Loading...
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/dma.h>
39#include <asm/fixmap.h>
40#include <asm/e820/api.h>
41#include <asm/apic.h>
42#include <asm/bugs.h>
43#include <asm/tlb.h>
44#include <asm/tlbflush.h>
45#include <asm/olpc_ofw.h>
46#include <asm/pgalloc.h>
47#include <asm/sections.h>
48#include <asm/paravirt.h>
49#include <asm/setup.h>
50#include <asm/set_memory.h>
51#include <asm/page_types.h>
52#include <asm/cpu_entry_area.h>
53#include <asm/init.h>
54#include <asm/pgtable_areas.h>
55#include <asm/numa.h>
56
57#include "mm_internal.h"
58
59unsigned long highstart_pfn, highend_pfn;
60
61bool __read_mostly __vmalloc_start_set = false;
62
63/*
64 * Creates a middle page table and puts a pointer to it in the
65 * given global directory entry. This only returns the gd entry
66 * in non-PAE compilation mode, since the middle layer is folded.
67 */
68static pmd_t * __init one_md_table_init(pgd_t *pgd)
69{
70 p4d_t *p4d;
71 pud_t *pud;
72 pmd_t *pmd_table;
73
74#ifdef CONFIG_X86_PAE
75 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
76 pmd_table = (pmd_t *)alloc_low_page();
77 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
78 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
79 p4d = p4d_offset(pgd, 0);
80 pud = pud_offset(p4d, 0);
81 BUG_ON(pmd_table != pmd_offset(pud, 0));
82
83 return pmd_table;
84 }
85#endif
86 p4d = p4d_offset(pgd, 0);
87 pud = pud_offset(p4d, 0);
88 pmd_table = pmd_offset(pud, 0);
89
90 return pmd_table;
91}
92
93/*
94 * Create a page table and place a pointer to it in a middle page
95 * directory entry:
96 */
97static pte_t * __init one_page_table_init(pmd_t *pmd)
98{
99 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
100 pte_t *page_table = (pte_t *)alloc_low_page();
101
102 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
103 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
104 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
105 }
106
107 return pte_offset_kernel(pmd, 0);
108}
109
110pmd_t * __init populate_extra_pmd(unsigned long vaddr)
111{
112 int pgd_idx = pgd_index(vaddr);
113 int pmd_idx = pmd_index(vaddr);
114
115 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
116}
117
118pte_t * __init populate_extra_pte(unsigned long vaddr)
119{
120 int pte_idx = pte_index(vaddr);
121 pmd_t *pmd;
122
123 pmd = populate_extra_pmd(vaddr);
124 return one_page_table_init(pmd) + pte_idx;
125}
126
127static unsigned long __init
128page_table_range_init_count(unsigned long start, unsigned long end)
129{
130 unsigned long count = 0;
131#ifdef CONFIG_HIGHMEM
132 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
133 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
134 int pgd_idx, pmd_idx;
135 unsigned long vaddr;
136
137 if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
138 return 0;
139
140 vaddr = start;
141 pgd_idx = pgd_index(vaddr);
142 pmd_idx = pmd_index(vaddr);
143
144 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
145 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
146 pmd_idx++) {
147 if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
148 (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
149 count++;
150 vaddr += PMD_SIZE;
151 }
152 pmd_idx = 0;
153 }
154#endif
155 return count;
156}
157
158static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
159 unsigned long vaddr, pte_t *lastpte,
160 void **adr)
161{
162#ifdef CONFIG_HIGHMEM
163 /*
164 * Something (early fixmap) may already have put a pte
165 * page here, which causes the page table allocation
166 * to become nonlinear. Attempt to fix it, and if it
167 * is still nonlinear then we have to bug.
168 */
169 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
170 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
171
172 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
173 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
174 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
175 pte_t *newpte;
176 int i;
177
178 BUG_ON(after_bootmem);
179 newpte = *adr;
180 for (i = 0; i < PTRS_PER_PTE; i++)
181 set_pte(newpte + i, pte[i]);
182 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
183
184 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
185 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
186 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
187 __flush_tlb_all();
188
189 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
190 pte = newpte;
191 }
192 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
193 && vaddr > fix_to_virt(FIX_KMAP_END)
194 && lastpte && lastpte + PTRS_PER_PTE != pte);
195#endif
196 return pte;
197}
198
199/*
200 * This function initializes a certain range of kernel virtual memory
201 * with new bootmem page tables, everywhere page tables are missing in
202 * the given range.
203 *
204 * NOTE: The pagetables are allocated contiguous on the physical space
205 * so we can cache the place of the first one and move around without
206 * checking the pgd every time.
207 */
208static void __init
209page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
210{
211 int pgd_idx, pmd_idx;
212 unsigned long vaddr;
213 pgd_t *pgd;
214 pmd_t *pmd;
215 pte_t *pte = NULL;
216 unsigned long count = page_table_range_init_count(start, end);
217 void *adr = NULL;
218
219 if (count)
220 adr = alloc_low_pages(count);
221
222 vaddr = start;
223 pgd_idx = pgd_index(vaddr);
224 pmd_idx = pmd_index(vaddr);
225 pgd = pgd_base + pgd_idx;
226
227 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
228 pmd = one_md_table_init(pgd);
229 pmd = pmd + pmd_index(vaddr);
230 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
231 pmd++, pmd_idx++) {
232 pte = page_table_kmap_check(one_page_table_init(pmd),
233 pmd, vaddr, pte, &adr);
234
235 vaddr += PMD_SIZE;
236 }
237 pmd_idx = 0;
238 }
239}
240
241static inline int is_x86_32_kernel_text(unsigned long addr)
242{
243 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
244 return 1;
245 return 0;
246}
247
248/*
249 * This maps the physical memory to kernel virtual address space, a total
250 * of max_low_pfn pages, by creating page tables starting from address
251 * PAGE_OFFSET:
252 */
253unsigned long __init
254kernel_physical_mapping_init(unsigned long start,
255 unsigned long end,
256 unsigned long page_size_mask,
257 pgprot_t prot)
258{
259 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
260 unsigned long last_map_addr = end;
261 unsigned long start_pfn, end_pfn;
262 pgd_t *pgd_base = swapper_pg_dir;
263 int pgd_idx, pmd_idx, pte_ofs;
264 unsigned long pfn;
265 pgd_t *pgd;
266 pmd_t *pmd;
267 pte_t *pte;
268 unsigned pages_2m, pages_4k;
269 int mapping_iter;
270
271 start_pfn = start >> PAGE_SHIFT;
272 end_pfn = end >> PAGE_SHIFT;
273
274 /*
275 * First iteration will setup identity mapping using large/small pages
276 * based on use_pse, with other attributes same as set by
277 * the early code in head_32.S
278 *
279 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
280 * as desired for the kernel identity mapping.
281 *
282 * This two pass mechanism conforms to the TLB app note which says:
283 *
284 * "Software should not write to a paging-structure entry in a way
285 * that would change, for any linear address, both the page size
286 * and either the page frame or attributes."
287 */
288 mapping_iter = 1;
289
290 if (!boot_cpu_has(X86_FEATURE_PSE))
291 use_pse = 0;
292
293repeat:
294 pages_2m = pages_4k = 0;
295 pfn = start_pfn;
296 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
297 pgd = pgd_base + pgd_idx;
298 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
299 pmd = one_md_table_init(pgd);
300
301 if (pfn >= end_pfn)
302 continue;
303#ifdef CONFIG_X86_PAE
304 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
305 pmd += pmd_idx;
306#else
307 pmd_idx = 0;
308#endif
309 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
310 pmd++, pmd_idx++) {
311 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
312
313 /*
314 * Map with big pages if possible, otherwise
315 * create normal page tables:
316 */
317 if (use_pse) {
318 unsigned int addr2;
319 pgprot_t prot = PAGE_KERNEL_LARGE;
320 /*
321 * first pass will use the same initial
322 * identity mapping attribute + _PAGE_PSE.
323 */
324 pgprot_t init_prot =
325 __pgprot(PTE_IDENT_ATTR |
326 _PAGE_PSE);
327
328 pfn &= PMD_MASK >> PAGE_SHIFT;
329 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
330 PAGE_OFFSET + PAGE_SIZE-1;
331
332 if (is_x86_32_kernel_text(addr) ||
333 is_x86_32_kernel_text(addr2))
334 prot = PAGE_KERNEL_LARGE_EXEC;
335
336 pages_2m++;
337 if (mapping_iter == 1)
338 set_pmd(pmd, pfn_pmd(pfn, init_prot));
339 else
340 set_pmd(pmd, pfn_pmd(pfn, prot));
341
342 pfn += PTRS_PER_PTE;
343 continue;
344 }
345 pte = one_page_table_init(pmd);
346
347 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
348 pte += pte_ofs;
349 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
350 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
351 pgprot_t prot = PAGE_KERNEL;
352 /*
353 * first pass will use the same initial
354 * identity mapping attribute.
355 */
356 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
357
358 if (is_x86_32_kernel_text(addr))
359 prot = PAGE_KERNEL_EXEC;
360
361 pages_4k++;
362 if (mapping_iter == 1) {
363 set_pte(pte, pfn_pte(pfn, init_prot));
364 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
365 } else
366 set_pte(pte, pfn_pte(pfn, prot));
367 }
368 }
369 }
370 if (mapping_iter == 1) {
371 /*
372 * update direct mapping page count only in the first
373 * iteration.
374 */
375 update_page_count(PG_LEVEL_2M, pages_2m);
376 update_page_count(PG_LEVEL_4K, pages_4k);
377
378 /*
379 * local global flush tlb, which will flush the previous
380 * mappings present in both small and large page TLB's.
381 */
382 __flush_tlb_all();
383
384 /*
385 * Second iteration will set the actual desired PTE attributes.
386 */
387 mapping_iter = 2;
388 goto repeat;
389 }
390 return last_map_addr;
391}
392
393#ifdef CONFIG_HIGHMEM
394static void __init permanent_kmaps_init(pgd_t *pgd_base)
395{
396 unsigned long vaddr = PKMAP_BASE;
397
398 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
399
400 pkmap_page_table = virt_to_kpte(vaddr);
401}
402
403void __init add_highpages_with_active_regions(int nid,
404 unsigned long start_pfn, unsigned long end_pfn)
405{
406 phys_addr_t start, end;
407 u64 i;
408
409 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
410 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
411 start_pfn, end_pfn);
412 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
413 start_pfn, end_pfn);
414 for ( ; pfn < e_pfn; pfn++)
415 if (pfn_valid(pfn))
416 free_highmem_page(pfn_to_page(pfn));
417 }
418}
419#else
420static inline void permanent_kmaps_init(pgd_t *pgd_base)
421{
422}
423#endif /* CONFIG_HIGHMEM */
424
425void __init sync_initial_page_table(void)
426{
427 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
428 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
429 KERNEL_PGD_PTRS);
430
431 /*
432 * sync back low identity map too. It is used for example
433 * in the 32-bit EFI stub.
434 */
435 clone_pgd_range(initial_page_table,
436 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
437 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
438}
439
440void __init native_pagetable_init(void)
441{
442 unsigned long pfn, va;
443 pgd_t *pgd, *base = swapper_pg_dir;
444 p4d_t *p4d;
445 pud_t *pud;
446 pmd_t *pmd;
447 pte_t *pte;
448
449 /*
450 * Remove any mappings which extend past the end of physical
451 * memory from the boot time page table.
452 * In virtual address space, we should have at least two pages
453 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
454 * definition. And max_low_pfn is set to VMALLOC_END physical
455 * address. If initial memory mapping is doing right job, we
456 * should have pte used near max_low_pfn or one pmd is not present.
457 */
458 for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
459 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
460 pgd = base + pgd_index(va);
461 if (!pgd_present(*pgd))
462 break;
463
464 p4d = p4d_offset(pgd, va);
465 pud = pud_offset(p4d, va);
466 pmd = pmd_offset(pud, va);
467 if (!pmd_present(*pmd))
468 break;
469
470 /* should not be large page here */
471 if (pmd_large(*pmd)) {
472 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",
473 pfn, pmd, __pa(pmd));
474 BUG_ON(1);
475 }
476
477 pte = pte_offset_kernel(pmd, va);
478 if (!pte_present(*pte))
479 break;
480
481 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
482 pfn, pmd, __pa(pmd), pte, __pa(pte));
483 pte_clear(NULL, va, pte);
484 }
485 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
486 paging_init();
487}
488
489/*
490 * Build a proper pagetable for the kernel mappings. Up until this
491 * point, we've been running on some set of pagetables constructed by
492 * the boot process.
493 *
494 * If we're booting on native hardware, this will be a pagetable
495 * constructed in arch/x86/kernel/head_32.S. The root of the
496 * pagetable will be swapper_pg_dir.
497 *
498 * If we're booting paravirtualized under a hypervisor, then there are
499 * more options: we may already be running PAE, and the pagetable may
500 * or may not be based in swapper_pg_dir. In any case,
501 * paravirt_pagetable_init() will set up swapper_pg_dir
502 * appropriately for the rest of the initialization to work.
503 *
504 * In general, pagetable_init() assumes that the pagetable may already
505 * be partially populated, and so it avoids stomping on any existing
506 * mappings.
507 */
508void __init early_ioremap_page_table_range_init(void)
509{
510 pgd_t *pgd_base = swapper_pg_dir;
511 unsigned long vaddr, end;
512
513 /*
514 * Fixed mappings, only the page table structure has to be
515 * created - mappings will be set by set_fixmap():
516 */
517 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
518 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
519 page_table_range_init(vaddr, end, pgd_base);
520 early_ioremap_reset();
521}
522
523static void __init pagetable_init(void)
524{
525 pgd_t *pgd_base = swapper_pg_dir;
526
527 permanent_kmaps_init(pgd_base);
528}
529
530#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
531/* Bits supported by the hardware: */
532pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
533/* Bits allowed in normal kernel mappings: */
534pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
535EXPORT_SYMBOL_GPL(__supported_pte_mask);
536/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
537EXPORT_SYMBOL(__default_kernel_pte_mask);
538
539/* user-defined highmem size */
540static unsigned int highmem_pages = -1;
541
542/*
543 * highmem=size forces highmem to be exactly 'size' bytes.
544 * This works even on boxes that have no highmem otherwise.
545 * This also works to reduce highmem size on bigger boxes.
546 */
547static int __init parse_highmem(char *arg)
548{
549 if (!arg)
550 return -EINVAL;
551
552 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
553 return 0;
554}
555early_param("highmem", parse_highmem);
556
557#define MSG_HIGHMEM_TOO_BIG \
558 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
559
560#define MSG_LOWMEM_TOO_SMALL \
561 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
562/*
563 * All of RAM fits into lowmem - but if user wants highmem
564 * artificially via the highmem=x boot parameter then create
565 * it:
566 */
567static void __init lowmem_pfn_init(void)
568{
569 /* max_low_pfn is 0, we already have early_res support */
570 max_low_pfn = max_pfn;
571
572 if (highmem_pages == -1)
573 highmem_pages = 0;
574#ifdef CONFIG_HIGHMEM
575 if (highmem_pages >= max_pfn) {
576 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
577 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
578 highmem_pages = 0;
579 }
580 if (highmem_pages) {
581 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
582 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
583 pages_to_mb(highmem_pages));
584 highmem_pages = 0;
585 }
586 max_low_pfn -= highmem_pages;
587 }
588#else
589 if (highmem_pages)
590 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
591#endif
592}
593
594#define MSG_HIGHMEM_TOO_SMALL \
595 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
596
597#define MSG_HIGHMEM_TRIMMED \
598 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
599/*
600 * We have more RAM than fits into lowmem - we try to put it into
601 * highmem, also taking the highmem=x boot parameter into account:
602 */
603static void __init highmem_pfn_init(void)
604{
605 max_low_pfn = MAXMEM_PFN;
606
607 if (highmem_pages == -1)
608 highmem_pages = max_pfn - MAXMEM_PFN;
609
610 if (highmem_pages + MAXMEM_PFN < max_pfn)
611 max_pfn = MAXMEM_PFN + highmem_pages;
612
613 if (highmem_pages + MAXMEM_PFN > max_pfn) {
614 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
615 pages_to_mb(max_pfn - MAXMEM_PFN),
616 pages_to_mb(highmem_pages));
617 highmem_pages = 0;
618 }
619#ifndef CONFIG_HIGHMEM
620 /* Maximum memory usable is what is directly addressable */
621 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
622 if (max_pfn > MAX_NONPAE_PFN)
623 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
624 else
625 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
626 max_pfn = MAXMEM_PFN;
627#else /* !CONFIG_HIGHMEM */
628#ifndef CONFIG_HIGHMEM64G
629 if (max_pfn > MAX_NONPAE_PFN) {
630 max_pfn = MAX_NONPAE_PFN;
631 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
632 }
633#endif /* !CONFIG_HIGHMEM64G */
634#endif /* !CONFIG_HIGHMEM */
635}
636
637/*
638 * Determine low and high memory ranges:
639 */
640void __init find_low_pfn_range(void)
641{
642 /* it could update max_pfn */
643
644 if (max_pfn <= MAXMEM_PFN)
645 lowmem_pfn_init();
646 else
647 highmem_pfn_init();
648}
649
650#ifndef CONFIG_NUMA
651void __init initmem_init(void)
652{
653#ifdef CONFIG_HIGHMEM
654 highstart_pfn = highend_pfn = max_pfn;
655 if (max_pfn > max_low_pfn)
656 highstart_pfn = max_low_pfn;
657 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
658 pages_to_mb(highend_pfn - highstart_pfn));
659 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
660#else
661 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
662#endif
663
664 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
665
666#ifdef CONFIG_FLATMEM
667 max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
668#endif
669 __vmalloc_start_set = true;
670
671 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
672 pages_to_mb(max_low_pfn));
673
674 setup_bootmem_allocator();
675}
676#endif /* !CONFIG_NUMA */
677
678void __init setup_bootmem_allocator(void)
679{
680 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
681 max_pfn_mapped<<PAGE_SHIFT);
682 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
683}
684
685/*
686 * paging_init() sets up the page tables - note that the first 8MB are
687 * already mapped by head.S.
688 *
689 * This routines also unmaps the page at virtual kernel address 0, so
690 * that we can trap those pesky NULL-reference errors in the kernel.
691 */
692void __init paging_init(void)
693{
694 pagetable_init();
695
696 __flush_tlb_all();
697
698 /*
699 * NOTE: at this point the bootmem allocator is fully available.
700 */
701 olpc_dt_build_devicetree();
702 sparse_init();
703 zone_sizes_init();
704}
705
706/*
707 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
708 * and also on some strange 486's. All 586+'s are OK. This used to involve
709 * black magic jumps to work around some nasty CPU bugs, but fortunately the
710 * switch to using exceptions got rid of all that.
711 */
712static void __init test_wp_bit(void)
713{
714 char z = 0;
715
716 printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
717
718 __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
719
720 if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
721 clear_fixmap(FIX_WP_TEST);
722 printk(KERN_CONT "Ok.\n");
723 return;
724 }
725
726 printk(KERN_CONT "No.\n");
727 panic("Linux doesn't support CPUs with broken WP.");
728}
729
730void __init mem_init(void)
731{
732 pci_iommu_alloc();
733
734#ifdef CONFIG_FLATMEM
735 BUG_ON(!mem_map);
736#endif
737 /*
738 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
739 * be done before memblock_free_all(). Memblock use free low memory for
740 * temporary data (see find_range_array()) and for this purpose can use
741 * pages that was already passed to the buddy allocator, hence marked as
742 * not accessible in the page tables when compiled with
743 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
744 * important here.
745 */
746 set_highmem_pages_init();
747
748 /* this will put all low memory onto the freelists */
749 memblock_free_all();
750
751 after_bootmem = 1;
752 x86_init.hyper.init_after_bootmem();
753
754 /*
755 * Check boundaries twice: Some fundamental inconsistencies can
756 * be detected at build time already.
757 */
758#define __FIXADDR_TOP (-PAGE_SIZE)
759#ifdef CONFIG_HIGHMEM
760 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
761 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
762#endif
763#define high_memory (-128UL << 20)
764 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
765#undef high_memory
766#undef __FIXADDR_TOP
767
768#ifdef CONFIG_HIGHMEM
769 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
770 BUG_ON(VMALLOC_END > PKMAP_BASE);
771#endif
772 BUG_ON(VMALLOC_START >= VMALLOC_END);
773 BUG_ON((unsigned long)high_memory > VMALLOC_START);
774
775 test_wp_bit();
776}
777
778int kernel_set_to_readonly __read_mostly;
779
780static void mark_nxdata_nx(void)
781{
782 /*
783 * When this called, init has already been executed and released,
784 * so everything past _etext should be NX.
785 */
786 unsigned long start = PFN_ALIGN(_etext);
787 /*
788 * This comes from is_x86_32_kernel_text upper limit. Also HPAGE where used:
789 */
790 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
791
792 if (__supported_pte_mask & _PAGE_NX)
793 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
794 set_memory_nx(start, size >> PAGE_SHIFT);
795}
796
797void mark_rodata_ro(void)
798{
799 unsigned long start = PFN_ALIGN(_text);
800 unsigned long size = (unsigned long)__end_rodata - start;
801
802 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
803 pr_info("Write protecting kernel text and read-only data: %luk\n",
804 size >> 10);
805
806 kernel_set_to_readonly = 1;
807
808#ifdef CONFIG_CPA_DEBUG
809 pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
810 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
811
812 pr_info("Testing CPA: write protecting again\n");
813 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
814#endif
815 mark_nxdata_nx();
816 if (__supported_pte_mask & _PAGE_NX)
817 debug_checkwx();
818}
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
56#include "mm_internal.h"
57
58unsigned long highstart_pfn, highend_pfn;
59
60static noinline int do_test_wp_bit(void);
61
62bool __read_mostly __vmalloc_start_set = false;
63
64/*
65 * Creates a middle page table and puts a pointer to it in the
66 * given global directory entry. This only returns the gd entry
67 * in non-PAE compilation mode, since the middle layer is folded.
68 */
69static pmd_t * __init one_md_table_init(pgd_t *pgd)
70{
71 pud_t *pud;
72 pmd_t *pmd_table;
73
74#ifdef CONFIG_X86_PAE
75 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
76 pmd_table = (pmd_t *)alloc_low_page();
77 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
78 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
79 pud = pud_offset(pgd, 0);
80 BUG_ON(pmd_table != pmd_offset(pud, 0));
81
82 return pmd_table;
83 }
84#endif
85 pud = pud_offset(pgd, 0);
86 pmd_table = pmd_offset(pud, 0);
87
88 return pmd_table;
89}
90
91/*
92 * Create a page table and place a pointer to it in a middle page
93 * directory entry:
94 */
95static pte_t * __init one_page_table_init(pmd_t *pmd)
96{
97 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
98 pte_t *page_table = (pte_t *)alloc_low_page();
99
100 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
101 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
102 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
103 }
104
105 return pte_offset_kernel(pmd, 0);
106}
107
108pmd_t * __init populate_extra_pmd(unsigned long vaddr)
109{
110 int pgd_idx = pgd_index(vaddr);
111 int pmd_idx = pmd_index(vaddr);
112
113 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
114}
115
116pte_t * __init populate_extra_pte(unsigned long vaddr)
117{
118 int pte_idx = pte_index(vaddr);
119 pmd_t *pmd;
120
121 pmd = populate_extra_pmd(vaddr);
122 return one_page_table_init(pmd) + pte_idx;
123}
124
125static unsigned long __init
126page_table_range_init_count(unsigned long start, unsigned long end)
127{
128 unsigned long count = 0;
129#ifdef CONFIG_HIGHMEM
130 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
131 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
132 int pgd_idx, pmd_idx;
133 unsigned long vaddr;
134
135 if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
136 return 0;
137
138 vaddr = start;
139 pgd_idx = pgd_index(vaddr);
140 pmd_idx = pmd_index(vaddr);
141
142 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
143 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
144 pmd_idx++) {
145 if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
146 (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
147 count++;
148 vaddr += PMD_SIZE;
149 }
150 pmd_idx = 0;
151 }
152#endif
153 return count;
154}
155
156static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
157 unsigned long vaddr, pte_t *lastpte,
158 void **adr)
159{
160#ifdef CONFIG_HIGHMEM
161 /*
162 * Something (early fixmap) may already have put a pte
163 * page here, which causes the page table allocation
164 * to become nonlinear. Attempt to fix it, and if it
165 * is still nonlinear then we have to bug.
166 */
167 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
168 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
169
170 if (pmd_idx_kmap_begin != pmd_idx_kmap_end
171 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
172 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
173 pte_t *newpte;
174 int i;
175
176 BUG_ON(after_bootmem);
177 newpte = *adr;
178 for (i = 0; i < PTRS_PER_PTE; i++)
179 set_pte(newpte + i, pte[i]);
180 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
181
182 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
183 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
184 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
185 __flush_tlb_all();
186
187 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
188 pte = newpte;
189 }
190 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
191 && vaddr > fix_to_virt(FIX_KMAP_END)
192 && lastpte && lastpte + PTRS_PER_PTE != pte);
193#endif
194 return pte;
195}
196
197/*
198 * This function initializes a certain range of kernel virtual memory
199 * with new bootmem page tables, everywhere page tables are missing in
200 * the given range.
201 *
202 * NOTE: The pagetables are allocated contiguous on the physical space
203 * so we can cache the place of the first one and move around without
204 * checking the pgd every time.
205 */
206static void __init
207page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
208{
209 int pgd_idx, pmd_idx;
210 unsigned long vaddr;
211 pgd_t *pgd;
212 pmd_t *pmd;
213 pte_t *pte = NULL;
214 unsigned long count = page_table_range_init_count(start, end);
215 void *adr = NULL;
216
217 if (count)
218 adr = alloc_low_pages(count);
219
220 vaddr = start;
221 pgd_idx = pgd_index(vaddr);
222 pmd_idx = pmd_index(vaddr);
223 pgd = pgd_base + pgd_idx;
224
225 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
226 pmd = one_md_table_init(pgd);
227 pmd = pmd + pmd_index(vaddr);
228 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
229 pmd++, pmd_idx++) {
230 pte = page_table_kmap_check(one_page_table_init(pmd),
231 pmd, vaddr, pte, &adr);
232
233 vaddr += PMD_SIZE;
234 }
235 pmd_idx = 0;
236 }
237}
238
239static inline int is_kernel_text(unsigned long addr)
240{
241 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
242 return 1;
243 return 0;
244}
245
246/*
247 * This maps the physical memory to kernel virtual address space, a total
248 * of max_low_pfn pages, by creating page tables starting from address
249 * PAGE_OFFSET:
250 */
251unsigned long __init
252kernel_physical_mapping_init(unsigned long start,
253 unsigned long end,
254 unsigned long page_size_mask)
255{
256 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
257 unsigned long last_map_addr = end;
258 unsigned long start_pfn, end_pfn;
259 pgd_t *pgd_base = swapper_pg_dir;
260 int pgd_idx, pmd_idx, pte_ofs;
261 unsigned long pfn;
262 pgd_t *pgd;
263 pmd_t *pmd;
264 pte_t *pte;
265 unsigned pages_2m, pages_4k;
266 int mapping_iter;
267
268 start_pfn = start >> PAGE_SHIFT;
269 end_pfn = end >> PAGE_SHIFT;
270
271 /*
272 * First iteration will setup identity mapping using large/small pages
273 * based on use_pse, with other attributes same as set by
274 * the early code in head_32.S
275 *
276 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
277 * as desired for the kernel identity mapping.
278 *
279 * This two pass mechanism conforms to the TLB app note which says:
280 *
281 * "Software should not write to a paging-structure entry in a way
282 * that would change, for any linear address, both the page size
283 * and either the page frame or attributes."
284 */
285 mapping_iter = 1;
286
287 if (!cpu_has_pse)
288 use_pse = 0;
289
290repeat:
291 pages_2m = pages_4k = 0;
292 pfn = start_pfn;
293 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
294 pgd = pgd_base + pgd_idx;
295 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
296 pmd = one_md_table_init(pgd);
297
298 if (pfn >= end_pfn)
299 continue;
300#ifdef CONFIG_X86_PAE
301 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
302 pmd += pmd_idx;
303#else
304 pmd_idx = 0;
305#endif
306 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
307 pmd++, pmd_idx++) {
308 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
309
310 /*
311 * Map with big pages if possible, otherwise
312 * create normal page tables:
313 */
314 if (use_pse) {
315 unsigned int addr2;
316 pgprot_t prot = PAGE_KERNEL_LARGE;
317 /*
318 * first pass will use the same initial
319 * identity mapping attribute + _PAGE_PSE.
320 */
321 pgprot_t init_prot =
322 __pgprot(PTE_IDENT_ATTR |
323 _PAGE_PSE);
324
325 pfn &= PMD_MASK >> PAGE_SHIFT;
326 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
327 PAGE_OFFSET + PAGE_SIZE-1;
328
329 if (is_kernel_text(addr) ||
330 is_kernel_text(addr2))
331 prot = PAGE_KERNEL_LARGE_EXEC;
332
333 pages_2m++;
334 if (mapping_iter == 1)
335 set_pmd(pmd, pfn_pmd(pfn, init_prot));
336 else
337 set_pmd(pmd, pfn_pmd(pfn, prot));
338
339 pfn += PTRS_PER_PTE;
340 continue;
341 }
342 pte = one_page_table_init(pmd);
343
344 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
345 pte += pte_ofs;
346 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
347 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
348 pgprot_t prot = PAGE_KERNEL;
349 /*
350 * first pass will use the same initial
351 * identity mapping attribute.
352 */
353 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
354
355 if (is_kernel_text(addr))
356 prot = PAGE_KERNEL_EXEC;
357
358 pages_4k++;
359 if (mapping_iter == 1) {
360 set_pte(pte, pfn_pte(pfn, init_prot));
361 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
362 } else
363 set_pte(pte, pfn_pte(pfn, prot));
364 }
365 }
366 }
367 if (mapping_iter == 1) {
368 /*
369 * update direct mapping page count only in the first
370 * iteration.
371 */
372 update_page_count(PG_LEVEL_2M, pages_2m);
373 update_page_count(PG_LEVEL_4K, pages_4k);
374
375 /*
376 * local global flush tlb, which will flush the previous
377 * mappings present in both small and large page TLB's.
378 */
379 __flush_tlb_all();
380
381 /*
382 * Second iteration will set the actual desired PTE attributes.
383 */
384 mapping_iter = 2;
385 goto repeat;
386 }
387 return last_map_addr;
388}
389
390pte_t *kmap_pte;
391
392static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
393{
394 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
395 vaddr), vaddr), vaddr);
396}
397
398static void __init kmap_init(void)
399{
400 unsigned long kmap_vstart;
401
402 /*
403 * Cache the first kmap pte:
404 */
405 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
406 kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
407}
408
409#ifdef CONFIG_HIGHMEM
410static void __init permanent_kmaps_init(pgd_t *pgd_base)
411{
412 unsigned long vaddr;
413 pgd_t *pgd;
414 pud_t *pud;
415 pmd_t *pmd;
416 pte_t *pte;
417
418 vaddr = PKMAP_BASE;
419 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
420
421 pgd = swapper_pg_dir + pgd_index(vaddr);
422 pud = pud_offset(pgd, vaddr);
423 pmd = pmd_offset(pud, vaddr);
424 pte = pte_offset_kernel(pmd, vaddr);
425 pkmap_page_table = pte;
426}
427
428void __init add_highpages_with_active_regions(int nid,
429 unsigned long start_pfn, unsigned long end_pfn)
430{
431 phys_addr_t start, end;
432 u64 i;
433
434 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
435 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
436 start_pfn, end_pfn);
437 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
438 start_pfn, end_pfn);
439 for ( ; pfn < e_pfn; pfn++)
440 if (pfn_valid(pfn))
441 free_highmem_page(pfn_to_page(pfn));
442 }
443}
444#else
445static inline void permanent_kmaps_init(pgd_t *pgd_base)
446{
447}
448#endif /* CONFIG_HIGHMEM */
449
450void __init native_pagetable_init(void)
451{
452 unsigned long pfn, va;
453 pgd_t *pgd, *base = swapper_pg_dir;
454 pud_t *pud;
455 pmd_t *pmd;
456 pte_t *pte;
457
458 /*
459 * Remove any mappings which extend past the end of physical
460 * memory from the boot time page table.
461 * In virtual address space, we should have at least two pages
462 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
463 * definition. And max_low_pfn is set to VMALLOC_END physical
464 * address. If initial memory mapping is doing right job, we
465 * should have pte used near max_low_pfn or one pmd is not present.
466 */
467 for (pfn = max_low_pfn; 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 /* should not be large page here */
479 if (pmd_large(*pmd)) {
480 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",
481 pfn, pmd, __pa(pmd));
482 BUG_ON(1);
483 }
484
485 pte = pte_offset_kernel(pmd, va);
486 if (!pte_present(*pte))
487 break;
488
489 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
490 pfn, pmd, __pa(pmd), pte, __pa(pte));
491 pte_clear(NULL, va, pte);
492 }
493 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
494 paging_init();
495}
496
497/*
498 * Build a proper pagetable for the kernel mappings. Up until this
499 * point, we've been running on some set of pagetables constructed by
500 * the boot process.
501 *
502 * If we're booting on native hardware, this will be a pagetable
503 * constructed in arch/x86/kernel/head_32.S. The root of the
504 * pagetable will be swapper_pg_dir.
505 *
506 * If we're booting paravirtualized under a hypervisor, then there are
507 * more options: we may already be running PAE, and the pagetable may
508 * or may not be based in swapper_pg_dir. In any case,
509 * paravirt_pagetable_init() will set up swapper_pg_dir
510 * appropriately for the rest of the initialization to work.
511 *
512 * In general, pagetable_init() assumes that the pagetable may already
513 * be partially populated, and so it avoids stomping on any existing
514 * mappings.
515 */
516void __init early_ioremap_page_table_range_init(void)
517{
518 pgd_t *pgd_base = swapper_pg_dir;
519 unsigned long vaddr, end;
520
521 /*
522 * Fixed mappings, only the page table structure has to be
523 * created - mappings will be set by set_fixmap():
524 */
525 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
526 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
527 page_table_range_init(vaddr, end, pgd_base);
528 early_ioremap_reset();
529}
530
531static void __init pagetable_init(void)
532{
533 pgd_t *pgd_base = swapper_pg_dir;
534
535 permanent_kmaps_init(pgd_base);
536}
537
538pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
539EXPORT_SYMBOL_GPL(__supported_pte_mask);
540
541/* user-defined highmem size */
542static unsigned int highmem_pages = -1;
543
544/*
545 * highmem=size forces highmem to be exactly 'size' bytes.
546 * This works even on boxes that have no highmem otherwise.
547 * This also works to reduce highmem size on bigger boxes.
548 */
549static int __init parse_highmem(char *arg)
550{
551 if (!arg)
552 return -EINVAL;
553
554 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
555 return 0;
556}
557early_param("highmem", parse_highmem);
558
559#define MSG_HIGHMEM_TOO_BIG \
560 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
561
562#define MSG_LOWMEM_TOO_SMALL \
563 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
564/*
565 * All of RAM fits into lowmem - but if user wants highmem
566 * artificially via the highmem=x boot parameter then create
567 * it:
568 */
569static void __init lowmem_pfn_init(void)
570{
571 /* max_low_pfn is 0, we already have early_res support */
572 max_low_pfn = max_pfn;
573
574 if (highmem_pages == -1)
575 highmem_pages = 0;
576#ifdef CONFIG_HIGHMEM
577 if (highmem_pages >= max_pfn) {
578 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
579 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
580 highmem_pages = 0;
581 }
582 if (highmem_pages) {
583 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
584 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
585 pages_to_mb(highmem_pages));
586 highmem_pages = 0;
587 }
588 max_low_pfn -= highmem_pages;
589 }
590#else
591 if (highmem_pages)
592 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
593#endif
594}
595
596#define MSG_HIGHMEM_TOO_SMALL \
597 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
598
599#define MSG_HIGHMEM_TRIMMED \
600 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
601/*
602 * We have more RAM than fits into lowmem - we try to put it into
603 * highmem, also taking the highmem=x boot parameter into account:
604 */
605static void __init highmem_pfn_init(void)
606{
607 max_low_pfn = MAXMEM_PFN;
608
609 if (highmem_pages == -1)
610 highmem_pages = max_pfn - MAXMEM_PFN;
611
612 if (highmem_pages + MAXMEM_PFN < max_pfn)
613 max_pfn = MAXMEM_PFN + highmem_pages;
614
615 if (highmem_pages + MAXMEM_PFN > max_pfn) {
616 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
617 pages_to_mb(max_pfn - MAXMEM_PFN),
618 pages_to_mb(highmem_pages));
619 highmem_pages = 0;
620 }
621#ifndef CONFIG_HIGHMEM
622 /* Maximum memory usable is what is directly addressable */
623 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
624 if (max_pfn > MAX_NONPAE_PFN)
625 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
626 else
627 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
628 max_pfn = MAXMEM_PFN;
629#else /* !CONFIG_HIGHMEM */
630#ifndef CONFIG_HIGHMEM64G
631 if (max_pfn > MAX_NONPAE_PFN) {
632 max_pfn = MAX_NONPAE_PFN;
633 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
634 }
635#endif /* !CONFIG_HIGHMEM64G */
636#endif /* !CONFIG_HIGHMEM */
637}
638
639/*
640 * Determine low and high memory ranges:
641 */
642void __init find_low_pfn_range(void)
643{
644 /* it could update max_pfn */
645
646 if (max_pfn <= MAXMEM_PFN)
647 lowmem_pfn_init();
648 else
649 highmem_pfn_init();
650}
651
652#ifndef CONFIG_NEED_MULTIPLE_NODES
653void __init initmem_init(void)
654{
655#ifdef CONFIG_HIGHMEM
656 highstart_pfn = highend_pfn = max_pfn;
657 if (max_pfn > max_low_pfn)
658 highstart_pfn = max_low_pfn;
659 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
660 pages_to_mb(highend_pfn - highstart_pfn));
661 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
662#else
663 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
664#endif
665
666 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
667 sparse_memory_present_with_active_regions(0);
668
669#ifdef CONFIG_FLATMEM
670 max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
671#endif
672 __vmalloc_start_set = true;
673
674 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
675 pages_to_mb(max_low_pfn));
676
677 setup_bootmem_allocator();
678}
679#endif /* !CONFIG_NEED_MULTIPLE_NODES */
680
681void __init setup_bootmem_allocator(void)
682{
683 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
684 max_pfn_mapped<<PAGE_SHIFT);
685 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
686}
687
688/*
689 * paging_init() sets up the page tables - note that the first 8MB are
690 * already mapped by head.S.
691 *
692 * This routines also unmaps the page at virtual kernel address 0, so
693 * that we can trap those pesky NULL-reference errors in the kernel.
694 */
695void __init paging_init(void)
696{
697 pagetable_init();
698
699 __flush_tlb_all();
700
701 kmap_init();
702
703 /*
704 * NOTE: at this point the bootmem allocator is fully available.
705 */
706 olpc_dt_build_devicetree();
707 sparse_memory_present_with_active_regions(MAX_NUMNODES);
708 sparse_init();
709 zone_sizes_init();
710}
711
712/*
713 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
714 * and also on some strange 486's. All 586+'s are OK. This used to involve
715 * black magic jumps to work around some nasty CPU bugs, but fortunately the
716 * switch to using exceptions got rid of all that.
717 */
718static void __init test_wp_bit(void)
719{
720 printk(KERN_INFO
721 "Checking if this processor honours the WP bit even in supervisor mode...");
722
723 /* Any page-aligned address will do, the test is non-destructive */
724 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_KERNEL_RO);
725 boot_cpu_data.wp_works_ok = do_test_wp_bit();
726 clear_fixmap(FIX_WP_TEST);
727
728 if (!boot_cpu_data.wp_works_ok) {
729 printk(KERN_CONT "No.\n");
730 panic("Linux doesn't support CPUs with broken WP.");
731 } else {
732 printk(KERN_CONT "Ok.\n");
733 }
734}
735
736void __init mem_init(void)
737{
738 pci_iommu_alloc();
739
740#ifdef CONFIG_FLATMEM
741 BUG_ON(!mem_map);
742#endif
743 /*
744 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
745 * be done before free_all_bootmem(). Memblock use free low memory for
746 * temporary data (see find_range_array()) and for this purpose can use
747 * pages that was already passed to the buddy allocator, hence marked as
748 * not accessible in the page tables when compiled with
749 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
750 * important here.
751 */
752 set_highmem_pages_init();
753
754 /* this will put all low memory onto the freelists */
755 free_all_bootmem();
756
757 after_bootmem = 1;
758
759 mem_init_print_info(NULL);
760 printk(KERN_INFO "virtual kernel memory layout:\n"
761 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
762#ifdef CONFIG_HIGHMEM
763 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
764#endif
765 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
766 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
767 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n"
768 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n"
769 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n",
770 FIXADDR_START, FIXADDR_TOP,
771 (FIXADDR_TOP - FIXADDR_START) >> 10,
772
773#ifdef CONFIG_HIGHMEM
774 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
775 (LAST_PKMAP*PAGE_SIZE) >> 10,
776#endif
777
778 VMALLOC_START, VMALLOC_END,
779 (VMALLOC_END - VMALLOC_START) >> 20,
780
781 (unsigned long)__va(0), (unsigned long)high_memory,
782 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
783
784 (unsigned long)&__init_begin, (unsigned long)&__init_end,
785 ((unsigned long)&__init_end -
786 (unsigned long)&__init_begin) >> 10,
787
788 (unsigned long)&_etext, (unsigned long)&_edata,
789 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
790
791 (unsigned long)&_text, (unsigned long)&_etext,
792 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
793
794 /*
795 * Check boundaries twice: Some fundamental inconsistencies can
796 * be detected at build time already.
797 */
798#define __FIXADDR_TOP (-PAGE_SIZE)
799#ifdef CONFIG_HIGHMEM
800 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
801 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
802#endif
803#define high_memory (-128UL << 20)
804 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
805#undef high_memory
806#undef __FIXADDR_TOP
807#ifdef CONFIG_RANDOMIZE_BASE
808 BUILD_BUG_ON(CONFIG_RANDOMIZE_BASE_MAX_OFFSET > KERNEL_IMAGE_SIZE);
809#endif
810
811#ifdef CONFIG_HIGHMEM
812 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
813 BUG_ON(VMALLOC_END > PKMAP_BASE);
814#endif
815 BUG_ON(VMALLOC_START >= VMALLOC_END);
816 BUG_ON((unsigned long)high_memory > VMALLOC_START);
817
818 if (boot_cpu_data.wp_works_ok < 0)
819 test_wp_bit();
820}
821
822#ifdef CONFIG_MEMORY_HOTPLUG
823int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
824{
825 struct pglist_data *pgdata = NODE_DATA(nid);
826 struct zone *zone = pgdata->node_zones +
827 zone_for_memory(nid, start, size, ZONE_HIGHMEM, for_device);
828 unsigned long start_pfn = start >> PAGE_SHIFT;
829 unsigned long nr_pages = size >> PAGE_SHIFT;
830
831 return __add_pages(nid, zone, start_pfn, nr_pages);
832}
833
834#ifdef CONFIG_MEMORY_HOTREMOVE
835int arch_remove_memory(u64 start, u64 size)
836{
837 unsigned long start_pfn = start >> PAGE_SHIFT;
838 unsigned long nr_pages = size >> PAGE_SHIFT;
839 struct zone *zone;
840
841 zone = page_zone(pfn_to_page(start_pfn));
842 return __remove_pages(zone, start_pfn, nr_pages);
843}
844#endif
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
871const int rodata_test_data = 0xC3;
872EXPORT_SYMBOL_GPL(rodata_test_data);
873
874int kernel_set_to_readonly __read_mostly;
875
876void set_kernel_text_rw(void)
877{
878 unsigned long start = PFN_ALIGN(_text);
879 unsigned long size = PFN_ALIGN(_etext) - start;
880
881 if (!kernel_set_to_readonly)
882 return;
883
884 pr_debug("Set kernel text: %lx - %lx for read write\n",
885 start, start+size);
886
887 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
888}
889
890void set_kernel_text_ro(void)
891{
892 unsigned long start = PFN_ALIGN(_text);
893 unsigned long size = PFN_ALIGN(_etext) - start;
894
895 if (!kernel_set_to_readonly)
896 return;
897
898 pr_debug("Set kernel text: %lx - %lx for read only\n",
899 start, start+size);
900
901 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
902}
903
904static void mark_nxdata_nx(void)
905{
906 /*
907 * When this called, init has already been executed and released,
908 * so everything past _etext should be NX.
909 */
910 unsigned long start = PFN_ALIGN(_etext);
911 /*
912 * This comes from is_kernel_text upper limit. Also HPAGE where used:
913 */
914 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
915
916 if (__supported_pte_mask & _PAGE_NX)
917 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
918 set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
919}
920
921void mark_rodata_ro(void)
922{
923 unsigned long start = PFN_ALIGN(_text);
924 unsigned long size = PFN_ALIGN(_etext) - start;
925
926 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
927 printk(KERN_INFO "Write protecting the kernel text: %luk\n",
928 size >> 10);
929
930 kernel_set_to_readonly = 1;
931
932#ifdef CONFIG_CPA_DEBUG
933 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
934 start, start+size);
935 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
936
937 printk(KERN_INFO "Testing CPA: write protecting again\n");
938 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
939#endif
940
941 start += size;
942 size = (unsigned long)__end_rodata - start;
943 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
944 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
945 size >> 10);
946 rodata_test();
947
948#ifdef CONFIG_CPA_DEBUG
949 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", 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 mark_nxdata_nx();
956 if (__supported_pte_mask & _PAGE_NX)
957 debug_checkwx();
958}