Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * Copyright (C) 1995 Linus Torvalds |
2d4a7167 | 3 | * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. |
f8eeb2e6 | 4 | * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar |
1da177e4 | 5 | */ |
a2bcd473 IM |
6 | #include <linux/magic.h> /* STACK_END_MAGIC */ |
7 | #include <linux/sched.h> /* test_thread_flag(), ... */ | |
8 | #include <linux/kdebug.h> /* oops_begin/end, ... */ | |
9 | #include <linux/module.h> /* search_exception_table */ | |
10 | #include <linux/bootmem.h> /* max_low_pfn */ | |
11 | #include <linux/kprobes.h> /* __kprobes, ... */ | |
12 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ | |
cdd6c482 | 13 | #include <linux/perf_event.h> /* perf_sw_event */ |
f672b49b | 14 | #include <linux/hugetlb.h> /* hstate_index_to_shift */ |
268bb0ce | 15 | #include <linux/prefetch.h> /* prefetchw */ |
56dd9470 | 16 | #include <linux/context_tracking.h> /* exception_enter(), ... */ |
2d4a7167 | 17 | |
a2bcd473 IM |
18 | #include <asm/traps.h> /* dotraplinkage, ... */ |
19 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f8561296 | 20 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
fab1167c | 21 | #include <asm/fixmap.h> /* VSYSCALL_START */ |
1da177e4 | 22 | |
d34603b0 SA |
23 | #define CREATE_TRACE_POINTS |
24 | #include <asm/trace/exceptions.h> | |
25 | ||
33cb5243 | 26 | /* |
2d4a7167 IM |
27 | * Page fault error code bits: |
28 | * | |
29 | * bit 0 == 0: no page found 1: protection fault | |
30 | * bit 1 == 0: read access 1: write access | |
31 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
32 | * bit 3 == 1: use of reserved bit detected | |
33 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 34 | */ |
2d4a7167 IM |
35 | enum x86_pf_error_code { |
36 | ||
37 | PF_PROT = 1 << 0, | |
38 | PF_WRITE = 1 << 1, | |
39 | PF_USER = 1 << 2, | |
40 | PF_RSVD = 1 << 3, | |
41 | PF_INSTR = 1 << 4, | |
42 | }; | |
66c58156 | 43 | |
b814d41f | 44 | /* |
b319eed0 IM |
45 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
46 | * handled by mmiotrace: | |
b814d41f | 47 | */ |
62c9295f MH |
48 | static inline int __kprobes |
49 | kmmio_fault(struct pt_regs *regs, unsigned long addr) | |
86069782 | 50 | { |
0fd0e3da PP |
51 | if (unlikely(is_kmmio_active())) |
52 | if (kmmio_handler(regs, addr) == 1) | |
53 | return -1; | |
0fd0e3da | 54 | return 0; |
86069782 PP |
55 | } |
56 | ||
e00b12e6 | 57 | static inline int __kprobes kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 58 | { |
74a0b576 CH |
59 | int ret = 0; |
60 | ||
61 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 62 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
63 | preempt_disable(); |
64 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
65 | ret = 1; | |
66 | preempt_enable(); | |
67 | } | |
1bd858a5 | 68 | |
74a0b576 | 69 | return ret; |
33cb5243 | 70 | } |
1bd858a5 | 71 | |
1dc85be0 | 72 | /* |
2d4a7167 IM |
73 | * Prefetch quirks: |
74 | * | |
75 | * 32-bit mode: | |
76 | * | |
77 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
78 | * Check that here and ignore it. | |
1dc85be0 | 79 | * |
2d4a7167 | 80 | * 64-bit mode: |
1dc85be0 | 81 | * |
2d4a7167 IM |
82 | * Sometimes the CPU reports invalid exceptions on prefetch. |
83 | * Check that here and ignore it. | |
84 | * | |
85 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 86 | */ |
107a0367 IM |
87 | static inline int |
88 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
89 | unsigned char opcode, int *prefetch) | |
90 | { | |
91 | unsigned char instr_hi = opcode & 0xf0; | |
92 | unsigned char instr_lo = opcode & 0x0f; | |
93 | ||
94 | switch (instr_hi) { | |
95 | case 0x20: | |
96 | case 0x30: | |
97 | /* | |
98 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
99 | * In X86_64 long mode, the CPU will signal invalid | |
100 | * opcode if some of these prefixes are present so | |
101 | * X86_64 will never get here anyway | |
102 | */ | |
103 | return ((instr_lo & 7) == 0x6); | |
104 | #ifdef CONFIG_X86_64 | |
105 | case 0x40: | |
106 | /* | |
107 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
108 | * Need to figure out under what instruction mode the | |
109 | * instruction was issued. Could check the LDT for lm, | |
110 | * but for now it's good enough to assume that long | |
111 | * mode only uses well known segments or kernel. | |
112 | */ | |
318f5a2a | 113 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
114 | #endif |
115 | case 0x60: | |
116 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
117 | return (instr_lo & 0xC) == 0x4; | |
118 | case 0xF0: | |
119 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
120 | return !instr_lo || (instr_lo>>1) == 1; | |
121 | case 0x00: | |
122 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
123 | if (probe_kernel_address(instr, opcode)) | |
124 | return 0; | |
125 | ||
126 | *prefetch = (instr_lo == 0xF) && | |
127 | (opcode == 0x0D || opcode == 0x18); | |
128 | return 0; | |
129 | default: | |
130 | return 0; | |
131 | } | |
132 | } | |
133 | ||
2d4a7167 IM |
134 | static int |
135 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 136 | { |
2d4a7167 | 137 | unsigned char *max_instr; |
ab2bf0c1 | 138 | unsigned char *instr; |
33cb5243 | 139 | int prefetch = 0; |
1da177e4 | 140 | |
3085354d IM |
141 | /* |
142 | * If it was a exec (instruction fetch) fault on NX page, then | |
143 | * do not ignore the fault: | |
144 | */ | |
66c58156 | 145 | if (error_code & PF_INSTR) |
1da177e4 | 146 | return 0; |
1dc85be0 | 147 | |
107a0367 | 148 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 149 | max_instr = instr + 15; |
1da177e4 | 150 | |
76381fee | 151 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
152 | return 0; |
153 | ||
107a0367 | 154 | while (instr < max_instr) { |
2d4a7167 | 155 | unsigned char opcode; |
1da177e4 | 156 | |
ab2bf0c1 | 157 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 158 | break; |
1da177e4 | 159 | |
1da177e4 LT |
160 | instr++; |
161 | ||
107a0367 | 162 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 163 | break; |
1da177e4 LT |
164 | } |
165 | return prefetch; | |
166 | } | |
167 | ||
2d4a7167 IM |
168 | static void |
169 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 170 | struct task_struct *tsk, int fault) |
c4aba4a8 | 171 | { |
f672b49b | 172 | unsigned lsb = 0; |
c4aba4a8 HH |
173 | siginfo_t info; |
174 | ||
2d4a7167 IM |
175 | info.si_signo = si_signo; |
176 | info.si_errno = 0; | |
177 | info.si_code = si_code; | |
178 | info.si_addr = (void __user *)address; | |
f672b49b AK |
179 | if (fault & VM_FAULT_HWPOISON_LARGE) |
180 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
181 | if (fault & VM_FAULT_HWPOISON) | |
182 | lsb = PAGE_SHIFT; | |
183 | info.si_addr_lsb = lsb; | |
2d4a7167 | 184 | |
c4aba4a8 HH |
185 | force_sig_info(si_signo, &info, tsk); |
186 | } | |
187 | ||
f2f13a85 IM |
188 | DEFINE_SPINLOCK(pgd_lock); |
189 | LIST_HEAD(pgd_list); | |
190 | ||
191 | #ifdef CONFIG_X86_32 | |
192 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 193 | { |
f2f13a85 IM |
194 | unsigned index = pgd_index(address); |
195 | pgd_t *pgd_k; | |
196 | pud_t *pud, *pud_k; | |
197 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 198 | |
f2f13a85 IM |
199 | pgd += index; |
200 | pgd_k = init_mm.pgd + index; | |
201 | ||
202 | if (!pgd_present(*pgd_k)) | |
203 | return NULL; | |
204 | ||
205 | /* | |
206 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
207 | * and redundant with the set_pmd() on non-PAE. As would | |
208 | * set_pud. | |
209 | */ | |
210 | pud = pud_offset(pgd, address); | |
211 | pud_k = pud_offset(pgd_k, address); | |
212 | if (!pud_present(*pud_k)) | |
213 | return NULL; | |
214 | ||
215 | pmd = pmd_offset(pud, address); | |
216 | pmd_k = pmd_offset(pud_k, address); | |
217 | if (!pmd_present(*pmd_k)) | |
218 | return NULL; | |
219 | ||
b8bcfe99 | 220 | if (!pmd_present(*pmd)) |
f2f13a85 | 221 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 222 | else |
f2f13a85 | 223 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
224 | |
225 | return pmd_k; | |
226 | } | |
227 | ||
228 | void vmalloc_sync_all(void) | |
229 | { | |
230 | unsigned long address; | |
231 | ||
232 | if (SHARED_KERNEL_PMD) | |
233 | return; | |
234 | ||
235 | for (address = VMALLOC_START & PMD_MASK; | |
236 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
237 | address += PMD_SIZE) { | |
f2f13a85 IM |
238 | struct page *page; |
239 | ||
a79e53d8 | 240 | spin_lock(&pgd_lock); |
f2f13a85 | 241 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 242 | spinlock_t *pgt_lock; |
f01f7c56 | 243 | pmd_t *ret; |
617d34d9 | 244 | |
a79e53d8 | 245 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
246 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
247 | ||
248 | spin_lock(pgt_lock); | |
249 | ret = vmalloc_sync_one(page_address(page), address); | |
250 | spin_unlock(pgt_lock); | |
251 | ||
252 | if (!ret) | |
f2f13a85 IM |
253 | break; |
254 | } | |
a79e53d8 | 255 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
256 | } |
257 | } | |
258 | ||
259 | /* | |
260 | * 32-bit: | |
261 | * | |
262 | * Handle a fault on the vmalloc or module mapping area | |
263 | */ | |
62c9295f | 264 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
265 | { |
266 | unsigned long pgd_paddr; | |
267 | pmd_t *pmd_k; | |
268 | pte_t *pte_k; | |
269 | ||
270 | /* Make sure we are in vmalloc area: */ | |
271 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
272 | return -1; | |
273 | ||
ebc8827f FW |
274 | WARN_ON_ONCE(in_nmi()); |
275 | ||
f2f13a85 IM |
276 | /* |
277 | * Synchronize this task's top level page-table | |
278 | * with the 'reference' page table. | |
279 | * | |
280 | * Do _not_ use "current" here. We might be inside | |
281 | * an interrupt in the middle of a task switch.. | |
282 | */ | |
283 | pgd_paddr = read_cr3(); | |
284 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
285 | if (!pmd_k) | |
286 | return -1; | |
287 | ||
288 | pte_k = pte_offset_kernel(pmd_k, address); | |
289 | if (!pte_present(*pte_k)) | |
290 | return -1; | |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
295 | /* | |
296 | * Did it hit the DOS screen memory VA from vm86 mode? | |
297 | */ | |
298 | static inline void | |
299 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
300 | struct task_struct *tsk) | |
301 | { | |
302 | unsigned long bit; | |
303 | ||
304 | if (!v8086_mode(regs)) | |
305 | return; | |
306 | ||
307 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
308 | if (bit < 32) | |
309 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 310 | } |
1da177e4 | 311 | |
087975b0 | 312 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 313 | { |
087975b0 AM |
314 | return pfn < max_low_pfn; |
315 | } | |
1156e098 | 316 | |
087975b0 AM |
317 | static void dump_pagetable(unsigned long address) |
318 | { | |
319 | pgd_t *base = __va(read_cr3()); | |
320 | pgd_t *pgd = &base[pgd_index(address)]; | |
321 | pmd_t *pmd; | |
322 | pte_t *pte; | |
2d4a7167 | 323 | |
1156e098 | 324 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
325 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
326 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
327 | goto out; | |
1156e098 | 328 | #endif |
087975b0 AM |
329 | pmd = pmd_offset(pud_offset(pgd, address), address); |
330 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
331 | |
332 | /* | |
333 | * We must not directly access the pte in the highpte | |
334 | * case if the page table is located in highmem. | |
335 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 336 | * it's allocated already: |
1156e098 | 337 | */ |
087975b0 AM |
338 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
339 | goto out; | |
1156e098 | 340 | |
087975b0 AM |
341 | pte = pte_offset_kernel(pmd, address); |
342 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
343 | out: | |
1156e098 | 344 | printk("\n"); |
f2f13a85 IM |
345 | } |
346 | ||
347 | #else /* CONFIG_X86_64: */ | |
348 | ||
349 | void vmalloc_sync_all(void) | |
350 | { | |
6afb5157 | 351 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
352 | } |
353 | ||
354 | /* | |
355 | * 64-bit: | |
356 | * | |
357 | * Handle a fault on the vmalloc area | |
358 | * | |
359 | * This assumes no large pages in there. | |
360 | */ | |
62c9295f | 361 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
362 | { |
363 | pgd_t *pgd, *pgd_ref; | |
364 | pud_t *pud, *pud_ref; | |
365 | pmd_t *pmd, *pmd_ref; | |
366 | pte_t *pte, *pte_ref; | |
367 | ||
368 | /* Make sure we are in vmalloc area: */ | |
369 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
370 | return -1; | |
371 | ||
ebc8827f FW |
372 | WARN_ON_ONCE(in_nmi()); |
373 | ||
f2f13a85 IM |
374 | /* |
375 | * Copy kernel mappings over when needed. This can also | |
376 | * happen within a race in page table update. In the later | |
377 | * case just flush: | |
378 | */ | |
379 | pgd = pgd_offset(current->active_mm, address); | |
380 | pgd_ref = pgd_offset_k(address); | |
381 | if (pgd_none(*pgd_ref)) | |
382 | return -1; | |
383 | ||
1160c277 | 384 | if (pgd_none(*pgd)) { |
f2f13a85 | 385 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
386 | arch_flush_lazy_mmu_mode(); |
387 | } else { | |
f2f13a85 | 388 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 389 | } |
f2f13a85 IM |
390 | |
391 | /* | |
392 | * Below here mismatches are bugs because these lower tables | |
393 | * are shared: | |
394 | */ | |
395 | ||
396 | pud = pud_offset(pgd, address); | |
397 | pud_ref = pud_offset(pgd_ref, address); | |
398 | if (pud_none(*pud_ref)) | |
399 | return -1; | |
400 | ||
401 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
402 | BUG(); | |
403 | ||
404 | pmd = pmd_offset(pud, address); | |
405 | pmd_ref = pmd_offset(pud_ref, address); | |
406 | if (pmd_none(*pmd_ref)) | |
407 | return -1; | |
408 | ||
409 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
410 | BUG(); | |
411 | ||
412 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
413 | if (!pte_present(*pte_ref)) | |
414 | return -1; | |
415 | ||
416 | pte = pte_offset_kernel(pmd, address); | |
417 | ||
418 | /* | |
419 | * Don't use pte_page here, because the mappings can point | |
420 | * outside mem_map, and the NUMA hash lookup cannot handle | |
421 | * that: | |
422 | */ | |
423 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
424 | BUG(); | |
425 | ||
426 | return 0; | |
427 | } | |
428 | ||
e05139f2 | 429 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 430 | static const char errata93_warning[] = |
ad361c98 JP |
431 | KERN_ERR |
432 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
433 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
434 | "******* Please consider a BIOS update.\n" | |
435 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 436 | #endif |
f2f13a85 IM |
437 | |
438 | /* | |
439 | * No vm86 mode in 64-bit mode: | |
440 | */ | |
441 | static inline void | |
442 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
443 | struct task_struct *tsk) | |
444 | { | |
445 | } | |
446 | ||
447 | static int bad_address(void *p) | |
448 | { | |
449 | unsigned long dummy; | |
450 | ||
451 | return probe_kernel_address((unsigned long *)p, dummy); | |
452 | } | |
453 | ||
454 | static void dump_pagetable(unsigned long address) | |
455 | { | |
087975b0 AM |
456 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
457 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
458 | pud_t *pud; |
459 | pmd_t *pmd; | |
460 | pte_t *pte; | |
461 | ||
2d4a7167 IM |
462 | if (bad_address(pgd)) |
463 | goto bad; | |
464 | ||
d646bce4 | 465 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
466 | |
467 | if (!pgd_present(*pgd)) | |
468 | goto out; | |
1da177e4 | 469 | |
d2ae5b5f | 470 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
471 | if (bad_address(pud)) |
472 | goto bad; | |
473 | ||
1da177e4 | 474 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 475 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 476 | goto out; |
1da177e4 LT |
477 | |
478 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
479 | if (bad_address(pmd)) |
480 | goto bad; | |
481 | ||
1da177e4 | 482 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
483 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
484 | goto out; | |
1da177e4 LT |
485 | |
486 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
487 | if (bad_address(pte)) |
488 | goto bad; | |
489 | ||
33cb5243 | 490 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 491 | out: |
1da177e4 LT |
492 | printk("\n"); |
493 | return; | |
494 | bad: | |
495 | printk("BAD\n"); | |
8c938f9f IM |
496 | } |
497 | ||
f2f13a85 | 498 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 499 | |
2d4a7167 IM |
500 | /* |
501 | * Workaround for K8 erratum #93 & buggy BIOS. | |
502 | * | |
503 | * BIOS SMM functions are required to use a specific workaround | |
504 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
505 | * | |
506 | * A lot of BIOS that didn't get tested properly miss this. | |
507 | * | |
508 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
509 | * Try to work around it here. | |
510 | * | |
511 | * Note we only handle faults in kernel here. | |
512 | * Does nothing on 32-bit. | |
fdfe8aa8 | 513 | */ |
33cb5243 | 514 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 515 | { |
e05139f2 JB |
516 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
517 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
518 | || boot_cpu_data.x86 != 0xf) | |
519 | return 0; | |
520 | ||
65ea5b03 | 521 | if (address != regs->ip) |
1da177e4 | 522 | return 0; |
2d4a7167 | 523 | |
33cb5243 | 524 | if ((address >> 32) != 0) |
1da177e4 | 525 | return 0; |
2d4a7167 | 526 | |
1da177e4 | 527 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
528 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
529 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 530 | printk_once(errata93_warning); |
65ea5b03 | 531 | regs->ip = address; |
1da177e4 LT |
532 | return 1; |
533 | } | |
fdfe8aa8 | 534 | #endif |
1da177e4 | 535 | return 0; |
33cb5243 | 536 | } |
1da177e4 | 537 | |
35f3266f | 538 | /* |
2d4a7167 IM |
539 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
540 | * to illegal addresses >4GB. | |
541 | * | |
542 | * We catch this in the page fault handler because these addresses | |
543 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
544 | * segment in LDT is compatibility mode. |
545 | */ | |
546 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
547 | { | |
548 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 549 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
550 | return 1; |
551 | #endif | |
552 | return 0; | |
553 | } | |
554 | ||
29caf2f9 HH |
555 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
556 | { | |
557 | #ifdef CONFIG_X86_F00F_BUG | |
558 | unsigned long nr; | |
2d4a7167 | 559 | |
29caf2f9 | 560 | /* |
2d4a7167 | 561 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 562 | */ |
e2604b49 | 563 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
564 | nr = (address - idt_descr.address) >> 3; |
565 | ||
566 | if (nr == 6) { | |
567 | do_invalid_op(regs, 0); | |
568 | return 1; | |
569 | } | |
570 | } | |
571 | #endif | |
572 | return 0; | |
573 | } | |
574 | ||
8f766149 IM |
575 | static const char nx_warning[] = KERN_CRIT |
576 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
577 | ||
2d4a7167 IM |
578 | static void |
579 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
580 | unsigned long address) | |
b3279c7f | 581 | { |
1156e098 HH |
582 | if (!oops_may_print()) |
583 | return; | |
584 | ||
1156e098 | 585 | if (error_code & PF_INSTR) { |
93809be8 | 586 | unsigned int level; |
2d4a7167 | 587 | |
1156e098 HH |
588 | pte_t *pte = lookup_address(address, &level); |
589 | ||
8f766149 | 590 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 591 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 592 | } |
1156e098 | 593 | |
19f0dda9 | 594 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 595 | if (address < PAGE_SIZE) |
19f0dda9 | 596 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 597 | else |
19f0dda9 | 598 | printk(KERN_CONT "paging request"); |
2d4a7167 | 599 | |
f294a8ce | 600 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 601 | printk(KERN_ALERT "IP:"); |
5f01c988 | 602 | printk_address(regs->ip); |
2d4a7167 | 603 | |
b3279c7f HH |
604 | dump_pagetable(address); |
605 | } | |
606 | ||
2d4a7167 IM |
607 | static noinline void |
608 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
609 | unsigned long address) | |
1da177e4 | 610 | { |
2d4a7167 IM |
611 | struct task_struct *tsk; |
612 | unsigned long flags; | |
613 | int sig; | |
614 | ||
615 | flags = oops_begin(); | |
616 | tsk = current; | |
617 | sig = SIGKILL; | |
1209140c | 618 | |
1da177e4 | 619 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 620 | tsk->comm, address); |
1da177e4 | 621 | dump_pagetable(address); |
2d4a7167 IM |
622 | |
623 | tsk->thread.cr2 = address; | |
51e7dc70 | 624 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
625 | tsk->thread.error_code = error_code; |
626 | ||
22f5991c | 627 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 628 | sig = 0; |
2d4a7167 | 629 | |
874d93d1 | 630 | oops_end(flags, regs, sig); |
1da177e4 LT |
631 | } |
632 | ||
2d4a7167 IM |
633 | static noinline void |
634 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 635 | unsigned long address, int signal, int si_code) |
92181f19 NP |
636 | { |
637 | struct task_struct *tsk = current; | |
19803078 | 638 | unsigned long *stackend; |
92181f19 NP |
639 | unsigned long flags; |
640 | int sig; | |
92181f19 | 641 | |
2d4a7167 | 642 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 643 | if (fixup_exception(regs)) { |
c026b359 PZ |
644 | /* |
645 | * Any interrupt that takes a fault gets the fixup. This makes | |
646 | * the below recursive fault logic only apply to a faults from | |
647 | * task context. | |
648 | */ | |
649 | if (in_interrupt()) | |
650 | return; | |
651 | ||
652 | /* | |
653 | * Per the above we're !in_interrupt(), aka. task context. | |
654 | * | |
655 | * In this case we need to make sure we're not recursively | |
656 | * faulting through the emulate_vsyscall() logic. | |
657 | */ | |
4fc34901 | 658 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 659 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
660 | tsk->thread.error_code = error_code | PF_USER; |
661 | tsk->thread.cr2 = address; | |
662 | ||
663 | /* XXX: hwpoison faults will set the wrong code. */ | |
664 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
665 | } | |
c026b359 PZ |
666 | |
667 | /* | |
668 | * Barring that, we can do the fixup and be happy. | |
669 | */ | |
92181f19 | 670 | return; |
4fc34901 | 671 | } |
92181f19 NP |
672 | |
673 | /* | |
2d4a7167 IM |
674 | * 32-bit: |
675 | * | |
676 | * Valid to do another page fault here, because if this fault | |
677 | * had been triggered by is_prefetch fixup_exception would have | |
678 | * handled it. | |
679 | * | |
680 | * 64-bit: | |
92181f19 | 681 | * |
2d4a7167 | 682 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
683 | */ |
684 | if (is_prefetch(regs, error_code, address)) | |
685 | return; | |
686 | ||
687 | if (is_errata93(regs, address)) | |
688 | return; | |
689 | ||
690 | /* | |
691 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 692 | * terminate things with extreme prejudice: |
92181f19 | 693 | */ |
92181f19 | 694 | flags = oops_begin(); |
92181f19 NP |
695 | |
696 | show_fault_oops(regs, error_code, address); | |
697 | ||
2d4a7167 | 698 | stackend = end_of_stack(tsk); |
0e7810be | 699 | if (tsk != &init_task && *stackend != STACK_END_MAGIC) |
b0f4c4b3 | 700 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 701 | |
1cc99544 | 702 | tsk->thread.cr2 = address; |
51e7dc70 | 703 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 704 | tsk->thread.error_code = error_code; |
92181f19 | 705 | |
92181f19 NP |
706 | sig = SIGKILL; |
707 | if (__die("Oops", regs, error_code)) | |
708 | sig = 0; | |
2d4a7167 | 709 | |
92181f19 | 710 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 711 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 712 | |
92181f19 | 713 | oops_end(flags, regs, sig); |
92181f19 NP |
714 | } |
715 | ||
2d4a7167 IM |
716 | /* |
717 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
718 | * sysctl is set: | |
719 | */ | |
720 | static inline void | |
721 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
722 | unsigned long address, struct task_struct *tsk) | |
723 | { | |
724 | if (!unhandled_signal(tsk, SIGSEGV)) | |
725 | return; | |
726 | ||
727 | if (!printk_ratelimit()) | |
728 | return; | |
729 | ||
a1a08d1c | 730 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
731 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
732 | tsk->comm, task_pid_nr(tsk), address, | |
733 | (void *)regs->ip, (void *)regs->sp, error_code); | |
734 | ||
735 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
736 | ||
737 | printk(KERN_CONT "\n"); | |
738 | } | |
739 | ||
740 | static void | |
741 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
742 | unsigned long address, int si_code) | |
92181f19 NP |
743 | { |
744 | struct task_struct *tsk = current; | |
745 | ||
746 | /* User mode accesses just cause a SIGSEGV */ | |
747 | if (error_code & PF_USER) { | |
748 | /* | |
2d4a7167 | 749 | * It's possible to have interrupts off here: |
92181f19 NP |
750 | */ |
751 | local_irq_enable(); | |
752 | ||
753 | /* | |
754 | * Valid to do another page fault here because this one came | |
2d4a7167 | 755 | * from user space: |
92181f19 NP |
756 | */ |
757 | if (is_prefetch(regs, error_code, address)) | |
758 | return; | |
759 | ||
760 | if (is_errata100(regs, address)) | |
761 | return; | |
762 | ||
3ae36655 AL |
763 | #ifdef CONFIG_X86_64 |
764 | /* | |
765 | * Instruction fetch faults in the vsyscall page might need | |
766 | * emulation. | |
767 | */ | |
768 | if (unlikely((error_code & PF_INSTR) && | |
769 | ((address & ~0xfff) == VSYSCALL_START))) { | |
770 | if (emulate_vsyscall(regs, address)) | |
771 | return; | |
772 | } | |
773 | #endif | |
e575a86f KC |
774 | /* Kernel addresses are always protection faults: */ |
775 | if (address >= TASK_SIZE) | |
776 | error_code |= PF_PROT; | |
3ae36655 | 777 | |
e575a86f | 778 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
779 | show_signal_msg(regs, error_code, address, tsk); |
780 | ||
2d4a7167 | 781 | tsk->thread.cr2 = address; |
e575a86f | 782 | tsk->thread.error_code = error_code; |
51e7dc70 | 783 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 784 | |
f672b49b | 785 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 786 | |
92181f19 NP |
787 | return; |
788 | } | |
789 | ||
790 | if (is_f00f_bug(regs, address)) | |
791 | return; | |
792 | ||
4fc34901 | 793 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
794 | } |
795 | ||
2d4a7167 IM |
796 | static noinline void |
797 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
798 | unsigned long address) | |
92181f19 NP |
799 | { |
800 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
801 | } | |
802 | ||
2d4a7167 IM |
803 | static void |
804 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
805 | unsigned long address, int si_code) | |
92181f19 NP |
806 | { |
807 | struct mm_struct *mm = current->mm; | |
808 | ||
809 | /* | |
810 | * Something tried to access memory that isn't in our memory map.. | |
811 | * Fix it, but check if it's kernel or user first.. | |
812 | */ | |
813 | up_read(&mm->mmap_sem); | |
814 | ||
815 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
816 | } | |
817 | ||
2d4a7167 IM |
818 | static noinline void |
819 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
820 | { |
821 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
822 | } | |
823 | ||
2d4a7167 IM |
824 | static noinline void |
825 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
826 | unsigned long address) | |
92181f19 NP |
827 | { |
828 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
829 | } | |
830 | ||
2d4a7167 | 831 | static void |
a6e04aa9 AK |
832 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
833 | unsigned int fault) | |
92181f19 NP |
834 | { |
835 | struct task_struct *tsk = current; | |
836 | struct mm_struct *mm = tsk->mm; | |
a6e04aa9 | 837 | int code = BUS_ADRERR; |
92181f19 NP |
838 | |
839 | up_read(&mm->mmap_sem); | |
840 | ||
2d4a7167 | 841 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 842 | if (!(error_code & PF_USER)) { |
4fc34901 | 843 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
844 | return; |
845 | } | |
2d4a7167 | 846 | |
cd1b68f0 | 847 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
848 | if (is_prefetch(regs, error_code, address)) |
849 | return; | |
2d4a7167 IM |
850 | |
851 | tsk->thread.cr2 = address; | |
852 | tsk->thread.error_code = error_code; | |
51e7dc70 | 853 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 854 | |
a6e04aa9 | 855 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 856 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
857 | printk(KERN_ERR |
858 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
859 | tsk->comm, tsk->pid, address); | |
860 | code = BUS_MCEERR_AR; | |
861 | } | |
862 | #endif | |
f672b49b | 863 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
864 | } |
865 | ||
3a13c4d7 | 866 | static noinline void |
2d4a7167 IM |
867 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
868 | unsigned long address, unsigned int fault) | |
92181f19 | 869 | { |
3a13c4d7 JW |
870 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
871 | up_read(¤t->mm->mmap_sem); | |
872 | no_context(regs, error_code, address, 0, 0); | |
873 | return; | |
b80ef10e | 874 | } |
b80ef10e | 875 | |
2d4a7167 | 876 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
877 | /* Kernel mode? Handle exceptions or die: */ |
878 | if (!(error_code & PF_USER)) { | |
879 | up_read(¤t->mm->mmap_sem); | |
4fc34901 AL |
880 | no_context(regs, error_code, address, |
881 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 882 | return; |
f8626854 AV |
883 | } |
884 | ||
c2d23f91 DR |
885 | up_read(¤t->mm->mmap_sem); |
886 | ||
887 | /* | |
888 | * We ran out of memory, call the OOM killer, and return the | |
889 | * userspace (which will retry the fault, or kill us if we got | |
890 | * oom-killed): | |
891 | */ | |
892 | pagefault_out_of_memory(); | |
2d4a7167 | 893 | } else { |
f672b49b AK |
894 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
895 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 896 | do_sigbus(regs, error_code, address, fault); |
2d4a7167 IM |
897 | else |
898 | BUG(); | |
899 | } | |
92181f19 NP |
900 | } |
901 | ||
d8b57bb7 TG |
902 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
903 | { | |
904 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
905 | return 0; | |
2d4a7167 | 906 | |
d8b57bb7 TG |
907 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
908 | return 0; | |
909 | ||
910 | return 1; | |
911 | } | |
912 | ||
5b727a3b | 913 | /* |
2d4a7167 IM |
914 | * Handle a spurious fault caused by a stale TLB entry. |
915 | * | |
916 | * This allows us to lazily refresh the TLB when increasing the | |
917 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
918 | * eagerly is very expensive since that implies doing a full | |
919 | * cross-processor TLB flush, even if no stale TLB entries exist | |
920 | * on other processors. | |
921 | * | |
5b727a3b JF |
922 | * There are no security implications to leaving a stale TLB when |
923 | * increasing the permissions on a page. | |
924 | */ | |
62c9295f | 925 | static noinline __kprobes int |
2d4a7167 | 926 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
927 | { |
928 | pgd_t *pgd; | |
929 | pud_t *pud; | |
930 | pmd_t *pmd; | |
931 | pte_t *pte; | |
3c3e5694 | 932 | int ret; |
5b727a3b JF |
933 | |
934 | /* Reserved-bit violation or user access to kernel space? */ | |
935 | if (error_code & (PF_USER | PF_RSVD)) | |
936 | return 0; | |
937 | ||
938 | pgd = init_mm.pgd + pgd_index(address); | |
939 | if (!pgd_present(*pgd)) | |
940 | return 0; | |
941 | ||
942 | pud = pud_offset(pgd, address); | |
943 | if (!pud_present(*pud)) | |
944 | return 0; | |
945 | ||
d8b57bb7 TG |
946 | if (pud_large(*pud)) |
947 | return spurious_fault_check(error_code, (pte_t *) pud); | |
948 | ||
5b727a3b JF |
949 | pmd = pmd_offset(pud, address); |
950 | if (!pmd_present(*pmd)) | |
951 | return 0; | |
952 | ||
d8b57bb7 TG |
953 | if (pmd_large(*pmd)) |
954 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
955 | ||
5b727a3b | 956 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 957 | if (!pte_present(*pte)) |
5b727a3b JF |
958 | return 0; |
959 | ||
3c3e5694 SR |
960 | ret = spurious_fault_check(error_code, pte); |
961 | if (!ret) | |
962 | return 0; | |
963 | ||
964 | /* | |
2d4a7167 IM |
965 | * Make sure we have permissions in PMD. |
966 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
967 | */ |
968 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
969 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 970 | |
3c3e5694 | 971 | return ret; |
5b727a3b JF |
972 | } |
973 | ||
abd4f750 | 974 | int show_unhandled_signals = 1; |
1da177e4 | 975 | |
2d4a7167 | 976 | static inline int |
68da336a | 977 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 978 | { |
68da336a | 979 | if (error_code & PF_WRITE) { |
2d4a7167 | 980 | /* write, present and write, not present: */ |
92181f19 NP |
981 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
982 | return 1; | |
2d4a7167 | 983 | return 0; |
92181f19 NP |
984 | } |
985 | ||
2d4a7167 IM |
986 | /* read, present: */ |
987 | if (unlikely(error_code & PF_PROT)) | |
988 | return 1; | |
989 | ||
990 | /* read, not present: */ | |
991 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
992 | return 1; | |
993 | ||
92181f19 NP |
994 | return 0; |
995 | } | |
996 | ||
0973a06c HS |
997 | static int fault_in_kernel_space(unsigned long address) |
998 | { | |
d9517346 | 999 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1000 | } |
1001 | ||
40d3cd66 PA |
1002 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1003 | { | |
4640c7ee PA |
1004 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1005 | return false; | |
1006 | ||
1007 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1008 | return false; | |
1009 | ||
40d3cd66 PA |
1010 | if (error_code & PF_USER) |
1011 | return false; | |
1012 | ||
1013 | if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) | |
1014 | return false; | |
1015 | ||
1016 | return true; | |
1017 | } | |
1018 | ||
1da177e4 LT |
1019 | /* |
1020 | * This routine handles page faults. It determines the address, | |
1021 | * and the problem, and then passes it off to one of the appropriate | |
1022 | * routines. | |
1da177e4 | 1023 | */ |
6ba3c97a | 1024 | static void __kprobes |
0ac09f9f JO |
1025 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1026 | unsigned long address) | |
1da177e4 | 1027 | { |
2d4a7167 | 1028 | struct vm_area_struct *vma; |
1da177e4 LT |
1029 | struct task_struct *tsk; |
1030 | struct mm_struct *mm; | |
f8c2ee22 | 1031 | int fault; |
759496ba | 1032 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1033 | |
a9ba9a3b AV |
1034 | tsk = current; |
1035 | mm = tsk->mm; | |
2d4a7167 | 1036 | |
f8561296 VN |
1037 | /* |
1038 | * Detect and handle instructions that would cause a page fault for | |
1039 | * both a tracked kernel page and a userspace page. | |
1040 | */ | |
1041 | if (kmemcheck_active(regs)) | |
1042 | kmemcheck_hide(regs); | |
5dfaf90f | 1043 | prefetchw(&mm->mmap_sem); |
f8561296 | 1044 | |
0fd0e3da | 1045 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1046 | return; |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * We fault-in kernel-space virtual memory on-demand. The | |
1050 | * 'reference' page table is init_mm.pgd. | |
1051 | * | |
1052 | * NOTE! We MUST NOT take any locks for this case. We may | |
1053 | * be in an interrupt or a critical region, and should | |
1054 | * only copy the information from the master page table, | |
1055 | * nothing more. | |
1056 | * | |
1057 | * This verifies that the fault happens in kernel space | |
1058 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1059 | * protection error (error_code & 9) == 0. |
1da177e4 | 1060 | */ |
0973a06c | 1061 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1062 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1063 | if (vmalloc_fault(address) >= 0) | |
1064 | return; | |
1065 | ||
1066 | if (kmemcheck_fault(regs, address, error_code)) | |
1067 | return; | |
1068 | } | |
5b727a3b | 1069 | |
2d4a7167 | 1070 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1071 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1072 | return; |
1073 | ||
2d4a7167 | 1074 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1075 | if (kprobes_fault(regs)) |
9be260a6 | 1076 | return; |
f8c2ee22 HH |
1077 | /* |
1078 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1079 | * fault we could otherwise deadlock: |
f8c2ee22 | 1080 | */ |
92181f19 | 1081 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1082 | |
92181f19 | 1083 | return; |
f8c2ee22 HH |
1084 | } |
1085 | ||
2d4a7167 | 1086 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1087 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1088 | return; |
8c914cb7 | 1089 | |
66c58156 | 1090 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1091 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1092 | |
4640c7ee PA |
1093 | if (unlikely(smap_violation(error_code, regs))) { |
1094 | bad_area_nosemaphore(regs, error_code, address); | |
1095 | return; | |
40d3cd66 PA |
1096 | } |
1097 | ||
1da177e4 | 1098 | /* |
2d4a7167 IM |
1099 | * If we're in an interrupt, have no user context or are running |
1100 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1101 | */ |
92181f19 NP |
1102 | if (unlikely(in_atomic() || !mm)) { |
1103 | bad_area_nosemaphore(regs, error_code, address); | |
1104 | return; | |
1105 | } | |
1da177e4 | 1106 | |
e00b12e6 PZ |
1107 | /* |
1108 | * It's safe to allow irq's after cr2 has been saved and the | |
1109 | * vmalloc fault has been handled. | |
1110 | * | |
1111 | * User-mode registers count as a user access even for any | |
1112 | * potential system fault or CPU buglet: | |
1113 | */ | |
1114 | if (user_mode_vm(regs)) { | |
1115 | local_irq_enable(); | |
1116 | error_code |= PF_USER; | |
1117 | flags |= FAULT_FLAG_USER; | |
1118 | } else { | |
1119 | if (regs->flags & X86_EFLAGS_IF) | |
1120 | local_irq_enable(); | |
1121 | } | |
1122 | ||
1123 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1124 | ||
759496ba JW |
1125 | if (error_code & PF_WRITE) |
1126 | flags |= FAULT_FLAG_WRITE; | |
1127 | ||
3a1dfe6e IM |
1128 | /* |
1129 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1130 | * addresses in user space. All other faults represent errors in |
1131 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1132 | * case of an erroneous fault occurring in a code path which already | |
1133 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1134 | * against the address space. Luckily the kernel only validly | |
1135 | * references user space from well defined areas of code, which are | |
1136 | * listed in the exceptions table. | |
1da177e4 LT |
1137 | * |
1138 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1139 | * the source reference check when there is a possibility of a |
1140 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1141 | * validate the source. If this is invalid we can skip the address | |
1142 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1143 | */ |
92181f19 | 1144 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1145 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1146 | !search_exception_tables(regs->ip)) { |
1147 | bad_area_nosemaphore(regs, error_code, address); | |
1148 | return; | |
1149 | } | |
d065bd81 | 1150 | retry: |
1da177e4 | 1151 | down_read(&mm->mmap_sem); |
01006074 PZ |
1152 | } else { |
1153 | /* | |
2d4a7167 IM |
1154 | * The above down_read_trylock() might have succeeded in |
1155 | * which case we'll have missed the might_sleep() from | |
1156 | * down_read(): | |
01006074 PZ |
1157 | */ |
1158 | might_sleep(); | |
1da177e4 LT |
1159 | } |
1160 | ||
1161 | vma = find_vma(mm, address); | |
92181f19 NP |
1162 | if (unlikely(!vma)) { |
1163 | bad_area(regs, error_code, address); | |
1164 | return; | |
1165 | } | |
1166 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1167 | goto good_area; |
92181f19 NP |
1168 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1169 | bad_area(regs, error_code, address); | |
1170 | return; | |
1171 | } | |
33cb5243 | 1172 | if (error_code & PF_USER) { |
6f4d368e HH |
1173 | /* |
1174 | * Accessing the stack below %sp is always a bug. | |
1175 | * The large cushion allows instructions like enter | |
2d4a7167 | 1176 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1177 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1178 | */ |
92181f19 NP |
1179 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1180 | bad_area(regs, error_code, address); | |
1181 | return; | |
1182 | } | |
1da177e4 | 1183 | } |
92181f19 NP |
1184 | if (unlikely(expand_stack(vma, address))) { |
1185 | bad_area(regs, error_code, address); | |
1186 | return; | |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * Ok, we have a good vm_area for this memory access, so | |
1191 | * we can handle it.. | |
1192 | */ | |
1da177e4 | 1193 | good_area: |
68da336a | 1194 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1195 | bad_area_access_error(regs, error_code, address); |
1196 | return; | |
1da177e4 LT |
1197 | } |
1198 | ||
1199 | /* | |
1200 | * If for any reason at all we couldn't handle the fault, | |
1201 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1202 | * the fault: |
1da177e4 | 1203 | */ |
d065bd81 | 1204 | fault = handle_mm_fault(mm, vma, address, flags); |
2d4a7167 | 1205 | |
3a13c4d7 JW |
1206 | /* |
1207 | * If we need to retry but a fatal signal is pending, handle the | |
1208 | * signal first. We do not need to release the mmap_sem because it | |
1209 | * would already be released in __lock_page_or_retry in mm/filemap.c. | |
1210 | */ | |
1211 | if (unlikely((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))) | |
1212 | return; | |
1213 | ||
1214 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
1215 | mm_fault_error(regs, error_code, address, fault); | |
1216 | return; | |
37b23e05 KM |
1217 | } |
1218 | ||
d065bd81 ML |
1219 | /* |
1220 | * Major/minor page fault accounting is only done on the | |
1221 | * initial attempt. If we go through a retry, it is extremely | |
1222 | * likely that the page will be found in page cache at that point. | |
1223 | */ | |
1224 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1225 | if (fault & VM_FAULT_MAJOR) { | |
1226 | tsk->maj_flt++; | |
a8b0ca17 | 1227 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
d065bd81 ML |
1228 | regs, address); |
1229 | } else { | |
1230 | tsk->min_flt++; | |
a8b0ca17 | 1231 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
d065bd81 ML |
1232 | regs, address); |
1233 | } | |
1234 | if (fault & VM_FAULT_RETRY) { | |
1235 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
1236 | * of starvation. */ | |
1237 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 1238 | flags |= FAULT_FLAG_TRIED; |
d065bd81 ML |
1239 | goto retry; |
1240 | } | |
ac17dc8e | 1241 | } |
d729ab35 | 1242 | |
8c938f9f IM |
1243 | check_v8086_mode(regs, address, tsk); |
1244 | ||
1da177e4 | 1245 | up_read(&mm->mmap_sem); |
1da177e4 | 1246 | } |
6ba3c97a FW |
1247 | |
1248 | dotraplinkage void __kprobes | |
1249 | do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1250 | { | |
6c1e0256 | 1251 | enum ctx_state prev_state; |
0ac09f9f JO |
1252 | /* Get the faulting address: */ |
1253 | unsigned long address = read_cr2(); | |
6c1e0256 FW |
1254 | |
1255 | prev_state = exception_enter(); | |
0ac09f9f | 1256 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1257 | exception_exit(prev_state); |
6ba3c97a | 1258 | } |
25c74b10 | 1259 | |
d34603b0 SA |
1260 | static void trace_page_fault_entries(struct pt_regs *regs, |
1261 | unsigned long error_code) | |
1262 | { | |
1263 | if (user_mode(regs)) | |
a4f61dec | 1264 | trace_page_fault_user(read_cr2(), regs, error_code); |
d34603b0 | 1265 | else |
a4f61dec | 1266 | trace_page_fault_kernel(read_cr2(), regs, error_code); |
d34603b0 SA |
1267 | } |
1268 | ||
25c74b10 SA |
1269 | dotraplinkage void __kprobes |
1270 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1271 | { | |
1272 | enum ctx_state prev_state; | |
0ac09f9f JO |
1273 | /* |
1274 | * The exception_enter and tracepoint processing could | |
1275 | * trigger another page faults (user space callchain | |
1276 | * reading) and destroy the original cr2 value, so read | |
1277 | * the faulting address now. | |
1278 | */ | |
1279 | unsigned long address = read_cr2(); | |
25c74b10 SA |
1280 | |
1281 | prev_state = exception_enter(); | |
d34603b0 | 1282 | trace_page_fault_entries(regs, error_code); |
0ac09f9f | 1283 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1284 | exception_exit(prev_state); |
1285 | } |