Merge branch 'linus' into x86/pat2
[linux-2.6.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/pfn.h>
25 #include <linux/poison.h>
26 #include <linux/bootmem.h>
27 #include <linux/slab.h>
28 #include <linux/proc_fs.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/initrd.h>
31 #include <linux/cpumask.h>
32
33 #include <asm/asm.h>
34 #include <asm/processor.h>
35 #include <asm/system.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
38 #include <asm/dma.h>
39 #include <asm/fixmap.h>
40 #include <asm/e820.h>
41 #include <asm/apic.h>
42 #include <asm/bugs.h>
43 #include <asm/tlb.h>
44 #include <asm/tlbflush.h>
45 #include <asm/pgalloc.h>
46 #include <asm/sections.h>
47 #include <asm/paravirt.h>
48 #include <asm/setup.h>
49 #include <asm/cacheflush.h>
50 #include <asm/smp.h>
51
52 unsigned int __VMALLOC_RESERVE = 128 << 20;
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
63 static unsigned long __initdata table_start;
64 static unsigned long __meminitdata table_end;
65 static unsigned long __meminitdata table_top;
66
67 static int __initdata after_init_bootmem;
68
69 static __init void *alloc_low_page(unsigned long *phys)
70 {
71         unsigned long pfn = table_end++;
72         void *adr;
73
74         if (pfn >= table_top)
75                 panic("alloc_low_page: ran out of memory");
76
77         adr = __va(pfn * PAGE_SIZE);
78         memset(adr, 0, PAGE_SIZE);
79         *phys  = pfn * PAGE_SIZE;
80         return adr;
81 }
82
83 /*
84  * Creates a middle page table and puts a pointer to it in the
85  * given global directory entry. This only returns the gd entry
86  * in non-PAE compilation mode, since the middle layer is folded.
87  */
88 static pmd_t * __init one_md_table_init(pgd_t *pgd)
89 {
90         pud_t *pud;
91         pmd_t *pmd_table;
92
93 #ifdef CONFIG_X86_PAE
94         unsigned long phys;
95         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
96                 if (after_init_bootmem)
97                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
98                 else
99                         pmd_table = (pmd_t *)alloc_low_page(&phys);
100                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
101                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
102                 pud = pud_offset(pgd, 0);
103                 BUG_ON(pmd_table != pmd_offset(pud, 0));
104         }
105 #endif
106         pud = pud_offset(pgd, 0);
107         pmd_table = pmd_offset(pud, 0);
108
109         return pmd_table;
110 }
111
112 /*
113  * Create a page table and place a pointer to it in a middle page
114  * directory entry:
115  */
116 static pte_t * __init one_page_table_init(pmd_t *pmd)
117 {
118         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
119                 pte_t *page_table = NULL;
120
121                 if (after_init_bootmem) {
122 #ifdef CONFIG_DEBUG_PAGEALLOC
123                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
124 #endif
125                         if (!page_table)
126                                 page_table =
127                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
128                 } else {
129                         unsigned long phys;
130                         page_table = (pte_t *)alloc_low_page(&phys);
131                 }
132
133                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
134                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
135                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
136         }
137
138         return pte_offset_kernel(pmd, 0);
139 }
140
141 /*
142  * This function initializes a certain range of kernel virtual memory
143  * with new bootmem page tables, everywhere page tables are missing in
144  * the given range.
145  *
146  * NOTE: The pagetables are allocated contiguous on the physical space
147  * so we can cache the place of the first one and move around without
148  * checking the pgd every time.
149  */
150 static void __init
151 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
152 {
153         int pgd_idx, pmd_idx;
154         unsigned long vaddr;
155         pgd_t *pgd;
156         pmd_t *pmd;
157
158         vaddr = start;
159         pgd_idx = pgd_index(vaddr);
160         pmd_idx = pmd_index(vaddr);
161         pgd = pgd_base + pgd_idx;
162
163         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
164                 pmd = one_md_table_init(pgd);
165                 pmd = pmd + pmd_index(vaddr);
166                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
167                                                         pmd++, pmd_idx++) {
168                         one_page_table_init(pmd);
169
170                         vaddr += PMD_SIZE;
171                 }
172                 pmd_idx = 0;
173         }
174 }
175
176 static inline int is_kernel_text(unsigned long addr)
177 {
178         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
179                 return 1;
180         return 0;
181 }
182
183 /*
184  * This maps the physical memory to kernel virtual address space, a total
185  * of max_low_pfn pages, by creating page tables starting from address
186  * PAGE_OFFSET:
187  */
188 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
189                                                 unsigned long start_pfn,
190                                                 unsigned long end_pfn,
191                                                 int use_pse)
192 {
193         int pgd_idx, pmd_idx, pte_ofs;
194         unsigned long pfn;
195         pgd_t *pgd;
196         pmd_t *pmd;
197         pte_t *pte;
198         unsigned pages_2m, pages_4k;
199         int mapping_iter;
200
201         /*
202          * First iteration will setup identity mapping using large/small pages
203          * based on use_pse, with other attributes same as set by
204          * the early code in head_32.S
205          *
206          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
207          * as desired for the kernel identity mapping.
208          *
209          * This two pass mechanism conforms to the TLB app note which says:
210          *
211          *     "Software should not write to a paging-structure entry in a way
212          *      that would change, for any linear address, both the page size
213          *      and either the page frame or attributes."
214          */
215         mapping_iter = 1;
216
217         if (!cpu_has_pse)
218                 use_pse = 0;
219
220 repeat:
221         pages_2m = pages_4k = 0;
222         pfn = start_pfn;
223         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
224         pgd = pgd_base + pgd_idx;
225         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
226                 pmd = one_md_table_init(pgd);
227
228                 if (pfn >= end_pfn)
229                         continue;
230 #ifdef CONFIG_X86_PAE
231                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
232                 pmd += pmd_idx;
233 #else
234                 pmd_idx = 0;
235 #endif
236                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
237                      pmd++, pmd_idx++) {
238                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
239
240                         /*
241                          * Map with big pages if possible, otherwise
242                          * create normal page tables:
243                          */
244                         if (use_pse) {
245                                 unsigned int addr2;
246                                 pgprot_t prot = PAGE_KERNEL_LARGE;
247                                 /*
248                                  * first pass will use the same initial
249                                  * identity mapping attribute + _PAGE_PSE.
250                                  */
251                                 pgprot_t init_prot =
252                                         __pgprot(PTE_IDENT_ATTR |
253                                                  _PAGE_PSE);
254
255                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
256                                         PAGE_OFFSET + PAGE_SIZE-1;
257
258                                 if (is_kernel_text(addr) ||
259                                     is_kernel_text(addr2))
260                                         prot = PAGE_KERNEL_LARGE_EXEC;
261
262                                 pages_2m++;
263                                 if (mapping_iter == 1)
264                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
265                                 else
266                                         set_pmd(pmd, pfn_pmd(pfn, prot));
267
268                                 pfn += PTRS_PER_PTE;
269                                 continue;
270                         }
271                         pte = one_page_table_init(pmd);
272
273                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
274                         pte += pte_ofs;
275                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
276                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
277                                 pgprot_t prot = PAGE_KERNEL;
278                                 /*
279                                  * first pass will use the same initial
280                                  * identity mapping attribute.
281                                  */
282                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
283
284                                 if (is_kernel_text(addr))
285                                         prot = PAGE_KERNEL_EXEC;
286
287                                 pages_4k++;
288                                 if (mapping_iter == 1)
289                                         set_pte(pte, pfn_pte(pfn, init_prot));
290                                 else
291                                         set_pte(pte, pfn_pte(pfn, prot));
292                         }
293                 }
294         }
295         if (mapping_iter == 1) {
296                 /*
297                  * update direct mapping page count only in the first
298                  * iteration.
299                  */
300                 update_page_count(PG_LEVEL_2M, pages_2m);
301                 update_page_count(PG_LEVEL_4K, pages_4k);
302
303                 /*
304                  * local global flush tlb, which will flush the previous
305                  * mappings present in both small and large page TLB's.
306                  */
307                 __flush_tlb_all();
308
309                 /*
310                  * Second iteration will set the actual desired PTE attributes.
311                  */
312                 mapping_iter = 2;
313                 goto repeat;
314         }
315 }
316
317 /*
318  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
319  * is valid. The argument is a physical page number.
320  *
321  *
322  * On x86, access has to be given to the first megabyte of ram because that area
323  * contains bios code and data regions used by X and dosemu and similar apps.
324  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
325  * mmio resources as well as potential bios/acpi data regions.
326  */
327 int devmem_is_allowed(unsigned long pagenr)
328 {
329         if (pagenr <= 256)
330                 return 1;
331         if (!page_is_ram(pagenr))
332                 return 1;
333         return 0;
334 }
335
336 #ifdef CONFIG_HIGHMEM
337 pte_t *kmap_pte;
338 pgprot_t kmap_prot;
339
340 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
341 {
342         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
343                         vaddr), vaddr), vaddr);
344 }
345
346 static void __init kmap_init(void)
347 {
348         unsigned long kmap_vstart;
349
350         /*
351          * Cache the first kmap pte:
352          */
353         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
354         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
355
356         kmap_prot = PAGE_KERNEL;
357 }
358
359 static void __init permanent_kmaps_init(pgd_t *pgd_base)
360 {
361         unsigned long vaddr;
362         pgd_t *pgd;
363         pud_t *pud;
364         pmd_t *pmd;
365         pte_t *pte;
366
367         vaddr = PKMAP_BASE;
368         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
369
370         pgd = swapper_pg_dir + pgd_index(vaddr);
371         pud = pud_offset(pgd, vaddr);
372         pmd = pmd_offset(pud, vaddr);
373         pte = pte_offset_kernel(pmd, vaddr);
374         pkmap_page_table = pte;
375 }
376
377 static void __init add_one_highpage_init(struct page *page, int pfn)
378 {
379         ClearPageReserved(page);
380         init_page_count(page);
381         __free_page(page);
382         totalhigh_pages++;
383 }
384
385 struct add_highpages_data {
386         unsigned long start_pfn;
387         unsigned long end_pfn;
388 };
389
390 static int __init add_highpages_work_fn(unsigned long start_pfn,
391                                          unsigned long end_pfn, void *datax)
392 {
393         int node_pfn;
394         struct page *page;
395         unsigned long final_start_pfn, final_end_pfn;
396         struct add_highpages_data *data;
397
398         data = (struct add_highpages_data *)datax;
399
400         final_start_pfn = max(start_pfn, data->start_pfn);
401         final_end_pfn = min(end_pfn, data->end_pfn);
402         if (final_start_pfn >= final_end_pfn)
403                 return 0;
404
405         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
406              node_pfn++) {
407                 if (!pfn_valid(node_pfn))
408                         continue;
409                 page = pfn_to_page(node_pfn);
410                 add_one_highpage_init(page, node_pfn);
411         }
412
413         return 0;
414
415 }
416
417 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
418                                               unsigned long end_pfn)
419 {
420         struct add_highpages_data data;
421
422         data.start_pfn = start_pfn;
423         data.end_pfn = end_pfn;
424
425         work_with_active_regions(nid, add_highpages_work_fn, &data);
426 }
427
428 #ifndef CONFIG_NUMA
429 static void __init set_highmem_pages_init(void)
430 {
431         add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
432
433         totalram_pages += totalhigh_pages;
434 }
435 #endif /* !CONFIG_NUMA */
436
437 #else
438 # define kmap_init()                            do { } while (0)
439 # define permanent_kmaps_init(pgd_base)         do { } while (0)
440 # define set_highmem_pages_init()       do { } while (0)
441 #endif /* CONFIG_HIGHMEM */
442
443 void __init native_pagetable_setup_start(pgd_t *base)
444 {
445         unsigned long pfn, va;
446         pgd_t *pgd;
447         pud_t *pud;
448         pmd_t *pmd;
449         pte_t *pte;
450
451         /*
452          * Remove any mappings which extend past the end of physical
453          * memory from the boot time page table:
454          */
455         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
456                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
457                 pgd = base + pgd_index(va);
458                 if (!pgd_present(*pgd))
459                         break;
460
461                 pud = pud_offset(pgd, va);
462                 pmd = pmd_offset(pud, va);
463                 if (!pmd_present(*pmd))
464                         break;
465
466                 pte = pte_offset_kernel(pmd, va);
467                 if (!pte_present(*pte))
468                         break;
469
470                 pte_clear(NULL, va, pte);
471         }
472         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
473 }
474
475 void __init native_pagetable_setup_done(pgd_t *base)
476 {
477 }
478
479 /*
480  * Build a proper pagetable for the kernel mappings.  Up until this
481  * point, we've been running on some set of pagetables constructed by
482  * the boot process.
483  *
484  * If we're booting on native hardware, this will be a pagetable
485  * constructed in arch/x86/kernel/head_32.S.  The root of the
486  * pagetable will be swapper_pg_dir.
487  *
488  * If we're booting paravirtualized under a hypervisor, then there are
489  * more options: we may already be running PAE, and the pagetable may
490  * or may not be based in swapper_pg_dir.  In any case,
491  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
492  * appropriately for the rest of the initialization to work.
493  *
494  * In general, pagetable_init() assumes that the pagetable may already
495  * be partially populated, and so it avoids stomping on any existing
496  * mappings.
497  */
498 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
499 {
500         unsigned long vaddr, end;
501
502         /*
503          * Fixed mappings, only the page table structure has to be
504          * created - mappings will be set by set_fixmap():
505          */
506         early_ioremap_clear();
507         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
508         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
509         page_table_range_init(vaddr, end, pgd_base);
510         early_ioremap_reset();
511 }
512
513 static void __init pagetable_init(void)
514 {
515         pgd_t *pgd_base = swapper_pg_dir;
516
517         permanent_kmaps_init(pgd_base);
518 }
519
520 #ifdef CONFIG_ACPI_SLEEP
521 /*
522  * ACPI suspend needs this for resume, because things like the intel-agp
523  * driver might have split up a kernel 4MB mapping.
524  */
525 char swsusp_pg_dir[PAGE_SIZE]
526         __attribute__ ((aligned(PAGE_SIZE)));
527
528 static inline void save_pg_dir(void)
529 {
530         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
531 }
532 #else /* !CONFIG_ACPI_SLEEP */
533 static inline void save_pg_dir(void)
534 {
535 }
536 #endif /* !CONFIG_ACPI_SLEEP */
537
538 void zap_low_mappings(void)
539 {
540         int i;
541
542         /*
543          * Zap initial low-memory mappings.
544          *
545          * Note that "pgd_clear()" doesn't do it for
546          * us, because pgd_clear() is a no-op on i386.
547          */
548         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
549 #ifdef CONFIG_X86_PAE
550                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
551 #else
552                 set_pgd(swapper_pg_dir+i, __pgd(0));
553 #endif
554         }
555         flush_tlb_all();
556 }
557
558 int nx_enabled;
559
560 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
561 EXPORT_SYMBOL_GPL(__supported_pte_mask);
562
563 #ifdef CONFIG_X86_PAE
564
565 static int disable_nx __initdata;
566
567 /*
568  * noexec = on|off
569  *
570  * Control non executable mappings.
571  *
572  * on      Enable
573  * off     Disable
574  */
575 static int __init noexec_setup(char *str)
576 {
577         if (!str || !strcmp(str, "on")) {
578                 if (cpu_has_nx) {
579                         __supported_pte_mask |= _PAGE_NX;
580                         disable_nx = 0;
581                 }
582         } else {
583                 if (!strcmp(str, "off")) {
584                         disable_nx = 1;
585                         __supported_pte_mask &= ~_PAGE_NX;
586                 } else {
587                         return -EINVAL;
588                 }
589         }
590
591         return 0;
592 }
593 early_param("noexec", noexec_setup);
594
595 static void __init set_nx(void)
596 {
597         unsigned int v[4], l, h;
598
599         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
600                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
601
602                 if ((v[3] & (1 << 20)) && !disable_nx) {
603                         rdmsr(MSR_EFER, l, h);
604                         l |= EFER_NX;
605                         wrmsr(MSR_EFER, l, h);
606                         nx_enabled = 1;
607                         __supported_pte_mask |= _PAGE_NX;
608                 }
609         }
610 }
611 #endif
612
613 /* user-defined highmem size */
614 static unsigned int highmem_pages = -1;
615
616 /*
617  * highmem=size forces highmem to be exactly 'size' bytes.
618  * This works even on boxes that have no highmem otherwise.
619  * This also works to reduce highmem size on bigger boxes.
620  */
621 static int __init parse_highmem(char *arg)
622 {
623         if (!arg)
624                 return -EINVAL;
625
626         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
627         return 0;
628 }
629 early_param("highmem", parse_highmem);
630
631 /*
632  * Determine low and high memory ranges:
633  */
634 void __init find_low_pfn_range(void)
635 {
636         /* it could update max_pfn */
637
638         /* max_low_pfn is 0, we already have early_res support */
639
640         max_low_pfn = max_pfn;
641         if (max_low_pfn > MAXMEM_PFN) {
642                 if (highmem_pages == -1)
643                         highmem_pages = max_pfn - MAXMEM_PFN;
644                 if (highmem_pages + MAXMEM_PFN < max_pfn)
645                         max_pfn = MAXMEM_PFN + highmem_pages;
646                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
647                         printk(KERN_WARNING "only %luMB highmem pages "
648                                 "available, ignoring highmem size of %uMB.\n",
649                                 pages_to_mb(max_pfn - MAXMEM_PFN),
650                                 pages_to_mb(highmem_pages));
651                         highmem_pages = 0;
652                 }
653                 max_low_pfn = MAXMEM_PFN;
654 #ifndef CONFIG_HIGHMEM
655                 /* Maximum memory usable is what is directly addressable */
656                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
657                                         MAXMEM>>20);
658                 if (max_pfn > MAX_NONPAE_PFN)
659                         printk(KERN_WARNING
660                                  "Use a HIGHMEM64G enabled kernel.\n");
661                 else
662                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
663                 max_pfn = MAXMEM_PFN;
664 #else /* !CONFIG_HIGHMEM */
665 #ifndef CONFIG_HIGHMEM64G
666                 if (max_pfn > MAX_NONPAE_PFN) {
667                         max_pfn = MAX_NONPAE_PFN;
668                         printk(KERN_WARNING "Warning only 4GB will be used."
669                                 "Use a HIGHMEM64G enabled kernel.\n");
670                 }
671 #endif /* !CONFIG_HIGHMEM64G */
672 #endif /* !CONFIG_HIGHMEM */
673         } else {
674                 if (highmem_pages == -1)
675                         highmem_pages = 0;
676 #ifdef CONFIG_HIGHMEM
677                 if (highmem_pages >= max_pfn) {
678                         printk(KERN_ERR "highmem size specified (%uMB) is "
679                                 "bigger than pages available (%luMB)!.\n",
680                                 pages_to_mb(highmem_pages),
681                                 pages_to_mb(max_pfn));
682                         highmem_pages = 0;
683                 }
684                 if (highmem_pages) {
685                         if (max_low_pfn - highmem_pages <
686                             64*1024*1024/PAGE_SIZE){
687                                 printk(KERN_ERR "highmem size %uMB results in "
688                                 "smaller than 64MB lowmem, ignoring it.\n"
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"
697                                         " kernel!\n");
698 #endif
699         }
700 }
701
702 #ifndef CONFIG_NEED_MULTIPLE_NODES
703 void __init initmem_init(unsigned long start_pfn,
704                                   unsigned long end_pfn)
705 {
706 #ifdef CONFIG_HIGHMEM
707         highstart_pfn = highend_pfn = max_pfn;
708         if (max_pfn > max_low_pfn)
709                 highstart_pfn = max_low_pfn;
710         memory_present(0, 0, highend_pfn);
711         e820_register_active_regions(0, 0, highend_pfn);
712         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
713                 pages_to_mb(highend_pfn - highstart_pfn));
714         num_physpages = highend_pfn;
715         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
716 #else
717         memory_present(0, 0, max_low_pfn);
718         e820_register_active_regions(0, 0, max_low_pfn);
719         num_physpages = max_low_pfn;
720         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
721 #endif
722 #ifdef CONFIG_FLATMEM
723         max_mapnr = num_physpages;
724 #endif
725         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
726                         pages_to_mb(max_low_pfn));
727
728         setup_bootmem_allocator();
729 }
730 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
731
732 static void __init zone_sizes_init(void)
733 {
734         unsigned long max_zone_pfns[MAX_NR_ZONES];
735         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
736         max_zone_pfns[ZONE_DMA] =
737                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
738         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
739 #ifdef CONFIG_HIGHMEM
740         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
741 #endif
742
743         free_area_init_nodes(max_zone_pfns);
744 }
745
746 void __init setup_bootmem_allocator(void)
747 {
748         int i;
749         unsigned long bootmap_size, bootmap;
750         /*
751          * Initialize the boot-time allocator (with low memory only):
752          */
753         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
754         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
755                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
756                                  PAGE_SIZE);
757         if (bootmap == -1L)
758                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
759         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
760
761         /* don't touch min_low_pfn */
762         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
763                                          min_low_pfn, max_low_pfn);
764         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
765                  max_pfn_mapped<<PAGE_SHIFT);
766         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
767                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
768         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
769                  bootmap, bootmap + bootmap_size);
770         for_each_online_node(i)
771                 free_bootmem_with_active_regions(i, max_low_pfn);
772         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
773
774         after_init_bootmem = 1;
775 }
776
777 static void __init find_early_table_space(unsigned long end, int use_pse)
778 {
779         unsigned long puds, pmds, ptes, tables, start;
780
781         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
782         tables = PAGE_ALIGN(puds * sizeof(pud_t));
783
784         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
785         tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
786
787         if (use_pse) {
788                 unsigned long extra;
789
790                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
791                 extra += PMD_SIZE;
792                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
793         } else
794                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
795
796         tables += PAGE_ALIGN(ptes * sizeof(pte_t));
797
798         /* for fixmap */
799         tables += PAGE_SIZE * 2;
800
801         /*
802          * RED-PEN putting page tables only on node 0 could
803          * cause a hotspot and fill up ZONE_DMA. The page tables
804          * need roughly 0.5KB per GB.
805          */
806         start = 0x7000;
807         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
808                                         tables, PAGE_SIZE);
809         if (table_start == -1UL)
810                 panic("Cannot find space for the kernel page tables");
811
812         table_start >>= PAGE_SHIFT;
813         table_end = table_start;
814         table_top = table_start + (tables>>PAGE_SHIFT);
815
816         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
817                 end, table_start << PAGE_SHIFT,
818                 (table_start << PAGE_SHIFT) + tables);
819 }
820
821 unsigned long __init_refok init_memory_mapping(unsigned long start,
822                                                 unsigned long end)
823 {
824         pgd_t *pgd_base = swapper_pg_dir;
825         unsigned long start_pfn, end_pfn;
826         unsigned long big_page_start;
827 #ifdef CONFIG_DEBUG_PAGEALLOC
828         /*
829          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
830          * This will simplify cpa(), which otherwise needs to support splitting
831          * large pages into small in interrupt context, etc.
832          */
833         int use_pse = 0;
834 #else
835         int use_pse = cpu_has_pse;
836 #endif
837
838         /*
839          * Find space for the kernel direct mapping tables.
840          */
841         if (!after_init_bootmem)
842                 find_early_table_space(end, use_pse);
843
844 #ifdef CONFIG_X86_PAE
845         set_nx();
846         if (nx_enabled)
847                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
848 #endif
849
850         /* Enable PSE if available */
851         if (cpu_has_pse)
852                 set_in_cr4(X86_CR4_PSE);
853
854         /* Enable PGE if available */
855         if (cpu_has_pge) {
856                 set_in_cr4(X86_CR4_PGE);
857                 __supported_pte_mask |= _PAGE_GLOBAL;
858         }
859
860         /*
861          * Don't use a large page for the first 2/4MB of memory
862          * because there are often fixed size MTRRs in there
863          * and overlapping MTRRs into large pages can cause
864          * slowdowns.
865          */
866         big_page_start = PMD_SIZE;
867
868         if (start < big_page_start) {
869                 start_pfn = start >> PAGE_SHIFT;
870                 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
871         } else {
872                 /* head is not big page alignment ? */
873                 start_pfn = start >> PAGE_SHIFT;
874                 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
875                                  << (PMD_SHIFT - PAGE_SHIFT);
876         }
877         if (start_pfn < end_pfn)
878                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
879
880         /* big page range */
881         start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
882                          << (PMD_SHIFT - PAGE_SHIFT);
883         if (start_pfn < (big_page_start >> PAGE_SHIFT))
884                 start_pfn =  big_page_start >> PAGE_SHIFT;
885         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
886         if (start_pfn < end_pfn)
887                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
888                                              use_pse);
889
890         /* tail is not big page alignment ? */
891         start_pfn = end_pfn;
892         if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
893                 end_pfn = end >> PAGE_SHIFT;
894                 if (start_pfn < end_pfn)
895                         kernel_physical_mapping_init(pgd_base, start_pfn,
896                                                          end_pfn, 0);
897         }
898
899         early_ioremap_page_table_range_init(pgd_base);
900
901         load_cr3(swapper_pg_dir);
902
903         __flush_tlb_all();
904
905         if (!after_init_bootmem)
906                 reserve_early(table_start << PAGE_SHIFT,
907                                  table_end << PAGE_SHIFT, "PGTABLE");
908
909         if (!after_init_bootmem)
910                 early_memtest(start, end);
911
912         return end >> PAGE_SHIFT;
913 }
914
915
916 /*
917  * paging_init() sets up the page tables - note that the first 8MB are
918  * already mapped by head.S.
919  *
920  * This routines also unmaps the page at virtual kernel address 0, so
921  * that we can trap those pesky NULL-reference errors in the kernel.
922  */
923 void __init paging_init(void)
924 {
925         pagetable_init();
926
927         __flush_tlb_all();
928
929         kmap_init();
930
931         /*
932          * NOTE: at this point the bootmem allocator is fully available.
933          */
934         sparse_init();
935         zone_sizes_init();
936 }
937
938 /*
939  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
940  * and also on some strange 486's. All 586+'s are OK. This used to involve
941  * black magic jumps to work around some nasty CPU bugs, but fortunately the
942  * switch to using exceptions got rid of all that.
943  */
944 static void __init test_wp_bit(void)
945 {
946         printk(KERN_INFO
947   "Checking if this processor honours the WP bit even in supervisor mode...");
948
949         /* Any page-aligned address will do, the test is non-destructive */
950         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
951         boot_cpu_data.wp_works_ok = do_test_wp_bit();
952         clear_fixmap(FIX_WP_TEST);
953
954         if (!boot_cpu_data.wp_works_ok) {
955                 printk(KERN_CONT "No.\n");
956 #ifdef CONFIG_X86_WP_WORKS_OK
957                 panic(
958   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
959 #endif
960         } else {
961                 printk(KERN_CONT "Ok.\n");
962         }
963 }
964
965 static struct kcore_list kcore_mem, kcore_vmalloc;
966
967 void __init mem_init(void)
968 {
969         int codesize, reservedpages, datasize, initsize;
970         int tmp;
971
972 #ifdef CONFIG_FLATMEM
973         BUG_ON(!mem_map);
974 #endif
975         /* this will put all low memory onto the freelists */
976         totalram_pages += free_all_bootmem();
977
978         reservedpages = 0;
979         for (tmp = 0; tmp < max_low_pfn; tmp++)
980                 /*
981                  * Only count reserved RAM pages:
982                  */
983                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
984                         reservedpages++;
985
986         set_highmem_pages_init();
987
988         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
989         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
990         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
991
992         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
993         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
994                    VMALLOC_END-VMALLOC_START);
995
996         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
997                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
998                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
999                 num_physpages << (PAGE_SHIFT-10),
1000                 codesize >> 10,
1001                 reservedpages << (PAGE_SHIFT-10),
1002                 datasize >> 10,
1003                 initsize >> 10,
1004                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1005                );
1006
1007         printk(KERN_INFO "virtual kernel memory layout:\n"
1008                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1009 #ifdef CONFIG_HIGHMEM
1010                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1011 #endif
1012                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1013                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1014                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1015                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1016                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
1017                 FIXADDR_START, FIXADDR_TOP,
1018                 (FIXADDR_TOP - FIXADDR_START) >> 10,
1019
1020 #ifdef CONFIG_HIGHMEM
1021                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1022                 (LAST_PKMAP*PAGE_SIZE) >> 10,
1023 #endif
1024
1025                 VMALLOC_START, VMALLOC_END,
1026                 (VMALLOC_END - VMALLOC_START) >> 20,
1027
1028                 (unsigned long)__va(0), (unsigned long)high_memory,
1029                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1030
1031                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1032                 ((unsigned long)&__init_end -
1033                  (unsigned long)&__init_begin) >> 10,
1034
1035                 (unsigned long)&_etext, (unsigned long)&_edata,
1036                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1037
1038                 (unsigned long)&_text, (unsigned long)&_etext,
1039                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1040
1041 #ifdef CONFIG_HIGHMEM
1042         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1043         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1044 #endif
1045         BUG_ON(VMALLOC_START                            > VMALLOC_END);
1046         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1047
1048         if (boot_cpu_data.wp_works_ok < 0)
1049                 test_wp_bit();
1050
1051         save_pg_dir();
1052         zap_low_mappings();
1053 }
1054
1055 #ifdef CONFIG_MEMORY_HOTPLUG
1056 int arch_add_memory(int nid, u64 start, u64 size)
1057 {
1058         struct pglist_data *pgdata = NODE_DATA(nid);
1059         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1060         unsigned long start_pfn = start >> PAGE_SHIFT;
1061         unsigned long nr_pages = size >> PAGE_SHIFT;
1062
1063         return __add_pages(zone, start_pfn, nr_pages);
1064 }
1065 #endif
1066
1067 /*
1068  * This function cannot be __init, since exceptions don't work in that
1069  * section.  Put this after the callers, so that it cannot be inlined.
1070  */
1071 static noinline int do_test_wp_bit(void)
1072 {
1073         char tmp_reg;
1074         int flag;
1075
1076         __asm__ __volatile__(
1077                 "       movb %0, %1     \n"
1078                 "1:     movb %1, %0     \n"
1079                 "       xorl %2, %2     \n"
1080                 "2:                     \n"
1081                 _ASM_EXTABLE(1b,2b)
1082                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1083                  "=q" (tmp_reg),
1084                  "=r" (flag)
1085                 :"2" (1)
1086                 :"memory");
1087
1088         return flag;
1089 }
1090
1091 #ifdef CONFIG_DEBUG_RODATA
1092 const int rodata_test_data = 0xC3;
1093 EXPORT_SYMBOL_GPL(rodata_test_data);
1094
1095 void mark_rodata_ro(void)
1096 {
1097         unsigned long start = PFN_ALIGN(_text);
1098         unsigned long size = PFN_ALIGN(_etext) - start;
1099
1100 #ifndef CONFIG_DYNAMIC_FTRACE
1101         /* Dynamic tracing modifies the kernel text section */
1102         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1103         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1104                 size >> 10);
1105
1106 #ifdef CONFIG_CPA_DEBUG
1107         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1108                 start, start+size);
1109         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1110
1111         printk(KERN_INFO "Testing CPA: write protecting again\n");
1112         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1113 #endif
1114 #endif /* CONFIG_DYNAMIC_FTRACE */
1115
1116         start += size;
1117         size = (unsigned long)__end_rodata - start;
1118         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1119         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1120                 size >> 10);
1121         rodata_test();
1122
1123 #ifdef CONFIG_CPA_DEBUG
1124         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1125         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1126
1127         printk(KERN_INFO "Testing CPA: write protecting again\n");
1128         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1129 #endif
1130 }
1131 #endif
1132
1133 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1134 {
1135 #ifdef CONFIG_DEBUG_PAGEALLOC
1136         /*
1137          * If debugging page accesses then do not free this memory but
1138          * mark them not present - any buggy init-section access will
1139          * create a kernel page fault:
1140          */
1141         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1142                 begin, PAGE_ALIGN(end));
1143         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1144 #else
1145         unsigned long addr;
1146
1147         /*
1148          * We just marked the kernel text read only above, now that
1149          * we are going to free part of that, we need to make that
1150          * writeable first.
1151          */
1152         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1153
1154         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1155                 ClearPageReserved(virt_to_page(addr));
1156                 init_page_count(virt_to_page(addr));
1157                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1158                 free_page(addr);
1159                 totalram_pages++;
1160         }
1161         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1162 #endif
1163 }
1164
1165 void free_initmem(void)
1166 {
1167         free_init_pages("unused kernel memory",
1168                         (unsigned long)(&__init_begin),
1169                         (unsigned long)(&__init_end));
1170 }
1171
1172 #ifdef CONFIG_BLK_DEV_INITRD
1173 void free_initrd_mem(unsigned long start, unsigned long end)
1174 {
1175         free_init_pages("initrd memory", start, end);
1176 }
1177 #endif
1178
1179 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1180                                    int flags)
1181 {
1182         return reserve_bootmem(phys, len, flags);
1183 }