]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 | 3 | * Copyright (C) 1995 Linus Torvalds |
2d4a7167 | 4 | * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. |
f8eeb2e6 | 5 | * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar |
1da177e4 | 6 | */ |
a2bcd473 | 7 | #include <linux/sched.h> /* test_thread_flag(), ... */ |
68db0cf1 | 8 | #include <linux/sched/task_stack.h> /* task_stack_*(), ... */ |
a2bcd473 | 9 | #include <linux/kdebug.h> /* oops_begin/end, ... */ |
4cdf8dbe | 10 | #include <linux/extable.h> /* search_exception_tables */ |
57c8a661 | 11 | #include <linux/memblock.h> /* max_low_pfn */ |
9326638c | 12 | #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ |
a2bcd473 | 13 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ |
cdd6c482 | 14 | #include <linux/perf_event.h> /* perf_sw_event */ |
f672b49b | 15 | #include <linux/hugetlb.h> /* hstate_index_to_shift */ |
268bb0ce | 16 | #include <linux/prefetch.h> /* prefetchw */ |
56dd9470 | 17 | #include <linux/context_tracking.h> /* exception_enter(), ... */ |
70ffdb93 | 18 | #include <linux/uaccess.h> /* faulthandler_disabled() */ |
3425d934 | 19 | #include <linux/efi.h> /* efi_recover_from_page_fault()*/ |
50a7ca3c | 20 | #include <linux/mm_types.h> |
2d4a7167 | 21 | |
019132ff | 22 | #include <asm/cpufeature.h> /* boot_cpu_has, ... */ |
a2bcd473 IM |
23 | #include <asm/traps.h> /* dotraplinkage, ... */ |
24 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f40c3300 AL |
25 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
26 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
ba3e127e | 27 | #include <asm/vm86.h> /* struct vm86 */ |
019132ff | 28 | #include <asm/mmu_context.h> /* vma_pkey() */ |
3425d934 | 29 | #include <asm/efi.h> /* efi_recover_from_page_fault()*/ |
a1a371c4 | 30 | #include <asm/desc.h> /* store_idt(), ... */ |
d876b673 | 31 | #include <asm/cpu_entry_area.h> /* exception stack */ |
186525bd | 32 | #include <asm/pgtable_areas.h> /* VMALLOC_START, ... */ |
1da177e4 | 33 | |
d34603b0 SA |
34 | #define CREATE_TRACE_POINTS |
35 | #include <asm/trace/exceptions.h> | |
36 | ||
b814d41f | 37 | /* |
b319eed0 IM |
38 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
39 | * handled by mmiotrace: | |
b814d41f | 40 | */ |
9326638c | 41 | static nokprobe_inline int |
62c9295f | 42 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 43 | { |
0fd0e3da PP |
44 | if (unlikely(is_kmmio_active())) |
45 | if (kmmio_handler(regs, addr) == 1) | |
46 | return -1; | |
0fd0e3da | 47 | return 0; |
86069782 PP |
48 | } |
49 | ||
1dc85be0 | 50 | /* |
2d4a7167 IM |
51 | * Prefetch quirks: |
52 | * | |
53 | * 32-bit mode: | |
54 | * | |
55 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
56 | * Check that here and ignore it. | |
1dc85be0 | 57 | * |
2d4a7167 | 58 | * 64-bit mode: |
1dc85be0 | 59 | * |
2d4a7167 IM |
60 | * Sometimes the CPU reports invalid exceptions on prefetch. |
61 | * Check that here and ignore it. | |
62 | * | |
63 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 64 | */ |
107a0367 IM |
65 | static inline int |
66 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
67 | unsigned char opcode, int *prefetch) | |
68 | { | |
69 | unsigned char instr_hi = opcode & 0xf0; | |
70 | unsigned char instr_lo = opcode & 0x0f; | |
71 | ||
72 | switch (instr_hi) { | |
73 | case 0x20: | |
74 | case 0x30: | |
75 | /* | |
76 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
77 | * In X86_64 long mode, the CPU will signal invalid | |
78 | * opcode if some of these prefixes are present so | |
79 | * X86_64 will never get here anyway | |
80 | */ | |
81 | return ((instr_lo & 7) == 0x6); | |
82 | #ifdef CONFIG_X86_64 | |
83 | case 0x40: | |
84 | /* | |
85 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
86 | * Need to figure out under what instruction mode the | |
87 | * instruction was issued. Could check the LDT for lm, | |
88 | * but for now it's good enough to assume that long | |
89 | * mode only uses well known segments or kernel. | |
90 | */ | |
318f5a2a | 91 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
92 | #endif |
93 | case 0x60: | |
94 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
95 | return (instr_lo & 0xC) == 0x4; | |
96 | case 0xF0: | |
97 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
98 | return !instr_lo || (instr_lo>>1) == 1; | |
99 | case 0x00: | |
100 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
101 | if (probe_kernel_address(instr, opcode)) | |
102 | return 0; | |
103 | ||
104 | *prefetch = (instr_lo == 0xF) && | |
105 | (opcode == 0x0D || opcode == 0x18); | |
106 | return 0; | |
107 | default: | |
108 | return 0; | |
109 | } | |
110 | } | |
111 | ||
2d4a7167 IM |
112 | static int |
113 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 114 | { |
2d4a7167 | 115 | unsigned char *max_instr; |
ab2bf0c1 | 116 | unsigned char *instr; |
33cb5243 | 117 | int prefetch = 0; |
1da177e4 | 118 | |
3085354d IM |
119 | /* |
120 | * If it was a exec (instruction fetch) fault on NX page, then | |
121 | * do not ignore the fault: | |
122 | */ | |
1067f030 | 123 | if (error_code & X86_PF_INSTR) |
1da177e4 | 124 | return 0; |
1dc85be0 | 125 | |
107a0367 | 126 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 127 | max_instr = instr + 15; |
1da177e4 | 128 | |
d31bf07f | 129 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
130 | return 0; |
131 | ||
107a0367 | 132 | while (instr < max_instr) { |
2d4a7167 | 133 | unsigned char opcode; |
1da177e4 | 134 | |
ab2bf0c1 | 135 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 136 | break; |
1da177e4 | 137 | |
1da177e4 LT |
138 | instr++; |
139 | ||
107a0367 | 140 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 141 | break; |
1da177e4 LT |
142 | } |
143 | return prefetch; | |
144 | } | |
145 | ||
f2f13a85 IM |
146 | DEFINE_SPINLOCK(pgd_lock); |
147 | LIST_HEAD(pgd_list); | |
148 | ||
149 | #ifdef CONFIG_X86_32 | |
150 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 151 | { |
f2f13a85 IM |
152 | unsigned index = pgd_index(address); |
153 | pgd_t *pgd_k; | |
e0c4f675 | 154 | p4d_t *p4d, *p4d_k; |
f2f13a85 IM |
155 | pud_t *pud, *pud_k; |
156 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 157 | |
f2f13a85 IM |
158 | pgd += index; |
159 | pgd_k = init_mm.pgd + index; | |
160 | ||
161 | if (!pgd_present(*pgd_k)) | |
162 | return NULL; | |
163 | ||
164 | /* | |
165 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
166 | * and redundant with the set_pmd() on non-PAE. As would | |
e0c4f675 | 167 | * set_p4d/set_pud. |
f2f13a85 | 168 | */ |
e0c4f675 KS |
169 | p4d = p4d_offset(pgd, address); |
170 | p4d_k = p4d_offset(pgd_k, address); | |
171 | if (!p4d_present(*p4d_k)) | |
172 | return NULL; | |
173 | ||
174 | pud = pud_offset(p4d, address); | |
175 | pud_k = pud_offset(p4d_k, address); | |
f2f13a85 IM |
176 | if (!pud_present(*pud_k)) |
177 | return NULL; | |
178 | ||
179 | pmd = pmd_offset(pud, address); | |
180 | pmd_k = pmd_offset(pud_k, address); | |
f2f13a85 | 181 | |
8e998fc2 | 182 | if (pmd_present(*pmd) != pmd_present(*pmd_k)) |
f2f13a85 | 183 | set_pmd(pmd, *pmd_k); |
8e998fc2 JR |
184 | |
185 | if (!pmd_present(*pmd_k)) | |
186 | return NULL; | |
b8bcfe99 | 187 | else |
51b75b5b | 188 | BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k)); |
f2f13a85 IM |
189 | |
190 | return pmd_k; | |
191 | } | |
192 | ||
193 | void vmalloc_sync_all(void) | |
194 | { | |
195 | unsigned long address; | |
196 | ||
197 | if (SHARED_KERNEL_PMD) | |
198 | return; | |
199 | ||
200 | for (address = VMALLOC_START & PMD_MASK; | |
9a62d200 | 201 | address >= TASK_SIZE_MAX && address < VMALLOC_END; |
f2f13a85 | 202 | address += PMD_SIZE) { |
f2f13a85 IM |
203 | struct page *page; |
204 | ||
a79e53d8 | 205 | spin_lock(&pgd_lock); |
f2f13a85 | 206 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 207 | spinlock_t *pgt_lock; |
617d34d9 | 208 | |
a79e53d8 | 209 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
210 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
211 | ||
212 | spin_lock(pgt_lock); | |
8e998fc2 | 213 | vmalloc_sync_one(page_address(page), address); |
617d34d9 | 214 | spin_unlock(pgt_lock); |
f2f13a85 | 215 | } |
a79e53d8 | 216 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
217 | } |
218 | } | |
219 | ||
220 | /* | |
221 | * 32-bit: | |
222 | * | |
223 | * Handle a fault on the vmalloc or module mapping area | |
224 | */ | |
9326638c | 225 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
226 | { |
227 | unsigned long pgd_paddr; | |
228 | pmd_t *pmd_k; | |
229 | pte_t *pte_k; | |
230 | ||
231 | /* Make sure we are in vmalloc area: */ | |
232 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
233 | return -1; | |
234 | ||
235 | /* | |
236 | * Synchronize this task's top level page-table | |
237 | * with the 'reference' page table. | |
238 | * | |
239 | * Do _not_ use "current" here. We might be inside | |
240 | * an interrupt in the middle of a task switch.. | |
241 | */ | |
6c690ee1 | 242 | pgd_paddr = read_cr3_pa(); |
f2f13a85 IM |
243 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); |
244 | if (!pmd_k) | |
245 | return -1; | |
246 | ||
18a95521 | 247 | if (pmd_large(*pmd_k)) |
f4eafd8b TK |
248 | return 0; |
249 | ||
f2f13a85 IM |
250 | pte_k = pte_offset_kernel(pmd_k, address); |
251 | if (!pte_present(*pte_k)) | |
252 | return -1; | |
253 | ||
254 | return 0; | |
255 | } | |
9326638c | 256 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
257 | |
258 | /* | |
259 | * Did it hit the DOS screen memory VA from vm86 mode? | |
260 | */ | |
261 | static inline void | |
262 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
263 | struct task_struct *tsk) | |
264 | { | |
9fda6a06 | 265 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
266 | unsigned long bit; |
267 | ||
9fda6a06 | 268 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
269 | return; |
270 | ||
271 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
272 | if (bit < 32) | |
9fda6a06 BG |
273 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
274 | #endif | |
33cb5243 | 275 | } |
1da177e4 | 276 | |
087975b0 | 277 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 278 | { |
087975b0 AM |
279 | return pfn < max_low_pfn; |
280 | } | |
1156e098 | 281 | |
087975b0 AM |
282 | static void dump_pagetable(unsigned long address) |
283 | { | |
6c690ee1 | 284 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 285 | pgd_t *pgd = &base[pgd_index(address)]; |
e0c4f675 KS |
286 | p4d_t *p4d; |
287 | pud_t *pud; | |
087975b0 AM |
288 | pmd_t *pmd; |
289 | pte_t *pte; | |
2d4a7167 | 290 | |
1156e098 | 291 | #ifdef CONFIG_X86_PAE |
39e48d9b | 292 | pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); |
087975b0 AM |
293 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) |
294 | goto out; | |
39e48d9b JB |
295 | #define pr_pde pr_cont |
296 | #else | |
297 | #define pr_pde pr_info | |
1156e098 | 298 | #endif |
e0c4f675 KS |
299 | p4d = p4d_offset(pgd, address); |
300 | pud = pud_offset(p4d, address); | |
301 | pmd = pmd_offset(pud, address); | |
39e48d9b JB |
302 | pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); |
303 | #undef pr_pde | |
1156e098 HH |
304 | |
305 | /* | |
306 | * We must not directly access the pte in the highpte | |
307 | * case if the page table is located in highmem. | |
308 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 309 | * it's allocated already: |
1156e098 | 310 | */ |
087975b0 AM |
311 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
312 | goto out; | |
1156e098 | 313 | |
087975b0 | 314 | pte = pte_offset_kernel(pmd, address); |
39e48d9b | 315 | pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); |
087975b0 | 316 | out: |
39e48d9b | 317 | pr_cont("\n"); |
f2f13a85 IM |
318 | } |
319 | ||
320 | #else /* CONFIG_X86_64: */ | |
321 | ||
322 | void vmalloc_sync_all(void) | |
323 | { | |
5372e155 | 324 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
325 | } |
326 | ||
327 | /* | |
328 | * 64-bit: | |
329 | * | |
330 | * Handle a fault on the vmalloc area | |
f2f13a85 | 331 | */ |
9326638c | 332 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 | 333 | { |
565977a3 TK |
334 | pgd_t *pgd, *pgd_k; |
335 | p4d_t *p4d, *p4d_k; | |
336 | pud_t *pud; | |
337 | pmd_t *pmd; | |
338 | pte_t *pte; | |
f2f13a85 IM |
339 | |
340 | /* Make sure we are in vmalloc area: */ | |
341 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
342 | return -1; | |
343 | ||
344 | /* | |
345 | * Copy kernel mappings over when needed. This can also | |
346 | * happen within a race in page table update. In the later | |
347 | * case just flush: | |
348 | */ | |
6c690ee1 | 349 | pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); |
565977a3 TK |
350 | pgd_k = pgd_offset_k(address); |
351 | if (pgd_none(*pgd_k)) | |
f2f13a85 IM |
352 | return -1; |
353 | ||
ed7588d5 | 354 | if (pgtable_l5_enabled()) { |
36b3a772 | 355 | if (pgd_none(*pgd)) { |
565977a3 | 356 | set_pgd(pgd, *pgd_k); |
36b3a772 AL |
357 | arch_flush_lazy_mmu_mode(); |
358 | } else { | |
565977a3 | 359 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k)); |
36b3a772 | 360 | } |
1160c277 | 361 | } |
f2f13a85 | 362 | |
b50858ce KS |
363 | /* With 4-level paging, copying happens on the p4d level. */ |
364 | p4d = p4d_offset(pgd, address); | |
565977a3 TK |
365 | p4d_k = p4d_offset(pgd_k, address); |
366 | if (p4d_none(*p4d_k)) | |
b50858ce KS |
367 | return -1; |
368 | ||
ed7588d5 | 369 | if (p4d_none(*p4d) && !pgtable_l5_enabled()) { |
565977a3 | 370 | set_p4d(p4d, *p4d_k); |
b50858ce KS |
371 | arch_flush_lazy_mmu_mode(); |
372 | } else { | |
565977a3 | 373 | BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k)); |
b50858ce KS |
374 | } |
375 | ||
36b3a772 | 376 | BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4); |
f2f13a85 | 377 | |
b50858ce | 378 | pud = pud_offset(p4d, address); |
565977a3 | 379 | if (pud_none(*pud)) |
f2f13a85 IM |
380 | return -1; |
381 | ||
18a95521 | 382 | if (pud_large(*pud)) |
f4eafd8b TK |
383 | return 0; |
384 | ||
f2f13a85 | 385 | pmd = pmd_offset(pud, address); |
565977a3 | 386 | if (pmd_none(*pmd)) |
f2f13a85 IM |
387 | return -1; |
388 | ||
18a95521 | 389 | if (pmd_large(*pmd)) |
f4eafd8b TK |
390 | return 0; |
391 | ||
f2f13a85 | 392 | pte = pte_offset_kernel(pmd, address); |
565977a3 TK |
393 | if (!pte_present(*pte)) |
394 | return -1; | |
f2f13a85 IM |
395 | |
396 | return 0; | |
397 | } | |
9326638c | 398 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 399 | |
e05139f2 | 400 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 401 | static const char errata93_warning[] = |
ad361c98 JP |
402 | KERN_ERR |
403 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
404 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
405 | "******* Please consider a BIOS update.\n" | |
406 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 407 | #endif |
f2f13a85 IM |
408 | |
409 | /* | |
410 | * No vm86 mode in 64-bit mode: | |
411 | */ | |
412 | static inline void | |
413 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
414 | struct task_struct *tsk) | |
415 | { | |
416 | } | |
417 | ||
418 | static int bad_address(void *p) | |
419 | { | |
420 | unsigned long dummy; | |
421 | ||
422 | return probe_kernel_address((unsigned long *)p, dummy); | |
423 | } | |
424 | ||
425 | static void dump_pagetable(unsigned long address) | |
426 | { | |
6c690ee1 | 427 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 428 | pgd_t *pgd = base + pgd_index(address); |
e0c4f675 | 429 | p4d_t *p4d; |
1da177e4 LT |
430 | pud_t *pud; |
431 | pmd_t *pmd; | |
432 | pte_t *pte; | |
433 | ||
2d4a7167 IM |
434 | if (bad_address(pgd)) |
435 | goto bad; | |
436 | ||
39e48d9b | 437 | pr_info("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
438 | |
439 | if (!pgd_present(*pgd)) | |
440 | goto out; | |
1da177e4 | 441 | |
e0c4f675 KS |
442 | p4d = p4d_offset(pgd, address); |
443 | if (bad_address(p4d)) | |
444 | goto bad; | |
445 | ||
39e48d9b | 446 | pr_cont("P4D %lx ", p4d_val(*p4d)); |
e0c4f675 KS |
447 | if (!p4d_present(*p4d) || p4d_large(*p4d)) |
448 | goto out; | |
449 | ||
450 | pud = pud_offset(p4d, address); | |
2d4a7167 IM |
451 | if (bad_address(pud)) |
452 | goto bad; | |
453 | ||
39e48d9b | 454 | pr_cont("PUD %lx ", pud_val(*pud)); |
b5360222 | 455 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 456 | goto out; |
1da177e4 LT |
457 | |
458 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
459 | if (bad_address(pmd)) |
460 | goto bad; | |
461 | ||
39e48d9b | 462 | pr_cont("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
463 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
464 | goto out; | |
1da177e4 LT |
465 | |
466 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
467 | if (bad_address(pte)) |
468 | goto bad; | |
469 | ||
39e48d9b | 470 | pr_cont("PTE %lx", pte_val(*pte)); |
2d4a7167 | 471 | out: |
39e48d9b | 472 | pr_cont("\n"); |
1da177e4 LT |
473 | return; |
474 | bad: | |
39e48d9b | 475 | pr_info("BAD\n"); |
8c938f9f IM |
476 | } |
477 | ||
f2f13a85 | 478 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 479 | |
2d4a7167 IM |
480 | /* |
481 | * Workaround for K8 erratum #93 & buggy BIOS. | |
482 | * | |
483 | * BIOS SMM functions are required to use a specific workaround | |
484 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
485 | * | |
486 | * A lot of BIOS that didn't get tested properly miss this. | |
487 | * | |
488 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
489 | * Try to work around it here. | |
490 | * | |
491 | * Note we only handle faults in kernel here. | |
492 | * Does nothing on 32-bit. | |
fdfe8aa8 | 493 | */ |
33cb5243 | 494 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 495 | { |
e05139f2 JB |
496 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
497 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
498 | || boot_cpu_data.x86 != 0xf) | |
499 | return 0; | |
500 | ||
65ea5b03 | 501 | if (address != regs->ip) |
1da177e4 | 502 | return 0; |
2d4a7167 | 503 | |
33cb5243 | 504 | if ((address >> 32) != 0) |
1da177e4 | 505 | return 0; |
2d4a7167 | 506 | |
1da177e4 | 507 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
508 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
509 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 510 | printk_once(errata93_warning); |
65ea5b03 | 511 | regs->ip = address; |
1da177e4 LT |
512 | return 1; |
513 | } | |
fdfe8aa8 | 514 | #endif |
1da177e4 | 515 | return 0; |
33cb5243 | 516 | } |
1da177e4 | 517 | |
35f3266f | 518 | /* |
2d4a7167 IM |
519 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
520 | * to illegal addresses >4GB. | |
521 | * | |
522 | * We catch this in the page fault handler because these addresses | |
523 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
524 | * segment in LDT is compatibility mode. |
525 | */ | |
526 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
527 | { | |
528 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 529 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
530 | return 1; |
531 | #endif | |
532 | return 0; | |
533 | } | |
534 | ||
29caf2f9 HH |
535 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
536 | { | |
537 | #ifdef CONFIG_X86_F00F_BUG | |
538 | unsigned long nr; | |
2d4a7167 | 539 | |
29caf2f9 | 540 | /* |
2d4a7167 | 541 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 542 | */ |
e2604b49 | 543 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
544 | nr = (address - idt_descr.address) >> 3; |
545 | ||
546 | if (nr == 6) { | |
547 | do_invalid_op(regs, 0); | |
548 | return 1; | |
549 | } | |
550 | } | |
551 | #endif | |
552 | return 0; | |
553 | } | |
554 | ||
a1a371c4 AL |
555 | static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index) |
556 | { | |
557 | u32 offset = (index >> 3) * sizeof(struct desc_struct); | |
558 | unsigned long addr; | |
559 | struct ldttss_desc desc; | |
560 | ||
561 | if (index == 0) { | |
562 | pr_alert("%s: NULL\n", name); | |
563 | return; | |
564 | } | |
565 | ||
566 | if (offset + sizeof(struct ldttss_desc) >= gdt->size) { | |
567 | pr_alert("%s: 0x%hx -- out of bounds\n", name, index); | |
568 | return; | |
569 | } | |
570 | ||
571 | if (probe_kernel_read(&desc, (void *)(gdt->address + offset), | |
572 | sizeof(struct ldttss_desc))) { | |
573 | pr_alert("%s: 0x%hx -- GDT entry is not readable\n", | |
574 | name, index); | |
575 | return; | |
576 | } | |
577 | ||
5ccd3528 | 578 | addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24); |
a1a371c4 AL |
579 | #ifdef CONFIG_X86_64 |
580 | addr |= ((u64)desc.base3 << 32); | |
581 | #endif | |
582 | pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n", | |
583 | name, index, addr, (desc.limit0 | (desc.limit1 << 16))); | |
584 | } | |
585 | ||
2d4a7167 | 586 | static void |
a2aa52ab | 587 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address) |
b3279c7f | 588 | { |
1156e098 HH |
589 | if (!oops_may_print()) |
590 | return; | |
591 | ||
1067f030 | 592 | if (error_code & X86_PF_INSTR) { |
93809be8 | 593 | unsigned int level; |
426e34cc MF |
594 | pgd_t *pgd; |
595 | pte_t *pte; | |
2d4a7167 | 596 | |
6c690ee1 | 597 | pgd = __va(read_cr3_pa()); |
426e34cc MF |
598 | pgd += pgd_index(address); |
599 | ||
600 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 601 | |
8f766149 | 602 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
d79d0d8a DV |
603 | pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", |
604 | from_kuid(&init_user_ns, current_uid())); | |
eff50c34 JK |
605 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
606 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 607 | (__read_cr4() & X86_CR4_SMEP)) |
d79d0d8a DV |
608 | pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", |
609 | from_kuid(&init_user_ns, current_uid())); | |
1156e098 | 610 | } |
1156e098 | 611 | |
f28b11a2 | 612 | if (address < PAGE_SIZE && !user_mode(regs)) |
ea2f8d60 | 613 | pr_alert("BUG: kernel NULL pointer dereference, address: %px\n", |
f28b11a2 SC |
614 | (void *)address); |
615 | else | |
ea2f8d60 | 616 | pr_alert("BUG: unable to handle page fault for address: %px\n", |
f28b11a2 | 617 | (void *)address); |
2d4a7167 | 618 | |
ea2f8d60 | 619 | pr_alert("#PF: %s %s in %s mode\n", |
18ea35c5 SC |
620 | (error_code & X86_PF_USER) ? "user" : "supervisor", |
621 | (error_code & X86_PF_INSTR) ? "instruction fetch" : | |
622 | (error_code & X86_PF_WRITE) ? "write access" : | |
623 | "read access", | |
624 | user_mode(regs) ? "user" : "kernel"); | |
625 | pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code, | |
626 | !(error_code & X86_PF_PROT) ? "not-present page" : | |
627 | (error_code & X86_PF_RSVD) ? "reserved bit violation" : | |
628 | (error_code & X86_PF_PK) ? "protection keys violation" : | |
629 | "permissions violation"); | |
a2aa52ab | 630 | |
a1a371c4 AL |
631 | if (!(error_code & X86_PF_USER) && user_mode(regs)) { |
632 | struct desc_ptr idt, gdt; | |
633 | u16 ldtr, tr; | |
634 | ||
a1a371c4 AL |
635 | /* |
636 | * This can happen for quite a few reasons. The more obvious | |
637 | * ones are faults accessing the GDT, or LDT. Perhaps | |
638 | * surprisingly, if the CPU tries to deliver a benign or | |
639 | * contributory exception from user code and gets a page fault | |
640 | * during delivery, the page fault can be delivered as though | |
641 | * it originated directly from user code. This could happen | |
642 | * due to wrong permissions on the IDT, GDT, LDT, TSS, or | |
643 | * kernel or IST stack. | |
644 | */ | |
645 | store_idt(&idt); | |
646 | ||
647 | /* Usable even on Xen PV -- it's just slow. */ | |
648 | native_store_gdt(&gdt); | |
649 | ||
650 | pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n", | |
651 | idt.address, idt.size, gdt.address, gdt.size); | |
652 | ||
653 | store_ldt(ldtr); | |
654 | show_ldttss(&gdt, "LDTR", ldtr); | |
655 | ||
656 | store_tr(tr); | |
657 | show_ldttss(&gdt, "TR", tr); | |
658 | } | |
659 | ||
b3279c7f HH |
660 | dump_pagetable(address); |
661 | } | |
662 | ||
2d4a7167 IM |
663 | static noinline void |
664 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
665 | unsigned long address) | |
1da177e4 | 666 | { |
2d4a7167 IM |
667 | struct task_struct *tsk; |
668 | unsigned long flags; | |
669 | int sig; | |
670 | ||
671 | flags = oops_begin(); | |
672 | tsk = current; | |
673 | sig = SIGKILL; | |
1209140c | 674 | |
1da177e4 | 675 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 676 | tsk->comm, address); |
1da177e4 | 677 | dump_pagetable(address); |
2d4a7167 | 678 | |
22f5991c | 679 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 680 | sig = 0; |
2d4a7167 | 681 | |
874d93d1 | 682 | oops_end(flags, regs, sig); |
1da177e4 LT |
683 | } |
684 | ||
e49d3cbe AL |
685 | static void set_signal_archinfo(unsigned long address, |
686 | unsigned long error_code) | |
687 | { | |
688 | struct task_struct *tsk = current; | |
689 | ||
690 | /* | |
691 | * To avoid leaking information about the kernel page | |
692 | * table layout, pretend that user-mode accesses to | |
693 | * kernel addresses are always protection faults. | |
e0a446ce AL |
694 | * |
695 | * NB: This means that failed vsyscalls with vsyscall=none | |
696 | * will have the PROT bit. This doesn't leak any | |
697 | * information and does not appear to cause any problems. | |
e49d3cbe AL |
698 | */ |
699 | if (address >= TASK_SIZE_MAX) | |
700 | error_code |= X86_PF_PROT; | |
701 | ||
702 | tsk->thread.trap_nr = X86_TRAP_PF; | |
703 | tsk->thread.error_code = error_code | X86_PF_USER; | |
704 | tsk->thread.cr2 = address; | |
705 | } | |
706 | ||
2d4a7167 IM |
707 | static noinline void |
708 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 709 | unsigned long address, int signal, int si_code) |
92181f19 NP |
710 | { |
711 | struct task_struct *tsk = current; | |
92181f19 NP |
712 | unsigned long flags; |
713 | int sig; | |
92181f19 | 714 | |
ebb53e25 AL |
715 | if (user_mode(regs)) { |
716 | /* | |
717 | * This is an implicit supervisor-mode access from user | |
718 | * mode. Bypass all the kernel-mode recovery code and just | |
719 | * OOPS. | |
720 | */ | |
721 | goto oops; | |
722 | } | |
723 | ||
2d4a7167 | 724 | /* Are we prepared to handle this kernel fault? */ |
81fd9c18 | 725 | if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) { |
c026b359 PZ |
726 | /* |
727 | * Any interrupt that takes a fault gets the fixup. This makes | |
728 | * the below recursive fault logic only apply to a faults from | |
729 | * task context. | |
730 | */ | |
731 | if (in_interrupt()) | |
732 | return; | |
733 | ||
734 | /* | |
735 | * Per the above we're !in_interrupt(), aka. task context. | |
736 | * | |
737 | * In this case we need to make sure we're not recursively | |
738 | * faulting through the emulate_vsyscall() logic. | |
739 | */ | |
2a53ccbc | 740 | if (current->thread.sig_on_uaccess_err && signal) { |
e49d3cbe | 741 | set_signal_archinfo(address, error_code); |
4fc34901 AL |
742 | |
743 | /* XXX: hwpoison faults will set the wrong code. */ | |
2e1661d2 | 744 | force_sig_fault(signal, si_code, (void __user *)address); |
4fc34901 | 745 | } |
c026b359 PZ |
746 | |
747 | /* | |
748 | * Barring that, we can do the fixup and be happy. | |
749 | */ | |
92181f19 | 750 | return; |
4fc34901 | 751 | } |
92181f19 | 752 | |
6271cfdf AL |
753 | #ifdef CONFIG_VMAP_STACK |
754 | /* | |
755 | * Stack overflow? During boot, we can fault near the initial | |
756 | * stack in the direct map, but that's not an overflow -- check | |
757 | * that we're in vmalloc space to avoid this. | |
758 | */ | |
759 | if (is_vmalloc_addr((void *)address) && | |
760 | (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || | |
761 | address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { | |
d876b673 | 762 | unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *); |
6271cfdf AL |
763 | /* |
764 | * We're likely to be running with very little stack space | |
765 | * left. It's plausible that we'd hit this condition but | |
766 | * double-fault even before we get this far, in which case | |
767 | * we're fine: the double-fault handler will deal with it. | |
768 | * | |
769 | * We don't want to make it all the way into the oops code | |
770 | * and then double-fault, though, because we're likely to | |
771 | * break the console driver and lose most of the stack dump. | |
772 | */ | |
773 | asm volatile ("movq %[stack], %%rsp\n\t" | |
774 | "call handle_stack_overflow\n\t" | |
775 | "1: jmp 1b" | |
f5caf621 | 776 | : ASM_CALL_CONSTRAINT |
6271cfdf AL |
777 | : "D" ("kernel stack overflow (page fault)"), |
778 | "S" (regs), "d" (address), | |
779 | [stack] "rm" (stack)); | |
780 | unreachable(); | |
781 | } | |
782 | #endif | |
783 | ||
92181f19 | 784 | /* |
2d4a7167 IM |
785 | * 32-bit: |
786 | * | |
787 | * Valid to do another page fault here, because if this fault | |
788 | * had been triggered by is_prefetch fixup_exception would have | |
789 | * handled it. | |
790 | * | |
791 | * 64-bit: | |
92181f19 | 792 | * |
2d4a7167 | 793 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
794 | */ |
795 | if (is_prefetch(regs, error_code, address)) | |
796 | return; | |
797 | ||
798 | if (is_errata93(regs, address)) | |
799 | return; | |
800 | ||
3425d934 SP |
801 | /* |
802 | * Buggy firmware could access regions which might page fault, try to | |
803 | * recover from such faults. | |
804 | */ | |
805 | if (IS_ENABLED(CONFIG_EFI)) | |
806 | efi_recover_from_page_fault(address); | |
807 | ||
ebb53e25 | 808 | oops: |
92181f19 NP |
809 | /* |
810 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 811 | * terminate things with extreme prejudice: |
92181f19 | 812 | */ |
92181f19 | 813 | flags = oops_begin(); |
92181f19 NP |
814 | |
815 | show_fault_oops(regs, error_code, address); | |
816 | ||
a70857e4 | 817 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 818 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 819 | |
92181f19 NP |
820 | sig = SIGKILL; |
821 | if (__die("Oops", regs, error_code)) | |
822 | sig = 0; | |
2d4a7167 | 823 | |
92181f19 | 824 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 825 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 826 | |
92181f19 | 827 | oops_end(flags, regs, sig); |
92181f19 NP |
828 | } |
829 | ||
2d4a7167 IM |
830 | /* |
831 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
832 | * sysctl is set: | |
833 | */ | |
834 | static inline void | |
835 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
836 | unsigned long address, struct task_struct *tsk) | |
837 | { | |
ba54d856 BP |
838 | const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; |
839 | ||
2d4a7167 IM |
840 | if (!unhandled_signal(tsk, SIGSEGV)) |
841 | return; | |
842 | ||
843 | if (!printk_ratelimit()) | |
844 | return; | |
845 | ||
10a7e9d8 | 846 | printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", |
ba54d856 | 847 | loglvl, tsk->comm, task_pid_nr(tsk), address, |
2d4a7167 IM |
848 | (void *)regs->ip, (void *)regs->sp, error_code); |
849 | ||
850 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
851 | ||
852 | printk(KERN_CONT "\n"); | |
ba54d856 | 853 | |
342db04a | 854 | show_opcodes(regs, loglvl); |
2d4a7167 IM |
855 | } |
856 | ||
02e983b7 DH |
857 | /* |
858 | * The (legacy) vsyscall page is the long page in the kernel portion | |
859 | * of the address space that has user-accessible permissions. | |
860 | */ | |
861 | static bool is_vsyscall_vaddr(unsigned long vaddr) | |
862 | { | |
3ae0ad92 | 863 | return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR); |
02e983b7 DH |
864 | } |
865 | ||
2d4a7167 IM |
866 | static void |
867 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
419ceeb1 | 868 | unsigned long address, u32 pkey, int si_code) |
92181f19 NP |
869 | { |
870 | struct task_struct *tsk = current; | |
871 | ||
872 | /* User mode accesses just cause a SIGSEGV */ | |
6ea59b07 | 873 | if (user_mode(regs) && (error_code & X86_PF_USER)) { |
92181f19 | 874 | /* |
2d4a7167 | 875 | * It's possible to have interrupts off here: |
92181f19 NP |
876 | */ |
877 | local_irq_enable(); | |
878 | ||
879 | /* | |
880 | * Valid to do another page fault here because this one came | |
2d4a7167 | 881 | * from user space: |
92181f19 NP |
882 | */ |
883 | if (is_prefetch(regs, error_code, address)) | |
884 | return; | |
885 | ||
886 | if (is_errata100(regs, address)) | |
887 | return; | |
888 | ||
dc4fac84 AL |
889 | /* |
890 | * To avoid leaking information about the kernel page table | |
891 | * layout, pretend that user-mode accesses to kernel addresses | |
892 | * are always protection faults. | |
893 | */ | |
894 | if (address >= TASK_SIZE_MAX) | |
1067f030 | 895 | error_code |= X86_PF_PROT; |
3ae36655 | 896 | |
e575a86f | 897 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
898 | show_signal_msg(regs, error_code, address, tsk); |
899 | ||
e49d3cbe | 900 | set_signal_archinfo(address, error_code); |
92181f19 | 901 | |
9db812db | 902 | if (si_code == SEGV_PKUERR) |
419ceeb1 | 903 | force_sig_pkuerr((void __user *)address, pkey); |
9db812db | 904 | |
2e1661d2 | 905 | force_sig_fault(SIGSEGV, si_code, (void __user *)address); |
2d4a7167 | 906 | |
92181f19 NP |
907 | return; |
908 | } | |
909 | ||
910 | if (is_f00f_bug(regs, address)) | |
911 | return; | |
912 | ||
4fc34901 | 913 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
914 | } |
915 | ||
2d4a7167 IM |
916 | static noinline void |
917 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
768fd9c6 | 918 | unsigned long address) |
92181f19 | 919 | { |
419ceeb1 | 920 | __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR); |
92181f19 NP |
921 | } |
922 | ||
2d4a7167 IM |
923 | static void |
924 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
419ceeb1 | 925 | unsigned long address, u32 pkey, int si_code) |
92181f19 NP |
926 | { |
927 | struct mm_struct *mm = current->mm; | |
92181f19 NP |
928 | /* |
929 | * Something tried to access memory that isn't in our memory map.. | |
930 | * Fix it, but check if it's kernel or user first.. | |
931 | */ | |
932 | up_read(&mm->mmap_sem); | |
933 | ||
aba1ecd3 | 934 | __bad_area_nosemaphore(regs, error_code, address, pkey, si_code); |
92181f19 NP |
935 | } |
936 | ||
2d4a7167 IM |
937 | static noinline void |
938 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 | 939 | { |
419ceeb1 | 940 | __bad_area(regs, error_code, address, 0, SEGV_MAPERR); |
92181f19 NP |
941 | } |
942 | ||
33a709b2 DH |
943 | static inline bool bad_area_access_from_pkeys(unsigned long error_code, |
944 | struct vm_area_struct *vma) | |
945 | { | |
07f146f5 DH |
946 | /* This code is always called on the current mm */ |
947 | bool foreign = false; | |
948 | ||
33a709b2 DH |
949 | if (!boot_cpu_has(X86_FEATURE_OSPKE)) |
950 | return false; | |
1067f030 | 951 | if (error_code & X86_PF_PK) |
33a709b2 | 952 | return true; |
07f146f5 | 953 | /* this checks permission keys on the VMA: */ |
1067f030 RN |
954 | if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), |
955 | (error_code & X86_PF_INSTR), foreign)) | |
07f146f5 | 956 | return true; |
33a709b2 | 957 | return false; |
92181f19 NP |
958 | } |
959 | ||
2d4a7167 IM |
960 | static noinline void |
961 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 962 | unsigned long address, struct vm_area_struct *vma) |
92181f19 | 963 | { |
019132ff DH |
964 | /* |
965 | * This OSPKE check is not strictly necessary at runtime. | |
966 | * But, doing it this way allows compiler optimizations | |
967 | * if pkeys are compiled out. | |
968 | */ | |
aba1ecd3 | 969 | if (bad_area_access_from_pkeys(error_code, vma)) { |
9db812db EB |
970 | /* |
971 | * A protection key fault means that the PKRU value did not allow | |
972 | * access to some PTE. Userspace can figure out what PKRU was | |
973 | * from the XSAVE state. This function captures the pkey from | |
974 | * the vma and passes it to userspace so userspace can discover | |
975 | * which protection key was set on the PTE. | |
976 | * | |
977 | * If we get here, we know that the hardware signaled a X86_PF_PK | |
978 | * fault and that there was a VMA once we got in the fault | |
979 | * handler. It does *not* guarantee that the VMA we find here | |
980 | * was the one that we faulted on. | |
981 | * | |
982 | * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); | |
983 | * 2. T1 : set PKRU to deny access to pkey=4, touches page | |
984 | * 3. T1 : faults... | |
985 | * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); | |
986 | * 5. T1 : enters fault handler, takes mmap_sem, etc... | |
987 | * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really | |
988 | * faulted on a pte with its pkey=4. | |
989 | */ | |
aba1ecd3 | 990 | u32 pkey = vma_pkey(vma); |
9db812db | 991 | |
419ceeb1 | 992 | __bad_area(regs, error_code, address, pkey, SEGV_PKUERR); |
aba1ecd3 | 993 | } else { |
419ceeb1 | 994 | __bad_area(regs, error_code, address, 0, SEGV_ACCERR); |
aba1ecd3 | 995 | } |
92181f19 NP |
996 | } |
997 | ||
2d4a7167 | 998 | static void |
a6e04aa9 | 999 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
3d353901 | 1000 | vm_fault_t fault) |
92181f19 | 1001 | { |
2d4a7167 | 1002 | /* Kernel mode? Handle exceptions or die: */ |
1067f030 | 1003 | if (!(error_code & X86_PF_USER)) { |
4fc34901 | 1004 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
1005 | return; |
1006 | } | |
2d4a7167 | 1007 | |
cd1b68f0 | 1008 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
1009 | if (is_prefetch(regs, error_code, address)) |
1010 | return; | |
2d4a7167 | 1011 | |
e49d3cbe | 1012 | set_signal_archinfo(address, error_code); |
2d4a7167 | 1013 | |
a6e04aa9 | 1014 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 1015 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
318759b4 | 1016 | struct task_struct *tsk = current; |
40e55394 EB |
1017 | unsigned lsb = 0; |
1018 | ||
1019 | pr_err( | |
a6e04aa9 AK |
1020 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", |
1021 | tsk->comm, tsk->pid, address); | |
40e55394 EB |
1022 | if (fault & VM_FAULT_HWPOISON_LARGE) |
1023 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
1024 | if (fault & VM_FAULT_HWPOISON) | |
1025 | lsb = PAGE_SHIFT; | |
f8eac901 | 1026 | force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); |
40e55394 | 1027 | return; |
a6e04aa9 AK |
1028 | } |
1029 | #endif | |
2e1661d2 | 1030 | force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); |
92181f19 NP |
1031 | } |
1032 | ||
3a13c4d7 | 1033 | static noinline void |
2d4a7167 | 1034 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
25c102d8 | 1035 | unsigned long address, vm_fault_t fault) |
92181f19 | 1036 | { |
1067f030 | 1037 | if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { |
3a13c4d7 JW |
1038 | no_context(regs, error_code, address, 0, 0); |
1039 | return; | |
b80ef10e | 1040 | } |
b80ef10e | 1041 | |
2d4a7167 | 1042 | if (fault & VM_FAULT_OOM) { |
f8626854 | 1043 | /* Kernel mode? Handle exceptions or die: */ |
1067f030 | 1044 | if (!(error_code & X86_PF_USER)) { |
4fc34901 AL |
1045 | no_context(regs, error_code, address, |
1046 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 1047 | return; |
f8626854 AV |
1048 | } |
1049 | ||
c2d23f91 DR |
1050 | /* |
1051 | * We ran out of memory, call the OOM killer, and return the | |
1052 | * userspace (which will retry the fault, or kill us if we got | |
1053 | * oom-killed): | |
1054 | */ | |
1055 | pagefault_out_of_memory(); | |
2d4a7167 | 1056 | } else { |
f672b49b AK |
1057 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
1058 | VM_FAULT_HWPOISON_LARGE)) | |
27274f73 | 1059 | do_sigbus(regs, error_code, address, fault); |
33692f27 | 1060 | else if (fault & VM_FAULT_SIGSEGV) |
768fd9c6 | 1061 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 IM |
1062 | else |
1063 | BUG(); | |
1064 | } | |
92181f19 NP |
1065 | } |
1066 | ||
8fed6200 | 1067 | static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte) |
d8b57bb7 | 1068 | { |
1067f030 | 1069 | if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) |
d8b57bb7 | 1070 | return 0; |
2d4a7167 | 1071 | |
1067f030 | 1072 | if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) |
d8b57bb7 TG |
1073 | return 0; |
1074 | ||
1075 | return 1; | |
1076 | } | |
1077 | ||
5b727a3b | 1078 | /* |
2d4a7167 IM |
1079 | * Handle a spurious fault caused by a stale TLB entry. |
1080 | * | |
1081 | * This allows us to lazily refresh the TLB when increasing the | |
1082 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
1083 | * eagerly is very expensive since that implies doing a full | |
1084 | * cross-processor TLB flush, even if no stale TLB entries exist | |
1085 | * on other processors. | |
1086 | * | |
31668511 DV |
1087 | * Spurious faults may only occur if the TLB contains an entry with |
1088 | * fewer permission than the page table entry. Non-present (P = 0) | |
1089 | * and reserved bit (R = 1) faults are never spurious. | |
1090 | * | |
5b727a3b JF |
1091 | * There are no security implications to leaving a stale TLB when |
1092 | * increasing the permissions on a page. | |
31668511 DV |
1093 | * |
1094 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
1095 | * | |
1096 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
1097 | * (Optional Invalidation). | |
5b727a3b | 1098 | */ |
9326638c | 1099 | static noinline int |
8fed6200 | 1100 | spurious_kernel_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
1101 | { |
1102 | pgd_t *pgd; | |
e0c4f675 | 1103 | p4d_t *p4d; |
5b727a3b JF |
1104 | pud_t *pud; |
1105 | pmd_t *pmd; | |
1106 | pte_t *pte; | |
3c3e5694 | 1107 | int ret; |
5b727a3b | 1108 | |
31668511 DV |
1109 | /* |
1110 | * Only writes to RO or instruction fetches from NX may cause | |
1111 | * spurious faults. | |
1112 | * | |
1113 | * These could be from user or supervisor accesses but the TLB | |
1114 | * is only lazily flushed after a kernel mapping protection | |
1115 | * change, so user accesses are not expected to cause spurious | |
1116 | * faults. | |
1117 | */ | |
1067f030 RN |
1118 | if (error_code != (X86_PF_WRITE | X86_PF_PROT) && |
1119 | error_code != (X86_PF_INSTR | X86_PF_PROT)) | |
5b727a3b JF |
1120 | return 0; |
1121 | ||
1122 | pgd = init_mm.pgd + pgd_index(address); | |
1123 | if (!pgd_present(*pgd)) | |
1124 | return 0; | |
1125 | ||
e0c4f675 KS |
1126 | p4d = p4d_offset(pgd, address); |
1127 | if (!p4d_present(*p4d)) | |
1128 | return 0; | |
1129 | ||
1130 | if (p4d_large(*p4d)) | |
8fed6200 | 1131 | return spurious_kernel_fault_check(error_code, (pte_t *) p4d); |
e0c4f675 KS |
1132 | |
1133 | pud = pud_offset(p4d, address); | |
5b727a3b JF |
1134 | if (!pud_present(*pud)) |
1135 | return 0; | |
1136 | ||
d8b57bb7 | 1137 | if (pud_large(*pud)) |
8fed6200 | 1138 | return spurious_kernel_fault_check(error_code, (pte_t *) pud); |
d8b57bb7 | 1139 | |
5b727a3b JF |
1140 | pmd = pmd_offset(pud, address); |
1141 | if (!pmd_present(*pmd)) | |
1142 | return 0; | |
1143 | ||
d8b57bb7 | 1144 | if (pmd_large(*pmd)) |
8fed6200 | 1145 | return spurious_kernel_fault_check(error_code, (pte_t *) pmd); |
d8b57bb7 | 1146 | |
5b727a3b | 1147 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 1148 | if (!pte_present(*pte)) |
5b727a3b JF |
1149 | return 0; |
1150 | ||
8fed6200 | 1151 | ret = spurious_kernel_fault_check(error_code, pte); |
3c3e5694 SR |
1152 | if (!ret) |
1153 | return 0; | |
1154 | ||
1155 | /* | |
2d4a7167 IM |
1156 | * Make sure we have permissions in PMD. |
1157 | * If not, then there's a bug in the page tables: | |
3c3e5694 | 1158 | */ |
8fed6200 | 1159 | ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd); |
3c3e5694 | 1160 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); |
2d4a7167 | 1161 | |
3c3e5694 | 1162 | return ret; |
5b727a3b | 1163 | } |
8fed6200 | 1164 | NOKPROBE_SYMBOL(spurious_kernel_fault); |
5b727a3b | 1165 | |
abd4f750 | 1166 | int show_unhandled_signals = 1; |
1da177e4 | 1167 | |
2d4a7167 | 1168 | static inline int |
68da336a | 1169 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1170 | { |
07f146f5 DH |
1171 | /* This is only called for the current mm, so: */ |
1172 | bool foreign = false; | |
e8c6226d DH |
1173 | |
1174 | /* | |
1175 | * Read or write was blocked by protection keys. This is | |
1176 | * always an unconditional error and can never result in | |
1177 | * a follow-up action to resolve the fault, like a COW. | |
1178 | */ | |
1067f030 | 1179 | if (error_code & X86_PF_PK) |
e8c6226d DH |
1180 | return 1; |
1181 | ||
07f146f5 DH |
1182 | /* |
1183 | * Make sure to check the VMA so that we do not perform | |
1067f030 | 1184 | * faults just to hit a X86_PF_PK as soon as we fill in a |
07f146f5 DH |
1185 | * page. |
1186 | */ | |
1067f030 RN |
1187 | if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), |
1188 | (error_code & X86_PF_INSTR), foreign)) | |
07f146f5 | 1189 | return 1; |
33a709b2 | 1190 | |
1067f030 | 1191 | if (error_code & X86_PF_WRITE) { |
2d4a7167 | 1192 | /* write, present and write, not present: */ |
92181f19 NP |
1193 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1194 | return 1; | |
2d4a7167 | 1195 | return 0; |
92181f19 NP |
1196 | } |
1197 | ||
2d4a7167 | 1198 | /* read, present: */ |
1067f030 | 1199 | if (unlikely(error_code & X86_PF_PROT)) |
2d4a7167 IM |
1200 | return 1; |
1201 | ||
1202 | /* read, not present: */ | |
1203 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1204 | return 1; | |
1205 | ||
92181f19 NP |
1206 | return 0; |
1207 | } | |
1208 | ||
0973a06c HS |
1209 | static int fault_in_kernel_space(unsigned long address) |
1210 | { | |
3ae0ad92 DH |
1211 | /* |
1212 | * On 64-bit systems, the vsyscall page is at an address above | |
1213 | * TASK_SIZE_MAX, but is not considered part of the kernel | |
1214 | * address space. | |
1215 | */ | |
1216 | if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address)) | |
1217 | return false; | |
1218 | ||
d9517346 | 1219 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1220 | } |
1221 | ||
1da177e4 | 1222 | /* |
8fed6200 DH |
1223 | * Called for all faults where 'address' is part of the kernel address |
1224 | * space. Might get called for faults that originate from *code* that | |
1225 | * ran in userspace or the kernel. | |
1da177e4 | 1226 | */ |
8fed6200 DH |
1227 | static void |
1228 | do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code, | |
1229 | unsigned long address) | |
1da177e4 | 1230 | { |
367e3f1d DH |
1231 | /* |
1232 | * Protection keys exceptions only happen on user pages. We | |
1233 | * have no user pages in the kernel portion of the address | |
1234 | * space, so do not expect them here. | |
1235 | */ | |
1236 | WARN_ON_ONCE(hw_error_code & X86_PF_PK); | |
1da177e4 LT |
1237 | |
1238 | /* | |
8fed6200 | 1239 | * We can fault-in kernel-space virtual memory on-demand. The |
1da177e4 LT |
1240 | * 'reference' page table is init_mm.pgd. |
1241 | * | |
1242 | * NOTE! We MUST NOT take any locks for this case. We may | |
1243 | * be in an interrupt or a critical region, and should | |
1244 | * only copy the information from the master page table, | |
1245 | * nothing more. | |
1246 | * | |
8fed6200 DH |
1247 | * Before doing this on-demand faulting, ensure that the |
1248 | * fault is not any of the following: | |
1249 | * 1. A fault on a PTE with a reserved bit set. | |
1250 | * 2. A fault caused by a user-mode access. (Do not demand- | |
1251 | * fault kernel memory due to user-mode accesses). | |
1252 | * 3. A fault caused by a page-level protection violation. | |
1253 | * (A demand fault would be on a non-present page which | |
1254 | * would have X86_PF_PROT==0). | |
1da177e4 | 1255 | */ |
8fed6200 DH |
1256 | if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { |
1257 | if (vmalloc_fault(address) >= 0) | |
5b727a3b | 1258 | return; |
8fed6200 | 1259 | } |
5b727a3b | 1260 | |
8fed6200 DH |
1261 | /* Was the fault spurious, caused by lazy TLB invalidation? */ |
1262 | if (spurious_kernel_fault(hw_error_code, address)) | |
1263 | return; | |
2d4a7167 | 1264 | |
8fed6200 | 1265 | /* kprobes don't want to hook the spurious faults: */ |
b98cca44 | 1266 | if (kprobe_page_fault(regs, X86_TRAP_PF)) |
92181f19 | 1267 | return; |
8fed6200 DH |
1268 | |
1269 | /* | |
1270 | * Note, despite being a "bad area", there are quite a few | |
1271 | * acceptable reasons to get here, such as erratum fixups | |
1272 | * and handling kernel code that can fault, like get_user(). | |
1273 | * | |
1274 | * Don't take the mm semaphore here. If we fixup a prefetch | |
1275 | * fault we could otherwise deadlock: | |
1276 | */ | |
ba9f6f89 | 1277 | bad_area_nosemaphore(regs, hw_error_code, address); |
8fed6200 DH |
1278 | } |
1279 | NOKPROBE_SYMBOL(do_kern_addr_fault); | |
1280 | ||
aa37c51b DH |
1281 | /* Handle faults in the user portion of the address space */ |
1282 | static inline | |
1283 | void do_user_addr_fault(struct pt_regs *regs, | |
1284 | unsigned long hw_error_code, | |
1285 | unsigned long address) | |
1da177e4 | 1286 | { |
2d4a7167 | 1287 | struct vm_area_struct *vma; |
1da177e4 LT |
1288 | struct task_struct *tsk; |
1289 | struct mm_struct *mm; | |
50a7ca3c | 1290 | vm_fault_t fault, major = 0; |
759496ba | 1291 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1292 | |
a9ba9a3b AV |
1293 | tsk = current; |
1294 | mm = tsk->mm; | |
f8c2ee22 | 1295 | |
2d4a7167 | 1296 | /* kprobes don't want to hook the spurious faults: */ |
b98cca44 | 1297 | if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF))) |
9be260a6 | 1298 | return; |
8c914cb7 | 1299 | |
5b0c2cac DH |
1300 | /* |
1301 | * Reserved bits are never expected to be set on | |
1302 | * entries in the user portion of the page tables. | |
1303 | */ | |
164477c2 DH |
1304 | if (unlikely(hw_error_code & X86_PF_RSVD)) |
1305 | pgtable_bad(regs, hw_error_code, address); | |
1da177e4 | 1306 | |
5b0c2cac | 1307 | /* |
e50928d7 AL |
1308 | * If SMAP is on, check for invalid kernel (supervisor) access to user |
1309 | * pages in the user address space. The odd case here is WRUSS, | |
1310 | * which, according to the preliminary documentation, does not respect | |
1311 | * SMAP and will have the USER bit set so, in all cases, SMAP | |
1312 | * enforcement appears to be consistent with the USER bit. | |
5b0c2cac | 1313 | */ |
a15781b5 AL |
1314 | if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) && |
1315 | !(hw_error_code & X86_PF_USER) && | |
e50928d7 | 1316 | !(regs->flags & X86_EFLAGS_AC))) |
a15781b5 | 1317 | { |
ba9f6f89 | 1318 | bad_area_nosemaphore(regs, hw_error_code, address); |
4640c7ee | 1319 | return; |
40d3cd66 PA |
1320 | } |
1321 | ||
1da177e4 | 1322 | /* |
2d4a7167 | 1323 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1324 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1325 | */ |
70ffdb93 | 1326 | if (unlikely(faulthandler_disabled() || !mm)) { |
ba9f6f89 | 1327 | bad_area_nosemaphore(regs, hw_error_code, address); |
92181f19 NP |
1328 | return; |
1329 | } | |
1da177e4 | 1330 | |
e00b12e6 PZ |
1331 | /* |
1332 | * It's safe to allow irq's after cr2 has been saved and the | |
1333 | * vmalloc fault has been handled. | |
1334 | * | |
1335 | * User-mode registers count as a user access even for any | |
1336 | * potential system fault or CPU buglet: | |
1337 | */ | |
f39b6f0e | 1338 | if (user_mode(regs)) { |
e00b12e6 | 1339 | local_irq_enable(); |
e00b12e6 PZ |
1340 | flags |= FAULT_FLAG_USER; |
1341 | } else { | |
1342 | if (regs->flags & X86_EFLAGS_IF) | |
1343 | local_irq_enable(); | |
1344 | } | |
1345 | ||
1346 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1347 | ||
0ed32f1a | 1348 | if (hw_error_code & X86_PF_WRITE) |
759496ba | 1349 | flags |= FAULT_FLAG_WRITE; |
0ed32f1a | 1350 | if (hw_error_code & X86_PF_INSTR) |
d61172b4 | 1351 | flags |= FAULT_FLAG_INSTRUCTION; |
759496ba | 1352 | |
3ae0ad92 | 1353 | #ifdef CONFIG_X86_64 |
3a1dfe6e | 1354 | /* |
918ce325 AL |
1355 | * Faults in the vsyscall page might need emulation. The |
1356 | * vsyscall page is at a high address (>PAGE_OFFSET), but is | |
1357 | * considered to be part of the user address space. | |
1da177e4 | 1358 | * |
3ae0ad92 DH |
1359 | * The vsyscall page does not have a "real" VMA, so do this |
1360 | * emulation before we go searching for VMAs. | |
e0a446ce AL |
1361 | * |
1362 | * PKRU never rejects instruction fetches, so we don't need | |
1363 | * to consider the PF_PK bit. | |
3ae0ad92 | 1364 | */ |
918ce325 AL |
1365 | if (is_vsyscall_vaddr(address)) { |
1366 | if (emulate_vsyscall(hw_error_code, regs, address)) | |
3ae0ad92 DH |
1367 | return; |
1368 | } | |
1369 | #endif | |
1370 | ||
3a1dfe6e | 1371 | /* |
88259744 DH |
1372 | * Kernel-mode access to the user address space should only occur |
1373 | * on well-defined single instructions listed in the exception | |
1374 | * tables. But, an erroneous kernel fault occurring outside one of | |
1375 | * those areas which also holds mmap_sem might deadlock attempting | |
1376 | * to validate the fault against the address space. | |
1da177e4 | 1377 | * |
88259744 DH |
1378 | * Only do the expensive exception table search when we might be at |
1379 | * risk of a deadlock. This happens if we | |
1380 | * 1. Failed to acquire mmap_sem, and | |
6344be60 | 1381 | * 2. The access did not originate in userspace. |
1da177e4 | 1382 | */ |
92181f19 | 1383 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
6344be60 | 1384 | if (!user_mode(regs) && !search_exception_tables(regs->ip)) { |
88259744 DH |
1385 | /* |
1386 | * Fault from code in kernel from | |
1387 | * which we do not expect faults. | |
1388 | */ | |
0ed32f1a | 1389 | bad_area_nosemaphore(regs, hw_error_code, address); |
92181f19 NP |
1390 | return; |
1391 | } | |
d065bd81 | 1392 | retry: |
1da177e4 | 1393 | down_read(&mm->mmap_sem); |
01006074 PZ |
1394 | } else { |
1395 | /* | |
2d4a7167 IM |
1396 | * The above down_read_trylock() might have succeeded in |
1397 | * which case we'll have missed the might_sleep() from | |
1398 | * down_read(): | |
01006074 PZ |
1399 | */ |
1400 | might_sleep(); | |
1da177e4 LT |
1401 | } |
1402 | ||
1403 | vma = find_vma(mm, address); | |
92181f19 | 1404 | if (unlikely(!vma)) { |
0ed32f1a | 1405 | bad_area(regs, hw_error_code, address); |
92181f19 NP |
1406 | return; |
1407 | } | |
1408 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1409 | goto good_area; |
92181f19 | 1410 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
0ed32f1a | 1411 | bad_area(regs, hw_error_code, address); |
92181f19 NP |
1412 | return; |
1413 | } | |
92181f19 | 1414 | if (unlikely(expand_stack(vma, address))) { |
0ed32f1a | 1415 | bad_area(regs, hw_error_code, address); |
92181f19 NP |
1416 | return; |
1417 | } | |
1418 | ||
1419 | /* | |
1420 | * Ok, we have a good vm_area for this memory access, so | |
1421 | * we can handle it.. | |
1422 | */ | |
1da177e4 | 1423 | good_area: |
0ed32f1a AL |
1424 | if (unlikely(access_error(hw_error_code, vma))) { |
1425 | bad_area_access_error(regs, hw_error_code, address, vma); | |
92181f19 | 1426 | return; |
1da177e4 LT |
1427 | } |
1428 | ||
1429 | /* | |
1430 | * If for any reason at all we couldn't handle the fault, | |
1431 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1432 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1433 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
cb0631fd VB |
1434 | * |
1435 | * Note that handle_userfault() may also release and reacquire mmap_sem | |
1436 | * (and not return with VM_FAULT_RETRY), when returning to userland to | |
1437 | * repeat the page fault later with a VM_FAULT_NOPAGE retval | |
1438 | * (potentially after handling any pending signal during the return to | |
1439 | * userland). The return to userland is identified whenever | |
1440 | * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. | |
1da177e4 | 1441 | */ |
dcddffd4 | 1442 | fault = handle_mm_fault(vma, address, flags); |
26178ec1 | 1443 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1444 | |
3a13c4d7 | 1445 | /* |
26178ec1 LT |
1446 | * If we need to retry the mmap_sem has already been released, |
1447 | * and if there is a fatal signal pending there is no guarantee | |
1448 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1449 | */ |
26178ec1 LT |
1450 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1451 | /* Retry at most once */ | |
1452 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1453 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1454 | flags |= FAULT_FLAG_TRIED; | |
1455 | if (!fatal_signal_pending(tsk)) | |
1456 | goto retry; | |
1457 | } | |
1458 | ||
1459 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1460 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1461 | return; |
1462 | ||
1463 | /* Not returning to user mode? Handle exceptions or die: */ | |
0ed32f1a | 1464 | no_context(regs, hw_error_code, address, SIGBUS, BUS_ADRERR); |
3a13c4d7 | 1465 | return; |
26178ec1 | 1466 | } |
3a13c4d7 | 1467 | |
26178ec1 | 1468 | up_read(&mm->mmap_sem); |
3a13c4d7 | 1469 | if (unlikely(fault & VM_FAULT_ERROR)) { |
0ed32f1a | 1470 | mm_fault_error(regs, hw_error_code, address, fault); |
3a13c4d7 | 1471 | return; |
37b23e05 KM |
1472 | } |
1473 | ||
d065bd81 | 1474 | /* |
26178ec1 LT |
1475 | * Major/minor page fault accounting. If any of the events |
1476 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1477 | */ |
26178ec1 LT |
1478 | if (major) { |
1479 | tsk->maj_flt++; | |
1480 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1481 | } else { | |
1482 | tsk->min_flt++; | |
1483 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1484 | } |
d729ab35 | 1485 | |
8c938f9f | 1486 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1487 | } |
aa37c51b DH |
1488 | NOKPROBE_SYMBOL(do_user_addr_fault); |
1489 | ||
1490 | /* | |
a0d14b89 PZ |
1491 | * Explicitly marked noinline such that the function tracer sees this as the |
1492 | * page_fault entry point. | |
aa37c51b DH |
1493 | */ |
1494 | static noinline void | |
1495 | __do_page_fault(struct pt_regs *regs, unsigned long hw_error_code, | |
1496 | unsigned long address) | |
1497 | { | |
1498 | prefetchw(¤t->mm->mmap_sem); | |
1499 | ||
1500 | if (unlikely(kmmio_fault(regs, address))) | |
1501 | return; | |
1502 | ||
1503 | /* Was the fault on kernel-controlled part of the address space? */ | |
1504 | if (unlikely(fault_in_kernel_space(address))) | |
1505 | do_kern_addr_fault(regs, hw_error_code, address); | |
1506 | else | |
1507 | do_user_addr_fault(regs, hw_error_code, address); | |
1508 | } | |
9326638c | 1509 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1510 | |
a0d14b89 PZ |
1511 | static __always_inline void |
1512 | trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code, | |
1513 | unsigned long address) | |
d34603b0 | 1514 | { |
a0d14b89 PZ |
1515 | if (!trace_pagefault_enabled()) |
1516 | return; | |
1517 | ||
d34603b0 | 1518 | if (user_mode(regs)) |
d4078e23 | 1519 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1520 | else |
d4078e23 | 1521 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1522 | } |
1523 | ||
a0d14b89 PZ |
1524 | dotraplinkage void |
1525 | do_page_fault(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
25c74b10 | 1526 | { |
d4078e23 | 1527 | enum ctx_state prev_state; |
25c74b10 SA |
1528 | |
1529 | prev_state = exception_enter(); | |
a0d14b89 | 1530 | trace_page_fault_entries(regs, error_code, address); |
0ac09f9f | 1531 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1532 | exception_exit(prev_state); |
1533 | } | |
11a7ffb0 | 1534 | NOKPROBE_SYMBOL(do_page_fault); |