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