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