Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[~shefty/rdma-dev.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@ucw.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/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/module.h>
31 #include <linux/memory.h>
32 #include <linux/memory_hotplug.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
35
36 #include <asm/processor.h>
37 #include <asm/bios_ebda.h>
38 #include <asm/system.h>
39 #include <asm/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
42 #include <asm/dma.h>
43 #include <asm/fixmap.h>
44 #include <asm/e820.h>
45 #include <asm/apic.h>
46 #include <asm/tlb.h>
47 #include <asm/mmu_context.h>
48 #include <asm/proto.h>
49 #include <asm/smp.h>
50 #include <asm/sections.h>
51 #include <asm/kdebug.h>
52 #include <asm/numa.h>
53 #include <asm/cacheflush.h>
54 #include <asm/init.h>
55 #include <asm/uv/uv.h>
56 #include <asm/setup.h>
57
58 static int __init parse_direct_gbpages_off(char *arg)
59 {
60         direct_gbpages = 0;
61         return 0;
62 }
63 early_param("nogbpages", parse_direct_gbpages_off);
64
65 static int __init parse_direct_gbpages_on(char *arg)
66 {
67         direct_gbpages = 1;
68         return 0;
69 }
70 early_param("gbpages", parse_direct_gbpages_on);
71
72 /*
73  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
74  * physical space so we can cache the place of the first one and move
75  * around without checking the pgd every time.
76  */
77
78 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
79 EXPORT_SYMBOL_GPL(__supported_pte_mask);
80
81 int force_personality32;
82
83 /*
84  * noexec32=on|off
85  * Control non executable heap for 32bit processes.
86  * To control the stack too use noexec=off
87  *
88  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
89  * off  PROT_READ implies PROT_EXEC
90  */
91 static int __init nonx32_setup(char *str)
92 {
93         if (!strcmp(str, "on"))
94                 force_personality32 &= ~READ_IMPLIES_EXEC;
95         else if (!strcmp(str, "off"))
96                 force_personality32 |= READ_IMPLIES_EXEC;
97         return 1;
98 }
99 __setup("noexec32=", nonx32_setup);
100
101 /*
102  * When memory was added/removed make sure all the processes MM have
103  * suitable PGD entries in the local PGD level page.
104  */
105 void sync_global_pgds(unsigned long start, unsigned long end)
106 {
107         unsigned long address;
108
109         for (address = start; address <= end; address += PGDIR_SIZE) {
110                 const pgd_t *pgd_ref = pgd_offset_k(address);
111                 struct page *page;
112
113                 if (pgd_none(*pgd_ref))
114                         continue;
115
116                 spin_lock(&pgd_lock);
117                 list_for_each_entry(page, &pgd_list, lru) {
118                         pgd_t *pgd;
119                         spinlock_t *pgt_lock;
120
121                         pgd = (pgd_t *)page_address(page) + pgd_index(address);
122                         /* the pgt_lock only for Xen */
123                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
124                         spin_lock(pgt_lock);
125
126                         if (pgd_none(*pgd))
127                                 set_pgd(pgd, *pgd_ref);
128                         else
129                                 BUG_ON(pgd_page_vaddr(*pgd)
130                                        != pgd_page_vaddr(*pgd_ref));
131
132                         spin_unlock(pgt_lock);
133                 }
134                 spin_unlock(&pgd_lock);
135         }
136 }
137
138 /*
139  * NOTE: This function is marked __ref because it calls __init function
140  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
141  */
142 static __ref void *spp_getpage(void)
143 {
144         void *ptr;
145
146         if (after_bootmem)
147                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
148         else
149                 ptr = alloc_bootmem_pages(PAGE_SIZE);
150
151         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
152                 panic("set_pte_phys: cannot allocate page data %s\n",
153                         after_bootmem ? "after bootmem" : "");
154         }
155
156         pr_debug("spp_getpage %p\n", ptr);
157
158         return ptr;
159 }
160
161 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
162 {
163         if (pgd_none(*pgd)) {
164                 pud_t *pud = (pud_t *)spp_getpage();
165                 pgd_populate(&init_mm, pgd, pud);
166                 if (pud != pud_offset(pgd, 0))
167                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
168                                pud, pud_offset(pgd, 0));
169         }
170         return pud_offset(pgd, vaddr);
171 }
172
173 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
174 {
175         if (pud_none(*pud)) {
176                 pmd_t *pmd = (pmd_t *) spp_getpage();
177                 pud_populate(&init_mm, pud, pmd);
178                 if (pmd != pmd_offset(pud, 0))
179                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
180                                pmd, pmd_offset(pud, 0));
181         }
182         return pmd_offset(pud, vaddr);
183 }
184
185 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
186 {
187         if (pmd_none(*pmd)) {
188                 pte_t *pte = (pte_t *) spp_getpage();
189                 pmd_populate_kernel(&init_mm, pmd, pte);
190                 if (pte != pte_offset_kernel(pmd, 0))
191                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
192         }
193         return pte_offset_kernel(pmd, vaddr);
194 }
195
196 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
197 {
198         pud_t *pud;
199         pmd_t *pmd;
200         pte_t *pte;
201
202         pud = pud_page + pud_index(vaddr);
203         pmd = fill_pmd(pud, vaddr);
204         pte = fill_pte(pmd, vaddr);
205
206         set_pte(pte, new_pte);
207
208         /*
209          * It's enough to flush this one mapping.
210          * (PGE mappings get flushed as well)
211          */
212         __flush_tlb_one(vaddr);
213 }
214
215 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
216 {
217         pgd_t *pgd;
218         pud_t *pud_page;
219
220         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
221
222         pgd = pgd_offset_k(vaddr);
223         if (pgd_none(*pgd)) {
224                 printk(KERN_ERR
225                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
226                 return;
227         }
228         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
229         set_pte_vaddr_pud(pud_page, vaddr, pteval);
230 }
231
232 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
233 {
234         pgd_t *pgd;
235         pud_t *pud;
236
237         pgd = pgd_offset_k(vaddr);
238         pud = fill_pud(pgd, vaddr);
239         return fill_pmd(pud, vaddr);
240 }
241
242 pte_t * __init populate_extra_pte(unsigned long vaddr)
243 {
244         pmd_t *pmd;
245
246         pmd = populate_extra_pmd(vaddr);
247         return fill_pte(pmd, vaddr);
248 }
249
250 /*
251  * Create large page table mappings for a range of physical addresses.
252  */
253 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
254                                                 pgprot_t prot)
255 {
256         pgd_t *pgd;
257         pud_t *pud;
258         pmd_t *pmd;
259
260         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
261         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
262                 pgd = pgd_offset_k((unsigned long)__va(phys));
263                 if (pgd_none(*pgd)) {
264                         pud = (pud_t *) spp_getpage();
265                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
266                                                 _PAGE_USER));
267                 }
268                 pud = pud_offset(pgd, (unsigned long)__va(phys));
269                 if (pud_none(*pud)) {
270                         pmd = (pmd_t *) spp_getpage();
271                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
272                                                 _PAGE_USER));
273                 }
274                 pmd = pmd_offset(pud, phys);
275                 BUG_ON(!pmd_none(*pmd));
276                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
277         }
278 }
279
280 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
281 {
282         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
283 }
284
285 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
286 {
287         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
288 }
289
290 /*
291  * The head.S code sets up the kernel high mapping:
292  *
293  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
294  *
295  * phys_addr holds the negative offset to the kernel, which is added
296  * to the compile time generated pmds. This results in invalid pmds up
297  * to the point where we hit the physaddr 0 mapping.
298  *
299  * We limit the mappings to the region from _text to _brk_end.  _brk_end
300  * is rounded up to the 2MB boundary. This catches the invalid pmds as
301  * well, as they are located before _text:
302  */
303 void __init cleanup_highmap(void)
304 {
305         unsigned long vaddr = __START_KERNEL_map;
306         unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
307         unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
308         pmd_t *pmd = level2_kernel_pgt;
309
310         for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
311                 if (pmd_none(*pmd))
312                         continue;
313                 if (vaddr < (unsigned long) _text || vaddr > end)
314                         set_pmd(pmd, __pmd(0));
315         }
316 }
317
318 static __ref void *alloc_low_page(unsigned long *phys)
319 {
320         unsigned long pfn = pgt_buf_end++;
321         void *adr;
322
323         if (after_bootmem) {
324                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
325                 *phys = __pa(adr);
326
327                 return adr;
328         }
329
330         if (pfn >= pgt_buf_top)
331                 panic("alloc_low_page: ran out of memory");
332
333         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
334         clear_page(adr);
335         *phys  = pfn * PAGE_SIZE;
336         return adr;
337 }
338
339 static __ref void *map_low_page(void *virt)
340 {
341         void *adr;
342         unsigned long phys, left;
343
344         if (after_bootmem)
345                 return virt;
346
347         phys = __pa(virt);
348         left = phys & (PAGE_SIZE - 1);
349         adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE);
350         adr = (void *)(((unsigned long)adr) | left);
351
352         return adr;
353 }
354
355 static __ref void unmap_low_page(void *adr)
356 {
357         if (after_bootmem)
358                 return;
359
360         early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE);
361 }
362
363 static unsigned long __meminit
364 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
365               pgprot_t prot)
366 {
367         unsigned pages = 0;
368         unsigned long last_map_addr = end;
369         int i;
370
371         pte_t *pte = pte_page + pte_index(addr);
372
373         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
374
375                 if (addr >= end) {
376                         if (!after_bootmem) {
377                                 for(; i < PTRS_PER_PTE; i++, pte++)
378                                         set_pte(pte, __pte(0));
379                         }
380                         break;
381                 }
382
383                 /*
384                  * We will re-use the existing mapping.
385                  * Xen for example has some special requirements, like mapping
386                  * pagetable pages as RO. So assume someone who pre-setup
387                  * these mappings are more intelligent.
388                  */
389                 if (pte_val(*pte)) {
390                         pages++;
391                         continue;
392                 }
393
394                 if (0)
395                         printk("   pte=%p addr=%lx pte=%016lx\n",
396                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
397                 pages++;
398                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
399                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
400         }
401
402         update_page_count(PG_LEVEL_4K, pages);
403
404         return last_map_addr;
405 }
406
407 static unsigned long __meminit
408 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
409               unsigned long page_size_mask, pgprot_t prot)
410 {
411         unsigned long pages = 0;
412         unsigned long last_map_addr = end;
413
414         int i = pmd_index(address);
415
416         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
417                 unsigned long pte_phys;
418                 pmd_t *pmd = pmd_page + pmd_index(address);
419                 pte_t *pte;
420                 pgprot_t new_prot = prot;
421
422                 if (address >= end) {
423                         if (!after_bootmem) {
424                                 for (; i < PTRS_PER_PMD; i++, pmd++)
425                                         set_pmd(pmd, __pmd(0));
426                         }
427                         break;
428                 }
429
430                 if (pmd_val(*pmd)) {
431                         if (!pmd_large(*pmd)) {
432                                 spin_lock(&init_mm.page_table_lock);
433                                 pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
434                                 last_map_addr = phys_pte_init(pte, address,
435                                                                 end, prot);
436                                 unmap_low_page(pte);
437                                 spin_unlock(&init_mm.page_table_lock);
438                                 continue;
439                         }
440                         /*
441                          * If we are ok with PG_LEVEL_2M mapping, then we will
442                          * use the existing mapping,
443                          *
444                          * Otherwise, we will split the large page mapping but
445                          * use the same existing protection bits except for
446                          * large page, so that we don't violate Intel's TLB
447                          * Application note (317080) which says, while changing
448                          * the page sizes, new and old translations should
449                          * not differ with respect to page frame and
450                          * attributes.
451                          */
452                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
453                                 pages++;
454                                 continue;
455                         }
456                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
457                 }
458
459                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
460                         pages++;
461                         spin_lock(&init_mm.page_table_lock);
462                         set_pte((pte_t *)pmd,
463                                 pfn_pte(address >> PAGE_SHIFT,
464                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
465                         spin_unlock(&init_mm.page_table_lock);
466                         last_map_addr = (address & PMD_MASK) + PMD_SIZE;
467                         continue;
468                 }
469
470                 pte = alloc_low_page(&pte_phys);
471                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
472                 unmap_low_page(pte);
473
474                 spin_lock(&init_mm.page_table_lock);
475                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
476                 spin_unlock(&init_mm.page_table_lock);
477         }
478         update_page_count(PG_LEVEL_2M, pages);
479         return last_map_addr;
480 }
481
482 static unsigned long __meminit
483 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
484                          unsigned long page_size_mask)
485 {
486         unsigned long pages = 0;
487         unsigned long last_map_addr = end;
488         int i = pud_index(addr);
489
490         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
491                 unsigned long pmd_phys;
492                 pud_t *pud = pud_page + pud_index(addr);
493                 pmd_t *pmd;
494                 pgprot_t prot = PAGE_KERNEL;
495
496                 if (addr >= end)
497                         break;
498
499                 if (!after_bootmem &&
500                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
501                         set_pud(pud, __pud(0));
502                         continue;
503                 }
504
505                 if (pud_val(*pud)) {
506                         if (!pud_large(*pud)) {
507                                 pmd = map_low_page(pmd_offset(pud, 0));
508                                 last_map_addr = phys_pmd_init(pmd, addr, end,
509                                                          page_size_mask, prot);
510                                 unmap_low_page(pmd);
511                                 __flush_tlb_all();
512                                 continue;
513                         }
514                         /*
515                          * If we are ok with PG_LEVEL_1G mapping, then we will
516                          * use the existing mapping.
517                          *
518                          * Otherwise, we will split the gbpage mapping but use
519                          * the same existing protection  bits except for large
520                          * page, so that we don't violate Intel's TLB
521                          * Application note (317080) which says, while changing
522                          * the page sizes, new and old translations should
523                          * not differ with respect to page frame and
524                          * attributes.
525                          */
526                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
527                                 pages++;
528                                 continue;
529                         }
530                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
531                 }
532
533                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
534                         pages++;
535                         spin_lock(&init_mm.page_table_lock);
536                         set_pte((pte_t *)pud,
537                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
538                         spin_unlock(&init_mm.page_table_lock);
539                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
540                         continue;
541                 }
542
543                 pmd = alloc_low_page(&pmd_phys);
544                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
545                                               prot);
546                 unmap_low_page(pmd);
547
548                 spin_lock(&init_mm.page_table_lock);
549                 pud_populate(&init_mm, pud, __va(pmd_phys));
550                 spin_unlock(&init_mm.page_table_lock);
551         }
552         __flush_tlb_all();
553
554         update_page_count(PG_LEVEL_1G, pages);
555
556         return last_map_addr;
557 }
558
559 unsigned long __meminit
560 kernel_physical_mapping_init(unsigned long start,
561                              unsigned long end,
562                              unsigned long page_size_mask)
563 {
564         bool pgd_changed = false;
565         unsigned long next, last_map_addr = end;
566         unsigned long addr;
567
568         start = (unsigned long)__va(start);
569         end = (unsigned long)__va(end);
570         addr = start;
571
572         for (; start < end; start = next) {
573                 pgd_t *pgd = pgd_offset_k(start);
574                 unsigned long pud_phys;
575                 pud_t *pud;
576
577                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
578                 if (next > end)
579                         next = end;
580
581                 if (pgd_val(*pgd)) {
582                         pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
583                         last_map_addr = phys_pud_init(pud, __pa(start),
584                                                  __pa(end), page_size_mask);
585                         unmap_low_page(pud);
586                         continue;
587                 }
588
589                 pud = alloc_low_page(&pud_phys);
590                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
591                                                  page_size_mask);
592                 unmap_low_page(pud);
593
594                 spin_lock(&init_mm.page_table_lock);
595                 pgd_populate(&init_mm, pgd, __va(pud_phys));
596                 spin_unlock(&init_mm.page_table_lock);
597                 pgd_changed = true;
598         }
599
600         if (pgd_changed)
601                 sync_global_pgds(addr, end);
602
603         __flush_tlb_all();
604
605         return last_map_addr;
606 }
607
608 #ifndef CONFIG_NUMA
609 void __init initmem_init(void)
610 {
611         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
612 }
613 #endif
614
615 void __init paging_init(void)
616 {
617         sparse_memory_present_with_active_regions(MAX_NUMNODES);
618         sparse_init();
619
620         /*
621          * clear the default setting with node 0
622          * note: don't use nodes_clear here, that is really clearing when
623          *       numa support is not compiled in, and later node_set_state
624          *       will not set it back.
625          */
626         node_clear_state(0, N_NORMAL_MEMORY);
627
628         zone_sizes_init();
629 }
630
631 /*
632  * Memory hotplug specific functions
633  */
634 #ifdef CONFIG_MEMORY_HOTPLUG
635 /*
636  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
637  * updating.
638  */
639 static void  update_end_of_memory_vars(u64 start, u64 size)
640 {
641         unsigned long end_pfn = PFN_UP(start + size);
642
643         if (end_pfn > max_pfn) {
644                 max_pfn = end_pfn;
645                 max_low_pfn = end_pfn;
646                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
647         }
648 }
649
650 /*
651  * Memory is added always to NORMAL zone. This means you will never get
652  * additional DMA/DMA32 memory.
653  */
654 int arch_add_memory(int nid, u64 start, u64 size)
655 {
656         struct pglist_data *pgdat = NODE_DATA(nid);
657         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
658         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
659         unsigned long nr_pages = size >> PAGE_SHIFT;
660         int ret;
661
662         last_mapped_pfn = init_memory_mapping(start, start + size);
663         if (last_mapped_pfn > max_pfn_mapped)
664                 max_pfn_mapped = last_mapped_pfn;
665
666         ret = __add_pages(nid, zone, start_pfn, nr_pages);
667         WARN_ON_ONCE(ret);
668
669         /* update max_pfn, max_low_pfn and high_memory */
670         update_end_of_memory_vars(start, size);
671
672         return ret;
673 }
674 EXPORT_SYMBOL_GPL(arch_add_memory);
675
676 #endif /* CONFIG_MEMORY_HOTPLUG */
677
678 static struct kcore_list kcore_vsyscall;
679
680 void __init mem_init(void)
681 {
682         long codesize, reservedpages, datasize, initsize;
683         unsigned long absent_pages;
684
685         pci_iommu_alloc();
686
687         /* clear_bss() already clear the empty_zero_page */
688
689         reservedpages = 0;
690
691         /* this will put all low memory onto the freelists */
692 #ifdef CONFIG_NUMA
693         totalram_pages = numa_free_all_bootmem();
694 #else
695         totalram_pages = free_all_bootmem();
696 #endif
697
698         absent_pages = absent_pages_in_range(0, max_pfn);
699         reservedpages = max_pfn - totalram_pages - absent_pages;
700         after_bootmem = 1;
701
702         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
703         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
704         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
705
706         /* Register memory areas for /proc/kcore */
707         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
708                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
709
710         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
711                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
712                 nr_free_pages() << (PAGE_SHIFT-10),
713                 max_pfn << (PAGE_SHIFT-10),
714                 codesize >> 10,
715                 absent_pages << (PAGE_SHIFT-10),
716                 reservedpages << (PAGE_SHIFT-10),
717                 datasize >> 10,
718                 initsize >> 10);
719 }
720
721 #ifdef CONFIG_DEBUG_RODATA
722 const int rodata_test_data = 0xC3;
723 EXPORT_SYMBOL_GPL(rodata_test_data);
724
725 int kernel_set_to_readonly;
726
727 void set_kernel_text_rw(void)
728 {
729         unsigned long start = PFN_ALIGN(_text);
730         unsigned long end = PFN_ALIGN(__stop___ex_table);
731
732         if (!kernel_set_to_readonly)
733                 return;
734
735         pr_debug("Set kernel text: %lx - %lx for read write\n",
736                  start, end);
737
738         /*
739          * Make the kernel identity mapping for text RW. Kernel text
740          * mapping will always be RO. Refer to the comment in
741          * static_protections() in pageattr.c
742          */
743         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
744 }
745
746 void set_kernel_text_ro(void)
747 {
748         unsigned long start = PFN_ALIGN(_text);
749         unsigned long end = PFN_ALIGN(__stop___ex_table);
750
751         if (!kernel_set_to_readonly)
752                 return;
753
754         pr_debug("Set kernel text: %lx - %lx for read only\n",
755                  start, end);
756
757         /*
758          * Set the kernel identity mapping for text RO.
759          */
760         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
761 }
762
763 void mark_rodata_ro(void)
764 {
765         unsigned long start = PFN_ALIGN(_text);
766         unsigned long rodata_start =
767                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
768         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
769         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
770         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
771         unsigned long data_start = (unsigned long) &_sdata;
772
773         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
774                (end - start) >> 10);
775         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
776
777         kernel_set_to_readonly = 1;
778
779         /*
780          * The rodata section (but not the kernel text!) should also be
781          * not-executable.
782          */
783         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
784
785         rodata_test();
786
787 #ifdef CONFIG_CPA_DEBUG
788         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
789         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
790
791         printk(KERN_INFO "Testing CPA: again\n");
792         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
793 #endif
794
795         free_init_pages("unused kernel memory",
796                         (unsigned long) page_address(virt_to_page(text_end)),
797                         (unsigned long)
798                                  page_address(virt_to_page(rodata_start)));
799         free_init_pages("unused kernel memory",
800                         (unsigned long) page_address(virt_to_page(rodata_end)),
801                         (unsigned long) page_address(virt_to_page(data_start)));
802 }
803
804 #endif
805
806 int kern_addr_valid(unsigned long addr)
807 {
808         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
809         pgd_t *pgd;
810         pud_t *pud;
811         pmd_t *pmd;
812         pte_t *pte;
813
814         if (above != 0 && above != -1UL)
815                 return 0;
816
817         pgd = pgd_offset_k(addr);
818         if (pgd_none(*pgd))
819                 return 0;
820
821         pud = pud_offset(pgd, addr);
822         if (pud_none(*pud))
823                 return 0;
824
825         pmd = pmd_offset(pud, addr);
826         if (pmd_none(*pmd))
827                 return 0;
828
829         if (pmd_large(*pmd))
830                 return pfn_valid(pmd_pfn(*pmd));
831
832         pte = pte_offset_kernel(pmd, addr);
833         if (pte_none(*pte))
834                 return 0;
835
836         return pfn_valid(pte_pfn(*pte));
837 }
838
839 /*
840  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
841  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
842  * not need special handling anymore:
843  */
844 static struct vm_area_struct gate_vma = {
845         .vm_start       = VSYSCALL_START,
846         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
847         .vm_page_prot   = PAGE_READONLY_EXEC,
848         .vm_flags       = VM_READ | VM_EXEC
849 };
850
851 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
852 {
853 #ifdef CONFIG_IA32_EMULATION
854         if (!mm || mm->context.ia32_compat)
855                 return NULL;
856 #endif
857         return &gate_vma;
858 }
859
860 int in_gate_area(struct mm_struct *mm, unsigned long addr)
861 {
862         struct vm_area_struct *vma = get_gate_vma(mm);
863
864         if (!vma)
865                 return 0;
866
867         return (addr >= vma->vm_start) && (addr < vma->vm_end);
868 }
869
870 /*
871  * Use this when you have no reliable mm, typically from interrupt
872  * context. It is less reliable than using a task's mm and may give
873  * false positives.
874  */
875 int in_gate_area_no_mm(unsigned long addr)
876 {
877         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
878 }
879
880 const char *arch_vma_name(struct vm_area_struct *vma)
881 {
882         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
883                 return "[vdso]";
884         if (vma == &gate_vma)
885                 return "[vsyscall]";
886         return NULL;
887 }
888
889 #ifdef CONFIG_X86_UV
890 unsigned long memory_block_size_bytes(void)
891 {
892         if (is_uv_system()) {
893                 printk(KERN_INFO "UV: memory block size 2GB\n");
894                 return 2UL * 1024 * 1024 * 1024;
895         }
896         return MIN_MEMORY_BLOCK_SIZE;
897 }
898 #endif
899
900 #ifdef CONFIG_SPARSEMEM_VMEMMAP
901 /*
902  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
903  */
904 static long __meminitdata addr_start, addr_end;
905 static void __meminitdata *p_start, *p_end;
906 static int __meminitdata node_start;
907
908 int __meminit
909 vmemmap_populate(struct page *start_page, unsigned long size, int node)
910 {
911         unsigned long addr = (unsigned long)start_page;
912         unsigned long end = (unsigned long)(start_page + size);
913         unsigned long next;
914         pgd_t *pgd;
915         pud_t *pud;
916         pmd_t *pmd;
917
918         for (; addr < end; addr = next) {
919                 void *p = NULL;
920
921                 pgd = vmemmap_pgd_populate(addr, node);
922                 if (!pgd)
923                         return -ENOMEM;
924
925                 pud = vmemmap_pud_populate(pgd, addr, node);
926                 if (!pud)
927                         return -ENOMEM;
928
929                 if (!cpu_has_pse) {
930                         next = (addr + PAGE_SIZE) & PAGE_MASK;
931                         pmd = vmemmap_pmd_populate(pud, addr, node);
932
933                         if (!pmd)
934                                 return -ENOMEM;
935
936                         p = vmemmap_pte_populate(pmd, addr, node);
937
938                         if (!p)
939                                 return -ENOMEM;
940
941                         addr_end = addr + PAGE_SIZE;
942                         p_end = p + PAGE_SIZE;
943                 } else {
944                         next = pmd_addr_end(addr, end);
945
946                         pmd = pmd_offset(pud, addr);
947                         if (pmd_none(*pmd)) {
948                                 pte_t entry;
949
950                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
951                                 if (!p)
952                                         return -ENOMEM;
953
954                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
955                                                 PAGE_KERNEL_LARGE);
956                                 set_pmd(pmd, __pmd(pte_val(entry)));
957
958                                 /* check to see if we have contiguous blocks */
959                                 if (p_end != p || node_start != node) {
960                                         if (p_start)
961                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
962                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
963                                         addr_start = addr;
964                                         node_start = node;
965                                         p_start = p;
966                                 }
967
968                                 addr_end = addr + PMD_SIZE;
969                                 p_end = p + PMD_SIZE;
970                         } else
971                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
972                 }
973
974         }
975         sync_global_pgds((unsigned long)start_page, end);
976         return 0;
977 }
978
979 void __meminit vmemmap_populate_print_last(void)
980 {
981         if (p_start) {
982                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
983                         addr_start, addr_end-1, p_start, p_end-1, node_start);
984                 p_start = NULL;
985                 p_end = NULL;
986                 node_start = 0;
987         }
988 }
989 #endif