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_*(), ... */ |
cedf03bd | 20 | #include <asm/vsyscall.h> |
1da177e4 | 21 | |
33cb5243 | 22 | /* |
2d4a7167 IM |
23 | * Page fault error code bits: |
24 | * | |
25 | * bit 0 == 0: no page found 1: protection fault | |
26 | * bit 1 == 0: read access 1: write access | |
27 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
28 | * bit 3 == 1: use of reserved bit detected | |
29 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 30 | */ |
2d4a7167 IM |
31 | enum x86_pf_error_code { |
32 | ||
33 | PF_PROT = 1 << 0, | |
34 | PF_WRITE = 1 << 1, | |
35 | PF_USER = 1 << 2, | |
36 | PF_RSVD = 1 << 3, | |
37 | PF_INSTR = 1 << 4, | |
38 | }; | |
66c58156 | 39 | |
b814d41f | 40 | /* |
b319eed0 IM |
41 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
42 | * handled by mmiotrace: | |
b814d41f | 43 | */ |
62c9295f MH |
44 | static inline int __kprobes |
45 | kmmio_fault(struct pt_regs *regs, unsigned long addr) | |
86069782 | 46 | { |
0fd0e3da PP |
47 | if (unlikely(is_kmmio_active())) |
48 | if (kmmio_handler(regs, addr) == 1) | |
49 | return -1; | |
0fd0e3da | 50 | return 0; |
86069782 PP |
51 | } |
52 | ||
62c9295f | 53 | static inline int __kprobes notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 54 | { |
74a0b576 CH |
55 | int ret = 0; |
56 | ||
57 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 58 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
59 | preempt_disable(); |
60 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
61 | ret = 1; | |
62 | preempt_enable(); | |
63 | } | |
1bd858a5 | 64 | |
74a0b576 | 65 | return ret; |
33cb5243 | 66 | } |
1bd858a5 | 67 | |
1dc85be0 | 68 | /* |
2d4a7167 IM |
69 | * Prefetch quirks: |
70 | * | |
71 | * 32-bit mode: | |
72 | * | |
73 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
74 | * Check that here and ignore it. | |
1dc85be0 | 75 | * |
2d4a7167 | 76 | * 64-bit mode: |
1dc85be0 | 77 | * |
2d4a7167 IM |
78 | * Sometimes the CPU reports invalid exceptions on prefetch. |
79 | * Check that here and ignore it. | |
80 | * | |
81 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 82 | */ |
107a0367 IM |
83 | static inline int |
84 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
85 | unsigned char opcode, int *prefetch) | |
86 | { | |
87 | unsigned char instr_hi = opcode & 0xf0; | |
88 | unsigned char instr_lo = opcode & 0x0f; | |
89 | ||
90 | switch (instr_hi) { | |
91 | case 0x20: | |
92 | case 0x30: | |
93 | /* | |
94 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
95 | * In X86_64 long mode, the CPU will signal invalid | |
96 | * opcode if some of these prefixes are present so | |
97 | * X86_64 will never get here anyway | |
98 | */ | |
99 | return ((instr_lo & 7) == 0x6); | |
100 | #ifdef CONFIG_X86_64 | |
101 | case 0x40: | |
102 | /* | |
103 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
104 | * Need to figure out under what instruction mode the | |
105 | * instruction was issued. Could check the LDT for lm, | |
106 | * but for now it's good enough to assume that long | |
107 | * mode only uses well known segments or kernel. | |
108 | */ | |
318f5a2a | 109 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
110 | #endif |
111 | case 0x60: | |
112 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
113 | return (instr_lo & 0xC) == 0x4; | |
114 | case 0xF0: | |
115 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
116 | return !instr_lo || (instr_lo>>1) == 1; | |
117 | case 0x00: | |
118 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
119 | if (probe_kernel_address(instr, opcode)) | |
120 | return 0; | |
121 | ||
122 | *prefetch = (instr_lo == 0xF) && | |
123 | (opcode == 0x0D || opcode == 0x18); | |
124 | return 0; | |
125 | default: | |
126 | return 0; | |
127 | } | |
128 | } | |
129 | ||
2d4a7167 IM |
130 | static int |
131 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 132 | { |
2d4a7167 | 133 | unsigned char *max_instr; |
ab2bf0c1 | 134 | unsigned char *instr; |
33cb5243 | 135 | int prefetch = 0; |
1da177e4 | 136 | |
3085354d IM |
137 | /* |
138 | * If it was a exec (instruction fetch) fault on NX page, then | |
139 | * do not ignore the fault: | |
140 | */ | |
66c58156 | 141 | if (error_code & PF_INSTR) |
1da177e4 | 142 | return 0; |
1dc85be0 | 143 | |
107a0367 | 144 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 145 | max_instr = instr + 15; |
1da177e4 | 146 | |
76381fee | 147 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
148 | return 0; |
149 | ||
107a0367 | 150 | while (instr < max_instr) { |
2d4a7167 | 151 | unsigned char opcode; |
1da177e4 | 152 | |
ab2bf0c1 | 153 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 154 | break; |
1da177e4 | 155 | |
1da177e4 LT |
156 | instr++; |
157 | ||
107a0367 | 158 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 159 | break; |
1da177e4 LT |
160 | } |
161 | return prefetch; | |
162 | } | |
163 | ||
2d4a7167 IM |
164 | static void |
165 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 166 | struct task_struct *tsk, int fault) |
c4aba4a8 | 167 | { |
f672b49b | 168 | unsigned lsb = 0; |
c4aba4a8 HH |
169 | siginfo_t info; |
170 | ||
2d4a7167 IM |
171 | info.si_signo = si_signo; |
172 | info.si_errno = 0; | |
173 | info.si_code = si_code; | |
174 | info.si_addr = (void __user *)address; | |
f672b49b AK |
175 | if (fault & VM_FAULT_HWPOISON_LARGE) |
176 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
177 | if (fault & VM_FAULT_HWPOISON) | |
178 | lsb = PAGE_SHIFT; | |
179 | info.si_addr_lsb = lsb; | |
2d4a7167 | 180 | |
c4aba4a8 HH |
181 | force_sig_info(si_signo, &info, tsk); |
182 | } | |
183 | ||
f2f13a85 IM |
184 | DEFINE_SPINLOCK(pgd_lock); |
185 | LIST_HEAD(pgd_list); | |
186 | ||
187 | #ifdef CONFIG_X86_32 | |
188 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 189 | { |
f2f13a85 IM |
190 | unsigned index = pgd_index(address); |
191 | pgd_t *pgd_k; | |
192 | pud_t *pud, *pud_k; | |
193 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 194 | |
f2f13a85 IM |
195 | pgd += index; |
196 | pgd_k = init_mm.pgd + index; | |
197 | ||
198 | if (!pgd_present(*pgd_k)) | |
199 | return NULL; | |
200 | ||
201 | /* | |
202 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
203 | * and redundant with the set_pmd() on non-PAE. As would | |
204 | * set_pud. | |
205 | */ | |
206 | pud = pud_offset(pgd, address); | |
207 | pud_k = pud_offset(pgd_k, address); | |
208 | if (!pud_present(*pud_k)) | |
209 | return NULL; | |
210 | ||
211 | pmd = pmd_offset(pud, address); | |
212 | pmd_k = pmd_offset(pud_k, address); | |
213 | if (!pmd_present(*pmd_k)) | |
214 | return NULL; | |
215 | ||
b8bcfe99 | 216 | if (!pmd_present(*pmd)) |
f2f13a85 | 217 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 218 | else |
f2f13a85 | 219 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
220 | |
221 | return pmd_k; | |
222 | } | |
223 | ||
224 | void vmalloc_sync_all(void) | |
225 | { | |
226 | unsigned long address; | |
227 | ||
228 | if (SHARED_KERNEL_PMD) | |
229 | return; | |
230 | ||
231 | for (address = VMALLOC_START & PMD_MASK; | |
232 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
233 | address += PMD_SIZE) { | |
f2f13a85 IM |
234 | struct page *page; |
235 | ||
a79e53d8 | 236 | spin_lock(&pgd_lock); |
f2f13a85 | 237 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 238 | spinlock_t *pgt_lock; |
f01f7c56 | 239 | pmd_t *ret; |
617d34d9 | 240 | |
a79e53d8 | 241 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
242 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
243 | ||
244 | spin_lock(pgt_lock); | |
245 | ret = vmalloc_sync_one(page_address(page), address); | |
246 | spin_unlock(pgt_lock); | |
247 | ||
248 | if (!ret) | |
f2f13a85 IM |
249 | break; |
250 | } | |
a79e53d8 | 251 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
252 | } |
253 | } | |
254 | ||
255 | /* | |
256 | * 32-bit: | |
257 | * | |
258 | * Handle a fault on the vmalloc or module mapping area | |
259 | */ | |
62c9295f | 260 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
261 | { |
262 | unsigned long pgd_paddr; | |
263 | pmd_t *pmd_k; | |
264 | pte_t *pte_k; | |
265 | ||
266 | /* Make sure we are in vmalloc area: */ | |
267 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
268 | return -1; | |
269 | ||
ebc8827f FW |
270 | WARN_ON_ONCE(in_nmi()); |
271 | ||
f2f13a85 IM |
272 | /* |
273 | * Synchronize this task's top level page-table | |
274 | * with the 'reference' page table. | |
275 | * | |
276 | * Do _not_ use "current" here. We might be inside | |
277 | * an interrupt in the middle of a task switch.. | |
278 | */ | |
279 | pgd_paddr = read_cr3(); | |
280 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
281 | if (!pmd_k) | |
282 | return -1; | |
283 | ||
284 | pte_k = pte_offset_kernel(pmd_k, address); | |
285 | if (!pte_present(*pte_k)) | |
286 | return -1; | |
287 | ||
288 | return 0; | |
289 | } | |
290 | ||
291 | /* | |
292 | * Did it hit the DOS screen memory VA from vm86 mode? | |
293 | */ | |
294 | static inline void | |
295 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
296 | struct task_struct *tsk) | |
297 | { | |
298 | unsigned long bit; | |
299 | ||
300 | if (!v8086_mode(regs)) | |
301 | return; | |
302 | ||
303 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
304 | if (bit < 32) | |
305 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 306 | } |
1da177e4 | 307 | |
087975b0 | 308 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 309 | { |
087975b0 AM |
310 | return pfn < max_low_pfn; |
311 | } | |
1156e098 | 312 | |
087975b0 AM |
313 | static void dump_pagetable(unsigned long address) |
314 | { | |
315 | pgd_t *base = __va(read_cr3()); | |
316 | pgd_t *pgd = &base[pgd_index(address)]; | |
317 | pmd_t *pmd; | |
318 | pte_t *pte; | |
2d4a7167 | 319 | |
1156e098 | 320 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
321 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
322 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
323 | goto out; | |
1156e098 | 324 | #endif |
087975b0 AM |
325 | pmd = pmd_offset(pud_offset(pgd, address), address); |
326 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
327 | |
328 | /* | |
329 | * We must not directly access the pte in the highpte | |
330 | * case if the page table is located in highmem. | |
331 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 332 | * it's allocated already: |
1156e098 | 333 | */ |
087975b0 AM |
334 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
335 | goto out; | |
1156e098 | 336 | |
087975b0 AM |
337 | pte = pte_offset_kernel(pmd, address); |
338 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
339 | out: | |
1156e098 | 340 | printk("\n"); |
f2f13a85 IM |
341 | } |
342 | ||
343 | #else /* CONFIG_X86_64: */ | |
344 | ||
345 | void vmalloc_sync_all(void) | |
346 | { | |
6afb5157 | 347 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
348 | } |
349 | ||
350 | /* | |
351 | * 64-bit: | |
352 | * | |
353 | * Handle a fault on the vmalloc area | |
354 | * | |
355 | * This assumes no large pages in there. | |
356 | */ | |
62c9295f | 357 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
358 | { |
359 | pgd_t *pgd, *pgd_ref; | |
360 | pud_t *pud, *pud_ref; | |
361 | pmd_t *pmd, *pmd_ref; | |
362 | pte_t *pte, *pte_ref; | |
363 | ||
364 | /* Make sure we are in vmalloc area: */ | |
365 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
366 | return -1; | |
367 | ||
ebc8827f FW |
368 | WARN_ON_ONCE(in_nmi()); |
369 | ||
f2f13a85 IM |
370 | /* |
371 | * Copy kernel mappings over when needed. This can also | |
372 | * happen within a race in page table update. In the later | |
373 | * case just flush: | |
374 | */ | |
375 | pgd = pgd_offset(current->active_mm, address); | |
376 | pgd_ref = pgd_offset_k(address); | |
377 | if (pgd_none(*pgd_ref)) | |
378 | return -1; | |
379 | ||
380 | if (pgd_none(*pgd)) | |
381 | set_pgd(pgd, *pgd_ref); | |
382 | else | |
383 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
384 | ||
385 | /* | |
386 | * Below here mismatches are bugs because these lower tables | |
387 | * are shared: | |
388 | */ | |
389 | ||
390 | pud = pud_offset(pgd, address); | |
391 | pud_ref = pud_offset(pgd_ref, address); | |
392 | if (pud_none(*pud_ref)) | |
393 | return -1; | |
394 | ||
395 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
396 | BUG(); | |
397 | ||
398 | pmd = pmd_offset(pud, address); | |
399 | pmd_ref = pmd_offset(pud_ref, address); | |
400 | if (pmd_none(*pmd_ref)) | |
401 | return -1; | |
402 | ||
403 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
404 | BUG(); | |
405 | ||
406 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
407 | if (!pte_present(*pte_ref)) | |
408 | return -1; | |
409 | ||
410 | pte = pte_offset_kernel(pmd, address); | |
411 | ||
412 | /* | |
413 | * Don't use pte_page here, because the mappings can point | |
414 | * outside mem_map, and the NUMA hash lookup cannot handle | |
415 | * that: | |
416 | */ | |
417 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
418 | BUG(); | |
419 | ||
420 | return 0; | |
421 | } | |
422 | ||
423 | static const char errata93_warning[] = | |
ad361c98 JP |
424 | KERN_ERR |
425 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
426 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
427 | "******* Please consider a BIOS update.\n" | |
428 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
f2f13a85 IM |
429 | |
430 | /* | |
431 | * No vm86 mode in 64-bit mode: | |
432 | */ | |
433 | static inline void | |
434 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
435 | struct task_struct *tsk) | |
436 | { | |
437 | } | |
438 | ||
439 | static int bad_address(void *p) | |
440 | { | |
441 | unsigned long dummy; | |
442 | ||
443 | return probe_kernel_address((unsigned long *)p, dummy); | |
444 | } | |
445 | ||
446 | static void dump_pagetable(unsigned long address) | |
447 | { | |
087975b0 AM |
448 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
449 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
450 | pud_t *pud; |
451 | pmd_t *pmd; | |
452 | pte_t *pte; | |
453 | ||
2d4a7167 IM |
454 | if (bad_address(pgd)) |
455 | goto bad; | |
456 | ||
d646bce4 | 457 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
458 | |
459 | if (!pgd_present(*pgd)) | |
460 | goto out; | |
1da177e4 | 461 | |
d2ae5b5f | 462 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
463 | if (bad_address(pud)) |
464 | goto bad; | |
465 | ||
1da177e4 | 466 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 467 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 468 | goto out; |
1da177e4 LT |
469 | |
470 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
471 | if (bad_address(pmd)) |
472 | goto bad; | |
473 | ||
1da177e4 | 474 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
475 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
476 | goto out; | |
1da177e4 LT |
477 | |
478 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
479 | if (bad_address(pte)) |
480 | goto bad; | |
481 | ||
33cb5243 | 482 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 483 | out: |
1da177e4 LT |
484 | printk("\n"); |
485 | return; | |
486 | bad: | |
487 | printk("BAD\n"); | |
8c938f9f IM |
488 | } |
489 | ||
f2f13a85 | 490 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 491 | |
2d4a7167 IM |
492 | /* |
493 | * Workaround for K8 erratum #93 & buggy BIOS. | |
494 | * | |
495 | * BIOS SMM functions are required to use a specific workaround | |
496 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
497 | * | |
498 | * A lot of BIOS that didn't get tested properly miss this. | |
499 | * | |
500 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
501 | * Try to work around it here. | |
502 | * | |
503 | * Note we only handle faults in kernel here. | |
504 | * Does nothing on 32-bit. | |
fdfe8aa8 | 505 | */ |
33cb5243 | 506 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 507 | { |
fdfe8aa8 | 508 | #ifdef CONFIG_X86_64 |
65ea5b03 | 509 | if (address != regs->ip) |
1da177e4 | 510 | return 0; |
2d4a7167 | 511 | |
33cb5243 | 512 | if ((address >> 32) != 0) |
1da177e4 | 513 | return 0; |
2d4a7167 | 514 | |
1da177e4 | 515 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
516 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
517 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 518 | printk_once(errata93_warning); |
65ea5b03 | 519 | regs->ip = address; |
1da177e4 LT |
520 | return 1; |
521 | } | |
fdfe8aa8 | 522 | #endif |
1da177e4 | 523 | return 0; |
33cb5243 | 524 | } |
1da177e4 | 525 | |
35f3266f | 526 | /* |
2d4a7167 IM |
527 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
528 | * to illegal addresses >4GB. | |
529 | * | |
530 | * We catch this in the page fault handler because these addresses | |
531 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
532 | * segment in LDT is compatibility mode. |
533 | */ | |
534 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
535 | { | |
536 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 537 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
538 | return 1; |
539 | #endif | |
540 | return 0; | |
541 | } | |
542 | ||
29caf2f9 HH |
543 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
544 | { | |
545 | #ifdef CONFIG_X86_F00F_BUG | |
546 | unsigned long nr; | |
2d4a7167 | 547 | |
29caf2f9 | 548 | /* |
2d4a7167 | 549 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 HH |
550 | */ |
551 | if (boot_cpu_data.f00f_bug) { | |
552 | nr = (address - idt_descr.address) >> 3; | |
553 | ||
554 | if (nr == 6) { | |
555 | do_invalid_op(regs, 0); | |
556 | return 1; | |
557 | } | |
558 | } | |
559 | #endif | |
560 | return 0; | |
561 | } | |
562 | ||
8f766149 IM |
563 | static const char nx_warning[] = KERN_CRIT |
564 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
565 | ||
2d4a7167 IM |
566 | static void |
567 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
568 | unsigned long address) | |
b3279c7f | 569 | { |
1156e098 HH |
570 | if (!oops_may_print()) |
571 | return; | |
572 | ||
1156e098 | 573 | if (error_code & PF_INSTR) { |
93809be8 | 574 | unsigned int level; |
2d4a7167 | 575 | |
1156e098 HH |
576 | pte_t *pte = lookup_address(address, &level); |
577 | ||
8f766149 IM |
578 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
579 | printk(nx_warning, current_uid()); | |
1156e098 | 580 | } |
1156e098 | 581 | |
19f0dda9 | 582 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 583 | if (address < PAGE_SIZE) |
19f0dda9 | 584 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 585 | else |
19f0dda9 | 586 | printk(KERN_CONT "paging request"); |
2d4a7167 | 587 | |
f294a8ce | 588 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 589 | printk(KERN_ALERT "IP:"); |
b3279c7f | 590 | printk_address(regs->ip, 1); |
2d4a7167 | 591 | |
b3279c7f HH |
592 | dump_pagetable(address); |
593 | } | |
594 | ||
2d4a7167 IM |
595 | static noinline void |
596 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
597 | unsigned long address) | |
1da177e4 | 598 | { |
2d4a7167 IM |
599 | struct task_struct *tsk; |
600 | unsigned long flags; | |
601 | int sig; | |
602 | ||
603 | flags = oops_begin(); | |
604 | tsk = current; | |
605 | sig = SIGKILL; | |
1209140c | 606 | |
1da177e4 | 607 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 608 | tsk->comm, address); |
1da177e4 | 609 | dump_pagetable(address); |
2d4a7167 IM |
610 | |
611 | tsk->thread.cr2 = address; | |
612 | tsk->thread.trap_no = 14; | |
613 | tsk->thread.error_code = error_code; | |
614 | ||
22f5991c | 615 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 616 | sig = 0; |
2d4a7167 | 617 | |
874d93d1 | 618 | oops_end(flags, regs, sig); |
1da177e4 LT |
619 | } |
620 | ||
2d4a7167 IM |
621 | static noinline void |
622 | no_context(struct pt_regs *regs, unsigned long error_code, | |
623 | unsigned long address) | |
92181f19 NP |
624 | { |
625 | struct task_struct *tsk = current; | |
19803078 | 626 | unsigned long *stackend; |
92181f19 NP |
627 | unsigned long flags; |
628 | int sig; | |
92181f19 | 629 | |
2d4a7167 | 630 | /* Are we prepared to handle this kernel fault? */ |
92181f19 NP |
631 | if (fixup_exception(regs)) |
632 | return; | |
633 | ||
634 | /* | |
2d4a7167 IM |
635 | * 32-bit: |
636 | * | |
637 | * Valid to do another page fault here, because if this fault | |
638 | * had been triggered by is_prefetch fixup_exception would have | |
639 | * handled it. | |
640 | * | |
641 | * 64-bit: | |
92181f19 | 642 | * |
2d4a7167 | 643 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
644 | */ |
645 | if (is_prefetch(regs, error_code, address)) | |
646 | return; | |
647 | ||
648 | if (is_errata93(regs, address)) | |
649 | return; | |
650 | ||
651 | /* | |
652 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 653 | * terminate things with extreme prejudice: |
92181f19 | 654 | */ |
92181f19 | 655 | flags = oops_begin(); |
92181f19 NP |
656 | |
657 | show_fault_oops(regs, error_code, address); | |
658 | ||
2d4a7167 | 659 | stackend = end_of_stack(tsk); |
0e7810be | 660 | if (tsk != &init_task && *stackend != STACK_END_MAGIC) |
19803078 IM |
661 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); |
662 | ||
1cc99544 IM |
663 | tsk->thread.cr2 = address; |
664 | tsk->thread.trap_no = 14; | |
665 | tsk->thread.error_code = error_code; | |
92181f19 | 666 | |
92181f19 NP |
667 | sig = SIGKILL; |
668 | if (__die("Oops", regs, error_code)) | |
669 | sig = 0; | |
2d4a7167 | 670 | |
92181f19 NP |
671 | /* Executive summary in case the body of the oops scrolled away */ |
672 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
2d4a7167 | 673 | |
92181f19 | 674 | oops_end(flags, regs, sig); |
92181f19 NP |
675 | } |
676 | ||
2d4a7167 IM |
677 | /* |
678 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
679 | * sysctl is set: | |
680 | */ | |
681 | static inline void | |
682 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
683 | unsigned long address, struct task_struct *tsk) | |
684 | { | |
685 | if (!unhandled_signal(tsk, SIGSEGV)) | |
686 | return; | |
687 | ||
688 | if (!printk_ratelimit()) | |
689 | return; | |
690 | ||
a1a08d1c | 691 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
692 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
693 | tsk->comm, task_pid_nr(tsk), address, | |
694 | (void *)regs->ip, (void *)regs->sp, error_code); | |
695 | ||
696 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
697 | ||
698 | printk(KERN_CONT "\n"); | |
699 | } | |
700 | ||
701 | static void | |
702 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
703 | unsigned long address, int si_code) | |
92181f19 NP |
704 | { |
705 | struct task_struct *tsk = current; | |
706 | ||
707 | /* User mode accesses just cause a SIGSEGV */ | |
708 | if (error_code & PF_USER) { | |
709 | /* | |
2d4a7167 | 710 | * It's possible to have interrupts off here: |
92181f19 NP |
711 | */ |
712 | local_irq_enable(); | |
713 | ||
714 | /* | |
715 | * Valid to do another page fault here because this one came | |
2d4a7167 | 716 | * from user space: |
92181f19 NP |
717 | */ |
718 | if (is_prefetch(regs, error_code, address)) | |
719 | return; | |
720 | ||
721 | if (is_errata100(regs, address)) | |
722 | return; | |
723 | ||
3ae36655 AL |
724 | #ifdef CONFIG_X86_64 |
725 | /* | |
726 | * Instruction fetch faults in the vsyscall page might need | |
727 | * emulation. | |
728 | */ | |
729 | if (unlikely((error_code & PF_INSTR) && | |
730 | ((address & ~0xfff) == VSYSCALL_START))) { | |
731 | if (emulate_vsyscall(regs, address)) | |
732 | return; | |
733 | } | |
734 | #endif | |
735 | ||
2d4a7167 IM |
736 | if (unlikely(show_unhandled_signals)) |
737 | show_signal_msg(regs, error_code, address, tsk); | |
738 | ||
739 | /* Kernel addresses are always protection faults: */ | |
740 | tsk->thread.cr2 = address; | |
741 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
742 | tsk->thread.trap_no = 14; | |
92181f19 | 743 | |
f672b49b | 744 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 745 | |
92181f19 NP |
746 | return; |
747 | } | |
748 | ||
749 | if (is_f00f_bug(regs, address)) | |
750 | return; | |
751 | ||
752 | no_context(regs, error_code, address); | |
753 | } | |
754 | ||
2d4a7167 IM |
755 | static noinline void |
756 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
757 | unsigned long address) | |
92181f19 NP |
758 | { |
759 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
760 | } | |
761 | ||
2d4a7167 IM |
762 | static void |
763 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
764 | unsigned long address, int si_code) | |
92181f19 NP |
765 | { |
766 | struct mm_struct *mm = current->mm; | |
767 | ||
768 | /* | |
769 | * Something tried to access memory that isn't in our memory map.. | |
770 | * Fix it, but check if it's kernel or user first.. | |
771 | */ | |
772 | up_read(&mm->mmap_sem); | |
773 | ||
774 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
775 | } | |
776 | ||
2d4a7167 IM |
777 | static noinline void |
778 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
779 | { |
780 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
781 | } | |
782 | ||
2d4a7167 IM |
783 | static noinline void |
784 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
785 | unsigned long address) | |
92181f19 NP |
786 | { |
787 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
788 | } | |
789 | ||
790 | /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */ | |
2d4a7167 IM |
791 | static void |
792 | out_of_memory(struct pt_regs *regs, unsigned long error_code, | |
793 | unsigned long address) | |
92181f19 NP |
794 | { |
795 | /* | |
796 | * We ran out of memory, call the OOM killer, and return the userspace | |
2d4a7167 | 797 | * (which will retry the fault, or kill us if we got oom-killed): |
92181f19 NP |
798 | */ |
799 | up_read(¤t->mm->mmap_sem); | |
2d4a7167 | 800 | |
92181f19 NP |
801 | pagefault_out_of_memory(); |
802 | } | |
803 | ||
2d4a7167 | 804 | static void |
a6e04aa9 AK |
805 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
806 | unsigned int fault) | |
92181f19 NP |
807 | { |
808 | struct task_struct *tsk = current; | |
809 | struct mm_struct *mm = tsk->mm; | |
a6e04aa9 | 810 | int code = BUS_ADRERR; |
92181f19 NP |
811 | |
812 | up_read(&mm->mmap_sem); | |
813 | ||
2d4a7167 | 814 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 815 | if (!(error_code & PF_USER)) { |
92181f19 | 816 | no_context(regs, error_code, address); |
96054569 LT |
817 | return; |
818 | } | |
2d4a7167 | 819 | |
cd1b68f0 | 820 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
821 | if (is_prefetch(regs, error_code, address)) |
822 | return; | |
2d4a7167 IM |
823 | |
824 | tsk->thread.cr2 = address; | |
825 | tsk->thread.error_code = error_code; | |
826 | tsk->thread.trap_no = 14; | |
827 | ||
a6e04aa9 | 828 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 829 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
830 | printk(KERN_ERR |
831 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
832 | tsk->comm, tsk->pid, address); | |
833 | code = BUS_MCEERR_AR; | |
834 | } | |
835 | #endif | |
f672b49b | 836 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
837 | } |
838 | ||
b80ef10e | 839 | static noinline int |
2d4a7167 IM |
840 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
841 | unsigned long address, unsigned int fault) | |
92181f19 | 842 | { |
b80ef10e KM |
843 | /* |
844 | * Pagefault was interrupted by SIGKILL. We have no reason to | |
845 | * continue pagefault. | |
846 | */ | |
847 | if (fatal_signal_pending(current)) { | |
848 | if (!(fault & VM_FAULT_RETRY)) | |
849 | up_read(¤t->mm->mmap_sem); | |
850 | if (!(error_code & PF_USER)) | |
851 | no_context(regs, error_code, address); | |
852 | return 1; | |
853 | } | |
854 | if (!(fault & VM_FAULT_ERROR)) | |
855 | return 0; | |
856 | ||
2d4a7167 | 857 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
858 | /* Kernel mode? Handle exceptions or die: */ |
859 | if (!(error_code & PF_USER)) { | |
860 | up_read(¤t->mm->mmap_sem); | |
861 | no_context(regs, error_code, address); | |
b80ef10e | 862 | return 1; |
f8626854 AV |
863 | } |
864 | ||
92181f19 | 865 | out_of_memory(regs, error_code, address); |
2d4a7167 | 866 | } else { |
f672b49b AK |
867 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
868 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 869 | do_sigbus(regs, error_code, address, fault); |
2d4a7167 IM |
870 | else |
871 | BUG(); | |
872 | } | |
b80ef10e | 873 | return 1; |
92181f19 NP |
874 | } |
875 | ||
d8b57bb7 TG |
876 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
877 | { | |
878 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
879 | return 0; | |
2d4a7167 | 880 | |
d8b57bb7 TG |
881 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
882 | return 0; | |
883 | ||
884 | return 1; | |
885 | } | |
886 | ||
5b727a3b | 887 | /* |
2d4a7167 IM |
888 | * Handle a spurious fault caused by a stale TLB entry. |
889 | * | |
890 | * This allows us to lazily refresh the TLB when increasing the | |
891 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
892 | * eagerly is very expensive since that implies doing a full | |
893 | * cross-processor TLB flush, even if no stale TLB entries exist | |
894 | * on other processors. | |
895 | * | |
5b727a3b JF |
896 | * There are no security implications to leaving a stale TLB when |
897 | * increasing the permissions on a page. | |
898 | */ | |
62c9295f | 899 | static noinline __kprobes int |
2d4a7167 | 900 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
901 | { |
902 | pgd_t *pgd; | |
903 | pud_t *pud; | |
904 | pmd_t *pmd; | |
905 | pte_t *pte; | |
3c3e5694 | 906 | int ret; |
5b727a3b JF |
907 | |
908 | /* Reserved-bit violation or user access to kernel space? */ | |
909 | if (error_code & (PF_USER | PF_RSVD)) | |
910 | return 0; | |
911 | ||
912 | pgd = init_mm.pgd + pgd_index(address); | |
913 | if (!pgd_present(*pgd)) | |
914 | return 0; | |
915 | ||
916 | pud = pud_offset(pgd, address); | |
917 | if (!pud_present(*pud)) | |
918 | return 0; | |
919 | ||
d8b57bb7 TG |
920 | if (pud_large(*pud)) |
921 | return spurious_fault_check(error_code, (pte_t *) pud); | |
922 | ||
5b727a3b JF |
923 | pmd = pmd_offset(pud, address); |
924 | if (!pmd_present(*pmd)) | |
925 | return 0; | |
926 | ||
d8b57bb7 TG |
927 | if (pmd_large(*pmd)) |
928 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
929 | ||
660a293e SL |
930 | /* |
931 | * Note: don't use pte_present() here, since it returns true | |
932 | * if the _PAGE_PROTNONE bit is set. However, this aliases the | |
933 | * _PAGE_GLOBAL bit, which for kernel pages give false positives | |
934 | * when CONFIG_DEBUG_PAGEALLOC is used. | |
935 | */ | |
5b727a3b | 936 | pte = pte_offset_kernel(pmd, address); |
660a293e | 937 | if (!(pte_flags(*pte) & _PAGE_PRESENT)) |
5b727a3b JF |
938 | return 0; |
939 | ||
3c3e5694 SR |
940 | ret = spurious_fault_check(error_code, pte); |
941 | if (!ret) | |
942 | return 0; | |
943 | ||
944 | /* | |
2d4a7167 IM |
945 | * Make sure we have permissions in PMD. |
946 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
947 | */ |
948 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
949 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 950 | |
3c3e5694 | 951 | return ret; |
5b727a3b JF |
952 | } |
953 | ||
abd4f750 | 954 | int show_unhandled_signals = 1; |
1da177e4 | 955 | |
2d4a7167 | 956 | static inline int |
68da336a | 957 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 958 | { |
68da336a | 959 | if (error_code & PF_WRITE) { |
2d4a7167 | 960 | /* write, present and write, not present: */ |
92181f19 NP |
961 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
962 | return 1; | |
2d4a7167 | 963 | return 0; |
92181f19 NP |
964 | } |
965 | ||
2d4a7167 IM |
966 | /* read, present: */ |
967 | if (unlikely(error_code & PF_PROT)) | |
968 | return 1; | |
969 | ||
970 | /* read, not present: */ | |
971 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
972 | return 1; | |
973 | ||
92181f19 NP |
974 | return 0; |
975 | } | |
976 | ||
0973a06c HS |
977 | static int fault_in_kernel_space(unsigned long address) |
978 | { | |
d9517346 | 979 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
980 | } |
981 | ||
1da177e4 LT |
982 | /* |
983 | * This routine handles page faults. It determines the address, | |
984 | * and the problem, and then passes it off to one of the appropriate | |
985 | * routines. | |
1da177e4 | 986 | */ |
c3731c68 IM |
987 | dotraplinkage void __kprobes |
988 | do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 | 989 | { |
2d4a7167 | 990 | struct vm_area_struct *vma; |
1da177e4 | 991 | struct task_struct *tsk; |
2d4a7167 | 992 | unsigned long address; |
1da177e4 | 993 | struct mm_struct *mm; |
f8c2ee22 | 994 | int fault; |
d065bd81 | 995 | int write = error_code & PF_WRITE; |
37b23e05 | 996 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE | |
d065bd81 | 997 | (write ? FAULT_FLAG_WRITE : 0); |
1da177e4 | 998 | |
a9ba9a3b AV |
999 | tsk = current; |
1000 | mm = tsk->mm; | |
2d4a7167 | 1001 | |
2d4a7167 | 1002 | /* Get the faulting address: */ |
f51c9452 | 1003 | address = read_cr2(); |
1da177e4 | 1004 | |
f8561296 VN |
1005 | /* |
1006 | * Detect and handle instructions that would cause a page fault for | |
1007 | * both a tracked kernel page and a userspace page. | |
1008 | */ | |
1009 | if (kmemcheck_active(regs)) | |
1010 | kmemcheck_hide(regs); | |
5dfaf90f | 1011 | prefetchw(&mm->mmap_sem); |
f8561296 | 1012 | |
0fd0e3da | 1013 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1014 | return; |
1da177e4 LT |
1015 | |
1016 | /* | |
1017 | * We fault-in kernel-space virtual memory on-demand. The | |
1018 | * 'reference' page table is init_mm.pgd. | |
1019 | * | |
1020 | * NOTE! We MUST NOT take any locks for this case. We may | |
1021 | * be in an interrupt or a critical region, and should | |
1022 | * only copy the information from the master page table, | |
1023 | * nothing more. | |
1024 | * | |
1025 | * This verifies that the fault happens in kernel space | |
1026 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1027 | * protection error (error_code & 9) == 0. |
1da177e4 | 1028 | */ |
0973a06c | 1029 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1030 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1031 | if (vmalloc_fault(address) >= 0) | |
1032 | return; | |
1033 | ||
1034 | if (kmemcheck_fault(regs, address, error_code)) | |
1035 | return; | |
1036 | } | |
5b727a3b | 1037 | |
2d4a7167 | 1038 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1039 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1040 | return; |
1041 | ||
2d4a7167 | 1042 | /* kprobes don't want to hook the spurious faults: */ |
9be260a6 MH |
1043 | if (notify_page_fault(regs)) |
1044 | return; | |
f8c2ee22 HH |
1045 | /* |
1046 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1047 | * fault we could otherwise deadlock: |
f8c2ee22 | 1048 | */ |
92181f19 | 1049 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1050 | |
92181f19 | 1051 | return; |
f8c2ee22 HH |
1052 | } |
1053 | ||
2d4a7167 | 1054 | /* kprobes don't want to hook the spurious faults: */ |
f8a6b2b9 | 1055 | if (unlikely(notify_page_fault(regs))) |
9be260a6 | 1056 | return; |
f8c2ee22 | 1057 | /* |
891cffbd LT |
1058 | * It's safe to allow irq's after cr2 has been saved and the |
1059 | * vmalloc fault has been handled. | |
1060 | * | |
1061 | * User-mode registers count as a user access even for any | |
2d4a7167 | 1062 | * potential system fault or CPU buglet: |
f8c2ee22 | 1063 | */ |
891cffbd LT |
1064 | if (user_mode_vm(regs)) { |
1065 | local_irq_enable(); | |
1066 | error_code |= PF_USER; | |
2d4a7167 IM |
1067 | } else { |
1068 | if (regs->flags & X86_EFLAGS_IF) | |
1069 | local_irq_enable(); | |
1070 | } | |
8c914cb7 | 1071 | |
66c58156 | 1072 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1073 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1074 | |
a8b0ca17 | 1075 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
7dd1fcc2 | 1076 | |
1da177e4 | 1077 | /* |
2d4a7167 IM |
1078 | * If we're in an interrupt, have no user context or are running |
1079 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1080 | */ |
92181f19 NP |
1081 | if (unlikely(in_atomic() || !mm)) { |
1082 | bad_area_nosemaphore(regs, error_code, address); | |
1083 | return; | |
1084 | } | |
1da177e4 | 1085 | |
3a1dfe6e IM |
1086 | /* |
1087 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1088 | * addresses in user space. All other faults represent errors in |
1089 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1090 | * case of an erroneous fault occurring in a code path which already | |
1091 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1092 | * against the address space. Luckily the kernel only validly | |
1093 | * references user space from well defined areas of code, which are | |
1094 | * listed in the exceptions table. | |
1da177e4 LT |
1095 | * |
1096 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1097 | * the source reference check when there is a possibility of a |
1098 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1099 | * validate the source. If this is invalid we can skip the address | |
1100 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1101 | */ |
92181f19 | 1102 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1103 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1104 | !search_exception_tables(regs->ip)) { |
1105 | bad_area_nosemaphore(regs, error_code, address); | |
1106 | return; | |
1107 | } | |
d065bd81 | 1108 | retry: |
1da177e4 | 1109 | down_read(&mm->mmap_sem); |
01006074 PZ |
1110 | } else { |
1111 | /* | |
2d4a7167 IM |
1112 | * The above down_read_trylock() might have succeeded in |
1113 | * which case we'll have missed the might_sleep() from | |
1114 | * down_read(): | |
01006074 PZ |
1115 | */ |
1116 | might_sleep(); | |
1da177e4 LT |
1117 | } |
1118 | ||
1119 | vma = find_vma(mm, address); | |
92181f19 NP |
1120 | if (unlikely(!vma)) { |
1121 | bad_area(regs, error_code, address); | |
1122 | return; | |
1123 | } | |
1124 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1125 | goto good_area; |
92181f19 NP |
1126 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1127 | bad_area(regs, error_code, address); | |
1128 | return; | |
1129 | } | |
33cb5243 | 1130 | if (error_code & PF_USER) { |
6f4d368e HH |
1131 | /* |
1132 | * Accessing the stack below %sp is always a bug. | |
1133 | * The large cushion allows instructions like enter | |
2d4a7167 | 1134 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1135 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1136 | */ |
92181f19 NP |
1137 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1138 | bad_area(regs, error_code, address); | |
1139 | return; | |
1140 | } | |
1da177e4 | 1141 | } |
92181f19 NP |
1142 | if (unlikely(expand_stack(vma, address))) { |
1143 | bad_area(regs, error_code, address); | |
1144 | return; | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * Ok, we have a good vm_area for this memory access, so | |
1149 | * we can handle it.. | |
1150 | */ | |
1da177e4 | 1151 | good_area: |
68da336a | 1152 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1153 | bad_area_access_error(regs, error_code, address); |
1154 | return; | |
1da177e4 LT |
1155 | } |
1156 | ||
1157 | /* | |
1158 | * If for any reason at all we couldn't handle the fault, | |
1159 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1160 | * the fault: |
1da177e4 | 1161 | */ |
d065bd81 | 1162 | fault = handle_mm_fault(mm, vma, address, flags); |
2d4a7167 | 1163 | |
b80ef10e KM |
1164 | if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) { |
1165 | if (mm_fault_error(regs, error_code, address, fault)) | |
1166 | return; | |
37b23e05 KM |
1167 | } |
1168 | ||
d065bd81 ML |
1169 | /* |
1170 | * Major/minor page fault accounting is only done on the | |
1171 | * initial attempt. If we go through a retry, it is extremely | |
1172 | * likely that the page will be found in page cache at that point. | |
1173 | */ | |
1174 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1175 | if (fault & VM_FAULT_MAJOR) { | |
1176 | tsk->maj_flt++; | |
a8b0ca17 | 1177 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
d065bd81 ML |
1178 | regs, address); |
1179 | } else { | |
1180 | tsk->min_flt++; | |
a8b0ca17 | 1181 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
d065bd81 ML |
1182 | regs, address); |
1183 | } | |
1184 | if (fault & VM_FAULT_RETRY) { | |
1185 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
1186 | * of starvation. */ | |
1187 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1188 | goto retry; | |
1189 | } | |
ac17dc8e | 1190 | } |
d729ab35 | 1191 | |
8c938f9f IM |
1192 | check_v8086_mode(regs, address, tsk); |
1193 | ||
1da177e4 | 1194 | up_read(&mm->mmap_sem); |
1da177e4 | 1195 | } |