x86: unify kernel_physical_mapping_init() function signatures
[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
53 unsigned long max_low_pfn_mapped;
54 unsigned long max_pfn_mapped;
55
56 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
57 unsigned long highstart_pfn, highend_pfn;
58
59 static noinline int do_test_wp_bit(void);
60
61
62 extern unsigned long __initdata e820_table_start;
63 extern unsigned long __meminitdata e820_table_end;
64 extern unsigned long __meminitdata e820_table_top;
65
66 static __init void *alloc_low_page(void)
67 {
68         unsigned long pfn = e820_table_end++;
69         void *adr;
70
71         if (pfn >= e820_table_top)
72                 panic("alloc_low_page: ran out of memory");
73
74         adr = __va(pfn * PAGE_SIZE);
75         memset(adr, 0, PAGE_SIZE);
76         return adr;
77 }
78
79 /*
80  * Creates a middle page table and puts a pointer to it in the
81  * given global directory entry. This only returns the gd entry
82  * in non-PAE compilation mode, since the middle layer is folded.
83  */
84 static pmd_t * __init one_md_table_init(pgd_t *pgd)
85 {
86         pud_t *pud;
87         pmd_t *pmd_table;
88
89 #ifdef CONFIG_X86_PAE
90         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
91                 if (after_bootmem)
92                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
93                 else
94                         pmd_table = (pmd_t *)alloc_low_page();
95                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
96                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
97                 pud = pud_offset(pgd, 0);
98                 BUG_ON(pmd_table != pmd_offset(pud, 0));
99
100                 return pmd_table;
101         }
102 #endif
103         pud = pud_offset(pgd, 0);
104         pmd_table = pmd_offset(pud, 0);
105
106         return pmd_table;
107 }
108
109 /*
110  * Create a page table and place a pointer to it in a middle page
111  * directory entry:
112  */
113 static pte_t * __init one_page_table_init(pmd_t *pmd)
114 {
115         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
116                 pte_t *page_table = NULL;
117
118                 if (after_bootmem) {
119 #ifdef CONFIG_DEBUG_PAGEALLOC
120                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
121 #endif
122                         if (!page_table)
123                                 page_table =
124                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
125                 } else
126                         page_table = (pte_t *)alloc_low_page();
127
128                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
129                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
130                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
131         }
132
133         return pte_offset_kernel(pmd, 0);
134 }
135
136 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
137                                            unsigned long vaddr, pte_t *lastpte)
138 {
139 #ifdef CONFIG_HIGHMEM
140         /*
141          * Something (early fixmap) may already have put a pte
142          * page here, which causes the page table allocation
143          * to become nonlinear. Attempt to fix it, and if it
144          * is still nonlinear then we have to bug.
145          */
146         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
147         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
148
149         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
150             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
151             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
152             && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
153                 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
154                 pte_t *newpte;
155                 int i;
156
157                 BUG_ON(after_bootmem);
158                 newpte = alloc_low_page();
159                 for (i = 0; i < PTRS_PER_PTE; i++)
160                         set_pte(newpte + i, pte[i]);
161
162                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
163                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
164                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
165                 __flush_tlb_all();
166
167                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
168                 pte = newpte;
169         }
170         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
171                && vaddr > fix_to_virt(FIX_KMAP_END)
172                && lastpte && lastpte + PTRS_PER_PTE != pte);
173 #endif
174         return pte;
175 }
176
177 /*
178  * This function initializes a certain range of kernel virtual memory
179  * with new bootmem page tables, everywhere page tables are missing in
180  * the given range.
181  *
182  * NOTE: The pagetables are allocated contiguous on the physical space
183  * so we can cache the place of the first one and move around without
184  * checking the pgd every time.
185  */
186 static void __init
187 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
188 {
189         int pgd_idx, pmd_idx;
190         unsigned long vaddr;
191         pgd_t *pgd;
192         pmd_t *pmd;
193         pte_t *pte = NULL;
194
195         vaddr = start;
196         pgd_idx = pgd_index(vaddr);
197         pmd_idx = pmd_index(vaddr);
198         pgd = pgd_base + pgd_idx;
199
200         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
201                 pmd = one_md_table_init(pgd);
202                 pmd = pmd + pmd_index(vaddr);
203                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
204                                                         pmd++, pmd_idx++) {
205                         pte = page_table_kmap_check(one_page_table_init(pmd),
206                                                     pmd, vaddr, pte);
207
208                         vaddr += PMD_SIZE;
209                 }
210                 pmd_idx = 0;
211         }
212 }
213
214 static inline int is_kernel_text(unsigned long addr)
215 {
216         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
217                 return 1;
218         return 0;
219 }
220
221 /*
222  * This maps the physical memory to kernel virtual address space, a total
223  * of max_low_pfn pages, by creating page tables starting from address
224  * PAGE_OFFSET:
225  */
226 unsigned long __init
227 kernel_physical_mapping_init(unsigned long start,
228                              unsigned long end,
229                              unsigned long page_size_mask)
230 {
231         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
232         unsigned long start_pfn, end_pfn;
233         pgd_t *pgd_base = swapper_pg_dir;
234         int pgd_idx, pmd_idx, pte_ofs;
235         unsigned long pfn;
236         pgd_t *pgd;
237         pmd_t *pmd;
238         pte_t *pte;
239         unsigned pages_2m, pages_4k;
240         int mapping_iter;
241
242         start_pfn = start >> PAGE_SHIFT;
243         end_pfn = end >> PAGE_SHIFT;
244
245         /*
246          * First iteration will setup identity mapping using large/small pages
247          * based on use_pse, with other attributes same as set by
248          * the early code in head_32.S
249          *
250          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
251          * as desired for the kernel identity mapping.
252          *
253          * This two pass mechanism conforms to the TLB app note which says:
254          *
255          *     "Software should not write to a paging-structure entry in a way
256          *      that would change, for any linear address, both the page size
257          *      and either the page frame or attributes."
258          */
259         mapping_iter = 1;
260
261         if (!cpu_has_pse)
262                 use_pse = 0;
263
264 repeat:
265         pages_2m = pages_4k = 0;
266         pfn = start_pfn;
267         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
268         pgd = pgd_base + pgd_idx;
269         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
270                 pmd = one_md_table_init(pgd);
271
272                 if (pfn >= end_pfn)
273                         continue;
274 #ifdef CONFIG_X86_PAE
275                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
276                 pmd += pmd_idx;
277 #else
278                 pmd_idx = 0;
279 #endif
280                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
281                      pmd++, pmd_idx++) {
282                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
283
284                         /*
285                          * Map with big pages if possible, otherwise
286                          * create normal page tables:
287                          */
288                         if (use_pse) {
289                                 unsigned int addr2;
290                                 pgprot_t prot = PAGE_KERNEL_LARGE;
291                                 /*
292                                  * first pass will use the same initial
293                                  * identity mapping attribute + _PAGE_PSE.
294                                  */
295                                 pgprot_t init_prot =
296                                         __pgprot(PTE_IDENT_ATTR |
297                                                  _PAGE_PSE);
298
299                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
300                                         PAGE_OFFSET + PAGE_SIZE-1;
301
302                                 if (is_kernel_text(addr) ||
303                                     is_kernel_text(addr2))
304                                         prot = PAGE_KERNEL_LARGE_EXEC;
305
306                                 pages_2m++;
307                                 if (mapping_iter == 1)
308                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
309                                 else
310                                         set_pmd(pmd, pfn_pmd(pfn, prot));
311
312                                 pfn += PTRS_PER_PTE;
313                                 continue;
314                         }
315                         pte = one_page_table_init(pmd);
316
317                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
318                         pte += pte_ofs;
319                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
320                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
321                                 pgprot_t prot = PAGE_KERNEL;
322                                 /*
323                                  * first pass will use the same initial
324                                  * identity mapping attribute.
325                                  */
326                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
327
328                                 if (is_kernel_text(addr))
329                                         prot = PAGE_KERNEL_EXEC;
330
331                                 pages_4k++;
332                                 if (mapping_iter == 1)
333                                         set_pte(pte, pfn_pte(pfn, init_prot));
334                                 else
335                                         set_pte(pte, pfn_pte(pfn, prot));
336                         }
337                 }
338         }
339         if (mapping_iter == 1) {
340                 /*
341                  * update direct mapping page count only in the first
342                  * iteration.
343                  */
344                 update_page_count(PG_LEVEL_2M, pages_2m);
345                 update_page_count(PG_LEVEL_4K, pages_4k);
346
347                 /*
348                  * local global flush tlb, which will flush the previous
349                  * mappings present in both small and large page TLB's.
350                  */
351                 __flush_tlb_all();
352
353                 /*
354                  * Second iteration will set the actual desired PTE attributes.
355                  */
356                 mapping_iter = 2;
357                 goto repeat;
358         }
359         return 0;
360 }
361
362 pte_t *kmap_pte;
363 pgprot_t kmap_prot;
364
365 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
366 {
367         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
368                         vaddr), vaddr), vaddr);
369 }
370
371 static void __init kmap_init(void)
372 {
373         unsigned long kmap_vstart;
374
375         /*
376          * Cache the first kmap pte:
377          */
378         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
379         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
380
381         kmap_prot = PAGE_KERNEL;
382 }
383
384 #ifdef CONFIG_HIGHMEM
385 static void __init permanent_kmaps_init(pgd_t *pgd_base)
386 {
387         unsigned long vaddr;
388         pgd_t *pgd;
389         pud_t *pud;
390         pmd_t *pmd;
391         pte_t *pte;
392
393         vaddr = PKMAP_BASE;
394         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
395
396         pgd = swapper_pg_dir + pgd_index(vaddr);
397         pud = pud_offset(pgd, vaddr);
398         pmd = pmd_offset(pud, vaddr);
399         pte = pte_offset_kernel(pmd, vaddr);
400         pkmap_page_table = pte;
401 }
402
403 static void __init add_one_highpage_init(struct page *page, int pfn)
404 {
405         ClearPageReserved(page);
406         init_page_count(page);
407         __free_page(page);
408         totalhigh_pages++;
409 }
410
411 struct add_highpages_data {
412         unsigned long start_pfn;
413         unsigned long end_pfn;
414 };
415
416 static int __init add_highpages_work_fn(unsigned long start_pfn,
417                                          unsigned long end_pfn, void *datax)
418 {
419         int node_pfn;
420         struct page *page;
421         unsigned long final_start_pfn, final_end_pfn;
422         struct add_highpages_data *data;
423
424         data = (struct add_highpages_data *)datax;
425
426         final_start_pfn = max(start_pfn, data->start_pfn);
427         final_end_pfn = min(end_pfn, data->end_pfn);
428         if (final_start_pfn >= final_end_pfn)
429                 return 0;
430
431         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
432              node_pfn++) {
433                 if (!pfn_valid(node_pfn))
434                         continue;
435                 page = pfn_to_page(node_pfn);
436                 add_one_highpage_init(page, node_pfn);
437         }
438
439         return 0;
440
441 }
442
443 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
444                                               unsigned long end_pfn)
445 {
446         struct add_highpages_data data;
447
448         data.start_pfn = start_pfn;
449         data.end_pfn = end_pfn;
450
451         work_with_active_regions(nid, add_highpages_work_fn, &data);
452 }
453
454 #else
455 static inline void permanent_kmaps_init(pgd_t *pgd_base)
456 {
457 }
458 #endif /* CONFIG_HIGHMEM */
459
460 void __init native_pagetable_setup_start(pgd_t *base)
461 {
462         unsigned long pfn, va;
463         pgd_t *pgd;
464         pud_t *pud;
465         pmd_t *pmd;
466         pte_t *pte;
467
468         /*
469          * Remove any mappings which extend past the end of physical
470          * memory from the boot time page table:
471          */
472         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
473                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
474                 pgd = base + pgd_index(va);
475                 if (!pgd_present(*pgd))
476                         break;
477
478                 pud = pud_offset(pgd, va);
479                 pmd = pmd_offset(pud, va);
480                 if (!pmd_present(*pmd))
481                         break;
482
483                 pte = pte_offset_kernel(pmd, va);
484                 if (!pte_present(*pte))
485                         break;
486
487                 pte_clear(NULL, va, pte);
488         }
489         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
490 }
491
492 void __init native_pagetable_setup_done(pgd_t *base)
493 {
494 }
495
496 /*
497  * Build a proper pagetable for the kernel mappings.  Up until this
498  * point, we've been running on some set of pagetables constructed by
499  * the boot process.
500  *
501  * If we're booting on native hardware, this will be a pagetable
502  * constructed in arch/x86/kernel/head_32.S.  The root of the
503  * pagetable will be swapper_pg_dir.
504  *
505  * If we're booting paravirtualized under a hypervisor, then there are
506  * more options: we may already be running PAE, and the pagetable may
507  * or may not be based in swapper_pg_dir.  In any case,
508  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
509  * appropriately for the rest of the initialization to work.
510  *
511  * In general, pagetable_init() assumes that the pagetable may already
512  * be partially populated, and so it avoids stomping on any existing
513  * mappings.
514  */
515 void __init early_ioremap_page_table_range_init(void)
516 {
517         pgd_t *pgd_base = swapper_pg_dir;
518         unsigned long vaddr, end;
519
520         /*
521          * Fixed mappings, only the page table structure has to be
522          * created - mappings will be set by set_fixmap():
523          */
524         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
525         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
526         page_table_range_init(vaddr, end, pgd_base);
527         early_ioremap_reset();
528 }
529
530 static void __init pagetable_init(void)
531 {
532         pgd_t *pgd_base = swapper_pg_dir;
533
534         permanent_kmaps_init(pgd_base);
535 }
536
537 #ifdef CONFIG_ACPI_SLEEP
538 /*
539  * ACPI suspend needs this for resume, because things like the intel-agp
540  * driver might have split up a kernel 4MB mapping.
541  */
542 char swsusp_pg_dir[PAGE_SIZE]
543         __attribute__ ((aligned(PAGE_SIZE)));
544
545 static inline void save_pg_dir(void)
546 {
547         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
548 }
549 #else /* !CONFIG_ACPI_SLEEP */
550 static inline void save_pg_dir(void)
551 {
552 }
553 #endif /* !CONFIG_ACPI_SLEEP */
554
555 void zap_low_mappings(void)
556 {
557         int i;
558
559         /*
560          * Zap initial low-memory mappings.
561          *
562          * Note that "pgd_clear()" doesn't do it for
563          * us, because pgd_clear() is a no-op on i386.
564          */
565         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
566 #ifdef CONFIG_X86_PAE
567                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
568 #else
569                 set_pgd(swapper_pg_dir+i, __pgd(0));
570 #endif
571         }
572         flush_tlb_all();
573 }
574
575 int nx_enabled;
576
577 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
578 EXPORT_SYMBOL_GPL(__supported_pte_mask);
579
580 #ifdef CONFIG_X86_PAE
581
582 static int disable_nx __initdata;
583
584 /*
585  * noexec = on|off
586  *
587  * Control non executable mappings.
588  *
589  * on      Enable
590  * off     Disable
591  */
592 static int __init noexec_setup(char *str)
593 {
594         if (!str || !strcmp(str, "on")) {
595                 if (cpu_has_nx) {
596                         __supported_pte_mask |= _PAGE_NX;
597                         disable_nx = 0;
598                 }
599         } else {
600                 if (!strcmp(str, "off")) {
601                         disable_nx = 1;
602                         __supported_pte_mask &= ~_PAGE_NX;
603                 } else {
604                         return -EINVAL;
605                 }
606         }
607
608         return 0;
609 }
610 early_param("noexec", noexec_setup);
611
612 static void __init set_nx(void)
613 {
614         unsigned int v[4], l, h;
615
616         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
617                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
618
619                 if ((v[3] & (1 << 20)) && !disable_nx) {
620                         rdmsr(MSR_EFER, l, h);
621                         l |= EFER_NX;
622                         wrmsr(MSR_EFER, l, h);
623                         nx_enabled = 1;
624                         __supported_pte_mask |= _PAGE_NX;
625                 }
626         }
627 }
628 #endif
629
630 /* user-defined highmem size */
631 static unsigned int highmem_pages = -1;
632
633 /*
634  * highmem=size forces highmem to be exactly 'size' bytes.
635  * This works even on boxes that have no highmem otherwise.
636  * This also works to reduce highmem size on bigger boxes.
637  */
638 static int __init parse_highmem(char *arg)
639 {
640         if (!arg)
641                 return -EINVAL;
642
643         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
644         return 0;
645 }
646 early_param("highmem", parse_highmem);
647
648 #define MSG_HIGHMEM_TOO_BIG \
649         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
650
651 #define MSG_LOWMEM_TOO_SMALL \
652         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
653 /*
654  * All of RAM fits into lowmem - but if user wants highmem
655  * artificially via the highmem=x boot parameter then create
656  * it:
657  */
658 void __init lowmem_pfn_init(void)
659 {
660         /* max_low_pfn is 0, we already have early_res support */
661         max_low_pfn = max_pfn;
662
663         if (highmem_pages == -1)
664                 highmem_pages = 0;
665 #ifdef CONFIG_HIGHMEM
666         if (highmem_pages >= max_pfn) {
667                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
668                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
669                 highmem_pages = 0;
670         }
671         if (highmem_pages) {
672                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
673                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
674                                 pages_to_mb(highmem_pages));
675                         highmem_pages = 0;
676                 }
677                 max_low_pfn -= highmem_pages;
678         }
679 #else
680         if (highmem_pages)
681                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
682 #endif
683 }
684
685 #define MSG_HIGHMEM_TOO_SMALL \
686         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
687
688 #define MSG_HIGHMEM_TRIMMED \
689         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
690 /*
691  * We have more RAM than fits into lowmem - we try to put it into
692  * highmem, also taking the highmem=x boot parameter into account:
693  */
694 void __init highmem_pfn_init(void)
695 {
696         max_low_pfn = MAXMEM_PFN;
697
698         if (highmem_pages == -1)
699                 highmem_pages = max_pfn - MAXMEM_PFN;
700
701         if (highmem_pages + MAXMEM_PFN < max_pfn)
702                 max_pfn = MAXMEM_PFN + highmem_pages;
703
704         if (highmem_pages + MAXMEM_PFN > max_pfn) {
705                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
706                         pages_to_mb(max_pfn - MAXMEM_PFN),
707                         pages_to_mb(highmem_pages));
708                 highmem_pages = 0;
709         }
710 #ifndef CONFIG_HIGHMEM
711         /* Maximum memory usable is what is directly addressable */
712         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
713         if (max_pfn > MAX_NONPAE_PFN)
714                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
715         else
716                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
717         max_pfn = MAXMEM_PFN;
718 #else /* !CONFIG_HIGHMEM */
719 #ifndef CONFIG_HIGHMEM64G
720         if (max_pfn > MAX_NONPAE_PFN) {
721                 max_pfn = MAX_NONPAE_PFN;
722                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
723         }
724 #endif /* !CONFIG_HIGHMEM64G */
725 #endif /* !CONFIG_HIGHMEM */
726 }
727
728 /*
729  * Determine low and high memory ranges:
730  */
731 void __init find_low_pfn_range(void)
732 {
733         /* it could update max_pfn */
734
735         if (max_pfn <= MAXMEM_PFN)
736                 lowmem_pfn_init();
737         else
738                 highmem_pfn_init();
739 }
740
741 #ifndef CONFIG_NEED_MULTIPLE_NODES
742 void __init initmem_init(unsigned long start_pfn,
743                                   unsigned long end_pfn)
744 {
745 #ifdef CONFIG_HIGHMEM
746         highstart_pfn = highend_pfn = max_pfn;
747         if (max_pfn > max_low_pfn)
748                 highstart_pfn = max_low_pfn;
749         memory_present(0, 0, highend_pfn);
750         e820_register_active_regions(0, 0, highend_pfn);
751         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
752                 pages_to_mb(highend_pfn - highstart_pfn));
753         num_physpages = highend_pfn;
754         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
755 #else
756         memory_present(0, 0, max_low_pfn);
757         e820_register_active_regions(0, 0, max_low_pfn);
758         num_physpages = max_low_pfn;
759         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
760 #endif
761 #ifdef CONFIG_FLATMEM
762         max_mapnr = num_physpages;
763 #endif
764         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
765                         pages_to_mb(max_low_pfn));
766
767         setup_bootmem_allocator();
768 }
769 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
770
771 static void __init zone_sizes_init(void)
772 {
773         unsigned long max_zone_pfns[MAX_NR_ZONES];
774         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
775         max_zone_pfns[ZONE_DMA] =
776                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
777         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
778 #ifdef CONFIG_HIGHMEM
779         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
780 #endif
781
782         free_area_init_nodes(max_zone_pfns);
783 }
784
785 static unsigned long __init setup_node_bootmem(int nodeid,
786                                  unsigned long start_pfn,
787                                  unsigned long end_pfn,
788                                  unsigned long bootmap)
789 {
790         unsigned long bootmap_size;
791
792         if (start_pfn > max_low_pfn)
793                 return bootmap;
794         if (end_pfn > max_low_pfn)
795                 end_pfn = max_low_pfn;
796
797         /* don't touch min_low_pfn */
798         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
799                                          bootmap >> PAGE_SHIFT,
800                                          start_pfn, end_pfn);
801         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
802                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
803         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
804                  nodeid, bootmap, bootmap + bootmap_size);
805         free_bootmem_with_active_regions(nodeid, end_pfn);
806         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
807
808         return bootmap + bootmap_size;
809 }
810
811 void __init setup_bootmem_allocator(void)
812 {
813         int nodeid;
814         unsigned long bootmap_size, bootmap;
815         /*
816          * Initialize the boot-time allocator (with low memory only):
817          */
818         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
819         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
820                                  PAGE_SIZE);
821         if (bootmap == -1L)
822                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
823         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
824
825         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
826                  max_pfn_mapped<<PAGE_SHIFT);
827         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
828
829 #ifdef CONFIG_NEED_MULTIPLE_NODES
830         for_each_online_node(nodeid)
831                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
832                                         node_end_pfn[nodeid], bootmap);
833 #else
834         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
835 #endif
836
837         after_bootmem = 1;
838 }
839
840 /*
841  * paging_init() sets up the page tables - note that the first 8MB are
842  * already mapped by head.S.
843  *
844  * This routines also unmaps the page at virtual kernel address 0, so
845  * that we can trap those pesky NULL-reference errors in the kernel.
846  */
847 void __init paging_init(void)
848 {
849         pagetable_init();
850
851         __flush_tlb_all();
852
853         kmap_init();
854
855         /*
856          * NOTE: at this point the bootmem allocator is fully available.
857          */
858         sparse_init();
859         zone_sizes_init();
860 }
861
862 /*
863  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
864  * and also on some strange 486's. All 586+'s are OK. This used to involve
865  * black magic jumps to work around some nasty CPU bugs, but fortunately the
866  * switch to using exceptions got rid of all that.
867  */
868 static void __init test_wp_bit(void)
869 {
870         printk(KERN_INFO
871   "Checking if this processor honours the WP bit even in supervisor mode...");
872
873         /* Any page-aligned address will do, the test is non-destructive */
874         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
875         boot_cpu_data.wp_works_ok = do_test_wp_bit();
876         clear_fixmap(FIX_WP_TEST);
877
878         if (!boot_cpu_data.wp_works_ok) {
879                 printk(KERN_CONT "No.\n");
880 #ifdef CONFIG_X86_WP_WORKS_OK
881                 panic(
882   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
883 #endif
884         } else {
885                 printk(KERN_CONT "Ok.\n");
886         }
887 }
888
889 static struct kcore_list kcore_mem, kcore_vmalloc;
890
891 void __init mem_init(void)
892 {
893         int codesize, reservedpages, datasize, initsize;
894         int tmp;
895
896         pci_iommu_alloc();
897
898 #ifdef CONFIG_FLATMEM
899         BUG_ON(!mem_map);
900 #endif
901         /* this will put all low memory onto the freelists */
902         totalram_pages += free_all_bootmem();
903
904         reservedpages = 0;
905         for (tmp = 0; tmp < max_low_pfn; tmp++)
906                 /*
907                  * Only count reserved RAM pages:
908                  */
909                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
910                         reservedpages++;
911
912         set_highmem_pages_init();
913
914         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
915         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
916         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
917
918         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
919         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
920                    VMALLOC_END-VMALLOC_START);
921
922         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
923                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
924                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
925                 num_physpages << (PAGE_SHIFT-10),
926                 codesize >> 10,
927                 reservedpages << (PAGE_SHIFT-10),
928                 datasize >> 10,
929                 initsize >> 10,
930                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
931                );
932
933         printk(KERN_INFO "virtual kernel memory layout:\n"
934                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
935 #ifdef CONFIG_HIGHMEM
936                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
937 #endif
938                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
939                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
940                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
941                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
942                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
943                 FIXADDR_START, FIXADDR_TOP,
944                 (FIXADDR_TOP - FIXADDR_START) >> 10,
945
946 #ifdef CONFIG_HIGHMEM
947                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
948                 (LAST_PKMAP*PAGE_SIZE) >> 10,
949 #endif
950
951                 VMALLOC_START, VMALLOC_END,
952                 (VMALLOC_END - VMALLOC_START) >> 20,
953
954                 (unsigned long)__va(0), (unsigned long)high_memory,
955                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
956
957                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
958                 ((unsigned long)&__init_end -
959                  (unsigned long)&__init_begin) >> 10,
960
961                 (unsigned long)&_etext, (unsigned long)&_edata,
962                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
963
964                 (unsigned long)&_text, (unsigned long)&_etext,
965                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
966
967         /*
968          * Check boundaries twice: Some fundamental inconsistencies can
969          * be detected at build time already.
970          */
971 #define __FIXADDR_TOP (-PAGE_SIZE)
972 #ifdef CONFIG_HIGHMEM
973         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
974         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
975 #endif
976 #define high_memory (-128UL << 20)
977         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
978 #undef high_memory
979 #undef __FIXADDR_TOP
980
981 #ifdef CONFIG_HIGHMEM
982         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
983         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
984 #endif
985         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
986         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
987
988         if (boot_cpu_data.wp_works_ok < 0)
989                 test_wp_bit();
990
991         save_pg_dir();
992         zap_low_mappings();
993 }
994
995 #ifdef CONFIG_MEMORY_HOTPLUG
996 int arch_add_memory(int nid, u64 start, u64 size)
997 {
998         struct pglist_data *pgdata = NODE_DATA(nid);
999         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1000         unsigned long start_pfn = start >> PAGE_SHIFT;
1001         unsigned long nr_pages = size >> PAGE_SHIFT;
1002
1003         return __add_pages(nid, zone, start_pfn, nr_pages);
1004 }
1005 #endif
1006
1007 /*
1008  * This function cannot be __init, since exceptions don't work in that
1009  * section.  Put this after the callers, so that it cannot be inlined.
1010  */
1011 static noinline int do_test_wp_bit(void)
1012 {
1013         char tmp_reg;
1014         int flag;
1015
1016         __asm__ __volatile__(
1017                 "       movb %0, %1     \n"
1018                 "1:     movb %1, %0     \n"
1019                 "       xorl %2, %2     \n"
1020                 "2:                     \n"
1021                 _ASM_EXTABLE(1b,2b)
1022                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1023                  "=q" (tmp_reg),
1024                  "=r" (flag)
1025                 :"2" (1)
1026                 :"memory");
1027
1028         return flag;
1029 }
1030
1031 #ifdef CONFIG_DEBUG_RODATA
1032 const int rodata_test_data = 0xC3;
1033 EXPORT_SYMBOL_GPL(rodata_test_data);
1034
1035 void mark_rodata_ro(void)
1036 {
1037         unsigned long start = PFN_ALIGN(_text);
1038         unsigned long size = PFN_ALIGN(_etext) - start;
1039
1040 #ifndef CONFIG_DYNAMIC_FTRACE
1041         /* Dynamic tracing modifies the kernel text section */
1042         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1043         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1044                 size >> 10);
1045
1046 #ifdef CONFIG_CPA_DEBUG
1047         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1048                 start, start+size);
1049         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1050
1051         printk(KERN_INFO "Testing CPA: write protecting again\n");
1052         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1053 #endif
1054 #endif /* CONFIG_DYNAMIC_FTRACE */
1055
1056         start += size;
1057         size = (unsigned long)__end_rodata - start;
1058         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1059         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1060                 size >> 10);
1061         rodata_test();
1062
1063 #ifdef CONFIG_CPA_DEBUG
1064         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1065         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1066
1067         printk(KERN_INFO "Testing CPA: write protecting again\n");
1068         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1069 #endif
1070 }
1071 #endif
1072
1073 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1074                                    int flags)
1075 {
1076         return reserve_bootmem(phys, len, flags);
1077 }