]>
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/sched.h> /* test_thread_flag(), ... */ |
7 | #include <linux/kdebug.h> /* oops_begin/end, ... */ | |
8 | #include <linux/module.h> /* search_exception_table */ | |
9 | #include <linux/bootmem.h> /* max_low_pfn */ | |
9326638c | 10 | #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ |
a2bcd473 | 11 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ |
cdd6c482 | 12 | #include <linux/perf_event.h> /* perf_sw_event */ |
f672b49b | 13 | #include <linux/hugetlb.h> /* hstate_index_to_shift */ |
268bb0ce | 14 | #include <linux/prefetch.h> /* prefetchw */ |
56dd9470 | 15 | #include <linux/context_tracking.h> /* exception_enter(), ... */ |
2d4a7167 | 16 | |
a2bcd473 IM |
17 | #include <asm/traps.h> /* dotraplinkage, ... */ |
18 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f8561296 | 19 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
f40c3300 AL |
20 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
21 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
1da177e4 | 22 | |
d34603b0 SA |
23 | #define CREATE_TRACE_POINTS |
24 | #include <asm/trace/exceptions.h> | |
25 | ||
33cb5243 | 26 | /* |
2d4a7167 IM |
27 | * Page fault error code bits: |
28 | * | |
29 | * bit 0 == 0: no page found 1: protection fault | |
30 | * bit 1 == 0: read access 1: write access | |
31 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
32 | * bit 3 == 1: use of reserved bit detected | |
33 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 34 | */ |
2d4a7167 IM |
35 | enum x86_pf_error_code { |
36 | ||
37 | PF_PROT = 1 << 0, | |
38 | PF_WRITE = 1 << 1, | |
39 | PF_USER = 1 << 2, | |
40 | PF_RSVD = 1 << 3, | |
41 | PF_INSTR = 1 << 4, | |
42 | }; | |
66c58156 | 43 | |
b814d41f | 44 | /* |
b319eed0 IM |
45 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
46 | * handled by mmiotrace: | |
b814d41f | 47 | */ |
9326638c | 48 | static nokprobe_inline int |
62c9295f | 49 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 50 | { |
0fd0e3da PP |
51 | if (unlikely(is_kmmio_active())) |
52 | if (kmmio_handler(regs, addr) == 1) | |
53 | return -1; | |
0fd0e3da | 54 | return 0; |
86069782 PP |
55 | } |
56 | ||
9326638c | 57 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 58 | { |
74a0b576 CH |
59 | int ret = 0; |
60 | ||
61 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 62 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
63 | preempt_disable(); |
64 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
65 | ret = 1; | |
66 | preempt_enable(); | |
67 | } | |
1bd858a5 | 68 | |
74a0b576 | 69 | return ret; |
33cb5243 | 70 | } |
1bd858a5 | 71 | |
1dc85be0 | 72 | /* |
2d4a7167 IM |
73 | * Prefetch quirks: |
74 | * | |
75 | * 32-bit mode: | |
76 | * | |
77 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
78 | * Check that here and ignore it. | |
1dc85be0 | 79 | * |
2d4a7167 | 80 | * 64-bit mode: |
1dc85be0 | 81 | * |
2d4a7167 IM |
82 | * Sometimes the CPU reports invalid exceptions on prefetch. |
83 | * Check that here and ignore it. | |
84 | * | |
85 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 86 | */ |
107a0367 IM |
87 | static inline int |
88 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
89 | unsigned char opcode, int *prefetch) | |
90 | { | |
91 | unsigned char instr_hi = opcode & 0xf0; | |
92 | unsigned char instr_lo = opcode & 0x0f; | |
93 | ||
94 | switch (instr_hi) { | |
95 | case 0x20: | |
96 | case 0x30: | |
97 | /* | |
98 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
99 | * In X86_64 long mode, the CPU will signal invalid | |
100 | * opcode if some of these prefixes are present so | |
101 | * X86_64 will never get here anyway | |
102 | */ | |
103 | return ((instr_lo & 7) == 0x6); | |
104 | #ifdef CONFIG_X86_64 | |
105 | case 0x40: | |
106 | /* | |
107 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
108 | * Need to figure out under what instruction mode the | |
109 | * instruction was issued. Could check the LDT for lm, | |
110 | * but for now it's good enough to assume that long | |
111 | * mode only uses well known segments or kernel. | |
112 | */ | |
318f5a2a | 113 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
114 | #endif |
115 | case 0x60: | |
116 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
117 | return (instr_lo & 0xC) == 0x4; | |
118 | case 0xF0: | |
119 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
120 | return !instr_lo || (instr_lo>>1) == 1; | |
121 | case 0x00: | |
122 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
123 | if (probe_kernel_address(instr, opcode)) | |
124 | return 0; | |
125 | ||
126 | *prefetch = (instr_lo == 0xF) && | |
127 | (opcode == 0x0D || opcode == 0x18); | |
128 | return 0; | |
129 | default: | |
130 | return 0; | |
131 | } | |
132 | } | |
133 | ||
2d4a7167 IM |
134 | static int |
135 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 136 | { |
2d4a7167 | 137 | unsigned char *max_instr; |
ab2bf0c1 | 138 | unsigned char *instr; |
33cb5243 | 139 | int prefetch = 0; |
1da177e4 | 140 | |
3085354d IM |
141 | /* |
142 | * If it was a exec (instruction fetch) fault on NX page, then | |
143 | * do not ignore the fault: | |
144 | */ | |
66c58156 | 145 | if (error_code & PF_INSTR) |
1da177e4 | 146 | return 0; |
1dc85be0 | 147 | |
107a0367 | 148 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 149 | max_instr = instr + 15; |
1da177e4 | 150 | |
76381fee | 151 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
152 | return 0; |
153 | ||
107a0367 | 154 | while (instr < max_instr) { |
2d4a7167 | 155 | unsigned char opcode; |
1da177e4 | 156 | |
ab2bf0c1 | 157 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 158 | break; |
1da177e4 | 159 | |
1da177e4 LT |
160 | instr++; |
161 | ||
107a0367 | 162 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 163 | break; |
1da177e4 LT |
164 | } |
165 | return prefetch; | |
166 | } | |
167 | ||
2d4a7167 IM |
168 | static void |
169 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 170 | struct task_struct *tsk, int fault) |
c4aba4a8 | 171 | { |
f672b49b | 172 | unsigned lsb = 0; |
c4aba4a8 HH |
173 | siginfo_t info; |
174 | ||
2d4a7167 IM |
175 | info.si_signo = si_signo; |
176 | info.si_errno = 0; | |
177 | info.si_code = si_code; | |
178 | info.si_addr = (void __user *)address; | |
f672b49b AK |
179 | if (fault & VM_FAULT_HWPOISON_LARGE) |
180 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
181 | if (fault & VM_FAULT_HWPOISON) | |
182 | lsb = PAGE_SHIFT; | |
183 | info.si_addr_lsb = lsb; | |
2d4a7167 | 184 | |
c4aba4a8 HH |
185 | force_sig_info(si_signo, &info, tsk); |
186 | } | |
187 | ||
f2f13a85 IM |
188 | DEFINE_SPINLOCK(pgd_lock); |
189 | LIST_HEAD(pgd_list); | |
190 | ||
191 | #ifdef CONFIG_X86_32 | |
192 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 193 | { |
f2f13a85 IM |
194 | unsigned index = pgd_index(address); |
195 | pgd_t *pgd_k; | |
196 | pud_t *pud, *pud_k; | |
197 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 198 | |
f2f13a85 IM |
199 | pgd += index; |
200 | pgd_k = init_mm.pgd + index; | |
201 | ||
202 | if (!pgd_present(*pgd_k)) | |
203 | return NULL; | |
204 | ||
205 | /* | |
206 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
207 | * and redundant with the set_pmd() on non-PAE. As would | |
208 | * set_pud. | |
209 | */ | |
210 | pud = pud_offset(pgd, address); | |
211 | pud_k = pud_offset(pgd_k, address); | |
212 | if (!pud_present(*pud_k)) | |
213 | return NULL; | |
214 | ||
215 | pmd = pmd_offset(pud, address); | |
216 | pmd_k = pmd_offset(pud_k, address); | |
217 | if (!pmd_present(*pmd_k)) | |
218 | return NULL; | |
219 | ||
b8bcfe99 | 220 | if (!pmd_present(*pmd)) |
f2f13a85 | 221 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 222 | else |
f2f13a85 | 223 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
224 | |
225 | return pmd_k; | |
226 | } | |
227 | ||
228 | void vmalloc_sync_all(void) | |
229 | { | |
230 | unsigned long address; | |
231 | ||
232 | if (SHARED_KERNEL_PMD) | |
233 | return; | |
234 | ||
235 | for (address = VMALLOC_START & PMD_MASK; | |
236 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
237 | address += PMD_SIZE) { | |
f2f13a85 IM |
238 | struct page *page; |
239 | ||
a79e53d8 | 240 | spin_lock(&pgd_lock); |
f2f13a85 | 241 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 242 | spinlock_t *pgt_lock; |
f01f7c56 | 243 | pmd_t *ret; |
617d34d9 | 244 | |
a79e53d8 | 245 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
246 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
247 | ||
248 | spin_lock(pgt_lock); | |
249 | ret = vmalloc_sync_one(page_address(page), address); | |
250 | spin_unlock(pgt_lock); | |
251 | ||
252 | if (!ret) | |
f2f13a85 IM |
253 | break; |
254 | } | |
a79e53d8 | 255 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
256 | } |
257 | } | |
258 | ||
259 | /* | |
260 | * 32-bit: | |
261 | * | |
262 | * Handle a fault on the vmalloc or module mapping area | |
263 | */ | |
9326638c | 264 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
265 | { |
266 | unsigned long pgd_paddr; | |
267 | pmd_t *pmd_k; | |
268 | pte_t *pte_k; | |
269 | ||
270 | /* Make sure we are in vmalloc area: */ | |
271 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
272 | return -1; | |
273 | ||
ebc8827f FW |
274 | WARN_ON_ONCE(in_nmi()); |
275 | ||
f2f13a85 IM |
276 | /* |
277 | * Synchronize this task's top level page-table | |
278 | * with the 'reference' page table. | |
279 | * | |
280 | * Do _not_ use "current" here. We might be inside | |
281 | * an interrupt in the middle of a task switch.. | |
282 | */ | |
283 | pgd_paddr = read_cr3(); | |
284 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
285 | if (!pmd_k) | |
286 | return -1; | |
287 | ||
288 | pte_k = pte_offset_kernel(pmd_k, address); | |
289 | if (!pte_present(*pte_k)) | |
290 | return -1; | |
291 | ||
292 | return 0; | |
293 | } | |
9326638c | 294 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
295 | |
296 | /* | |
297 | * Did it hit the DOS screen memory VA from vm86 mode? | |
298 | */ | |
299 | static inline void | |
300 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
301 | struct task_struct *tsk) | |
302 | { | |
303 | unsigned long bit; | |
304 | ||
305 | if (!v8086_mode(regs)) | |
306 | return; | |
307 | ||
308 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
309 | if (bit < 32) | |
310 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 311 | } |
1da177e4 | 312 | |
087975b0 | 313 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 314 | { |
087975b0 AM |
315 | return pfn < max_low_pfn; |
316 | } | |
1156e098 | 317 | |
087975b0 AM |
318 | static void dump_pagetable(unsigned long address) |
319 | { | |
320 | pgd_t *base = __va(read_cr3()); | |
321 | pgd_t *pgd = &base[pgd_index(address)]; | |
322 | pmd_t *pmd; | |
323 | pte_t *pte; | |
2d4a7167 | 324 | |
1156e098 | 325 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
326 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
327 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
328 | goto out; | |
1156e098 | 329 | #endif |
087975b0 AM |
330 | pmd = pmd_offset(pud_offset(pgd, address), address); |
331 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
332 | |
333 | /* | |
334 | * We must not directly access the pte in the highpte | |
335 | * case if the page table is located in highmem. | |
336 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 337 | * it's allocated already: |
1156e098 | 338 | */ |
087975b0 AM |
339 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
340 | goto out; | |
1156e098 | 341 | |
087975b0 AM |
342 | pte = pte_offset_kernel(pmd, address); |
343 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
344 | out: | |
1156e098 | 345 | printk("\n"); |
f2f13a85 IM |
346 | } |
347 | ||
348 | #else /* CONFIG_X86_64: */ | |
349 | ||
350 | void vmalloc_sync_all(void) | |
351 | { | |
9661d5bc | 352 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); |
f2f13a85 IM |
353 | } |
354 | ||
355 | /* | |
356 | * 64-bit: | |
357 | * | |
358 | * Handle a fault on the vmalloc area | |
359 | * | |
360 | * This assumes no large pages in there. | |
361 | */ | |
9326638c | 362 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
363 | { |
364 | pgd_t *pgd, *pgd_ref; | |
365 | pud_t *pud, *pud_ref; | |
366 | pmd_t *pmd, *pmd_ref; | |
367 | pte_t *pte, *pte_ref; | |
368 | ||
369 | /* Make sure we are in vmalloc area: */ | |
370 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
371 | return -1; | |
372 | ||
ebc8827f FW |
373 | WARN_ON_ONCE(in_nmi()); |
374 | ||
f2f13a85 IM |
375 | /* |
376 | * Copy kernel mappings over when needed. This can also | |
377 | * happen within a race in page table update. In the later | |
378 | * case just flush: | |
379 | */ | |
380 | pgd = pgd_offset(current->active_mm, address); | |
381 | pgd_ref = pgd_offset_k(address); | |
382 | if (pgd_none(*pgd_ref)) | |
383 | return -1; | |
384 | ||
1160c277 | 385 | if (pgd_none(*pgd)) { |
f2f13a85 | 386 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
387 | arch_flush_lazy_mmu_mode(); |
388 | } else { | |
f2f13a85 | 389 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 390 | } |
f2f13a85 IM |
391 | |
392 | /* | |
393 | * Below here mismatches are bugs because these lower tables | |
394 | * are shared: | |
395 | */ | |
396 | ||
397 | pud = pud_offset(pgd, address); | |
398 | pud_ref = pud_offset(pgd_ref, address); | |
399 | if (pud_none(*pud_ref)) | |
400 | return -1; | |
401 | ||
402 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
403 | BUG(); | |
404 | ||
405 | pmd = pmd_offset(pud, address); | |
406 | pmd_ref = pmd_offset(pud_ref, address); | |
407 | if (pmd_none(*pmd_ref)) | |
408 | return -1; | |
409 | ||
410 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
411 | BUG(); | |
412 | ||
413 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
414 | if (!pte_present(*pte_ref)) | |
415 | return -1; | |
416 | ||
417 | pte = pte_offset_kernel(pmd, address); | |
418 | ||
419 | /* | |
420 | * Don't use pte_page here, because the mappings can point | |
421 | * outside mem_map, and the NUMA hash lookup cannot handle | |
422 | * that: | |
423 | */ | |
424 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
425 | BUG(); | |
426 | ||
427 | return 0; | |
428 | } | |
9326638c | 429 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 430 | |
e05139f2 | 431 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 432 | static const char errata93_warning[] = |
ad361c98 JP |
433 | KERN_ERR |
434 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
435 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
436 | "******* Please consider a BIOS update.\n" | |
437 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 438 | #endif |
f2f13a85 IM |
439 | |
440 | /* | |
441 | * No vm86 mode in 64-bit mode: | |
442 | */ | |
443 | static inline void | |
444 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
445 | struct task_struct *tsk) | |
446 | { | |
447 | } | |
448 | ||
449 | static int bad_address(void *p) | |
450 | { | |
451 | unsigned long dummy; | |
452 | ||
453 | return probe_kernel_address((unsigned long *)p, dummy); | |
454 | } | |
455 | ||
456 | static void dump_pagetable(unsigned long address) | |
457 | { | |
087975b0 AM |
458 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
459 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
460 | pud_t *pud; |
461 | pmd_t *pmd; | |
462 | pte_t *pte; | |
463 | ||
2d4a7167 IM |
464 | if (bad_address(pgd)) |
465 | goto bad; | |
466 | ||
d646bce4 | 467 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
468 | |
469 | if (!pgd_present(*pgd)) | |
470 | goto out; | |
1da177e4 | 471 | |
d2ae5b5f | 472 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
473 | if (bad_address(pud)) |
474 | goto bad; | |
475 | ||
1da177e4 | 476 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 477 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 478 | goto out; |
1da177e4 LT |
479 | |
480 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
481 | if (bad_address(pmd)) |
482 | goto bad; | |
483 | ||
1da177e4 | 484 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
485 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
486 | goto out; | |
1da177e4 LT |
487 | |
488 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
489 | if (bad_address(pte)) |
490 | goto bad; | |
491 | ||
33cb5243 | 492 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 493 | out: |
1da177e4 LT |
494 | printk("\n"); |
495 | return; | |
496 | bad: | |
497 | printk("BAD\n"); | |
8c938f9f IM |
498 | } |
499 | ||
f2f13a85 | 500 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 501 | |
2d4a7167 IM |
502 | /* |
503 | * Workaround for K8 erratum #93 & buggy BIOS. | |
504 | * | |
505 | * BIOS SMM functions are required to use a specific workaround | |
506 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
507 | * | |
508 | * A lot of BIOS that didn't get tested properly miss this. | |
509 | * | |
510 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
511 | * Try to work around it here. | |
512 | * | |
513 | * Note we only handle faults in kernel here. | |
514 | * Does nothing on 32-bit. | |
fdfe8aa8 | 515 | */ |
33cb5243 | 516 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 517 | { |
e05139f2 JB |
518 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
519 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
520 | || boot_cpu_data.x86 != 0xf) | |
521 | return 0; | |
522 | ||
65ea5b03 | 523 | if (address != regs->ip) |
1da177e4 | 524 | return 0; |
2d4a7167 | 525 | |
33cb5243 | 526 | if ((address >> 32) != 0) |
1da177e4 | 527 | return 0; |
2d4a7167 | 528 | |
1da177e4 | 529 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
530 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
531 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 532 | printk_once(errata93_warning); |
65ea5b03 | 533 | regs->ip = address; |
1da177e4 LT |
534 | return 1; |
535 | } | |
fdfe8aa8 | 536 | #endif |
1da177e4 | 537 | return 0; |
33cb5243 | 538 | } |
1da177e4 | 539 | |
35f3266f | 540 | /* |
2d4a7167 IM |
541 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
542 | * to illegal addresses >4GB. | |
543 | * | |
544 | * We catch this in the page fault handler because these addresses | |
545 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
546 | * segment in LDT is compatibility mode. |
547 | */ | |
548 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
549 | { | |
550 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 551 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
552 | return 1; |
553 | #endif | |
554 | return 0; | |
555 | } | |
556 | ||
29caf2f9 HH |
557 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
558 | { | |
559 | #ifdef CONFIG_X86_F00F_BUG | |
560 | unsigned long nr; | |
2d4a7167 | 561 | |
29caf2f9 | 562 | /* |
2d4a7167 | 563 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 564 | */ |
e2604b49 | 565 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
566 | nr = (address - idt_descr.address) >> 3; |
567 | ||
568 | if (nr == 6) { | |
569 | do_invalid_op(regs, 0); | |
570 | return 1; | |
571 | } | |
572 | } | |
573 | #endif | |
574 | return 0; | |
575 | } | |
576 | ||
8f766149 IM |
577 | static const char nx_warning[] = KERN_CRIT |
578 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
579 | static const char smep_warning[] = KERN_CRIT |
580 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 581 | |
2d4a7167 IM |
582 | static void |
583 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
584 | unsigned long address) | |
b3279c7f | 585 | { |
1156e098 HH |
586 | if (!oops_may_print()) |
587 | return; | |
588 | ||
1156e098 | 589 | if (error_code & PF_INSTR) { |
93809be8 | 590 | unsigned int level; |
426e34cc MF |
591 | pgd_t *pgd; |
592 | pte_t *pte; | |
2d4a7167 | 593 | |
426e34cc MF |
594 | pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
595 | pgd += pgd_index(address); | |
596 | ||
597 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 598 | |
8f766149 | 599 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 600 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
601 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
602 | (pgd_flags(*pgd) & _PAGE_USER) && | |
603 | (read_cr4() & X86_CR4_SMEP)) | |
604 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); | |
1156e098 | 605 | } |
1156e098 | 606 | |
19f0dda9 | 607 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 608 | if (address < PAGE_SIZE) |
19f0dda9 | 609 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 610 | else |
19f0dda9 | 611 | printk(KERN_CONT "paging request"); |
2d4a7167 | 612 | |
f294a8ce | 613 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 614 | printk(KERN_ALERT "IP:"); |
5f01c988 | 615 | printk_address(regs->ip); |
2d4a7167 | 616 | |
b3279c7f HH |
617 | dump_pagetable(address); |
618 | } | |
619 | ||
2d4a7167 IM |
620 | static noinline void |
621 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
622 | unsigned long address) | |
1da177e4 | 623 | { |
2d4a7167 IM |
624 | struct task_struct *tsk; |
625 | unsigned long flags; | |
626 | int sig; | |
627 | ||
628 | flags = oops_begin(); | |
629 | tsk = current; | |
630 | sig = SIGKILL; | |
1209140c | 631 | |
1da177e4 | 632 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 633 | tsk->comm, address); |
1da177e4 | 634 | dump_pagetable(address); |
2d4a7167 IM |
635 | |
636 | tsk->thread.cr2 = address; | |
51e7dc70 | 637 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
638 | tsk->thread.error_code = error_code; |
639 | ||
22f5991c | 640 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 641 | sig = 0; |
2d4a7167 | 642 | |
874d93d1 | 643 | oops_end(flags, regs, sig); |
1da177e4 LT |
644 | } |
645 | ||
2d4a7167 IM |
646 | static noinline void |
647 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 648 | unsigned long address, int signal, int si_code) |
92181f19 NP |
649 | { |
650 | struct task_struct *tsk = current; | |
92181f19 NP |
651 | unsigned long flags; |
652 | int sig; | |
92181f19 | 653 | |
2d4a7167 | 654 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 655 | if (fixup_exception(regs)) { |
c026b359 PZ |
656 | /* |
657 | * Any interrupt that takes a fault gets the fixup. This makes | |
658 | * the below recursive fault logic only apply to a faults from | |
659 | * task context. | |
660 | */ | |
661 | if (in_interrupt()) | |
662 | return; | |
663 | ||
664 | /* | |
665 | * Per the above we're !in_interrupt(), aka. task context. | |
666 | * | |
667 | * In this case we need to make sure we're not recursively | |
668 | * faulting through the emulate_vsyscall() logic. | |
669 | */ | |
4fc34901 | 670 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 671 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
672 | tsk->thread.error_code = error_code | PF_USER; |
673 | tsk->thread.cr2 = address; | |
674 | ||
675 | /* XXX: hwpoison faults will set the wrong code. */ | |
676 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
677 | } | |
c026b359 PZ |
678 | |
679 | /* | |
680 | * Barring that, we can do the fixup and be happy. | |
681 | */ | |
92181f19 | 682 | return; |
4fc34901 | 683 | } |
92181f19 NP |
684 | |
685 | /* | |
2d4a7167 IM |
686 | * 32-bit: |
687 | * | |
688 | * Valid to do another page fault here, because if this fault | |
689 | * had been triggered by is_prefetch fixup_exception would have | |
690 | * handled it. | |
691 | * | |
692 | * 64-bit: | |
92181f19 | 693 | * |
2d4a7167 | 694 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
695 | */ |
696 | if (is_prefetch(regs, error_code, address)) | |
697 | return; | |
698 | ||
699 | if (is_errata93(regs, address)) | |
700 | return; | |
701 | ||
702 | /* | |
703 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 704 | * terminate things with extreme prejudice: |
92181f19 | 705 | */ |
92181f19 | 706 | flags = oops_begin(); |
92181f19 NP |
707 | |
708 | show_fault_oops(regs, error_code, address); | |
709 | ||
a70857e4 | 710 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 711 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 712 | |
1cc99544 | 713 | tsk->thread.cr2 = address; |
51e7dc70 | 714 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 715 | tsk->thread.error_code = error_code; |
92181f19 | 716 | |
92181f19 NP |
717 | sig = SIGKILL; |
718 | if (__die("Oops", regs, error_code)) | |
719 | sig = 0; | |
2d4a7167 | 720 | |
92181f19 | 721 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 722 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 723 | |
92181f19 | 724 | oops_end(flags, regs, sig); |
92181f19 NP |
725 | } |
726 | ||
2d4a7167 IM |
727 | /* |
728 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
729 | * sysctl is set: | |
730 | */ | |
731 | static inline void | |
732 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
733 | unsigned long address, struct task_struct *tsk) | |
734 | { | |
735 | if (!unhandled_signal(tsk, SIGSEGV)) | |
736 | return; | |
737 | ||
738 | if (!printk_ratelimit()) | |
739 | return; | |
740 | ||
a1a08d1c | 741 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
742 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
743 | tsk->comm, task_pid_nr(tsk), address, | |
744 | (void *)regs->ip, (void *)regs->sp, error_code); | |
745 | ||
746 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
747 | ||
748 | printk(KERN_CONT "\n"); | |
749 | } | |
750 | ||
751 | static void | |
752 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
753 | unsigned long address, int si_code) | |
92181f19 NP |
754 | { |
755 | struct task_struct *tsk = current; | |
756 | ||
757 | /* User mode accesses just cause a SIGSEGV */ | |
758 | if (error_code & PF_USER) { | |
759 | /* | |
2d4a7167 | 760 | * It's possible to have interrupts off here: |
92181f19 NP |
761 | */ |
762 | local_irq_enable(); | |
763 | ||
764 | /* | |
765 | * Valid to do another page fault here because this one came | |
2d4a7167 | 766 | * from user space: |
92181f19 NP |
767 | */ |
768 | if (is_prefetch(regs, error_code, address)) | |
769 | return; | |
770 | ||
771 | if (is_errata100(regs, address)) | |
772 | return; | |
773 | ||
3ae36655 AL |
774 | #ifdef CONFIG_X86_64 |
775 | /* | |
776 | * Instruction fetch faults in the vsyscall page might need | |
777 | * emulation. | |
778 | */ | |
779 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 780 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
781 | if (emulate_vsyscall(regs, address)) |
782 | return; | |
783 | } | |
784 | #endif | |
e575a86f KC |
785 | /* Kernel addresses are always protection faults: */ |
786 | if (address >= TASK_SIZE) | |
787 | error_code |= PF_PROT; | |
3ae36655 | 788 | |
e575a86f | 789 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
790 | show_signal_msg(regs, error_code, address, tsk); |
791 | ||
2d4a7167 | 792 | tsk->thread.cr2 = address; |
e575a86f | 793 | tsk->thread.error_code = error_code; |
51e7dc70 | 794 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 795 | |
f672b49b | 796 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 797 | |
92181f19 NP |
798 | return; |
799 | } | |
800 | ||
801 | if (is_f00f_bug(regs, address)) | |
802 | return; | |
803 | ||
4fc34901 | 804 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
805 | } |
806 | ||
2d4a7167 IM |
807 | static noinline void |
808 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
809 | unsigned long address) | |
92181f19 NP |
810 | { |
811 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
812 | } | |
813 | ||
2d4a7167 IM |
814 | static void |
815 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
816 | unsigned long address, int si_code) | |
92181f19 NP |
817 | { |
818 | struct mm_struct *mm = current->mm; | |
819 | ||
820 | /* | |
821 | * Something tried to access memory that isn't in our memory map.. | |
822 | * Fix it, but check if it's kernel or user first.. | |
823 | */ | |
824 | up_read(&mm->mmap_sem); | |
825 | ||
826 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
827 | } | |
828 | ||
2d4a7167 IM |
829 | static noinline void |
830 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
831 | { |
832 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
833 | } | |
834 | ||
2d4a7167 IM |
835 | static noinline void |
836 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
837 | unsigned long address) | |
92181f19 NP |
838 | { |
839 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
840 | } | |
841 | ||
2d4a7167 | 842 | static void |
a6e04aa9 AK |
843 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
844 | unsigned int fault) | |
92181f19 NP |
845 | { |
846 | struct task_struct *tsk = current; | |
a6e04aa9 | 847 | int code = BUS_ADRERR; |
92181f19 | 848 | |
2d4a7167 | 849 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 850 | if (!(error_code & PF_USER)) { |
4fc34901 | 851 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
852 | return; |
853 | } | |
2d4a7167 | 854 | |
cd1b68f0 | 855 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
856 | if (is_prefetch(regs, error_code, address)) |
857 | return; | |
2d4a7167 IM |
858 | |
859 | tsk->thread.cr2 = address; | |
860 | tsk->thread.error_code = error_code; | |
51e7dc70 | 861 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 862 | |
a6e04aa9 | 863 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 864 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
865 | printk(KERN_ERR |
866 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
867 | tsk->comm, tsk->pid, address); | |
868 | code = BUS_MCEERR_AR; | |
869 | } | |
870 | #endif | |
f672b49b | 871 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
872 | } |
873 | ||
3a13c4d7 | 874 | static noinline void |
2d4a7167 IM |
875 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
876 | unsigned long address, unsigned int fault) | |
92181f19 | 877 | { |
3a13c4d7 | 878 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
3a13c4d7 JW |
879 | no_context(regs, error_code, address, 0, 0); |
880 | return; | |
b80ef10e | 881 | } |
b80ef10e | 882 | |
2d4a7167 | 883 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
884 | /* Kernel mode? Handle exceptions or die: */ |
885 | if (!(error_code & PF_USER)) { | |
4fc34901 AL |
886 | no_context(regs, error_code, address, |
887 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 888 | return; |
f8626854 AV |
889 | } |
890 | ||
c2d23f91 DR |
891 | /* |
892 | * We ran out of memory, call the OOM killer, and return the | |
893 | * userspace (which will retry the fault, or kill us if we got | |
894 | * oom-killed): | |
895 | */ | |
896 | pagefault_out_of_memory(); | |
2d4a7167 | 897 | } else { |
f672b49b AK |
898 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
899 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 900 | do_sigbus(regs, error_code, address, fault); |
2d4a7167 IM |
901 | else |
902 | BUG(); | |
903 | } | |
92181f19 NP |
904 | } |
905 | ||
d8b57bb7 TG |
906 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
907 | { | |
908 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
909 | return 0; | |
2d4a7167 | 910 | |
d8b57bb7 TG |
911 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
912 | return 0; | |
913 | ||
914 | return 1; | |
915 | } | |
916 | ||
5b727a3b | 917 | /* |
2d4a7167 IM |
918 | * Handle a spurious fault caused by a stale TLB entry. |
919 | * | |
920 | * This allows us to lazily refresh the TLB when increasing the | |
921 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
922 | * eagerly is very expensive since that implies doing a full | |
923 | * cross-processor TLB flush, even if no stale TLB entries exist | |
924 | * on other processors. | |
925 | * | |
31668511 DV |
926 | * Spurious faults may only occur if the TLB contains an entry with |
927 | * fewer permission than the page table entry. Non-present (P = 0) | |
928 | * and reserved bit (R = 1) faults are never spurious. | |
929 | * | |
5b727a3b JF |
930 | * There are no security implications to leaving a stale TLB when |
931 | * increasing the permissions on a page. | |
31668511 DV |
932 | * |
933 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
934 | * | |
935 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
936 | * (Optional Invalidation). | |
5b727a3b | 937 | */ |
9326638c | 938 | static noinline int |
2d4a7167 | 939 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
940 | { |
941 | pgd_t *pgd; | |
942 | pud_t *pud; | |
943 | pmd_t *pmd; | |
944 | pte_t *pte; | |
3c3e5694 | 945 | int ret; |
5b727a3b | 946 | |
31668511 DV |
947 | /* |
948 | * Only writes to RO or instruction fetches from NX may cause | |
949 | * spurious faults. | |
950 | * | |
951 | * These could be from user or supervisor accesses but the TLB | |
952 | * is only lazily flushed after a kernel mapping protection | |
953 | * change, so user accesses are not expected to cause spurious | |
954 | * faults. | |
955 | */ | |
956 | if (error_code != (PF_WRITE | PF_PROT) | |
957 | && error_code != (PF_INSTR | PF_PROT)) | |
5b727a3b JF |
958 | return 0; |
959 | ||
960 | pgd = init_mm.pgd + pgd_index(address); | |
961 | if (!pgd_present(*pgd)) | |
962 | return 0; | |
963 | ||
964 | pud = pud_offset(pgd, address); | |
965 | if (!pud_present(*pud)) | |
966 | return 0; | |
967 | ||
d8b57bb7 TG |
968 | if (pud_large(*pud)) |
969 | return spurious_fault_check(error_code, (pte_t *) pud); | |
970 | ||
5b727a3b JF |
971 | pmd = pmd_offset(pud, address); |
972 | if (!pmd_present(*pmd)) | |
973 | return 0; | |
974 | ||
d8b57bb7 TG |
975 | if (pmd_large(*pmd)) |
976 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
977 | ||
5b727a3b | 978 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 979 | if (!pte_present(*pte)) |
5b727a3b JF |
980 | return 0; |
981 | ||
3c3e5694 SR |
982 | ret = spurious_fault_check(error_code, pte); |
983 | if (!ret) | |
984 | return 0; | |
985 | ||
986 | /* | |
2d4a7167 IM |
987 | * Make sure we have permissions in PMD. |
988 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
989 | */ |
990 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
991 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 992 | |
3c3e5694 | 993 | return ret; |
5b727a3b | 994 | } |
9326638c | 995 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 996 | |
abd4f750 | 997 | int show_unhandled_signals = 1; |
1da177e4 | 998 | |
2d4a7167 | 999 | static inline int |
68da336a | 1000 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1001 | { |
68da336a | 1002 | if (error_code & PF_WRITE) { |
2d4a7167 | 1003 | /* write, present and write, not present: */ |
92181f19 NP |
1004 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1005 | return 1; | |
2d4a7167 | 1006 | return 0; |
92181f19 NP |
1007 | } |
1008 | ||
2d4a7167 IM |
1009 | /* read, present: */ |
1010 | if (unlikely(error_code & PF_PROT)) | |
1011 | return 1; | |
1012 | ||
1013 | /* read, not present: */ | |
1014 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1015 | return 1; | |
1016 | ||
92181f19 NP |
1017 | return 0; |
1018 | } | |
1019 | ||
0973a06c HS |
1020 | static int fault_in_kernel_space(unsigned long address) |
1021 | { | |
d9517346 | 1022 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1023 | } |
1024 | ||
40d3cd66 PA |
1025 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1026 | { | |
4640c7ee PA |
1027 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1028 | return false; | |
1029 | ||
1030 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1031 | return false; | |
1032 | ||
40d3cd66 PA |
1033 | if (error_code & PF_USER) |
1034 | return false; | |
1035 | ||
1036 | if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) | |
1037 | return false; | |
1038 | ||
1039 | return true; | |
1040 | } | |
1041 | ||
1da177e4 LT |
1042 | /* |
1043 | * This routine handles page faults. It determines the address, | |
1044 | * and the problem, and then passes it off to one of the appropriate | |
1045 | * routines. | |
d4078e23 PZ |
1046 | * |
1047 | * This function must have noinline because both callers | |
1048 | * {,trace_}do_page_fault() have notrace on. Having this an actual function | |
1049 | * guarantees there's a function trace entry. | |
1da177e4 | 1050 | */ |
9326638c | 1051 | static noinline void |
0ac09f9f JO |
1052 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1053 | unsigned long address) | |
1da177e4 | 1054 | { |
2d4a7167 | 1055 | struct vm_area_struct *vma; |
1da177e4 LT |
1056 | struct task_struct *tsk; |
1057 | struct mm_struct *mm; | |
26178ec1 | 1058 | int fault, major = 0; |
759496ba | 1059 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1060 | |
a9ba9a3b AV |
1061 | tsk = current; |
1062 | mm = tsk->mm; | |
2d4a7167 | 1063 | |
f8561296 VN |
1064 | /* |
1065 | * Detect and handle instructions that would cause a page fault for | |
1066 | * both a tracked kernel page and a userspace page. | |
1067 | */ | |
1068 | if (kmemcheck_active(regs)) | |
1069 | kmemcheck_hide(regs); | |
5dfaf90f | 1070 | prefetchw(&mm->mmap_sem); |
f8561296 | 1071 | |
0fd0e3da | 1072 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1073 | return; |
1da177e4 LT |
1074 | |
1075 | /* | |
1076 | * We fault-in kernel-space virtual memory on-demand. The | |
1077 | * 'reference' page table is init_mm.pgd. | |
1078 | * | |
1079 | * NOTE! We MUST NOT take any locks for this case. We may | |
1080 | * be in an interrupt or a critical region, and should | |
1081 | * only copy the information from the master page table, | |
1082 | * nothing more. | |
1083 | * | |
1084 | * This verifies that the fault happens in kernel space | |
1085 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1086 | * protection error (error_code & 9) == 0. |
1da177e4 | 1087 | */ |
0973a06c | 1088 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1089 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1090 | if (vmalloc_fault(address) >= 0) | |
1091 | return; | |
1092 | ||
1093 | if (kmemcheck_fault(regs, address, error_code)) | |
1094 | return; | |
1095 | } | |
5b727a3b | 1096 | |
2d4a7167 | 1097 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1098 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1099 | return; |
1100 | ||
2d4a7167 | 1101 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1102 | if (kprobes_fault(regs)) |
9be260a6 | 1103 | return; |
f8c2ee22 HH |
1104 | /* |
1105 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1106 | * fault we could otherwise deadlock: |
f8c2ee22 | 1107 | */ |
92181f19 | 1108 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1109 | |
92181f19 | 1110 | return; |
f8c2ee22 HH |
1111 | } |
1112 | ||
2d4a7167 | 1113 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1114 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1115 | return; |
8c914cb7 | 1116 | |
66c58156 | 1117 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1118 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1119 | |
4640c7ee PA |
1120 | if (unlikely(smap_violation(error_code, regs))) { |
1121 | bad_area_nosemaphore(regs, error_code, address); | |
1122 | return; | |
40d3cd66 PA |
1123 | } |
1124 | ||
1da177e4 | 1125 | /* |
2d4a7167 IM |
1126 | * If we're in an interrupt, have no user context or are running |
1127 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1128 | */ |
92181f19 NP |
1129 | if (unlikely(in_atomic() || !mm)) { |
1130 | bad_area_nosemaphore(regs, error_code, address); | |
1131 | return; | |
1132 | } | |
1da177e4 | 1133 | |
e00b12e6 PZ |
1134 | /* |
1135 | * It's safe to allow irq's after cr2 has been saved and the | |
1136 | * vmalloc fault has been handled. | |
1137 | * | |
1138 | * User-mode registers count as a user access even for any | |
1139 | * potential system fault or CPU buglet: | |
1140 | */ | |
1141 | if (user_mode_vm(regs)) { | |
1142 | local_irq_enable(); | |
1143 | error_code |= PF_USER; | |
1144 | flags |= FAULT_FLAG_USER; | |
1145 | } else { | |
1146 | if (regs->flags & X86_EFLAGS_IF) | |
1147 | local_irq_enable(); | |
1148 | } | |
1149 | ||
1150 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1151 | ||
759496ba JW |
1152 | if (error_code & PF_WRITE) |
1153 | flags |= FAULT_FLAG_WRITE; | |
1154 | ||
3a1dfe6e IM |
1155 | /* |
1156 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1157 | * addresses in user space. All other faults represent errors in |
1158 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1159 | * case of an erroneous fault occurring in a code path which already | |
1160 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1161 | * against the address space. Luckily the kernel only validly | |
1162 | * references user space from well defined areas of code, which are | |
1163 | * listed in the exceptions table. | |
1da177e4 LT |
1164 | * |
1165 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1166 | * the source reference check when there is a possibility of a |
1167 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1168 | * validate the source. If this is invalid we can skip the address | |
1169 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1170 | */ |
92181f19 | 1171 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1172 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1173 | !search_exception_tables(regs->ip)) { |
1174 | bad_area_nosemaphore(regs, error_code, address); | |
1175 | return; | |
1176 | } | |
d065bd81 | 1177 | retry: |
1da177e4 | 1178 | down_read(&mm->mmap_sem); |
01006074 PZ |
1179 | } else { |
1180 | /* | |
2d4a7167 IM |
1181 | * The above down_read_trylock() might have succeeded in |
1182 | * which case we'll have missed the might_sleep() from | |
1183 | * down_read(): | |
01006074 PZ |
1184 | */ |
1185 | might_sleep(); | |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | vma = find_vma(mm, address); | |
92181f19 NP |
1189 | if (unlikely(!vma)) { |
1190 | bad_area(regs, error_code, address); | |
1191 | return; | |
1192 | } | |
1193 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1194 | goto good_area; |
92181f19 NP |
1195 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1196 | bad_area(regs, error_code, address); | |
1197 | return; | |
1198 | } | |
33cb5243 | 1199 | if (error_code & PF_USER) { |
6f4d368e HH |
1200 | /* |
1201 | * Accessing the stack below %sp is always a bug. | |
1202 | * The large cushion allows instructions like enter | |
2d4a7167 | 1203 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1204 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1205 | */ |
92181f19 NP |
1206 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1207 | bad_area(regs, error_code, address); | |
1208 | return; | |
1209 | } | |
1da177e4 | 1210 | } |
92181f19 NP |
1211 | if (unlikely(expand_stack(vma, address))) { |
1212 | bad_area(regs, error_code, address); | |
1213 | return; | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * Ok, we have a good vm_area for this memory access, so | |
1218 | * we can handle it.. | |
1219 | */ | |
1da177e4 | 1220 | good_area: |
68da336a | 1221 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1222 | bad_area_access_error(regs, error_code, address); |
1223 | return; | |
1da177e4 LT |
1224 | } |
1225 | ||
1226 | /* | |
1227 | * If for any reason at all we couldn't handle the fault, | |
1228 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1229 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1230 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
1da177e4 | 1231 | */ |
d065bd81 | 1232 | fault = handle_mm_fault(mm, vma, address, flags); |
26178ec1 | 1233 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1234 | |
3a13c4d7 | 1235 | /* |
26178ec1 LT |
1236 | * If we need to retry the mmap_sem has already been released, |
1237 | * and if there is a fatal signal pending there is no guarantee | |
1238 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1239 | */ |
26178ec1 LT |
1240 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1241 | /* Retry at most once */ | |
1242 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1243 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1244 | flags |= FAULT_FLAG_TRIED; | |
1245 | if (!fatal_signal_pending(tsk)) | |
1246 | goto retry; | |
1247 | } | |
1248 | ||
1249 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1250 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1251 | return; |
1252 | ||
1253 | /* Not returning to user mode? Handle exceptions or die: */ | |
1254 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1255 | return; |
26178ec1 | 1256 | } |
3a13c4d7 | 1257 | |
26178ec1 | 1258 | up_read(&mm->mmap_sem); |
3a13c4d7 JW |
1259 | if (unlikely(fault & VM_FAULT_ERROR)) { |
1260 | mm_fault_error(regs, error_code, address, fault); | |
1261 | return; | |
37b23e05 KM |
1262 | } |
1263 | ||
d065bd81 | 1264 | /* |
26178ec1 LT |
1265 | * Major/minor page fault accounting. If any of the events |
1266 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1267 | */ |
26178ec1 LT |
1268 | if (major) { |
1269 | tsk->maj_flt++; | |
1270 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1271 | } else { | |
1272 | tsk->min_flt++; | |
1273 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1274 | } |
d729ab35 | 1275 | |
8c938f9f | 1276 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1277 | } |
9326638c | 1278 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1279 | |
9326638c | 1280 | dotraplinkage void notrace |
6ba3c97a FW |
1281 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1282 | { | |
d4078e23 | 1283 | unsigned long address = read_cr2(); /* Get the faulting address */ |
6c1e0256 | 1284 | enum ctx_state prev_state; |
d4078e23 PZ |
1285 | |
1286 | /* | |
1287 | * We must have this function tagged with __kprobes, notrace and call | |
1288 | * read_cr2() before calling anything else. To avoid calling any kind | |
1289 | * of tracing machinery before we've observed the CR2 value. | |
1290 | * | |
1291 | * exception_{enter,exit}() contain all sorts of tracepoints. | |
1292 | */ | |
6c1e0256 FW |
1293 | |
1294 | prev_state = exception_enter(); | |
0ac09f9f | 1295 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1296 | exception_exit(prev_state); |
6ba3c97a | 1297 | } |
9326638c | 1298 | NOKPROBE_SYMBOL(do_page_fault); |
25c74b10 | 1299 | |
d4078e23 | 1300 | #ifdef CONFIG_TRACING |
9326638c MH |
1301 | static nokprobe_inline void |
1302 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1303 | unsigned long error_code) | |
d34603b0 SA |
1304 | { |
1305 | if (user_mode(regs)) | |
d4078e23 | 1306 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1307 | else |
d4078e23 | 1308 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1309 | } |
1310 | ||
9326638c | 1311 | dotraplinkage void notrace |
25c74b10 SA |
1312 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1313 | { | |
0ac09f9f JO |
1314 | /* |
1315 | * The exception_enter and tracepoint processing could | |
1316 | * trigger another page faults (user space callchain | |
1317 | * reading) and destroy the original cr2 value, so read | |
1318 | * the faulting address now. | |
1319 | */ | |
1320 | unsigned long address = read_cr2(); | |
d4078e23 | 1321 | enum ctx_state prev_state; |
25c74b10 SA |
1322 | |
1323 | prev_state = exception_enter(); | |
d4078e23 | 1324 | trace_page_fault_entries(address, regs, error_code); |
0ac09f9f | 1325 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1326 | exception_exit(prev_state); |
1327 | } | |
9326638c | 1328 | NOKPROBE_SYMBOL(trace_do_page_fault); |
d4078e23 | 1329 | #endif /* CONFIG_TRACING */ |