Commit | Line | Data |
---|---|---|
c1cc1552 CM |
1 | /* |
2 | * Based on arch/arm/mm/mmu.c | |
3 | * | |
4 | * Copyright (C) 1995-2005 Russell King | |
5 | * Copyright (C) 2012 ARM Ltd. | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | |
18 | */ | |
19 | ||
5a9e3e15 | 20 | #include <linux/cache.h> |
c1cc1552 CM |
21 | #include <linux/export.h> |
22 | #include <linux/kernel.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/init.h> | |
98d2e153 TA |
25 | #include <linux/ioport.h> |
26 | #include <linux/kexec.h> | |
61bd93ce | 27 | #include <linux/libfdt.h> |
c1cc1552 CM |
28 | #include <linux/mman.h> |
29 | #include <linux/nodemask.h> | |
30 | #include <linux/memblock.h> | |
31 | #include <linux/fs.h> | |
2475ff9d | 32 | #include <linux/io.h> |
2077be67 | 33 | #include <linux/mm.h> |
c1cc1552 | 34 | |
21ab99c2 | 35 | #include <asm/barrier.h> |
c1cc1552 | 36 | #include <asm/cputype.h> |
af86e597 | 37 | #include <asm/fixmap.h> |
068a17a5 | 38 | #include <asm/kasan.h> |
b433dce0 | 39 | #include <asm/kernel-pgtable.h> |
c1cc1552 CM |
40 | #include <asm/sections.h> |
41 | #include <asm/setup.h> | |
42 | #include <asm/sizes.h> | |
43 | #include <asm/tlb.h> | |
c79b954b | 44 | #include <asm/memblock.h> |
c1cc1552 | 45 | #include <asm/mmu_context.h> |
1404d6f1 | 46 | #include <asm/ptdump.h> |
c1cc1552 | 47 | |
c0951366 | 48 | #define NO_BLOCK_MAPPINGS BIT(0) |
d27cfa1f | 49 | #define NO_CONT_MAPPINGS BIT(1) |
c0951366 | 50 | |
dd006da2 AB |
51 | u64 idmap_t0sz = TCR_T0SZ(VA_BITS); |
52 | ||
5a9e3e15 | 53 | u64 kimage_voffset __ro_after_init; |
a7f8de16 AB |
54 | EXPORT_SYMBOL(kimage_voffset); |
55 | ||
c1cc1552 CM |
56 | /* |
57 | * Empty_zero_page is a special page that is used for zero-initialized data | |
58 | * and COW. | |
59 | */ | |
5227cfa7 | 60 | unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss; |
c1cc1552 CM |
61 | EXPORT_SYMBOL(empty_zero_page); |
62 | ||
f9040773 AB |
63 | static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss; |
64 | static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused; | |
65 | static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused; | |
66 | ||
c1cc1552 CM |
67 | pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, |
68 | unsigned long size, pgprot_t vma_prot) | |
69 | { | |
70 | if (!pfn_valid(pfn)) | |
71 | return pgprot_noncached(vma_prot); | |
72 | else if (file->f_flags & O_SYNC) | |
73 | return pgprot_writecombine(vma_prot); | |
74 | return vma_prot; | |
75 | } | |
76 | EXPORT_SYMBOL(phys_mem_access_prot); | |
77 | ||
f4710445 | 78 | static phys_addr_t __init early_pgtable_alloc(void) |
c1cc1552 | 79 | { |
7142392d SP |
80 | phys_addr_t phys; |
81 | void *ptr; | |
82 | ||
21ab99c2 | 83 | phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE); |
f4710445 MR |
84 | |
85 | /* | |
86 | * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE | |
87 | * slot will be free, so we can (ab)use the FIX_PTE slot to initialise | |
88 | * any level of table. | |
89 | */ | |
90 | ptr = pte_set_fixmap(phys); | |
91 | ||
21ab99c2 MR |
92 | memset(ptr, 0, PAGE_SIZE); |
93 | ||
f4710445 MR |
94 | /* |
95 | * Implicit barriers also ensure the zeroed page is visible to the page | |
96 | * table walker | |
97 | */ | |
98 | pte_clear_fixmap(); | |
99 | ||
100 | return phys; | |
c1cc1552 CM |
101 | } |
102 | ||
e98216b5 AB |
103 | static bool pgattr_change_is_safe(u64 old, u64 new) |
104 | { | |
105 | /* | |
106 | * The following mapping attributes may be updated in live | |
107 | * kernel mappings without the need for break-before-make. | |
108 | */ | |
109 | static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE; | |
110 | ||
141d1497 AB |
111 | /* creating or taking down mappings is always safe */ |
112 | if (old == 0 || new == 0) | |
113 | return true; | |
114 | ||
115 | /* live contiguous mappings may not be manipulated at all */ | |
116 | if ((old | new) & PTE_CONT) | |
117 | return false; | |
118 | ||
119 | return ((old ^ new) & ~mask) == 0; | |
e98216b5 AB |
120 | } |
121 | ||
d27cfa1f AB |
122 | static void init_pte(pmd_t *pmd, unsigned long addr, unsigned long end, |
123 | phys_addr_t phys, pgprot_t prot) | |
c1cc1552 CM |
124 | { |
125 | pte_t *pte; | |
126 | ||
f4710445 | 127 | pte = pte_set_fixmap_offset(pmd, addr); |
c1cc1552 | 128 | do { |
e98216b5 AB |
129 | pte_t old_pte = *pte; |
130 | ||
e393cf40 | 131 | set_pte(pte, pfn_pte(__phys_to_pfn(phys), prot)); |
e98216b5 AB |
132 | |
133 | /* | |
134 | * After the PTE entry has been populated once, we | |
135 | * only allow updates to the permission attributes. | |
136 | */ | |
137 | BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte))); | |
138 | ||
e393cf40 | 139 | phys += PAGE_SIZE; |
667c2759 | 140 | } while (pte++, addr += PAGE_SIZE, addr != end); |
f4710445 MR |
141 | |
142 | pte_clear_fixmap(); | |
c1cc1552 CM |
143 | } |
144 | ||
d27cfa1f AB |
145 | static void alloc_init_cont_pte(pmd_t *pmd, unsigned long addr, |
146 | unsigned long end, phys_addr_t phys, | |
147 | pgprot_t prot, | |
148 | phys_addr_t (*pgtable_alloc)(void), | |
149 | int flags) | |
c1cc1552 | 150 | { |
c1cc1552 CM |
151 | unsigned long next; |
152 | ||
d27cfa1f AB |
153 | BUG_ON(pmd_sect(*pmd)); |
154 | if (pmd_none(*pmd)) { | |
155 | phys_addr_t pte_phys; | |
132233a7 | 156 | BUG_ON(!pgtable_alloc); |
d27cfa1f AB |
157 | pte_phys = pgtable_alloc(); |
158 | __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE); | |
c1cc1552 | 159 | } |
d27cfa1f AB |
160 | BUG_ON(pmd_bad(*pmd)); |
161 | ||
162 | do { | |
163 | pgprot_t __prot = prot; | |
164 | ||
165 | next = pte_cont_addr_end(addr, end); | |
166 | ||
167 | /* use a contiguous mapping if the range is suitably aligned */ | |
168 | if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) && | |
169 | (flags & NO_CONT_MAPPINGS) == 0) | |
170 | __prot = __pgprot(pgprot_val(prot) | PTE_CONT); | |
171 | ||
172 | init_pte(pmd, addr, next, phys, __prot); | |
173 | ||
174 | phys += next - addr; | |
175 | } while (addr = next, addr != end); | |
176 | } | |
177 | ||
178 | static void init_pmd(pud_t *pud, unsigned long addr, unsigned long end, | |
179 | phys_addr_t phys, pgprot_t prot, | |
180 | phys_addr_t (*pgtable_alloc)(void), int flags) | |
181 | { | |
182 | unsigned long next; | |
183 | pmd_t *pmd; | |
c1cc1552 | 184 | |
f4710445 | 185 | pmd = pmd_set_fixmap_offset(pud, addr); |
c1cc1552 | 186 | do { |
e98216b5 AB |
187 | pmd_t old_pmd = *pmd; |
188 | ||
c1cc1552 | 189 | next = pmd_addr_end(addr, end); |
e98216b5 | 190 | |
c1cc1552 | 191 | /* try section mapping first */ |
83863f25 | 192 | if (((addr | next | phys) & ~SECTION_MASK) == 0 && |
c0951366 | 193 | (flags & NO_BLOCK_MAPPINGS) == 0) { |
d81bbe6d | 194 | pmd_set_huge(pmd, phys, prot); |
e98216b5 | 195 | |
a55f9929 | 196 | /* |
e98216b5 AB |
197 | * After the PMD entry has been populated once, we |
198 | * only allow updates to the permission attributes. | |
a55f9929 | 199 | */ |
e98216b5 AB |
200 | BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd), |
201 | pmd_val(*pmd))); | |
a55f9929 | 202 | } else { |
d27cfa1f AB |
203 | alloc_init_cont_pte(pmd, addr, next, phys, prot, |
204 | pgtable_alloc, flags); | |
e98216b5 AB |
205 | |
206 | BUG_ON(pmd_val(old_pmd) != 0 && | |
207 | pmd_val(old_pmd) != pmd_val(*pmd)); | |
a55f9929 | 208 | } |
c1cc1552 CM |
209 | phys += next - addr; |
210 | } while (pmd++, addr = next, addr != end); | |
f4710445 MR |
211 | |
212 | pmd_clear_fixmap(); | |
c1cc1552 CM |
213 | } |
214 | ||
d27cfa1f AB |
215 | static void alloc_init_cont_pmd(pud_t *pud, unsigned long addr, |
216 | unsigned long end, phys_addr_t phys, | |
217 | pgprot_t prot, | |
218 | phys_addr_t (*pgtable_alloc)(void), int flags) | |
219 | { | |
220 | unsigned long next; | |
221 | ||
222 | /* | |
223 | * Check for initial section mappings in the pgd/pud. | |
224 | */ | |
225 | BUG_ON(pud_sect(*pud)); | |
226 | if (pud_none(*pud)) { | |
227 | phys_addr_t pmd_phys; | |
228 | BUG_ON(!pgtable_alloc); | |
229 | pmd_phys = pgtable_alloc(); | |
230 | __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE); | |
231 | } | |
232 | BUG_ON(pud_bad(*pud)); | |
233 | ||
234 | do { | |
235 | pgprot_t __prot = prot; | |
236 | ||
237 | next = pmd_cont_addr_end(addr, end); | |
238 | ||
239 | /* use a contiguous mapping if the range is suitably aligned */ | |
240 | if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) && | |
241 | (flags & NO_CONT_MAPPINGS) == 0) | |
242 | __prot = __pgprot(pgprot_val(prot) | PTE_CONT); | |
243 | ||
244 | init_pmd(pud, addr, next, phys, __prot, pgtable_alloc, flags); | |
245 | ||
246 | phys += next - addr; | |
247 | } while (addr = next, addr != end); | |
248 | } | |
249 | ||
da141706 LA |
250 | static inline bool use_1G_block(unsigned long addr, unsigned long next, |
251 | unsigned long phys) | |
252 | { | |
253 | if (PAGE_SHIFT != 12) | |
254 | return false; | |
255 | ||
256 | if (((addr | next | phys) & ~PUD_MASK) != 0) | |
257 | return false; | |
258 | ||
259 | return true; | |
260 | } | |
261 | ||
11509a30 | 262 | static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end, |
da141706 | 263 | phys_addr_t phys, pgprot_t prot, |
53e1b329 | 264 | phys_addr_t (*pgtable_alloc)(void), |
c0951366 | 265 | int flags) |
c1cc1552 | 266 | { |
c79b954b | 267 | pud_t *pud; |
c1cc1552 CM |
268 | unsigned long next; |
269 | ||
c79b954b | 270 | if (pgd_none(*pgd)) { |
132233a7 LA |
271 | phys_addr_t pud_phys; |
272 | BUG_ON(!pgtable_alloc); | |
273 | pud_phys = pgtable_alloc(); | |
f4710445 | 274 | __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE); |
c79b954b JL |
275 | } |
276 | BUG_ON(pgd_bad(*pgd)); | |
277 | ||
f4710445 | 278 | pud = pud_set_fixmap_offset(pgd, addr); |
c1cc1552 | 279 | do { |
e98216b5 AB |
280 | pud_t old_pud = *pud; |
281 | ||
c1cc1552 | 282 | next = pud_addr_end(addr, end); |
206a2a73 SC |
283 | |
284 | /* | |
285 | * For 4K granule only, attempt to put down a 1GB block | |
286 | */ | |
c0951366 AB |
287 | if (use_1G_block(addr, next, phys) && |
288 | (flags & NO_BLOCK_MAPPINGS) == 0) { | |
c661cb1c | 289 | pud_set_huge(pud, phys, prot); |
206a2a73 SC |
290 | |
291 | /* | |
e98216b5 AB |
292 | * After the PUD entry has been populated once, we |
293 | * only allow updates to the permission attributes. | |
206a2a73 | 294 | */ |
e98216b5 AB |
295 | BUG_ON(!pgattr_change_is_safe(pud_val(old_pud), |
296 | pud_val(*pud))); | |
206a2a73 | 297 | } else { |
d27cfa1f AB |
298 | alloc_init_cont_pmd(pud, addr, next, phys, prot, |
299 | pgtable_alloc, flags); | |
e98216b5 AB |
300 | |
301 | BUG_ON(pud_val(old_pud) != 0 && | |
302 | pud_val(old_pud) != pud_val(*pud)); | |
206a2a73 | 303 | } |
c1cc1552 CM |
304 | phys += next - addr; |
305 | } while (pud++, addr = next, addr != end); | |
f4710445 MR |
306 | |
307 | pud_clear_fixmap(); | |
c1cc1552 CM |
308 | } |
309 | ||
40f87d31 AB |
310 | static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys, |
311 | unsigned long virt, phys_addr_t size, | |
312 | pgprot_t prot, | |
313 | phys_addr_t (*pgtable_alloc)(void), | |
c0951366 | 314 | int flags) |
c1cc1552 CM |
315 | { |
316 | unsigned long addr, length, end, next; | |
40f87d31 | 317 | pgd_t *pgd = pgd_offset_raw(pgdir, virt); |
c1cc1552 | 318 | |
cc5d2b3b MR |
319 | /* |
320 | * If the virtual and physical address don't have the same offset | |
321 | * within a page, we cannot map the region as the caller expects. | |
322 | */ | |
323 | if (WARN_ON((phys ^ virt) & ~PAGE_MASK)) | |
324 | return; | |
325 | ||
9c4e08a3 | 326 | phys &= PAGE_MASK; |
c1cc1552 CM |
327 | addr = virt & PAGE_MASK; |
328 | length = PAGE_ALIGN(size + (virt & ~PAGE_MASK)); | |
329 | ||
c1cc1552 CM |
330 | end = addr + length; |
331 | do { | |
332 | next = pgd_addr_end(addr, end); | |
53e1b329 | 333 | alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc, |
c0951366 | 334 | flags); |
c1cc1552 CM |
335 | phys += next - addr; |
336 | } while (pgd++, addr = next, addr != end); | |
337 | } | |
338 | ||
1378dc3d | 339 | static phys_addr_t pgd_pgtable_alloc(void) |
da141706 | 340 | { |
21ab99c2 | 341 | void *ptr = (void *)__get_free_page(PGALLOC_GFP); |
1378dc3d AB |
342 | if (!ptr || !pgtable_page_ctor(virt_to_page(ptr))) |
343 | BUG(); | |
21ab99c2 MR |
344 | |
345 | /* Ensure the zeroed page is visible to the page table walker */ | |
346 | dsb(ishst); | |
f4710445 | 347 | return __pa(ptr); |
da141706 LA |
348 | } |
349 | ||
132233a7 LA |
350 | /* |
351 | * This function can only be used to modify existing table entries, | |
352 | * without allocating new levels of table. Note that this permits the | |
353 | * creation of new section or page entries. | |
354 | */ | |
355 | static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt, | |
da141706 | 356 | phys_addr_t size, pgprot_t prot) |
d7ecbddf MS |
357 | { |
358 | if (virt < VMALLOC_START) { | |
359 | pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n", | |
360 | &phys, virt); | |
361 | return; | |
362 | } | |
d27cfa1f AB |
363 | __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, |
364 | NO_CONT_MAPPINGS); | |
d7ecbddf MS |
365 | } |
366 | ||
8ce837ce AB |
367 | void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys, |
368 | unsigned long virt, phys_addr_t size, | |
f14c66ce | 369 | pgprot_t prot, bool page_mappings_only) |
8ce837ce | 370 | { |
c0951366 AB |
371 | int flags = 0; |
372 | ||
1378dc3d AB |
373 | BUG_ON(mm == &init_mm); |
374 | ||
c0951366 | 375 | if (page_mappings_only) |
d27cfa1f | 376 | flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; |
c0951366 | 377 | |
11509a30 | 378 | __create_pgd_mapping(mm->pgd, phys, virt, size, prot, |
c0951366 | 379 | pgd_pgtable_alloc, flags); |
d7ecbddf MS |
380 | } |
381 | ||
aa8c09be AB |
382 | static void update_mapping_prot(phys_addr_t phys, unsigned long virt, |
383 | phys_addr_t size, pgprot_t prot) | |
da141706 LA |
384 | { |
385 | if (virt < VMALLOC_START) { | |
aa8c09be | 386 | pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n", |
da141706 LA |
387 | &phys, virt); |
388 | return; | |
389 | } | |
390 | ||
d27cfa1f AB |
391 | __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, |
392 | NO_CONT_MAPPINGS); | |
aa8c09be AB |
393 | |
394 | /* flush the TLBs after updating live kernel mappings */ | |
395 | flush_tlb_kernel_range(virt, virt + size); | |
da141706 LA |
396 | } |
397 | ||
98d2e153 TA |
398 | static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, |
399 | phys_addr_t end, pgprot_t prot, int flags) | |
400 | { | |
401 | __create_pgd_mapping(pgd, start, __phys_to_virt(start), end - start, | |
402 | prot, early_pgtable_alloc, flags); | |
403 | } | |
404 | ||
405 | void __init mark_linear_text_alias_ro(void) | |
406 | { | |
407 | /* | |
408 | * Remove the write permissions from the linear alias of .text/.rodata | |
409 | */ | |
410 | update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text), | |
411 | (unsigned long)__init_begin - (unsigned long)_text, | |
412 | PAGE_KERNEL_RO); | |
413 | } | |
414 | ||
415 | static void __init map_mem(pgd_t *pgd) | |
da141706 | 416 | { |
eac8017f MC |
417 | phys_addr_t kernel_start = __pa_symbol(_text); |
418 | phys_addr_t kernel_end = __pa_symbol(__init_begin); | |
98d2e153 | 419 | struct memblock_region *reg; |
c0951366 AB |
420 | int flags = 0; |
421 | ||
422 | if (debug_pagealloc_enabled()) | |
d27cfa1f | 423 | flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS; |
068a17a5 | 424 | |
da141706 | 425 | /* |
f9040773 AB |
426 | * Take care not to create a writable alias for the |
427 | * read-only text and rodata sections of the kernel image. | |
98d2e153 TA |
428 | * So temporarily mark them as NOMAP to skip mappings in |
429 | * the following for-loop | |
da141706 | 430 | */ |
98d2e153 TA |
431 | memblock_mark_nomap(kernel_start, kernel_end - kernel_start); |
432 | #ifdef CONFIG_KEXEC_CORE | |
433 | if (crashk_res.end) | |
434 | memblock_mark_nomap(crashk_res.start, | |
435 | resource_size(&crashk_res)); | |
436 | #endif | |
068a17a5 | 437 | |
98d2e153 TA |
438 | /* map all the memory banks */ |
439 | for_each_memblock(memory, reg) { | |
440 | phys_addr_t start = reg->base; | |
441 | phys_addr_t end = start + reg->size; | |
da141706 | 442 | |
98d2e153 TA |
443 | if (start >= end) |
444 | break; | |
445 | if (memblock_is_nomap(reg)) | |
446 | continue; | |
447 | ||
448 | __map_memblock(pgd, start, end, PAGE_KERNEL, flags); | |
449 | } | |
f9040773 AB |
450 | |
451 | /* | |
5ea5306c AB |
452 | * Map the linear alias of the [_text, __init_begin) interval |
453 | * as non-executable now, and remove the write permission in | |
454 | * mark_linear_text_alias_ro() below (which will be called after | |
455 | * alternative patching has completed). This makes the contents | |
456 | * of the region accessible to subsystems such as hibernate, | |
457 | * but protects it from inadvertent modification or execution. | |
d27cfa1f AB |
458 | * Note that contiguous mappings cannot be remapped in this way, |
459 | * so we should avoid them here. | |
f9040773 | 460 | */ |
98d2e153 TA |
461 | __map_memblock(pgd, kernel_start, kernel_end, |
462 | PAGE_KERNEL, NO_CONT_MAPPINGS); | |
463 | memblock_clear_nomap(kernel_start, kernel_end - kernel_start); | |
da141706 | 464 | |
98d2e153 | 465 | #ifdef CONFIG_KEXEC_CORE |
5ea5306c | 466 | /* |
98d2e153 TA |
467 | * Use page-level mappings here so that we can shrink the region |
468 | * in page granularity and put back unused memory to buddy system | |
469 | * through /sys/kernel/kexec_crash_size interface. | |
5ea5306c | 470 | */ |
98d2e153 TA |
471 | if (crashk_res.end) { |
472 | __map_memblock(pgd, crashk_res.start, crashk_res.end + 1, | |
473 | PAGE_KERNEL, | |
474 | NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS); | |
475 | memblock_clear_nomap(crashk_res.start, | |
476 | resource_size(&crashk_res)); | |
c1cc1552 | 477 | } |
98d2e153 | 478 | #endif |
c1cc1552 CM |
479 | } |
480 | ||
da141706 LA |
481 | void mark_rodata_ro(void) |
482 | { | |
2f39b5f9 | 483 | unsigned long section_size; |
f9040773 | 484 | |
2f39b5f9 | 485 | /* |
9fdc14c5 AB |
486 | * mark .rodata as read only. Use __init_begin rather than __end_rodata |
487 | * to cover NOTES and EXCEPTION_TABLE. | |
2f39b5f9 | 488 | */ |
9fdc14c5 | 489 | section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata; |
aa8c09be | 490 | update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata, |
2f39b5f9 | 491 | section_size, PAGE_KERNEL_RO); |
e98216b5 | 492 | |
1404d6f1 | 493 | debug_checkwx(); |
da141706 | 494 | } |
da141706 | 495 | |
2c09ec06 | 496 | static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end, |
d27cfa1f AB |
497 | pgprot_t prot, struct vm_struct *vma, |
498 | int flags) | |
068a17a5 | 499 | { |
2077be67 | 500 | phys_addr_t pa_start = __pa_symbol(va_start); |
068a17a5 MR |
501 | unsigned long size = va_end - va_start; |
502 | ||
503 | BUG_ON(!PAGE_ALIGNED(pa_start)); | |
504 | BUG_ON(!PAGE_ALIGNED(size)); | |
505 | ||
506 | __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot, | |
d27cfa1f | 507 | early_pgtable_alloc, flags); |
f9040773 AB |
508 | |
509 | vma->addr = va_start; | |
510 | vma->phys_addr = pa_start; | |
511 | vma->size = size; | |
512 | vma->flags = VM_MAP; | |
513 | vma->caller = __builtin_return_address(0); | |
514 | ||
515 | vm_area_add_early(vma); | |
068a17a5 MR |
516 | } |
517 | ||
28b066da AB |
518 | static int __init parse_rodata(char *arg) |
519 | { | |
520 | return strtobool(arg, &rodata_enabled); | |
521 | } | |
522 | early_param("rodata", parse_rodata); | |
523 | ||
068a17a5 MR |
524 | /* |
525 | * Create fine-grained mappings for the kernel. | |
526 | */ | |
527 | static void __init map_kernel(pgd_t *pgd) | |
528 | { | |
2ebe088b AB |
529 | static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext, |
530 | vmlinux_initdata, vmlinux_data; | |
068a17a5 | 531 | |
28b066da AB |
532 | /* |
533 | * External debuggers may need to write directly to the text | |
534 | * mapping to install SW breakpoints. Allow this (only) when | |
535 | * explicitly requested with rodata=off. | |
536 | */ | |
537 | pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; | |
538 | ||
d27cfa1f AB |
539 | /* |
540 | * Only rodata will be remapped with different permissions later on, | |
541 | * all other segments are allowed to use contiguous mappings. | |
542 | */ | |
543 | map_kernel_segment(pgd, _text, _etext, text_prot, &vmlinux_text, 0); | |
2ebe088b | 544 | map_kernel_segment(pgd, __start_rodata, __inittext_begin, PAGE_KERNEL, |
d27cfa1f | 545 | &vmlinux_rodata, NO_CONT_MAPPINGS); |
2ebe088b | 546 | map_kernel_segment(pgd, __inittext_begin, __inittext_end, text_prot, |
d27cfa1f | 547 | &vmlinux_inittext, 0); |
2ebe088b | 548 | map_kernel_segment(pgd, __initdata_begin, __initdata_end, PAGE_KERNEL, |
d27cfa1f AB |
549 | &vmlinux_initdata, 0); |
550 | map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data, 0); | |
068a17a5 | 551 | |
f9040773 AB |
552 | if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) { |
553 | /* | |
554 | * The fixmap falls in a separate pgd to the kernel, and doesn't | |
555 | * live in the carveout for the swapper_pg_dir. We can simply | |
556 | * re-use the existing dir for the fixmap. | |
557 | */ | |
558 | set_pgd(pgd_offset_raw(pgd, FIXADDR_START), | |
559 | *pgd_offset_k(FIXADDR_START)); | |
560 | } else if (CONFIG_PGTABLE_LEVELS > 3) { | |
561 | /* | |
562 | * The fixmap shares its top level pgd entry with the kernel | |
563 | * mapping. This can really only occur when we are running | |
564 | * with 16k/4 levels, so we can simply reuse the pud level | |
565 | * entry instead. | |
566 | */ | |
567 | BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); | |
568 | set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START), | |
2077be67 | 569 | __pud(__pa_symbol(bm_pmd) | PUD_TYPE_TABLE)); |
f9040773 AB |
570 | pud_clear_fixmap(); |
571 | } else { | |
572 | BUG(); | |
573 | } | |
068a17a5 MR |
574 | |
575 | kasan_copy_shadow(pgd); | |
576 | } | |
577 | ||
c1cc1552 CM |
578 | /* |
579 | * paging_init() sets up the page tables, initialises the zone memory | |
580 | * maps and sets up the zero page. | |
581 | */ | |
582 | void __init paging_init(void) | |
583 | { | |
068a17a5 MR |
584 | phys_addr_t pgd_phys = early_pgtable_alloc(); |
585 | pgd_t *pgd = pgd_set_fixmap(pgd_phys); | |
586 | ||
587 | map_kernel(pgd); | |
588 | map_mem(pgd); | |
589 | ||
590 | /* | |
591 | * We want to reuse the original swapper_pg_dir so we don't have to | |
592 | * communicate the new address to non-coherent secondaries in | |
593 | * secondary_entry, and so cpu_switch_mm can generate the address with | |
594 | * adrp+add rather than a load from some global variable. | |
595 | * | |
596 | * To do this we need to go via a temporary pgd. | |
597 | */ | |
598 | cpu_replace_ttbr1(__va(pgd_phys)); | |
12f043ff | 599 | memcpy(swapper_pg_dir, pgd, PGD_SIZE); |
2077be67 | 600 | cpu_replace_ttbr1(lm_alias(swapper_pg_dir)); |
068a17a5 MR |
601 | |
602 | pgd_clear_fixmap(); | |
603 | memblock_free(pgd_phys, PAGE_SIZE); | |
604 | ||
605 | /* | |
606 | * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd | |
607 | * allocated with it. | |
608 | */ | |
2077be67 | 609 | memblock_free(__pa_symbol(swapper_pg_dir) + PAGE_SIZE, |
068a17a5 | 610 | SWAPPER_DIR_SIZE - PAGE_SIZE); |
c1cc1552 CM |
611 | } |
612 | ||
c1cc1552 CM |
613 | /* |
614 | * Check whether a kernel address is valid (derived from arch/x86/). | |
615 | */ | |
616 | int kern_addr_valid(unsigned long addr) | |
617 | { | |
618 | pgd_t *pgd; | |
619 | pud_t *pud; | |
620 | pmd_t *pmd; | |
621 | pte_t *pte; | |
622 | ||
623 | if ((((long)addr) >> VA_BITS) != -1UL) | |
624 | return 0; | |
625 | ||
626 | pgd = pgd_offset_k(addr); | |
627 | if (pgd_none(*pgd)) | |
628 | return 0; | |
629 | ||
630 | pud = pud_offset(pgd, addr); | |
631 | if (pud_none(*pud)) | |
632 | return 0; | |
633 | ||
206a2a73 SC |
634 | if (pud_sect(*pud)) |
635 | return pfn_valid(pud_pfn(*pud)); | |
636 | ||
c1cc1552 CM |
637 | pmd = pmd_offset(pud, addr); |
638 | if (pmd_none(*pmd)) | |
639 | return 0; | |
640 | ||
da6e4cb6 DA |
641 | if (pmd_sect(*pmd)) |
642 | return pfn_valid(pmd_pfn(*pmd)); | |
643 | ||
c1cc1552 CM |
644 | pte = pte_offset_kernel(pmd, addr); |
645 | if (pte_none(*pte)) | |
646 | return 0; | |
647 | ||
648 | return pfn_valid(pte_pfn(*pte)); | |
649 | } | |
650 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
b433dce0 | 651 | #if !ARM64_SWAPPER_USES_SECTION_MAPS |
0aad818b | 652 | int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node) |
c1cc1552 | 653 | { |
0aad818b | 654 | return vmemmap_populate_basepages(start, end, node); |
c1cc1552 | 655 | } |
b433dce0 | 656 | #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */ |
0aad818b | 657 | int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node) |
c1cc1552 | 658 | { |
0aad818b | 659 | unsigned long addr = start; |
c1cc1552 CM |
660 | unsigned long next; |
661 | pgd_t *pgd; | |
662 | pud_t *pud; | |
663 | pmd_t *pmd; | |
664 | ||
665 | do { | |
666 | next = pmd_addr_end(addr, end); | |
667 | ||
668 | pgd = vmemmap_pgd_populate(addr, node); | |
669 | if (!pgd) | |
670 | return -ENOMEM; | |
671 | ||
672 | pud = vmemmap_pud_populate(pgd, addr, node); | |
673 | if (!pud) | |
674 | return -ENOMEM; | |
675 | ||
676 | pmd = pmd_offset(pud, addr); | |
677 | if (pmd_none(*pmd)) { | |
678 | void *p = NULL; | |
679 | ||
680 | p = vmemmap_alloc_block_buf(PMD_SIZE, node); | |
681 | if (!p) | |
682 | return -ENOMEM; | |
683 | ||
a501e324 | 684 | set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL)); |
c1cc1552 CM |
685 | } else |
686 | vmemmap_verify((pte_t *)pmd, node, addr, next); | |
687 | } while (addr = next, addr != end); | |
688 | ||
689 | return 0; | |
690 | } | |
691 | #endif /* CONFIG_ARM64_64K_PAGES */ | |
0aad818b | 692 | void vmemmap_free(unsigned long start, unsigned long end) |
0197518c TC |
693 | { |
694 | } | |
c1cc1552 | 695 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |
af86e597 | 696 | |
af86e597 LA |
697 | static inline pud_t * fixmap_pud(unsigned long addr) |
698 | { | |
699 | pgd_t *pgd = pgd_offset_k(addr); | |
700 | ||
701 | BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd)); | |
702 | ||
157962f5 | 703 | return pud_offset_kimg(pgd, addr); |
af86e597 LA |
704 | } |
705 | ||
706 | static inline pmd_t * fixmap_pmd(unsigned long addr) | |
707 | { | |
708 | pud_t *pud = fixmap_pud(addr); | |
709 | ||
710 | BUG_ON(pud_none(*pud) || pud_bad(*pud)); | |
711 | ||
157962f5 | 712 | return pmd_offset_kimg(pud, addr); |
af86e597 LA |
713 | } |
714 | ||
715 | static inline pte_t * fixmap_pte(unsigned long addr) | |
716 | { | |
157962f5 | 717 | return &bm_pte[pte_index(addr)]; |
af86e597 LA |
718 | } |
719 | ||
2077be67 LA |
720 | /* |
721 | * The p*d_populate functions call virt_to_phys implicitly so they can't be used | |
722 | * directly on kernel symbols (bm_p*d). This function is called too early to use | |
723 | * lm_alias so __p*d_populate functions must be used to populate with the | |
724 | * physical address from __pa_symbol. | |
725 | */ | |
af86e597 LA |
726 | void __init early_fixmap_init(void) |
727 | { | |
728 | pgd_t *pgd; | |
729 | pud_t *pud; | |
730 | pmd_t *pmd; | |
731 | unsigned long addr = FIXADDR_START; | |
732 | ||
733 | pgd = pgd_offset_k(addr); | |
f80fb3a3 | 734 | if (CONFIG_PGTABLE_LEVELS > 3 && |
2077be67 | 735 | !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa_symbol(bm_pud))) { |
f9040773 AB |
736 | /* |
737 | * We only end up here if the kernel mapping and the fixmap | |
738 | * share the top level pgd entry, which should only happen on | |
739 | * 16k/4 levels configurations. | |
740 | */ | |
741 | BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES)); | |
742 | pud = pud_offset_kimg(pgd, addr); | |
743 | } else { | |
2077be67 LA |
744 | if (pgd_none(*pgd)) |
745 | __pgd_populate(pgd, __pa_symbol(bm_pud), PUD_TYPE_TABLE); | |
f9040773 AB |
746 | pud = fixmap_pud(addr); |
747 | } | |
2077be67 LA |
748 | if (pud_none(*pud)) |
749 | __pud_populate(pud, __pa_symbol(bm_pmd), PMD_TYPE_TABLE); | |
157962f5 | 750 | pmd = fixmap_pmd(addr); |
2077be67 | 751 | __pmd_populate(pmd, __pa_symbol(bm_pte), PMD_TYPE_TABLE); |
af86e597 LA |
752 | |
753 | /* | |
754 | * The boot-ioremap range spans multiple pmds, for which | |
157962f5 | 755 | * we are not prepared: |
af86e597 LA |
756 | */ |
757 | BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT) | |
758 | != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT)); | |
759 | ||
760 | if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN))) | |
761 | || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) { | |
762 | WARN_ON(1); | |
763 | pr_warn("pmd %p != %p, %p\n", | |
764 | pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)), | |
765 | fixmap_pmd(fix_to_virt(FIX_BTMAP_END))); | |
766 | pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n", | |
767 | fix_to_virt(FIX_BTMAP_BEGIN)); | |
768 | pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n", | |
769 | fix_to_virt(FIX_BTMAP_END)); | |
770 | ||
771 | pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END); | |
772 | pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN); | |
773 | } | |
774 | } | |
775 | ||
776 | void __set_fixmap(enum fixed_addresses idx, | |
777 | phys_addr_t phys, pgprot_t flags) | |
778 | { | |
779 | unsigned long addr = __fix_to_virt(idx); | |
780 | pte_t *pte; | |
781 | ||
b63dbef9 | 782 | BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses); |
af86e597 LA |
783 | |
784 | pte = fixmap_pte(addr); | |
785 | ||
786 | if (pgprot_val(flags)) { | |
787 | set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags)); | |
788 | } else { | |
789 | pte_clear(&init_mm, addr, pte); | |
790 | flush_tlb_kernel_range(addr, addr+PAGE_SIZE); | |
791 | } | |
792 | } | |
61bd93ce | 793 | |
f80fb3a3 | 794 | void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot) |
61bd93ce AB |
795 | { |
796 | const u64 dt_virt_base = __fix_to_virt(FIX_FDT); | |
f80fb3a3 | 797 | int offset; |
61bd93ce AB |
798 | void *dt_virt; |
799 | ||
800 | /* | |
801 | * Check whether the physical FDT address is set and meets the minimum | |
802 | * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be | |
04a84810 AB |
803 | * at least 8 bytes so that we can always access the magic and size |
804 | * fields of the FDT header after mapping the first chunk, double check | |
805 | * here if that is indeed the case. | |
61bd93ce AB |
806 | */ |
807 | BUILD_BUG_ON(MIN_FDT_ALIGN < 8); | |
808 | if (!dt_phys || dt_phys % MIN_FDT_ALIGN) | |
809 | return NULL; | |
810 | ||
811 | /* | |
812 | * Make sure that the FDT region can be mapped without the need to | |
813 | * allocate additional translation table pages, so that it is safe | |
132233a7 | 814 | * to call create_mapping_noalloc() this early. |
61bd93ce AB |
815 | * |
816 | * On 64k pages, the FDT will be mapped using PTEs, so we need to | |
817 | * be in the same PMD as the rest of the fixmap. | |
818 | * On 4k pages, we'll use section mappings for the FDT so we only | |
819 | * have to be in the same PUD. | |
820 | */ | |
821 | BUILD_BUG_ON(dt_virt_base % SZ_2M); | |
822 | ||
b433dce0 SP |
823 | BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT != |
824 | __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT); | |
61bd93ce | 825 | |
b433dce0 | 826 | offset = dt_phys % SWAPPER_BLOCK_SIZE; |
61bd93ce AB |
827 | dt_virt = (void *)dt_virt_base + offset; |
828 | ||
829 | /* map the first chunk so we can read the size from the header */ | |
132233a7 LA |
830 | create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), |
831 | dt_virt_base, SWAPPER_BLOCK_SIZE, prot); | |
61bd93ce | 832 | |
04a84810 | 833 | if (fdt_magic(dt_virt) != FDT_MAGIC) |
61bd93ce AB |
834 | return NULL; |
835 | ||
f80fb3a3 AB |
836 | *size = fdt_totalsize(dt_virt); |
837 | if (*size > MAX_FDT_SIZE) | |
61bd93ce AB |
838 | return NULL; |
839 | ||
f80fb3a3 | 840 | if (offset + *size > SWAPPER_BLOCK_SIZE) |
132233a7 | 841 | create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base, |
f80fb3a3 | 842 | round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot); |
61bd93ce | 843 | |
f80fb3a3 AB |
844 | return dt_virt; |
845 | } | |
61bd93ce | 846 | |
f80fb3a3 AB |
847 | void *__init fixmap_remap_fdt(phys_addr_t dt_phys) |
848 | { | |
849 | void *dt_virt; | |
850 | int size; | |
851 | ||
852 | dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO); | |
853 | if (!dt_virt) | |
854 | return NULL; | |
855 | ||
856 | memblock_reserve(dt_phys, size); | |
61bd93ce AB |
857 | return dt_virt; |
858 | } | |
324420bf AB |
859 | |
860 | int __init arch_ioremap_pud_supported(void) | |
861 | { | |
862 | /* only 4k granule supports level 1 block mappings */ | |
863 | return IS_ENABLED(CONFIG_ARM64_4K_PAGES); | |
864 | } | |
865 | ||
866 | int __init arch_ioremap_pmd_supported(void) | |
867 | { | |
868 | return 1; | |
869 | } | |
870 | ||
871 | int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot) | |
872 | { | |
873 | BUG_ON(phys & ~PUD_MASK); | |
874 | set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot)))); | |
875 | return 1; | |
876 | } | |
877 | ||
878 | int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot) | |
879 | { | |
880 | BUG_ON(phys & ~PMD_MASK); | |
881 | set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot)))); | |
882 | return 1; | |
883 | } | |
884 | ||
885 | int pud_clear_huge(pud_t *pud) | |
886 | { | |
887 | if (!pud_sect(*pud)) | |
888 | return 0; | |
889 | pud_clear(pud); | |
890 | return 1; | |
891 | } | |
892 | ||
893 | int pmd_clear_huge(pmd_t *pmd) | |
894 | { | |
895 | if (!pmd_sect(*pmd)) | |
896 | return 0; | |
897 | pmd_clear(pmd); | |
898 | return 1; | |
899 | } |