]>
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(), ... */ |
70ffdb93 | 16 | #include <linux/uaccess.h> /* faulthandler_disabled() */ |
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 */ | |
ba3e127e | 23 | #include <asm/vm86.h> /* struct vm86 */ |
1da177e4 | 24 | |
d34603b0 SA |
25 | #define CREATE_TRACE_POINTS |
26 | #include <asm/trace/exceptions.h> | |
27 | ||
33cb5243 | 28 | /* |
2d4a7167 IM |
29 | * Page fault error code bits: |
30 | * | |
31 | * bit 0 == 0: no page found 1: protection fault | |
32 | * bit 1 == 0: read access 1: write access | |
33 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
34 | * bit 3 == 1: use of reserved bit detected | |
35 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 36 | */ |
2d4a7167 IM |
37 | enum x86_pf_error_code { |
38 | ||
39 | PF_PROT = 1 << 0, | |
40 | PF_WRITE = 1 << 1, | |
41 | PF_USER = 1 << 2, | |
42 | PF_RSVD = 1 << 3, | |
43 | PF_INSTR = 1 << 4, | |
44 | }; | |
66c58156 | 45 | |
b814d41f | 46 | /* |
b319eed0 IM |
47 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
48 | * handled by mmiotrace: | |
b814d41f | 49 | */ |
9326638c | 50 | static nokprobe_inline int |
62c9295f | 51 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 52 | { |
0fd0e3da PP |
53 | if (unlikely(is_kmmio_active())) |
54 | if (kmmio_handler(regs, addr) == 1) | |
55 | return -1; | |
0fd0e3da | 56 | return 0; |
86069782 PP |
57 | } |
58 | ||
9326638c | 59 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 60 | { |
74a0b576 CH |
61 | int ret = 0; |
62 | ||
63 | /* kprobe_running() needs smp_processor_id() */ | |
f39b6f0e | 64 | if (kprobes_built_in() && !user_mode(regs)) { |
74a0b576 CH |
65 | preempt_disable(); |
66 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
67 | ret = 1; | |
68 | preempt_enable(); | |
69 | } | |
1bd858a5 | 70 | |
74a0b576 | 71 | return ret; |
33cb5243 | 72 | } |
1bd858a5 | 73 | |
1dc85be0 | 74 | /* |
2d4a7167 IM |
75 | * Prefetch quirks: |
76 | * | |
77 | * 32-bit mode: | |
78 | * | |
79 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
80 | * Check that here and ignore it. | |
1dc85be0 | 81 | * |
2d4a7167 | 82 | * 64-bit mode: |
1dc85be0 | 83 | * |
2d4a7167 IM |
84 | * Sometimes the CPU reports invalid exceptions on prefetch. |
85 | * Check that here and ignore it. | |
86 | * | |
87 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 88 | */ |
107a0367 IM |
89 | static inline int |
90 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
91 | unsigned char opcode, int *prefetch) | |
92 | { | |
93 | unsigned char instr_hi = opcode & 0xf0; | |
94 | unsigned char instr_lo = opcode & 0x0f; | |
95 | ||
96 | switch (instr_hi) { | |
97 | case 0x20: | |
98 | case 0x30: | |
99 | /* | |
100 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
101 | * In X86_64 long mode, the CPU will signal invalid | |
102 | * opcode if some of these prefixes are present so | |
103 | * X86_64 will never get here anyway | |
104 | */ | |
105 | return ((instr_lo & 7) == 0x6); | |
106 | #ifdef CONFIG_X86_64 | |
107 | case 0x40: | |
108 | /* | |
109 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
110 | * Need to figure out under what instruction mode the | |
111 | * instruction was issued. Could check the LDT for lm, | |
112 | * but for now it's good enough to assume that long | |
113 | * mode only uses well known segments or kernel. | |
114 | */ | |
318f5a2a | 115 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
116 | #endif |
117 | case 0x60: | |
118 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
119 | return (instr_lo & 0xC) == 0x4; | |
120 | case 0xF0: | |
121 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
122 | return !instr_lo || (instr_lo>>1) == 1; | |
123 | case 0x00: | |
124 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
125 | if (probe_kernel_address(instr, opcode)) | |
126 | return 0; | |
127 | ||
128 | *prefetch = (instr_lo == 0xF) && | |
129 | (opcode == 0x0D || opcode == 0x18); | |
130 | return 0; | |
131 | default: | |
132 | return 0; | |
133 | } | |
134 | } | |
135 | ||
2d4a7167 IM |
136 | static int |
137 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 138 | { |
2d4a7167 | 139 | unsigned char *max_instr; |
ab2bf0c1 | 140 | unsigned char *instr; |
33cb5243 | 141 | int prefetch = 0; |
1da177e4 | 142 | |
3085354d IM |
143 | /* |
144 | * If it was a exec (instruction fetch) fault on NX page, then | |
145 | * do not ignore the fault: | |
146 | */ | |
66c58156 | 147 | if (error_code & PF_INSTR) |
1da177e4 | 148 | return 0; |
1dc85be0 | 149 | |
107a0367 | 150 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 151 | max_instr = instr + 15; |
1da177e4 | 152 | |
d31bf07f | 153 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
154 | return 0; |
155 | ||
107a0367 | 156 | while (instr < max_instr) { |
2d4a7167 | 157 | unsigned char opcode; |
1da177e4 | 158 | |
ab2bf0c1 | 159 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 160 | break; |
1da177e4 | 161 | |
1da177e4 LT |
162 | instr++; |
163 | ||
107a0367 | 164 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 165 | break; |
1da177e4 LT |
166 | } |
167 | return prefetch; | |
168 | } | |
169 | ||
2d4a7167 IM |
170 | static void |
171 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 172 | struct task_struct *tsk, int fault) |
c4aba4a8 | 173 | { |
f672b49b | 174 | unsigned lsb = 0; |
c4aba4a8 HH |
175 | siginfo_t info; |
176 | ||
2d4a7167 IM |
177 | info.si_signo = si_signo; |
178 | info.si_errno = 0; | |
179 | info.si_code = si_code; | |
180 | info.si_addr = (void __user *)address; | |
f672b49b AK |
181 | if (fault & VM_FAULT_HWPOISON_LARGE) |
182 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
183 | if (fault & VM_FAULT_HWPOISON) | |
184 | lsb = PAGE_SHIFT; | |
185 | info.si_addr_lsb = lsb; | |
2d4a7167 | 186 | |
c4aba4a8 HH |
187 | force_sig_info(si_signo, &info, tsk); |
188 | } | |
189 | ||
f2f13a85 IM |
190 | DEFINE_SPINLOCK(pgd_lock); |
191 | LIST_HEAD(pgd_list); | |
192 | ||
193 | #ifdef CONFIG_X86_32 | |
194 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 195 | { |
f2f13a85 IM |
196 | unsigned index = pgd_index(address); |
197 | pgd_t *pgd_k; | |
198 | pud_t *pud, *pud_k; | |
199 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 200 | |
f2f13a85 IM |
201 | pgd += index; |
202 | pgd_k = init_mm.pgd + index; | |
203 | ||
204 | if (!pgd_present(*pgd_k)) | |
205 | return NULL; | |
206 | ||
207 | /* | |
208 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
209 | * and redundant with the set_pmd() on non-PAE. As would | |
210 | * set_pud. | |
211 | */ | |
212 | pud = pud_offset(pgd, address); | |
213 | pud_k = pud_offset(pgd_k, address); | |
214 | if (!pud_present(*pud_k)) | |
215 | return NULL; | |
216 | ||
217 | pmd = pmd_offset(pud, address); | |
218 | pmd_k = pmd_offset(pud_k, address); | |
219 | if (!pmd_present(*pmd_k)) | |
220 | return NULL; | |
221 | ||
b8bcfe99 | 222 | if (!pmd_present(*pmd)) |
f2f13a85 | 223 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 224 | else |
f2f13a85 | 225 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
226 | |
227 | return pmd_k; | |
228 | } | |
229 | ||
230 | void vmalloc_sync_all(void) | |
231 | { | |
232 | unsigned long address; | |
233 | ||
234 | if (SHARED_KERNEL_PMD) | |
235 | return; | |
236 | ||
237 | for (address = VMALLOC_START & PMD_MASK; | |
238 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
239 | address += PMD_SIZE) { | |
f2f13a85 IM |
240 | struct page *page; |
241 | ||
a79e53d8 | 242 | spin_lock(&pgd_lock); |
f2f13a85 | 243 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 244 | spinlock_t *pgt_lock; |
f01f7c56 | 245 | pmd_t *ret; |
617d34d9 | 246 | |
a79e53d8 | 247 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
248 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
249 | ||
250 | spin_lock(pgt_lock); | |
251 | ret = vmalloc_sync_one(page_address(page), address); | |
252 | spin_unlock(pgt_lock); | |
253 | ||
254 | if (!ret) | |
f2f13a85 IM |
255 | break; |
256 | } | |
a79e53d8 | 257 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
258 | } |
259 | } | |
260 | ||
261 | /* | |
262 | * 32-bit: | |
263 | * | |
264 | * Handle a fault on the vmalloc or module mapping area | |
265 | */ | |
9326638c | 266 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
267 | { |
268 | unsigned long pgd_paddr; | |
269 | pmd_t *pmd_k; | |
270 | pte_t *pte_k; | |
271 | ||
272 | /* Make sure we are in vmalloc area: */ | |
273 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
274 | return -1; | |
275 | ||
ebc8827f FW |
276 | WARN_ON_ONCE(in_nmi()); |
277 | ||
f2f13a85 IM |
278 | /* |
279 | * Synchronize this task's top level page-table | |
280 | * with the 'reference' page table. | |
281 | * | |
282 | * Do _not_ use "current" here. We might be inside | |
283 | * an interrupt in the middle of a task switch.. | |
284 | */ | |
285 | pgd_paddr = read_cr3(); | |
286 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
287 | if (!pmd_k) | |
288 | return -1; | |
289 | ||
f4eafd8b TK |
290 | if (pmd_huge(*pmd_k)) |
291 | return 0; | |
292 | ||
f2f13a85 IM |
293 | pte_k = pte_offset_kernel(pmd_k, address); |
294 | if (!pte_present(*pte_k)) | |
295 | return -1; | |
296 | ||
297 | return 0; | |
298 | } | |
9326638c | 299 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
300 | |
301 | /* | |
302 | * Did it hit the DOS screen memory VA from vm86 mode? | |
303 | */ | |
304 | static inline void | |
305 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
306 | struct task_struct *tsk) | |
307 | { | |
9fda6a06 | 308 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
309 | unsigned long bit; |
310 | ||
9fda6a06 | 311 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
312 | return; |
313 | ||
314 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
315 | if (bit < 32) | |
9fda6a06 BG |
316 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
317 | #endif | |
33cb5243 | 318 | } |
1da177e4 | 319 | |
087975b0 | 320 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 321 | { |
087975b0 AM |
322 | return pfn < max_low_pfn; |
323 | } | |
1156e098 | 324 | |
087975b0 AM |
325 | static void dump_pagetable(unsigned long address) |
326 | { | |
327 | pgd_t *base = __va(read_cr3()); | |
328 | pgd_t *pgd = &base[pgd_index(address)]; | |
329 | pmd_t *pmd; | |
330 | pte_t *pte; | |
2d4a7167 | 331 | |
1156e098 | 332 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
333 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
334 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
335 | goto out; | |
1156e098 | 336 | #endif |
087975b0 AM |
337 | pmd = pmd_offset(pud_offset(pgd, address), address); |
338 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
339 | |
340 | /* | |
341 | * We must not directly access the pte in the highpte | |
342 | * case if the page table is located in highmem. | |
343 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 344 | * it's allocated already: |
1156e098 | 345 | */ |
087975b0 AM |
346 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
347 | goto out; | |
1156e098 | 348 | |
087975b0 AM |
349 | pte = pte_offset_kernel(pmd, address); |
350 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
351 | out: | |
1156e098 | 352 | printk("\n"); |
f2f13a85 IM |
353 | } |
354 | ||
355 | #else /* CONFIG_X86_64: */ | |
356 | ||
357 | void vmalloc_sync_all(void) | |
358 | { | |
9661d5bc | 359 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); |
f2f13a85 IM |
360 | } |
361 | ||
362 | /* | |
363 | * 64-bit: | |
364 | * | |
365 | * Handle a fault on the vmalloc area | |
f2f13a85 | 366 | */ |
9326638c | 367 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
368 | { |
369 | pgd_t *pgd, *pgd_ref; | |
370 | pud_t *pud, *pud_ref; | |
371 | pmd_t *pmd, *pmd_ref; | |
372 | pte_t *pte, *pte_ref; | |
373 | ||
374 | /* Make sure we are in vmalloc area: */ | |
375 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
376 | return -1; | |
377 | ||
ebc8827f FW |
378 | WARN_ON_ONCE(in_nmi()); |
379 | ||
f2f13a85 IM |
380 | /* |
381 | * Copy kernel mappings over when needed. This can also | |
382 | * happen within a race in page table update. In the later | |
383 | * case just flush: | |
384 | */ | |
385 | pgd = pgd_offset(current->active_mm, address); | |
386 | pgd_ref = pgd_offset_k(address); | |
387 | if (pgd_none(*pgd_ref)) | |
388 | return -1; | |
389 | ||
1160c277 | 390 | if (pgd_none(*pgd)) { |
f2f13a85 | 391 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
392 | arch_flush_lazy_mmu_mode(); |
393 | } else { | |
f2f13a85 | 394 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 395 | } |
f2f13a85 IM |
396 | |
397 | /* | |
398 | * Below here mismatches are bugs because these lower tables | |
399 | * are shared: | |
400 | */ | |
401 | ||
402 | pud = pud_offset(pgd, address); | |
403 | pud_ref = pud_offset(pgd_ref, address); | |
404 | if (pud_none(*pud_ref)) | |
405 | return -1; | |
406 | ||
f4eafd8b | 407 | if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref)) |
f2f13a85 IM |
408 | BUG(); |
409 | ||
f4eafd8b TK |
410 | if (pud_huge(*pud)) |
411 | return 0; | |
412 | ||
f2f13a85 IM |
413 | pmd = pmd_offset(pud, address); |
414 | pmd_ref = pmd_offset(pud_ref, address); | |
415 | if (pmd_none(*pmd_ref)) | |
416 | return -1; | |
417 | ||
f4eafd8b | 418 | if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref)) |
f2f13a85 IM |
419 | BUG(); |
420 | ||
f4eafd8b TK |
421 | if (pmd_huge(*pmd)) |
422 | return 0; | |
423 | ||
f2f13a85 IM |
424 | pte_ref = pte_offset_kernel(pmd_ref, address); |
425 | if (!pte_present(*pte_ref)) | |
426 | return -1; | |
427 | ||
428 | pte = pte_offset_kernel(pmd, address); | |
429 | ||
430 | /* | |
431 | * Don't use pte_page here, because the mappings can point | |
432 | * outside mem_map, and the NUMA hash lookup cannot handle | |
433 | * that: | |
434 | */ | |
435 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
436 | BUG(); | |
437 | ||
438 | return 0; | |
439 | } | |
9326638c | 440 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 441 | |
e05139f2 | 442 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 443 | static const char errata93_warning[] = |
ad361c98 JP |
444 | KERN_ERR |
445 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
446 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
447 | "******* Please consider a BIOS update.\n" | |
448 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 449 | #endif |
f2f13a85 IM |
450 | |
451 | /* | |
452 | * No vm86 mode in 64-bit mode: | |
453 | */ | |
454 | static inline void | |
455 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
456 | struct task_struct *tsk) | |
457 | { | |
458 | } | |
459 | ||
460 | static int bad_address(void *p) | |
461 | { | |
462 | unsigned long dummy; | |
463 | ||
464 | return probe_kernel_address((unsigned long *)p, dummy); | |
465 | } | |
466 | ||
467 | static void dump_pagetable(unsigned long address) | |
468 | { | |
087975b0 AM |
469 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
470 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
471 | pud_t *pud; |
472 | pmd_t *pmd; | |
473 | pte_t *pte; | |
474 | ||
2d4a7167 IM |
475 | if (bad_address(pgd)) |
476 | goto bad; | |
477 | ||
d646bce4 | 478 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
479 | |
480 | if (!pgd_present(*pgd)) | |
481 | goto out; | |
1da177e4 | 482 | |
d2ae5b5f | 483 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
484 | if (bad_address(pud)) |
485 | goto bad; | |
486 | ||
1da177e4 | 487 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 488 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 489 | goto out; |
1da177e4 LT |
490 | |
491 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
492 | if (bad_address(pmd)) |
493 | goto bad; | |
494 | ||
1da177e4 | 495 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
496 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
497 | goto out; | |
1da177e4 LT |
498 | |
499 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
500 | if (bad_address(pte)) |
501 | goto bad; | |
502 | ||
33cb5243 | 503 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 504 | out: |
1da177e4 LT |
505 | printk("\n"); |
506 | return; | |
507 | bad: | |
508 | printk("BAD\n"); | |
8c938f9f IM |
509 | } |
510 | ||
f2f13a85 | 511 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 512 | |
2d4a7167 IM |
513 | /* |
514 | * Workaround for K8 erratum #93 & buggy BIOS. | |
515 | * | |
516 | * BIOS SMM functions are required to use a specific workaround | |
517 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
518 | * | |
519 | * A lot of BIOS that didn't get tested properly miss this. | |
520 | * | |
521 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
522 | * Try to work around it here. | |
523 | * | |
524 | * Note we only handle faults in kernel here. | |
525 | * Does nothing on 32-bit. | |
fdfe8aa8 | 526 | */ |
33cb5243 | 527 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 528 | { |
e05139f2 JB |
529 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
530 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
531 | || boot_cpu_data.x86 != 0xf) | |
532 | return 0; | |
533 | ||
65ea5b03 | 534 | if (address != regs->ip) |
1da177e4 | 535 | return 0; |
2d4a7167 | 536 | |
33cb5243 | 537 | if ((address >> 32) != 0) |
1da177e4 | 538 | return 0; |
2d4a7167 | 539 | |
1da177e4 | 540 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
541 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
542 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 543 | printk_once(errata93_warning); |
65ea5b03 | 544 | regs->ip = address; |
1da177e4 LT |
545 | return 1; |
546 | } | |
fdfe8aa8 | 547 | #endif |
1da177e4 | 548 | return 0; |
33cb5243 | 549 | } |
1da177e4 | 550 | |
35f3266f | 551 | /* |
2d4a7167 IM |
552 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
553 | * to illegal addresses >4GB. | |
554 | * | |
555 | * We catch this in the page fault handler because these addresses | |
556 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
557 | * segment in LDT is compatibility mode. |
558 | */ | |
559 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
560 | { | |
561 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 562 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
563 | return 1; |
564 | #endif | |
565 | return 0; | |
566 | } | |
567 | ||
29caf2f9 HH |
568 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
569 | { | |
570 | #ifdef CONFIG_X86_F00F_BUG | |
571 | unsigned long nr; | |
2d4a7167 | 572 | |
29caf2f9 | 573 | /* |
2d4a7167 | 574 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 575 | */ |
e2604b49 | 576 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
577 | nr = (address - idt_descr.address) >> 3; |
578 | ||
579 | if (nr == 6) { | |
580 | do_invalid_op(regs, 0); | |
581 | return 1; | |
582 | } | |
583 | } | |
584 | #endif | |
585 | return 0; | |
586 | } | |
587 | ||
8f766149 IM |
588 | static const char nx_warning[] = KERN_CRIT |
589 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
590 | static const char smep_warning[] = KERN_CRIT |
591 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 592 | |
2d4a7167 IM |
593 | static void |
594 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
595 | unsigned long address) | |
b3279c7f | 596 | { |
1156e098 HH |
597 | if (!oops_may_print()) |
598 | return; | |
599 | ||
1156e098 | 600 | if (error_code & PF_INSTR) { |
93809be8 | 601 | unsigned int level; |
426e34cc MF |
602 | pgd_t *pgd; |
603 | pte_t *pte; | |
2d4a7167 | 604 | |
426e34cc MF |
605 | pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
606 | pgd += pgd_index(address); | |
607 | ||
608 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 609 | |
8f766149 | 610 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 611 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
612 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
613 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 614 | (__read_cr4() & X86_CR4_SMEP)) |
eff50c34 | 615 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 616 | } |
1156e098 | 617 | |
19f0dda9 | 618 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 619 | if (address < PAGE_SIZE) |
19f0dda9 | 620 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 621 | else |
19f0dda9 | 622 | printk(KERN_CONT "paging request"); |
2d4a7167 | 623 | |
f294a8ce | 624 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 625 | printk(KERN_ALERT "IP:"); |
5f01c988 | 626 | printk_address(regs->ip); |
2d4a7167 | 627 | |
b3279c7f HH |
628 | dump_pagetable(address); |
629 | } | |
630 | ||
2d4a7167 IM |
631 | static noinline void |
632 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
633 | unsigned long address) | |
1da177e4 | 634 | { |
2d4a7167 IM |
635 | struct task_struct *tsk; |
636 | unsigned long flags; | |
637 | int sig; | |
638 | ||
639 | flags = oops_begin(); | |
640 | tsk = current; | |
641 | sig = SIGKILL; | |
1209140c | 642 | |
1da177e4 | 643 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 644 | tsk->comm, address); |
1da177e4 | 645 | dump_pagetable(address); |
2d4a7167 IM |
646 | |
647 | tsk->thread.cr2 = address; | |
51e7dc70 | 648 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
649 | tsk->thread.error_code = error_code; |
650 | ||
22f5991c | 651 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 652 | sig = 0; |
2d4a7167 | 653 | |
874d93d1 | 654 | oops_end(flags, regs, sig); |
1da177e4 LT |
655 | } |
656 | ||
2d4a7167 IM |
657 | static noinline void |
658 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 659 | unsigned long address, int signal, int si_code) |
92181f19 NP |
660 | { |
661 | struct task_struct *tsk = current; | |
92181f19 NP |
662 | unsigned long flags; |
663 | int sig; | |
92181f19 | 664 | |
2d4a7167 | 665 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 666 | if (fixup_exception(regs)) { |
c026b359 PZ |
667 | /* |
668 | * Any interrupt that takes a fault gets the fixup. This makes | |
669 | * the below recursive fault logic only apply to a faults from | |
670 | * task context. | |
671 | */ | |
672 | if (in_interrupt()) | |
673 | return; | |
674 | ||
675 | /* | |
676 | * Per the above we're !in_interrupt(), aka. task context. | |
677 | * | |
678 | * In this case we need to make sure we're not recursively | |
679 | * faulting through the emulate_vsyscall() logic. | |
680 | */ | |
4fc34901 | 681 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 682 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
683 | tsk->thread.error_code = error_code | PF_USER; |
684 | tsk->thread.cr2 = address; | |
685 | ||
686 | /* XXX: hwpoison faults will set the wrong code. */ | |
687 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
688 | } | |
c026b359 PZ |
689 | |
690 | /* | |
691 | * Barring that, we can do the fixup and be happy. | |
692 | */ | |
92181f19 | 693 | return; |
4fc34901 | 694 | } |
92181f19 NP |
695 | |
696 | /* | |
2d4a7167 IM |
697 | * 32-bit: |
698 | * | |
699 | * Valid to do another page fault here, because if this fault | |
700 | * had been triggered by is_prefetch fixup_exception would have | |
701 | * handled it. | |
702 | * | |
703 | * 64-bit: | |
92181f19 | 704 | * |
2d4a7167 | 705 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
706 | */ |
707 | if (is_prefetch(regs, error_code, address)) | |
708 | return; | |
709 | ||
710 | if (is_errata93(regs, address)) | |
711 | return; | |
712 | ||
713 | /* | |
714 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 715 | * terminate things with extreme prejudice: |
92181f19 | 716 | */ |
92181f19 | 717 | flags = oops_begin(); |
92181f19 NP |
718 | |
719 | show_fault_oops(regs, error_code, address); | |
720 | ||
a70857e4 | 721 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 722 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 723 | |
1cc99544 | 724 | tsk->thread.cr2 = address; |
51e7dc70 | 725 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 726 | tsk->thread.error_code = error_code; |
92181f19 | 727 | |
92181f19 NP |
728 | sig = SIGKILL; |
729 | if (__die("Oops", regs, error_code)) | |
730 | sig = 0; | |
2d4a7167 | 731 | |
92181f19 | 732 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 733 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 734 | |
92181f19 | 735 | oops_end(flags, regs, sig); |
92181f19 NP |
736 | } |
737 | ||
2d4a7167 IM |
738 | /* |
739 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
740 | * sysctl is set: | |
741 | */ | |
742 | static inline void | |
743 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
744 | unsigned long address, struct task_struct *tsk) | |
745 | { | |
746 | if (!unhandled_signal(tsk, SIGSEGV)) | |
747 | return; | |
748 | ||
749 | if (!printk_ratelimit()) | |
750 | return; | |
751 | ||
a1a08d1c | 752 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
753 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
754 | tsk->comm, task_pid_nr(tsk), address, | |
755 | (void *)regs->ip, (void *)regs->sp, error_code); | |
756 | ||
757 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
758 | ||
759 | printk(KERN_CONT "\n"); | |
760 | } | |
761 | ||
762 | static void | |
763 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
764 | unsigned long address, int si_code) | |
92181f19 NP |
765 | { |
766 | struct task_struct *tsk = current; | |
767 | ||
768 | /* User mode accesses just cause a SIGSEGV */ | |
769 | if (error_code & PF_USER) { | |
770 | /* | |
2d4a7167 | 771 | * It's possible to have interrupts off here: |
92181f19 NP |
772 | */ |
773 | local_irq_enable(); | |
774 | ||
775 | /* | |
776 | * Valid to do another page fault here because this one came | |
2d4a7167 | 777 | * from user space: |
92181f19 NP |
778 | */ |
779 | if (is_prefetch(regs, error_code, address)) | |
780 | return; | |
781 | ||
782 | if (is_errata100(regs, address)) | |
783 | return; | |
784 | ||
3ae36655 AL |
785 | #ifdef CONFIG_X86_64 |
786 | /* | |
787 | * Instruction fetch faults in the vsyscall page might need | |
788 | * emulation. | |
789 | */ | |
790 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 791 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
792 | if (emulate_vsyscall(regs, address)) |
793 | return; | |
794 | } | |
795 | #endif | |
e575a86f KC |
796 | /* Kernel addresses are always protection faults: */ |
797 | if (address >= TASK_SIZE) | |
798 | error_code |= PF_PROT; | |
3ae36655 | 799 | |
e575a86f | 800 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
801 | show_signal_msg(regs, error_code, address, tsk); |
802 | ||
2d4a7167 | 803 | tsk->thread.cr2 = address; |
e575a86f | 804 | tsk->thread.error_code = error_code; |
51e7dc70 | 805 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 806 | |
f672b49b | 807 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 808 | |
92181f19 NP |
809 | return; |
810 | } | |
811 | ||
812 | if (is_f00f_bug(regs, address)) | |
813 | return; | |
814 | ||
4fc34901 | 815 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
816 | } |
817 | ||
2d4a7167 IM |
818 | static noinline void |
819 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
820 | unsigned long address) | |
92181f19 NP |
821 | { |
822 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
823 | } | |
824 | ||
2d4a7167 IM |
825 | static void |
826 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
827 | unsigned long address, int si_code) | |
92181f19 NP |
828 | { |
829 | struct mm_struct *mm = current->mm; | |
830 | ||
831 | /* | |
832 | * Something tried to access memory that isn't in our memory map.. | |
833 | * Fix it, but check if it's kernel or user first.. | |
834 | */ | |
835 | up_read(&mm->mmap_sem); | |
836 | ||
837 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
838 | } | |
839 | ||
2d4a7167 IM |
840 | static noinline void |
841 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
842 | { |
843 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
844 | } | |
845 | ||
2d4a7167 IM |
846 | static noinline void |
847 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
848 | unsigned long address) | |
92181f19 NP |
849 | { |
850 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
851 | } | |
852 | ||
2d4a7167 | 853 | static void |
a6e04aa9 AK |
854 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
855 | unsigned int fault) | |
92181f19 NP |
856 | { |
857 | struct task_struct *tsk = current; | |
a6e04aa9 | 858 | int code = BUS_ADRERR; |
92181f19 | 859 | |
2d4a7167 | 860 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 861 | if (!(error_code & PF_USER)) { |
4fc34901 | 862 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
863 | return; |
864 | } | |
2d4a7167 | 865 | |
cd1b68f0 | 866 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
867 | if (is_prefetch(regs, error_code, address)) |
868 | return; | |
2d4a7167 IM |
869 | |
870 | tsk->thread.cr2 = address; | |
871 | tsk->thread.error_code = error_code; | |
51e7dc70 | 872 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 873 | |
a6e04aa9 | 874 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 875 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
876 | printk(KERN_ERR |
877 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
878 | tsk->comm, tsk->pid, address); | |
879 | code = BUS_MCEERR_AR; | |
880 | } | |
881 | #endif | |
f672b49b | 882 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
883 | } |
884 | ||
3a13c4d7 | 885 | static noinline void |
2d4a7167 IM |
886 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
887 | unsigned long address, unsigned int fault) | |
92181f19 | 888 | { |
3a13c4d7 | 889 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
3a13c4d7 JW |
890 | no_context(regs, error_code, address, 0, 0); |
891 | return; | |
b80ef10e | 892 | } |
b80ef10e | 893 | |
2d4a7167 | 894 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
895 | /* Kernel mode? Handle exceptions or die: */ |
896 | if (!(error_code & PF_USER)) { | |
4fc34901 AL |
897 | no_context(regs, error_code, address, |
898 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 899 | return; |
f8626854 AV |
900 | } |
901 | ||
c2d23f91 DR |
902 | /* |
903 | * We ran out of memory, call the OOM killer, and return the | |
904 | * userspace (which will retry the fault, or kill us if we got | |
905 | * oom-killed): | |
906 | */ | |
907 | pagefault_out_of_memory(); | |
2d4a7167 | 908 | } else { |
f672b49b AK |
909 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
910 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 911 | do_sigbus(regs, error_code, address, fault); |
33692f27 LT |
912 | else if (fault & VM_FAULT_SIGSEGV) |
913 | bad_area_nosemaphore(regs, error_code, address); | |
2d4a7167 IM |
914 | else |
915 | BUG(); | |
916 | } | |
92181f19 NP |
917 | } |
918 | ||
d8b57bb7 TG |
919 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
920 | { | |
921 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
922 | return 0; | |
2d4a7167 | 923 | |
d8b57bb7 TG |
924 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
925 | return 0; | |
926 | ||
927 | return 1; | |
928 | } | |
929 | ||
5b727a3b | 930 | /* |
2d4a7167 IM |
931 | * Handle a spurious fault caused by a stale TLB entry. |
932 | * | |
933 | * This allows us to lazily refresh the TLB when increasing the | |
934 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
935 | * eagerly is very expensive since that implies doing a full | |
936 | * cross-processor TLB flush, even if no stale TLB entries exist | |
937 | * on other processors. | |
938 | * | |
31668511 DV |
939 | * Spurious faults may only occur if the TLB contains an entry with |
940 | * fewer permission than the page table entry. Non-present (P = 0) | |
941 | * and reserved bit (R = 1) faults are never spurious. | |
942 | * | |
5b727a3b JF |
943 | * There are no security implications to leaving a stale TLB when |
944 | * increasing the permissions on a page. | |
31668511 DV |
945 | * |
946 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
947 | * | |
948 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
949 | * (Optional Invalidation). | |
5b727a3b | 950 | */ |
9326638c | 951 | static noinline int |
2d4a7167 | 952 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
953 | { |
954 | pgd_t *pgd; | |
955 | pud_t *pud; | |
956 | pmd_t *pmd; | |
957 | pte_t *pte; | |
3c3e5694 | 958 | int ret; |
5b727a3b | 959 | |
31668511 DV |
960 | /* |
961 | * Only writes to RO or instruction fetches from NX may cause | |
962 | * spurious faults. | |
963 | * | |
964 | * These could be from user or supervisor accesses but the TLB | |
965 | * is only lazily flushed after a kernel mapping protection | |
966 | * change, so user accesses are not expected to cause spurious | |
967 | * faults. | |
968 | */ | |
969 | if (error_code != (PF_WRITE | PF_PROT) | |
970 | && error_code != (PF_INSTR | PF_PROT)) | |
5b727a3b JF |
971 | return 0; |
972 | ||
973 | pgd = init_mm.pgd + pgd_index(address); | |
974 | if (!pgd_present(*pgd)) | |
975 | return 0; | |
976 | ||
977 | pud = pud_offset(pgd, address); | |
978 | if (!pud_present(*pud)) | |
979 | return 0; | |
980 | ||
d8b57bb7 TG |
981 | if (pud_large(*pud)) |
982 | return spurious_fault_check(error_code, (pte_t *) pud); | |
983 | ||
5b727a3b JF |
984 | pmd = pmd_offset(pud, address); |
985 | if (!pmd_present(*pmd)) | |
986 | return 0; | |
987 | ||
d8b57bb7 TG |
988 | if (pmd_large(*pmd)) |
989 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
990 | ||
5b727a3b | 991 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 992 | if (!pte_present(*pte)) |
5b727a3b JF |
993 | return 0; |
994 | ||
3c3e5694 SR |
995 | ret = spurious_fault_check(error_code, pte); |
996 | if (!ret) | |
997 | return 0; | |
998 | ||
999 | /* | |
2d4a7167 IM |
1000 | * Make sure we have permissions in PMD. |
1001 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
1002 | */ |
1003 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
1004 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 1005 | |
3c3e5694 | 1006 | return ret; |
5b727a3b | 1007 | } |
9326638c | 1008 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 1009 | |
abd4f750 | 1010 | int show_unhandled_signals = 1; |
1da177e4 | 1011 | |
2d4a7167 | 1012 | static inline int |
68da336a | 1013 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1014 | { |
68da336a | 1015 | if (error_code & PF_WRITE) { |
2d4a7167 | 1016 | /* write, present and write, not present: */ |
92181f19 NP |
1017 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1018 | return 1; | |
2d4a7167 | 1019 | return 0; |
92181f19 NP |
1020 | } |
1021 | ||
2d4a7167 IM |
1022 | /* read, present: */ |
1023 | if (unlikely(error_code & PF_PROT)) | |
1024 | return 1; | |
1025 | ||
1026 | /* read, not present: */ | |
1027 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1028 | return 1; | |
1029 | ||
92181f19 NP |
1030 | return 0; |
1031 | } | |
1032 | ||
0973a06c HS |
1033 | static int fault_in_kernel_space(unsigned long address) |
1034 | { | |
d9517346 | 1035 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1036 | } |
1037 | ||
40d3cd66 PA |
1038 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1039 | { | |
4640c7ee PA |
1040 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1041 | return false; | |
1042 | ||
1043 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1044 | return false; | |
1045 | ||
40d3cd66 PA |
1046 | if (error_code & PF_USER) |
1047 | return false; | |
1048 | ||
f39b6f0e | 1049 | if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) |
40d3cd66 PA |
1050 | return false; |
1051 | ||
1052 | return true; | |
1053 | } | |
1054 | ||
1da177e4 LT |
1055 | /* |
1056 | * This routine handles page faults. It determines the address, | |
1057 | * and the problem, and then passes it off to one of the appropriate | |
1058 | * routines. | |
d4078e23 PZ |
1059 | * |
1060 | * This function must have noinline because both callers | |
1061 | * {,trace_}do_page_fault() have notrace on. Having this an actual function | |
1062 | * guarantees there's a function trace entry. | |
1da177e4 | 1063 | */ |
9326638c | 1064 | static noinline void |
0ac09f9f JO |
1065 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1066 | unsigned long address) | |
1da177e4 | 1067 | { |
2d4a7167 | 1068 | struct vm_area_struct *vma; |
1da177e4 LT |
1069 | struct task_struct *tsk; |
1070 | struct mm_struct *mm; | |
26178ec1 | 1071 | int fault, major = 0; |
759496ba | 1072 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1073 | |
a9ba9a3b AV |
1074 | tsk = current; |
1075 | mm = tsk->mm; | |
2d4a7167 | 1076 | |
f8561296 VN |
1077 | /* |
1078 | * Detect and handle instructions that would cause a page fault for | |
1079 | * both a tracked kernel page and a userspace page. | |
1080 | */ | |
1081 | if (kmemcheck_active(regs)) | |
1082 | kmemcheck_hide(regs); | |
5dfaf90f | 1083 | prefetchw(&mm->mmap_sem); |
f8561296 | 1084 | |
0fd0e3da | 1085 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1086 | return; |
1da177e4 LT |
1087 | |
1088 | /* | |
1089 | * We fault-in kernel-space virtual memory on-demand. The | |
1090 | * 'reference' page table is init_mm.pgd. | |
1091 | * | |
1092 | * NOTE! We MUST NOT take any locks for this case. We may | |
1093 | * be in an interrupt or a critical region, and should | |
1094 | * only copy the information from the master page table, | |
1095 | * nothing more. | |
1096 | * | |
1097 | * This verifies that the fault happens in kernel space | |
1098 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1099 | * protection error (error_code & 9) == 0. |
1da177e4 | 1100 | */ |
0973a06c | 1101 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1102 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1103 | if (vmalloc_fault(address) >= 0) | |
1104 | return; | |
1105 | ||
1106 | if (kmemcheck_fault(regs, address, error_code)) | |
1107 | return; | |
1108 | } | |
5b727a3b | 1109 | |
2d4a7167 | 1110 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1111 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1112 | return; |
1113 | ||
2d4a7167 | 1114 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1115 | if (kprobes_fault(regs)) |
9be260a6 | 1116 | return; |
f8c2ee22 HH |
1117 | /* |
1118 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1119 | * fault we could otherwise deadlock: |
f8c2ee22 | 1120 | */ |
92181f19 | 1121 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1122 | |
92181f19 | 1123 | return; |
f8c2ee22 HH |
1124 | } |
1125 | ||
2d4a7167 | 1126 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1127 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1128 | return; |
8c914cb7 | 1129 | |
66c58156 | 1130 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1131 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1132 | |
4640c7ee PA |
1133 | if (unlikely(smap_violation(error_code, regs))) { |
1134 | bad_area_nosemaphore(regs, error_code, address); | |
1135 | return; | |
40d3cd66 PA |
1136 | } |
1137 | ||
1da177e4 | 1138 | /* |
2d4a7167 | 1139 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1140 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1141 | */ |
70ffdb93 | 1142 | if (unlikely(faulthandler_disabled() || !mm)) { |
92181f19 NP |
1143 | bad_area_nosemaphore(regs, error_code, address); |
1144 | return; | |
1145 | } | |
1da177e4 | 1146 | |
e00b12e6 PZ |
1147 | /* |
1148 | * It's safe to allow irq's after cr2 has been saved and the | |
1149 | * vmalloc fault has been handled. | |
1150 | * | |
1151 | * User-mode registers count as a user access even for any | |
1152 | * potential system fault or CPU buglet: | |
1153 | */ | |
f39b6f0e | 1154 | if (user_mode(regs)) { |
e00b12e6 PZ |
1155 | local_irq_enable(); |
1156 | error_code |= PF_USER; | |
1157 | flags |= FAULT_FLAG_USER; | |
1158 | } else { | |
1159 | if (regs->flags & X86_EFLAGS_IF) | |
1160 | local_irq_enable(); | |
1161 | } | |
1162 | ||
1163 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1164 | ||
759496ba JW |
1165 | if (error_code & PF_WRITE) |
1166 | flags |= FAULT_FLAG_WRITE; | |
1167 | ||
3a1dfe6e IM |
1168 | /* |
1169 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1170 | * addresses in user space. All other faults represent errors in |
1171 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1172 | * case of an erroneous fault occurring in a code path which already | |
1173 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1174 | * against the address space. Luckily the kernel only validly | |
1175 | * references user space from well defined areas of code, which are | |
1176 | * listed in the exceptions table. | |
1da177e4 LT |
1177 | * |
1178 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1179 | * the source reference check when there is a possibility of a |
1180 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1181 | * validate the source. If this is invalid we can skip the address | |
1182 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1183 | */ |
92181f19 | 1184 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1185 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1186 | !search_exception_tables(regs->ip)) { |
1187 | bad_area_nosemaphore(regs, error_code, address); | |
1188 | return; | |
1189 | } | |
d065bd81 | 1190 | retry: |
1da177e4 | 1191 | down_read(&mm->mmap_sem); |
01006074 PZ |
1192 | } else { |
1193 | /* | |
2d4a7167 IM |
1194 | * The above down_read_trylock() might have succeeded in |
1195 | * which case we'll have missed the might_sleep() from | |
1196 | * down_read(): | |
01006074 PZ |
1197 | */ |
1198 | might_sleep(); | |
1da177e4 LT |
1199 | } |
1200 | ||
1201 | vma = find_vma(mm, address); | |
92181f19 NP |
1202 | if (unlikely(!vma)) { |
1203 | bad_area(regs, error_code, address); | |
1204 | return; | |
1205 | } | |
1206 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1207 | goto good_area; |
92181f19 NP |
1208 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1209 | bad_area(regs, error_code, address); | |
1210 | return; | |
1211 | } | |
33cb5243 | 1212 | if (error_code & PF_USER) { |
6f4d368e HH |
1213 | /* |
1214 | * Accessing the stack below %sp is always a bug. | |
1215 | * The large cushion allows instructions like enter | |
2d4a7167 | 1216 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1217 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1218 | */ |
92181f19 NP |
1219 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1220 | bad_area(regs, error_code, address); | |
1221 | return; | |
1222 | } | |
1da177e4 | 1223 | } |
92181f19 NP |
1224 | if (unlikely(expand_stack(vma, address))) { |
1225 | bad_area(regs, error_code, address); | |
1226 | return; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Ok, we have a good vm_area for this memory access, so | |
1231 | * we can handle it.. | |
1232 | */ | |
1da177e4 | 1233 | good_area: |
68da336a | 1234 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1235 | bad_area_access_error(regs, error_code, address); |
1236 | return; | |
1da177e4 LT |
1237 | } |
1238 | ||
1239 | /* | |
1240 | * If for any reason at all we couldn't handle the fault, | |
1241 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1242 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1243 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
1da177e4 | 1244 | */ |
d065bd81 | 1245 | fault = handle_mm_fault(mm, vma, address, flags); |
26178ec1 | 1246 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1247 | |
3a13c4d7 | 1248 | /* |
26178ec1 LT |
1249 | * If we need to retry the mmap_sem has already been released, |
1250 | * and if there is a fatal signal pending there is no guarantee | |
1251 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1252 | */ |
26178ec1 LT |
1253 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1254 | /* Retry at most once */ | |
1255 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1256 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1257 | flags |= FAULT_FLAG_TRIED; | |
1258 | if (!fatal_signal_pending(tsk)) | |
1259 | goto retry; | |
1260 | } | |
1261 | ||
1262 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1263 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1264 | return; |
1265 | ||
1266 | /* Not returning to user mode? Handle exceptions or die: */ | |
1267 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1268 | return; |
26178ec1 | 1269 | } |
3a13c4d7 | 1270 | |
26178ec1 | 1271 | up_read(&mm->mmap_sem); |
3a13c4d7 JW |
1272 | if (unlikely(fault & VM_FAULT_ERROR)) { |
1273 | mm_fault_error(regs, error_code, address, fault); | |
1274 | return; | |
37b23e05 KM |
1275 | } |
1276 | ||
d065bd81 | 1277 | /* |
26178ec1 LT |
1278 | * Major/minor page fault accounting. If any of the events |
1279 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1280 | */ |
26178ec1 LT |
1281 | if (major) { |
1282 | tsk->maj_flt++; | |
1283 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1284 | } else { | |
1285 | tsk->min_flt++; | |
1286 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1287 | } |
d729ab35 | 1288 | |
8c938f9f | 1289 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1290 | } |
9326638c | 1291 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1292 | |
9326638c | 1293 | dotraplinkage void notrace |
6ba3c97a FW |
1294 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1295 | { | |
d4078e23 | 1296 | unsigned long address = read_cr2(); /* Get the faulting address */ |
6c1e0256 | 1297 | enum ctx_state prev_state; |
d4078e23 PZ |
1298 | |
1299 | /* | |
1300 | * We must have this function tagged with __kprobes, notrace and call | |
1301 | * read_cr2() before calling anything else. To avoid calling any kind | |
1302 | * of tracing machinery before we've observed the CR2 value. | |
1303 | * | |
1304 | * exception_{enter,exit}() contain all sorts of tracepoints. | |
1305 | */ | |
6c1e0256 FW |
1306 | |
1307 | prev_state = exception_enter(); | |
0ac09f9f | 1308 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1309 | exception_exit(prev_state); |
6ba3c97a | 1310 | } |
9326638c | 1311 | NOKPROBE_SYMBOL(do_page_fault); |
25c74b10 | 1312 | |
d4078e23 | 1313 | #ifdef CONFIG_TRACING |
9326638c MH |
1314 | static nokprobe_inline void |
1315 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1316 | unsigned long error_code) | |
d34603b0 SA |
1317 | { |
1318 | if (user_mode(regs)) | |
d4078e23 | 1319 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1320 | else |
d4078e23 | 1321 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1322 | } |
1323 | ||
9326638c | 1324 | dotraplinkage void notrace |
25c74b10 SA |
1325 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1326 | { | |
0ac09f9f JO |
1327 | /* |
1328 | * The exception_enter and tracepoint processing could | |
1329 | * trigger another page faults (user space callchain | |
1330 | * reading) and destroy the original cr2 value, so read | |
1331 | * the faulting address now. | |
1332 | */ | |
1333 | unsigned long address = read_cr2(); | |
d4078e23 | 1334 | enum ctx_state prev_state; |
25c74b10 SA |
1335 | |
1336 | prev_state = exception_enter(); | |
d4078e23 | 1337 | trace_page_fault_entries(address, regs, error_code); |
0ac09f9f | 1338 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1339 | exception_exit(prev_state); |
1340 | } | |
9326638c | 1341 | NOKPROBE_SYMBOL(trace_do_page_fault); |
d4078e23 | 1342 | #endif /* CONFIG_TRACING */ |