6e7d5a42a09ae19fa997aff8b104c3c9ffec3f04
[linux-3.10.git] / arch / x86 / mm / init_64.c
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
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>
24 #include <linux/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/module.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/nmi.h>
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>
45 #include <asm/sections.h>
46 #include <asm/kdebug.h>
47 #include <asm/numa.h>
48 #include <asm/cacheflush.h>
49
50 const struct dma_mapping_ops *dma_ops;
51 EXPORT_SYMBOL(dma_ops);
52
53 static unsigned long dma_reserve __initdata;
54
55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
56
57 int direct_gbpages __meminitdata
58 #ifdef CONFIG_DIRECT_GBPAGES
59                                 = 1
60 #endif
61 ;
62
63 static int __init parse_direct_gbpages_off(char *arg)
64 {
65         direct_gbpages = 0;
66         return 0;
67 }
68 early_param("nogbpages", parse_direct_gbpages_off);
69
70 static int __init parse_direct_gbpages_on(char *arg)
71 {
72         direct_gbpages = 1;
73         return 0;
74 }
75 early_param("gbpages", parse_direct_gbpages_on);
76
77 /*
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
83 void show_mem(void)
84 {
85         long i, total = 0, reserved = 0;
86         long shared = 0, cached = 0;
87         struct page *page;
88         pg_data_t *pgdat;
89
90         printk(KERN_INFO "Mem-info:\n");
91         show_free_areas();
92         printk(KERN_INFO "Free swap:       %6ldkB\n",
93                 nr_swap_pages << (PAGE_SHIFT-10));
94
95         for_each_online_pgdat(pgdat) {
96                 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))
102                                 touch_nmi_watchdog();
103
104                         if (!pfn_valid(pgdat->node_start_pfn + i))
105                                 continue;
106
107                         page = pfn_to_page(pgdat->node_start_pfn + i);
108                         total++;
109                         if (PageReserved(page))
110                                 reserved++;
111                         else if (PageSwapCache(page))
112                                 cached++;
113                         else if (page_count(page))
114                                 shared += page_count(page) - 1;
115                 }
116         }
117         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);
121 }
122
123 int after_bootmem;
124
125 static __init void *spp_getpage(void)
126 {
127         void *ptr;
128
129         if (after_bootmem)
130                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
131         else
132                 ptr = alloc_bootmem_pages(PAGE_SIZE);
133
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         }
138
139         pr_debug("spp_getpage %p\n", ptr);
140
141         return ptr;
142 }
143
144 static __init void
145 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
146 {
147         pgd_t *pgd;
148         pud_t *pud;
149         pmd_t *pmd;
150         pte_t *pte, new_pte;
151
152         pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
153
154         pgd = pgd_offset_k(vaddr);
155         if (pgd_none(*pgd)) {
156                 printk(KERN_ERR
157                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
158                 return;
159         }
160         pud = pud_offset(pgd, vaddr);
161         if (pud_none(*pud)) {
162                 pmd = (pmd_t *) spp_getpage();
163                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
164                 if (pmd != pmd_offset(pud, 0)) {
165                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
166                                 pmd, pmd_offset(pud, 0));
167                         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)) {
175                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
176                         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
194 /*
195  * The head.S code sets up the kernel high mapping:
196  *
197  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
198  *
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  */
207 void __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
222 /* NOTE: this is meant to be run only at boot */
223 void __init
224 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
225 {
226         unsigned long address = __fix_to_virt(idx);
227
228         if (idx >= __end_of_fixed_addresses) {
229                 printk(KERN_ERR "Invalid __set_fixmap\n");
230                 return;
231         }
232         set_pte_phys(address, phys, prot);
233 }
234
235 static unsigned long __initdata table_start;
236 static unsigned long __meminitdata table_end;
237
238 static __meminit void *alloc_low_page(unsigned long *phys)
239 {
240         unsigned long pfn = table_end++;
241         void *adr;
242
243         if (after_bootmem) {
244                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
245                 *phys = __pa(adr);
246
247                 return adr;
248         }
249
250         if (pfn >= end_pfn)
251                 panic("alloc_low_page: ran out of memory");
252
253         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
254         memset(adr, 0, PAGE_SIZE);
255         *phys  = pfn * PAGE_SIZE;
256         return adr;
257 }
258
259 static __meminit void unmap_low_page(void *adr)
260 {
261         if (after_bootmem)
262                 return;
263
264         early_iounmap(adr, PAGE_SIZE);
265 }
266
267 /* Must run before zap_low_mappings */
268 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
269 {
270         pmd_t *pmd, *last_pmd;
271         unsigned long vaddr;
272         int i, pmds;
273
274         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;
278
279         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
280                 for (i = 0; i < pmds; i++) {
281                         if (pmd_present(pmd[i]))
282                                 goto continue_outer_loop;
283                 }
284                 vaddr += addr & ~PMD_MASK;
285                 addr &= PMD_MASK;
286
287                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
288                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
289                 __flush_tlb_all();
290
291                 return (void *)vaddr;
292 continue_outer_loop:
293                 ;
294         }
295         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
296
297         return NULL;
298 }
299
300 /*
301  * To avoid virtual aliases later:
302  */
303 __meminit void early_iounmap(void *addr, unsigned long size)
304 {
305         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);
312
313         for (i = 0; i < pmds; i++)
314                 pmd_clear(pmd + i);
315
316         __flush_tlb_all();
317 }
318
319 static void __meminit
320 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
321 {
322         int i = pmd_index(address);
323
324         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
325                 pmd_t *pmd = pmd_page + pmd_index(address);
326
327                 if (address >= end) {
328                         if (!after_bootmem) {
329                                 for (; i < PTRS_PER_PMD; i++, pmd++)
330                                         set_pmd(pmd, __pmd(0));
331                         }
332                         break;
333                 }
334
335                 if (pmd_val(*pmd))
336                         continue;
337
338                 set_pte((pte_t *)pmd,
339                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
340         }
341 }
342
343 static void __meminit
344 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
345 {
346         pmd_t *pmd = pmd_offset(pud, 0);
347         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();
351 }
352
353 static void __meminit
354 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
355 {
356         int i = pud_index(addr);
357
358         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
359                 unsigned long pmd_phys;
360                 pud_t *pud = pud_page + pud_index(addr);
361                 pmd_t *pmd;
362
363                 if (addr >= end)
364                         break;
365
366                 if (!after_bootmem &&
367                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
368                         set_pud(pud, __pud(0));
369                         continue;
370                 }
371
372                 if (pud_val(*pud)) {
373                         phys_pmd_update(pud, addr, end);
374                         continue;
375                 }
376
377                 pmd = alloc_low_page(&pmd_phys);
378
379                 spin_lock(&init_mm.page_table_lock);
380                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
381                 phys_pmd_init(pmd, addr, end);
382                 spin_unlock(&init_mm.page_table_lock);
383
384                 unmap_low_page(pmd);
385         }
386         __flush_tlb_all();
387 }
388
389 static void __init find_early_table_space(unsigned long end)
390 {
391         unsigned long puds, pmds, tables, start;
392
393         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
394         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
395         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
396                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
397
398         /*
399          * RED-PEN putting page tables only on node 0 could
400          * cause a hotspot and fill up ZONE_DMA. The page tables
401          * need roughly 0.5KB per GB.
402          */
403         start = 0x8000;
404         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
405         if (table_start == -1UL)
406                 panic("Cannot find space for the kernel page tables");
407
408         table_start >>= PAGE_SHIFT;
409         table_end = table_start;
410
411         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
412                 end, table_start << PAGE_SHIFT,
413                 (table_start << PAGE_SHIFT) + tables);
414 }
415
416 /*
417  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
418  * This runs before bootmem is initialized and gets pages directly from
419  * the physical memory. To access them they are temporarily mapped.
420  */
421 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
422 {
423         unsigned long next;
424
425         pr_debug("init_memory_mapping\n");
426
427         /*
428          * Find space for the kernel direct mapping tables.
429          *
430          * Later we should allocate these tables in the local node of the
431          * memory mapped. Unfortunately this is done currently before the
432          * nodes are discovered.
433          */
434         if (!after_bootmem)
435                 find_early_table_space(end);
436
437         start = (unsigned long)__va(start);
438         end = (unsigned long)__va(end);
439
440         for (; start < end; start = next) {
441                 pgd_t *pgd = pgd_offset_k(start);
442                 unsigned long pud_phys;
443                 pud_t *pud;
444
445                 if (after_bootmem)
446                         pud = pud_offset(pgd, start & PGDIR_MASK);
447                 else
448                         pud = alloc_low_page(&pud_phys);
449
450                 next = start + PGDIR_SIZE;
451                 if (next > end)
452                         next = end;
453                 phys_pud_init(pud, __pa(start), __pa(next));
454                 if (!after_bootmem)
455                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
456                 unmap_low_page(pud);
457         }
458
459         if (!after_bootmem)
460                 mmu_cr4_features = read_cr4();
461         __flush_tlb_all();
462
463         if (!after_bootmem)
464                 reserve_early(table_start << PAGE_SHIFT,
465                                  table_end << PAGE_SHIFT, "PGTABLE");
466 }
467
468 #ifndef CONFIG_NUMA
469 void __init paging_init(void)
470 {
471         unsigned long max_zone_pfns[MAX_NR_ZONES];
472
473         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
474         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
475         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
476         max_zone_pfns[ZONE_NORMAL] = end_pfn;
477
478         memory_present(0, 0, end_pfn);
479         sparse_init();
480         free_area_init_nodes(max_zone_pfns);
481 }
482 #endif
483
484 /*
485  * Memory hotplug specific functions
486  */
487 void online_page(struct page *page)
488 {
489         ClearPageReserved(page);
490         init_page_count(page);
491         __free_page(page);
492         totalram_pages++;
493         num_physpages++;
494 }
495
496 #ifdef CONFIG_MEMORY_HOTPLUG
497 /*
498  * Memory is added always to NORMAL zone. This means you will never get
499  * additional DMA/DMA32 memory.
500  */
501 int arch_add_memory(int nid, u64 start, u64 size)
502 {
503         struct pglist_data *pgdat = NODE_DATA(nid);
504         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
505         unsigned long start_pfn = start >> PAGE_SHIFT;
506         unsigned long nr_pages = size >> PAGE_SHIFT;
507         int ret;
508
509         init_memory_mapping(start, start + size-1);
510
511         ret = __add_pages(zone, start_pfn, nr_pages);
512         WARN_ON(1);
513
514         return ret;
515 }
516 EXPORT_SYMBOL_GPL(arch_add_memory);
517
518 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
519 int memory_add_physaddr_to_nid(u64 start)
520 {
521         return 0;
522 }
523 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
524 #endif
525
526 #endif /* CONFIG_MEMORY_HOTPLUG */
527
528 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
529                          kcore_modules, kcore_vsyscall;
530
531 void __init mem_init(void)
532 {
533         long codesize, reservedpages, datasize, initsize;
534
535         pci_iommu_alloc();
536
537         /* clear_bss() already clear the empty_zero_page */
538
539         reservedpages = 0;
540
541         /* this will put all low memory onto the freelists */
542 #ifdef CONFIG_NUMA
543         totalram_pages = numa_free_all_bootmem();
544 #else
545         totalram_pages = free_all_bootmem();
546 #endif
547         reservedpages = end_pfn - totalram_pages -
548                                         absent_pages_in_range(0, end_pfn);
549         after_bootmem = 1;
550
551         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
552         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
553         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
554
555         /* Register memory areas for /proc/kcore */
556         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
557         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
558                    VMALLOC_END-VMALLOC_START);
559         kclist_add(&kcore_kernel, &_stext, _end - _stext);
560         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
561         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
562                                  VSYSCALL_END - VSYSCALL_START);
563
564         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
565                                 "%ldk reserved, %ldk data, %ldk init)\n",
566                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
567                 end_pfn << (PAGE_SHIFT-10),
568                 codesize >> 10,
569                 reservedpages << (PAGE_SHIFT-10),
570                 datasize >> 10,
571                 initsize >> 10);
572
573         cpa_init();
574 }
575
576 void free_init_pages(char *what, unsigned long begin, unsigned long end)
577 {
578         unsigned long addr = begin;
579
580         if (addr >= end)
581                 return;
582
583         /*
584          * If debugging page accesses then do not free this memory but
585          * mark them not present - any buggy init-section access will
586          * create a kernel page fault:
587          */
588 #ifdef CONFIG_DEBUG_PAGEALLOC
589         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
590                 begin, PAGE_ALIGN(end));
591         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
592 #else
593         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
594
595         for (; addr < end; addr += PAGE_SIZE) {
596                 ClearPageReserved(virt_to_page(addr));
597                 init_page_count(virt_to_page(addr));
598                 memset((void *)(addr & ~(PAGE_SIZE-1)),
599                         POISON_FREE_INITMEM, PAGE_SIZE);
600                 free_page(addr);
601                 totalram_pages++;
602         }
603 #endif
604 }
605
606 void free_initmem(void)
607 {
608         free_init_pages("unused kernel memory",
609                         (unsigned long)(&__init_begin),
610                         (unsigned long)(&__init_end));
611 }
612
613 #ifdef CONFIG_DEBUG_RODATA
614 const int rodata_test_data = 0xC3;
615 EXPORT_SYMBOL_GPL(rodata_test_data);
616
617 void mark_rodata_ro(void)
618 {
619         unsigned long start = (unsigned long)_stext, end;
620
621 #ifdef CONFIG_HOTPLUG_CPU
622         /* It must still be possible to apply SMP alternatives. */
623         if (num_possible_cpus() > 1)
624                 start = (unsigned long)_etext;
625 #endif
626
627 #ifdef CONFIG_KPROBES
628         start = (unsigned long)__start_rodata;
629 #endif
630
631         end = (unsigned long)__end_rodata;
632         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
633         end &= PAGE_MASK;
634         if (end <= start)
635                 return;
636
637
638         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
639                (end - start) >> 10);
640         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
641
642         /*
643          * The rodata section (but not the kernel text!) should also be
644          * not-executable.
645          */
646         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
647         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
648
649         rodata_test();
650
651 #ifdef CONFIG_CPA_DEBUG
652         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
653         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
654
655         printk(KERN_INFO "Testing CPA: again\n");
656         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
657 #endif
658 }
659 #endif
660
661 #ifdef CONFIG_BLK_DEV_INITRD
662 void free_initrd_mem(unsigned long start, unsigned long end)
663 {
664         free_init_pages("initrd memory", start, end);
665 }
666 #endif
667
668 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
669 {
670 #ifdef CONFIG_NUMA
671         int nid = phys_to_nid(phys);
672 #endif
673         unsigned long pfn = phys >> PAGE_SHIFT;
674
675         if (pfn >= end_pfn) {
676                 /*
677                  * This can happen with kdump kernels when accessing
678                  * firmware tables:
679                  */
680                 if (pfn < end_pfn_map)
681                         return;
682
683                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
684                                 phys, len);
685                 return;
686         }
687
688         /* Should check here against the e820 map to avoid double free */
689 #ifdef CONFIG_NUMA
690         reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
691 #else
692         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
693 #endif
694         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
695                 dma_reserve += len / PAGE_SIZE;
696                 set_dma_reserve(dma_reserve);
697         }
698 }
699
700 int kern_addr_valid(unsigned long addr)
701 {
702         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
703         pgd_t *pgd;
704         pud_t *pud;
705         pmd_t *pmd;
706         pte_t *pte;
707
708         if (above != 0 && above != -1UL)
709                 return 0;
710
711         pgd = pgd_offset_k(addr);
712         if (pgd_none(*pgd))
713                 return 0;
714
715         pud = pud_offset(pgd, addr);
716         if (pud_none(*pud))
717                 return 0;
718
719         pmd = pmd_offset(pud, addr);
720         if (pmd_none(*pmd))
721                 return 0;
722
723         if (pmd_large(*pmd))
724                 return pfn_valid(pmd_pfn(*pmd));
725
726         pte = pte_offset_kernel(pmd, addr);
727         if (pte_none(*pte))
728                 return 0;
729
730         return pfn_valid(pte_pfn(*pte));
731 }
732
733 /*
734  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
735  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
736  * not need special handling anymore:
737  */
738 static struct vm_area_struct gate_vma = {
739         .vm_start       = VSYSCALL_START,
740         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
741         .vm_page_prot   = PAGE_READONLY_EXEC,
742         .vm_flags       = VM_READ | VM_EXEC
743 };
744
745 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
746 {
747 #ifdef CONFIG_IA32_EMULATION
748         if (test_tsk_thread_flag(tsk, TIF_IA32))
749                 return NULL;
750 #endif
751         return &gate_vma;
752 }
753
754 int in_gate_area(struct task_struct *task, unsigned long addr)
755 {
756         struct vm_area_struct *vma = get_gate_vma(task);
757
758         if (!vma)
759                 return 0;
760
761         return (addr >= vma->vm_start) && (addr < vma->vm_end);
762 }
763
764 /*
765  * Use this when you have no reliable task/vma, typically from interrupt
766  * context. It is less reliable than using the task's vma and may give
767  * false positives:
768  */
769 int in_gate_area_no_task(unsigned long addr)
770 {
771         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
772 }
773
774 const char *arch_vma_name(struct vm_area_struct *vma)
775 {
776         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
777                 return "[vdso]";
778         if (vma == &gate_vma)
779                 return "[vsyscall]";
780         return NULL;
781 }
782
783 #ifdef CONFIG_SPARSEMEM_VMEMMAP
784 /*
785  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
786  */
787 int __meminit
788 vmemmap_populate(struct page *start_page, unsigned long size, int node)
789 {
790         unsigned long addr = (unsigned long)start_page;
791         unsigned long end = (unsigned long)(start_page + size);
792         unsigned long next;
793         pgd_t *pgd;
794         pud_t *pud;
795         pmd_t *pmd;
796
797         for (; addr < end; addr = next) {
798                 next = pmd_addr_end(addr, end);
799
800                 pgd = vmemmap_pgd_populate(addr, node);
801                 if (!pgd)
802                         return -ENOMEM;
803
804                 pud = vmemmap_pud_populate(pgd, addr, node);
805                 if (!pud)
806                         return -ENOMEM;
807
808                 pmd = pmd_offset(pud, addr);
809                 if (pmd_none(*pmd)) {
810                         pte_t entry;
811                         void *p;
812
813                         p = vmemmap_alloc_block(PMD_SIZE, node);
814                         if (!p)
815                                 return -ENOMEM;
816
817                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
818                                                         PAGE_KERNEL_LARGE);
819                         set_pmd(pmd, __pmd(pte_val(entry)));
820
821                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
822                                 addr, addr + PMD_SIZE - 1, p, node);
823                 } else {
824                         vmemmap_verify((pte_t *)pmd, node, addr, next);
825                 }
826         }
827         return 0;
828 }
829 #endif