Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * | |
6 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
7 | */ | |
8 | ||
1da177e4 LT |
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> | |
cfb80c9e | 24 | #include <linux/pci.h> |
6fb14755 | 25 | #include <linux/pfn.h> |
c9cf5528 | 26 | #include <linux/poison.h> |
a9ce6bc1 | 27 | #include <linux/memblock.h> |
1da177e4 | 28 | #include <linux/proc_fs.h> |
05039b92 | 29 | #include <linux/memory_hotplug.h> |
27d99f7e | 30 | #include <linux/initrd.h> |
55b2355e | 31 | #include <linux/cpumask.h> |
5a0e3ad6 | 32 | #include <linux/gfp.h> |
1da177e4 | 33 | |
f832ff18 | 34 | #include <asm/asm.h> |
46eaa670 | 35 | #include <asm/bios_ebda.h> |
1da177e4 | 36 | #include <asm/processor.h> |
7c0f6ba6 | 37 | #include <linux/uaccess.h> |
1da177e4 LT |
38 | #include <asm/dma.h> |
39 | #include <asm/fixmap.h> | |
66441bd3 | 40 | #include <asm/e820/api.h> |
1da177e4 | 41 | #include <asm/apic.h> |
8550eb99 | 42 | #include <asm/bugs.h> |
1da177e4 LT |
43 | #include <asm/tlb.h> |
44 | #include <asm/tlbflush.h> | |
c10d1e26 | 45 | #include <asm/olpc_ofw.h> |
a5a19c63 | 46 | #include <asm/pgalloc.h> |
1da177e4 | 47 | #include <asm/sections.h> |
551889a6 | 48 | #include <asm/setup.h> |
d1163651 | 49 | #include <asm/set_memory.h> |
2b72394e | 50 | #include <asm/page_types.h> |
92a0f81d | 51 | #include <asm/cpu_entry_area.h> |
4fcb2083 | 52 | #include <asm/init.h> |
186525bd | 53 | #include <asm/pgtable_areas.h> |
0cd39f46 | 54 | #include <asm/numa.h> |
1da177e4 | 55 | |
5c51bdbe YL |
56 | #include "mm_internal.h" |
57 | ||
1da177e4 LT |
58 | unsigned long highstart_pfn, highend_pfn; |
59 | ||
dc16ecf7 | 60 | bool __read_mostly __vmalloc_start_set = false; |
4e29684c | 61 | |
1da177e4 LT |
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 | { | |
e0c4f675 | 69 | p4d_t *p4d; |
1da177e4 LT |
70 | pud_t *pud; |
71 | pmd_t *pmd_table; | |
8550eb99 | 72 | |
1da177e4 | 73 | #ifdef CONFIG_X86_PAE |
b239fb25 | 74 | if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { |
4e37a890 | 75 | pmd_table = (pmd_t *)alloc_low_page(); |
b239fb25 | 76 | set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
e0c4f675 KS |
77 | p4d = p4d_offset(pgd, 0); |
78 | pud = pud_offset(p4d, 0); | |
8550eb99 | 79 | BUG_ON(pmd_table != pmd_offset(pud, 0)); |
a376f30a Z |
80 | |
81 | return pmd_table; | |
b239fb25 JF |
82 | } |
83 | #endif | |
e0c4f675 KS |
84 | p4d = p4d_offset(pgd, 0); |
85 | pud = pud_offset(p4d, 0); | |
1da177e4 | 86 | pmd_table = pmd_offset(pud, 0); |
8550eb99 | 87 | |
1da177e4 LT |
88 | return pmd_table; |
89 | } | |
90 | ||
91 | /* | |
92 | * Create a page table and place a pointer to it in a middle page | |
8550eb99 | 93 | * directory entry: |
1da177e4 LT |
94 | */ |
95 | static pte_t * __init one_page_table_init(pmd_t *pmd) | |
96 | { | |
b239fb25 | 97 | if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { |
4e37a890 | 98 | pte_t *page_table = (pte_t *)alloc_low_page(); |
b239fb25 | 99 | |
1da177e4 | 100 | set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); |
b239fb25 | 101 | BUG_ON(page_table != pte_offset_kernel(pmd, 0)); |
1da177e4 | 102 | } |
509a80c4 | 103 | |
1da177e4 LT |
104 | return pte_offset_kernel(pmd, 0); |
105 | } | |
106 | ||
458a3e64 | 107 | pmd_t * __init populate_extra_pmd(unsigned long vaddr) |
11124411 TH |
108 | { |
109 | int pgd_idx = pgd_index(vaddr); | |
110 | int pmd_idx = pmd_index(vaddr); | |
458a3e64 TH |
111 | |
112 | return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; | |
113 | } | |
114 | ||
115 | pte_t * __init populate_extra_pte(unsigned long vaddr) | |
116 | { | |
117 | int pte_idx = pte_index(vaddr); | |
11124411 TH |
118 | pmd_t *pmd; |
119 | ||
458a3e64 TH |
120 | pmd = populate_extra_pmd(vaddr); |
121 | return one_page_table_init(pmd) + pte_idx; | |
11124411 TH |
122 | } |
123 | ||
719272c4 YL |
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); | |
9962eea9 | 139 | pmd_idx = pmd_index(vaddr); |
719272c4 YL |
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 | ||
a3c6018e | 155 | static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, |
719272c4 YL |
156 | unsigned long vaddr, pte_t *lastpte, |
157 | void **adr) | |
a3c6018e JB |
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 | |
719272c4 | 171 | && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) { |
a3c6018e JB |
172 | pte_t *newpte; |
173 | int i; | |
174 | ||
c464573c | 175 | BUG_ON(after_bootmem); |
719272c4 | 176 | newpte = *adr; |
a3c6018e JB |
177 | for (i = 0; i < PTRS_PER_PTE; i++) |
178 | set_pte(newpte + i, pte[i]); | |
719272c4 | 179 | *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE); |
a3c6018e | 180 | |
a3c6018e JB |
181 | set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); |
182 | BUG_ON(newpte != pte_offset_kernel(pmd, 0)); | |
183 | __flush_tlb_all(); | |
184 | ||
a3c6018e JB |
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 | ||
1da177e4 | 194 | /* |
8550eb99 | 195 | * This function initializes a certain range of kernel virtual memory |
1da177e4 LT |
196 | * with new bootmem page tables, everywhere page tables are missing in |
197 | * the given range. | |
8550eb99 IM |
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 | |
1da177e4 LT |
201 | * checking the pgd every time. |
202 | */ | |
8550eb99 IM |
203 | static void __init |
204 | page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) | |
1da177e4 | 205 | { |
1da177e4 LT |
206 | int pgd_idx, pmd_idx; |
207 | unsigned long vaddr; | |
8550eb99 IM |
208 | pgd_t *pgd; |
209 | pmd_t *pmd; | |
a3c6018e | 210 | pte_t *pte = NULL; |
719272c4 YL |
211 | unsigned long count = page_table_range_init_count(start, end); |
212 | void *adr = NULL; | |
213 | ||
214 | if (count) | |
215 | adr = alloc_low_pages(count); | |
1da177e4 LT |
216 | |
217 | vaddr = start; | |
218 | pgd_idx = pgd_index(vaddr); | |
219 | pmd_idx = pmd_index(vaddr); | |
220 | pgd = pgd_base + pgd_idx; | |
221 | ||
222 | for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { | |
b239fb25 JF |
223 | pmd = one_md_table_init(pgd); |
224 | pmd = pmd + pmd_index(vaddr); | |
8550eb99 IM |
225 | for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); |
226 | pmd++, pmd_idx++) { | |
a3c6018e | 227 | pte = page_table_kmap_check(one_page_table_init(pmd), |
719272c4 | 228 | pmd, vaddr, pte, &adr); |
1da177e4 LT |
229 | |
230 | vaddr += PMD_SIZE; | |
231 | } | |
232 | pmd_idx = 0; | |
233 | } | |
234 | } | |
235 | ||
0a96c902 | 236 | static inline int is_x86_32_kernel_text(unsigned long addr) |
1da177e4 | 237 | { |
5bd5a452 | 238 | if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end) |
1da177e4 LT |
239 | return 1; |
240 | return 0; | |
241 | } | |
242 | ||
243 | /* | |
8550eb99 IM |
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: | |
1da177e4 | 247 | */ |
e53fb04f PE |
248 | unsigned long __init |
249 | kernel_physical_mapping_init(unsigned long start, | |
250 | unsigned long end, | |
c164fbb4 LG |
251 | unsigned long page_size_mask, |
252 | pgprot_t prot) | |
1da177e4 | 253 | { |
e53fb04f | 254 | int use_pse = page_size_mask == (1<<PG_LEVEL_2M); |
c1fd1b43 | 255 | unsigned long last_map_addr = end; |
e53fb04f | 256 | unsigned long start_pfn, end_pfn; |
e7179853 | 257 | pgd_t *pgd_base = swapper_pg_dir; |
8550eb99 | 258 | int pgd_idx, pmd_idx, pte_ofs; |
1da177e4 LT |
259 | unsigned long pfn; |
260 | pgd_t *pgd; | |
261 | pmd_t *pmd; | |
262 | pte_t *pte; | |
a2699e47 SS |
263 | unsigned pages_2m, pages_4k; |
264 | int mapping_iter; | |
265 | ||
e53fb04f PE |
266 | start_pfn = start >> PAGE_SHIFT; |
267 | end_pfn = end >> PAGE_SHIFT; | |
268 | ||
a2699e47 SS |
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; | |
1da177e4 | 284 | |
16bf9226 | 285 | if (!boot_cpu_has(X86_FEATURE_PSE)) |
a04ad82d | 286 | use_pse = 0; |
1da177e4 | 287 | |
a2699e47 SS |
288 | repeat: |
289 | pages_2m = pages_4k = 0; | |
a04ad82d YL |
290 | pfn = start_pfn; |
291 | pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); | |
292 | pgd = pgd_base + pgd_idx; | |
1da177e4 LT |
293 | for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { |
294 | pmd = one_md_table_init(pgd); | |
8550eb99 | 295 | |
a04ad82d YL |
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; | |
f3f20de8 | 305 | pmd++, pmd_idx++) { |
8550eb99 | 306 | unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; |
1da177e4 | 307 | |
8550eb99 IM |
308 | /* |
309 | * Map with big pages if possible, otherwise | |
310 | * create normal page tables: | |
311 | */ | |
a04ad82d | 312 | if (use_pse) { |
8550eb99 | 313 | unsigned int addr2; |
f3f20de8 | 314 | pgprot_t prot = PAGE_KERNEL_LARGE; |
a2699e47 SS |
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); | |
f3f20de8 | 322 | |
960ddb4f | 323 | pfn &= PMD_MASK >> PAGE_SHIFT; |
8550eb99 | 324 | addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + |
f3f20de8 JF |
325 | PAGE_OFFSET + PAGE_SIZE-1; |
326 | ||
0a96c902 KW |
327 | if (is_x86_32_kernel_text(addr) || |
328 | is_x86_32_kernel_text(addr2)) | |
f3f20de8 JF |
329 | prot = PAGE_KERNEL_LARGE_EXEC; |
330 | ||
ce0c0e50 | 331 | pages_2m++; |
a2699e47 SS |
332 | if (mapping_iter == 1) |
333 | set_pmd(pmd, pfn_pmd(pfn, init_prot)); | |
334 | else | |
335 | set_pmd(pmd, pfn_pmd(pfn, prot)); | |
b239fb25 | 336 | |
1da177e4 | 337 | pfn += PTRS_PER_PTE; |
8550eb99 IM |
338 | continue; |
339 | } | |
340 | pte = one_page_table_init(pmd); | |
1da177e4 | 341 | |
a04ad82d YL |
342 | pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); |
343 | pte += pte_ofs; | |
344 | for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; | |
8550eb99 IM |
345 | pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { |
346 | pgprot_t prot = PAGE_KERNEL; | |
a2699e47 SS |
347 | /* |
348 | * first pass will use the same initial | |
349 | * identity mapping attribute. | |
350 | */ | |
351 | pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); | |
f3f20de8 | 352 | |
0a96c902 | 353 | if (is_x86_32_kernel_text(addr)) |
8550eb99 | 354 | prot = PAGE_KERNEL_EXEC; |
f3f20de8 | 355 | |
ce0c0e50 | 356 | pages_4k++; |
c1fd1b43 | 357 | if (mapping_iter == 1) { |
a2699e47 | 358 | set_pte(pte, pfn_pte(pfn, init_prot)); |
c1fd1b43 PE |
359 | last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; |
360 | } else | |
a2699e47 | 361 | set_pte(pte, pfn_pte(pfn, prot)); |
1da177e4 LT |
362 | } |
363 | } | |
364 | } | |
a2699e47 SS |
365 | if (mapping_iter == 1) { |
366 | /* | |
367 | * update direct mapping page count only in the first | |
368 | * iteration. | |
369 | */ | |
370 | update_page_count(PG_LEVEL_2M, pages_2m); | |
371 | update_page_count(PG_LEVEL_4K, 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 | } | |
c1fd1b43 | 385 | return last_map_addr; |
ae531c26 AV |
386 | } |
387 | ||
fd940934 | 388 | #ifdef CONFIG_HIGHMEM |
1da177e4 LT |
389 | static void __init permanent_kmaps_init(pgd_t *pgd_base) |
390 | { | |
e05c7b1f | 391 | unsigned long vaddr = PKMAP_BASE; |
1da177e4 | 392 | |
1da177e4 LT |
393 | page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); |
394 | ||
e05c7b1f | 395 | pkmap_page_table = virt_to_kpte(vaddr); |
1da177e4 LT |
396 | } |
397 | ||
1d931264 YL |
398 | void __init add_highpages_with_active_regions(int nid, |
399 | unsigned long start_pfn, unsigned long end_pfn) | |
1da177e4 | 400 | { |
8a9ca34c TH |
401 | phys_addr_t start, end; |
402 | u64 i; | |
403 | ||
fc6daaf9 | 404 | for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) { |
8a9ca34c TH |
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)) | |
5e7ccf86 | 411 | free_highmem_page(pfn_to_page(pfn)); |
23be8c7d | 412 | } |
b5bc6c0e | 413 | } |
1da177e4 | 414 | #else |
e8e32326 IB |
415 | static inline void permanent_kmaps_init(pgd_t *pgd_base) |
416 | { | |
417 | } | |
1da177e4 LT |
418 | #endif /* CONFIG_HIGHMEM */ |
419 | ||
945fd17a TG |
420 | void __init sync_initial_page_table(void) |
421 | { | |
422 | clone_pgd_range(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(initial_page_table, | |
431 | swapper_pg_dir + KERNEL_PGD_BOUNDARY, | |
432 | min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); | |
433 | } | |
434 | ||
7737b215 | 435 | void __init native_pagetable_init(void) |
1da177e4 | 436 | { |
551889a6 | 437 | unsigned long pfn, va; |
73090f89 | 438 | pgd_t *pgd, *base = swapper_pg_dir; |
e0c4f675 | 439 | p4d_t *p4d; |
551889a6 IC |
440 | pud_t *pud; |
441 | pmd_t *pmd; | |
442 | pte_t *pte; | |
b239fb25 JF |
443 | |
444 | /* | |
551889a6 | 445 | * Remove any mappings which extend past the end of physical |
11ed9e92 YL |
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. | |
b239fb25 | 452 | */ |
11ed9e92 | 453 | for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) { |
551889a6 IC |
454 | va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); |
455 | pgd = base + pgd_index(va); | |
456 | if (!pgd_present(*pgd)) | |
457 | break; | |
458 | ||
e0c4f675 KS |
459 | p4d = p4d_offset(pgd, va); |
460 | pud = pud_offset(p4d, va); | |
551889a6 IC |
461 | pmd = pmd_offset(pud, va); |
462 | if (!pmd_present(*pmd)) | |
463 | break; | |
464 | ||
11ed9e92 YL |
465 | /* should not be large page here */ |
466 | if (pmd_large(*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 | ||
551889a6 IC |
472 | pte = pte_offset_kernel(pmd, va); |
473 | if (!pte_present(*pte)) | |
474 | break; | |
475 | ||
11ed9e92 YL |
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)); | |
551889a6 IC |
478 | pte_clear(NULL, va, pte); |
479 | } | |
843b8ed2 | 480 | paging_init(); |
b239fb25 JF |
481 | } |
482 | ||
b239fb25 JF |
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 | * | |
78841cd1 JG |
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. | |
b239fb25 JF |
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 | */ | |
f765090a | 495 | void __init early_ioremap_page_table_range_init(void) |
b239fb25 | 496 | { |
e7179853 | 497 | pgd_t *pgd_base = swapper_pg_dir; |
8550eb99 | 498 | unsigned long vaddr, end; |
b239fb25 | 499 | |
1da177e4 LT |
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; | |
b239fb25 JF |
505 | end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; |
506 | page_table_range_init(vaddr, end, pgd_base); | |
beacfaac | 507 | early_ioremap_reset(); |
e7b37895 YL |
508 | } |
509 | ||
510 | static void __init pagetable_init(void) | |
511 | { | |
512 | pgd_t *pgd_base = swapper_pg_dir; | |
513 | ||
1da177e4 | 514 | permanent_kmaps_init(pgd_base); |
1da177e4 LT |
515 | } |
516 | ||
8a57f484 DH |
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; | |
6fdc05d4 | 522 | EXPORT_SYMBOL_GPL(__supported_pte_mask); |
8a57f484 DH |
523 | /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */ |
524 | EXPORT_SYMBOL(__default_kernel_pte_mask); | |
6fdc05d4 | 525 | |
90d967e0 YL |
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(arg, &arg) >> PAGE_SHIFT; | |
540 | return 0; | |
541 | } | |
542 | early_param("highmem", parse_highmem); | |
543 | ||
4769843b IM |
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" | |
90d967e0 | 549 | /* |
4769843b IM |
550 | * All of RAM fits into lowmem - but if user wants highmem |
551 | * artificially via the highmem=x boot parameter then create | |
552 | * it: | |
90d967e0 | 553 | */ |
f836e35a | 554 | static void __init lowmem_pfn_init(void) |
90d967e0 | 555 | { |
346cafec | 556 | /* max_low_pfn is 0, we already have early_res support */ |
90d967e0 | 557 | max_low_pfn = max_pfn; |
d88316c2 | 558 | |
4769843b IM |
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, | |
90d967e0 YL |
570 | pages_to_mb(highmem_pages)); |
571 | highmem_pages = 0; | |
572 | } | |
4769843b IM |
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 | */ | |
f836e35a | 590 | static void __init highmem_pfn_init(void) |
4769843b | 591 | { |
d88316c2 IM |
592 | max_low_pfn = MAXMEM_PFN; |
593 | ||
4769843b IM |
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 | } | |
90d967e0 | 606 | #ifndef CONFIG_HIGHMEM |
4769843b IM |
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; | |
90d967e0 YL |
614 | #else /* !CONFIG_HIGHMEM */ |
615 | #ifndef CONFIG_HIGHMEM64G | |
4769843b IM |
616 | if (max_pfn > MAX_NONPAE_PFN) { |
617 | max_pfn = MAX_NONPAE_PFN; | |
618 | printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); | |
619 | } | |
90d967e0 YL |
620 | #endif /* !CONFIG_HIGHMEM64G */ |
621 | #endif /* !CONFIG_HIGHMEM */ | |
4769843b IM |
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 | ||
d88316c2 | 631 | if (max_pfn <= MAXMEM_PFN) |
4769843b | 632 | lowmem_pfn_init(); |
d88316c2 IM |
633 | else |
634 | highmem_pfn_init(); | |
90d967e0 YL |
635 | } |
636 | ||
a9ee6cf5 | 637 | #ifndef CONFIG_NUMA |
d8fc3afc | 638 | void __init initmem_init(void) |
b2ac82a0 | 639 | { |
b2ac82a0 YL |
640 | #ifdef CONFIG_HIGHMEM |
641 | highstart_pfn = highend_pfn = max_pfn; | |
642 | if (max_pfn > max_low_pfn) | |
643 | highstart_pfn = max_low_pfn; | |
b2ac82a0 YL |
644 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", |
645 | pages_to_mb(highend_pfn - highstart_pfn)); | |
b2ac82a0 YL |
646 | high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; |
647 | #else | |
b2ac82a0 YL |
648 | high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; |
649 | #endif | |
0608f70c | 650 | |
d7dc899a | 651 | memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0); |
0608f70c | 652 | |
b2ac82a0 | 653 | #ifdef CONFIG_FLATMEM |
46a84132 | 654 | max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn; |
b2ac82a0 | 655 | #endif |
dc16ecf7 JF |
656 | __vmalloc_start_set = true; |
657 | ||
b2ac82a0 YL |
658 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", |
659 | pages_to_mb(max_low_pfn)); | |
660 | ||
661 | setup_bootmem_allocator(); | |
b2ac82a0 | 662 | } |
a9ee6cf5 | 663 | #endif /* !CONFIG_NUMA */ |
b2ac82a0 | 664 | |
b2ac82a0 YL |
665 | void __init setup_bootmem_allocator(void) |
666 | { | |
b2ac82a0 YL |
667 | printk(KERN_INFO " mapped low ram: 0 - %08lx\n", |
668 | max_pfn_mapped<<PAGE_SHIFT); | |
fc5efe39 | 669 | printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); |
4e29684c YL |
670 | } |
671 | ||
1da177e4 LT |
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 | { | |
1da177e4 LT |
681 | pagetable_init(); |
682 | ||
1da177e4 LT |
683 | __flush_tlb_all(); |
684 | ||
11cd0bc1 YL |
685 | /* |
686 | * NOTE: at this point the bootmem allocator is fully available. | |
687 | */ | |
c10d1e26 | 688 | olpc_dt_build_devicetree(); |
11cd0bc1 YL |
689 | sparse_init(); |
690 | zone_sizes_init(); | |
1da177e4 LT |
691 | } |
692 | ||
693 | /* | |
694 | * Test if the WP bit works in supervisor mode. It isn't supported on 386's | |
f7f17a67 DV |
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. | |
1da177e4 | 698 | */ |
1da177e4 LT |
699 | static void __init test_wp_bit(void) |
700 | { | |
4af17110 | 701 | char z = 0; |
6415813b | 702 | |
952a6c2c | 703 | printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode..."); |
1da177e4 | 704 | |
4af17110 | 705 | __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO); |
1da177e4 | 706 | |
fe557319 | 707 | if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) { |
952a6c2c BP |
708 | clear_fixmap(FIX_WP_TEST); |
709 | printk(KERN_CONT "Ok.\n"); | |
710 | return; | |
1da177e4 | 711 | } |
4af17110 | 712 | |
952a6c2c BP |
713 | printk(KERN_CONT "No.\n"); |
714 | panic("Linux doesn't support CPUs with broken WP."); | |
1da177e4 LT |
715 | } |
716 | ||
1da177e4 LT |
717 | void __init mem_init(void) |
718 | { | |
cfb80c9e JF |
719 | pci_iommu_alloc(); |
720 | ||
05b79bdc | 721 | #ifdef CONFIG_FLATMEM |
8d8f3cbe | 722 | BUG_ON(!mem_map); |
1da177e4 | 723 | #endif |
855c743a SG |
724 | /* |
725 | * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to | |
c6ffc5ca | 726 | * be done before memblock_free_all(). Memblock use free low memory for |
855c743a SG |
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 | ||
1da177e4 | 735 | /* this will put all low memory onto the freelists */ |
c6ffc5ca | 736 | memblock_free_all(); |
1da177e4 | 737 | |
4e37a890 | 738 | after_bootmem = 1; |
6f84f8d1 | 739 | x86_init.hyper.init_after_bootmem(); |
4e37a890 | 740 | |
beeb4195 JB |
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 | ||
052e7994 | 755 | #ifdef CONFIG_HIGHMEM |
8550eb99 IM |
756 | BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); |
757 | BUG_ON(VMALLOC_END > PKMAP_BASE); | |
052e7994 | 758 | #endif |
beeb4195 | 759 | BUG_ON(VMALLOC_START >= VMALLOC_END); |
8550eb99 | 760 | BUG_ON((unsigned long)high_memory > VMALLOC_START); |
052e7994 | 761 | |
6415813b | 762 | test_wp_bit(); |
1da177e4 LT |
763 | } |
764 | ||
502f6604 | 765 | int kernel_set_to_readonly __read_mostly; |
16239630 | 766 | |
5bd5a452 MC |
767 | static void mark_nxdata_nx(void) |
768 | { | |
769 | /* | |
770 | * When this called, init has already been executed and released, | |
0d2eb44f | 771 | * so everything past _etext should be NX. |
5bd5a452 MC |
772 | */ |
773 | unsigned long start = PFN_ALIGN(_etext); | |
774 | /* | |
0a96c902 | 775 | * This comes from is_x86_32_kernel_text upper limit. Also HPAGE where used: |
5bd5a452 MC |
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); | |
185be151 | 781 | set_memory_nx(start, size >> PAGE_SHIFT); |
5bd5a452 MC |
782 | } |
783 | ||
63aaf308 AV |
784 | void mark_rodata_ro(void) |
785 | { | |
6fb14755 | 786 | unsigned long start = PFN_ALIGN(_text); |
2a25dc7c | 787 | unsigned long size = (unsigned long)__end_rodata - start; |
63aaf308 | 788 | |
4e4eee0e | 789 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
2a25dc7c | 790 | pr_info("Write protecting kernel text and read-only data: %luk\n", |
4e4eee0e | 791 | size >> 10); |
0c42f392 | 792 | |
16239630 SR |
793 | kernel_set_to_readonly = 1; |
794 | ||
0c42f392 | 795 | #ifdef CONFIG_CPA_DEBUG |
2a25dc7c | 796 | pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size); |
6d238cc4 | 797 | set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 798 | |
2a25dc7c | 799 | pr_info("Testing CPA: write protecting again\n"); |
6d238cc4 | 800 | set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); |
0c42f392 | 801 | #endif |
5bd5a452 | 802 | mark_nxdata_nx(); |
e1a58320 SS |
803 | if (__supported_pte_mask & _PAGE_NX) |
804 | debug_checkwx(); | |
63aaf308 | 805 | } |