 if (direct_gbpages) {
379 set_pte((pte_t *)pud,
380 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
Keith Mannthey6ad91652006-09-26 10:52:36 +0200381 continue;
382 }
383
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200384 pmd = alloc_low_page(&pmd_phys);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100385
Matt Tolentino44df75e2006-01-17 07:03:41 +0100386 spin_lock(&init_mm.page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
Keith Mannthey6ad91652006-09-26 10:52:36 +0200388 phys_pmd_init(pmd, addr, end);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100389 spin_unlock(&init_mm.page_table_lock);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100390
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200391 unmap_low_page(pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 }
Andi Kleen1a2b4412008-01-30 13:33:54 +0100393 __flush_tlb_all();
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100394}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396static void __init find_early_table_space(unsigned long end)
397{
Andi Kleen6c5acd12006-01-11 22:46:57 +0100398 unsigned long puds, pmds, tables, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
Andi Kleenef925762008-04-17 17:40:45 +0200401 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
402 if (!direct_gbpages) {
403 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
404 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100407 /*
408 * RED-PEN putting page tables only on node 0 could
409 * cause a hotspot and fill up ZONE_DMA. The page tables
410 * need roughly 0.5KB per GB.
411 */
412 start = 0x8000;
Yinghai Lu24a5da72008-02-01 17:49:41 +0100413 table_start = find_e820_area(start, end, tables, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (table_start == -1UL)
415 panic("Cannot find space for the kernel page tables");
416
417 table_start >>= PAGE_SHIFT;
418 table_end = table_start;
Matt Tolentino44df75e2006-01-17 07:03:41 +0100419
420 early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
Jan Beulich5f51e132006-06-26 13:59:02 +0200421 end, table_start << PAGE_SHIFT,
422 (table_start << PAGE_SHIFT) + tables);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Andi Kleenef925762008-04-17 17:40:45 +0200425static void __init init_gbpages(void)
426{
427 if (direct_gbpages && cpu_has_gbpages)
428 printk(KERN_INFO "Using GB pages for direct mapping\n");
429 else
430 direct_gbpages = 0;
431}
432
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100433/*
434 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
435 * This runs before bootmem is initialized and gets pages directly from
436 * the physical memory. To access them they are temporarily mapped.
437 */
KAMEZAWA Hiroyukib6fd6ec2007-11-28 16:21:58 -0800438void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100439{
440 unsigned long next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100442 pr_debug("init_memory_mapping\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100444 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 * Find space for the kernel direct mapping tables.
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100446 *
447 * Later we should allocate these tables in the local node of the
448 * memory mapped. Unfortunately this is done currently before the
449 * nodes are discovered.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
Andi Kleenef925762008-04-17 17:40:45 +0200451 if (!after_bootmem) {
452 init_gbpages();
Matt Tolentino44df75e2006-01-17 07:03:41 +0100453 find_early_table_space(end);
Andi Kleenef925762008-04-17 17:40:45 +0200454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 start = (unsigned long)__va(start);
457 end = (unsigned long)__va(end);
458
459 for (; start < end; start = next) {
Matt Tolentino44df75e2006-01-17 07:03:41 +0100460 pgd_t *pgd = pgd_offset_k(start);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100461 unsigned long pud_phys;
Matt Tolentino44df75e2006-01-17 07:03:41 +0100462 pud_t *pud;
463
464 if (after_bootmem)
Andi Kleend2ae5b52006-06-26 13:57:56 +0200465 pud = pud_offset(pgd, start & PGDIR_MASK);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100466 else
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200467 pud = alloc_low_page(&pud_phys);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 next = start + PGDIR_SIZE;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100470 if (next > end)
471 next = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 phys_pud_init(pud, __pa(start), __pa(next));
Matt Tolentino44df75e2006-01-17 07:03:41 +0100473 if (!after_bootmem)
474 set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
Vivek Goyaldafe41e2007-05-02 19:27:06 +0200475 unmap_low_page(pud);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
Matt Tolentino44df75e2006-01-17 07:03:41 +0100478 if (!after_bootmem)
Glauber de Oliveira Costaf51c9452007-07-22 11:12:29 +0200479 mmu_cr4_features = read_cr4();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 __flush_tlb_all();
Andi Kleen75175272008-01-30 13:33:17 +0100481
Yinghai Lu24a5da72008-02-01 17:49:41 +0100482 if (!after_bootmem)
483 reserve_early(table_start << PAGE_SHIFT,
484 table_end << PAGE_SHIFT, "PGTABLE");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Matt Tolentino2b976902005-06-23 00:08:06 -0700487#ifndef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488void __init paging_init(void)
489{
Mel Gorman6391af12006-10-11 01:20:39 -0700490 unsigned long max_zone_pfns[MAX_NR_ZONES];
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100491
Mel Gorman6391af12006-10-11 01:20:39 -0700492 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
493 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
494 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
495 max_zone_pfns[ZONE_NORMAL] = end_pfn;
496
Matt Tolentino44df75e2006-01-17 07:03:41 +0100497 memory_present(0, 0, end_pfn);
498 sparse_init();
Mel Gorman5cb248a2006-09-27 01:49:52 -0700499 free_area_init_nodes(max_zone_pfns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500}
501#endif
502
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100503/*
Matt Tolentino44df75e2006-01-17 07:03:41 +0100504 * Memory hotplug specific functions
Matt Tolentino44df75e2006-01-17 07:03:41 +0100505 */
Matt Tolentino44df75e2006-01-17 07:03:41 +0100506void online_page(struct page *page)
507{
508 ClearPageReserved(page);
Nick Piggin7835e982006-03-22 00:08:40 -0800509 init_page_count(page);
Matt Tolentino44df75e2006-01-17 07:03:41 +0100510 __free_page(page);
511 totalram_pages++;
512 num_physpages++;
513}
514
Yasunori Gotobc02af92006-06-27 02:53:30 -0700515#ifdef CONFIG_MEMORY_HOTPLUG
516/*
Yasunori Gotobc02af92006-06-27 02:53:30 -0700517 * Memory is added always to NORMAL zone. This means you will never get
518 * additional DMA/DMA32 memory.
519 */
520int arch_add_memory(int nid, u64 start, u64 size)
521{
522 struct pglist_data *pgdat = NODE_DATA(nid);
Christoph Lameter776ed982006-09-25 23:31:09 -0700523 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700524 unsigned long start_pfn = start >> PAGE_SHIFT;
525 unsigned long nr_pages = size >> PAGE_SHIFT;
526 int ret;
527
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100528 init_memory_mapping(start, start + size-1);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700529
Yasunori Gotobc02af92006-06-27 02:53:30 -0700530 ret = __add_pages(zone, start_pfn, nr_pages);
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100531 WARN_ON(1);
Yasunori Gotobc02af92006-06-27 02:53:30 -0700532
Yasunori Gotobc02af92006-06-27 02:53:30 -0700533 return ret;
Yasunori Gotobc02af92006-06-27 02:53:30 -0700534}
535EXPORT_SYMBOL_GPL(arch_add_memory);
536
Yasunori Goto82432292006-11-18 22:19:40 -0800537#if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
Keith Mannthey4942e992006-09-30 23:27:06 -0700538int memory_add_physaddr_to_nid(u64 start)
539{
540 return 0;
541}
Keith Mannthey8c2676a2006-09-30 23:27:07 -0700542EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
Keith Mannthey4942e992006-09-30 23:27:06 -0700543#endif
544
Keith Mannthey45e0b782006-09-30 23:27:09 -0700545#endif /* CONFIG_MEMORY_HOTPLUG */
546
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100547static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
548 kcore_modules, kcore_vsyscall;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550void __init mem_init(void)
551{
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200552 long codesize, reservedpages, datasize, initsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Jon Mason0dc243a2006-06-26 13:58:11 +0200554 pci_iommu_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Yinghai Lu48ddb152008-01-30 13:32:36 +0100556 /* clear_bss() already clear the empty_zero_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 reservedpages = 0;
559
560 /* this will put all low memory onto the freelists */
Matt Tolentino2b976902005-06-23 00:08:06 -0700561#ifdef CONFIG_NUMA
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200562 totalram_pages = numa_free_all_bootmem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563#else
Andi Kleen0a43e4b2005-09-12 18:49:24 +0200564 totalram_pages = free_all_bootmem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565#endif
Mel Gorman5cb248a2006-09-27 01:49:52 -0700566 reservedpages = end_pfn - totalram_pages -
567 absent_pages_in_range(0, end_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 after_bootmem = 1;
569
570 codesize = (unsigned long) &_etext - (unsigned long) &_text;
571 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
572 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
573
574 /* Register memory areas for /proc/kcore */
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100575 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
576 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 VMALLOC_END-VMALLOC_START);
578 kclist_add(&kcore_kernel, &_stext, _end - _stext);
579 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100580 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 VSYSCALL_END - VSYSCALL_START);
582
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100583 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100584 "%ldk reserved, %ldk data, %ldk init)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
586 end_pfn << (PAGE_SHIFT-10),
587 codesize >> 10,
588 reservedpages << (PAGE_SHIFT-10),
589 datasize >> 10,
590 initsize >> 10);
Thomas Gleixner76ebd052008-02-09 23:24:09 +0100591
592 cpa_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593}
594
Gerd Hoffmannd167a512006-06-26 13:56:16 +0200595void free_init_pages(char *what, unsigned long begin, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Thomas Gleixnerbfc734b2008-02-09 23:24:09 +0100597 unsigned long addr = begin;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Thomas Gleixnerbfc734b2008-02-09 23:24:09 +0100599 if (addr >= end)
Gerd Hoffmannd167a512006-06-26 13:56:16 +0200600 return;
601
Ingo Molnaree01f112008-01-30 13:34:09 +0100602 /*
603 * If debugging page accesses then do not free this memory but
604 * mark them not present - any buggy init-section access will
605 * create a kernel page fault:
606 */
607#ifdef CONFIG_DEBUG_PAGEALLOC
608 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
609 begin, PAGE_ALIGN(end));
610 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
611#else
Jan Beulich6fb14752007-05-02 19:27:10 +0200612 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100613
Thomas Gleixnerbfc734b2008-02-09 23:24:09 +0100614 for (; addr < end; addr += PAGE_SIZE) {
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700615 ClearPageReserved(virt_to_page(addr));
616 init_page_count(virt_to_page(addr));
617 memset((void *)(addr & ~(PAGE_SIZE-1)),
618 POISON_FREE_INITMEM, PAGE_SIZE);
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700619 free_page(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 totalram_pages++;
621 }
Ingo Molnaree01f112008-01-30 13:34:09 +0100622#endif
Gerd Hoffmannd167a512006-06-26 13:56:16 +0200623}
624
625void free_initmem(void)
626{
Gerd Hoffmannd167a512006-06-26 13:56:16 +0200627 free_init_pages("unused kernel memory",
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700628 (unsigned long)(&__init_begin),
629 (unsigned long)(&__init_end));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630}
631
Arjan van de Ven67df1972006-01-06 00:12:04 -0800632#ifdef CONFIG_DEBUG_RODATA
Arjan van de Venedeed302008-01-30 13:34:08 +0100633const int rodata_test_data = 0xC3;
634EXPORT_SYMBOL_GPL(rodata_test_data);
Arjan van de Ven67df1972006-01-06 00:12:04 -0800635
Arjan van de Ven67df1972006-01-06 00:12:04 -0800636void mark_rodata_ro(void)
637{
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700638 unsigned long start = (unsigned long)_stext, end;
Arjan van de Ven67df1972006-01-06 00:12:04 -0800639
Linus Torvalds602033e2007-07-26 12:07:21 -0700640#ifdef CONFIG_HOTPLUG_CPU
641 /* It must still be possible to apply SMP alternatives. */
642 if (num_possible_cpus() > 1)
643 start = (unsigned long)_etext;
644#endif
645
646#ifdef CONFIG_KPROBES
647 start = (unsigned long)__start_rodata;
648#endif
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100649
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700650 end = (unsigned long)__end_rodata;
651 start = (start + PAGE_SIZE - 1) & PAGE_MASK;
652 end &= PAGE_MASK;
653 if (end <= start)
654 return;
655
Arjan van de Ven67df1972006-01-06 00:12:04 -0800656
Jan Beulich6fb14752007-05-02 19:27:10 +0200657 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700658 (end - start) >> 10);
Arjan van de Ven984bb802008-02-06 22:39:45 +0100659 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
660
661 /*
662 * The rodata section (but not the kernel text!) should also be
663 * not-executable.
664 */
665 start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
666 set_memory_nx(start, (end - start) >> PAGE_SHIFT);
Arjan van de Ven67df1972006-01-06 00:12:04 -0800667
Arjan van de Ven1a487252008-01-30 13:34:09 +0100668 rodata_test();
669
Andi Kleen0c42f392008-01-30 13:33:42 +0100670#ifdef CONFIG_CPA_DEBUG
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100671 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
Arjan van de Ven6d238cc2008-01-30 13:34:06 +0100672 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
Andi Kleen0c42f392008-01-30 13:33:42 +0100673
Ingo Molnar10f22dd2008-01-30 13:34:10 +0100674 printk(KERN_INFO "Testing CPA: again\n");
Arjan van de Ven6d238cc2008-01-30 13:34:06 +0100675 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
Andi Kleen0c42f392008-01-30 13:33:42 +0100676#endif
Arjan van de Ven67df1972006-01-06 00:12:04 -0800677}
678#endif
679
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680#ifdef CONFIG_BLK_DEV_INITRD
681void free_initrd_mem(unsigned long start, unsigned long end)
682{
Linus Torvaldse3ebadd2007-05-07 08:44:24 -0700683 free_init_pages("initrd memory", start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685#endif
686
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100687void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
688{
Matt Tolentino2b976902005-06-23 00:08:06 -0700689#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 int nid = phys_to_nid(phys);
Andi Kleen5e58a022006-11-14 16:57:46 +0100691#endif
692 unsigned long pfn = phys >> PAGE_SHIFT;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100693
Andi Kleen5e58a022006-11-14 16:57:46 +0100694 if (pfn >= end_pfn) {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100695 /*
696 * This can happen with kdump kernels when accessing
697 * firmware tables:
698 */
Andi Kleen5e58a022006-11-14 16:57:46 +0100699 if (pfn < end_pfn_map)
700 return;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100701
Andi Kleen5e58a022006-11-14 16:57:46 +0100702 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
703 phys, len);
704 return;
705 }
706
707 /* Should check here against the e820 map to avoid double free */
708#ifdef CONFIG_NUMA
Bernhard Walle72a7fe32008-02-07 00:15:17 -0800709 reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100710#else
Bernhard Walle72a7fe32008-02-07 00:15:17 -0800711 reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712#endif
Mel Gorman0e0b8642006-09-27 01:49:56 -0700713 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
Andi Kleene18c6872005-11-05 17:25:53 +0100714 dma_reserve += len / PAGE_SIZE;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700715 set_dma_reserve(dma_reserve);
716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100719int kern_addr_valid(unsigned long addr)
720{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100722 pgd_t *pgd;
723 pud_t *pud;
724 pmd_t *pmd;
725 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
727 if (above != 0 && above != -1UL)
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100728 return 0;
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 pgd = pgd_offset_k(addr);
731 if (pgd_none(*pgd))
732 return 0;
733
734 pud = pud_offset(pgd, addr);
735 if (pud_none(*pud))
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100736 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
738 pmd = pmd_offset(pud, addr);
739 if (pmd_none(*pmd))
740 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (pmd_large(*pmd))
743 return pfn_valid(pmd_pfn(*pmd));
744
745 pte = pte_offset_kernel(pmd, addr);
746 if (pte_none(*pte))
747 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 return pfn_valid(pte_pfn(*pte));
750}
751
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100752/*
753 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
754 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
755 * not need special handling anymore:
756 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757static struct vm_area_struct gate_vma = {
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100758 .vm_start = VSYSCALL_START,
759 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
760 .vm_page_prot = PAGE_READONLY_EXEC,
761 .vm_flags = VM_READ | VM_EXEC
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762};
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
765{
766#ifdef CONFIG_IA32_EMULATION
Andi Kleen1e014412005-04-16 15:24:55 -0700767 if (test_tsk_thread_flag(tsk, TIF_IA32))
768 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769#endif
770 return &gate_vma;
771}
772
773int in_gate_area(struct task_struct *task, unsigned long addr)
774{
775 struct vm_area_struct *vma = get_gate_vma(task);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100776
Andi Kleen1e014412005-04-16 15:24:55 -0700777 if (!vma)
778 return 0;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return (addr >= vma->vm_start) && (addr < vma->vm_end);
781}
782
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100783/*
784 * Use this when you have no reliable task/vma, typically from interrupt
785 * context. It is less reliable than using the task's vma and may give
786 * false positives:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
788int in_gate_area_no_task(unsigned long addr)
789{
Andi Kleen1e014412005-04-16 15:24:55 -0700790 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
Zou Nan hai2e1c49d2007-06-01 00:46:28 -0700792
Andi Kleen2aae9502007-07-21 17:10:01 +0200793const char *arch_vma_name(struct vm_area_struct *vma)
794{
795 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
796 return "[vdso]";
797 if (vma == &gate_vma)
798 return "[vsyscall]";
799 return NULL;
800}
Christoph Lameter0889eba2007-10-16 01:24:15 -0700801
802#ifdef CONFIG_SPARSEMEM_VMEMMAP
803/*
804 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
805 */
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100806int __meminit
807vmemmap_populate(struct page *start_page, unsigned long size, int node)
Christoph Lameter0889eba2007-10-16 01:24:15 -0700808{
809 unsigned long addr = (unsigned long)start_page;
810 unsigned long end = (unsigned long)(start_page + size);
811 unsigned long next;
812 pgd_t *pgd;
813 pud_t *pud;
814 pmd_t *pmd;
815
816 for (; addr < end; addr = next) {
817 next = pmd_addr_end(addr, end);
818
819 pgd = vmemmap_pgd_populate(addr, node);
820 if (!pgd)
821 return -ENOMEM;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100822
Christoph Lameter0889eba2007-10-16 01:24:15 -0700823 pud = vmemmap_pud_populate(pgd, addr, node);
824 if (!pud)
825 return -ENOMEM;
826
827 pmd = pmd_offset(pud, addr);
828 if (pmd_none(*pmd)) {
829 pte_t entry;
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100830 void *p;
831
832 p = vmemmap_alloc_block(PMD_SIZE, node);
Christoph Lameter0889eba2007-10-16 01:24:15 -0700833 if (!p)
834 return -ENOMEM;
835
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100836 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
837 PAGE_KERNEL_LARGE);
Christoph Lameter0889eba2007-10-16 01:24:15 -0700838 set_pmd(pmd, __pmd(pte_val(entry)));
839
840 printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
841 addr, addr + PMD_SIZE - 1, p, node);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100842 } else {
Christoph Lameter0889eba2007-10-16 01:24:15 -0700843 vmemmap_verify((pte_t *)pmd, node, addr, next);
Thomas Gleixner14a62c32008-01-30 13:34:10 +0100844 }
Christoph Lameter0889eba2007-10-16 01:24:15 -0700845 }
Christoph Lameter0889eba2007-10-16 01:24:15 -0700846 return 0;
847}
848#endif