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 */ | |
9326638c | 11 | #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ |
a2bcd473 | 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_*(), ... */ |
f40c3300 AL |
21 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
22 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
1da177e4 | 23 | |
d34603b0 SA |
24 | #define CREATE_TRACE_POINTS |
25 | #include <asm/trace/exceptions.h> | |
26 | ||
33cb5243 | 27 | /* |
2d4a7167 IM |
28 | * Page fault error code bits: |
29 | * | |
30 | * bit 0 == 0: no page found 1: protection fault | |
31 | * bit 1 == 0: read access 1: write access | |
32 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
33 | * bit 3 == 1: use of reserved bit detected | |
34 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 35 | */ |
2d4a7167 IM |
36 | enum x86_pf_error_code { |
37 | ||
38 | PF_PROT = 1 << 0, | |
39 | PF_WRITE = 1 << 1, | |
40 | PF_USER = 1 << 2, | |
41 | PF_RSVD = 1 << 3, | |
42 | PF_INSTR = 1 << 4, | |
43 | }; | |
66c58156 | 44 | |
b814d41f | 45 | /* |
b319eed0 IM |
46 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
47 | * handled by mmiotrace: | |
b814d41f | 48 | */ |
9326638c | 49 | static nokprobe_inline int |
62c9295f | 50 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 51 | { |
0fd0e3da PP |
52 | if (unlikely(is_kmmio_active())) |
53 | if (kmmio_handler(regs, addr) == 1) | |
54 | return -1; | |
0fd0e3da | 55 | return 0; |
86069782 PP |
56 | } |
57 | ||
9326638c | 58 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 59 | { |
74a0b576 CH |
60 | int ret = 0; |
61 | ||
62 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 63 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
64 | preempt_disable(); |
65 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
66 | ret = 1; | |
67 | preempt_enable(); | |
68 | } | |
1bd858a5 | 69 | |
74a0b576 | 70 | return ret; |
33cb5243 | 71 | } |
1bd858a5 | 72 | |
1dc85be0 | 73 | /* |
2d4a7167 IM |
74 | * Prefetch quirks: |
75 | * | |
76 | * 32-bit mode: | |
77 | * | |
78 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
79 | * Check that here and ignore it. | |
1dc85be0 | 80 | * |
2d4a7167 | 81 | * 64-bit mode: |
1dc85be0 | 82 | * |
2d4a7167 IM |
83 | * Sometimes the CPU reports invalid exceptions on prefetch. |
84 | * Check that here and ignore it. | |
85 | * | |
86 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 87 | */ |
107a0367 IM |
88 | static inline int |
89 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
90 | unsigned char opcode, int *prefetch) | |
91 | { | |
92 | unsigned char instr_hi = opcode & 0xf0; | |
93 | unsigned char instr_lo = opcode & 0x0f; | |
94 | ||
95 | switch (instr_hi) { | |
96 | case 0x20: | |
97 | case 0x30: | |
98 | /* | |
99 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
100 | * In X86_64 long mode, the CPU will signal invalid | |
101 | * opcode if some of these prefixes are present so | |
102 | * X86_64 will never get here anyway | |
103 | */ | |
104 | return ((instr_lo & 7) == 0x6); | |
105 | #ifdef CONFIG_X86_64 | |
106 | case 0x40: | |
107 | /* | |
108 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
109 | * Need to figure out under what instruction mode the | |
110 | * instruction was issued. Could check the LDT for lm, | |
111 | * but for now it's good enough to assume that long | |
112 | * mode only uses well known segments or kernel. | |
113 | */ | |
318f5a2a | 114 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
115 | #endif |
116 | case 0x60: | |
117 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
118 | return (instr_lo & 0xC) == 0x4; | |
119 | case 0xF0: | |
120 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
121 | return !instr_lo || (instr_lo>>1) == 1; | |
122 | case 0x00: | |
123 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
124 | if (probe_kernel_address(instr, opcode)) | |
125 | return 0; | |
126 | ||
127 | *prefetch = (instr_lo == 0xF) && | |
128 | (opcode == 0x0D || opcode == 0x18); | |
129 | return 0; | |
130 | default: | |
131 | return 0; | |
132 | } | |
133 | } | |
134 | ||
2d4a7167 IM |
135 | static int |
136 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 137 | { |
2d4a7167 | 138 | unsigned char *max_instr; |
ab2bf0c1 | 139 | unsigned char *instr; |
33cb5243 | 140 | int prefetch = 0; |
1da177e4 | 141 | |
3085354d IM |
142 | /* |
143 | * If it was a exec (instruction fetch) fault on NX page, then | |
144 | * do not ignore the fault: | |
145 | */ | |
66c58156 | 146 | if (error_code & PF_INSTR) |
1da177e4 | 147 | return 0; |
1dc85be0 | 148 | |
107a0367 | 149 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 150 | max_instr = instr + 15; |
1da177e4 | 151 | |
76381fee | 152 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
153 | return 0; |
154 | ||
107a0367 | 155 | while (instr < max_instr) { |
2d4a7167 | 156 | unsigned char opcode; |
1da177e4 | 157 | |
ab2bf0c1 | 158 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 159 | break; |
1da177e4 | 160 | |
1da177e4 LT |
161 | instr++; |
162 | ||
107a0367 | 163 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 164 | break; |
1da177e4 LT |
165 | } |
166 | return prefetch; | |
167 | } | |
168 | ||
2d4a7167 IM |
169 | static void |
170 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 171 | struct task_struct *tsk, int fault) |
c4aba4a8 | 172 | { |
f672b49b | 173 | unsigned lsb = 0; |
c4aba4a8 HH |
174 | siginfo_t info; |
175 | ||
2d4a7167 IM |
176 | info.si_signo = si_signo; |
177 | info.si_errno = 0; | |
178 | info.si_code = si_code; | |
179 | info.si_addr = (void __user *)address; | |
f672b49b AK |
180 | if (fault & VM_FAULT_HWPOISON_LARGE) |
181 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
182 | if (fault & VM_FAULT_HWPOISON) | |
183 | lsb = PAGE_SHIFT; | |
184 | info.si_addr_lsb = lsb; | |
2d4a7167 | 185 | |
c4aba4a8 HH |
186 | force_sig_info(si_signo, &info, tsk); |
187 | } | |
188 | ||
f2f13a85 IM |
189 | DEFINE_SPINLOCK(pgd_lock); |
190 | LIST_HEAD(pgd_list); | |
191 | ||
192 | #ifdef CONFIG_X86_32 | |
193 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 194 | { |
f2f13a85 IM |
195 | unsigned index = pgd_index(address); |
196 | pgd_t *pgd_k; | |
197 | pud_t *pud, *pud_k; | |
198 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 199 | |
f2f13a85 IM |
200 | pgd += index; |
201 | pgd_k = init_mm.pgd + index; | |
202 | ||
203 | if (!pgd_present(*pgd_k)) | |
204 | return NULL; | |
205 | ||
206 | /* | |
207 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
208 | * and redundant with the set_pmd() on non-PAE. As would | |
209 | * set_pud. | |
210 | */ | |
211 | pud = pud_offset(pgd, address); | |
212 | pud_k = pud_offset(pgd_k, address); | |
213 | if (!pud_present(*pud_k)) | |
214 | return NULL; | |
215 | ||
216 | pmd = pmd_offset(pud, address); | |
217 | pmd_k = pmd_offset(pud_k, address); | |
218 | if (!pmd_present(*pmd_k)) | |
219 | return NULL; | |
220 | ||
b8bcfe99 | 221 | if (!pmd_present(*pmd)) |
f2f13a85 | 222 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 223 | else |
f2f13a85 | 224 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
225 | |
226 | return pmd_k; | |
227 | } | |
228 | ||
229 | void vmalloc_sync_all(void) | |
230 | { | |
231 | unsigned long address; | |
232 | ||
233 | if (SHARED_KERNEL_PMD) | |
234 | return; | |
235 | ||
236 | for (address = VMALLOC_START & PMD_MASK; | |
237 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
238 | address += PMD_SIZE) { | |
f2f13a85 IM |
239 | struct page *page; |
240 | ||
a79e53d8 | 241 | spin_lock(&pgd_lock); |
f2f13a85 | 242 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 243 | spinlock_t *pgt_lock; |
f01f7c56 | 244 | pmd_t *ret; |
617d34d9 | 245 | |
a79e53d8 | 246 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
247 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
248 | ||
249 | spin_lock(pgt_lock); | |
250 | ret = vmalloc_sync_one(page_address(page), address); | |
251 | spin_unlock(pgt_lock); | |
252 | ||
253 | if (!ret) | |
f2f13a85 IM |
254 | break; |
255 | } | |
a79e53d8 | 256 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
257 | } |
258 | } | |
259 | ||
260 | /* | |
261 | * 32-bit: | |
262 | * | |
263 | * Handle a fault on the vmalloc or module mapping area | |
264 | */ | |
9326638c | 265 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
266 | { |
267 | unsigned long pgd_paddr; | |
268 | pmd_t *pmd_k; | |
269 | pte_t *pte_k; | |
270 | ||
271 | /* Make sure we are in vmalloc area: */ | |
272 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
273 | return -1; | |
274 | ||
ebc8827f FW |
275 | WARN_ON_ONCE(in_nmi()); |
276 | ||
f2f13a85 IM |
277 | /* |
278 | * Synchronize this task's top level page-table | |
279 | * with the 'reference' page table. | |
280 | * | |
281 | * Do _not_ use "current" here. We might be inside | |
282 | * an interrupt in the middle of a task switch.. | |
283 | */ | |
284 | pgd_paddr = read_cr3(); | |
285 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
286 | if (!pmd_k) | |
287 | return -1; | |
288 | ||
289 | pte_k = pte_offset_kernel(pmd_k, address); | |
290 | if (!pte_present(*pte_k)) | |
291 | return -1; | |
292 | ||
293 | return 0; | |
294 | } | |
9326638c | 295 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
296 | |
297 | /* | |
298 | * Did it hit the DOS screen memory VA from vm86 mode? | |
299 | */ | |
300 | static inline void | |
301 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
302 | struct task_struct *tsk) | |
303 | { | |
304 | unsigned long bit; | |
305 | ||
306 | if (!v8086_mode(regs)) | |
307 | return; | |
308 | ||
309 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
310 | if (bit < 32) | |
311 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 312 | } |
1da177e4 | 313 | |
087975b0 | 314 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 315 | { |
087975b0 AM |
316 | return pfn < max_low_pfn; |
317 | } | |
1156e098 | 318 | |
087975b0 AM |
319 | static void dump_pagetable(unsigned long address) |
320 | { | |
321 | pgd_t *base = __va(read_cr3()); | |
322 | pgd_t *pgd = &base[pgd_index(address)]; | |
323 | pmd_t *pmd; | |
324 | pte_t *pte; | |
2d4a7167 | 325 | |
1156e098 | 326 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
327 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
328 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
329 | goto out; | |
1156e098 | 330 | #endif |
087975b0 AM |
331 | pmd = pmd_offset(pud_offset(pgd, address), address); |
332 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
333 | |
334 | /* | |
335 | * We must not directly access the pte in the highpte | |
336 | * case if the page table is located in highmem. | |
337 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 338 | * it's allocated already: |
1156e098 | 339 | */ |
087975b0 AM |
340 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
341 | goto out; | |
1156e098 | 342 | |
087975b0 AM |
343 | pte = pte_offset_kernel(pmd, address); |
344 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
345 | out: | |
1156e098 | 346 | printk("\n"); |
f2f13a85 IM |
347 | } |
348 | ||
349 | #else /* CONFIG_X86_64: */ | |
350 | ||
351 | void vmalloc_sync_all(void) | |
352 | { | |
6afb5157 | 353 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
354 | } |
355 | ||
356 | /* | |
357 | * 64-bit: | |
358 | * | |
359 | * Handle a fault on the vmalloc area | |
360 | * | |
361 | * This assumes no large pages in there. | |
362 | */ | |
9326638c | 363 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
364 | { |
365 | pgd_t *pgd, *pgd_ref; | |
366 | pud_t *pud, *pud_ref; | |
367 | pmd_t *pmd, *pmd_ref; | |
368 | pte_t *pte, *pte_ref; | |
369 | ||
370 | /* Make sure we are in vmalloc area: */ | |
371 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
372 | return -1; | |
373 | ||
ebc8827f FW |
374 | WARN_ON_ONCE(in_nmi()); |
375 | ||
f2f13a85 IM |
376 | /* |
377 | * Copy kernel mappings over when needed. This can also | |
378 | * happen within a race in page table update. In the later | |
379 | * case just flush: | |
380 | */ | |
381 | pgd = pgd_offset(current->active_mm, address); | |
382 | pgd_ref = pgd_offset_k(address); | |
383 | if (pgd_none(*pgd_ref)) | |
384 | return -1; | |
385 | ||
1160c277 | 386 | if (pgd_none(*pgd)) { |
f2f13a85 | 387 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
388 | arch_flush_lazy_mmu_mode(); |
389 | } else { | |
f2f13a85 | 390 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 391 | } |
f2f13a85 IM |
392 | |
393 | /* | |
394 | * Below here mismatches are bugs because these lower tables | |
395 | * are shared: | |
396 | */ | |
397 | ||
398 | pud = pud_offset(pgd, address); | |
399 | pud_ref = pud_offset(pgd_ref, address); | |
400 | if (pud_none(*pud_ref)) | |
401 | return -1; | |
402 | ||
403 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
404 | BUG(); | |
405 | ||
406 | pmd = pmd_offset(pud, address); | |
407 | pmd_ref = pmd_offset(pud_ref, address); | |
408 | if (pmd_none(*pmd_ref)) | |
409 | return -1; | |
410 | ||
411 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
412 | BUG(); | |
413 | ||
414 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
415 | if (!pte_present(*pte_ref)) | |
416 | return -1; | |
417 | ||
418 | pte = pte_offset_kernel(pmd, address); | |
419 | ||
420 | /* | |
421 | * Don't use pte_page here, because the mappings can point | |
422 | * outside mem_map, and the NUMA hash lookup cannot handle | |
423 | * that: | |
424 | */ | |
425 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
426 | BUG(); | |
427 | ||
428 | return 0; | |
429 | } | |
9326638c | 430 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 431 | |
e05139f2 | 432 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 433 | static const char errata93_warning[] = |
ad361c98 JP |
434 | KERN_ERR |
435 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
436 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
437 | "******* Please consider a BIOS update.\n" | |
438 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 439 | #endif |
f2f13a85 IM |
440 | |
441 | /* | |
442 | * No vm86 mode in 64-bit mode: | |
443 | */ | |
444 | static inline void | |
445 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
446 | struct task_struct *tsk) | |
447 | { | |
448 | } | |
449 | ||
450 | static int bad_address(void *p) | |
451 | { | |
452 | unsigned long dummy; | |
453 | ||
454 | return probe_kernel_address((unsigned long *)p, dummy); | |
455 | } | |
456 | ||
457 | static void dump_pagetable(unsigned long address) | |
458 | { | |
087975b0 AM |
459 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
460 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
461 | pud_t *pud; |
462 | pmd_t *pmd; | |
463 | pte_t *pte; | |
464 | ||
2d4a7167 IM |
465 | if (bad_address(pgd)) |
466 | goto bad; | |
467 | ||
d646bce4 | 468 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
469 | |
470 | if (!pgd_present(*pgd)) | |
471 | goto out; | |
1da177e4 | 472 | |
d2ae5b5f | 473 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
474 | if (bad_address(pud)) |
475 | goto bad; | |
476 | ||
1da177e4 | 477 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 478 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 479 | goto out; |
1da177e4 LT |
480 | |
481 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
482 | if (bad_address(pmd)) |
483 | goto bad; | |
484 | ||
1da177e4 | 485 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
486 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
487 | goto out; | |
1da177e4 LT |
488 | |
489 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
490 | if (bad_address(pte)) |
491 | goto bad; | |
492 | ||
33cb5243 | 493 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 494 | out: |
1da177e4 LT |
495 | printk("\n"); |
496 | return; | |
497 | bad: | |
498 | printk("BAD\n"); | |
8c938f9f IM |
499 | } |
500 | ||
f2f13a85 | 501 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 502 | |
2d4a7167 IM |
503 | /* |
504 | * Workaround for K8 erratum #93 & buggy BIOS. | |
505 | * | |
506 | * BIOS SMM functions are required to use a specific workaround | |
507 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
508 | * | |
509 | * A lot of BIOS that didn't get tested properly miss this. | |
510 | * | |
511 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
512 | * Try to work around it here. | |
513 | * | |
514 | * Note we only handle faults in kernel here. | |
515 | * Does nothing on 32-bit. | |
fdfe8aa8 | 516 | */ |
33cb5243 | 517 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 518 | { |
e05139f2 JB |
519 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
520 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
521 | || boot_cpu_data.x86 != 0xf) | |
522 | return 0; | |
523 | ||
65ea5b03 | 524 | if (address != regs->ip) |
1da177e4 | 525 | return 0; |
2d4a7167 | 526 | |
33cb5243 | 527 | if ((address >> 32) != 0) |
1da177e4 | 528 | return 0; |
2d4a7167 | 529 | |
1da177e4 | 530 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
531 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
532 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 533 | printk_once(errata93_warning); |
65ea5b03 | 534 | regs->ip = address; |
1da177e4 LT |
535 | return 1; |
536 | } | |
fdfe8aa8 | 537 | #endif |
1da177e4 | 538 | return 0; |
33cb5243 | 539 | } |
1da177e4 | 540 | |
35f3266f | 541 | /* |
2d4a7167 IM |
542 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
543 | * to illegal addresses >4GB. | |
544 | * | |
545 | * We catch this in the page fault handler because these addresses | |
546 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
547 | * segment in LDT is compatibility mode. |
548 | */ | |
549 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
550 | { | |
551 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 552 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
553 | return 1; |
554 | #endif | |
555 | return 0; | |
556 | } | |
557 | ||
29caf2f9 HH |
558 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
559 | { | |
560 | #ifdef CONFIG_X86_F00F_BUG | |
561 | unsigned long nr; | |
2d4a7167 | 562 | |
29caf2f9 | 563 | /* |
2d4a7167 | 564 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 565 | */ |
e2604b49 | 566 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
567 | nr = (address - idt_descr.address) >> 3; |
568 | ||
569 | if (nr == 6) { | |
570 | do_invalid_op(regs, 0); | |
571 | return 1; | |
572 | } | |
573 | } | |
574 | #endif | |
575 | return 0; | |
576 | } | |
577 | ||
8f766149 IM |
578 | static const char nx_warning[] = KERN_CRIT |
579 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
580 | static const char smep_warning[] = KERN_CRIT |
581 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 582 | |
2d4a7167 IM |
583 | static void |
584 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
585 | unsigned long address) | |
b3279c7f | 586 | { |
1156e098 HH |
587 | if (!oops_may_print()) |
588 | return; | |
589 | ||
1156e098 | 590 | if (error_code & PF_INSTR) { |
93809be8 | 591 | unsigned int level; |
426e34cc MF |
592 | pgd_t *pgd; |
593 | pte_t *pte; | |
2d4a7167 | 594 | |
426e34cc MF |
595 | pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
596 | pgd += pgd_index(address); | |
597 | ||
598 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 599 | |
8f766149 | 600 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 601 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
602 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
603 | (pgd_flags(*pgd) & _PAGE_USER) && | |
604 | (read_cr4() & X86_CR4_SMEP)) | |
605 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); | |
1156e098 | 606 | } |
1156e098 | 607 | |
19f0dda9 | 608 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 609 | if (address < PAGE_SIZE) |
19f0dda9 | 610 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 611 | else |
19f0dda9 | 612 | printk(KERN_CONT "paging request"); |
2d4a7167 | 613 | |
f294a8ce | 614 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 615 | printk(KERN_ALERT "IP:"); |
5f01c988 | 616 | printk_address(regs->ip); |
2d4a7167 | 617 | |
b3279c7f HH |
618 | dump_pagetable(address); |
619 | } | |
620 | ||
2d4a7167 IM |
621 | static noinline void |
622 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
623 | unsigned long address) | |
1da177e4 | 624 | { |
2d4a7167 IM |
625 | struct task_struct *tsk; |
626 | unsigned long flags; | |
627 | int sig; | |
628 | ||
629 | flags = oops_begin(); | |
630 | tsk = current; | |
631 | sig = SIGKILL; | |
1209140c | 632 | |
1da177e4 | 633 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 634 | tsk->comm, address); |
1da177e4 | 635 | dump_pagetable(address); |
2d4a7167 IM |
636 | |
637 | tsk->thread.cr2 = address; | |
51e7dc70 | 638 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
639 | tsk->thread.error_code = error_code; |
640 | ||
22f5991c | 641 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 642 | sig = 0; |
2d4a7167 | 643 | |
874d93d1 | 644 | oops_end(flags, regs, sig); |
1da177e4 LT |
645 | } |
646 | ||
2d4a7167 IM |
647 | static noinline void |
648 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 649 | unsigned long address, int signal, int si_code) |
92181f19 NP |
650 | { |
651 | struct task_struct *tsk = current; | |
19803078 | 652 | unsigned long *stackend; |
92181f19 NP |
653 | unsigned long flags; |
654 | int sig; | |
92181f19 | 655 | |
2d4a7167 | 656 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 657 | if (fixup_exception(regs)) { |
c026b359 PZ |
658 | /* |
659 | * Any interrupt that takes a fault gets the fixup. This makes | |
660 | * the below recursive fault logic only apply to a faults from | |
661 | * task context. | |
662 | */ | |
663 | if (in_interrupt()) | |
664 | return; | |
665 | ||
666 | /* | |
667 | * Per the above we're !in_interrupt(), aka. task context. | |
668 | * | |
669 | * In this case we need to make sure we're not recursively | |
670 | * faulting through the emulate_vsyscall() logic. | |
671 | */ | |
4fc34901 | 672 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 673 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
674 | tsk->thread.error_code = error_code | PF_USER; |
675 | tsk->thread.cr2 = address; | |
676 | ||
677 | /* XXX: hwpoison faults will set the wrong code. */ | |
678 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
679 | } | |
c026b359 PZ |
680 | |
681 | /* | |
682 | * Barring that, we can do the fixup and be happy. | |
683 | */ | |
92181f19 | 684 | return; |
4fc34901 | 685 | } |
92181f19 NP |
686 | |
687 | /* | |
2d4a7167 IM |
688 | * 32-bit: |
689 | * | |
690 | * Valid to do another page fault here, because if this fault | |
691 | * had been triggered by is_prefetch fixup_exception would have | |
692 | * handled it. | |
693 | * | |
694 | * 64-bit: | |
92181f19 | 695 | * |
2d4a7167 | 696 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
697 | */ |
698 | if (is_prefetch(regs, error_code, address)) | |
699 | return; | |
700 | ||
701 | if (is_errata93(regs, address)) | |
702 | return; | |
703 | ||
704 | /* | |
705 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 706 | * terminate things with extreme prejudice: |
92181f19 | 707 | */ |
92181f19 | 708 | flags = oops_begin(); |
92181f19 NP |
709 | |
710 | show_fault_oops(regs, error_code, address); | |
711 | ||
2d4a7167 | 712 | stackend = end_of_stack(tsk); |
0e7810be | 713 | if (tsk != &init_task && *stackend != STACK_END_MAGIC) |
b0f4c4b3 | 714 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 715 | |
1cc99544 | 716 | tsk->thread.cr2 = address; |
51e7dc70 | 717 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 718 | tsk->thread.error_code = error_code; |
92181f19 | 719 | |
92181f19 NP |
720 | sig = SIGKILL; |
721 | if (__die("Oops", regs, error_code)) | |
722 | sig = 0; | |
2d4a7167 | 723 | |
92181f19 | 724 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 725 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 726 | |
92181f19 | 727 | oops_end(flags, regs, sig); |
92181f19 NP |
728 | } |
729 | ||
2d4a7167 IM |
730 | /* |
731 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
732 | * sysctl is set: | |
733 | */ | |
734 | static inline void | |
735 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
736 | unsigned long address, struct task_struct *tsk) | |
737 | { | |
738 | if (!unhandled_signal(tsk, SIGSEGV)) | |
739 | return; | |
740 | ||
741 | if (!printk_ratelimit()) | |
742 | return; | |
743 | ||
a1a08d1c | 744 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
745 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
746 | tsk->comm, task_pid_nr(tsk), address, | |
747 | (void *)regs->ip, (void *)regs->sp, error_code); | |
748 | ||
749 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
750 | ||
751 | printk(KERN_CONT "\n"); | |
752 | } | |
753 | ||
754 | static void | |
755 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
756 | unsigned long address, int si_code) | |
92181f19 NP |
757 | { |
758 | struct task_struct *tsk = current; | |
759 | ||
760 | /* User mode accesses just cause a SIGSEGV */ | |
761 | if (error_code & PF_USER) { | |
762 | /* | |
2d4a7167 | 763 | * It's possible to have interrupts off here: |
92181f19 NP |
764 | */ |
765 | local_irq_enable(); | |
766 | ||
767 | /* | |
768 | * Valid to do another page fault here because this one came | |
2d4a7167 | 769 | * from user space: |
92181f19 NP |
770 | */ |
771 | if (is_prefetch(regs, error_code, address)) | |
772 | return; | |
773 | ||
774 | if (is_errata100(regs, address)) | |
775 | return; | |
776 | ||
3ae36655 AL |
777 | #ifdef CONFIG_X86_64 |
778 | /* | |
779 | * Instruction fetch faults in the vsyscall page might need | |
780 | * emulation. | |
781 | */ | |
782 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 783 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
784 | if (emulate_vsyscall(regs, address)) |
785 | return; | |
786 | } | |
787 | #endif | |
e575a86f KC |
788 | /* Kernel addresses are always protection faults: */ |
789 | if (address >= TASK_SIZE) | |
790 | error_code |= PF_PROT; | |
3ae36655 | 791 | |
e575a86f | 792 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
793 | show_signal_msg(regs, error_code, address, tsk); |
794 | ||
2d4a7167 | 795 | tsk->thread.cr2 = address; |
e575a86f | 796 | tsk->thread.error_code = error_code; |
51e7dc70 | 797 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 798 | |
f672b49b | 799 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 800 | |
92181f19 NP |
801 | return; |
802 | } | |
803 | ||
804 | if (is_f00f_bug(regs, address)) | |
805 | return; | |
806 | ||
4fc34901 | 807 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
808 | } |
809 | ||
2d4a7167 IM |
810 | static noinline void |
811 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
812 | unsigned long address) | |
92181f19 NP |
813 | { |
814 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
815 | } | |
816 | ||
2d4a7167 IM |
817 | static void |
818 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
819 | unsigned long address, int si_code) | |
92181f19 NP |
820 | { |
821 | struct mm_struct *mm = current->mm; | |
822 | ||
823 | /* | |
824 | * Something tried to access memory that isn't in our memory map.. | |
825 | * Fix it, but check if it's kernel or user first.. | |
826 | */ | |
827 | up_read(&mm->mmap_sem); | |
828 | ||
829 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
830 | } | |
831 | ||
2d4a7167 IM |
832 | static noinline void |
833 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
834 | { |
835 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
836 | } | |
837 | ||
2d4a7167 IM |
838 | static noinline void |
839 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
840 | unsigned long address) | |
92181f19 NP |
841 | { |
842 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
843 | } | |
844 | ||
2d4a7167 | 845 | static void |
a6e04aa9 AK |
846 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
847 | unsigned int fault) | |
92181f19 NP |
848 | { |
849 | struct task_struct *tsk = current; | |
850 | struct mm_struct *mm = tsk->mm; | |
a6e04aa9 | 851 | int code = BUS_ADRERR; |
92181f19 NP |
852 | |
853 | up_read(&mm->mmap_sem); | |
854 | ||
2d4a7167 | 855 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 856 | if (!(error_code & PF_USER)) { |
4fc34901 | 857 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
858 | return; |
859 | } | |
2d4a7167 | 860 | |
cd1b68f0 | 861 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
862 | if (is_prefetch(regs, error_code, address)) |
863 | return; | |
2d4a7167 IM |
864 | |
865 | tsk->thread.cr2 = address; | |
866 | tsk->thread.error_code = error_code; | |
51e7dc70 | 867 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 868 | |
a6e04aa9 | 869 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 870 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
871 | printk(KERN_ERR |
872 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
873 | tsk->comm, tsk->pid, address); | |
874 | code = BUS_MCEERR_AR; | |
875 | } | |
876 | #endif | |
f672b49b | 877 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
878 | } |
879 | ||
3a13c4d7 | 880 | static noinline void |
2d4a7167 IM |
881 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
882 | unsigned long address, unsigned int fault) | |
92181f19 | 883 | { |
3a13c4d7 JW |
884 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
885 | up_read(¤t->mm->mmap_sem); | |
886 | no_context(regs, error_code, address, 0, 0); | |
887 | return; | |
b80ef10e | 888 | } |
b80ef10e | 889 | |
2d4a7167 | 890 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
891 | /* Kernel mode? Handle exceptions or die: */ |
892 | if (!(error_code & PF_USER)) { | |
893 | up_read(¤t->mm->mmap_sem); | |
4fc34901 AL |
894 | no_context(regs, error_code, address, |
895 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 896 | return; |
f8626854 AV |
897 | } |
898 | ||
c2d23f91 DR |
899 | up_read(¤t->mm->mmap_sem); |
900 | ||
901 | /* | |
902 | * We ran out of memory, call the OOM killer, and return the | |
903 | * userspace (which will retry the fault, or kill us if we got | |
904 | * oom-killed): | |
905 | */ | |
906 | pagefault_out_of_memory(); | |
2d4a7167 | 907 | } else { |
f672b49b AK |
908 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
909 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 910 | do_sigbus(regs, error_code, address, fault); |
2d4a7167 IM |
911 | else |
912 | BUG(); | |
913 | } | |
92181f19 NP |
914 | } |
915 | ||
d8b57bb7 TG |
916 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
917 | { | |
918 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
919 | return 0; | |
2d4a7167 | 920 | |
d8b57bb7 TG |
921 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
922 | return 0; | |
923 | ||
924 | return 1; | |
925 | } | |
926 | ||
5b727a3b | 927 | /* |
2d4a7167 IM |
928 | * Handle a spurious fault caused by a stale TLB entry. |
929 | * | |
930 | * This allows us to lazily refresh the TLB when increasing the | |
931 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
932 | * eagerly is very expensive since that implies doing a full | |
933 | * cross-processor TLB flush, even if no stale TLB entries exist | |
934 | * on other processors. | |
935 | * | |
5b727a3b JF |
936 | * There are no security implications to leaving a stale TLB when |
937 | * increasing the permissions on a page. | |
938 | */ | |
9326638c | 939 | static noinline int |
2d4a7167 | 940 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
941 | { |
942 | pgd_t *pgd; | |
943 | pud_t *pud; | |
944 | pmd_t *pmd; | |
945 | pte_t *pte; | |
3c3e5694 | 946 | int ret; |
5b727a3b JF |
947 | |
948 | /* Reserved-bit violation or user access to kernel space? */ | |
949 | if (error_code & (PF_USER | PF_RSVD)) | |
950 | return 0; | |
951 | ||
952 | pgd = init_mm.pgd + pgd_index(address); | |
953 | if (!pgd_present(*pgd)) | |
954 | return 0; | |
955 | ||
956 | pud = pud_offset(pgd, address); | |
957 | if (!pud_present(*pud)) | |
958 | return 0; | |
959 | ||
d8b57bb7 TG |
960 | if (pud_large(*pud)) |
961 | return spurious_fault_check(error_code, (pte_t *) pud); | |
962 | ||
5b727a3b JF |
963 | pmd = pmd_offset(pud, address); |
964 | if (!pmd_present(*pmd)) | |
965 | return 0; | |
966 | ||
d8b57bb7 TG |
967 | if (pmd_large(*pmd)) |
968 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
969 | ||
5b727a3b | 970 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 971 | if (!pte_present(*pte)) |
5b727a3b JF |
972 | return 0; |
973 | ||
3c3e5694 SR |
974 | ret = spurious_fault_check(error_code, pte); |
975 | if (!ret) | |
976 | return 0; | |
977 | ||
978 | /* | |
2d4a7167 IM |
979 | * Make sure we have permissions in PMD. |
980 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
981 | */ |
982 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
983 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 984 | |
3c3e5694 | 985 | return ret; |
5b727a3b | 986 | } |
9326638c | 987 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 988 | |
abd4f750 | 989 | int show_unhandled_signals = 1; |
1da177e4 | 990 | |
2d4a7167 | 991 | static inline int |
68da336a | 992 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 993 | { |
68da336a | 994 | if (error_code & PF_WRITE) { |
2d4a7167 | 995 | /* write, present and write, not present: */ |
92181f19 NP |
996 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
997 | return 1; | |
2d4a7167 | 998 | return 0; |
92181f19 NP |
999 | } |
1000 | ||
2d4a7167 IM |
1001 | /* read, present: */ |
1002 | if (unlikely(error_code & PF_PROT)) | |
1003 | return 1; | |
1004 | ||
1005 | /* read, not present: */ | |
1006 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1007 | return 1; | |
1008 | ||
92181f19 NP |
1009 | return 0; |
1010 | } | |
1011 | ||
0973a06c HS |
1012 | static int fault_in_kernel_space(unsigned long address) |
1013 | { | |
d9517346 | 1014 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1015 | } |
1016 | ||
40d3cd66 PA |
1017 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1018 | { | |
4640c7ee PA |
1019 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1020 | return false; | |
1021 | ||
1022 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1023 | return false; | |
1024 | ||
40d3cd66 PA |
1025 | if (error_code & PF_USER) |
1026 | return false; | |
1027 | ||
1028 | if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) | |
1029 | return false; | |
1030 | ||
1031 | return true; | |
1032 | } | |
1033 | ||
1da177e4 LT |
1034 | /* |
1035 | * This routine handles page faults. It determines the address, | |
1036 | * and the problem, and then passes it off to one of the appropriate | |
1037 | * routines. | |
d4078e23 PZ |
1038 | * |
1039 | * This function must have noinline because both callers | |
1040 | * {,trace_}do_page_fault() have notrace on. Having this an actual function | |
1041 | * guarantees there's a function trace entry. | |
1da177e4 | 1042 | */ |
9326638c | 1043 | static noinline void |
0ac09f9f JO |
1044 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1045 | unsigned long address) | |
1da177e4 | 1046 | { |
2d4a7167 | 1047 | struct vm_area_struct *vma; |
1da177e4 LT |
1048 | struct task_struct *tsk; |
1049 | struct mm_struct *mm; | |
f8c2ee22 | 1050 | int fault; |
759496ba | 1051 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1052 | |
a9ba9a3b AV |
1053 | tsk = current; |
1054 | mm = tsk->mm; | |
2d4a7167 | 1055 | |
f8561296 VN |
1056 | /* |
1057 | * Detect and handle instructions that would cause a page fault for | |
1058 | * both a tracked kernel page and a userspace page. | |
1059 | */ | |
1060 | if (kmemcheck_active(regs)) | |
1061 | kmemcheck_hide(regs); | |
5dfaf90f | 1062 | prefetchw(&mm->mmap_sem); |
f8561296 | 1063 | |
0fd0e3da | 1064 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1065 | return; |
1da177e4 LT |
1066 | |
1067 | /* | |
1068 | * We fault-in kernel-space virtual memory on-demand. The | |
1069 | * 'reference' page table is init_mm.pgd. | |
1070 | * | |
1071 | * NOTE! We MUST NOT take any locks for this case. We may | |
1072 | * be in an interrupt or a critical region, and should | |
1073 | * only copy the information from the master page table, | |
1074 | * nothing more. | |
1075 | * | |
1076 | * This verifies that the fault happens in kernel space | |
1077 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1078 | * protection error (error_code & 9) == 0. |
1da177e4 | 1079 | */ |
0973a06c | 1080 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1081 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1082 | if (vmalloc_fault(address) >= 0) | |
1083 | return; | |
1084 | ||
1085 | if (kmemcheck_fault(regs, address, error_code)) | |
1086 | return; | |
1087 | } | |
5b727a3b | 1088 | |
2d4a7167 | 1089 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1090 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1091 | return; |
1092 | ||
2d4a7167 | 1093 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1094 | if (kprobes_fault(regs)) |
9be260a6 | 1095 | return; |
f8c2ee22 HH |
1096 | /* |
1097 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1098 | * fault we could otherwise deadlock: |
f8c2ee22 | 1099 | */ |
92181f19 | 1100 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1101 | |
92181f19 | 1102 | return; |
f8c2ee22 HH |
1103 | } |
1104 | ||
2d4a7167 | 1105 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1106 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1107 | return; |
8c914cb7 | 1108 | |
66c58156 | 1109 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1110 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1111 | |
4640c7ee PA |
1112 | if (unlikely(smap_violation(error_code, regs))) { |
1113 | bad_area_nosemaphore(regs, error_code, address); | |
1114 | return; | |
40d3cd66 PA |
1115 | } |
1116 | ||
1da177e4 | 1117 | /* |
2d4a7167 IM |
1118 | * If we're in an interrupt, have no user context or are running |
1119 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1120 | */ |
92181f19 NP |
1121 | if (unlikely(in_atomic() || !mm)) { |
1122 | bad_area_nosemaphore(regs, error_code, address); | |
1123 | return; | |
1124 | } | |
1da177e4 | 1125 | |
e00b12e6 PZ |
1126 | /* |
1127 | * It's safe to allow irq's after cr2 has been saved and the | |
1128 | * vmalloc fault has been handled. | |
1129 | * | |
1130 | * User-mode registers count as a user access even for any | |
1131 | * potential system fault or CPU buglet: | |
1132 | */ | |
1133 | if (user_mode_vm(regs)) { | |
1134 | local_irq_enable(); | |
1135 | error_code |= PF_USER; | |
1136 | flags |= FAULT_FLAG_USER; | |
1137 | } else { | |
1138 | if (regs->flags & X86_EFLAGS_IF) | |
1139 | local_irq_enable(); | |
1140 | } | |
1141 | ||
1142 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1143 | ||
759496ba JW |
1144 | if (error_code & PF_WRITE) |
1145 | flags |= FAULT_FLAG_WRITE; | |
1146 | ||
3a1dfe6e IM |
1147 | /* |
1148 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1149 | * addresses in user space. All other faults represent errors in |
1150 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1151 | * case of an erroneous fault occurring in a code path which already | |
1152 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1153 | * against the address space. Luckily the kernel only validly | |
1154 | * references user space from well defined areas of code, which are | |
1155 | * listed in the exceptions table. | |
1da177e4 LT |
1156 | * |
1157 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1158 | * the source reference check when there is a possibility of a |
1159 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1160 | * validate the source. If this is invalid we can skip the address | |
1161 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1162 | */ |
92181f19 | 1163 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1164 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1165 | !search_exception_tables(regs->ip)) { |
1166 | bad_area_nosemaphore(regs, error_code, address); | |
1167 | return; | |
1168 | } | |
d065bd81 | 1169 | retry: |
1da177e4 | 1170 | down_read(&mm->mmap_sem); |
01006074 PZ |
1171 | } else { |
1172 | /* | |
2d4a7167 IM |
1173 | * The above down_read_trylock() might have succeeded in |
1174 | * which case we'll have missed the might_sleep() from | |
1175 | * down_read(): | |
01006074 PZ |
1176 | */ |
1177 | might_sleep(); | |
1da177e4 LT |
1178 | } |
1179 | ||
1180 | vma = find_vma(mm, address); | |
92181f19 NP |
1181 | if (unlikely(!vma)) { |
1182 | bad_area(regs, error_code, address); | |
1183 | return; | |
1184 | } | |
1185 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1186 | goto good_area; |
92181f19 NP |
1187 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1188 | bad_area(regs, error_code, address); | |
1189 | return; | |
1190 | } | |
33cb5243 | 1191 | if (error_code & PF_USER) { |
6f4d368e HH |
1192 | /* |
1193 | * Accessing the stack below %sp is always a bug. | |
1194 | * The large cushion allows instructions like enter | |
2d4a7167 | 1195 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1196 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1197 | */ |
92181f19 NP |
1198 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1199 | bad_area(regs, error_code, address); | |
1200 | return; | |
1201 | } | |
1da177e4 | 1202 | } |
92181f19 NP |
1203 | if (unlikely(expand_stack(vma, address))) { |
1204 | bad_area(regs, error_code, address); | |
1205 | return; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Ok, we have a good vm_area for this memory access, so | |
1210 | * we can handle it.. | |
1211 | */ | |
1da177e4 | 1212 | good_area: |
68da336a | 1213 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1214 | bad_area_access_error(regs, error_code, address); |
1215 | return; | |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * If for any reason at all we couldn't handle the fault, | |
1220 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1221 | * the fault: |
1da177e4 | 1222 | */ |
d065bd81 | 1223 | fault = handle_mm_fault(mm, vma, address, flags); |
2d4a7167 | 1224 | |
3a13c4d7 JW |
1225 | /* |
1226 | * If we need to retry but a fatal signal is pending, handle the | |
1227 | * signal first. We do not need to release the mmap_sem because it | |
1228 | * would already be released in __lock_page_or_retry in mm/filemap.c. | |
1229 | */ | |
1230 | if (unlikely((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))) | |
1231 | return; | |
1232 | ||
1233 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
1234 | mm_fault_error(regs, error_code, address, fault); | |
1235 | return; | |
37b23e05 KM |
1236 | } |
1237 | ||
d065bd81 ML |
1238 | /* |
1239 | * Major/minor page fault accounting is only done on the | |
1240 | * initial attempt. If we go through a retry, it is extremely | |
1241 | * likely that the page will be found in page cache at that point. | |
1242 | */ | |
1243 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1244 | if (fault & VM_FAULT_MAJOR) { | |
1245 | tsk->maj_flt++; | |
a8b0ca17 | 1246 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, |
d065bd81 ML |
1247 | regs, address); |
1248 | } else { | |
1249 | tsk->min_flt++; | |
a8b0ca17 | 1250 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, |
d065bd81 ML |
1251 | regs, address); |
1252 | } | |
1253 | if (fault & VM_FAULT_RETRY) { | |
1254 | /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk | |
1255 | * of starvation. */ | |
1256 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
45cac65b | 1257 | flags |= FAULT_FLAG_TRIED; |
d065bd81 ML |
1258 | goto retry; |
1259 | } | |
ac17dc8e | 1260 | } |
d729ab35 | 1261 | |
8c938f9f IM |
1262 | check_v8086_mode(regs, address, tsk); |
1263 | ||
1da177e4 | 1264 | up_read(&mm->mmap_sem); |
1da177e4 | 1265 | } |
9326638c | 1266 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1267 | |
9326638c | 1268 | dotraplinkage void notrace |
6ba3c97a FW |
1269 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1270 | { | |
d4078e23 | 1271 | unsigned long address = read_cr2(); /* Get the faulting address */ |
6c1e0256 | 1272 | enum ctx_state prev_state; |
d4078e23 PZ |
1273 | |
1274 | /* | |
1275 | * We must have this function tagged with __kprobes, notrace and call | |
1276 | * read_cr2() before calling anything else. To avoid calling any kind | |
1277 | * of tracing machinery before we've observed the CR2 value. | |
1278 | * | |
1279 | * exception_{enter,exit}() contain all sorts of tracepoints. | |
1280 | */ | |
6c1e0256 FW |
1281 | |
1282 | prev_state = exception_enter(); | |
0ac09f9f | 1283 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1284 | exception_exit(prev_state); |
6ba3c97a | 1285 | } |
9326638c | 1286 | NOKPROBE_SYMBOL(do_page_fault); |
25c74b10 | 1287 | |
d4078e23 | 1288 | #ifdef CONFIG_TRACING |
9326638c MH |
1289 | static nokprobe_inline void |
1290 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1291 | unsigned long error_code) | |
d34603b0 SA |
1292 | { |
1293 | if (user_mode(regs)) | |
d4078e23 | 1294 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1295 | else |
d4078e23 | 1296 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1297 | } |
1298 | ||
9326638c | 1299 | dotraplinkage void notrace |
25c74b10 SA |
1300 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1301 | { | |
0ac09f9f JO |
1302 | /* |
1303 | * The exception_enter and tracepoint processing could | |
1304 | * trigger another page faults (user space callchain | |
1305 | * reading) and destroy the original cr2 value, so read | |
1306 | * the faulting address now. | |
1307 | */ | |
1308 | unsigned long address = read_cr2(); | |
d4078e23 | 1309 | enum ctx_state prev_state; |
25c74b10 SA |
1310 | |
1311 | prev_state = exception_enter(); | |
d4078e23 | 1312 | trace_page_fault_entries(address, regs, error_code); |
0ac09f9f | 1313 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1314 | exception_exit(prev_state); |
1315 | } | |
9326638c | 1316 | NOKPROBE_SYMBOL(trace_do_page_fault); |
d4078e23 | 1317 | #endif /* CONFIG_TRACING */ |