Merge branch 'x86/doc' into x86/core
[linux-2.6] / 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/slab.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33
34 #include <asm/asm.h>
35 #include <asm/bios_ebda.h>
36 #include <asm/processor.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
40 #include <asm/dma.h>
41 #include <asm/fixmap.h>
42 #include <asm/e820.h>
43 #include <asm/apic.h>
44 #include <asm/bugs.h>
45 #include <asm/tlb.h>
46 #include <asm/tlbflush.h>
47 #include <asm/pgalloc.h>
48 #include <asm/sections.h>
49 #include <asm/paravirt.h>
50 #include <asm/setup.h>
51 #include <asm/cacheflush.h>
52 #include <asm/init.h>
53
54 unsigned long max_low_pfn_mapped;
55 unsigned long max_pfn_mapped;
56
57 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
58 unsigned long highstart_pfn, highend_pfn;
59
60 static noinline int do_test_wp_bit(void);
61
62 bool __read_mostly __vmalloc_start_set = false;
63
64 static __init void *alloc_low_page(void)
65 {
66         unsigned long pfn = e820_table_end++;
67         void *adr;
68
69         if (pfn >= e820_table_top)
70                 panic("alloc_low_page: ran out of memory");
71
72         adr = __va(pfn * PAGE_SIZE);
73         memset(adr, 0, PAGE_SIZE);
74         return adr;
75 }
76
77 /*
78  * Creates a middle page table and puts a pointer to it in the
79  * given global directory entry. This only returns the gd entry
80  * in non-PAE compilation mode, since the middle layer is folded.
81  */
82 static pmd_t * __init one_md_table_init(pgd_t *pgd)
83 {
84         pud_t *pud;
85         pmd_t *pmd_table;
86
87 #ifdef CONFIG_X86_PAE
88         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
89                 if (after_bootmem)
90                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
91                 else
92                         pmd_table = (pmd_t *)alloc_low_page();
93                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
94                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
95                 pud = pud_offset(pgd, 0);
96                 BUG_ON(pmd_table != pmd_offset(pud, 0));
97
98                 return pmd_table;
99         }
100 #endif
101         pud = pud_offset(pgd, 0);
102         pmd_table = pmd_offset(pud, 0);
103
104         return pmd_table;
105 }
106
107 /*
108  * Create a page table and place a pointer to it in a middle page
109  * directory entry:
110  */
111 static pte_t * __init one_page_table_init(pmd_t *pmd)
112 {
113         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
114                 pte_t *page_table = NULL;
115
116                 if (after_bootmem) {
117 #ifdef CONFIG_DEBUG_PAGEALLOC
118                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
119 #endif
120                         if (!page_table)
121                                 page_table =
122                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
123                 } else
124                         page_table = (pte_t *)alloc_low_page();
125
126                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
127                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
128                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
129         }
130
131         return pte_offset_kernel(pmd, 0);
132 }
133
134 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
135 {
136         int pgd_idx = pgd_index(vaddr);
137         int pmd_idx = pmd_index(vaddr);
138
139         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
140 }
141
142 pte_t * __init populate_extra_pte(unsigned long vaddr)
143 {
144         int pte_idx = pte_index(vaddr);
145         pmd_t *pmd;
146
147         pmd = populate_extra_pmd(vaddr);
148         return one_page_table_init(pmd) + pte_idx;
149 }
150
151 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
152                                            unsigned long vaddr, pte_t *lastpte)
153 {
154 #ifdef CONFIG_HIGHMEM
155         /*
156          * Something (early fixmap) may already have put a pte
157          * page here, which causes the page table allocation
158          * to become nonlinear. Attempt to fix it, and if it
159          * is still nonlinear then we have to bug.
160          */
161         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
162         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
163
164         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
165             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
166             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
167             && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
168                 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
169                 pte_t *newpte;
170                 int i;
171
172                 BUG_ON(after_bootmem);
173                 newpte = alloc_low_page();
174                 for (i = 0; i < PTRS_PER_PTE; i++)
175                         set_pte(newpte + i, pte[i]);
176
177                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
178                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
179                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
180                 __flush_tlb_all();
181
182                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
183                 pte = newpte;
184         }
185         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
186                && vaddr > fix_to_virt(FIX_KMAP_END)
187                && lastpte && lastpte + PTRS_PER_PTE != pte);
188 #endif
189         return pte;
190 }
191
192 /*
193  * This function initializes a certain range of kernel virtual memory
194  * with new bootmem page tables, everywhere page tables are missing in
195  * the given range.
196  *
197  * NOTE: The pagetables are allocated contiguous on the physical space
198  * so we can cache the place of the first one and move around without
199  * checking the pgd every time.
200  */
201 static void __init
202 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
203 {
204         int pgd_idx, pmd_idx;
205         unsigned long vaddr;
206         pgd_t *pgd;
207         pmd_t *pmd;
208         pte_t *pte = NULL;
209
210         vaddr = start;
211         pgd_idx = pgd_index(vaddr);
212         pmd_idx = pmd_index(vaddr);
213         pgd = pgd_base + pgd_idx;
214
215         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
216                 pmd = one_md_table_init(pgd);
217                 pmd = pmd + pmd_index(vaddr);
218                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
219                                                         pmd++, pmd_idx++) {
220                         pte = page_table_kmap_check(one_page_table_init(pmd),
221                                                     pmd, vaddr, pte);
222
223                         vaddr += PMD_SIZE;
224                 }
225                 pmd_idx = 0;
226         }
227 }
228
229 static inline int is_kernel_text(unsigned long addr)
230 {
231         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
232                 return 1;
233         return 0;
234 }
235
236 /*
237  * This maps the physical memory to kernel virtual address space, a total
238  * of max_low_pfn pages, by creating page tables starting from address
239  * PAGE_OFFSET:
240  */
241 unsigned long __init
242 kernel_physical_mapping_init(unsigned long start,
243                              unsigned long end,
244                              unsigned long page_size_mask)
245 {
246         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
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                                 else
350                                         set_pte(pte, pfn_pte(pfn, prot));
351                         }
352                 }
353         }
354         if (mapping_iter == 1) {
355                 /*
356                  * update direct mapping page count only in the first
357                  * iteration.
358                  */
359                 update_page_count(PG_LEVEL_2M, pages_2m);
360                 update_page_count(PG_LEVEL_4K, pages_4k);
361
362                 /*
363                  * local global flush tlb, which will flush the previous
364                  * mappings present in both small and large page TLB's.
365                  */
366                 __flush_tlb_all();
367
368                 /*
369                  * Second iteration will set the actual desired PTE attributes.
370                  */
371                 mapping_iter = 2;
372                 goto repeat;
373         }
374         return 0;
375 }
376
377 pte_t *kmap_pte;
378 pgprot_t kmap_prot;
379
380 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
381 {
382         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
383                         vaddr), vaddr), vaddr);
384 }
385
386 static void __init kmap_init(void)
387 {
388         unsigned long kmap_vstart;
389
390         /*
391          * Cache the first kmap pte:
392          */
393         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
394         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
395
396         kmap_prot = PAGE_KERNEL;
397 }
398
399 #ifdef CONFIG_HIGHMEM
400 static void __init permanent_kmaps_init(pgd_t *pgd_base)
401 {
402         unsigned long vaddr;
403         pgd_t *pgd;
404         pud_t *pud;
405         pmd_t *pmd;
406         pte_t *pte;
407
408         vaddr = PKMAP_BASE;
409         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
410
411         pgd = swapper_pg_dir + pgd_index(vaddr);
412         pud = pud_offset(pgd, vaddr);
413         pmd = pmd_offset(pud, vaddr);
414         pte = pte_offset_kernel(pmd, vaddr);
415         pkmap_page_table = pte;
416 }
417
418 static void __init add_one_highpage_init(struct page *page, int pfn)
419 {
420         ClearPageReserved(page);
421         init_page_count(page);
422         __free_page(page);
423         totalhigh_pages++;
424 }
425
426 struct add_highpages_data {
427         unsigned long start_pfn;
428         unsigned long end_pfn;
429 };
430
431 static int __init add_highpages_work_fn(unsigned long start_pfn,
432                                          unsigned long end_pfn, void *datax)
433 {
434         int node_pfn;
435         struct page *page;
436         unsigned long final_start_pfn, final_end_pfn;
437         struct add_highpages_data *data;
438
439         data = (struct add_highpages_data *)datax;
440
441         final_start_pfn = max(start_pfn, data->start_pfn);
442         final_end_pfn = min(end_pfn, data->end_pfn);
443         if (final_start_pfn >= final_end_pfn)
444                 return 0;
445
446         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
447              node_pfn++) {
448                 if (!pfn_valid(node_pfn))
449                         continue;
450                 page = pfn_to_page(node_pfn);
451                 add_one_highpage_init(page, node_pfn);
452         }
453
454         return 0;
455
456 }
457
458 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
459                                               unsigned long end_pfn)
460 {
461         struct add_highpages_data data;
462
463         data.start_pfn = start_pfn;
464         data.end_pfn = end_pfn;
465
466         work_with_active_regions(nid, add_highpages_work_fn, &data);
467 }
468
469 #else
470 static inline void permanent_kmaps_init(pgd_t *pgd_base)
471 {
472 }
473 #endif /* CONFIG_HIGHMEM */
474
475 void __init native_pagetable_setup_start(pgd_t *base)
476 {
477         unsigned long pfn, va;
478         pgd_t *pgd;
479         pud_t *pud;
480         pmd_t *pmd;
481         pte_t *pte;
482
483         /*
484          * Remove any mappings which extend past the end of physical
485          * memory from the boot time page table:
486          */
487         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
488                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
489                 pgd = base + pgd_index(va);
490                 if (!pgd_present(*pgd))
491                         break;
492
493                 pud = pud_offset(pgd, va);
494                 pmd = pmd_offset(pud, va);
495                 if (!pmd_present(*pmd))
496                         break;
497
498                 pte = pte_offset_kernel(pmd, va);
499                 if (!pte_present(*pte))
500                         break;
501
502                 pte_clear(NULL, va, pte);
503         }
504         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
505 }
506
507 void __init native_pagetable_setup_done(pgd_t *base)
508 {
509 }
510
511 /*
512  * Build a proper pagetable for the kernel mappings.  Up until this
513  * point, we've been running on some set of pagetables constructed by
514  * the boot process.
515  *
516  * If we're booting on native hardware, this will be a pagetable
517  * constructed in arch/x86/kernel/head_32.S.  The root of the
518  * pagetable will be swapper_pg_dir.
519  *
520  * If we're booting paravirtualized under a hypervisor, then there are
521  * more options: we may already be running PAE, and the pagetable may
522  * or may not be based in swapper_pg_dir.  In any case,
523  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
524  * appropriately for the rest of the initialization to work.
525  *
526  * In general, pagetable_init() assumes that the pagetable may already
527  * be partially populated, and so it avoids stomping on any existing
528  * mappings.
529  */
530 void __init early_ioremap_page_table_range_init(void)
531 {
532         pgd_t *pgd_base = swapper_pg_dir;
533         unsigned long vaddr, end;
534
535         /*
536          * Fixed mappings, only the page table structure has to be
537          * created - mappings will be set by set_fixmap():
538          */
539         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
540         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
541         page_table_range_init(vaddr, end, pgd_base);
542         early_ioremap_reset();
543 }
544
545 static void __init pagetable_init(void)
546 {
547         pgd_t *pgd_base = swapper_pg_dir;
548
549         permanent_kmaps_init(pgd_base);
550 }
551
552 #ifdef CONFIG_ACPI_SLEEP
553 /*
554  * ACPI suspend needs this for resume, because things like the intel-agp
555  * driver might have split up a kernel 4MB mapping.
556  */
557 char swsusp_pg_dir[PAGE_SIZE]
558         __attribute__ ((aligned(PAGE_SIZE)));
559
560 static inline void save_pg_dir(void)
561 {
562         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
563 }
564 #else /* !CONFIG_ACPI_SLEEP */
565 static inline void save_pg_dir(void)
566 {
567 }
568 #endif /* !CONFIG_ACPI_SLEEP */
569
570 void zap_low_mappings(void)
571 {
572         int i;
573
574         /*
575          * Zap initial low-memory mappings.
576          *
577          * Note that "pgd_clear()" doesn't do it for
578          * us, because pgd_clear() is a no-op on i386.
579          */
580         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
581 #ifdef CONFIG_X86_PAE
582                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
583 #else
584                 set_pgd(swapper_pg_dir+i, __pgd(0));
585 #endif
586         }
587         flush_tlb_all();
588 }
589
590 int nx_enabled;
591
592 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
593 EXPORT_SYMBOL_GPL(__supported_pte_mask);
594
595 #ifdef CONFIG_X86_PAE
596
597 static int disable_nx __initdata;
598
599 /*
600  * noexec = on|off
601  *
602  * Control non executable mappings.
603  *
604  * on      Enable
605  * off     Disable
606  */
607 static int __init noexec_setup(char *str)
608 {
609         if (!str || !strcmp(str, "on")) {
610                 if (cpu_has_nx) {
611                         __supported_pte_mask |= _PAGE_NX;
612                         disable_nx = 0;
613                 }
614         } else {
615                 if (!strcmp(str, "off")) {
616                         disable_nx = 1;
617                         __supported_pte_mask &= ~_PAGE_NX;
618                 } else {
619                         return -EINVAL;
620                 }
621         }
622
623         return 0;
624 }
625 early_param("noexec", noexec_setup);
626
627 void __init set_nx(void)
628 {
629         unsigned int v[4], l, h;
630
631         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
632                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
633
634                 if ((v[3] & (1 << 20)) && !disable_nx) {
635                         rdmsr(MSR_EFER, l, h);
636                         l |= EFER_NX;
637                         wrmsr(MSR_EFER, l, h);
638                         nx_enabled = 1;
639                         __supported_pte_mask |= _PAGE_NX;
640                 }
641         }
642 }
643 #endif
644
645 /* user-defined highmem size */
646 static unsigned int highmem_pages = -1;
647
648 /*
649  * highmem=size forces highmem to be exactly 'size' bytes.
650  * This works even on boxes that have no highmem otherwise.
651  * This also works to reduce highmem size on bigger boxes.
652  */
653 static int __init parse_highmem(char *arg)
654 {
655         if (!arg)
656                 return -EINVAL;
657
658         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
659         return 0;
660 }
661 early_param("highmem", parse_highmem);
662
663 #define MSG_HIGHMEM_TOO_BIG \
664         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
665
666 #define MSG_LOWMEM_TOO_SMALL \
667         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
668 /*
669  * All of RAM fits into lowmem - but if user wants highmem
670  * artificially via the highmem=x boot parameter then create
671  * it:
672  */
673 void __init lowmem_pfn_init(void)
674 {
675         /* max_low_pfn is 0, we already have early_res support */
676         max_low_pfn = max_pfn;
677
678         if (highmem_pages == -1)
679                 highmem_pages = 0;
680 #ifdef CONFIG_HIGHMEM
681         if (highmem_pages >= max_pfn) {
682                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
683                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
684                 highmem_pages = 0;
685         }
686         if (highmem_pages) {
687                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
688                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
689                                 pages_to_mb(highmem_pages));
690                         highmem_pages = 0;
691                 }
692                 max_low_pfn -= highmem_pages;
693         }
694 #else
695         if (highmem_pages)
696                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
697 #endif
698 }
699
700 #define MSG_HIGHMEM_TOO_SMALL \
701         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
702
703 #define MSG_HIGHMEM_TRIMMED \
704         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
705 /*
706  * We have more RAM than fits into lowmem - we try to put it into
707  * highmem, also taking the highmem=x boot parameter into account:
708  */
709 void __init highmem_pfn_init(void)
710 {
711         max_low_pfn = MAXMEM_PFN;
712
713         if (highmem_pages == -1)
714                 highmem_pages = max_pfn - MAXMEM_PFN;
715
716         if (highmem_pages + MAXMEM_PFN < max_pfn)
717                 max_pfn = MAXMEM_PFN + highmem_pages;
718
719         if (highmem_pages + MAXMEM_PFN > max_pfn) {
720                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
721                         pages_to_mb(max_pfn - MAXMEM_PFN),
722                         pages_to_mb(highmem_pages));
723                 highmem_pages = 0;
724         }
725 #ifndef CONFIG_HIGHMEM
726         /* Maximum memory usable is what is directly addressable */
727         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
728         if (max_pfn > MAX_NONPAE_PFN)
729                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
730         else
731                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
732         max_pfn = MAXMEM_PFN;
733 #else /* !CONFIG_HIGHMEM */
734 #ifndef CONFIG_HIGHMEM64G
735         if (max_pfn > MAX_NONPAE_PFN) {
736                 max_pfn = MAX_NONPAE_PFN;
737                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
738         }
739 #endif /* !CONFIG_HIGHMEM64G */
740 #endif /* !CONFIG_HIGHMEM */
741 }
742
743 /*
744  * Determine low and high memory ranges:
745  */
746 void __init find_low_pfn_range(void)
747 {
748         /* it could update max_pfn */
749
750         if (max_pfn <= MAXMEM_PFN)
751                 lowmem_pfn_init();
752         else
753                 highmem_pfn_init();
754 }
755
756 #ifndef CONFIG_NEED_MULTIPLE_NODES
757 void __init initmem_init(unsigned long start_pfn,
758                                   unsigned long end_pfn)
759 {
760 #ifdef CONFIG_HIGHMEM
761         highstart_pfn = highend_pfn = max_pfn;
762         if (max_pfn > max_low_pfn)
763                 highstart_pfn = max_low_pfn;
764         memory_present(0, 0, highend_pfn);
765         e820_register_active_regions(0, 0, highend_pfn);
766         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
767                 pages_to_mb(highend_pfn - highstart_pfn));
768         num_physpages = highend_pfn;
769         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
770 #else
771         memory_present(0, 0, max_low_pfn);
772         e820_register_active_regions(0, 0, max_low_pfn);
773         num_physpages = max_low_pfn;
774         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
775 #endif
776 #ifdef CONFIG_FLATMEM
777         max_mapnr = num_physpages;
778 #endif
779         __vmalloc_start_set = true;
780
781         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
782                         pages_to_mb(max_low_pfn));
783
784         setup_bootmem_allocator();
785 }
786 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
787
788 static void __init zone_sizes_init(void)
789 {
790         unsigned long max_zone_pfns[MAX_NR_ZONES];
791         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
792         max_zone_pfns[ZONE_DMA] =
793                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
794         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
795 #ifdef CONFIG_HIGHMEM
796         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
797 #endif
798
799         free_area_init_nodes(max_zone_pfns);
800 }
801
802 static unsigned long __init setup_node_bootmem(int nodeid,
803                                  unsigned long start_pfn,
804                                  unsigned long end_pfn,
805                                  unsigned long bootmap)
806 {
807         unsigned long bootmap_size;
808
809         if (start_pfn > max_low_pfn)
810                 return bootmap;
811         if (end_pfn > max_low_pfn)
812                 end_pfn = max_low_pfn;
813
814         /* don't touch min_low_pfn */
815         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
816                                          bootmap >> PAGE_SHIFT,
817                                          start_pfn, end_pfn);
818         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
819                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
820         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
821                  nodeid, bootmap, bootmap + bootmap_size);
822         free_bootmem_with_active_regions(nodeid, end_pfn);
823         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
824
825         return bootmap + bootmap_size;
826 }
827
828 void __init setup_bootmem_allocator(void)
829 {
830         int nodeid;
831         unsigned long bootmap_size, bootmap;
832         /*
833          * Initialize the boot-time allocator (with low memory only):
834          */
835         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
836         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
837                                  PAGE_SIZE);
838         if (bootmap == -1L)
839                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
840         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
841
842         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
843                  max_pfn_mapped<<PAGE_SHIFT);
844         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
845
846 #ifdef CONFIG_NEED_MULTIPLE_NODES
847         for_each_online_node(nodeid)
848                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
849                                         node_end_pfn[nodeid], bootmap);
850 #else
851         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
852 #endif
853
854         after_bootmem = 1;
855 }
856
857 /*
858  * paging_init() sets up the page tables - note that the first 8MB are
859  * already mapped by head.S.
860  *
861  * This routines also unmaps the page at virtual kernel address 0, so
862  * that we can trap those pesky NULL-reference errors in the kernel.
863  */
864 void __init paging_init(void)
865 {
866         pagetable_init();
867
868         __flush_tlb_all();
869
870         kmap_init();
871
872         /*
873          * NOTE: at this point the bootmem allocator is fully available.
874          */
875         sparse_init();
876         zone_sizes_init();
877 }
878
879 /*
880  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
881  * and also on some strange 486's. All 586+'s are OK. This used to involve
882  * black magic jumps to work around some nasty CPU bugs, but fortunately the
883  * switch to using exceptions got rid of all that.
884  */
885 static void __init test_wp_bit(void)
886 {
887         printk(KERN_INFO
888   "Checking if this processor honours the WP bit even in supervisor mode...");
889
890         /* Any page-aligned address will do, the test is non-destructive */
891         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
892         boot_cpu_data.wp_works_ok = do_test_wp_bit();
893         clear_fixmap(FIX_WP_TEST);
894
895         if (!boot_cpu_data.wp_works_ok) {
896                 printk(KERN_CONT "No.\n");
897 #ifdef CONFIG_X86_WP_WORKS_OK
898                 panic(
899   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
900 #endif
901         } else {
902                 printk(KERN_CONT "Ok.\n");
903         }
904 }
905
906 static struct kcore_list kcore_mem, kcore_vmalloc;
907
908 void __init mem_init(void)
909 {
910         int codesize, reservedpages, datasize, initsize;
911         int tmp;
912
913         pci_iommu_alloc();
914
915 #ifdef CONFIG_FLATMEM
916         BUG_ON(!mem_map);
917 #endif
918         /* this will put all low memory onto the freelists */
919         totalram_pages += free_all_bootmem();
920
921         reservedpages = 0;
922         for (tmp = 0; tmp < max_low_pfn; tmp++)
923                 /*
924                  * Only count reserved RAM pages:
925                  */
926                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
927                         reservedpages++;
928
929         set_highmem_pages_init();
930
931         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
932         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
933         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
934
935         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
936         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
937                    VMALLOC_END-VMALLOC_START);
938
939         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
940                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
941                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
942                 num_physpages << (PAGE_SHIFT-10),
943                 codesize >> 10,
944                 reservedpages << (PAGE_SHIFT-10),
945                 datasize >> 10,
946                 initsize >> 10,
947                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
948                );
949
950         printk(KERN_INFO "virtual kernel memory layout:\n"
951                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
952 #ifdef CONFIG_HIGHMEM
953                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
954 #endif
955                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
956                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
957                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
958                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
959                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
960                 FIXADDR_START, FIXADDR_TOP,
961                 (FIXADDR_TOP - FIXADDR_START) >> 10,
962
963 #ifdef CONFIG_HIGHMEM
964                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
965                 (LAST_PKMAP*PAGE_SIZE) >> 10,
966 #endif
967
968                 VMALLOC_START, VMALLOC_END,
969                 (VMALLOC_END - VMALLOC_START) >> 20,
970
971                 (unsigned long)__va(0), (unsigned long)high_memory,
972                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
973
974                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
975                 ((unsigned long)&__init_end -
976                  (unsigned long)&__init_begin) >> 10,
977
978                 (unsigned long)&_etext, (unsigned long)&_edata,
979                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
980
981                 (unsigned long)&_text, (unsigned long)&_etext,
982                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
983
984         /*
985          * Check boundaries twice: Some fundamental inconsistencies can
986          * be detected at build time already.
987          */
988 #define __FIXADDR_TOP (-PAGE_SIZE)
989 #ifdef CONFIG_HIGHMEM
990         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
991         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
992 #endif
993 #define high_memory (-128UL << 20)
994         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
995 #undef high_memory
996 #undef __FIXADDR_TOP
997
998 #ifdef CONFIG_HIGHMEM
999         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1000         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1001 #endif
1002         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
1003         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1004
1005         if (boot_cpu_data.wp_works_ok < 0)
1006                 test_wp_bit();
1007
1008         save_pg_dir();
1009         zap_low_mappings();
1010 }
1011
1012 #ifdef CONFIG_MEMORY_HOTPLUG
1013 int arch_add_memory(int nid, u64 start, u64 size)
1014 {
1015         struct pglist_data *pgdata = NODE_DATA(nid);
1016         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1017         unsigned long start_pfn = start >> PAGE_SHIFT;
1018         unsigned long nr_pages = size >> PAGE_SHIFT;
1019
1020         return __add_pages(nid, zone, start_pfn, nr_pages);
1021 }
1022 #endif
1023
1024 /*
1025  * This function cannot be __init, since exceptions don't work in that
1026  * section.  Put this after the callers, so that it cannot be inlined.
1027  */
1028 static noinline int do_test_wp_bit(void)
1029 {
1030         char tmp_reg;
1031         int flag;
1032
1033         __asm__ __volatile__(
1034                 "       movb %0, %1     \n"
1035                 "1:     movb %1, %0     \n"
1036                 "       xorl %2, %2     \n"
1037                 "2:                     \n"
1038                 _ASM_EXTABLE(1b,2b)
1039                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1040                  "=q" (tmp_reg),
1041                  "=r" (flag)
1042                 :"2" (1)
1043                 :"memory");
1044
1045         return flag;
1046 }
1047
1048 #ifdef CONFIG_DEBUG_RODATA
1049 const int rodata_test_data = 0xC3;
1050 EXPORT_SYMBOL_GPL(rodata_test_data);
1051
1052 void mark_rodata_ro(void)
1053 {
1054         unsigned long start = PFN_ALIGN(_text);
1055         unsigned long size = PFN_ALIGN(_etext) - start;
1056
1057 #ifndef CONFIG_DYNAMIC_FTRACE
1058         /* Dynamic tracing modifies the kernel text section */
1059         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1060         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1061                 size >> 10);
1062
1063 #ifdef CONFIG_CPA_DEBUG
1064         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1065                 start, start+size);
1066         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1067
1068         printk(KERN_INFO "Testing CPA: write protecting again\n");
1069         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1070 #endif
1071 #endif /* CONFIG_DYNAMIC_FTRACE */
1072
1073         start += size;
1074         size = (unsigned long)__end_rodata - start;
1075         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1076         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1077                 size >> 10);
1078         rodata_test();
1079
1080 #ifdef CONFIG_CPA_DEBUG
1081         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1082         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1083
1084         printk(KERN_INFO "Testing CPA: write protecting again\n");
1085         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1086 #endif
1087 }
1088 #endif
1089
1090 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1091                                    int flags)
1092 {
1093         return reserve_bootmem(phys, len, flags);
1094 }