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/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
57unsigned long highstart_pfn, highend_pfn;
58
59static noinline int do_test_wp_bit(void);
60
61bool __read_mostly __vmalloc_start_set = false;
62
63static __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 */
81static 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 */
110static 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
133pmd_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
141pte_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
150static 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 */
200static void __init
201page_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
228static 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 */
240unsigned long __init
241kernel_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
279repeat:
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
378pte_t *kmap_pte;
379pgprot_t kmap_prot;
380
381static 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
387static 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
401static 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
419static 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
427void __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
450static inline void permanent_kmaps_init(pgd_t *pgd_base)
451{
452}
453#endif /* CONFIG_HIGHMEM */
454
455void __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
487void __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 */
510void __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
525static void __init pagetable_init(void)
526{
527 pgd_t *pgd_base = swapper_pg_dir;
528
529 permanent_kmaps_init(pgd_base);
530}
531
532pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
533EXPORT_SYMBOL_GPL(__supported_pte_mask);
534
535/* user-defined highmem size */
536static 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 */
543static 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}
551early_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 */
563void __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 */
599void __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 */
636void __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
647void __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
677static 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#ifdef CONFIG_ZONE_DMA
682 max_zone_pfns[ZONE_DMA] =
683 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
684#endif
685 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
686#ifdef CONFIG_HIGHMEM
687 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
688#endif
689
690 free_area_init_nodes(max_zone_pfns);
691}
692
693void __init setup_bootmem_allocator(void)
694{
695 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
696 max_pfn_mapped<<PAGE_SHIFT);
697 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
698
699 after_bootmem = 1;
700}
701
702/*
703 * paging_init() sets up the page tables - note that the first 8MB are
704 * already mapped by head.S.
705 *
706 * This routines also unmaps the page at virtual kernel address 0, so
707 * that we can trap those pesky NULL-reference errors in the kernel.
708 */
709void __init paging_init(void)
710{
711 pagetable_init();
712
713 __flush_tlb_all();
714
715 kmap_init();
716
717 /*
718 * NOTE: at this point the bootmem allocator is fully available.
719 */
720 olpc_dt_build_devicetree();
721 sparse_memory_present_with_active_regions(MAX_NUMNODES);
722 sparse_init();
723 zone_sizes_init();
724}
725
726/*
727 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
728 * and also on some strange 486's. All 586+'s are OK. This used to involve
729 * black magic jumps to work around some nasty CPU bugs, but fortunately the
730 * switch to using exceptions got rid of all that.
731 */
732static void __init test_wp_bit(void)
733{
734 printk(KERN_INFO
735 "Checking if this processor honours the WP bit even in supervisor mode...");
736
737 /* Any page-aligned address will do, the test is non-destructive */
738 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
739 boot_cpu_data.wp_works_ok = do_test_wp_bit();
740 clear_fixmap(FIX_WP_TEST);
741
742 if (!boot_cpu_data.wp_works_ok) {
743 printk(KERN_CONT "No.\n");
744#ifdef CONFIG_X86_WP_WORKS_OK
745 panic(
746 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
747#endif
748 } else {
749 printk(KERN_CONT "Ok.\n");
750 }
751}
752
753void __init mem_init(void)
754{
755 int codesize, reservedpages, datasize, initsize;
756 int tmp;
757
758 pci_iommu_alloc();
759
760#ifdef CONFIG_FLATMEM
761 BUG_ON(!mem_map);
762#endif
763 /* this will put all low memory onto the freelists */
764 totalram_pages += free_all_bootmem();
765
766 reservedpages = 0;
767 for (tmp = 0; tmp < max_low_pfn; tmp++)
768 /*
769 * Only count reserved RAM pages:
770 */
771 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
772 reservedpages++;
773
774 set_highmem_pages_init();
775
776 codesize = (unsigned long) &_etext - (unsigned long) &_text;
777 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
778 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
779
780 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
781 "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
782 nr_free_pages() << (PAGE_SHIFT-10),
783 num_physpages << (PAGE_SHIFT-10),
784 codesize >> 10,
785 reservedpages << (PAGE_SHIFT-10),
786 datasize >> 10,
787 initsize >> 10,
788 totalhigh_pages << (PAGE_SHIFT-10));
789
790 printk(KERN_INFO "virtual kernel memory layout:\n"
791 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
792#ifdef CONFIG_HIGHMEM
793 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
794#endif
795 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
796 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
797 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
798 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
799 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
800 FIXADDR_START, FIXADDR_TOP,
801 (FIXADDR_TOP - FIXADDR_START) >> 10,
802
803#ifdef CONFIG_HIGHMEM
804 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
805 (LAST_PKMAP*PAGE_SIZE) >> 10,
806#endif
807
808 VMALLOC_START, VMALLOC_END,
809 (VMALLOC_END - VMALLOC_START) >> 20,
810
811 (unsigned long)__va(0), (unsigned long)high_memory,
812 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
813
814 (unsigned long)&__init_begin, (unsigned long)&__init_end,
815 ((unsigned long)&__init_end -
816 (unsigned long)&__init_begin) >> 10,
817
818 (unsigned long)&_etext, (unsigned long)&_edata,
819 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
820
821 (unsigned long)&_text, (unsigned long)&_etext,
822 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
823
824 /*
825 * Check boundaries twice: Some fundamental inconsistencies can
826 * be detected at build time already.
827 */
828#define __FIXADDR_TOP (-PAGE_SIZE)
829#ifdef CONFIG_HIGHMEM
830 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
831 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
832#endif
833#define high_memory (-128UL << 20)
834 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
835#undef high_memory
836#undef __FIXADDR_TOP
837
838#ifdef CONFIG_HIGHMEM
839 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
840 BUG_ON(VMALLOC_END > PKMAP_BASE);
841#endif
842 BUG_ON(VMALLOC_START >= VMALLOC_END);
843 BUG_ON((unsigned long)high_memory > VMALLOC_START);
844
845 if (boot_cpu_data.wp_works_ok < 0)
846 test_wp_bit();
847}
848
849#ifdef CONFIG_MEMORY_HOTPLUG
850int arch_add_memory(int nid, u64 start, u64 size)
851{
852 struct pglist_data *pgdata = NODE_DATA(nid);
853 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
854 unsigned long start_pfn = start >> PAGE_SHIFT;
855 unsigned long nr_pages = size >> PAGE_SHIFT;
856
857 return __add_pages(nid, zone, start_pfn, nr_pages);
858}
859#endif
860
861/*
862 * This function cannot be __init, since exceptions don't work in that
863 * section. Put this after the callers, so that it cannot be inlined.
864 */
865static noinline int do_test_wp_bit(void)
866{
867 char tmp_reg;
868 int flag;
869
870 __asm__ __volatile__(
871 " movb %0, %1 \n"
872 "1: movb %1, %0 \n"
873 " xorl %2, %2 \n"
874 "2: \n"
875 _ASM_EXTABLE(1b,2b)
876 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
877 "=q" (tmp_reg),
878 "=r" (flag)
879 :"2" (1)
880 :"memory");
881
882 return flag;
883}
884
885#ifdef CONFIG_DEBUG_RODATA
886const int rodata_test_data = 0xC3;
887EXPORT_SYMBOL_GPL(rodata_test_data);
888
889int kernel_set_to_readonly __read_mostly;
890
891void set_kernel_text_rw(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 write\n",
900 start, start+size);
901
902 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
903}
904
905void set_kernel_text_ro(void)
906{
907 unsigned long start = PFN_ALIGN(_text);
908 unsigned long size = PFN_ALIGN(_etext) - start;
909
910 if (!kernel_set_to_readonly)
911 return;
912
913 pr_debug("Set kernel text: %lx - %lx for read only\n",
914 start, start+size);
915
916 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
917}
918
919static void mark_nxdata_nx(void)
920{
921 /*
922 * When this called, init has already been executed and released,
923 * so everything past _etext should be NX.
924 */
925 unsigned long start = PFN_ALIGN(_etext);
926 /*
927 * This comes from is_kernel_text upper limit. Also HPAGE where used:
928 */
929 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
930
931 if (__supported_pte_mask & _PAGE_NX)
932 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
933 set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
934}
935
936void mark_rodata_ro(void)
937{
938 unsigned long start = PFN_ALIGN(_text);
939 unsigned long size = PFN_ALIGN(_etext) - start;
940
941 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
942 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
943 size >> 10);
944
945 kernel_set_to_readonly = 1;
946
947#ifdef CONFIG_CPA_DEBUG
948 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
949 start, start+size);
950 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
951
952 printk(KERN_INFO "Testing CPA: write protecting again\n");
953 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
954#endif
955
956 start += size;
957 size = (unsigned long)__end_rodata - start;
958 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
959 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
960 size >> 10);
961 rodata_test();
962
963#ifdef CONFIG_CPA_DEBUG
964 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
965 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
966
967 printk(KERN_INFO "Testing CPA: write protecting again\n");
968 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
969#endif
970 mark_nxdata_nx();
971}
972#endif
973
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 *
6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
7 */
8
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/memblock.h>
28#include <linux/proc_fs.h>
29#include <linux/memory_hotplug.h>
30#include <linux/initrd.h>
31#include <linux/cpumask.h>
32#include <linux/gfp.h>
33
34#include <asm/asm.h>
35#include <asm/bios_ebda.h>
36#include <asm/processor.h>
37#include <linux/uaccess.h>
38#include <asm/pgtable.h>
39#include <asm/dma.h>
40#include <asm/fixmap.h>
41#include <asm/e820/api.h>
42#include <asm/apic.h>
43#include <asm/bugs.h>
44#include <asm/tlb.h>
45#include <asm/tlbflush.h>
46#include <asm/olpc_ofw.h>
47#include <asm/pgalloc.h>
48#include <asm/sections.h>
49#include <asm/paravirt.h>
50#include <asm/setup.h>
51#include <asm/set_memory.h>
52#include <asm/page_types.h>
53#include <asm/cpu_entry_area.h>
54#include <asm/init.h>
55
56#include "mm_internal.h"
57
58unsigned long highstart_pfn, highend_pfn;
59
60bool __read_mostly __vmalloc_start_set = false;
61
62/*
63 * Creates a middle page table and puts a pointer to it in the
64 * given global directory entry. This only returns the gd entry
65 * in non-PAE compilation mode, since the middle layer is folded.
66 */
67static pmd_t * __init one_md_table_init(pgd_t *pgd)
68{
69 p4d_t *p4d;
70 pud_t *pud;
71 pmd_t *pmd_table;
72
73#ifdef CONFIG_X86_PAE
74 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
75 pmd_table = (pmd_t *)alloc_low_page();
76 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
77 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
78 p4d = p4d_offset(pgd, 0);
79 pud = pud_offset(p4d, 0);
80 BUG_ON(pmd_table != pmd_offset(pud, 0));
81
82 return pmd_table;
83 }
84#endif
85 p4d = p4d_offset(pgd, 0);
86 pud = pud_offset(p4d, 0);
87 pmd_table = pmd_offset(pud, 0);
88
89 return pmd_table;
90}
91
92/*
93 * Create a page table and place a pointer to it in a middle page
94 * directory entry:
95 */
96static pte_t * __init one_page_table_init(pmd_t *pmd)
97{
98 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
99 pte_t *page_table = (pte_t *)alloc_low_page();
100
101 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
102 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
103 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
104 }
105
106 return pte_offset_kernel(pmd, 0);
107}
108
109pmd_t * __init populate_extra_pmd(unsigned long vaddr)
110{
111 int pgd_idx = pgd_index(vaddr);
112 int pmd_idx = pmd_index(vaddr);
113
114 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
115}
116
117pte_t * __init populate_extra_pte(unsigned long vaddr)
118{
119 int pte_idx = pte_index(vaddr);
120 pmd_t *pmd;
121
122 pmd = populate_extra_pmd(vaddr);
123 return one_page_table_init(pmd) + pte_idx;
124}
125
126static unsigned long __init
127page_table_range_init_count(unsigned long start, unsigned long end)
128{
129 unsigned long count = 0;
130#ifdef CONFIG_HIGHMEM
131 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
132 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
133 int pgd_idx, pmd_idx;
134 unsigned long vaddr;
135
136 if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
137 return 0;
138
139 vaddr = start;
140 pgd_idx = pgd_index(vaddr);
141 pmd_idx = pmd_index(vaddr);
142
143 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
144 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
145 pmd_idx++) {
146 if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
147 (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
148 count++;
149 vaddr += PMD_SIZE;
150 }
151 pmd_idx = 0;
152 }
153#endif
154 return count;
155}
156
157static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
158 unsigned long vaddr, pte_t *lastpte,
159 void **adr)
160{
161#ifdef CONFIG_HIGHMEM
162 /*
163 * Something (early fixmap) may already have put a pte
164 * page here, which causes the page table allocation
165 * to become nonlinear. Attempt to fix it, and if it
166 * is still nonlinear then we have to bug.
167 */
168 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
169 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
170
171 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
172 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
173 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
174 pte_t *newpte;
175 int i;
176
177 BUG_ON(after_bootmem);
178 newpte = *adr;
179 for (i = 0; i < PTRS_PER_PTE; i++)
180 set_pte(newpte + i, pte[i]);
181 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
182
183 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
184 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
185 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
186 __flush_tlb_all();
187
188 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
189 pte = newpte;
190 }
191 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
192 && vaddr > fix_to_virt(FIX_KMAP_END)
193 && lastpte && lastpte + PTRS_PER_PTE != pte);
194#endif
195 return pte;
196}
197
198/*
199 * This function initializes a certain range of kernel virtual memory
200 * with new bootmem page tables, everywhere page tables are missing in
201 * the given range.
202 *
203 * NOTE: The pagetables are allocated contiguous on the physical space
204 * so we can cache the place of the first one and move around without
205 * checking the pgd every time.
206 */
207static void __init
208page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
209{
210 int pgd_idx, pmd_idx;
211 unsigned long vaddr;
212 pgd_t *pgd;
213 pmd_t *pmd;
214 pte_t *pte = NULL;
215 unsigned long count = page_table_range_init_count(start, end);
216 void *adr = NULL;
217
218 if (count)
219 adr = alloc_low_pages(count);
220
221 vaddr = start;
222 pgd_idx = pgd_index(vaddr);
223 pmd_idx = pmd_index(vaddr);
224 pgd = pgd_base + pgd_idx;
225
226 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
227 pmd = one_md_table_init(pgd);
228 pmd = pmd + pmd_index(vaddr);
229 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
230 pmd++, pmd_idx++) {
231 pte = page_table_kmap_check(one_page_table_init(pmd),
232 pmd, vaddr, pte, &adr);
233
234 vaddr += PMD_SIZE;
235 }
236 pmd_idx = 0;
237 }
238}
239
240static inline int is_kernel_text(unsigned long addr)
241{
242 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
243 return 1;
244 return 0;
245}
246
247/*
248 * This maps the physical memory to kernel virtual address space, a total
249 * of max_low_pfn pages, by creating page tables starting from address
250 * PAGE_OFFSET:
251 */
252unsigned long __init
253kernel_physical_mapping_init(unsigned long start,
254 unsigned long end,
255 unsigned long page_size_mask)
256{
257 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
258 unsigned long last_map_addr = end;
259 unsigned long start_pfn, end_pfn;
260 pgd_t *pgd_base = swapper_pg_dir;
261 int pgd_idx, pmd_idx, pte_ofs;
262 unsigned long pfn;
263 pgd_t *pgd;
264 pmd_t *pmd;
265 pte_t *pte;
266 unsigned pages_2m, pages_4k;
267 int mapping_iter;
268
269 start_pfn = start >> PAGE_SHIFT;
270 end_pfn = end >> PAGE_SHIFT;
271
272 /*
273 * First iteration will setup identity mapping using large/small pages
274 * based on use_pse, with other attributes same as set by
275 * the early code in head_32.S
276 *
277 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
278 * as desired for the kernel identity mapping.
279 *
280 * This two pass mechanism conforms to the TLB app note which says:
281 *
282 * "Software should not write to a paging-structure entry in a way
283 * that would change, for any linear address, both the page size
284 * and either the page frame or attributes."
285 */
286 mapping_iter = 1;
287
288 if (!boot_cpu_has(X86_FEATURE_PSE))
289 use_pse = 0;
290
291repeat:
292 pages_2m = pages_4k = 0;
293 pfn = start_pfn;
294 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
295 pgd = pgd_base + pgd_idx;
296 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
297 pmd = one_md_table_init(pgd);
298
299 if (pfn >= end_pfn)
300 continue;
301#ifdef CONFIG_X86_PAE
302 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
303 pmd += pmd_idx;
304#else
305 pmd_idx = 0;
306#endif
307 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
308 pmd++, pmd_idx++) {
309 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
310
311 /*
312 * Map with big pages if possible, otherwise
313 * create normal page tables:
314 */
315 if (use_pse) {
316 unsigned int addr2;
317 pgprot_t prot = PAGE_KERNEL_LARGE;
318 /*
319 * first pass will use the same initial
320 * identity mapping attribute + _PAGE_PSE.
321 */
322 pgprot_t init_prot =
323 __pgprot(PTE_IDENT_ATTR |
324 _PAGE_PSE);
325
326 pfn &= PMD_MASK >> PAGE_SHIFT;
327 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
328 PAGE_OFFSET + PAGE_SIZE-1;
329
330 if (is_kernel_text(addr) ||
331 is_kernel_text(addr2))
332 prot = PAGE_KERNEL_LARGE_EXEC;
333
334 pages_2m++;
335 if (mapping_iter == 1)
336 set_pmd(pmd, pfn_pmd(pfn, init_prot));
337 else
338 set_pmd(pmd, pfn_pmd(pfn, prot));
339
340 pfn += PTRS_PER_PTE;
341 continue;
342 }
343 pte = one_page_table_init(pmd);
344
345 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
346 pte += pte_ofs;
347 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
348 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
349 pgprot_t prot = PAGE_KERNEL;
350 /*
351 * first pass will use the same initial
352 * identity mapping attribute.
353 */
354 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
355
356 if (is_kernel_text(addr))
357 prot = PAGE_KERNEL_EXEC;
358
359 pages_4k++;
360 if (mapping_iter == 1) {
361 set_pte(pte, pfn_pte(pfn, init_prot));
362 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
363 } else
364 set_pte(pte, pfn_pte(pfn, prot));
365 }
366 }
367 }
368 if (mapping_iter == 1) {
369 /*
370 * update direct mapping page count only in the first
371 * iteration.
372 */
373 update_page_count(PG_LEVEL_2M, pages_2m);
374 update_page_count(PG_LEVEL_4K, pages_4k);
375
376 /*
377 * local global flush tlb, which will flush the previous
378 * mappings present in both small and large page TLB's.
379 */
380 __flush_tlb_all();
381
382 /*
383 * Second iteration will set the actual desired PTE attributes.
384 */
385 mapping_iter = 2;
386 goto repeat;
387 }
388 return last_map_addr;
389}
390
391pte_t *kmap_pte;
392
393static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
394{
395 pgd_t *pgd = pgd_offset_k(vaddr);
396 p4d_t *p4d = p4d_offset(pgd, vaddr);
397 pud_t *pud = pud_offset(p4d, vaddr);
398 pmd_t *pmd = pmd_offset(pud, vaddr);
399 return pte_offset_kernel(pmd, vaddr);
400}
401
402static void __init kmap_init(void)
403{
404 unsigned long kmap_vstart;
405
406 /*
407 * Cache the first kmap pte:
408 */
409 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
410 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
411}
412
413#ifdef CONFIG_HIGHMEM
414static void __init permanent_kmaps_init(pgd_t *pgd_base)
415{
416 unsigned long vaddr;
417 pgd_t *pgd;
418 p4d_t *p4d;
419 pud_t *pud;
420 pmd_t *pmd;
421 pte_t *pte;
422
423 vaddr = PKMAP_BASE;
424 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
425
426 pgd = swapper_pg_dir + pgd_index(vaddr);
427 p4d = p4d_offset(pgd, vaddr);
428 pud = pud_offset(p4d, vaddr);
429 pmd = pmd_offset(pud, vaddr);
430 pte = pte_offset_kernel(pmd, vaddr);
431 pkmap_page_table = pte;
432}
433
434void __init add_highpages_with_active_regions(int nid,
435 unsigned long start_pfn, unsigned long end_pfn)
436{
437 phys_addr_t start, end;
438 u64 i;
439
440 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
441 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
442 start_pfn, end_pfn);
443 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
444 start_pfn, end_pfn);
445 for ( ; pfn < e_pfn; pfn++)
446 if (pfn_valid(pfn))
447 free_highmem_page(pfn_to_page(pfn));
448 }
449}
450#else
451static inline void permanent_kmaps_init(pgd_t *pgd_base)
452{
453}
454#endif /* CONFIG_HIGHMEM */
455
456void __init sync_initial_page_table(void)
457{
458 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
459 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
460 KERNEL_PGD_PTRS);
461
462 /*
463 * sync back low identity map too. It is used for example
464 * in the 32-bit EFI stub.
465 */
466 clone_pgd_range(initial_page_table,
467 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
468 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
469}
470
471void __init native_pagetable_init(void)
472{
473 unsigned long pfn, va;
474 pgd_t *pgd, *base = swapper_pg_dir;
475 p4d_t *p4d;
476 pud_t *pud;
477 pmd_t *pmd;
478 pte_t *pte;
479
480 /*
481 * Remove any mappings which extend past the end of physical
482 * memory from the boot time page table.
483 * In virtual address space, we should have at least two pages
484 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
485 * definition. And max_low_pfn is set to VMALLOC_END physical
486 * address. If initial memory mapping is doing right job, we
487 * should have pte used near max_low_pfn or one pmd is not present.
488 */
489 for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
490 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
491 pgd = base + pgd_index(va);
492 if (!pgd_present(*pgd))
493 break;
494
495 p4d = p4d_offset(pgd, va);
496 pud = pud_offset(p4d, va);
497 pmd = pmd_offset(pud, va);
498 if (!pmd_present(*pmd))
499 break;
500
501 /* should not be large page here */
502 if (pmd_large(*pmd)) {
503 pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
504 pfn, pmd, __pa(pmd));
505 BUG_ON(1);
506 }
507
508 pte = pte_offset_kernel(pmd, va);
509 if (!pte_present(*pte))
510 break;
511
512 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
513 pfn, pmd, __pa(pmd), pte, __pa(pte));
514 pte_clear(NULL, va, pte);
515 }
516 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
517 paging_init();
518}
519
520/*
521 * Build a proper pagetable for the kernel mappings. Up until this
522 * point, we've been running on some set of pagetables constructed by
523 * the boot process.
524 *
525 * If we're booting on native hardware, this will be a pagetable
526 * constructed in arch/x86/kernel/head_32.S. The root of the
527 * pagetable will be swapper_pg_dir.
528 *
529 * If we're booting paravirtualized under a hypervisor, then there are
530 * more options: we may already be running PAE, and the pagetable may
531 * or may not be based in swapper_pg_dir. In any case,
532 * paravirt_pagetable_init() will set up swapper_pg_dir
533 * appropriately for the rest of the initialization to work.
534 *
535 * In general, pagetable_init() assumes that the pagetable may already
536 * be partially populated, and so it avoids stomping on any existing
537 * mappings.
538 */
539void __init early_ioremap_page_table_range_init(void)
540{
541 pgd_t *pgd_base = swapper_pg_dir;
542 unsigned long vaddr, end;
543
544 /*
545 * Fixed mappings, only the page table structure has to be
546 * created - mappings will be set by set_fixmap():
547 */
548 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
549 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
550 page_table_range_init(vaddr, end, pgd_base);
551 early_ioremap_reset();
552}
553
554static void __init pagetable_init(void)
555{
556 pgd_t *pgd_base = swapper_pg_dir;
557
558 permanent_kmaps_init(pgd_base);
559}
560
561#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
562/* Bits supported by the hardware: */
563pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
564/* Bits allowed in normal kernel mappings: */
565pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
566EXPORT_SYMBOL_GPL(__supported_pte_mask);
567/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
568EXPORT_SYMBOL(__default_kernel_pte_mask);
569
570/* user-defined highmem size */
571static unsigned int highmem_pages = -1;
572
573/*
574 * highmem=size forces highmem to be exactly 'size' bytes.
575 * This works even on boxes that have no highmem otherwise.
576 * This also works to reduce highmem size on bigger boxes.
577 */
578static int __init parse_highmem(char *arg)
579{
580 if (!arg)
581 return -EINVAL;
582
583 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
584 return 0;
585}
586early_param("highmem", parse_highmem);
587
588#define MSG_HIGHMEM_TOO_BIG \
589 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
590
591#define MSG_LOWMEM_TOO_SMALL \
592 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
593/*
594 * All of RAM fits into lowmem - but if user wants highmem
595 * artificially via the highmem=x boot parameter then create
596 * it:
597 */
598static void __init lowmem_pfn_init(void)
599{
600 /* max_low_pfn is 0, we already have early_res support */
601 max_low_pfn = max_pfn;
602
603 if (highmem_pages == -1)
604 highmem_pages = 0;
605#ifdef CONFIG_HIGHMEM
606 if (highmem_pages >= max_pfn) {
607 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
608 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
609 highmem_pages = 0;
610 }
611 if (highmem_pages) {
612 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
613 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
614 pages_to_mb(highmem_pages));
615 highmem_pages = 0;
616 }
617 max_low_pfn -= highmem_pages;
618 }
619#else
620 if (highmem_pages)
621 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
622#endif
623}
624
625#define MSG_HIGHMEM_TOO_SMALL \
626 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
627
628#define MSG_HIGHMEM_TRIMMED \
629 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
630/*
631 * We have more RAM than fits into lowmem - we try to put it into
632 * highmem, also taking the highmem=x boot parameter into account:
633 */
634static void __init highmem_pfn_init(void)
635{
636 max_low_pfn = MAXMEM_PFN;
637
638 if (highmem_pages == -1)
639 highmem_pages = max_pfn - MAXMEM_PFN;
640
641 if (highmem_pages + MAXMEM_PFN < max_pfn)
642 max_pfn = MAXMEM_PFN + highmem_pages;
643
644 if (highmem_pages + MAXMEM_PFN > max_pfn) {
645 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
646 pages_to_mb(max_pfn - MAXMEM_PFN),
647 pages_to_mb(highmem_pages));
648 highmem_pages = 0;
649 }
650#ifndef CONFIG_HIGHMEM
651 /* Maximum memory usable is what is directly addressable */
652 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
653 if (max_pfn > MAX_NONPAE_PFN)
654 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
655 else
656 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
657 max_pfn = MAXMEM_PFN;
658#else /* !CONFIG_HIGHMEM */
659#ifndef CONFIG_HIGHMEM64G
660 if (max_pfn > MAX_NONPAE_PFN) {
661 max_pfn = MAX_NONPAE_PFN;
662 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
663 }
664#endif /* !CONFIG_HIGHMEM64G */
665#endif /* !CONFIG_HIGHMEM */
666}
667
668/*
669 * Determine low and high memory ranges:
670 */
671void __init find_low_pfn_range(void)
672{
673 /* it could update max_pfn */
674
675 if (max_pfn <= MAXMEM_PFN)
676 lowmem_pfn_init();
677 else
678 highmem_pfn_init();
679}
680
681#ifndef CONFIG_NEED_MULTIPLE_NODES
682void __init initmem_init(void)
683{
684#ifdef CONFIG_HIGHMEM
685 highstart_pfn = highend_pfn = max_pfn;
686 if (max_pfn > max_low_pfn)
687 highstart_pfn = max_low_pfn;
688 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
689 pages_to_mb(highend_pfn - highstart_pfn));
690 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
691#else
692 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
693#endif
694
695 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
696 sparse_memory_present_with_active_regions(0);
697
698#ifdef CONFIG_FLATMEM
699 max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
700#endif
701 __vmalloc_start_set = true;
702
703 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
704 pages_to_mb(max_low_pfn));
705
706 setup_bootmem_allocator();
707}
708#endif /* !CONFIG_NEED_MULTIPLE_NODES */
709
710void __init setup_bootmem_allocator(void)
711{
712 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
713 max_pfn_mapped<<PAGE_SHIFT);
714 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
715}
716
717/*
718 * paging_init() sets up the page tables - note that the first 8MB are
719 * already mapped by head.S.
720 *
721 * This routines also unmaps the page at virtual kernel address 0, so
722 * that we can trap those pesky NULL-reference errors in the kernel.
723 */
724void __init paging_init(void)
725{
726 pagetable_init();
727
728 __flush_tlb_all();
729
730 kmap_init();
731
732 /*
733 * NOTE: at this point the bootmem allocator is fully available.
734 */
735 olpc_dt_build_devicetree();
736 sparse_memory_present_with_active_regions(MAX_NUMNODES);
737 sparse_init();
738 zone_sizes_init();
739}
740
741/*
742 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
743 * and also on some strange 486's. All 586+'s are OK. This used to involve
744 * black magic jumps to work around some nasty CPU bugs, but fortunately the
745 * switch to using exceptions got rid of all that.
746 */
747static void __init test_wp_bit(void)
748{
749 char z = 0;
750
751 printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
752
753 __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
754
755 if (probe_kernel_write((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
756 clear_fixmap(FIX_WP_TEST);
757 printk(KERN_CONT "Ok.\n");
758 return;
759 }
760
761 printk(KERN_CONT "No.\n");
762 panic("Linux doesn't support CPUs with broken WP.");
763}
764
765void __init mem_init(void)
766{
767 pci_iommu_alloc();
768
769#ifdef CONFIG_FLATMEM
770 BUG_ON(!mem_map);
771#endif
772 /*
773 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
774 * be done before memblock_free_all(). Memblock use free low memory for
775 * temporary data (see find_range_array()) and for this purpose can use
776 * pages that was already passed to the buddy allocator, hence marked as
777 * not accessible in the page tables when compiled with
778 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
779 * important here.
780 */
781 set_highmem_pages_init();
782
783 /* this will put all low memory onto the freelists */
784 memblock_free_all();
785
786 after_bootmem = 1;
787 x86_init.hyper.init_after_bootmem();
788
789 mem_init_print_info(NULL);
790 printk(KERN_INFO "virtual kernel memory layout:\n"
791 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
792 " cpu_entry : 0x%08lx - 0x%08lx (%4ld kB)\n"
793#ifdef CONFIG_HIGHMEM
794 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
795#endif
796 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
797 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
798 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
799 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
800 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
801 FIXADDR_START, FIXADDR_TOP,
802 (FIXADDR_TOP - FIXADDR_START) >> 10,
803
804 CPU_ENTRY_AREA_BASE,
805 CPU_ENTRY_AREA_BASE + CPU_ENTRY_AREA_MAP_SIZE,
806 CPU_ENTRY_AREA_MAP_SIZE >> 10,
807
808#ifdef CONFIG_HIGHMEM
809 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
810 (LAST_PKMAP*PAGE_SIZE) >> 10,
811#endif
812
813 VMALLOC_START, VMALLOC_END,
814 (VMALLOC_END - VMALLOC_START) >> 20,
815
816 (unsigned long)__va(0), (unsigned long)high_memory,
817 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
818
819 (unsigned long)&__init_begin, (unsigned long)&__init_end,
820 ((unsigned long)&__init_end -
821 (unsigned long)&__init_begin) >> 10,
822
823 (unsigned long)&_etext, (unsigned long)&_edata,
824 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
825
826 (unsigned long)&_text, (unsigned long)&_etext,
827 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
828
829 /*
830 * Check boundaries twice: Some fundamental inconsistencies can
831 * be detected at build time already.
832 */
833#define __FIXADDR_TOP (-PAGE_SIZE)
834#ifdef CONFIG_HIGHMEM
835 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
836 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
837#endif
838#define high_memory (-128UL << 20)
839 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
840#undef high_memory
841#undef __FIXADDR_TOP
842
843#ifdef CONFIG_HIGHMEM
844 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
845 BUG_ON(VMALLOC_END > PKMAP_BASE);
846#endif
847 BUG_ON(VMALLOC_START >= VMALLOC_END);
848 BUG_ON((unsigned long)high_memory > VMALLOC_START);
849
850 test_wp_bit();
851}
852
853#ifdef CONFIG_MEMORY_HOTPLUG
854int arch_add_memory(int nid, u64 start, u64 size,
855 struct mhp_restrictions *restrictions)
856{
857 unsigned long start_pfn = start >> PAGE_SHIFT;
858 unsigned long nr_pages = size >> PAGE_SHIFT;
859
860 return __add_pages(nid, start_pfn, nr_pages, restrictions);
861}
862
863void arch_remove_memory(int nid, u64 start, u64 size,
864 struct vmem_altmap *altmap)
865{
866 unsigned long start_pfn = start >> PAGE_SHIFT;
867 unsigned long nr_pages = size >> PAGE_SHIFT;
868 struct zone *zone;
869
870 zone = page_zone(pfn_to_page(start_pfn));
871 __remove_pages(zone, start_pfn, nr_pages, altmap);
872}
873#endif
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_memory_nx(start, size >> PAGE_SHIFT);
920}
921
922void mark_rodata_ro(void)
923{
924 unsigned long start = PFN_ALIGN(_text);
925 unsigned long size = (unsigned long)__end_rodata - start;
926
927 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
928 pr_info("Write protecting kernel text and read-only data: %luk\n",
929 size >> 10);
930
931 kernel_set_to_readonly = 1;
932
933#ifdef CONFIG_CPA_DEBUG
934 pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
935 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
936
937 pr_info("Testing CPA: write protecting again\n");
938 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
939#endif
940 mark_nxdata_nx();
941 if (__supported_pte_mask & _PAGE_NX)
942 debug_checkwx();
943}