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/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}
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}