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
59 unsigned long highstart_pfn, highend_pfn;
60
61 bool __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 */
one_md_table_init(pgd_t * pgd)68 static 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 */
one_page_table_init(pmd_t * pmd)97 static 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
populate_extra_pmd(unsigned long vaddr)110 pmd_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
populate_extra_pte(unsigned long vaddr)118 pte_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
127 static unsigned long __init
page_table_range_init_count(unsigned long start,unsigned long end)128 page_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
page_table_kmap_check(pte_t * pte,pmd_t * pmd,unsigned long vaddr,pte_t * lastpte,void ** adr)158 static 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 */
208 static void __init
page_table_range_init(unsigned long start,unsigned long end,pgd_t * pgd_base)209 page_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
241 /*
242 * The <linux/kallsyms.h> already defines is_kernel_text,
243 * using '__' prefix not to get in conflict.
244 */
__is_kernel_text(unsigned long addr)245 static inline int __is_kernel_text(unsigned long addr)
246 {
247 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
248 return 1;
249 return 0;
250 }
251
252 /*
253 * This maps the physical memory to kernel virtual address space, a total
254 * of max_low_pfn pages, by creating page tables starting from address
255 * PAGE_OFFSET:
256 */
257 unsigned long __init
kernel_physical_mapping_init(unsigned long start,unsigned long end,unsigned long page_size_mask,pgprot_t prot)258 kernel_physical_mapping_init(unsigned long start,
259 unsigned long end,
260 unsigned long page_size_mask,
261 pgprot_t prot)
262 {
263 int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
264 unsigned long last_map_addr = end;
265 unsigned long start_pfn, end_pfn;
266 pgd_t *pgd_base = swapper_pg_dir;
267 int pgd_idx, pmd_idx, pte_ofs;
268 unsigned long pfn;
269 pgd_t *pgd;
270 pmd_t *pmd;
271 pte_t *pte;
272 unsigned pages_2m, pages_4k;
273 int mapping_iter;
274
275 start_pfn = start >> PAGE_SHIFT;
276 end_pfn = end >> PAGE_SHIFT;
277
278 /*
279 * First iteration will setup identity mapping using large/small pages
280 * based on use_pse, with other attributes same as set by
281 * the early code in head_32.S
282 *
283 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
284 * as desired for the kernel identity mapping.
285 *
286 * This two pass mechanism conforms to the TLB app note which says:
287 *
288 * "Software should not write to a paging-structure entry in a way
289 * that would change, for any linear address, both the page size
290 * and either the page frame or attributes."
291 */
292 mapping_iter = 1;
293
294 if (!boot_cpu_has(X86_FEATURE_PSE))
295 use_pse = 0;
296
297 repeat:
298 pages_2m = pages_4k = 0;
299 pfn = start_pfn;
300 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
301 pgd = pgd_base + pgd_idx;
302 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
303 pmd = one_md_table_init(pgd);
304
305 if (pfn >= end_pfn)
306 continue;
307 #ifdef CONFIG_X86_PAE
308 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
309 pmd += pmd_idx;
310 #else
311 pmd_idx = 0;
312 #endif
313 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
314 pmd++, pmd_idx++) {
315 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
316
317 /*
318 * Map with big pages if possible, otherwise
319 * create normal page tables:
320 */
321 if (use_pse) {
322 unsigned int addr2;
323 pgprot_t prot = PAGE_KERNEL_LARGE;
324 /*
325 * first pass will use the same initial
326 * identity mapping attribute + _PAGE_PSE.
327 */
328 pgprot_t init_prot =
329 __pgprot(PTE_IDENT_ATTR |
330 _PAGE_PSE);
331
332 pfn &= PMD_MASK >> PAGE_SHIFT;
333 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
334 PAGE_OFFSET + PAGE_SIZE-1;
335
336 if (__is_kernel_text(addr) ||
337 __is_kernel_text(addr2))
338 prot = PAGE_KERNEL_LARGE_EXEC;
339
340 pages_2m++;
341 if (mapping_iter == 1)
342 set_pmd(pmd, pfn_pmd(pfn, init_prot));
343 else
344 set_pmd(pmd, pfn_pmd(pfn, prot));
345
346 pfn += PTRS_PER_PTE;
347 continue;
348 }
349 pte = one_page_table_init(pmd);
350
351 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
352 pte += pte_ofs;
353 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
354 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
355 pgprot_t prot = PAGE_KERNEL;
356 /*
357 * first pass will use the same initial
358 * identity mapping attribute.
359 */
360 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
361
362 if (__is_kernel_text(addr))
363 prot = PAGE_KERNEL_EXEC;
364
365 pages_4k++;
366 if (mapping_iter == 1) {
367 set_pte(pte, pfn_pte(pfn, init_prot));
368 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
369 } else
370 set_pte(pte, pfn_pte(pfn, prot));
371 }
372 }
373 }
374 if (mapping_iter == 1) {
375 /*
376 * update direct mapping page count only in the first
377 * iteration.
378 */
379 update_page_count(PG_LEVEL_2M, pages_2m);
380 update_page_count(PG_LEVEL_4K, pages_4k);
381
382 /*
383 * local global flush tlb, which will flush the previous
384 * mappings present in both small and large page TLB's.
385 */
386 __flush_tlb_all();
387
388 /*
389 * Second iteration will set the actual desired PTE attributes.
390 */
391 mapping_iter = 2;
392 goto repeat;
393 }
394 return last_map_addr;
395 }
396
397 #ifdef CONFIG_HIGHMEM
permanent_kmaps_init(pgd_t * pgd_base)398 static void __init permanent_kmaps_init(pgd_t *pgd_base)
399 {
400 unsigned long vaddr = PKMAP_BASE;
401
402 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
403
404 pkmap_page_table = virt_to_kpte(vaddr);
405 }
406
add_highpages_with_active_regions(int nid,unsigned long start_pfn,unsigned long end_pfn)407 void __init add_highpages_with_active_regions(int nid,
408 unsigned long start_pfn, unsigned long end_pfn)
409 {
410 phys_addr_t start, end;
411 u64 i;
412
413 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
414 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
415 start_pfn, end_pfn);
416 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
417 start_pfn, end_pfn);
418 for ( ; pfn < e_pfn; pfn++)
419 if (pfn_valid(pfn))
420 free_highmem_page(pfn_to_page(pfn));
421 }
422 }
423 #else
permanent_kmaps_init(pgd_t * pgd_base)424 static inline void permanent_kmaps_init(pgd_t *pgd_base)
425 {
426 }
427 #endif /* CONFIG_HIGHMEM */
428
sync_initial_page_table(void)429 void __init sync_initial_page_table(void)
430 {
431 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
432 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
433 KERNEL_PGD_PTRS);
434
435 /*
436 * sync back low identity map too. It is used for example
437 * in the 32-bit EFI stub.
438 */
439 clone_pgd_range(initial_page_table,
440 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
441 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
442 }
443
native_pagetable_init(void)444 void __init native_pagetable_init(void)
445 {
446 unsigned long pfn, va;
447 pgd_t *pgd, *base = swapper_pg_dir;
448 p4d_t *p4d;
449 pud_t *pud;
450 pmd_t *pmd;
451 pte_t *pte;
452
453 /*
454 * Remove any mappings which extend past the end of physical
455 * memory from the boot time page table.
456 * In virtual address space, we should have at least two pages
457 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
458 * definition. And max_low_pfn is set to VMALLOC_END physical
459 * address. If initial memory mapping is doing right job, we
460 * should have pte used near max_low_pfn or one pmd is not present.
461 */
462 for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
463 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
464 pgd = base + pgd_index(va);
465 if (!pgd_present(*pgd))
466 break;
467
468 p4d = p4d_offset(pgd, va);
469 pud = pud_offset(p4d, va);
470 pmd = pmd_offset(pud, va);
471 if (!pmd_present(*pmd))
472 break;
473
474 /* should not be large page here */
475 if (pmd_large(*pmd)) {
476 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",
477 pfn, pmd, __pa(pmd));
478 BUG_ON(1);
479 }
480
481 pte = pte_offset_kernel(pmd, va);
482 if (!pte_present(*pte))
483 break;
484
485 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
486 pfn, pmd, __pa(pmd), pte, __pa(pte));
487 pte_clear(NULL, va, pte);
488 }
489 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
490 paging_init();
491 }
492
493 /*
494 * Build a proper pagetable for the kernel mappings. Up until this
495 * point, we've been running on some set of pagetables constructed by
496 * the boot process.
497 *
498 * If we're booting on native hardware, this will be a pagetable
499 * constructed in arch/x86/kernel/head_32.S. The root of the
500 * pagetable will be swapper_pg_dir.
501 *
502 * If we're booting paravirtualized under a hypervisor, then there are
503 * more options: we may already be running PAE, and the pagetable may
504 * or may not be based in swapper_pg_dir. In any case,
505 * paravirt_pagetable_init() will set up swapper_pg_dir
506 * appropriately for the rest of the initialization to work.
507 *
508 * In general, pagetable_init() assumes that the pagetable may already
509 * be partially populated, and so it avoids stomping on any existing
510 * mappings.
511 */
early_ioremap_page_table_range_init(void)512 void __init early_ioremap_page_table_range_init(void)
513 {
514 pgd_t *pgd_base = swapper_pg_dir;
515 unsigned long vaddr, end;
516
517 /*
518 * Fixed mappings, only the page table structure has to be
519 * created - mappings will be set by set_fixmap():
520 */
521 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
522 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
523 page_table_range_init(vaddr, end, pgd_base);
524 early_ioremap_reset();
525 }
526
pagetable_init(void)527 static void __init pagetable_init(void)
528 {
529 pgd_t *pgd_base = swapper_pg_dir;
530
531 permanent_kmaps_init(pgd_base);
532 }
533
534 #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
535 /* Bits supported by the hardware: */
536 pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
537 /* Bits allowed in normal kernel mappings: */
538 pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
539 EXPORT_SYMBOL_GPL(__supported_pte_mask);
540 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
541 EXPORT_SYMBOL(__default_kernel_pte_mask);
542
543 /* user-defined highmem size */
544 static unsigned int highmem_pages = -1;
545
546 /*
547 * highmem=size forces highmem to be exactly 'size' bytes.
548 * This works even on boxes that have no highmem otherwise.
549 * This also works to reduce highmem size on bigger boxes.
550 */
parse_highmem(char * arg)551 static int __init parse_highmem(char *arg)
552 {
553 if (!arg)
554 return -EINVAL;
555
556 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
557 return 0;
558 }
559 early_param("highmem", parse_highmem);
560
561 #define MSG_HIGHMEM_TOO_BIG \
562 "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
563
564 #define MSG_LOWMEM_TOO_SMALL \
565 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
566 /*
567 * All of RAM fits into lowmem - but if user wants highmem
568 * artificially via the highmem=x boot parameter then create
569 * it:
570 */
lowmem_pfn_init(void)571 static void __init lowmem_pfn_init(void)
572 {
573 /* max_low_pfn is 0, we already have early_res support */
574 max_low_pfn = max_pfn;
575
576 if (highmem_pages == -1)
577 highmem_pages = 0;
578 #ifdef CONFIG_HIGHMEM
579 if (highmem_pages >= max_pfn) {
580 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
581 pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
582 highmem_pages = 0;
583 }
584 if (highmem_pages) {
585 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
586 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
587 pages_to_mb(highmem_pages));
588 highmem_pages = 0;
589 }
590 max_low_pfn -= highmem_pages;
591 }
592 #else
593 if (highmem_pages)
594 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
595 #endif
596 }
597
598 #define MSG_HIGHMEM_TOO_SMALL \
599 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
600
601 #define MSG_HIGHMEM_TRIMMED \
602 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
603 /*
604 * We have more RAM than fits into lowmem - we try to put it into
605 * highmem, also taking the highmem=x boot parameter into account:
606 */
highmem_pfn_init(void)607 static void __init highmem_pfn_init(void)
608 {
609 max_low_pfn = MAXMEM_PFN;
610
611 if (highmem_pages == -1)
612 highmem_pages = max_pfn - MAXMEM_PFN;
613
614 if (highmem_pages + MAXMEM_PFN < max_pfn)
615 max_pfn = MAXMEM_PFN + highmem_pages;
616
617 if (highmem_pages + MAXMEM_PFN > max_pfn) {
618 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
619 pages_to_mb(max_pfn - MAXMEM_PFN),
620 pages_to_mb(highmem_pages));
621 highmem_pages = 0;
622 }
623 #ifndef CONFIG_HIGHMEM
624 /* Maximum memory usable is what is directly addressable */
625 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
626 if (max_pfn > MAX_NONPAE_PFN)
627 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
628 else
629 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
630 max_pfn = MAXMEM_PFN;
631 #else /* !CONFIG_HIGHMEM */
632 #ifndef CONFIG_HIGHMEM64G
633 if (max_pfn > MAX_NONPAE_PFN) {
634 max_pfn = MAX_NONPAE_PFN;
635 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
636 }
637 #endif /* !CONFIG_HIGHMEM64G */
638 #endif /* !CONFIG_HIGHMEM */
639 }
640
641 /*
642 * Determine low and high memory ranges:
643 */
find_low_pfn_range(void)644 void __init find_low_pfn_range(void)
645 {
646 /* it could update max_pfn */
647
648 if (max_pfn <= MAXMEM_PFN)
649 lowmem_pfn_init();
650 else
651 highmem_pfn_init();
652 }
653
654 #ifndef CONFIG_NUMA
initmem_init(void)655 void __init initmem_init(void)
656 {
657 #ifdef CONFIG_HIGHMEM
658 highstart_pfn = highend_pfn = max_pfn;
659 if (max_pfn > max_low_pfn)
660 highstart_pfn = max_low_pfn;
661 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
662 pages_to_mb(highend_pfn - highstart_pfn));
663 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
664 #else
665 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
666 #endif
667
668 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
669
670 #ifdef CONFIG_FLATMEM
671 max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
672 #endif
673 __vmalloc_start_set = true;
674
675 printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
676 pages_to_mb(max_low_pfn));
677
678 setup_bootmem_allocator();
679 }
680 #endif /* !CONFIG_NUMA */
681
setup_bootmem_allocator(void)682 void __init setup_bootmem_allocator(void)
683 {
684 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
685 max_pfn_mapped<<PAGE_SHIFT);
686 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
687 }
688
689 /*
690 * paging_init() sets up the page tables - note that the first 8MB are
691 * already mapped by head.S.
692 *
693 * This routines also unmaps the page at virtual kernel address 0, so
694 * that we can trap those pesky NULL-reference errors in the kernel.
695 */
paging_init(void)696 void __init paging_init(void)
697 {
698 pagetable_init();
699
700 __flush_tlb_all();
701
702 /*
703 * NOTE: at this point the bootmem allocator is fully available.
704 */
705 olpc_dt_build_devicetree();
706 sparse_init();
707 zone_sizes_init();
708 }
709
710 /*
711 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
712 * and also on some strange 486's. All 586+'s are OK. This used to involve
713 * black magic jumps to work around some nasty CPU bugs, but fortunately the
714 * switch to using exceptions got rid of all that.
715 */
test_wp_bit(void)716 static void __init test_wp_bit(void)
717 {
718 char z = 0;
719
720 printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
721
722 __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
723
724 if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
725 clear_fixmap(FIX_WP_TEST);
726 printk(KERN_CONT "Ok.\n");
727 return;
728 }
729
730 printk(KERN_CONT "No.\n");
731 panic("Linux doesn't support CPUs with broken WP.");
732 }
733
mem_init(void)734 void __init mem_init(void)
735 {
736 pci_iommu_alloc();
737
738 #ifdef CONFIG_FLATMEM
739 BUG_ON(!mem_map);
740 #endif
741 /*
742 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
743 * be done before memblock_free_all(). Memblock use free low memory for
744 * temporary data (see find_range_array()) and for this purpose can use
745 * pages that was already passed to the buddy allocator, hence marked as
746 * not accessible in the page tables when compiled with
747 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
748 * important here.
749 */
750 set_highmem_pages_init();
751
752 /* this will put all low memory onto the freelists */
753 memblock_free_all();
754
755 after_bootmem = 1;
756 x86_init.hyper.init_after_bootmem();
757
758 /*
759 * Check boundaries twice: Some fundamental inconsistencies can
760 * be detected at build time already.
761 */
762 #define __FIXADDR_TOP (-PAGE_SIZE)
763 #ifdef CONFIG_HIGHMEM
764 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
765 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE);
766 #endif
767 #define high_memory (-128UL << 20)
768 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END);
769 #undef high_memory
770 #undef __FIXADDR_TOP
771
772 #ifdef CONFIG_HIGHMEM
773 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
774 BUG_ON(VMALLOC_END > PKMAP_BASE);
775 #endif
776 BUG_ON(VMALLOC_START >= VMALLOC_END);
777 BUG_ON((unsigned long)high_memory > VMALLOC_START);
778
779 test_wp_bit();
780 }
781
782 #ifdef CONFIG_MEMORY_HOTPLUG
arch_add_memory(int nid,u64 start,u64 size,struct mhp_params * params)783 int arch_add_memory(int nid, u64 start, u64 size,
784 struct mhp_params *params)
785 {
786 unsigned long start_pfn = start >> PAGE_SHIFT;
787 unsigned long nr_pages = size >> PAGE_SHIFT;
788 int ret;
789
790 /*
791 * The page tables were already mapped at boot so if the caller
792 * requests a different mapping type then we must change all the
793 * pages with __set_memory_prot().
794 */
795 if (params->pgprot.pgprot != PAGE_KERNEL.pgprot) {
796 ret = __set_memory_prot(start, nr_pages, params->pgprot);
797 if (ret)
798 return ret;
799 }
800
801 return __add_pages(nid, start_pfn, nr_pages, params);
802 }
803
arch_remove_memory(u64 start,u64 size,struct vmem_altmap * altmap)804 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
805 {
806 unsigned long start_pfn = start >> PAGE_SHIFT;
807 unsigned long nr_pages = size >> PAGE_SHIFT;
808
809 __remove_pages(start_pfn, nr_pages, altmap);
810 }
811 #endif
812
813 int kernel_set_to_readonly __read_mostly;
814
mark_nxdata_nx(void)815 static void mark_nxdata_nx(void)
816 {
817 /*
818 * When this called, init has already been executed and released,
819 * so everything past _etext should be NX.
820 */
821 unsigned long start = PFN_ALIGN(_etext);
822 /*
823 * This comes from __is_kernel_text upper limit. Also HPAGE where used:
824 */
825 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
826
827 if (__supported_pte_mask & _PAGE_NX)
828 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
829 set_memory_nx(start, size >> PAGE_SHIFT);
830 }
831
mark_rodata_ro(void)832 void mark_rodata_ro(void)
833 {
834 unsigned long start = PFN_ALIGN(_text);
835 unsigned long size = (unsigned long)__end_rodata - start;
836
837 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
838 pr_info("Write protecting kernel text and read-only data: %luk\n",
839 size >> 10);
840
841 kernel_set_to_readonly = 1;
842
843 #ifdef CONFIG_CPA_DEBUG
844 pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
845 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
846
847 pr_info("Testing CPA: write protecting again\n");
848 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
849 #endif
850 mark_nxdata_nx();
851 if (__supported_pte_mask & _PAGE_NX)
852 debug_checkwx();
853 }
854