]> bbs.cooldavid.org Git - net-next-2.6.git/blame - arch/x86/mm/init_64.c
x86: clean up apic_32/64.c
[net-next-2.6.git] / arch / x86 / mm / init_64.c
CommitLineData
1da177e4
LT
1/*
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
1da177e4
LT
9#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>
59170891 24#include <linux/pci.h>
6fb14755 25#include <linux/pfn.h>
c9cf5528 26#include <linux/poison.h>
17a941d8 27#include <linux/dma-mapping.h>
44df75e6
MT
28#include <linux/module.h>
29#include <linux/memory_hotplug.h>
ae32b129 30#include <linux/nmi.h>
1da177e4
LT
31
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>
2bc0414e 45#include <asm/sections.h>
718fc13b 46#include <asm/kdebug.h>
aaa64e04 47#include <asm/numa.h>
1da177e4
LT
48
49#ifndef Dprintk
50#define Dprintk(x...)
51#endif
52
e6584504 53const struct dma_mapping_ops* dma_ops;
17a941d8
MBY
54EXPORT_SYMBOL(dma_ops);
55
e18c6874
AK
56static unsigned long dma_reserve __initdata;
57
1da177e4
LT
58DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
59
60/*
61 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
62 * physical space so we can cache the place of the first one and move
63 * around without checking the pgd every time.
64 */
65
66void show_mem(void)
67{
e92343cc
AK
68 long i, total = 0, reserved = 0;
69 long shared = 0, cached = 0;
1da177e4
LT
70 pg_data_t *pgdat;
71 struct page *page;
72
e92343cc 73 printk(KERN_INFO "Mem-info:\n");
1da177e4 74 show_free_areas();
e92343cc 75 printk(KERN_INFO "Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
1da177e4 76
ec936fc5 77 for_each_online_pgdat(pgdat) {
1da177e4 78 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
ae32b129
KR
79 /* this loop can take a while with 256 GB and 4k pages
80 so update the NMI watchdog */
81 if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) {
82 touch_nmi_watchdog();
83 }
12710a56
BP
84 if (!pfn_valid(pgdat->node_start_pfn + i))
85 continue;
1da177e4
LT
86 page = pfn_to_page(pgdat->node_start_pfn + i);
87 total++;
e92343cc
AK
88 if (PageReserved(page))
89 reserved++;
90 else if (PageSwapCache(page))
91 cached++;
92 else if (page_count(page))
93 shared += page_count(page) - 1;
1da177e4
LT
94 }
95 }
e92343cc
AK
96 printk(KERN_INFO "%lu pages of RAM\n", total);
97 printk(KERN_INFO "%lu reserved pages\n",reserved);
98 printk(KERN_INFO "%lu pages shared\n",shared);
99 printk(KERN_INFO "%lu pages swap cached\n",cached);
1da177e4
LT
100}
101
1da177e4
LT
102int after_bootmem;
103
5f44a669 104static __init void *spp_getpage(void)
1da177e4
LT
105{
106 void *ptr;
107 if (after_bootmem)
108 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
109 else
110 ptr = alloc_bootmem_pages(PAGE_SIZE);
111 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK))
112 panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":"");
113
114 Dprintk("spp_getpage %p\n", ptr);
115 return ptr;
116}
117
5f44a669 118static __init void set_pte_phys(unsigned long vaddr,
1da177e4
LT
119 unsigned long phys, pgprot_t prot)
120{
121 pgd_t *pgd;
122 pud_t *pud;
123 pmd_t *pmd;
124 pte_t *pte, new_pte;
125
126 Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys);
127
128 pgd = pgd_offset_k(vaddr);
129 if (pgd_none(*pgd)) {
130 printk("PGD FIXMAP MISSING, it should be setup in head.S!\n");
131 return;
132 }
133 pud = pud_offset(pgd, vaddr);
134 if (pud_none(*pud)) {
135 pmd = (pmd_t *) spp_getpage();
136 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
137 if (pmd != pmd_offset(pud, 0)) {
138 printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0));
139 return;
140 }
141 }
142 pmd = pmd_offset(pud, vaddr);
143 if (pmd_none(*pmd)) {
144 pte = (pte_t *) spp_getpage();
145 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
146 if (pte != pte_offset_kernel(pmd, 0)) {
147 printk("PAGETABLE BUG #02!\n");
148 return;
149 }
150 }
151 new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
152
153 pte = pte_offset_kernel(pmd, vaddr);
154 if (!pte_none(*pte) &&
155 pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
156 pte_ERROR(*pte);
157 set_pte(pte, new_pte);
158
159 /*
160 * It's enough to flush this one mapping.
161 * (PGE mappings get flushed as well)
162 */
163 __flush_tlb_one(vaddr);
164}
165
166/* NOTE: this is meant to be run only at boot */
5f44a669
AK
167void __init
168__set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
1da177e4
LT
169{
170 unsigned long address = __fix_to_virt(idx);
171
172 if (idx >= __end_of_fixed_addresses) {
173 printk("Invalid __set_fixmap\n");
174 return;
175 }
176 set_pte_phys(address, phys, prot);
177}
178
a3142c8e 179unsigned long __meminitdata table_start, table_end;
1da177e4 180
dafe41ee 181static __meminit void *alloc_low_page(unsigned long *phys)
1da177e4 182{
dafe41ee 183 unsigned long pfn = table_end++;
1da177e4
LT
184 void *adr;
185
44df75e6
MT
186 if (after_bootmem) {
187 adr = (void *)get_zeroed_page(GFP_ATOMIC);
188 *phys = __pa(adr);
189 return adr;
190 }
191
1da177e4
LT
192 if (pfn >= end_pfn)
193 panic("alloc_low_page: ran out of memory");
dafe41ee
VG
194
195 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
44df75e6 196 memset(adr, 0, PAGE_SIZE);
dafe41ee
VG
197 *phys = pfn * PAGE_SIZE;
198 return adr;
199}
1da177e4 200
dafe41ee 201static __meminit void unmap_low_page(void *adr)
1da177e4 202{
44df75e6
MT
203
204 if (after_bootmem)
205 return;
206
dafe41ee 207 early_iounmap(adr, PAGE_SIZE);
1da177e4
LT
208}
209
f2d3efed 210/* Must run before zap_low_mappings */
a3142c8e 211__meminit void *early_ioremap(unsigned long addr, unsigned long size)
f2d3efed 212{
dafe41ee
VG
213 unsigned long vaddr;
214 pmd_t *pmd, *last_pmd;
215 int i, pmds;
216
217 pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
218 vaddr = __START_KERNEL_map;
219 pmd = level2_kernel_pgt;
220 last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
221 for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
222 for (i = 0; i < pmds; i++) {
223 if (pmd_present(pmd[i]))
224 goto next;
225 }
226 vaddr += addr & ~PMD_MASK;
227 addr &= PMD_MASK;
228 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
929fd589 229 set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
dafe41ee
VG
230 __flush_tlb();
231 return (void *)vaddr;
232 next:
233 ;
f2d3efed 234 }
dafe41ee
VG
235 printk("early_ioremap(0x%lx, %lu) failed\n", addr, size);
236 return NULL;
f2d3efed
AK
237}
238
239/* To avoid virtual aliases later */
a3142c8e 240__meminit void early_iounmap(void *addr, unsigned long size)
f2d3efed 241{
dafe41ee
VG
242 unsigned long vaddr;
243 pmd_t *pmd;
244 int i, pmds;
245
246 vaddr = (unsigned long)addr;
247 pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
248 pmd = level2_kernel_pgt + pmd_index(vaddr);
249 for (i = 0; i < pmds; i++)
250 pmd_clear(pmd + i);
f2d3efed
AK
251 __flush_tlb();
252}
253
44df75e6 254static void __meminit
6ad91658 255phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
44df75e6 256{
6ad91658 257 int i = pmd_index(address);
44df75e6 258
6ad91658 259 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
44df75e6 260 unsigned long entry;
6ad91658 261 pmd_t *pmd = pmd_page + pmd_index(address);
44df75e6 262
5f51e139
JB
263 if (address >= end) {
264 if (!after_bootmem)
265 for (; i < PTRS_PER_PMD; i++, pmd++)
266 set_pmd(pmd, __pmd(0));
44df75e6
MT
267 break;
268 }
6ad91658
KM
269
270 if (pmd_val(*pmd))
271 continue;
272
40842bf5 273 entry = __PAGE_KERNEL_LARGE|_PAGE_GLOBAL|address;
44df75e6
MT
274 entry &= __supported_pte_mask;
275 set_pmd(pmd, __pmd(entry));
276 }
277}
278
279static void __meminit
280phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
281{
6ad91658
KM
282 pmd_t *pmd = pmd_offset(pud,0);
283 spin_lock(&init_mm.page_table_lock);
284 phys_pmd_init(pmd, address, end);
285 spin_unlock(&init_mm.page_table_lock);
286 __flush_tlb_all();
44df75e6
MT
287}
288
6ad91658 289static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
1da177e4 290{
6ad91658 291 int i = pud_index(addr);
44df75e6 292
44df75e6 293
6ad91658 294 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE ) {
6ad91658
KM
295 unsigned long pmd_phys;
296 pud_t *pud = pud_page + pud_index(addr);
1da177e4
LT
297 pmd_t *pmd;
298
6ad91658 299 if (addr >= end)
1da177e4 300 break;
1da177e4 301
6ad91658 302 if (!after_bootmem && !e820_any_mapped(addr,addr+PUD_SIZE,0)) {
1da177e4
LT
303 set_pud(pud, __pud(0));
304 continue;
305 }
306
6ad91658
KM
307 if (pud_val(*pud)) {
308 phys_pmd_update(pud, addr, end);
309 continue;
310 }
311
dafe41ee 312 pmd = alloc_low_page(&pmd_phys);
44df75e6 313 spin_lock(&init_mm.page_table_lock);
1da177e4 314 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
6ad91658 315 phys_pmd_init(pmd, addr, end);
44df75e6 316 spin_unlock(&init_mm.page_table_lock);
dafe41ee 317 unmap_low_page(pmd);
1da177e4
LT
318 }
319 __flush_tlb();
320}
321
322static void __init find_early_table_space(unsigned long end)
323{
6c5acd16 324 unsigned long puds, pmds, tables, start;
1da177e4
LT
325
326 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
327 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
328 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
329 round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
330
ee408c79
AK
331 /* RED-PEN putting page tables only on node 0 could
332 cause a hotspot and fill up ZONE_DMA. The page tables
333 need roughly 0.5KB per GB. */
334 start = 0x8000;
335 table_start = find_e820_area(start, end, tables);
1da177e4
LT
336 if (table_start == -1UL)
337 panic("Cannot find space for the kernel page tables");
338
339 table_start >>= PAGE_SHIFT;
340 table_end = table_start;
44df75e6
MT
341
342 early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
5f51e139
JB
343 end, table_start << PAGE_SHIFT,
344 (table_start << PAGE_SHIFT) + tables);
1da177e4
LT
345}
346
347/* Setup the direct mapping of the physical memory at PAGE_OFFSET.
348 This runs before bootmem is initialized and gets pages directly from the
349 physical memory. To access them they are temporarily mapped. */
b6fd6ecb 350void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
1da177e4
LT
351{
352 unsigned long next;
353
354 Dprintk("init_memory_mapping\n");
355
356 /*
357 * Find space for the kernel direct mapping tables.
358 * Later we should allocate these tables in the local node of the memory
359 * mapped. Unfortunately this is done currently before the nodes are
360 * discovered.
361 */
44df75e6
MT
362 if (!after_bootmem)
363 find_early_table_space(end);
1da177e4
LT
364
365 start = (unsigned long)__va(start);
366 end = (unsigned long)__va(end);
367
368 for (; start < end; start = next) {
1da177e4 369 unsigned long pud_phys;
44df75e6
MT
370 pgd_t *pgd = pgd_offset_k(start);
371 pud_t *pud;
372
373 if (after_bootmem)
d2ae5b5f 374 pud = pud_offset(pgd, start & PGDIR_MASK);
44df75e6 375 else
dafe41ee 376 pud = alloc_low_page(&pud_phys);
44df75e6 377
1da177e4
LT
378 next = start + PGDIR_SIZE;
379 if (next > end)
380 next = end;
381 phys_pud_init(pud, __pa(start), __pa(next));
44df75e6
MT
382 if (!after_bootmem)
383 set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
dafe41ee 384 unmap_low_page(pud);
1da177e4
LT
385 }
386
44df75e6 387 if (!after_bootmem)
f51c9452 388 mmu_cr4_features = read_cr4();
1da177e4 389 __flush_tlb_all();
1da177e4
LT
390}
391
2b97690f 392#ifndef CONFIG_NUMA
1da177e4
LT
393void __init paging_init(void)
394{
6391af17
MG
395 unsigned long max_zone_pfns[MAX_NR_ZONES];
396 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
397 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
398 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
399 max_zone_pfns[ZONE_NORMAL] = end_pfn;
400
44df75e6
MT
401 memory_present(0, 0, end_pfn);
402 sparse_init();
5cb248ab 403 free_area_init_nodes(max_zone_pfns);
1da177e4
LT
404}
405#endif
406
407/* Unmap a kernel mapping if it exists. This is useful to avoid prefetches
408 from the CPU leading to inconsistent cache lines. address and size
409 must be aligned to 2MB boundaries.
410 Does nothing when the mapping doesn't exist. */
411void __init clear_kernel_mapping(unsigned long address, unsigned long size)
412{
413 unsigned long end = address + size;
414
415 BUG_ON(address & ~LARGE_PAGE_MASK);
416 BUG_ON(size & ~LARGE_PAGE_MASK);
417
418 for (; address < end; address += LARGE_PAGE_SIZE) {
419 pgd_t *pgd = pgd_offset_k(address);
420 pud_t *pud;
421 pmd_t *pmd;
422 if (pgd_none(*pgd))
423 continue;
424 pud = pud_offset(pgd, address);
425 if (pud_none(*pud))
426 continue;
427 pmd = pmd_offset(pud, address);
428 if (!pmd || pmd_none(*pmd))
429 continue;
430 if (0 == (pmd_val(*pmd) & _PAGE_PSE)) {
431 /* Could handle this, but it should not happen currently. */
432 printk(KERN_ERR
433 "clear_kernel_mapping: mapping has been split. will leak memory\n");
434 pmd_ERROR(*pmd);
435 }
436 set_pmd(pmd, __pmd(0));
437 }
438 __flush_tlb_all();
439}
440
44df75e6
MT
441/*
442 * Memory hotplug specific functions
44df75e6 443 */
44df75e6
MT
444void online_page(struct page *page)
445{
446 ClearPageReserved(page);
7835e98b 447 init_page_count(page);
44df75e6
MT
448 __free_page(page);
449 totalram_pages++;
450 num_physpages++;
451}
452
bc02af93 453#ifdef CONFIG_MEMORY_HOTPLUG
9d99aaa3
AK
454/*
455 * Memory is added always to NORMAL zone. This means you will never get
456 * additional DMA/DMA32 memory.
457 */
bc02af93 458int arch_add_memory(int nid, u64 start, u64 size)
44df75e6 459{
bc02af93 460 struct pglist_data *pgdat = NODE_DATA(nid);
776ed98b 461 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
44df75e6
MT
462 unsigned long start_pfn = start >> PAGE_SHIFT;
463 unsigned long nr_pages = size >> PAGE_SHIFT;
464 int ret;
465
45e0b78b
KM
466 init_memory_mapping(start, (start + size -1));
467
44df75e6
MT
468 ret = __add_pages(zone, start_pfn, nr_pages);
469 if (ret)
470 goto error;
471
44df75e6
MT
472 return ret;
473error:
474 printk("%s: Problem encountered in __add_pages!\n", __func__);
475 return ret;
476}
bc02af93 477EXPORT_SYMBOL_GPL(arch_add_memory);
44df75e6 478
8243229f 479#if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
4942e998
KM
480int memory_add_physaddr_to_nid(u64 start)
481{
482 return 0;
483}
8c2676a5 484EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
4942e998
KM
485#endif
486
45e0b78b
KM
487#endif /* CONFIG_MEMORY_HOTPLUG */
488
1da177e4
LT
489static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules,
490 kcore_vsyscall;
491
492void __init mem_init(void)
493{
0a43e4bf 494 long codesize, reservedpages, datasize, initsize;
1da177e4 495
0dc243ae 496 pci_iommu_alloc();
1da177e4 497
1da177e4
LT
498 /* clear the zero-page */
499 memset(empty_zero_page, 0, PAGE_SIZE);
500
501 reservedpages = 0;
502
503 /* this will put all low memory onto the freelists */
2b97690f 504#ifdef CONFIG_NUMA
0a43e4bf 505 totalram_pages = numa_free_all_bootmem();
1da177e4 506#else
0a43e4bf 507 totalram_pages = free_all_bootmem();
1da177e4 508#endif
5cb248ab
MG
509 reservedpages = end_pfn - totalram_pages -
510 absent_pages_in_range(0, end_pfn);
1da177e4
LT
511
512 after_bootmem = 1;
513
514 codesize = (unsigned long) &_etext - (unsigned long) &_text;
515 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
516 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
517
518 /* Register memory areas for /proc/kcore */
519 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
520 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
521 VMALLOC_END-VMALLOC_START);
522 kclist_add(&kcore_kernel, &_stext, _end - _stext);
523 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
524 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
525 VSYSCALL_END - VSYSCALL_START);
526
0a43e4bf 527 printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n",
1da177e4
LT
528 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
529 end_pfn << (PAGE_SHIFT-10),
530 codesize >> 10,
531 reservedpages << (PAGE_SHIFT-10),
532 datasize >> 10,
533 initsize >> 10);
1da177e4
LT
534}
535
d167a518 536void free_init_pages(char *what, unsigned long begin, unsigned long end)
1da177e4
LT
537{
538 unsigned long addr;
539
d167a518
GH
540 if (begin >= end)
541 return;
542
6fb14755 543 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
d167a518 544 for (addr = begin; addr < end; addr += PAGE_SIZE) {
e3ebadd9
LT
545 ClearPageReserved(virt_to_page(addr));
546 init_page_count(virt_to_page(addr));
547 memset((void *)(addr & ~(PAGE_SIZE-1)),
548 POISON_FREE_INITMEM, PAGE_SIZE);
6fb14755
JB
549 if (addr >= __START_KERNEL_map)
550 change_page_attr_addr(addr, 1, __pgprot(0));
e3ebadd9 551 free_page(addr);
1da177e4
LT
552 totalram_pages++;
553 }
6fb14755
JB
554 if (addr > __START_KERNEL_map)
555 global_flush_tlb();
d167a518
GH
556}
557
558void free_initmem(void)
559{
d167a518 560 free_init_pages("unused kernel memory",
e3ebadd9
LT
561 (unsigned long)(&__init_begin),
562 (unsigned long)(&__init_end));
1da177e4
LT
563}
564
67df197b
AV
565#ifdef CONFIG_DEBUG_RODATA
566
67df197b
AV
567void mark_rodata_ro(void)
568{
e3ebadd9 569 unsigned long start = (unsigned long)_stext, end;
67df197b 570
602033ed
LT
571#ifdef CONFIG_HOTPLUG_CPU
572 /* It must still be possible to apply SMP alternatives. */
573 if (num_possible_cpus() > 1)
574 start = (unsigned long)_etext;
575#endif
576
577#ifdef CONFIG_KPROBES
578 start = (unsigned long)__start_rodata;
579#endif
580
e3ebadd9
LT
581 end = (unsigned long)__end_rodata;
582 start = (start + PAGE_SIZE - 1) & PAGE_MASK;
583 end &= PAGE_MASK;
584 if (end <= start)
585 return;
586
587 change_page_attr_addr(start, (end - start) >> PAGE_SHIFT, PAGE_KERNEL_RO);
67df197b 588
6fb14755 589 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
e3ebadd9 590 (end - start) >> 10);
67df197b
AV
591
592 /*
593 * change_page_attr_addr() requires a global_flush_tlb() call after it.
594 * We do this after the printk so that if something went wrong in the
595 * change, the printk gets out at least to give a better debug hint
596 * of who is the culprit.
597 */
598 global_flush_tlb();
599}
600#endif
601
1da177e4
LT
602#ifdef CONFIG_BLK_DEV_INITRD
603void free_initrd_mem(unsigned long start, unsigned long end)
604{
e3ebadd9 605 free_init_pages("initrd memory", start, end);
1da177e4
LT
606}
607#endif
608
609void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
610{
2b97690f 611#ifdef CONFIG_NUMA
1da177e4 612 int nid = phys_to_nid(phys);
5e58a02a
AK
613#endif
614 unsigned long pfn = phys >> PAGE_SHIFT;
615 if (pfn >= end_pfn) {
616 /* This can happen with kdump kernels when accessing firmware
617 tables. */
618 if (pfn < end_pfn_map)
619 return;
620 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
621 phys, len);
622 return;
623 }
624
625 /* Should check here against the e820 map to avoid double free */
626#ifdef CONFIG_NUMA
1da177e4
LT
627 reserve_bootmem_node(NODE_DATA(nid), phys, len);
628#else
629 reserve_bootmem(phys, len);
630#endif
0e0b864e 631 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
e18c6874 632 dma_reserve += len / PAGE_SIZE;
0e0b864e
MG
633 set_dma_reserve(dma_reserve);
634 }
1da177e4
LT
635}
636
637int kern_addr_valid(unsigned long addr)
638{
639 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
640 pgd_t *pgd;
641 pud_t *pud;
642 pmd_t *pmd;
643 pte_t *pte;
644
645 if (above != 0 && above != -1UL)
646 return 0;
647
648 pgd = pgd_offset_k(addr);
649 if (pgd_none(*pgd))
650 return 0;
651
652 pud = pud_offset(pgd, addr);
653 if (pud_none(*pud))
654 return 0;
655
656 pmd = pmd_offset(pud, addr);
657 if (pmd_none(*pmd))
658 return 0;
659 if (pmd_large(*pmd))
660 return pfn_valid(pmd_pfn(*pmd));
661
662 pte = pte_offset_kernel(pmd, addr);
663 if (pte_none(*pte))
664 return 0;
665 return pfn_valid(pte_pfn(*pte));
666}
667
103efcd9 668/* A pseudo VMA to allow ptrace access for the vsyscall page. This only
1e014410
AK
669 covers the 64bit vsyscall page now. 32bit has a real VMA now and does
670 not need special handling anymore. */
1da177e4
LT
671
672static struct vm_area_struct gate_vma = {
673 .vm_start = VSYSCALL_START,
103efcd9
EP
674 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES << PAGE_SHIFT),
675 .vm_page_prot = PAGE_READONLY_EXEC,
676 .vm_flags = VM_READ | VM_EXEC
1da177e4
LT
677};
678
1da177e4
LT
679struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
680{
681#ifdef CONFIG_IA32_EMULATION
1e014410
AK
682 if (test_tsk_thread_flag(tsk, TIF_IA32))
683 return NULL;
1da177e4
LT
684#endif
685 return &gate_vma;
686}
687
688int in_gate_area(struct task_struct *task, unsigned long addr)
689{
690 struct vm_area_struct *vma = get_gate_vma(task);
1e014410
AK
691 if (!vma)
692 return 0;
1da177e4
LT
693 return (addr >= vma->vm_start) && (addr < vma->vm_end);
694}
695
696/* Use this when you have no reliable task/vma, typically from interrupt
697 * context. It is less reliable than using the task's vma and may give
698 * false positives.
699 */
700int in_gate_area_no_task(unsigned long addr)
701{
1e014410 702 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
1da177e4 703}
2e1c49db 704
2aae950b
AK
705const char *arch_vma_name(struct vm_area_struct *vma)
706{
707 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
708 return "[vdso]";
709 if (vma == &gate_vma)
710 return "[vsyscall]";
711 return NULL;
712}
0889eba5
CL
713
714#ifdef CONFIG_SPARSEMEM_VMEMMAP
715/*
716 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
717 */
718int __meminit vmemmap_populate(struct page *start_page,
719 unsigned long size, int node)
720{
721 unsigned long addr = (unsigned long)start_page;
722 unsigned long end = (unsigned long)(start_page + size);
723 unsigned long next;
724 pgd_t *pgd;
725 pud_t *pud;
726 pmd_t *pmd;
727
728 for (; addr < end; addr = next) {
729 next = pmd_addr_end(addr, end);
730
731 pgd = vmemmap_pgd_populate(addr, node);
732 if (!pgd)
733 return -ENOMEM;
734 pud = vmemmap_pud_populate(pgd, addr, node);
735 if (!pud)
736 return -ENOMEM;
737
738 pmd = pmd_offset(pud, addr);
739 if (pmd_none(*pmd)) {
740 pte_t entry;
741 void *p = vmemmap_alloc_block(PMD_SIZE, node);
742 if (!p)
743 return -ENOMEM;
744
27ec161f 745 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL_LARGE);
0889eba5
CL
746 set_pmd(pmd, __pmd(pte_val(entry)));
747
748 printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
749 addr, addr + PMD_SIZE - 1, p, node);
750 } else
751 vmemmap_verify((pte_t *)pmd, node, addr, next);
752 }
753
754 return 0;
755}
756#endif