Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[~shefty/rdma-dev.git] / arch / x86 / mm / init_32.c
1 /*
2  *
3  *  Copyright (C) 1995  Linus Torvalds
4  *
5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6  */
7
8 #include <linux/module.h>
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/hugetlb.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/bootmem.h>
28 #include <linux/memblock.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33 #include <linux/gfp.h>
34
35 #include <asm/asm.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/processor.h>
38 #include <asm/system.h>
39 #include <asm/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/dma.h>
42 #include <asm/fixmap.h>
43 #include <asm/e820.h>
44 #include <asm/apic.h>
45 #include <asm/bugs.h>
46 #include <asm/tlb.h>
47 #include <asm/tlbflush.h>
48 #include <asm/olpc_ofw.h>
49 #include <asm/pgalloc.h>
50 #include <asm/sections.h>
51 #include <asm/paravirt.h>
52 #include <asm/setup.h>
53 #include <asm/cacheflush.h>
54 #include <asm/page_types.h>
55 #include <asm/init.h>
56
57 unsigned long highstart_pfn, highend_pfn;
58
59 static noinline int do_test_wp_bit(void);
60
61 bool __read_mostly __vmalloc_start_set = false;
62
63 static __init void *alloc_low_page(void)
64 {
65         unsigned long pfn = pgt_buf_end++;
66         void *adr;
67
68         if (pfn >= pgt_buf_top)
69                 panic("alloc_low_page: ran out of memory");
70
71         adr = __va(pfn * PAGE_SIZE);
72         clear_page(adr);
73         return adr;
74 }
75
76 /*
77  * Creates a middle page table and puts a pointer to it in the
78  * given global directory entry. This only returns the gd entry
79  * in non-PAE compilation mode, since the middle layer is folded.
80  */
81 static pmd_t * __init one_md_table_init(pgd_t *pgd)
82 {
83         pud_t *pud;
84         pmd_t *pmd_table;
85
86 #ifdef CONFIG_X86_PAE
87         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
88                 if (after_bootmem)
89                         pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE);
90                 else
91                         pmd_table = (pmd_t *)alloc_low_page();
92                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
93                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
94                 pud = pud_offset(pgd, 0);
95                 BUG_ON(pmd_table != pmd_offset(pud, 0));
96
97                 return pmd_table;
98         }
99 #endif
100         pud = pud_offset(pgd, 0);
101         pmd_table = pmd_offset(pud, 0);
102
103         return pmd_table;
104 }
105
106 /*
107  * Create a page table and place a pointer to it in a middle page
108  * directory entry:
109  */
110 static pte_t * __init one_page_table_init(pmd_t *pmd)
111 {
112         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
113                 pte_t *page_table = NULL;
114
115                 if (after_bootmem) {
116 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
117                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
118 #endif
119                         if (!page_table)
120                                 page_table =
121                                 (pte_t *)alloc_bootmem_pages(PAGE_SIZE);
122                 } else
123                         page_table = (pte_t *)alloc_low_page();
124
125                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
126                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
127                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
128         }
129
130         return pte_offset_kernel(pmd, 0);
131 }
132
133 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
134 {
135         int pgd_idx = pgd_index(vaddr);
136         int pmd_idx = pmd_index(vaddr);
137
138         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
139 }
140
141 pte_t * __init populate_extra_pte(unsigned long vaddr)
142 {
143         int pte_idx = pte_index(vaddr);
144         pmd_t *pmd;
145
146         pmd = populate_extra_pmd(vaddr);
147         return one_page_table_init(pmd) + pte_idx;
148 }
149
150 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
151                                            unsigned long vaddr, pte_t *lastpte)
152 {
153 #ifdef CONFIG_HIGHMEM
154         /*
155          * Something (early fixmap) may already have put a pte
156          * page here, which causes the page table allocation
157          * to become nonlinear. Attempt to fix it, and if it
158          * is still nonlinear then we have to bug.
159          */
160         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
161         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
162
163         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
164             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
165             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
166             && ((__pa(pte) >> PAGE_SHIFT) < pgt_buf_start
167                 || (__pa(pte) >> PAGE_SHIFT) >= pgt_buf_end)) {
168                 pte_t *newpte;
169                 int i;
170
171                 BUG_ON(after_bootmem);
172                 newpte = alloc_low_page();
173                 for (i = 0; i < PTRS_PER_PTE; i++)
174                         set_pte(newpte + i, pte[i]);
175
176                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
177                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
178                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
179                 __flush_tlb_all();
180
181                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
182                 pte = newpte;
183         }
184         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
185                && vaddr > fix_to_virt(FIX_KMAP_END)
186                && lastpte && lastpte + PTRS_PER_PTE != pte);
187 #endif
188         return pte;
189 }
190
191 /*
192  * This function initializes a certain range of kernel virtual memory
193  * with new bootmem page tables, everywhere page tables are missing in
194  * the given range.
195  *
196  * NOTE: The pagetables are allocated contiguous on the physical space
197  * so we can cache the place of the first one and move around without
198  * checking the pgd every time.
199  */
200 static void __init
201 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
202 {
203         int pgd_idx, pmd_idx;
204         unsigned long vaddr;
205         pgd_t *pgd;
206         pmd_t *pmd;
207         pte_t *pte = NULL;
208
209         vaddr = start;
210         pgd_idx = pgd_index(vaddr);
211         pmd_idx = pmd_index(vaddr);
212         pgd = pgd_base + pgd_idx;
213
214         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
215                 pmd = one_md_table_init(pgd);
216                 pmd = pmd + pmd_index(vaddr);
217                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
218                                                         pmd++, pmd_idx++) {
219                         pte = page_table_kmap_check(one_page_table_init(pmd),
220                                                     pmd, vaddr, pte);
221
222                         vaddr += PMD_SIZE;
223                 }
224                 pmd_idx = 0;
225         }
226 }
227
228 static inline int is_kernel_text(unsigned long addr)
229 {
230         if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
231                 return 1;
232         return 0;
233 }
234
235 /*
236  * This maps the physical memory to kernel virtual address space, a total
237  * of max_low_pfn pages, by creating page tables starting from address
238  * PAGE_OFFSET:
239  */
240 unsigned long __init
241 kernel_physical_mapping_init(unsigned long start,
242                              unsigned long end,
243                              unsigned long page_size_mask)
244 {
245         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
246         unsigned long last_map_addr = end;
247         unsigned long start_pfn, end_pfn;
248         pgd_t *pgd_base = swapper_pg_dir;
249         int pgd_idx, pmd_idx, pte_ofs;
250         unsigned long pfn;
251         pgd_t *pgd;
252         pmd_t *pmd;
253         pte_t *pte;
254         unsigned pages_2m, pages_4k;
255         int mapping_iter;
256
257         start_pfn = start >> PAGE_SHIFT;
258         end_pfn = end >> PAGE_SHIFT;
259
260         /*
261          * First iteration will setup identity mapping using large/small pages
262          * based on use_pse, with other attributes same as set by
263          * the early code in head_32.S
264          *
265          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
266          * as desired for the kernel identity mapping.
267          *
268          * This two pass mechanism conforms to the TLB app note which says:
269          *
270          *     "Software should not write to a paging-structure entry in a way
271          *      that would change, for any linear address, both the page size
272          *      and either the page frame or attributes."
273          */
274         mapping_iter = 1;
275
276         if (!cpu_has_pse)
277                 use_pse = 0;
278
279 repeat:
280         pages_2m = pages_4k = 0;
281         pfn = start_pfn;
282         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
283         pgd = pgd_base + pgd_idx;
284         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
285                 pmd = one_md_table_init(pgd);
286
287                 if (pfn >= end_pfn)
288                         continue;
289 #ifdef CONFIG_X86_PAE
290                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
291                 pmd += pmd_idx;
292 #else
293                 pmd_idx = 0;
294 #endif
295                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
296                      pmd++, pmd_idx++) {
297                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
298
299                         /*
300                          * Map with big pages if possible, otherwise
301                          * create normal page tables:
302                          */
303                         if (use_pse) {
304                                 unsigned int addr2;
305                                 pgprot_t prot = PAGE_KERNEL_LARGE;
306                                 /*
307                                  * first pass will use the same initial
308                                  * identity mapping attribute + _PAGE_PSE.
309                                  */
310                                 pgprot_t init_prot =
311                                         __pgprot(PTE_IDENT_ATTR |
312                                                  _PAGE_PSE);
313
314                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
315                                         PAGE_OFFSET + PAGE_SIZE-1;
316
317                                 if (is_kernel_text(addr) ||
318                                     is_kernel_text(addr2))
319                                         prot = PAGE_KERNEL_LARGE_EXEC;
320
321                                 pages_2m++;
322                                 if (mapping_iter == 1)
323                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
324                                 else
325                                         set_pmd(pmd, pfn_pmd(pfn, prot));
326
327                                 pfn += PTRS_PER_PTE;
328                                 continue;
329                         }
330                         pte = one_page_table_init(pmd);
331
332                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
333                         pte += pte_ofs;
334                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
335                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
336                                 pgprot_t prot = PAGE_KERNEL;
337                                 /*
338                                  * first pass will use the same initial
339                                  * identity mapping attribute.
340                                  */
341                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
342
343                                 if (is_kernel_text(addr))
344                                         prot = PAGE_KERNEL_EXEC;
345
346                                 pages_4k++;
347                                 if (mapping_iter == 1) {
348                                         set_pte(pte, pfn_pte(pfn, init_prot));
349                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
350                                 } else
351                                         set_pte(pte, pfn_pte(pfn, prot));
352                         }
353                 }
354         }
355         if (mapping_iter == 1) {
356                 /*
357                  * update direct mapping page count only in the first
358                  * iteration.
359                  */
360                 update_page_count(PG_LEVEL_2M, pages_2m);
361                 update_page_count(PG_LEVEL_4K, pages_4k);
362
363                 /*
364                  * local global flush tlb, which will flush the previous
365                  * mappings present in both small and large page TLB's.
366                  */
367                 __flush_tlb_all();
368
369                 /*
370                  * Second iteration will set the actual desired PTE attributes.
371                  */
372                 mapping_iter = 2;
373                 goto repeat;
374         }
375         return last_map_addr;
376 }
377
378 pte_t *kmap_pte;
379 pgprot_t kmap_prot;
380
381 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
382 {
383         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
384                         vaddr), vaddr), vaddr);
385 }
386
387 static void __init kmap_init(void)
388 {
389         unsigned long kmap_vstart;
390
391         /*
392          * Cache the first kmap pte:
393          */
394         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
395         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
396
397         kmap_prot = PAGE_KERNEL;
398 }
399
400 #ifdef CONFIG_HIGHMEM
401 static void __init permanent_kmaps_init(pgd_t *pgd_base)
402 {
403         unsigned long vaddr;
404         pgd_t *pgd;
405         pud_t *pud;
406         pmd_t *pmd;
407         pte_t *pte;
408
409         vaddr = PKMAP_BASE;
410         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
411
412         pgd = swapper_pg_dir + pgd_index(vaddr);
413         pud = pud_offset(pgd, vaddr);
414         pmd = pmd_offset(pud, vaddr);
415         pte = pte_offset_kernel(pmd, vaddr);
416         pkmap_page_table = pte;
417 }
418
419 static void __init add_one_highpage_init(struct page *page)
420 {
421         ClearPageReserved(page);
422         init_page_count(page);
423         __free_page(page);
424         totalhigh_pages++;
425 }
426
427 void __init add_highpages_with_active_regions(int nid,
428                          unsigned long start_pfn, unsigned long end_pfn)
429 {
430         phys_addr_t start, end;
431         u64 i;
432
433         for_each_free_mem_range(i, nid, &start, &end, NULL) {
434                 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
435                                             start_pfn, end_pfn);
436                 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
437                                               start_pfn, end_pfn);
438                 for ( ; pfn < e_pfn; pfn++)
439                         if (pfn_valid(pfn))
440                                 add_one_highpage_init(pfn_to_page(pfn));
441         }
442 }
443 #else
444 static inline void permanent_kmaps_init(pgd_t *pgd_base)
445 {
446 }
447 #endif /* CONFIG_HIGHMEM */
448
449 void __init native_pagetable_setup_start(pgd_t *base)
450 {
451         unsigned long pfn, va;
452         pgd_t *pgd;
453         pud_t *pud;
454         pmd_t *pmd;
455         pte_t *pte;
456
457         /*
458          * Remove any mappings which extend past the end of physical
459          * memory from the boot time page table:
460          */
461         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
462                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
463                 pgd = base + pgd_index(va);
464                 if (!pgd_present(*pgd))
465                         break;
466
467                 pud = pud_offset(pgd, va);
468                 pmd = pmd_offset(pud, va);
469                 if (!pmd_present(*pmd))
470                         break;
471
472                 pte = pte_offset_kernel(pmd, va);
473                 if (!pte_present(*pte))
474                         break;
475
476                 pte_clear(NULL, va, pte);
477         }
478         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
479 }
480
481 void __init native_pagetable_setup_done(pgd_t *base)
482 {
483 }
484
485 /*
486  * Build a proper pagetable for the kernel mappings.  Up until this
487  * point, we've been running on some set of pagetables constructed by
488  * the boot process.
489  *
490  * If we're booting on native hardware, this will be a pagetable
491  * constructed in arch/x86/kernel/head_32.S.  The root of the
492  * pagetable will be swapper_pg_dir.
493  *
494  * If we're booting paravirtualized under a hypervisor, then there are
495  * more options: we may already be running PAE, and the pagetable may
496  * or may not be based in swapper_pg_dir.  In any case,
497  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
498  * appropriately for the rest of the initialization to work.
499  *
500  * In general, pagetable_init() assumes that the pagetable may already
501  * be partially populated, and so it avoids stomping on any existing
502  * mappings.
503  */
504 void __init early_ioremap_page_table_range_init(void)
505 {
506         pgd_t *pgd_base = swapper_pg_dir;
507         unsigned long vaddr, end;
508
509         /*
510          * Fixed mappings, only the page table structure has to be
511          * created - mappings will be set by set_fixmap():
512          */
513         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
514         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
515         page_table_range_init(vaddr, end, pgd_base);
516         early_ioremap_reset();
517 }
518
519 static void __init pagetable_init(void)
520 {
521         pgd_t *pgd_base = swapper_pg_dir;
522
523         permanent_kmaps_init(pgd_base);
524 }
525
526 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
527 EXPORT_SYMBOL_GPL(__supported_pte_mask);
528
529 /* user-defined highmem size */
530 static unsigned int highmem_pages = -1;
531
532 /*
533  * highmem=size forces highmem to be exactly 'size' bytes.
534  * This works even on boxes that have no highmem otherwise.
535  * This also works to reduce highmem size on bigger boxes.
536  */
537 static int __init parse_highmem(char *arg)
538 {
539         if (!arg)
540                 return -EINVAL;
541
542         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
543         return 0;
544 }
545 early_param("highmem", parse_highmem);
546
547 #define MSG_HIGHMEM_TOO_BIG \
548         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
549
550 #define MSG_LOWMEM_TOO_SMALL \
551         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
552 /*
553  * All of RAM fits into lowmem - but if user wants highmem
554  * artificially via the highmem=x boot parameter then create
555  * it:
556  */
557 void __init lowmem_pfn_init(void)
558 {
559         /* max_low_pfn is 0, we already have early_res support */
560         max_low_pfn = max_pfn;
561
562         if (highmem_pages == -1)
563                 highmem_pages = 0;
564 #ifdef CONFIG_HIGHMEM
565         if (highmem_pages >= max_pfn) {
566                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
567                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
568                 highmem_pages = 0;
569         }
570         if (highmem_pages) {
571                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
572                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
573                                 pages_to_mb(highmem_pages));
574                         highmem_pages = 0;
575                 }
576                 max_low_pfn -= highmem_pages;
577         }
578 #else
579         if (highmem_pages)
580                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
581 #endif
582 }
583
584 #define MSG_HIGHMEM_TOO_SMALL \
585         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
586
587 #define MSG_HIGHMEM_TRIMMED \
588         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
589 /*
590  * We have more RAM than fits into lowmem - we try to put it into
591  * highmem, also taking the highmem=x boot parameter into account:
592  */
593 void __init highmem_pfn_init(void)
594 {
595         max_low_pfn = MAXMEM_PFN;
596
597         if (highmem_pages == -1)
598                 highmem_pages = max_pfn - MAXMEM_PFN;
599
600         if (highmem_pages + MAXMEM_PFN < max_pfn)
601                 max_pfn = MAXMEM_PFN + highmem_pages;
602
603         if (highmem_pages + MAXMEM_PFN > max_pfn) {
604                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
605                         pages_to_mb(max_pfn - MAXMEM_PFN),
606                         pages_to_mb(highmem_pages));
607                 highmem_pages = 0;
608         }
609 #ifndef CONFIG_HIGHMEM
610         /* Maximum memory usable is what is directly addressable */
611         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
612         if (max_pfn > MAX_NONPAE_PFN)
613                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
614         else
615                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
616         max_pfn = MAXMEM_PFN;
617 #else /* !CONFIG_HIGHMEM */
618 #ifndef CONFIG_HIGHMEM64G
619         if (max_pfn > MAX_NONPAE_PFN) {
620                 max_pfn = MAX_NONPAE_PFN;
621                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
622         }
623 #endif /* !CONFIG_HIGHMEM64G */
624 #endif /* !CONFIG_HIGHMEM */
625 }
626
627 /*
628  * Determine low and high memory ranges:
629  */
630 void __init find_low_pfn_range(void)
631 {
632         /* it could update max_pfn */
633
634         if (max_pfn <= MAXMEM_PFN)
635                 lowmem_pfn_init();
636         else
637                 highmem_pfn_init();
638 }
639
640 #ifndef CONFIG_NEED_MULTIPLE_NODES
641 void __init initmem_init(void)
642 {
643 #ifdef CONFIG_HIGHMEM
644         highstart_pfn = highend_pfn = max_pfn;
645         if (max_pfn > max_low_pfn)
646                 highstart_pfn = max_low_pfn;
647         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
648                 pages_to_mb(highend_pfn - highstart_pfn));
649         num_physpages = highend_pfn;
650         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
651 #else
652         num_physpages = max_low_pfn;
653         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
654 #endif
655
656         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
657         sparse_memory_present_with_active_regions(0);
658
659 #ifdef CONFIG_FLATMEM
660         max_mapnr = num_physpages;
661 #endif
662         __vmalloc_start_set = true;
663
664         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
665                         pages_to_mb(max_low_pfn));
666
667         setup_bootmem_allocator();
668 }
669 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
670
671 void __init setup_bootmem_allocator(void)
672 {
673         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
674                  max_pfn_mapped<<PAGE_SHIFT);
675         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
676
677         after_bootmem = 1;
678 }
679
680 /*
681  * paging_init() sets up the page tables - note that the first 8MB are
682  * already mapped by head.S.
683  *
684  * This routines also unmaps the page at virtual kernel address 0, so
685  * that we can trap those pesky NULL-reference errors in the kernel.
686  */
687 void __init paging_init(void)
688 {
689         pagetable_init();
690
691         __flush_tlb_all();
692
693         kmap_init();
694
695         /*
696          * NOTE: at this point the bootmem allocator is fully available.
697          */
698         olpc_dt_build_devicetree();
699         sparse_memory_present_with_active_regions(MAX_NUMNODES);
700         sparse_init();
701         zone_sizes_init();
702 }
703
704 /*
705  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
706  * and also on some strange 486's. All 586+'s are OK. This used to involve
707  * black magic jumps to work around some nasty CPU bugs, but fortunately the
708  * switch to using exceptions got rid of all that.
709  */
710 static void __init test_wp_bit(void)
711 {
712         printk(KERN_INFO
713   "Checking if this processor honours the WP bit even in supervisor mode...");
714
715         /* Any page-aligned address will do, the test is non-destructive */
716         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
717         boot_cpu_data.wp_works_ok = do_test_wp_bit();
718         clear_fixmap(FIX_WP_TEST);
719
720         if (!boot_cpu_data.wp_works_ok) {
721                 printk(KERN_CONT "No.\n");
722 #ifdef CONFIG_X86_WP_WORKS_OK
723                 panic(
724   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
725 #endif
726         } else {
727                 printk(KERN_CONT "Ok.\n");
728         }
729 }
730
731 void __init mem_init(void)
732 {
733         int codesize, reservedpages, datasize, initsize;
734         int tmp;
735
736         pci_iommu_alloc();
737
738 #ifdef CONFIG_FLATMEM
739         BUG_ON(!mem_map);
740 #endif
741         /*
742          * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
743          * be done before free_all_bootmem(). Memblock use free low memory for
744          * temporary data (see find_range_array()) and for this purpose can use
745          * pages that was already passed to the buddy allocator, hence marked as
746          * not accessible in the page tables when compiled with
747          * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
748          * important here.
749          */
750         set_highmem_pages_init();
751
752         /* this will put all low memory onto the freelists */
753         totalram_pages += free_all_bootmem();
754
755         reservedpages = 0;
756         for (tmp = 0; tmp < max_low_pfn; tmp++)
757                 /*
758                  * Only count reserved RAM pages:
759                  */
760                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
761                         reservedpages++;
762
763         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
764         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
765         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
766
767         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
768                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
769                 nr_free_pages() << (PAGE_SHIFT-10),
770                 num_physpages << (PAGE_SHIFT-10),
771                 codesize >> 10,
772                 reservedpages << (PAGE_SHIFT-10),
773                 datasize >> 10,
774                 initsize >> 10,
775                 totalhigh_pages << (PAGE_SHIFT-10));
776
777         printk(KERN_INFO "virtual kernel memory layout:\n"
778                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
779 #ifdef CONFIG_HIGHMEM
780                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
781 #endif
782                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
783                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
784                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
785                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
786                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
787                 FIXADDR_START, FIXADDR_TOP,
788                 (FIXADDR_TOP - FIXADDR_START) >> 10,
789
790 #ifdef CONFIG_HIGHMEM
791                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
792                 (LAST_PKMAP*PAGE_SIZE) >> 10,
793 #endif
794
795                 VMALLOC_START, VMALLOC_END,
796                 (VMALLOC_END - VMALLOC_START) >> 20,
797
798                 (unsigned long)__va(0), (unsigned long)high_memory,
799                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
800
801                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
802                 ((unsigned long)&__init_end -
803                  (unsigned long)&__init_begin) >> 10,
804
805                 (unsigned long)&_etext, (unsigned long)&_edata,
806                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
807
808                 (unsigned long)&_text, (unsigned long)&_etext,
809                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
810
811         /*
812          * Check boundaries twice: Some fundamental inconsistencies can
813          * be detected at build time already.
814          */
815 #define __FIXADDR_TOP (-PAGE_SIZE)
816 #ifdef CONFIG_HIGHMEM
817         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
818         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
819 #endif
820 #define high_memory (-128UL << 20)
821         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
822 #undef high_memory
823 #undef __FIXADDR_TOP
824
825 #ifdef CONFIG_HIGHMEM
826         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
827         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
828 #endif
829         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
830         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
831
832         if (boot_cpu_data.wp_works_ok < 0)
833                 test_wp_bit();
834 }
835
836 #ifdef CONFIG_MEMORY_HOTPLUG
837 int arch_add_memory(int nid, u64 start, u64 size)
838 {
839         struct pglist_data *pgdata = NODE_DATA(nid);
840         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
841         unsigned long start_pfn = start >> PAGE_SHIFT;
842         unsigned long nr_pages = size >> PAGE_SHIFT;
843
844         return __add_pages(nid, zone, start_pfn, nr_pages);
845 }
846 #endif
847
848 /*
849  * This function cannot be __init, since exceptions don't work in that
850  * section.  Put this after the callers, so that it cannot be inlined.
851  */
852 static noinline int do_test_wp_bit(void)
853 {
854         char tmp_reg;
855         int flag;
856
857         __asm__ __volatile__(
858                 "       movb %0, %1     \n"
859                 "1:     movb %1, %0     \n"
860                 "       xorl %2, %2     \n"
861                 "2:                     \n"
862                 _ASM_EXTABLE(1b,2b)
863                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
864                  "=q" (tmp_reg),
865                  "=r" (flag)
866                 :"2" (1)
867                 :"memory");
868
869         return flag;
870 }
871
872 #ifdef CONFIG_DEBUG_RODATA
873 const int rodata_test_data = 0xC3;
874 EXPORT_SYMBOL_GPL(rodata_test_data);
875
876 int kernel_set_to_readonly __read_mostly;
877
878 void set_kernel_text_rw(void)
879 {
880         unsigned long start = PFN_ALIGN(_text);
881         unsigned long size = PFN_ALIGN(_etext) - start;
882
883         if (!kernel_set_to_readonly)
884                 return;
885
886         pr_debug("Set kernel text: %lx - %lx for read write\n",
887                  start, start+size);
888
889         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
890 }
891
892 void set_kernel_text_ro(void)
893 {
894         unsigned long start = PFN_ALIGN(_text);
895         unsigned long size = PFN_ALIGN(_etext) - start;
896
897         if (!kernel_set_to_readonly)
898                 return;
899
900         pr_debug("Set kernel text: %lx - %lx for read only\n",
901                  start, start+size);
902
903         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
904 }
905
906 static void mark_nxdata_nx(void)
907 {
908         /*
909          * When this called, init has already been executed and released,
910          * so everything past _etext should be NX.
911          */
912         unsigned long start = PFN_ALIGN(_etext);
913         /*
914          * This comes from is_kernel_text upper limit. Also HPAGE where used:
915          */
916         unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
917
918         if (__supported_pte_mask & _PAGE_NX)
919                 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
920         set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
921 }
922
923 void mark_rodata_ro(void)
924 {
925         unsigned long start = PFN_ALIGN(_text);
926         unsigned long size = PFN_ALIGN(_etext) - start;
927
928         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
929         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
930                 size >> 10);
931
932         kernel_set_to_readonly = 1;
933
934 #ifdef CONFIG_CPA_DEBUG
935         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
936                 start, start+size);
937         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
938
939         printk(KERN_INFO "Testing CPA: write protecting again\n");
940         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
941 #endif
942
943         start += size;
944         size = (unsigned long)__end_rodata - start;
945         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
946         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
947                 size >> 10);
948         rodata_test();
949
950 #ifdef CONFIG_CPA_DEBUG
951         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
952         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
953
954         printk(KERN_INFO "Testing CPA: write protecting again\n");
955         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
956 #endif
957         mark_nxdata_nx();
958 }
959 #endif
960