1 /*
2 *
3 * Copyright (C) 1995 Linus Torvalds
4 *
5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6 */
7
8 #include <linux/module.h>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/mm.h>
18 #include <linux/hugetlb.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/bootmem.h>
28 #include <linux/memblock.h>
29 #include <linux/proc_fs.h>
30 #include <linux/memory_hotplug.h>
31 #include <linux/initrd.h>
32 #include <linux/cpumask.h>
33 #include <linux/gfp.h>
34
35 #include <asm/asm.h>
36 #include <asm/bios_ebda.h>
37 #include <asm/processor.h>
38 #include <asm/system.h>
39 #include <asm/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/dma.h>
42 #include <asm/fixmap.h>
43 #include <asm/e820.h>
44 #include <asm/apic.h>
45 #include <asm/bugs.h>
46 #include <asm/tlb.h>
47 #include <asm/tlbflush.h>
48 #include <asm/olpc_ofw.h>
49 #include <asm/pgalloc.h>
50 #include <asm/sections.h>
51 #include <asm/paravirt.h>
52 #include <asm/setup.h>
53 #include <asm/cacheflush.h>
54 #include <asm/page_types.h>
55 #include <asm/init.h>
56
57 unsigned long highstart_pfn, highend_pfn;
58
59 static noinline int do_test_wp_bit(void);
60
61 bool __read_mostly __vmalloc_start_set = false;
62
alloc_low_page(void)63 static __init void *alloc_low_page(void)
64 {
65 unsigned long pfn = pgt_buf_end++;
66 void *adr;
67
68 if (pfn >= pgt_buf_top)
69 panic("alloc_low_page: ran out of memory");
70
71 adr = __va(pfn * PAGE_SIZE);
72 clear_page(adr);
73 return adr;
74 }
75
76 /*
77 * Creates a middle page table and puts a pointer to it in the
78 * given global directory entry. This only returns the gd entry
79 * in non-PAE compilation mode, since the middle layer is folded.
80 */
one_md_table_init(pgd_t * pgd)81 static pmd_t * __init one_md_table_init(pgd_t *pgd)
82 {
83 pud_t *pud;
84 pmd_t *pmd_table;
85
86 #ifdef CONFIG_X86_PAE
87 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
88 if (after_bootmem)
89 pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE);
90 else
91 pmd_table = (pmd_t *)alloc_low_page();
92 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
93 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
94 pud = pud_offset(pgd, 0);
95 BUG_ON(pmd_table != pmd_offset(pud, 0));
96
97 return pmd_table;
98 }
99 #endif
100 pud = pud_offset(pgd, 0);
101 pmd_table = pmd_offset(pud, 0);
102
103 return pmd_table;
104 }
105
106 /*
107 * Create a page table and place a pointer to it in a middle page
108 * directory entry:
109 */
one_page_table_init(pmd_t * pmd)110 static pte_t * __init one_page_table_init(pmd_t *pmd)
111 {
112 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
113 pte_t *page_table = NULL;
114
115 if (after_bootmem) {
116 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
117 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
118 #endif
119 if (!page_table)
120 page_table =
121 (pte_t *)alloc_bootmem_pages(PAGE_SIZE);
122 } else
123 page_table = (pte_t *)alloc_low_page();
124
125 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
126 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
127 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
128 }
129
130 return pte_offset_kernel(pmd, 0);
131 }
132
populate_extra_pmd(unsigned long vaddr)133 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
134 {
135 int pgd_idx = pgd_index(vaddr);
136 int pmd_idx = pmd_index(vaddr);
137
138 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
139 }
140
populate_extra_pte(unsigned long vaddr)141 pte_t * __init populate_extra_pte(unsigned long vaddr)
142 {
143 int pte_idx = pte_index(vaddr);
144 pmd_t *pmd;
145
146 pmd = populate_extra_pmd(vaddr);
147 return one_page_table_init(pmd) + pte_idx;
148 }
149
page_table_kmap_check(pte_t * pte,pmd_t * pmd,unsigned long vaddr,pte_t * lastpte)150 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
151 unsigned long vaddr, pte_t *lastpte)
152 {
153 #ifdef CONFIG_HIGHMEM
154 /*
155 * Something (early fixmap) may already have put a pte
156 * page here, which causes the page table allocation
157 * to become nonlinear. Attempt to fix it, and if it
158 * is still nonlinear then we have to bug.
159 */
160 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
161 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
162
163 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
164 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
165 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
166 && ((__pa(pte) >> PAGE_SHIFT) < pgt_buf_start
167 || (__pa(pte) >> PAGE_SHIFT) >= pgt_buf_end)) {
168 pte_t *newpte;
169 int i;
170
171 BUG_ON(after_bootmem);
172 newpte = alloc_low_page();
173 for (i = 0; i < PTRS_PER_PTE; i++)
174 set_pte(newpte + i, pte[i]);
175
176 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
177 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
178 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
179 __flush_tlb_all();
180
181 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
182 pte = newpte;
183 }
184 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
185 && vaddr > fix_to_virt(FIX_KMAP_END)
186 && lastpte && lastpte + PTRS_PER_PTE != pte);
187 #endif
188 return pte;
189 }
190
191 /*
192 * This function initializes a certain range of kernel virtual memory
193 * with new bootmem page tables, everywhere page tables are missing in
194 * the given range.
195 *
196 * NOTE: The pagetables are allocated contiguous on the physical space
197 * so we can cache the place of the first one and move around without
198 * checking the pgd every time.
199 */
200 static void __init
page_table_range_init(unsigned long start,unsigned long end,pgd_t * pgd_base)201 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
202 {
203 int pgd_idx, pmd_idx;
204 unsigned long vaddr;
205 pgd_t *pgd;
206 pmd_t *pmd;
207 pte_t *pte = NULL;
208
209 vaddr = start;
210 pgd_idx = pgd_index(vaddr);
211 pmd_idx = pmd_index(vaddr);
212 pgd = pgd_base + pgd_idx;
213
214 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
215 pmd = one_md_table_init(pgd);
216 pmd = pmd + pmd_index(vaddr);
217 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
218 pmd++, pmd_idx++) {
219 pte = page_table_kmap_check(one_page_table_init(pmd),
220 pmd, vaddr, pte);
221
222 vaddr += PMD_SIZE;
223 }
224 pmd_idx = 0;
225 }
226 }
227
is_kernel_text(unsigned long addr)228 static inline int is_kernel_text(unsigned long addr)
229 {
230 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
231 return 1;
232 return 0;
233 }
234
235 /*
236 * This maps the physical memory to kernel virtual address space, a total
237 * of max_low_pfn pages, by creating page tables starting from address
238 * PAGE_OFFSET:
239 */
240 unsigned long __init
kernel_physical_mapping_init(unsigned long start,unsigned long end,unsigned long page_size_mask)241 kernel_physical_mapping_init(unsigned long start,
242 unsigned long end,
243 unsigned long page_size_mask)
244 {
245 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
246 unsigned long last_map_addr = end;
247 unsigned long start_pfn, end_pfn;
248 pgd_t *pgd_base = swapper_pg_dir;
249 int pgd_idx, pmd_idx, pte_ofs;
250 unsigned long pfn;
251 pgd_t *pgd;
252 pmd_t *pmd;
253 pte_t *pte;
254 unsigned pages_2m, pages_4k;
255 int mapping_iter;
256
257 start_pfn = start >> PAGE_SHIFT;
258 end_pfn = end >> PAGE_SHIFT;
259
260 /*
261 * First iteration will setup identity mapping using large/small pages
262 * based on use_pse, with other attributes same as set by
263 * the early code in head_32.S
264 *
265 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
266 * as desired for the kernel identity mapping.
267 *
268 * This two pass mechanism conforms to the TLB app note which says:
269 *
270 * "Software should not write to a paging-structure entry in a way
271 * that would change, for any linear address, both the page size
272 * and either the page frame or attributes."
273 */
274 mapping_iter = 1;
275
276 if (!cpu_has_pse)
277 use_pse = 0;
278
279 repeat:
280 pages_2m = pages_4k = 0;
281 pfn = start_pfn;
282 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
283 pgd = pgd_base + pgd_idx;
284 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
285 pmd = one_md_table_init(pgd);
286
287 if (pfn >= end_pfn)
288 continue;
289 #ifdef CONFIG_X86_PAE
290 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
291 pmd += pmd_idx;
292 #else
293 pmd_idx = 0;
294 #endif
295 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
296 pmd++, pmd_idx++) {
297 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
298
299 /*
300 * Map with big pages if possible, otherwise
301 * create normal page tables:
302 */
303 if (use_pse) {
304 unsigned int addr2;
305 pgprot_t prot = PAGE_KERNEL_LARGE;
306 /*
307 * first pass will use the same initial
308 * identity mapping attribute + _PAGE_PSE.
309 */
310 pgprot_t init_prot =
311 __pgprot(PTE_IDENT_ATTR |
312 _PAGE_PSE);
313
314 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
315 PAGE_OFFSET + PAGE_SIZE-1;
316
317 if (is_kernel_text(addr) ||
318 is_kernel_text(addr2))
319 prot = PAGE_KERNEL_LARGE_EXEC;
320
321 pages_2m++;
322 if (mapping_iter == 1)
323 set_pmd(pmd, pfn_pmd(pfn, init_prot));
324 else
325 set_pmd(pmd, pfn_pmd(pfn, prot));
326
327 pfn += PTRS_PER_PTE;
328 continue;
329 }
330 pte = one_page_table_init(pmd);
331
332 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
333 pte += pte_ofs;
334 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
335 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
336 pgprot_t prot = PAGE_KERNEL;
337 /*
338 * first pass will use the same initial
339 * identity mapping attribute.
340 */
341 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
342
343 if (is_kernel_text(addr))
344 prot = PAGE_KERNEL_EXEC;
345
346 pages_4k++;
347 if (mapping_iter == 1) {
348 set_pte(pte, pfn_pte(pfn, init_prot));
349 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
350 } else
351 set_pte(pte, pfn_pte(pfn, prot));
352 }
353 }
354 }
355 if (mapping_iter == 1) {
356 /*
357 * update direct mapping page count only in the first
358 * iteration.
359 */
360 update_page_count(PG_LEVEL_2M, pages_2m);
361 update_page_count(PG_LEVEL_4K, pages_4k);
362
363 /*
364 * local global flush tlb, which will flush the previous
365 * mappings present in both small and large page TLB's.
366 */
367 __flush_tlb_all();
368
369 /*
370 * Second iteration will set the actual desired PTE attributes.
371 */
372 mapping_iter = 2;
373 goto repeat;
374 }
375 return last_map_addr;
376 }
377
378 pte_t *kmap_pte;
379 pgprot_t kmap_prot;
380
kmap_get_fixmap_pte(unsigned long vaddr)381 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
382 {
383 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
384 vaddr), vaddr), vaddr);
385 }
386
kmap_init(void)387 static void __init kmap_init(void)
388 {
389 unsigned long kmap_vstart;
390
391 /*
392 * Cache the first kmap pte:
393 */
394 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
395 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
396
397 kmap_prot = PAGE_KERNEL;
398 }
399
400 #ifdef CONFIG_HIGHMEM
permanent_kmaps_init(pgd_t * pgd_base)401 static void __init permanent_kmaps_init(pgd_t *pgd_base)
402 {
403 unsigned long vaddr;
404 pgd_t *pgd;
405 pud_t *pud;
406 pmd_t *pmd;
407 pte_t *pte;
408
409 vaddr = PKMAP_BASE;
410 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
411
412 pgd = swapper_pg_dir + pgd_index(vaddr);
413 pud = pud_offset(pgd, vaddr);
414 pmd = pmd_offset(pud, vaddr);
415 pte = pte_offset_kernel(pmd, vaddr);
416 pkmap_page_table = pte;
417 }
418
add_one_highpage_init(struct page * page)419 static void __init add_one_highpage_init(struct page *page)
420 {
421 ClearPageReserved(page);
422 init_page_count(page);
423 __free_page(page);
424 totalhigh_pages++;
425 }
426
add_highpages_with_active_regions(int nid,unsigned long start_pfn,unsigned long end_pfn)427 void __init add_highpages_with_active_regions(int nid,
428 unsigned long start_pfn, unsigned long end_pfn)
429 {
430 struct range *range;
431 int nr_range;
432 int i;
433
434 nr_range = __get_free_all_memory_range(&range, nid, start_pfn, end_pfn);
435
436 for (i = 0; i < nr_range; i++) {
437 struct page *page;
438 int node_pfn;
439
440 for (node_pfn = range[i].start; node_pfn < range[i].end;
441 node_pfn++) {
442 if (!pfn_valid(node_pfn))
443 continue;
444 page = pfn_to_page(node_pfn);
445 add_one_highpage_init(page);
446 }
447 }
448 }
449 #else
permanent_kmaps_init(pgd_t * pgd_base)450 static inline void permanent_kmaps_init(pgd_t *pgd_base)
451 {
452 }
453 #endif /* CONFIG_HIGHMEM */
454
native_pagetable_setup_start(pgd_t * base)455 void __init native_pagetable_setup_start(pgd_t *base)
456 {
457 unsigned long pfn, va;
458 pgd_t *pgd;
459 pud_t *pud;
460 pmd_t *pmd;
461 pte_t *pte;
462
463 /*
464 * Remove any mappings which extend past the end of physical
465 * memory from the boot time page table:
466 */
467 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
468 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
469 pgd = base + pgd_index(va);
470 if (!pgd_present(*pgd))
471 break;
472
473 pud = pud_offset(pgd, va);
474 pmd = pmd_offset(pud, va);
475 if (!pmd_present(*pmd))
476 break;
477
478 pte = pte_offset_kernel(pmd, va);
479 if (!pte_present(*pte))
480 break;
481
482 pte_clear(NULL, va, pte);
483 }
484 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
485 }
486
native_pagetable_setup_done(pgd_t * base)487 void __init native_pagetable_setup_done(pgd_t *base)
488 {
489 }
490
491 /*
492 * Build a proper pagetable for the kernel mappings. Up until this
493 * point, we've been running on some set of pagetables constructed by
494 * the boot process.
495 *
496 * If we're booting on native hardware, this will be a pagetable
497 * constructed in arch/x86/kernel/head_32.S. The root of the
498 * pagetable will be swapper_pg_dir.
499 *
500 * If we're booting paravirtualized under a hypervisor, then there are
501 * more options: we may already be running PAE, and the pagetable may
502 * or may not be based in swapper_pg_dir. In any case,
503 * paravirt_pagetable_setup_start() will set up swapper_pg_dir
504 * appropriately for the rest of the initialization to work.
505 *
506 * In general, pagetable_init() assumes that the pagetable may already
507 * be partially populated, and so it avoids stomping on any existing
508 * mappings.
509 */
early_ioremap_page_table_range_init(void)510 void __init early_ioremap_page_table_range_init(void)
511 {
512 pgd_t *pgd_base = swapper_pg_dir;
513 unsigned long vaddr, end;
514
515 /*
516 * Fixed mappings, only the page table structure has to be
517 * created - mappings will be set by set_fixmap():
518 */
519 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
520 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
521 page_table_range_init(vaddr, end, pgd_base);
522 early_ioremap_reset();
523 }
524
pagetable_init(void)525 static void __init pagetable_init(void)
526 {
527 pgd_t *pgd_base = swapper_pg_dir;
528
529 permanent_kmaps_init(pgd_base);
530 }
531
532 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
533 EXPORT_SYMBOL_GPL(__supported_pte_mask);
534
535 /* user-defined highmem size */
536 static unsigned int highmem_pages = -1;
537
538 /*
539 * highmem=size forces highmem to be exactly 'size' bytes.
540 * This works even on boxes that have no highmem otherwise.
541 * This also works to reduce highmem size on bigger boxes.
542 */
parse_highmem(char * arg)543 static int __init parse_highmem(char *arg)
544 {
545 if (!arg)
546 return -EINVAL;
547
548 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
549 return 0;
550 }
551 early_param("highmem", parse_highmem);
552
553 #define MSG_HIGHMEM_TOO_BIG \
554 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
555
556 #define MSG_LOWMEM_TOO_SMALL \
557 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
558 /*
559 * All of RAM fits into lowmem - but if user wants highmem
560 * artificially via the highmem=x boot parameter then create
561 * it:
562 */
lowmem_pfn_init(void)563 void __init lowmem_pfn_init(void)
564 {
565 /* max_low_pfn is 0, we already have early_res support */
566 max_low_pfn = max_pfn;
567
568 if (highmem_pages == -1)
569 highmem_pages = 0;
570 #ifdef CONFIG_HIGHMEM
571 if (highmem_pages >= max_pfn) {
572 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
573 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
574 highmem_pages = 0;
575 }
576 if (highmem_pages) {
577 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
578 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
579 pages_to_mb(highmem_pages));
580 highmem_pages = 0;
581 }
582 max_low_pfn -= highmem_pages;
583 }
584 #else
585 if (highmem_pages)
586 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
587 #endif
588 }
589
590 #define MSG_HIGHMEM_TOO_SMALL \
591 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
592
593 #define MSG_HIGHMEM_TRIMMED \
594 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
595 /*
596 * We have more RAM than fits into lowmem - we try to put it into
597 * highmem, also taking the highmem=x boot parameter into account:
598 */
highmem_pfn_init(void)599 void __init highmem_pfn_init(void)
600 {
601 max_low_pfn = MAXMEM_PFN;
602
603 if (highmem_pages == -1)
604 highmem_pages = max_pfn - MAXMEM_PFN;
605
606 if (highmem_pages + MAXMEM_PFN < max_pfn)
607 max_pfn = MAXMEM_PFN + highmem_pages;
608
609 if (highmem_pages + MAXMEM_PFN > max_pfn) {
610 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
611 pages_to_mb(max_pfn - MAXMEM_PFN),
612 pages_to_mb(highmem_pages));
613 highmem_pages = 0;
614 }
615 #ifndef CONFIG_HIGHMEM
616 /* Maximum memory usable is what is directly addressable */
617 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
618 if (max_pfn > MAX_NONPAE_PFN)
619 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
620 else
621 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
622 max_pfn = MAXMEM_PFN;
623 #else /* !CONFIG_HIGHMEM */
624 #ifndef CONFIG_HIGHMEM64G
625 if (max_pfn > MAX_NONPAE_PFN) {
626 max_pfn = MAX_NONPAE_PFN;
627 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
628 }
629 #endif /* !CONFIG_HIGHMEM64G */
630 #endif /* !CONFIG_HIGHMEM */
631 }
632
633 /*
634 * Determine low and high memory ranges:
635 */
find_low_pfn_range(void)636 void __init find_low_pfn_range(void)
637 {
638 /* it could update max_pfn */
639
640 if (max_pfn <= MAXMEM_PFN)
641 lowmem_pfn_init();
642 else
643 highmem_pfn_init();
644 }
645
646 #ifndef CONFIG_NEED_MULTIPLE_NODES
initmem_init(void)647 void __init initmem_init(void)
648 {
649 #ifdef CONFIG_HIGHMEM
650 highstart_pfn = highend_pfn = max_pfn;
651 if (max_pfn > max_low_pfn)
652 highstart_pfn = max_low_pfn;
653 memblock_x86_register_active_regions(0, 0, highend_pfn);
654 sparse_memory_present_with_active_regions(0);
655 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
656 pages_to_mb(highend_pfn - highstart_pfn));
657 num_physpages = highend_pfn;
658 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
659 #else
660 memblock_x86_register_active_regions(0, 0, max_low_pfn);
661 sparse_memory_present_with_active_regions(0);
662 num_physpages = max_low_pfn;
663 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
664 #endif
665 #ifdef CONFIG_FLATMEM
666 max_mapnr = num_physpages;
667 #endif
668 __vmalloc_start_set = true;
669
670 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
671 pages_to_mb(max_low_pfn));
672
673 setup_bootmem_allocator();
674 }
675 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
676
zone_sizes_init(void)677 static void __init zone_sizes_init(void)
678 {
679 unsigned long max_zone_pfns[MAX_NR_ZONES];
680 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
681 max_zone_pfns[ZONE_DMA] =
682 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
683 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
684 #ifdef CONFIG_HIGHMEM
685 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
686 #endif
687
688 free_area_init_nodes(max_zone_pfns);
689 }
690
setup_bootmem_allocator(void)691 void __init setup_bootmem_allocator(void)
692 {
693 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
694 max_pfn_mapped<<PAGE_SHIFT);
695 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
696
697 after_bootmem = 1;
698 }
699
700 /*
701 * paging_init() sets up the page tables - note that the first 8MB are
702 * already mapped by head.S.
703 *
704 * This routines also unmaps the page at virtual kernel address 0, so
705 * that we can trap those pesky NULL-reference errors in the kernel.
706 */
paging_init(void)707 void __init paging_init(void)
708 {
709 pagetable_init();
710
711 __flush_tlb_all();
712
713 kmap_init();
714
715 /*
716 * NOTE: at this point the bootmem allocator is fully available.
717 */
718 olpc_dt_build_devicetree();
719 sparse_init();
720 zone_sizes_init();
721 }
722
723 /*
724 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
725 * and also on some strange 486's. All 586+'s are OK. This used to involve
726 * black magic jumps to work around some nasty CPU bugs, but fortunately the
727 * switch to using exceptions got rid of all that.
728 */
test_wp_bit(void)729 static void __init test_wp_bit(void)
730 {
731 printk(KERN_INFO
732 "Checking if this processor honours the WP bit even in supervisor mode...");
733
734 /* Any page-aligned address will do, the test is non-destructive */
735 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
736 boot_cpu_data.wp_works_ok = do_test_wp_bit();
737 clear_fixmap(FIX_WP_TEST);
738
739 if (!boot_cpu_data.wp_works_ok) {
740 printk(KERN_CONT "No.\n");
741 #ifdef CONFIG_X86_WP_WORKS_OK
742 panic(
743 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
744 #endif
745 } else {
746 printk(KERN_CONT "Ok.\n");
747 }
748 }
749
mem_init(void)750 void __init mem_init(void)
751 {
752 int codesize, reservedpages, datasize, initsize;
753 int tmp;
754
755 pci_iommu_alloc();
756
757 #ifdef CONFIG_FLATMEM
758 BUG_ON(!mem_map);
759 #endif
760 /* this will put all low memory onto the freelists */
761 totalram_pages += free_all_bootmem();
762
763 reservedpages = 0;
764 for (tmp = 0; tmp < max_low_pfn; tmp++)
765 /*
766 * Only count reserved RAM pages:
767 */
768 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
769 reservedpages++;
770
771 set_highmem_pages_init();
772
773 codesize = (unsigned long) &_etext - (unsigned long) &_text;
774 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
775 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
776
777 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
778 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
779 nr_free_pages() << (PAGE_SHIFT-10),
780 num_physpages << (PAGE_SHIFT-10),
781 codesize >> 10,
782 reservedpages << (PAGE_SHIFT-10),
783 datasize >> 10,
784 initsize >> 10,
785 totalhigh_pages << (PAGE_SHIFT-10));
786
787 printk(KERN_INFO "virtual kernel memory layout:\n"
788 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
789 #ifdef CONFIG_HIGHMEM
790 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
791 #endif
792 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
793 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
794 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
795 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
796 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
797 FIXADDR_START, FIXADDR_TOP,
798 (FIXADDR_TOP - FIXADDR_START) >> 10,
799
800 #ifdef CONFIG_HIGHMEM
801 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
802 (LAST_PKMAP*PAGE_SIZE) >> 10,
803 #endif
804
805 VMALLOC_START, VMALLOC_END,
806 (VMALLOC_END - VMALLOC_START) >> 20,
807
808 (unsigned long)__va(0), (unsigned long)high_memory,
809 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
810
811 (unsigned long)&__init_begin, (unsigned long)&__init_end,
812 ((unsigned long)&__init_end -
813 (unsigned long)&__init_begin) >> 10,
814
815 (unsigned long)&_etext, (unsigned long)&_edata,
816 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
817
818 (unsigned long)&_text, (unsigned long)&_etext,
819 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
820
821 /*
822 * Check boundaries twice: Some fundamental inconsistencies can
823 * be detected at build time already.
824 */
825 #define __FIXADDR_TOP (-PAGE_SIZE)
826 #ifdef CONFIG_HIGHMEM
827 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
828 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
829 #endif
830 #define high_memory (-128UL << 20)
831 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
832 #undef high_memory
833 #undef __FIXADDR_TOP
834
835 #ifdef CONFIG_HIGHMEM
836 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
837 BUG_ON(VMALLOC_END > PKMAP_BASE);
838 #endif
839 BUG_ON(VMALLOC_START >= VMALLOC_END);
840 BUG_ON((unsigned long)high_memory > VMALLOC_START);
841
842 if (boot_cpu_data.wp_works_ok < 0)
843 test_wp_bit();
844 }
845
846 #ifdef CONFIG_MEMORY_HOTPLUG
arch_add_memory(int nid,u64 start,u64 size)847 int arch_add_memory(int nid, u64 start, u64 size)
848 {
849 struct pglist_data *pgdata = NODE_DATA(nid);
850 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
851 unsigned long start_pfn = start >> PAGE_SHIFT;
852 unsigned long nr_pages = size >> PAGE_SHIFT;
853
854 return __add_pages(nid, zone, start_pfn, nr_pages);
855 }
856 #endif
857
858 /*
859 * This function cannot be __init, since exceptions don't work in that
860 * section. Put this after the callers, so that it cannot be inlined.
861 */
do_test_wp_bit(void)862 static noinline int do_test_wp_bit(void)
863 {
864 char tmp_reg;
865 int flag;
866
867 __asm__ __volatile__(
868 " movb %0, %1 \n"
869 "1: movb %1, %0 \n"
870 " xorl %2, %2 \n"
871 "2: \n"
872 _ASM_EXTABLE(1b,2b)
873 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
874 "=q" (tmp_reg),
875 "=r" (flag)
876 :"2" (1)
877 :"memory");
878
879 return flag;
880 }
881
882 #ifdef CONFIG_DEBUG_RODATA
883 const int rodata_test_data = 0xC3;
884 EXPORT_SYMBOL_GPL(rodata_test_data);
885
886 int kernel_set_to_readonly __read_mostly;
887
set_kernel_text_rw(void)888 void set_kernel_text_rw(void)
889 {
890 unsigned long start = PFN_ALIGN(_text);
891 unsigned long size = PFN_ALIGN(_etext) - start;
892
893 if (!kernel_set_to_readonly)
894 return;
895
896 pr_debug("Set kernel text: %lx - %lx for read write\n",
897 start, start+size);
898
899 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
900 }
901
set_kernel_text_ro(void)902 void set_kernel_text_ro(void)
903 {
904 unsigned long start = PFN_ALIGN(_text);
905 unsigned long size = PFN_ALIGN(_etext) - start;
906
907 if (!kernel_set_to_readonly)
908 return;
909
910 pr_debug("Set kernel text: %lx - %lx for read only\n",
911 start, start+size);
912
913 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
914 }
915
mark_nxdata_nx(void)916 static void mark_nxdata_nx(void)
917 {
918 /*
919 * When this called, init has already been executed and released,
920 * so everything past _etext should be NX.
921 */
922 unsigned long start = PFN_ALIGN(_etext);
923 /*
924 * This comes from is_kernel_text upper limit. Also HPAGE where used:
925 */
926 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
927
928 if (__supported_pte_mask & _PAGE_NX)
929 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
930 set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
931 }
932
mark_rodata_ro(void)933 void mark_rodata_ro(void)
934 {
935 unsigned long start = PFN_ALIGN(_text);
936 unsigned long size = PFN_ALIGN(_etext) - start;
937
938 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
939 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
940 size >> 10);
941
942 kernel_set_to_readonly = 1;
943
944 #ifdef CONFIG_CPA_DEBUG
945 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
946 start, start+size);
947 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
948
949 printk(KERN_INFO "Testing CPA: write protecting again\n");
950 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
951 #endif
952
953 start += size;
954 size = (unsigned long)__end_rodata - start;
955 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
956 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
957 size >> 10);
958 rodata_test();
959
960 #ifdef CONFIG_CPA_DEBUG
961 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
962 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
963
964 printk(KERN_INFO "Testing CPA: write protecting again\n");
965 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
966 #endif
967 mark_nxdata_nx();
968 }
969 #endif
970
971