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 */ |
2d4a7167 | 16 | |
a2bcd473 IM |
17 | #include <asm/traps.h> /* dotraplinkage, ... */ |
18 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f8561296 | 19 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
fab1167c | 20 | #include <asm/fixmap.h> /* VSYSCALL_START */ |
91d1aa43 | 21 | #include <asm/context_tracking.h> /* exception_enter(), ... */ |
1da177e4 | 22 | |
33cb5243 | 23 | /* |
2d4a7167 IM |
24 | * Page fault error code bits: |
25 | * | |
26 | * bit 0 == 0: no page found 1: protection fault | |
27 | * bit 1 == 0: read access 1: write access | |
28 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
29 | * bit 3 == 1: use of reserved bit detected | |
30 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 31 | */ |
2d4a7167 IM |
32 | enum x86_pf_error_code { |
33 | ||
34 | PF_PROT = 1 << 0, | |
35 | PF_WRITE = 1 << 1, | |
36 | PF_USER = 1 << 2, | |
37 | PF_RSVD = 1 << 3, | |
38 | PF_INSTR = 1 << 4, | |
39 | }; | |
66c58156 | 40 | |
b814d41f | 41 | /* |
b319eed0 IM |
42 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
43 | * handled by mmiotrace: | |
b814d41f | 44 | */ |
62c9295f MH |
45 | static inline int __kprobes |
46 | kmmio_fault(struct pt_regs *regs, unsigned long addr) | |
86069782 | 47 | { |
0fd0e3da PP |
48 | if (unlikely(is_kmmio_active())) |
49 | if (kmmio_handler(regs, addr) == 1) | |
50 | return -1; | |
0fd0e3da | 51 | return 0; |
86069782 PP |
52 | } |
53 | ||
62c9295f | 54 | static inline int __kprobes notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 55 | { |
74a0b576 CH |
56 | int ret = 0; |
57 | ||
58 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 59 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
60 | preempt_disable(); |
61 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
62 | ret = 1; | |
63 | preempt_enable(); | |
64 | } | |
1bd858a5 | 65 | |
74a0b576 | 66 | return ret; |
33cb5243 | 67 | } |
1bd858a5 | 68 | |
1dc85be0 | 69 | /* |
2d4a7167 IM |
70 | * Prefetch quirks: |
71 | * | |
72 | * 32-bit mode: | |
73 | * | |
74 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
75 | * Check that here and ignore it. | |
1dc85be0 | 76 | * |
2d4a7167 | 77 | * 64-bit mode: |
1dc85be0 | 78 | * |
2d4a7167 IM |
79 | * Sometimes the CPU reports invalid exceptions on prefetch. |
80 | * Check that here and ignore it. | |
81 | * | |
82 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 83 | */ |
107a0367 IM |
84 | static inline int |
85 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
86 | unsigned char opcode, int *prefetch) | |
87 | { | |
88 | unsigned char instr_hi = opcode & 0xf0; | |
89 | unsigned char instr_lo = opcode & 0x0f; | |
90 | ||
91 | switch (instr_hi) { | |
92 | case 0x20: | |
93 | case 0x30: | |
94 | /* | |
95 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
96 | * In X86_64 long mode, the CPU will signal invalid | |
97 | * opcode if some of these prefixes are present so | |
98 | * X86_64 will never get here anyway | |
99 | */ | |
100 | return ((instr_lo & 7) == 0x6); | |
101 | #ifdef CONFIG_X86_64 | |
102 | case 0x40: | |
103 | /* | |
104 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
105 | * Need to figure out under what instruction mode the | |
106 | * instruction was issued. Could check the LDT for lm, | |
107 | * but for now it's good enough to assume that long | |
108 | * mode only uses well known segments or kernel. | |
109 | */ | |
318f5a2a | 110 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
111 | #endif |
112 | case 0x60: | |
113 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
114 | return (instr_lo & 0xC) == 0x4; | |
115 | case 0xF0: | |
116 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
117 | return !instr_lo || (instr_lo>>1) == 1; | |
118 | case 0x00: | |
119 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
120 | if (probe_kernel_address(instr, opcode)) | |
121 | return 0; | |
122 | ||
123 | *prefetch = (instr_lo == 0xF) && | |
124 | (opcode == 0x0D || opcode == 0x18); | |
125 | return 0; | |
126 | default: | |
127 | return 0; | |
128 | } | |
129 | } | |
130 | ||
2d4a7167 IM |
131 | static int |
132 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 133 | { |
2d4a7167 | 134 | unsigned char *max_instr; |
ab2bf0c1 | 135 | unsigned char *instr; |
33cb5243 | 136 | int prefetch = 0; |
1da177e4 | 137 | |
3085354d IM |
138 | /* |
139 | * If it was a exec (instruction fetch) fault on NX page, then | |
140 | * do not ignore the fault: | |
141 | */ | |
66c58156 | 142 | if (error_code & PF_INSTR) |
1da177e4 | 143 | return 0; |
1dc85be0 | 144 | |
107a0367 | 145 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 146 | max_instr = instr + 15; |
1da177e4 | 147 | |
76381fee | 148 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
149 | return 0; |
150 | ||
107a0367 | 151 | while (instr < max_instr) { |
2d4a7167 | 152 | unsigned char opcode; |
1da177e4 | 153 | |
ab2bf0c1 | 154 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 155 | break; |
1da177e4 | 156 | |
1da177e4 LT |
157 | instr++; |
158 | ||
107a0367 | 159 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 160 | break; |
1da177e4 LT |
161 | } |
162 | return prefetch; | |
163 | } | |
164 | ||
2d4a7167 IM |
165 | static void |
166 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 167 | struct task_struct *tsk, int fault) |
c4aba4a8 | 168 | { |
f672b49b | 169 | unsigned lsb = 0; |
c4aba4a8 HH |
170 | siginfo_t info; |
171 | ||
2d4a7167 IM |
172 | info.si_signo = si_signo; |
173 | info.si_errno = 0; | |
174 | info.si_code = si_code; | |
175 | info.si_addr = (void __user *)address; | |
f672b49b AK |
176 | if (fault & VM_FAULT_HWPOISON_LARGE) |
177 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
178 | if (fault & VM_FAULT_HWPOISON) | |
179 | lsb = PAGE_SHIFT; | |
180 | info.si_addr_lsb = lsb; | |
2d4a7167 | 181 | |
c4aba4a8 HH |
182 | force_sig_info(si_signo, &info, tsk); |
183 | } | |
184 | ||
f2f13a85 IM |
185 | DEFINE_SPINLOCK(pgd_lock); |
186 | LIST_HEAD(pgd_list); | |
187 | ||
188 | #ifdef CONFIG_X86_32 | |
189 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 190 | { |
f2f13a85 IM |
191 | unsigned index = pgd_index(address); |
192 | pgd_t *pgd_k; | |
193 | pud_t *pud, *pud_k; | |
194 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 195 | |
f2f13a85 IM |
196 | pgd += index; |
197 | pgd_k = init_mm.pgd + index; | |
198 | ||
199 | if (!pgd_present(*pgd_k)) | |
200 | return NULL; | |
201 | ||
202 | /* | |
203 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
204 | * and redundant with the set_pmd() on non-PAE. As would | |
205 | * set_pud. | |
206 | */ | |
207 | pud = pud_offset(pgd, address); | |
208 | pud_k = pud_offset(pgd_k, address); | |
209 | if (!pud_present(*pud_k)) | |
210 | return NULL; | |
211 | ||
212 | pmd = pmd_offset(pud, address); | |
213 | pmd_k = pmd_offset(pud_k, address); | |
214 | if (!pmd_present(*pmd_k)) | |
215 | return NULL; | |
216 | ||
b8bcfe99 | 217 | if (!pmd_present(*pmd)) |
f2f13a85 | 218 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 219 | else |
f2f13a85 | 220 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
221 | |
222 | return pmd_k; | |
223 | } | |
224 | ||
225 | void vmalloc_sync_all(void) | |
226 | { | |
227 | unsigned long address; | |
228 | ||
229 | if (SHARED_KERNEL_PMD) | |
230 | return; | |
231 | ||
232 | for (address = VMALLOC_START & PMD_MASK; | |
233 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
234 | address += PMD_SIZE) { | |
f2f13a85 IM |
235 | struct page *page; |
236 | ||
a79e53d8 | 237 | spin_lock(&pgd_lock); |
f2f13a85 | 238 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 239 | spinlock_t *pgt_lock; |
f01f7c56 | 240 | pmd_t *ret; |
617d34d9 | 241 | |
a79e53d8 | 242 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
243 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
244 | ||
245 | spin_lock(pgt_lock); | |
246 | ret = vmalloc_sync_one(page_address(page), address); | |
247 | spin_unlock(pgt_lock); | |
248 | ||
249 | if (!ret) | |
f2f13a85 IM |
250 | break; |
251 | } | |
a79e53d8 | 252 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
253 | } |
254 | } | |
255 | ||
256 | /* | |
257 | * 32-bit: | |
258 | * | |
259 | * Handle a fault on the vmalloc or module mapping area | |
260 | */ | |
62c9295f | 261 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
262 | { |
263 | unsigned long pgd_paddr; | |
264 | pmd_t *pmd_k; | |
265 | pte_t *pte_k; | |
266 | ||
267 | /* Make sure we are in vmalloc area: */ | |
268 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
269 | return -1; | |
270 | ||
ebc8827f FW |
271 | WARN_ON_ONCE(in_nmi()); |
272 | ||
f2f13a85 IM |
273 | /* |
274 | * Synchronize this task's top level page-table | |
275 | * with the 'reference' page table. | |
276 | * | |
277 | * Do _not_ use "current" here. We might be inside | |
278 | * an interrupt in the middle of a task switch.. | |
279 | */ | |
280 | pgd_paddr = read_cr3(); | |
281 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
282 | if (!pmd_k) | |
283 | return -1; | |
284 | ||
285 | pte_k = pte_offset_kernel(pmd_k, address); | |
286 | if (!pte_present(*pte_k)) | |
287 | return -1; | |
288 | ||
289 | return 0; | |
290 | } | |
291 | ||
292 | /* | |
293 | * Did it hit the DOS screen memory VA from vm86 mode? | |
294 | */ | |
295 | static inline void | |
296 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
297 | struct task_struct *tsk) | |
298 | { | |
299 | unsigned long bit; | |
300 | ||
301 | if (!v8086_mode(regs)) | |
302 | return; | |
303 | ||
304 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
305 | if (bit < 32) | |
306 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 307 | } |
1da177e4 | 308 | |
087975b0 | 309 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 310 | { |
087975b0 AM |
311 | return pfn < max_low_pfn; |
312 | } | |
1156e098 | 313 | |
087975b0 AM |
314 | static void dump_pagetable(unsigned long address) |
315 | { | |
316 | pgd_t *base = __va(read_cr3()); | |
317 | pgd_t *pgd = &base[pgd_index(address)]; | |
318 | pmd_t *pmd; | |
319 | pte_t *pte; | |
2d4a7167 | 320 | |
1156e098 | 321 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
322 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
323 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
324 | goto out; | |
1156e098 | 325 | #endif |
087975b0 AM |
326 | pmd = pmd_offset(pud_offset(pgd, address), address); |
327 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
328 | |
329 | /* | |
330 | * We must not directly access the pte in the highpte | |
331 | * case if the page table is located in highmem. | |
332 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 333 | * it's allocated already: |
1156e098 | 334 | */ |
087975b0 AM |
335 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
336 | goto out; | |
1156e098 | 337 | |
087975b0 AM |
338 | pte = pte_offset_kernel(pmd, address); |
339 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
340 | out: | |
1156e098 | 341 | printk("\n"); |
f2f13a85 IM |
342 | } |
343 | ||
344 | #else /* CONFIG_X86_64: */ | |
345 | ||
346 | void vmalloc_sync_all(void) | |
347 | { | |
6afb5157 | 348 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
349 | } |
350 | ||
351 | /* | |
352 | * 64-bit: | |
353 | * | |
354 | * Handle a fault on the vmalloc area | |
355 | * | |
356 | * This assumes no large pages in there. | |
357 | */ | |
62c9295f | 358 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
359 | { |
360 | pgd_t *pgd, *pgd_ref; | |
361 | pud_t *pud, *pud_ref; | |
362 | pmd_t *pmd, *pmd_ref; | |
363 | pte_t *pte, *pte_ref; | |
364 | ||
365 | /* Make sure we are in vmalloc area: */ | |
366 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
367 | return -1; | |
368 | ||
ebc8827f FW |
369 | WARN_ON_ONCE(in_nmi()); |
370 | ||
f2f13a85 IM |
371 | /* |
372 | * Copy kernel mappings over when needed. This can also | |
373 | * happen within a race in page table update. In the later | |
374 | * case just flush: | |
375 | */ | |
376 | pgd = pgd_offset(current->active_mm, address); | |
377 | pgd_ref = pgd_offset_k(address); | |
378 | if (pgd_none(*pgd_ref)) | |
379 | return -1; | |
380 | ||
381 | if (pgd_none(*pgd)) | |
382 | set_pgd(pgd, *pgd_ref); | |
383 | else | |
384 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
385 | ||
386 | /* | |
387 | * Below here mismatches are bugs because these lower tables | |
388 | * are shared: | |
389 | */ | |
390 | ||
391 | pud = pud_offset(pgd, address); | |
392 | pud_ref = pud_offset(pgd_ref, address); | |
393 | if (pud_none(*pud_ref)) | |
394 | return -1; | |
395 | ||
396 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
397 | BUG(); | |
398 | ||
399 | pmd = pmd_offset(pud, address); | |
400 | pmd_ref = pmd_offset(pud_ref, address); | |
401 | if (pmd_none(*pmd_ref)) | |
402 | return -1; | |
403 | ||
404 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
405 | BUG(); | |
406 | ||
407 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
408 | if (!pte_present(*pte_ref)) | |
409 | return -1; | |
410 | ||
411 | pte = pte_offset_kernel(pmd, address); | |
412 | ||
413 | /* | |
414 | * Don't use pte_page here, because the mappings can point | |
415 | * outside mem_map, and the NUMA hash lookup cannot handle | |
416 | * that: | |
417 | */ | |
418 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
419 | BUG(); | |
420 | ||
421 | return 0; | |
422 | } | |
423 | ||
e05139f2 | 424 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 425 | static const char errata93_warning[] = |
ad361c98 JP |
426 | KERN_ERR |
427 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
428 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
429 | "******* Please consider a BIOS update.\n" | |
430 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 431 | #endif |
f2f13a85 IM |
432 | |
433 | /* | |
434 | * No vm86 mode in 64-bit mode: | |
435 | */ | |
436 | static inline void | |
437 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
438 | struct task_struct *tsk) | |
439 | { | |
440 | } | |
441 | ||
442 | static int bad_address(void *p) | |
443 | { | |
444 | unsigned long dummy; | |
445 | ||
446 | return probe_kernel_address((unsigned long *)p, dummy); | |
447 | } | |
448 | ||
449 | static void dump_pagetable(unsigned long address) | |
450 | { | |
087975b0 AM |
451 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
452 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
453 | pud_t *pud; |
454 | pmd_t *pmd; | |
455 | pte_t *pte; | |
456 | ||
2d4a7167 IM |
457 | if (bad_address(pgd)) |
458 | goto bad; | |
459 | ||
d646bce4 | 460 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
461 | |
462 | if (!pgd_present(*pgd)) | |
463 | goto out; | |
1da177e4 | 464 | |
d2ae5b5f | 465 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
466 | if (bad_address(pud)) |
467 | goto bad; | |
468 | ||
1da177e4 | 469 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 470 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 471 | goto out; |
1da177e4 LT |
472 | |
473 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
474 | if (bad_address(pmd)) |
475 | goto bad; | |
476 | ||
1da177e4 | 477 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
478 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
479 | goto out; | |
1da177e4 LT |
480 | |
481 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
482 | if (bad_address(pte)) |
483 | goto bad; | |
484 | ||
33cb5243 | 485 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 486 | out: |
1da177e4 LT |
487 | printk("\n"); |
488 | return; | |
489 | bad: | |
490 | printk("BAD\n"); | |
8c938f9f IM |
491 | } |
492 | ||
f2f13a85 | 493 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 494 | |
2d4a7167 IM |
495 | /* |
496 | * Workaround for K8 erratum #93 & buggy BIOS. | |
497 | * | |
498 | * BIOS SMM functions are required to use a specific workaround | |
499 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
500 | * | |
501 | * A lot of BIOS that didn't get tested properly miss this. | |
502 | * | |
503 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
504 | * Try to work around it here. | |
505 | * | |
506 | * Note we only handle faults in kernel here. | |
507 | * Does nothing on 32-bit. | |
fdfe8aa8 | 508 | */ |
33cb5243 | 509 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 510 | { |
e05139f2 JB |
511 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
512 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
513 | || boot_cpu_data.x86 != 0xf) | |
514 | return 0; | |
515 | ||
65ea5b03 | 516 | if (address != regs->ip) |
1da177e4 | 517 | return 0; |
2d4a7167 | 518 | |
33cb5243 | 519 | if ((address >> 32) != 0) |
1da177e4 | 520 | return 0; |
2d4a7167 | 521 | |
1da177e4 | 522 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
523 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
524 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 525 | printk_once(errata93_warning); |
65ea5b03 | 526 | regs->ip = address; |
1da177e4 LT |
527 | return 1; |
528 | } | |
fdfe8aa8 | 529 | #endif |
1da177e4 | 530 | return 0; |
33cb5243 | 531 | } |
1da177e4 | 532 | |
35f3266f | 533 | /* |
2d4a7167 IM |
534 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
535 | * to illegal addresses >4GB. | |
536 | * | |
537 | * We catch this in the page fault handler because these addresses | |
538 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
539 | * segment in LDT is compatibility mode. |
540 | */ | |
541 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
542 | { | |
543 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 544 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
545 | return 1; |
546 | #endif | |
547 | return 0; | |
548 | } | |
549 | ||
29caf2f9 HH |
550 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
551 | { | |
552 | #ifdef CONFIG_X86_F00F_BUG | |
553 | unsigned long nr; | |
2d4a7167 | 554 | |
29caf2f9 | 555 | /* |
2d4a7167 | 556 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 HH |
557 | */ |
558 | if (boot_cpu_data.f00f_bug) { | |
559 | nr = (address - idt_descr.address) >> 3; | |
560 | ||
561 | if (nr == 6) { | |
562 | do_invalid_op(regs, 0); | |
563 | return 1; | |
564 | } | |
565 | } | |
566 | #endif | |
567 | return 0; | |
568 | } | |
569 | ||
8f766149 IM |
570 | static const char nx_warning[] = KERN_CRIT |
571 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
572 | ||
2d4a7167 IM |
573 | static void |
574 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
575 | unsigned long address) | |
b3279c7f | 576 | { |
1156e098 HH |
577 | if (!oops_may_print()) |
578 | return; | |
579 | ||
1156e098 | 580 | if (error_code & PF_INSTR) { |
93809be8 | 581 | unsigned int level; |
2d4a7167 | 582 | |
1156e098 HH |
583 | pte_t *pte = lookup_address(address, &level); |
584 | ||
8f766149 | 585 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 586 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 587 | } |
1156e098 | 588 | |
19f0dda9 | 589 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 590 | if (address < PAGE_SIZE) |
19f0dda9 | 591 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 592 | else |
19f0dda9 | 593 | printk(KERN_CONT "paging request"); |
2d4a7167 | 594 | |
f294a8ce | 595 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 596 | printk(KERN_ALERT "IP:"); |
b3279c7f | 597 | printk_address(regs->ip, 1); |
2d4a7167 | 598 | |
b3279c7f HH |
599 | dump_pagetable(address); |
600 | } | |
601 | ||
2d4a7167 IM |
602 | static noinline void |
603 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
604 | unsigned long address) | |
1da177e4 | 605 | { |
2d4a7167 IM |
606 | struct task_struct *tsk; |
607 | unsigned long flags; | |
608 | int sig; | |
609 | ||
610 | flags = oops_begin(); | |
611 | tsk = current; | |
612 | sig = SIGKILL; | |
1209140c | 613 | |
1da177e4 | 614 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 615 | tsk->comm, address); |
1da177e4 | 616 | dump_pagetable(address); |
2d4a7167 IM |
617 | |
618 | tsk->thread.cr2 = address; | |
51e7dc70 | 619 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
620 | tsk->thread.error_code = error_code; |
621 | ||
22f5991c | 622 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 623 | sig = 0; |
2d4a7167 | 624 | |
874d93d1 | 625 | oops_end(flags, regs, sig); |
1da177e4 LT |
626 | } |
627 | ||
2d4a7167 IM |
628 | static noinline void |
629 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 630 | unsigned long address, int signal, int si_code) |
92181f19 NP |
631 | { |
632 | struct task_struct *tsk = current; | |
19803078 | 633 | unsigned long *stackend; |
92181f19 NP |
634 | unsigned long flags; |
635 | int sig; | |
92181f19 | 636 | |
2d4a7167 | 637 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 AL |
638 | if (fixup_exception(regs)) { |
639 | if (current_thread_info()->sig_on_uaccess_error && signal) { | |
51e7dc70 | 640 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
641 | tsk->thread.error_code = error_code | PF_USER; |
642 | tsk->thread.cr2 = address; | |
643 | ||
644 | /* XXX: hwpoison faults will set the wrong code. */ | |
645 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
646 | } | |
92181f19 | 647 | return; |
4fc34901 | 648 | } |
92181f19 NP |
649 | |
650 | /* | |
2d4a7167 IM |
651 | * 32-bit: |
652 | * | |
653 | * Valid to do another page fault here, because if this fault | |
654 | * had been triggered by is_prefetch fixup_exception would have | |
655 | * handled it. | |
656 | * | |
657 | * 64-bit: | |
92181f19 | 658 | * |
2d4a7167 | 659 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
660 | */ |
661 | if (is_prefetch(regs, error_code, address)) | |
662 | return; | |
663 | ||
664 | if (is_errata93(regs, address)) | |
665 | return; | |
666 | ||
667 | /* | |
668 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 669 | * terminate things with extreme prejudice: |
92181f19 | 670 | */ |
92181f19 | 671 | flags = oops_begin(); |
92181f19 NP |
672 | |
673 | show_fault_oops(regs, error_code, address); | |
674 | ||
2d4a7167 | 675 | stackend = end_of_stack(tsk); |
0e7810be | 676 | if (tsk != &init_task && *stackend != STACK_END_MAGIC) |
b0f4c4b3 | 677 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 678 | |
1cc99544 | 679 | tsk->thread.cr2 = address; |
51e7dc70 | 680 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 681 | tsk->thread.error_code = error_code; |
92181f19 | 682 | |
92181f19 NP |
683 | sig = SIGKILL; |
684 | if (__die("Oops", regs, error_code)) | |
685 | sig = 0; | |
2d4a7167 | 686 | |
92181f19 | 687 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 688 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 689 | |
92181f19 | 690 | oops_end(flags, regs, sig); |
92181f19 NP |
691 | } |
692 | ||
2d4a7167 IM |
693 | /* |
694 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
695 | * sysctl is set: | |
696 | */ | |
697 | static inline void | |
698 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
699 | unsigned long address, struct task_struct *tsk) | |
700 | { | |
701 | if (!unhandled_signal(tsk, SIGSEGV)) | |
702 | return; | |
703 | ||
704 | if (!printk_ratelimit()) | |
705 | return; | |
706 | ||
a1a08d1c | 707 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
708 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
709 | tsk->comm, task_pid_nr(tsk), address, | |
710 | (void *)regs->ip, (void *)regs->sp, error_code); | |
711 | ||
712 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
713 | ||
714 | printk(KERN_CONT "\n"); | |
715 | } | |
716 | ||
717 | static void | |
718 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
719 | unsigned long address, int si_code) | |
92181f19 NP |
720 | { |
721 | struct task_struct *tsk = current; | |
722 | ||
723 | /* User mode accesses just cause a SIGSEGV */ | |
724 | if (error_code & PF_USER) { | |
725 | /* | |
2d4a7167 | 726 | * It's possible to have interrupts off here: |
92181f19 NP |
727 | */ |
728 | local_irq_enable(); | |
729 | ||
730 | /* | |
731 | * Valid to do another page fault here because this one came | |
2d4a7167 | 732 | * from user space: |
92181f19 NP |
733 | */ |
734 | if (is_prefetch(regs, error_code, address)) | |
735 | return; | |
736 | ||
737 | if (is_errata100(regs, address)) | |
738 | return; | |
739 | ||
3ae36655 AL |
740 | #ifdef CONFIG_X86_64 |
741 | /* | |
742 | * Instruction fetch faults in the vsyscall page might need | |
743 | * emulation. | |
744 | */ | |
745 | if (unlikely((error_code & PF_INSTR) && | |
746 | ((address & ~0xfff) == VSYSCALL_START))) { | |
747 | if (emulate_vsyscall(regs, address)) | |
748 | return; | |
749 | } | |
750 | #endif | |
e575a86f KC |
751 | /* Kernel addresses are always protection faults: */ |
752 | if (address >= TASK_SIZE) | |
753 | error_code |= PF_PROT; | |
3ae36655 | 754 | |
e575a86f | 755 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
756 | show_signal_msg(regs, error_code, address, tsk); |
757 | ||
2d4a7167 | 758 | tsk->thread.cr2 = address; |
e575a86f | 759 | tsk->thread.error_code = error_code; |
51e7dc70 | 760 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 761 | |
f672b49b | 762 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 763 | |
92181f19 NP |
764 | return; |
765 | } | |
766 | ||
767 | if (is_f00f_bug(regs, address)) | |
768 | return; | |
769 | ||
4fc34901 | 770 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
771 | } |
772 | ||
2d4a7167 IM |
773 | static noinline void |
774 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
775 | unsigned long address) | |
92181f19 NP |
776 | { |
777 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
778 | } | |
779 | ||
2d4a7167 IM |
780 | static void |
781 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
782 | unsigned long address, int si_code) | |
92181f19 NP |
783 | { |
784 | struct mm_struct *mm = current->mm; | |
785 | ||
786 | /* | |
787 | * Something tried to access memory that isn't in our memory map.. | |
788 | * Fix it, but check if it's kernel or user first.. | |
789 | */ | |
790 | up_read(&mm->mmap_sem); | |
791 | ||
792 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
793 | } | |
794 | ||
2d4a7167 IM |
795 | static noinline void |
796 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
797 | { |
798 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
799 | } | |
800 | ||
2d4a7167 IM |
801 | static noinline void |
802 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
803 | unsigned long address) | |
92181f19 NP |
804 | { |
805 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
806 | } | |
807 | ||
2d4a7167 | 808 | static void |
a6e04aa9 AK |
809 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
810 | unsigned int fault) | |
92181f19 NP |
811 | { |
812 | struct task_struct *tsk = current; | |
813 | struct mm_struct *mm = tsk->mm; | |
a6e04aa9 | 814 | int code = BUS_ADRERR; |
92181f19 NP |
815 | |
816 | up_read(&mm->mmap_sem); | |
817 | ||
2d4a7167 | 818 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 819 | if (!(error_code & PF_USER)) { |
4fc34901 | 820 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
821 | return; |
822 | } | |
2d4a7167 | 823 | |
cd1b68f0 | 824 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
825 | if (is_prefetch(regs, error_code, address)) |
826 | return; | |
2d4a7167 IM |
827 | |
828 | tsk->thread.cr2 = address; | |
829 | tsk->thread.error_code = error_code; | |
51e7dc70 | 830 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 831 | |
a6e04aa9 | 832 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 833 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
834 | printk(KERN_ERR |
835 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
836 | tsk->comm, tsk->pid, address); | |
837 | code = BUS_MCEERR_AR; | |
838 | } | |
839 | #endif | |
f672b49b | 840 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
841 | } |
842 | ||
b80ef10e | 843 | static noinline int |
2d4a7167 IM |
844 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
845 | unsigned long address, unsigned int fault) | |
92181f19 | 846 | { |
b80ef10e KM |
847 | /* |
848 | * Pagefault was interrupted by SIGKILL. We have no reason to | |
849 | * continue pagefault. | |
850 | */ | |
851 | if (fatal_signal_pending(current)) { | |
852 | if (!(fault & VM_FAULT_RETRY)) | |
853 | up_read(¤t->mm->mmap_sem); | |
854 | if (!(error_code & PF_USER)) | |
4fc34901 | 855 | no_context(regs, error_code, address, 0, 0); |
b80ef10e KM |
856 | return 1; |
857 | } | |
858 | if (!(fault & VM_FAULT_ERROR)) | |
859 | return 0; | |
860 | ||
2d4a7167 | 861 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
862 | /* Kernel mode? Handle exceptions or die: */ |
863 | if (!(error_code & PF_USER)) { | |
864 | up_read(¤t->mm->mmap_sem); | |
4fc34901 AL |
865 | no_context(regs, error_code, address, |
866 | SIGSEGV, SEGV_MAPERR); | |
b80ef10e | 867 | return 1; |
f8626854 AV |
868 | } |
869 | ||
c2d23f91 DR |
870 | up_read(¤t->mm->mmap_sem); |
871 | ||
872 | /* | |
873 | * We ran out of memory, call the OOM killer, and return the | |
874 | * userspace (which will retry the fault, or kill us if we got | |
875 | * oom-killed): | |
876 | */ | |
877 | pagefault_out_of_memory(); | |
2d4a7167 | 878 | } else { |
f672b49b AK |
879 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
880 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 881 | do_sigbus(regs, error_code, address, fault); |
2d4a7167 IM |
882 | else |
883 | BUG(); | |
884 | } | |
b80ef10e | 885 | return 1; |
92181f19 NP |
886 | } |
887 | ||
d8b57bb7 TG |
888 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
889 | { | |
890 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
891 | return 0; | |
2d4a7167 | 892 | |
d8b57bb7 TG |
893 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
894 | return 0; | |
895 | ||
896 | return 1; | |
897 | } | |
898 | ||
5b727a3b | 899 | /* |
2d4a7167 IM |
900 | * Handle a spurious fault caused by a stale TLB entry. |
901 | * | |
902 | * This allows us to lazily refresh the TLB when increasing the | |
903 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
904 | * eagerly is very expensive since that implies doing a full | |
905 | * cross-processor TLB flush, even if no stale TLB entries exist | |
906 | * on other processors. | |
907 | * | |
5b727a3b JF |
908 | * There are no security implications to leaving a stale TLB when |
909 | * increasing the permissions on a page. | |
910 | */ | |
62c9295f | 911 | static noinline __kprobes int |
2d4a7167 | 912 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
913 | { |
914 | pgd_t *pgd; | |
915 | pud_t *pud; | |
916 | pmd_t *pmd; | |
917 | pte_t *pte; | |
3c3e5694 | 918 | int ret; |
5b727a3b JF |
919 | |
920 | /* Reserved-bit violation or user access to kernel space? */ | |
921 | if (error_code & (PF_USER | PF_RSVD)) | |
922 | return 0; | |
923 | ||
924 | pgd = init_mm.pgd + pgd_index(address); | |
925 | if (!pgd_present(*pgd)) | |
926 | return 0; | |
927 | ||
928 | pud = pud_offset(pgd, address); | |
929 | if (!pud_present(*pud)) | |
930 | return 0; | |
931 | ||
d8b57bb7 TG |
932 | if (pud_large(*pud)) |
933 | return spurious_fault_check(error_code, (pte_t *) pud); | |
934 | ||
5b727a3b JF |
935 | pmd = pmd_offset(pud, address); |
936 | if (!pmd_present(*pmd)) | |
937 | return 0; | |
938 | ||
d8b57bb7 TG |
939 | if (pmd_large(*pmd)) |
940 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
941 | ||
5b727a3b | 942 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 943 | if (!pte_present(*pte)) |
5b727a3b JF |
944 | return 0; |
945 | ||
3c3e5694 SR |
946 | ret = spurious_fault_check(error_code, pte); |
947 | if (!ret) | |
948 | return 0; | |
949 | ||
950 | /* | |
2d4a7167 IM |
951 | * Make sure we have permissions in PMD. |
952 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
953 | */ |
954 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
955 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 956 | |
3c3e5694 | 957 | return ret; |
5b727a3b JF |
958 | } |
959 | ||
abd4f750 | 960 | int show_unhandled_signals = 1; |
1da177e4 | 961 | |
2d4a7167 | 962 | static inline int |
68da336a | 963 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 964 | { |
68da336a | 965 | if (error_code & PF_WRITE) { |
2d4a7167 | 966 | /* write, present and write, not present: */ |
92181f19 NP |
967 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
968 | return 1; | |
2d4a7167 | 969 | return 0; |
92181f19 NP |
970 | } |
971 | ||
2d4a7167 IM |
972 | /* read, present: */ |
973 | if (unlikely(error_code & PF_PROT)) | |
974 | return 1; | |
975 | ||
976 | /* read, not present: */ | |
977 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
978 | return 1; | |
979 | ||
92181f19 NP |
980 | return 0; |
981 | } | |
982 | ||
0973a06c HS |
983 | static int fault_in_kernel_space(unsigned long address) |
984 | { | |
d9517346 | 985 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
986 | } |
987 | ||
40d3cd66 PA |
988 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
989 | { | |
990 | if (error_code & PF_USER) | |
991 | return false; | |
992 | ||
993 | if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) | |
994 | return false; | |
995 | ||
996 | return true; | |
997 | } | |
998 | ||
1da177e4 LT |
999 | /* |
1000 | * This routine handles page faults. It determines the address, | |
1001 | * and the problem, and then passes it off to one of the appropriate | |
1002 | * routines. | |
1da177e4 | 1003 | */ |
6ba3c97a FW |
1004 | static void __kprobes |
1005 | __do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 | 1006 | { |
2d4a7167 | 1007 | struct vm_area_struct *vma; |
1da177e4 | 1008 | struct task_struct *tsk; |
2d4a7167 | 1009 | unsigned long address; |
1da177e4 | 1010 | struct mm_struct *mm; |
f8c2ee22 | 1011 | int fault; |
d065bd81 | 1012 | int write = error_code & PF_WRITE; |
37b23e05 | 1013 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE | |
d065bd81 | 1014 | (write ? FAULT_FLAG_WRITE : 0); |
1da177e4 | 1015 | |
a9ba9a3b AV |
1016 | tsk = current; |
1017 | mm = tsk->mm; | |
2d4a7167 | 1018 | |
2d4a7167 | 1019 | /* Get the faulting address: */ |
f51c9452 | 1020 | address = read_cr2(); |
1da177e4 | 1021 | |
f8561296 VN |
1022 | /* |
1023 | * Detect and handle instructions that would cause a page fault for | |
1024 | * both a tracked kernel page and a userspace page. | |
1025 | */ | |
1026 | if (kmemcheck_active(regs)) | |
1027 | kmemcheck_hide(regs); | |
5dfaf90f | 1028 | prefetchw(&mm->mmap_sem); |
f8561296 | 1029 | |
0fd0e3da | 1030 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1031 | return; |
1da177e4 LT |
1032 | |
1033 | /* | |
1034 | * We fault-in kernel-space virtual memory on-demand. The | |
1035 | * 'reference' page table is init_mm.pgd. | |
1036 | * | |
1037 | * NOTE! We MUST NOT take any locks for this case. We may | |
1038 | * be in an interrupt or a critical region, and should | |
1039 | * only copy the information from the master page table, | |
1040 | * nothing more. | |
1041 | * | |
1042 | * This verifies that the fault happens in kernel space | |
1043 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1044 | * protection error (error_code & 9) == 0. |
1da177e4 | 1045 | */ |
0973a06c | 1046 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1047 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1048 | if (vmalloc_fault(address) >= 0) | |
1049 | return; | |
1050 | ||
1051 | if (kmemcheck_fault(regs, address, error_code)) | |
1052 | return; | |
1053 | } | |
5b727a3b | 1054 | |
2d4a7167 | 1055 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1056 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1057 | return; |
1058 | ||
2d4a7167 | 1059 | /* kprobes don't want to hook the spurious faults: */ |
9be260a6 MH |
1060 | if (notify_page_fault(regs)) |
1061 | return; | |
f8c2ee22 HH |
1062 | /* |
1063 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1064 | * fault we could otherwise deadlock: |
f8c2ee22 | 1065 | */ |
92181f19 | 1066 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1067 | |
92181f19 | 1068 | return; |
f8c2ee22 HH |
1069 | } |
1070 | ||
2d4a7167 | 1071 | /* kprobes don't want to hook the spurious faults: */ |
f8a6b2b9 | 1072 | if (unlikely(notify_page_fault(regs))) |
9be260a6 | 1073 | return; |
f8c2ee22 | 1074 | /* |
891cffbd LT |
1075 | * It's safe to allow irq's after cr2 has been saved and the |
1076 | * vmalloc fault has been handled. | |
1077 | * | |
1078 | * User-mode registers count as a user access even for any | |
2d4a7167 | 1079 | * potential system fault or CPU buglet: |
f8c2ee22 | 1080 | */ |
891cffbd LT |
1081 | if (user_mode_vm(regs)) { |
1082 | local_irq_enable(); | |
1083 | error_code |= PF_USER; | |
2d4a7167 IM |
1084 | } else { |
1085 | if (regs->flags & X86_EFLAGS_IF) | |
1086 | local_irq_enable(); | |
1087 | } | |
8c914cb7 | 1088 | |
66c58156 | 1089 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1090 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1091 | |
40d3cd66 PA |
1092 | if (static_cpu_has(X86_FEATURE_SMAP)) { |
1093 | if (unlikely(smap_violation(error_code, regs))) { | |
1094 | bad_area_nosemaphore(regs, error_code, address); | |
1095 | return; | |
1096 | } | |
1097 | } | |
1098 | ||
a8b0ca17 | 1099 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
7dd1fcc2 | 1100 | |
1da177e4 | 1101 | /* |
2d4a7167 IM |
1102 | * If we're in an interrupt, have no user context or are running |
1103 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1104 | */ |
92181f19 NP |
1105 | if (unlikely(in_atomic() || !mm)) { |
1106 | bad_area_nosemaphore(regs, error_code, address); | |
1107 | return; | |
1108 | } | |
1da177e4 | 1109 | |
3a1dfe6e IM |
1110 | /* |
1111 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1112 | * addresses in user space. All other faults represent errors in |
1113 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1114 | * case of an erroneous fault occurring in a code path which already | |
1115 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1116 | * against the address space. Luckily the kernel only validly | |
1117 | * references user space from well defined areas of code, which are | |
1118 | * listed in the exceptions table. | |
1da177e4 LT |
1119 | * |
1120 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1121 | * the source reference check when there is a possibility of a |
1122 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1123 | * validate the source. If this is invalid we can skip the address | |
1124 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1125 | */ |
92181f19 | 1126 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1127 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1128 | !search_exception_tables(regs->ip)) { |
1129 | bad_area_nosemaphore(regs, error_code, address); | |
1130 | return; | |
1131 | } | |
d065bd81 | 1132 | retry: |
1da177e4 | 1133 | down_read(&mm->mmap_sem); |
01006074 PZ |
1134 | } else { |
1135 | /* | |
2d4a7167 IM |
1136 | * The above down_read_trylock() might have succeeded in |
1137 | * which case we'll have missed the might_sleep() from | |
1138 | * down_read(): | |
01006074 PZ |
1139 | */ |
1140 | might_sleep(); | |
1da177e4 LT |
1141 | } |
1142 | ||
1143 | vma = find_vma(mm, address); | |
92181f19 NP |
1144 | if (unlikely(!vma)) { |
1145 | bad_area(regs, error_code, address); | |
1146 | return; | |
1147 | } | |
1148 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1149 | goto good_area; |
92181f19 NP |
1150 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1151 | bad_area(regs, error_code, address); | |
1152 | return; | |
1153 | } | |
33cb5243 | 1154 | if (error_code & PF_USER) { |
6f4d368e HH |
1155 | /* |
1156 | * Accessing the stack below %sp is always a bug. | |
1157 | * The large cushion allows instructions like enter | |
2d4a7167 | 1158 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1159 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1160 | */ |
92181f19 NP |
1161 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1162 | bad_area(regs, error_code, address); | |
1163 | return; | |
1164 | } | |
1da177e4 | 1165 | } |
92181f19 NP |
1166 | if (unlikely(expand_stack(vma, address))) { |
1167 | bad_area(regs, error_code, address); | |
1168 | return; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * Ok, we have a good vm_area for this memory access, so | |
1173 | * we can handle it.. | |
1174 | */ | |
1da177e4 | 1175 | good_area: |
68da336a | 1176 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1177 | bad_area_access_error(regs, error_code, address); |
1178 | return; | |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /* | |
1182 | * If for any reason at all we couldn't handle the fault, | |
1183 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1184 | * the fault: |
1da177e4 | 1185 | */ |
d065bd81 | 1186 | fault = handle_mm_fault(mm, vma, address, flags); |
2d4a7167 | 1187 | |
b80ef10e KM |
1188 | if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) { |
1189 | if (mm_fault_error(regs, error_code, address, fault)) | |
1190 | return; | |
37b23e05 KM |
1191 | } |
1192 | ||
d065bd81 ML |
1193 | /* |
1194 | * Major/minor page fault accounting is only done on the | |
1195 | * initial attempt. If we go through a retry, it is extremely | |
1196 | * likely that the page will be found in page cache at that point. | |
1197 | */ | |
1198 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1199 | if (fault & VM_FAULT_MAJOR) { | |
1200 | tsk->maj_flt++; | |
a8b0ca17 | 1201 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
d065bd81 ML |
1202 | regs, address); |
1203 | } else { | |
1204 | tsk->min_flt++; | |
a8b0ca17 | 1205 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
d065bd81 ML |
1206 | regs, address); |
1207 | } | |
1208 | if (fault & VM_FAULT_RETRY) { | |
1209 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
1210 | * of starvation. */ | |
1211 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 1212 | flags |= FAULT_FLAG_TRIED; |
d065bd81 ML |
1213 | goto retry; |
1214 | } | |
ac17dc8e | 1215 | } |
d729ab35 | 1216 | |
8c938f9f IM |
1217 | check_v8086_mode(regs, address, tsk); |
1218 | ||
1da177e4 | 1219 | up_read(&mm->mmap_sem); |
1da177e4 | 1220 | } |
6ba3c97a FW |
1221 | |
1222 | dotraplinkage void __kprobes | |
1223 | do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1224 | { | |
1225 | exception_enter(regs); | |
1226 | __do_page_fault(regs, error_code); | |
1227 | exception_exit(regs); | |
1228 | } |