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/smp.h>
53
54 unsigned int __VMALLOC_RESERVE = 128 << 20;
55
56 unsigned long max_low_pfn_mapped;
57 unsigned long max_pfn_mapped;
58
59 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
60 unsigned long highstart_pfn, highend_pfn;
61
62 static noinline int do_test_wp_bit(void);
63
64
65 static unsigned long __initdata table_start;
66 static unsigned long __meminitdata table_end;
67 static unsigned long __meminitdata table_top;
68
69 static int __initdata after_init_bootmem;
70
alloc_low_page(void)71 static __init void *alloc_low_page(void)
72 {
73 unsigned long pfn = table_end++;
74 void *adr;
75
76 if (pfn >= table_top)
77 panic("alloc_low_page: ran out of memory");
78
79 adr = __va(pfn * PAGE_SIZE);
80 memset(adr, 0, PAGE_SIZE);
81 return adr;
82 }
83
84 /*
85 * Creates a middle page table and puts a pointer to it in the
86 * given global directory entry. This only returns the gd entry
87 * in non-PAE compilation mode, since the middle layer is folded.
88 */
one_md_table_init(pgd_t * pgd)89 static pmd_t * __init one_md_table_init(pgd_t *pgd)
90 {
91 pud_t *pud;
92 pmd_t *pmd_table;
93
94 #ifdef CONFIG_X86_PAE
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();
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 return pmd_table;
106 }
107 #endif
108 pud = pud_offset(pgd, 0);
109 pmd_table = pmd_offset(pud, 0);
110
111 return pmd_table;
112 }
113
114 /*
115 * Create a page table and place a pointer to it in a middle page
116 * directory entry:
117 */
one_page_table_init(pmd_t * pmd)118 static pte_t * __init one_page_table_init(pmd_t *pmd)
119 {
120 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
121 pte_t *page_table = NULL;
122
123 if (after_init_bootmem) {
124 #ifdef CONFIG_DEBUG_PAGEALLOC
125 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
126 #endif
127 if (!page_table)
128 page_table =
129 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
130 } else
131 page_table = (pte_t *)alloc_low_page();
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
page_table_kmap_check(pte_t * pte,pmd_t * pmd,unsigned long vaddr,pte_t * lastpte)141 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
142 unsigned long vaddr, pte_t *lastpte)
143 {
144 #ifdef CONFIG_HIGHMEM
145 /*
146 * Something (early fixmap) may already have put a pte
147 * page here, which causes the page table allocation
148 * to become nonlinear. Attempt to fix it, and if it
149 * is still nonlinear then we have to bug.
150 */
151 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
152 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
153
154 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
155 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
156 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
157 && ((__pa(pte) >> PAGE_SHIFT) < table_start
158 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
159 pte_t *newpte;
160 int i;
161
162 BUG_ON(after_init_bootmem);
163 newpte = alloc_low_page();
164 for (i = 0; i < PTRS_PER_PTE; i++)
165 set_pte(newpte + i, pte[i]);
166
167 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
168 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
169 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
170 __flush_tlb_all();
171
172 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
173 pte = newpte;
174 }
175 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
176 && vaddr > fix_to_virt(FIX_KMAP_END)
177 && lastpte && lastpte + PTRS_PER_PTE != pte);
178 #endif
179 return pte;
180 }
181
182 /*
183 * This function initializes a certain range of kernel virtual memory
184 * with new bootmem page tables, everywhere page tables are missing in
185 * the given range.
186 *
187 * NOTE: The pagetables are allocated contiguous on the physical space
188 * so we can cache the place of the first one and move around without
189 * checking the pgd every time.
190 */
191 static void __init
page_table_range_init(unsigned long start,unsigned long end,pgd_t * pgd_base)192 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
193 {
194 int pgd_idx, pmd_idx;
195 unsigned long vaddr;
196 pgd_t *pgd;
197 pmd_t *pmd;
198 pte_t *pte = NULL;
199
200 vaddr = start;
201 pgd_idx = pgd_index(vaddr);
202 pmd_idx = pmd_index(vaddr);
203 pgd = pgd_base + pgd_idx;
204
205 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
206 pmd = one_md_table_init(pgd);
207 pmd = pmd + pmd_index(vaddr);
208 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
209 pmd++, pmd_idx++) {
210 pte = page_table_kmap_check(one_page_table_init(pmd),
211 pmd, vaddr, pte);
212
213 vaddr += PMD_SIZE;
214 }
215 pmd_idx = 0;
216 }
217 }
218
is_kernel_text(unsigned long addr)219 static inline int is_kernel_text(unsigned long addr)
220 {
221 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
222 return 1;
223 return 0;
224 }
225
226 /*
227 * This maps the physical memory to kernel virtual address space, a total
228 * of max_low_pfn pages, by creating page tables starting from address
229 * PAGE_OFFSET:
230 */
kernel_physical_mapping_init(pgd_t * pgd_base,unsigned long start_pfn,unsigned long end_pfn,int use_pse)231 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
232 unsigned long start_pfn,
233 unsigned long end_pfn,
234 int use_pse)
235 {
236 int pgd_idx, pmd_idx, pte_ofs;
237 unsigned long pfn;
238 pgd_t *pgd;
239 pmd_t *pmd;
240 pte_t *pte;
241 unsigned pages_2m, pages_4k;
242 int mapping_iter;
243
244 /*
245 * First iteration will setup identity mapping using large/small pages
246 * based on use_pse, with other attributes same as set by
247 * the early code in head_32.S
248 *
249 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
250 * as desired for the kernel identity mapping.
251 *
252 * This two pass mechanism conforms to the TLB app note which says:
253 *
254 * "Software should not write to a paging-structure entry in a way
255 * that would change, for any linear address, both the page size
256 * and either the page frame or attributes."
257 */
258 mapping_iter = 1;
259
260 if (!cpu_has_pse)
261 use_pse = 0;
262
263 repeat:
264 pages_2m = pages_4k = 0;
265 pfn = start_pfn;
266 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
267 pgd = pgd_base + pgd_idx;
268 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
269 pmd = one_md_table_init(pgd);
270
271 if (pfn >= end_pfn)
272 continue;
273 #ifdef CONFIG_X86_PAE
274 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
275 pmd += pmd_idx;
276 #else
277 pmd_idx = 0;
278 #endif
279 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
280 pmd++, pmd_idx++) {
281 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
282
283 /*
284 * Map with big pages if possible, otherwise
285 * create normal page tables:
286 */
287 if (use_pse) {
288 unsigned int addr2;
289 pgprot_t prot = PAGE_KERNEL_LARGE;
290 /*
291 * first pass will use the same initial
292 * identity mapping attribute + _PAGE_PSE.
293 */
294 pgprot_t init_prot =
295 __pgprot(PTE_IDENT_ATTR |
296 _PAGE_PSE);
297
298 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
299 PAGE_OFFSET + PAGE_SIZE-1;
300
301 if (is_kernel_text(addr) ||
302 is_kernel_text(addr2))
303 prot = PAGE_KERNEL_LARGE_EXEC;
304
305 pages_2m++;
306 if (mapping_iter == 1)
307 set_pmd(pmd, pfn_pmd(pfn, init_prot));
308 else
309 set_pmd(pmd, pfn_pmd(pfn, prot));
310
311 pfn += PTRS_PER_PTE;
312 continue;
313 }
314 pte = one_page_table_init(pmd);
315
316 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
317 pte += pte_ofs;
318 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
319 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
320 pgprot_t prot = PAGE_KERNEL;
321 /*
322 * first pass will use the same initial
323 * identity mapping attribute.
324 */
325 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
326
327 if (is_kernel_text(addr))
328 prot = PAGE_KERNEL_EXEC;
329
330 pages_4k++;
331 if (mapping_iter == 1)
332 set_pte(pte, pfn_pte(pfn, init_prot));
333 else
334 set_pte(pte, pfn_pte(pfn, prot));
335 }
336 }
337 }
338 if (mapping_iter == 1) {
339 /*
340 * update direct mapping page count only in the first
341 * iteration.
342 */
343 update_page_count(PG_LEVEL_2M, pages_2m);
344 update_page_count(PG_LEVEL_4K, pages_4k);
345
346 /*
347 * local global flush tlb, which will flush the previous
348 * mappings present in both small and large page TLB's.
349 */
350 __flush_tlb_all();
351
352 /*
353 * Second iteration will set the actual desired PTE attributes.
354 */
355 mapping_iter = 2;
356 goto repeat;
357 }
358 }
359
360 /*
361 * devmem_is_allowed() checks to see if /dev/mem access to a certain address
362 * is valid. The argument is a physical page number.
363 *
364 *
365 * On x86, access has to be given to the first megabyte of ram because that area
366 * contains bios code and data regions used by X and dosemu and similar apps.
367 * Access has to be given to non-kernel-ram areas as well, these contain the PCI
368 * mmio resources as well as potential bios/acpi data regions.
369 */
devmem_is_allowed(unsigned long pagenr)370 int devmem_is_allowed(unsigned long pagenr)
371 {
372 if (pagenr <= 256)
373 return 1;
374 if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
375 return 0;
376 if (!page_is_ram(pagenr))
377 return 1;
378 return 0;
379 }
380
381 pte_t *kmap_pte;
382 pgprot_t kmap_prot;
383
kmap_get_fixmap_pte(unsigned long vaddr)384 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
385 {
386 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
387 vaddr), vaddr), vaddr);
388 }
389
kmap_init(void)390 static void __init kmap_init(void)
391 {
392 unsigned long kmap_vstart;
393
394 /*
395 * Cache the first kmap pte:
396 */
397 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
398 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
399
400 kmap_prot = PAGE_KERNEL;
401 }
402
403 #ifdef CONFIG_HIGHMEM
permanent_kmaps_init(pgd_t * pgd_base)404 static void __init permanent_kmaps_init(pgd_t *pgd_base)
405 {
406 unsigned long vaddr;
407 pgd_t *pgd;
408 pud_t *pud;
409 pmd_t *pmd;
410 pte_t *pte;
411
412 vaddr = PKMAP_BASE;
413 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
414
415 pgd = swapper_pg_dir + pgd_index(vaddr);
416 pud = pud_offset(pgd, vaddr);
417 pmd = pmd_offset(pud, vaddr);
418 pte = pte_offset_kernel(pmd, vaddr);
419 pkmap_page_table = pte;
420 }
421
add_one_highpage_init(struct page * page,int pfn)422 static void __init add_one_highpage_init(struct page *page, int pfn)
423 {
424 ClearPageReserved(page);
425 init_page_count(page);
426 __free_page(page);
427 totalhigh_pages++;
428 }
429
430 struct add_highpages_data {
431 unsigned long start_pfn;
432 unsigned long end_pfn;
433 };
434
add_highpages_work_fn(unsigned long start_pfn,unsigned long end_pfn,void * datax)435 static int __init add_highpages_work_fn(unsigned long start_pfn,
436 unsigned long end_pfn, void *datax)
437 {
438 int node_pfn;
439 struct page *page;
440 unsigned long final_start_pfn, final_end_pfn;
441 struct add_highpages_data *data;
442
443 data = (struct add_highpages_data *)datax;
444
445 final_start_pfn = max(start_pfn, data->start_pfn);
446 final_end_pfn = min(end_pfn, data->end_pfn);
447 if (final_start_pfn >= final_end_pfn)
448 return 0;
449
450 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
451 node_pfn++) {
452 if (!pfn_valid(node_pfn))
453 continue;
454 page = pfn_to_page(node_pfn);
455 add_one_highpage_init(page, node_pfn);
456 }
457
458 return 0;
459
460 }
461
add_highpages_with_active_regions(int nid,unsigned long start_pfn,unsigned long end_pfn)462 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
463 unsigned long end_pfn)
464 {
465 struct add_highpages_data data;
466
467 data.start_pfn = start_pfn;
468 data.end_pfn = end_pfn;
469
470 work_with_active_regions(nid, add_highpages_work_fn, &data);
471 }
472
473 #ifndef CONFIG_NUMA
set_highmem_pages_init(void)474 static void __init set_highmem_pages_init(void)
475 {
476 add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
477
478 totalram_pages += totalhigh_pages;
479 }
480 #endif /* !CONFIG_NUMA */
481
482 #else
permanent_kmaps_init(pgd_t * pgd_base)483 static inline void permanent_kmaps_init(pgd_t *pgd_base)
484 {
485 }
set_highmem_pages_init(void)486 static inline void set_highmem_pages_init(void)
487 {
488 }
489 #endif /* CONFIG_HIGHMEM */
490
native_pagetable_setup_start(pgd_t * base)491 void __init native_pagetable_setup_start(pgd_t *base)
492 {
493 unsigned long pfn, va;
494 pgd_t *pgd;
495 pud_t *pud;
496 pmd_t *pmd;
497 pte_t *pte;
498
499 /*
500 * Remove any mappings which extend past the end of physical
501 * memory from the boot time page table:
502 */
503 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
504 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
505 pgd = base + pgd_index(va);
506 if (!pgd_present(*pgd))
507 break;
508
509 pud = pud_offset(pgd, va);
510 pmd = pmd_offset(pud, va);
511 if (!pmd_present(*pmd))
512 break;
513
514 pte = pte_offset_kernel(pmd, va);
515 if (!pte_present(*pte))
516 break;
517
518 pte_clear(NULL, va, pte);
519 }
520 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
521 }
522
native_pagetable_setup_done(pgd_t * base)523 void __init native_pagetable_setup_done(pgd_t *base)
524 {
525 }
526
527 /*
528 * Build a proper pagetable for the kernel mappings. Up until this
529 * point, we've been running on some set of pagetables constructed by
530 * the boot process.
531 *
532 * If we're booting on native hardware, this will be a pagetable
533 * constructed in arch/x86/kernel/head_32.S. The root of the
534 * pagetable will be swapper_pg_dir.
535 *
536 * If we're booting paravirtualized under a hypervisor, then there are
537 * more options: we may already be running PAE, and the pagetable may
538 * or may not be based in swapper_pg_dir. In any case,
539 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
540 * appropriately for the rest of the initialization to work.
541 *
542 * In general, pagetable_init() assumes that the pagetable may already
543 * be partially populated, and so it avoids stomping on any existing
544 * mappings.
545 */
early_ioremap_page_table_range_init(pgd_t * pgd_base)546 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
547 {
548 unsigned long vaddr, end;
549
550 /*
551 * Fixed mappings, only the page table structure has to be
552 * created - mappings will be set by set_fixmap():
553 */
554 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
555 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
556 page_table_range_init(vaddr, end, pgd_base);
557 early_ioremap_reset();
558 }
559
pagetable_init(void)560 static void __init pagetable_init(void)
561 {
562 pgd_t *pgd_base = swapper_pg_dir;
563
564 permanent_kmaps_init(pgd_base);
565 }
566
567 #ifdef CONFIG_ACPI_SLEEP
568 /*
569 * ACPI suspend needs this for resume, because things like the intel-agp
570 * driver might have split up a kernel 4MB mapping.
571 */
572 char swsusp_pg_dir[PAGE_SIZE]
573 __attribute__ ((aligned(PAGE_SIZE)));
574
save_pg_dir(void)575 static inline void save_pg_dir(void)
576 {
577 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
578 }
579 #else /* !CONFIG_ACPI_SLEEP */
save_pg_dir(void)580 static inline void save_pg_dir(void)
581 {
582 }
583 #endif /* !CONFIG_ACPI_SLEEP */
584
zap_low_mappings(void)585 void zap_low_mappings(void)
586 {
587 int i;
588
589 /*
590 * Zap initial low-memory mappings.
591 *
592 * Note that "pgd_clear()" doesn't do it for
593 * us, because pgd_clear() is a no-op on i386.
594 */
595 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
596 #ifdef CONFIG_X86_PAE
597 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
598 #else
599 set_pgd(swapper_pg_dir+i, __pgd(0));
600 #endif
601 }
602 flush_tlb_all();
603 }
604
605 int nx_enabled;
606
607 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
608 EXPORT_SYMBOL_GPL(__supported_pte_mask);
609
610 #ifdef CONFIG_X86_PAE
611
612 static int disable_nx __initdata;
613
614 /*
615 * noexec = on|off
616 *
617 * Control non executable mappings.
618 *
619 * on Enable
620 * off Disable
621 */
noexec_setup(char * str)622 static int __init noexec_setup(char *str)
623 {
624 if (!str || !strcmp(str, "on")) {
625 if (cpu_has_nx) {
626 __supported_pte_mask |= _PAGE_NX;
627 disable_nx = 0;
628 }
629 } else {
630 if (!strcmp(str, "off")) {
631 disable_nx = 1;
632 __supported_pte_mask &= ~_PAGE_NX;
633 } else {
634 return -EINVAL;
635 }
636 }
637
638 return 0;
639 }
640 early_param("noexec", noexec_setup);
641
set_nx(void)642 static void __init set_nx(void)
643 {
644 unsigned int v[4], l, h;
645
646 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
647 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
648
649 if ((v[3] & (1 << 20)) && !disable_nx) {
650 rdmsr(MSR_EFER, l, h);
651 l |= EFER_NX;
652 wrmsr(MSR_EFER, l, h);
653 nx_enabled = 1;
654 __supported_pte_mask |= _PAGE_NX;
655 }
656 }
657 }
658 #endif
659
660 /* user-defined highmem size */
661 static unsigned int highmem_pages = -1;
662
663 /*
664 * highmem=size forces highmem to be exactly 'size' bytes.
665 * This works even on boxes that have no highmem otherwise.
666 * This also works to reduce highmem size on bigger boxes.
667 */
parse_highmem(char * arg)668 static int __init parse_highmem(char *arg)
669 {
670 if (!arg)
671 return -EINVAL;
672
673 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
674 return 0;
675 }
676 early_param("highmem", parse_highmem);
677
678 /*
679 * Determine low and high memory ranges:
680 */
find_low_pfn_range(void)681 void __init find_low_pfn_range(void)
682 {
683 /* it could update max_pfn */
684
685 /* max_low_pfn is 0, we already have early_res support */
686
687 max_low_pfn = max_pfn;
688 if (max_low_pfn > MAXMEM_PFN) {
689 if (highmem_pages == -1)
690 highmem_pages = max_pfn - MAXMEM_PFN;
691 if (highmem_pages + MAXMEM_PFN < max_pfn)
692 max_pfn = MAXMEM_PFN + highmem_pages;
693 if (highmem_pages + MAXMEM_PFN > max_pfn) {
694 printk(KERN_WARNING "only %luMB highmem pages "
695 "available, ignoring highmem size of %uMB.\n",
696 pages_to_mb(max_pfn - MAXMEM_PFN),
697 pages_to_mb(highmem_pages));
698 highmem_pages = 0;
699 }
700 max_low_pfn = MAXMEM_PFN;
701 #ifndef CONFIG_HIGHMEM
702 /* Maximum memory usable is what is directly addressable */
703 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
704 MAXMEM>>20);
705 if (max_pfn > MAX_NONPAE_PFN)
706 printk(KERN_WARNING
707 "Use a HIGHMEM64G enabled kernel.\n");
708 else
709 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
710 max_pfn = MAXMEM_PFN;
711 #else /* !CONFIG_HIGHMEM */
712 #ifndef CONFIG_HIGHMEM64G
713 if (max_pfn > MAX_NONPAE_PFN) {
714 max_pfn = MAX_NONPAE_PFN;
715 printk(KERN_WARNING "Warning only 4GB will be used."
716 "Use a HIGHMEM64G enabled kernel.\n");
717 }
718 #endif /* !CONFIG_HIGHMEM64G */
719 #endif /* !CONFIG_HIGHMEM */
720 } else {
721 if (highmem_pages == -1)
722 highmem_pages = 0;
723 #ifdef CONFIG_HIGHMEM
724 if (highmem_pages >= max_pfn) {
725 printk(KERN_ERR "highmem size specified (%uMB) is "
726 "bigger than pages available (%luMB)!.\n",
727 pages_to_mb(highmem_pages),
728 pages_to_mb(max_pfn));
729 highmem_pages = 0;
730 }
731 if (highmem_pages) {
732 if (max_low_pfn - highmem_pages <
733 64*1024*1024/PAGE_SIZE){
734 printk(KERN_ERR "highmem size %uMB results in "
735 "smaller than 64MB lowmem, ignoring it.\n"
736 , pages_to_mb(highmem_pages));
737 highmem_pages = 0;
738 }
739 max_low_pfn -= highmem_pages;
740 }
741 #else
742 if (highmem_pages)
743 printk(KERN_ERR "ignoring highmem size on non-highmem"
744 " kernel!\n");
745 #endif
746 }
747 }
748
749 #ifndef CONFIG_NEED_MULTIPLE_NODES
initmem_init(unsigned long start_pfn,unsigned long end_pfn)750 void __init initmem_init(unsigned long start_pfn,
751 unsigned long end_pfn)
752 {
753 #ifdef CONFIG_HIGHMEM
754 highstart_pfn = highend_pfn = max_pfn;
755 if (max_pfn > max_low_pfn)
756 highstart_pfn = max_low_pfn;
757 memory_present(0, 0, highend_pfn);
758 e820_register_active_regions(0, 0, highend_pfn);
759 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
760 pages_to_mb(highend_pfn - highstart_pfn));
761 num_physpages = highend_pfn;
762 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
763 #else
764 memory_present(0, 0, max_low_pfn);
765 e820_register_active_regions(0, 0, max_low_pfn);
766 num_physpages = max_low_pfn;
767 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
768 #endif
769 #ifdef CONFIG_FLATMEM
770 max_mapnr = num_physpages;
771 #endif
772 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
773 pages_to_mb(max_low_pfn));
774
775 setup_bootmem_allocator();
776 }
777 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
778
zone_sizes_init(void)779 static void __init zone_sizes_init(void)
780 {
781 unsigned long max_zone_pfns[MAX_NR_ZONES];
782 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
783 max_zone_pfns[ZONE_DMA] =
784 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
785 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
786 #ifdef CONFIG_HIGHMEM
787 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
788 #endif
789
790 free_area_init_nodes(max_zone_pfns);
791 }
792
setup_bootmem_allocator(void)793 void __init setup_bootmem_allocator(void)
794 {
795 int i;
796 unsigned long bootmap_size, bootmap;
797 /*
798 * Initialize the boot-time allocator (with low memory only):
799 */
800 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
801 bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
802 max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
803 PAGE_SIZE);
804 if (bootmap == -1L)
805 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
806 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
807
808 /* don't touch min_low_pfn */
809 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
810 min_low_pfn, max_low_pfn);
811 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
812 max_pfn_mapped<<PAGE_SHIFT);
813 printk(KERN_INFO " low ram: %08lx - %08lx\n",
814 min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
815 printk(KERN_INFO " bootmap %08lx - %08lx\n",
816 bootmap, bootmap + bootmap_size);
817 for_each_online_node(i)
818 free_bootmem_with_active_regions(i, max_low_pfn);
819 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
820
821 after_init_bootmem = 1;
822 }
823
find_early_table_space(unsigned long end,int use_pse)824 static void __init find_early_table_space(unsigned long end, int use_pse)
825 {
826 unsigned long puds, pmds, ptes, tables, start;
827
828 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
829 tables = PAGE_ALIGN(puds * sizeof(pud_t));
830
831 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
832 tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
833
834 if (use_pse) {
835 unsigned long extra;
836
837 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
838 extra += PMD_SIZE;
839 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
840 } else
841 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
842
843 tables += PAGE_ALIGN(ptes * sizeof(pte_t));
844
845 /* for fixmap */
846 tables += PAGE_ALIGN(__end_of_fixed_addresses * sizeof(pte_t));
847
848 /*
849 * RED-PEN putting page tables only on node 0 could
850 * cause a hotspot and fill up ZONE_DMA. The page tables
851 * need roughly 0.5KB per GB.
852 */
853 start = 0x7000;
854 table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
855 tables, PAGE_SIZE);
856 if (table_start == -1UL)
857 panic("Cannot find space for the kernel page tables");
858
859 table_start >>= PAGE_SHIFT;
860 table_end = table_start;
861 table_top = table_start + (tables>>PAGE_SHIFT);
862
863 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
864 end, table_start << PAGE_SHIFT,
865 (table_start << PAGE_SHIFT) + tables);
866 }
867
init_memory_mapping(unsigned long start,unsigned long end)868 unsigned long __init_refok init_memory_mapping(unsigned long start,
869 unsigned long end)
870 {
871 pgd_t *pgd_base = swapper_pg_dir;
872 unsigned long start_pfn, end_pfn;
873 unsigned long big_page_start;
874 #ifdef CONFIG_DEBUG_PAGEALLOC
875 /*
876 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
877 * This will simplify cpa(), which otherwise needs to support splitting
878 * large pages into small in interrupt context, etc.
879 */
880 int use_pse = 0;
881 #else
882 int use_pse = cpu_has_pse;
883 #endif
884
885 /*
886 * Find space for the kernel direct mapping tables.
887 */
888 if (!after_init_bootmem)
889 find_early_table_space(end, use_pse);
890
891 #ifdef CONFIG_X86_PAE
892 set_nx();
893 if (nx_enabled)
894 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
895 #endif
896
897 /* Enable PSE if available */
898 if (cpu_has_pse)
899 set_in_cr4(X86_CR4_PSE);
900
901 /* Enable PGE if available */
902 if (cpu_has_pge) {
903 set_in_cr4(X86_CR4_PGE);
904 __supported_pte_mask |= _PAGE_GLOBAL;
905 }
906
907 /*
908 * Don't use a large page for the first 2/4MB of memory
909 * because there are often fixed size MTRRs in there
910 * and overlapping MTRRs into large pages can cause
911 * slowdowns.
912 */
913 big_page_start = PMD_SIZE;
914
915 if (start < big_page_start) {
916 start_pfn = start >> PAGE_SHIFT;
917 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
918 } else {
919 /* head is not big page alignment ? */
920 start_pfn = start >> PAGE_SHIFT;
921 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
922 << (PMD_SHIFT - PAGE_SHIFT);
923 }
924 if (start_pfn < end_pfn)
925 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
926
927 /* big page range */
928 start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
929 << (PMD_SHIFT - PAGE_SHIFT);
930 if (start_pfn < (big_page_start >> PAGE_SHIFT))
931 start_pfn = big_page_start >> PAGE_SHIFT;
932 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
933 if (start_pfn < end_pfn)
934 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
935 use_pse);
936
937 /* tail is not big page alignment ? */
938 start_pfn = end_pfn;
939 if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
940 end_pfn = end >> PAGE_SHIFT;
941 if (start_pfn < end_pfn)
942 kernel_physical_mapping_init(pgd_base, start_pfn,
943 end_pfn, 0);
944 }
945
946 early_ioremap_page_table_range_init(pgd_base);
947
948 load_cr3(swapper_pg_dir);
949
950 __flush_tlb_all();
951
952 if (!after_init_bootmem)
953 reserve_early(table_start << PAGE_SHIFT,
954 table_end << PAGE_SHIFT, "PGTABLE");
955
956 if (!after_init_bootmem)
957 early_memtest(start, end);
958
959 return end >> PAGE_SHIFT;
960 }
961
962
963 /*
964 * paging_init() sets up the page tables - note that the first 8MB are
965 * already mapped by head.S.
966 *
967 * This routines also unmaps the page at virtual kernel address 0, so
968 * that we can trap those pesky NULL-reference errors in the kernel.
969 */
paging_init(void)970 void __init paging_init(void)
971 {
972 pagetable_init();
973
974 __flush_tlb_all();
975
976 kmap_init();
977
978 /*
979 * NOTE: at this point the bootmem allocator is fully available.
980 */
981 sparse_init();
982 zone_sizes_init();
983 }
984
985 /*
986 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
987 * and also on some strange 486's. All 586+'s are OK. This used to involve
988 * black magic jumps to work around some nasty CPU bugs, but fortunately the
989 * switch to using exceptions got rid of all that.
990 */
test_wp_bit(void)991 static void __init test_wp_bit(void)
992 {
993 printk(KERN_INFO
994 "Checking if this processor honours the WP bit even in supervisor mode...");
995
996 /* Any page-aligned address will do, the test is non-destructive */
997 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
998 boot_cpu_data.wp_works_ok = do_test_wp_bit();
999 clear_fixmap(FIX_WP_TEST);
1000
1001 if (!boot_cpu_data.wp_works_ok) {
1002 printk(KERN_CONT "No.\n");
1003 #ifdef CONFIG_X86_WP_WORKS_OK
1004 panic(
1005 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1006 #endif
1007 } else {
1008 printk(KERN_CONT "Ok.\n");
1009 }
1010 }
1011
1012 static struct kcore_list kcore_mem, kcore_vmalloc;
1013
mem_init(void)1014 void __init mem_init(void)
1015 {
1016 int codesize, reservedpages, datasize, initsize;
1017 int tmp;
1018
1019 pci_iommu_alloc();
1020
1021 #ifdef CONFIG_FLATMEM
1022 BUG_ON(!mem_map);
1023 #endif
1024 /* this will put all low memory onto the freelists */
1025 totalram_pages += free_all_bootmem();
1026
1027 reservedpages = 0;
1028 for (tmp = 0; tmp < max_low_pfn; tmp++)
1029 /*
1030 * Only count reserved RAM pages:
1031 */
1032 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1033 reservedpages++;
1034
1035 set_highmem_pages_init();
1036
1037 codesize = (unsigned long) &_etext - (unsigned long) &_text;
1038 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
1039 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
1040
1041 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1042 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1043 VMALLOC_END-VMALLOC_START);
1044
1045 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1046 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1047 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1048 num_physpages << (PAGE_SHIFT-10),
1049 codesize >> 10,
1050 reservedpages << (PAGE_SHIFT-10),
1051 datasize >> 10,
1052 initsize >> 10,
1053 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1054 );
1055
1056 printk(KERN_INFO "virtual kernel memory layout:\n"
1057 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1058 #ifdef CONFIG_HIGHMEM
1059 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
1060 #endif
1061 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
1062 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
1063 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
1064 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
1065 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
1066 FIXADDR_START, FIXADDR_TOP,
1067 (FIXADDR_TOP - FIXADDR_START) >> 10,
1068
1069 #ifdef CONFIG_HIGHMEM
1070 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1071 (LAST_PKMAP*PAGE_SIZE) >> 10,
1072 #endif
1073
1074 VMALLOC_START, VMALLOC_END,
1075 (VMALLOC_END - VMALLOC_START) >> 20,
1076
1077 (unsigned long)__va(0), (unsigned long)high_memory,
1078 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1079
1080 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1081 ((unsigned long)&__init_end -
1082 (unsigned long)&__init_begin) >> 10,
1083
1084 (unsigned long)&_etext, (unsigned long)&_edata,
1085 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1086
1087 (unsigned long)&_text, (unsigned long)&_etext,
1088 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1089
1090 /*
1091 * Check boundaries twice: Some fundamental inconsistencies can
1092 * be detected at build time already.
1093 */
1094 #define __FIXADDR_TOP (-PAGE_SIZE)
1095 #ifdef CONFIG_HIGHMEM
1096 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1097 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
1098 #endif
1099 #define high_memory (-128UL << 20)
1100 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
1101 #undef high_memory
1102 #undef __FIXADDR_TOP
1103
1104 #ifdef CONFIG_HIGHMEM
1105 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
1106 BUG_ON(VMALLOC_END > PKMAP_BASE);
1107 #endif
1108 BUG_ON(VMALLOC_START >= VMALLOC_END);
1109 BUG_ON((unsigned long)high_memory > VMALLOC_START);
1110
1111 if (boot_cpu_data.wp_works_ok < 0)
1112 test_wp_bit();
1113
1114 save_pg_dir();
1115 zap_low_mappings();
1116 }
1117
1118 #ifdef CONFIG_MEMORY_HOTPLUG
arch_add_memory(int nid,u64 start,u64 size)1119 int arch_add_memory(int nid, u64 start, u64 size)
1120 {
1121 struct pglist_data *pgdata = NODE_DATA(nid);
1122 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1123 unsigned long start_pfn = start >> PAGE_SHIFT;
1124 unsigned long nr_pages = size >> PAGE_SHIFT;
1125
1126 return __add_pages(nid, zone, start_pfn, nr_pages);
1127 }
1128 #endif
1129
1130 /*
1131 * This function cannot be __init, since exceptions don't work in that
1132 * section. Put this after the callers, so that it cannot be inlined.
1133 */
do_test_wp_bit(void)1134 static noinline int do_test_wp_bit(void)
1135 {
1136 char tmp_reg;
1137 int flag;
1138
1139 __asm__ __volatile__(
1140 " movb %0, %1 \n"
1141 "1: movb %1, %0 \n"
1142 " xorl %2, %2 \n"
1143 "2: \n"
1144 _ASM_EXTABLE(1b,2b)
1145 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1146 "=q" (tmp_reg),
1147 "=r" (flag)
1148 :"2" (1)
1149 :"memory");
1150
1151 return flag;
1152 }
1153
1154 #ifdef CONFIG_DEBUG_RODATA
1155 const int rodata_test_data = 0xC3;
1156 EXPORT_SYMBOL_GPL(rodata_test_data);
1157
mark_rodata_ro(void)1158 void mark_rodata_ro(void)
1159 {
1160 unsigned long start = PFN_ALIGN(_text);
1161 unsigned long size = PFN_ALIGN(_etext) - start;
1162
1163 #ifndef CONFIG_DYNAMIC_FTRACE
1164 /* Dynamic tracing modifies the kernel text section */
1165 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1166 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1167 size >> 10);
1168
1169 #ifdef CONFIG_CPA_DEBUG
1170 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1171 start, start+size);
1172 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1173
1174 printk(KERN_INFO "Testing CPA: write protecting again\n");
1175 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1176 #endif
1177 #endif /* CONFIG_DYNAMIC_FTRACE */
1178
1179 start += size;
1180 size = (unsigned long)__end_rodata - start;
1181 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1182 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1183 size >> 10);
1184 rodata_test();
1185
1186 #ifdef CONFIG_CPA_DEBUG
1187 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1188 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1189
1190 printk(KERN_INFO "Testing CPA: write protecting again\n");
1191 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1192 #endif
1193 }
1194 #endif
1195
free_init_pages(char * what,unsigned long begin,unsigned long end)1196 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1197 {
1198 #ifdef CONFIG_DEBUG_PAGEALLOC
1199 /*
1200 * If debugging page accesses then do not free this memory but
1201 * mark them not present - any buggy init-section access will
1202 * create a kernel page fault:
1203 */
1204 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1205 begin, PAGE_ALIGN(end));
1206 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1207 #else
1208 unsigned long addr;
1209
1210 /*
1211 * We just marked the kernel text read only above, now that
1212 * we are going to free part of that, we need to make that
1213 * writeable first.
1214 */
1215 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1216
1217 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1218 ClearPageReserved(virt_to_page(addr));
1219 init_page_count(virt_to_page(addr));
1220 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1221 free_page(addr);
1222 totalram_pages++;
1223 }
1224 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1225 #endif
1226 }
1227
free_initmem(void)1228 void free_initmem(void)
1229 {
1230 free_init_pages("unused kernel memory",
1231 (unsigned long)(&__init_begin),
1232 (unsigned long)(&__init_end));
1233 }
1234
1235 #ifdef CONFIG_BLK_DEV_INITRD
free_initrd_mem(unsigned long start,unsigned long end)1236 void free_initrd_mem(unsigned long start, unsigned long end)
1237 {
1238 free_init_pages("initrd memory", start, end);
1239 }
1240 #endif
1241
reserve_bootmem_generic(unsigned long phys,unsigned long len,int flags)1242 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1243 int flags)
1244 {
1245 return reserve_bootmem(phys, len, flags);
1246 }
1247