]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 | 3 | * Copyright (C) 1995 Linus Torvalds |
2d4a7167 | 4 | * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. |
f8eeb2e6 | 5 | * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar |
1da177e4 | 6 | */ |
a2bcd473 | 7 | #include <linux/sched.h> /* test_thread_flag(), ... */ |
68db0cf1 | 8 | #include <linux/sched/task_stack.h> /* task_stack_*(), ... */ |
a2bcd473 | 9 | #include <linux/kdebug.h> /* oops_begin/end, ... */ |
4cdf8dbe | 10 | #include <linux/extable.h> /* search_exception_tables */ |
a2bcd473 | 11 | #include <linux/bootmem.h> /* max_low_pfn */ |
9326638c | 12 | #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ |
a2bcd473 | 13 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ |
cdd6c482 | 14 | #include <linux/perf_event.h> /* perf_sw_event */ |
f672b49b | 15 | #include <linux/hugetlb.h> /* hstate_index_to_shift */ |
268bb0ce | 16 | #include <linux/prefetch.h> /* prefetchw */ |
56dd9470 | 17 | #include <linux/context_tracking.h> /* exception_enter(), ... */ |
70ffdb93 | 18 | #include <linux/uaccess.h> /* faulthandler_disabled() */ |
2d4a7167 | 19 | |
019132ff | 20 | #include <asm/cpufeature.h> /* boot_cpu_has, ... */ |
a2bcd473 IM |
21 | #include <asm/traps.h> /* dotraplinkage, ... */ |
22 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f40c3300 AL |
23 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
24 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
ba3e127e | 25 | #include <asm/vm86.h> /* struct vm86 */ |
019132ff | 26 | #include <asm/mmu_context.h> /* vma_pkey() */ |
1da177e4 | 27 | |
d34603b0 SA |
28 | #define CREATE_TRACE_POINTS |
29 | #include <asm/trace/exceptions.h> | |
30 | ||
b814d41f | 31 | /* |
b319eed0 IM |
32 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
33 | * handled by mmiotrace: | |
b814d41f | 34 | */ |
9326638c | 35 | static nokprobe_inline int |
62c9295f | 36 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 37 | { |
0fd0e3da PP |
38 | if (unlikely(is_kmmio_active())) |
39 | if (kmmio_handler(regs, addr) == 1) | |
40 | return -1; | |
0fd0e3da | 41 | return 0; |
86069782 PP |
42 | } |
43 | ||
9326638c | 44 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 45 | { |
74a0b576 CH |
46 | int ret = 0; |
47 | ||
48 | /* kprobe_running() needs smp_processor_id() */ | |
f39b6f0e | 49 | if (kprobes_built_in() && !user_mode(regs)) { |
74a0b576 CH |
50 | preempt_disable(); |
51 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
52 | ret = 1; | |
53 | preempt_enable(); | |
54 | } | |
1bd858a5 | 55 | |
74a0b576 | 56 | return ret; |
33cb5243 | 57 | } |
1bd858a5 | 58 | |
1dc85be0 | 59 | /* |
2d4a7167 IM |
60 | * Prefetch quirks: |
61 | * | |
62 | * 32-bit mode: | |
63 | * | |
64 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
65 | * Check that here and ignore it. | |
1dc85be0 | 66 | * |
2d4a7167 | 67 | * 64-bit mode: |
1dc85be0 | 68 | * |
2d4a7167 IM |
69 | * Sometimes the CPU reports invalid exceptions on prefetch. |
70 | * Check that here and ignore it. | |
71 | * | |
72 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 73 | */ |
107a0367 IM |
74 | static inline int |
75 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
76 | unsigned char opcode, int *prefetch) | |
77 | { | |
78 | unsigned char instr_hi = opcode & 0xf0; | |
79 | unsigned char instr_lo = opcode & 0x0f; | |
80 | ||
81 | switch (instr_hi) { | |
82 | case 0x20: | |
83 | case 0x30: | |
84 | /* | |
85 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
86 | * In X86_64 long mode, the CPU will signal invalid | |
87 | * opcode if some of these prefixes are present so | |
88 | * X86_64 will never get here anyway | |
89 | */ | |
90 | return ((instr_lo & 7) == 0x6); | |
91 | #ifdef CONFIG_X86_64 | |
92 | case 0x40: | |
93 | /* | |
94 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
95 | * Need to figure out under what instruction mode the | |
96 | * instruction was issued. Could check the LDT for lm, | |
97 | * but for now it's good enough to assume that long | |
98 | * mode only uses well known segments or kernel. | |
99 | */ | |
318f5a2a | 100 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
101 | #endif |
102 | case 0x60: | |
103 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
104 | return (instr_lo & 0xC) == 0x4; | |
105 | case 0xF0: | |
106 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
107 | return !instr_lo || (instr_lo>>1) == 1; | |
108 | case 0x00: | |
109 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
110 | if (probe_kernel_address(instr, opcode)) | |
111 | return 0; | |
112 | ||
113 | *prefetch = (instr_lo == 0xF) && | |
114 | (opcode == 0x0D || opcode == 0x18); | |
115 | return 0; | |
116 | default: | |
117 | return 0; | |
118 | } | |
119 | } | |
120 | ||
2d4a7167 IM |
121 | static int |
122 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 123 | { |
2d4a7167 | 124 | unsigned char *max_instr; |
ab2bf0c1 | 125 | unsigned char *instr; |
33cb5243 | 126 | int prefetch = 0; |
1da177e4 | 127 | |
3085354d IM |
128 | /* |
129 | * If it was a exec (instruction fetch) fault on NX page, then | |
130 | * do not ignore the fault: | |
131 | */ | |
1067f030 | 132 | if (error_code & X86_PF_INSTR) |
1da177e4 | 133 | return 0; |
1dc85be0 | 134 | |
107a0367 | 135 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 136 | max_instr = instr + 15; |
1da177e4 | 137 | |
d31bf07f | 138 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
139 | return 0; |
140 | ||
107a0367 | 141 | while (instr < max_instr) { |
2d4a7167 | 142 | unsigned char opcode; |
1da177e4 | 143 | |
ab2bf0c1 | 144 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 145 | break; |
1da177e4 | 146 | |
1da177e4 LT |
147 | instr++; |
148 | ||
107a0367 | 149 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 150 | break; |
1da177e4 LT |
151 | } |
152 | return prefetch; | |
153 | } | |
154 | ||
019132ff DH |
155 | /* |
156 | * A protection key fault means that the PKRU value did not allow | |
157 | * access to some PTE. Userspace can figure out what PKRU was | |
158 | * from the XSAVE state, and this function fills out a field in | |
159 | * siginfo so userspace can discover which protection key was set | |
160 | * on the PTE. | |
161 | * | |
1067f030 | 162 | * If we get here, we know that the hardware signaled a X86_PF_PK |
019132ff DH |
163 | * fault and that there was a VMA once we got in the fault |
164 | * handler. It does *not* guarantee that the VMA we find here | |
165 | * was the one that we faulted on. | |
166 | * | |
167 | * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); | |
168 | * 2. T1 : set PKRU to deny access to pkey=4, touches page | |
169 | * 3. T1 : faults... | |
170 | * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); | |
171 | * 5. T1 : enters fault handler, takes mmap_sem, etc... | |
172 | * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really | |
173 | * faulted on a pte with its pkey=4. | |
174 | */ | |
a3c4fb7c | 175 | static void fill_sig_info_pkey(int si_code, siginfo_t *info, u32 *pkey) |
019132ff DH |
176 | { |
177 | /* This is effectively an #ifdef */ | |
178 | if (!boot_cpu_has(X86_FEATURE_OSPKE)) | |
179 | return; | |
180 | ||
181 | /* Fault not from Protection Keys: nothing to do */ | |
182 | if (si_code != SEGV_PKUERR) | |
183 | return; | |
184 | /* | |
185 | * force_sig_info_fault() is called from a number of | |
186 | * contexts, some of which have a VMA and some of which | |
1067f030 | 187 | * do not. The X86_PF_PK handing happens after we have a |
019132ff DH |
188 | * valid VMA, so we should never reach this without a |
189 | * valid VMA. | |
190 | */ | |
a3c4fb7c | 191 | if (!pkey) { |
019132ff DH |
192 | WARN_ONCE(1, "PKU fault with no VMA passed in"); |
193 | info->si_pkey = 0; | |
194 | return; | |
195 | } | |
196 | /* | |
197 | * si_pkey should be thought of as a strong hint, but not | |
198 | * absolutely guranteed to be 100% accurate because of | |
199 | * the race explained above. | |
200 | */ | |
a3c4fb7c | 201 | info->si_pkey = *pkey; |
019132ff DH |
202 | } |
203 | ||
2d4a7167 IM |
204 | static void |
205 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
a3c4fb7c | 206 | struct task_struct *tsk, u32 *pkey, int fault) |
c4aba4a8 | 207 | { |
f672b49b | 208 | unsigned lsb = 0; |
c4aba4a8 HH |
209 | siginfo_t info; |
210 | ||
2d4a7167 IM |
211 | info.si_signo = si_signo; |
212 | info.si_errno = 0; | |
213 | info.si_code = si_code; | |
214 | info.si_addr = (void __user *)address; | |
f672b49b AK |
215 | if (fault & VM_FAULT_HWPOISON_LARGE) |
216 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
217 | if (fault & VM_FAULT_HWPOISON) | |
218 | lsb = PAGE_SHIFT; | |
219 | info.si_addr_lsb = lsb; | |
2d4a7167 | 220 | |
a3c4fb7c | 221 | fill_sig_info_pkey(si_code, &info, pkey); |
019132ff | 222 | |
c4aba4a8 HH |
223 | force_sig_info(si_signo, &info, tsk); |
224 | } | |
225 | ||
f2f13a85 IM |
226 | DEFINE_SPINLOCK(pgd_lock); |
227 | LIST_HEAD(pgd_list); | |
228 | ||
229 | #ifdef CONFIG_X86_32 | |
230 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 231 | { |
f2f13a85 IM |
232 | unsigned index = pgd_index(address); |
233 | pgd_t *pgd_k; | |
e0c4f675 | 234 | p4d_t *p4d, *p4d_k; |
f2f13a85 IM |
235 | pud_t *pud, *pud_k; |
236 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 237 | |
f2f13a85 IM |
238 | pgd += index; |
239 | pgd_k = init_mm.pgd + index; | |
240 | ||
241 | if (!pgd_present(*pgd_k)) | |
242 | return NULL; | |
243 | ||
244 | /* | |
245 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
246 | * and redundant with the set_pmd() on non-PAE. As would | |
e0c4f675 | 247 | * set_p4d/set_pud. |
f2f13a85 | 248 | */ |
e0c4f675 KS |
249 | p4d = p4d_offset(pgd, address); |
250 | p4d_k = p4d_offset(pgd_k, address); | |
251 | if (!p4d_present(*p4d_k)) | |
252 | return NULL; | |
253 | ||
254 | pud = pud_offset(p4d, address); | |
255 | pud_k = pud_offset(p4d_k, address); | |
f2f13a85 IM |
256 | if (!pud_present(*pud_k)) |
257 | return NULL; | |
258 | ||
259 | pmd = pmd_offset(pud, address); | |
260 | pmd_k = pmd_offset(pud_k, address); | |
261 | if (!pmd_present(*pmd_k)) | |
262 | return NULL; | |
263 | ||
b8bcfe99 | 264 | if (!pmd_present(*pmd)) |
f2f13a85 | 265 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 266 | else |
f2f13a85 | 267 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
268 | |
269 | return pmd_k; | |
270 | } | |
271 | ||
272 | void vmalloc_sync_all(void) | |
273 | { | |
274 | unsigned long address; | |
275 | ||
276 | if (SHARED_KERNEL_PMD) | |
277 | return; | |
278 | ||
279 | for (address = VMALLOC_START & PMD_MASK; | |
dc4fac84 | 280 | address >= TASK_SIZE_MAX && address < FIXADDR_TOP; |
f2f13a85 | 281 | address += PMD_SIZE) { |
f2f13a85 IM |
282 | struct page *page; |
283 | ||
a79e53d8 | 284 | spin_lock(&pgd_lock); |
f2f13a85 | 285 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 286 | spinlock_t *pgt_lock; |
f01f7c56 | 287 | pmd_t *ret; |
617d34d9 | 288 | |
a79e53d8 | 289 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
290 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
291 | ||
292 | spin_lock(pgt_lock); | |
293 | ret = vmalloc_sync_one(page_address(page), address); | |
294 | spin_unlock(pgt_lock); | |
295 | ||
296 | if (!ret) | |
f2f13a85 IM |
297 | break; |
298 | } | |
a79e53d8 | 299 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
300 | } |
301 | } | |
302 | ||
303 | /* | |
304 | * 32-bit: | |
305 | * | |
306 | * Handle a fault on the vmalloc or module mapping area | |
307 | */ | |
9326638c | 308 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
309 | { |
310 | unsigned long pgd_paddr; | |
311 | pmd_t *pmd_k; | |
312 | pte_t *pte_k; | |
313 | ||
314 | /* Make sure we are in vmalloc area: */ | |
315 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
316 | return -1; | |
317 | ||
ebc8827f FW |
318 | WARN_ON_ONCE(in_nmi()); |
319 | ||
f2f13a85 IM |
320 | /* |
321 | * Synchronize this task's top level page-table | |
322 | * with the 'reference' page table. | |
323 | * | |
324 | * Do _not_ use "current" here. We might be inside | |
325 | * an interrupt in the middle of a task switch.. | |
326 | */ | |
6c690ee1 | 327 | pgd_paddr = read_cr3_pa(); |
f2f13a85 IM |
328 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); |
329 | if (!pmd_k) | |
330 | return -1; | |
331 | ||
f4eafd8b TK |
332 | if (pmd_huge(*pmd_k)) |
333 | return 0; | |
334 | ||
f2f13a85 IM |
335 | pte_k = pte_offset_kernel(pmd_k, address); |
336 | if (!pte_present(*pte_k)) | |
337 | return -1; | |
338 | ||
339 | return 0; | |
340 | } | |
9326638c | 341 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
342 | |
343 | /* | |
344 | * Did it hit the DOS screen memory VA from vm86 mode? | |
345 | */ | |
346 | static inline void | |
347 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
348 | struct task_struct *tsk) | |
349 | { | |
9fda6a06 | 350 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
351 | unsigned long bit; |
352 | ||
9fda6a06 | 353 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
354 | return; |
355 | ||
356 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
357 | if (bit < 32) | |
9fda6a06 BG |
358 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
359 | #endif | |
33cb5243 | 360 | } |
1da177e4 | 361 | |
087975b0 | 362 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 363 | { |
087975b0 AM |
364 | return pfn < max_low_pfn; |
365 | } | |
1156e098 | 366 | |
087975b0 AM |
367 | static void dump_pagetable(unsigned long address) |
368 | { | |
6c690ee1 | 369 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 370 | pgd_t *pgd = &base[pgd_index(address)]; |
e0c4f675 KS |
371 | p4d_t *p4d; |
372 | pud_t *pud; | |
087975b0 AM |
373 | pmd_t *pmd; |
374 | pte_t *pte; | |
2d4a7167 | 375 | |
1156e098 | 376 | #ifdef CONFIG_X86_PAE |
39e48d9b | 377 | pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); |
087975b0 AM |
378 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) |
379 | goto out; | |
39e48d9b JB |
380 | #define pr_pde pr_cont |
381 | #else | |
382 | #define pr_pde pr_info | |
1156e098 | 383 | #endif |
e0c4f675 KS |
384 | p4d = p4d_offset(pgd, address); |
385 | pud = pud_offset(p4d, address); | |
386 | pmd = pmd_offset(pud, address); | |
39e48d9b JB |
387 | pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); |
388 | #undef pr_pde | |
1156e098 HH |
389 | |
390 | /* | |
391 | * We must not directly access the pte in the highpte | |
392 | * case if the page table is located in highmem. | |
393 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 394 | * it's allocated already: |
1156e098 | 395 | */ |
087975b0 AM |
396 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
397 | goto out; | |
1156e098 | 398 | |
087975b0 | 399 | pte = pte_offset_kernel(pmd, address); |
39e48d9b | 400 | pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); |
087975b0 | 401 | out: |
39e48d9b | 402 | pr_cont("\n"); |
f2f13a85 IM |
403 | } |
404 | ||
405 | #else /* CONFIG_X86_64: */ | |
406 | ||
407 | void vmalloc_sync_all(void) | |
408 | { | |
5372e155 | 409 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
410 | } |
411 | ||
412 | /* | |
413 | * 64-bit: | |
414 | * | |
415 | * Handle a fault on the vmalloc area | |
f2f13a85 | 416 | */ |
9326638c | 417 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
418 | { |
419 | pgd_t *pgd, *pgd_ref; | |
b50858ce | 420 | p4d_t *p4d, *p4d_ref; |
f2f13a85 IM |
421 | pud_t *pud, *pud_ref; |
422 | pmd_t *pmd, *pmd_ref; | |
423 | pte_t *pte, *pte_ref; | |
424 | ||
425 | /* Make sure we are in vmalloc area: */ | |
426 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
427 | return -1; | |
428 | ||
ebc8827f FW |
429 | WARN_ON_ONCE(in_nmi()); |
430 | ||
f2f13a85 IM |
431 | /* |
432 | * Copy kernel mappings over when needed. This can also | |
433 | * happen within a race in page table update. In the later | |
434 | * case just flush: | |
435 | */ | |
6c690ee1 | 436 | pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); |
f2f13a85 IM |
437 | pgd_ref = pgd_offset_k(address); |
438 | if (pgd_none(*pgd_ref)) | |
439 | return -1; | |
440 | ||
1160c277 | 441 | if (pgd_none(*pgd)) { |
f2f13a85 | 442 | set_pgd(pgd, *pgd_ref); |
1160c277 | 443 | arch_flush_lazy_mmu_mode(); |
b50858ce KS |
444 | } else if (CONFIG_PGTABLE_LEVELS > 4) { |
445 | /* | |
446 | * With folded p4d, pgd_none() is always false, so the pgd may | |
447 | * point to an empty page table entry and pgd_page_vaddr() | |
448 | * will return garbage. | |
449 | * | |
450 | * We will do the correct sanity check on the p4d level. | |
451 | */ | |
f2f13a85 | 452 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 453 | } |
f2f13a85 | 454 | |
b50858ce KS |
455 | /* With 4-level paging, copying happens on the p4d level. */ |
456 | p4d = p4d_offset(pgd, address); | |
457 | p4d_ref = p4d_offset(pgd_ref, address); | |
458 | if (p4d_none(*p4d_ref)) | |
459 | return -1; | |
460 | ||
461 | if (p4d_none(*p4d)) { | |
462 | set_p4d(p4d, *p4d_ref); | |
463 | arch_flush_lazy_mmu_mode(); | |
464 | } else { | |
465 | BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref)); | |
466 | } | |
467 | ||
f2f13a85 IM |
468 | /* |
469 | * Below here mismatches are bugs because these lower tables | |
470 | * are shared: | |
471 | */ | |
472 | ||
b50858ce KS |
473 | pud = pud_offset(p4d, address); |
474 | pud_ref = pud_offset(p4d_ref, address); | |
f2f13a85 IM |
475 | if (pud_none(*pud_ref)) |
476 | return -1; | |
477 | ||
f4eafd8b | 478 | if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref)) |
f2f13a85 IM |
479 | BUG(); |
480 | ||
f4eafd8b TK |
481 | if (pud_huge(*pud)) |
482 | return 0; | |
483 | ||
f2f13a85 IM |
484 | pmd = pmd_offset(pud, address); |
485 | pmd_ref = pmd_offset(pud_ref, address); | |
486 | if (pmd_none(*pmd_ref)) | |
487 | return -1; | |
488 | ||
f4eafd8b | 489 | if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref)) |
f2f13a85 IM |
490 | BUG(); |
491 | ||
f4eafd8b TK |
492 | if (pmd_huge(*pmd)) |
493 | return 0; | |
494 | ||
f2f13a85 IM |
495 | pte_ref = pte_offset_kernel(pmd_ref, address); |
496 | if (!pte_present(*pte_ref)) | |
497 | return -1; | |
498 | ||
499 | pte = pte_offset_kernel(pmd, address); | |
500 | ||
501 | /* | |
502 | * Don't use pte_page here, because the mappings can point | |
503 | * outside mem_map, and the NUMA hash lookup cannot handle | |
504 | * that: | |
505 | */ | |
506 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
507 | BUG(); | |
508 | ||
509 | return 0; | |
510 | } | |
9326638c | 511 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 512 | |
e05139f2 | 513 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 514 | static const char errata93_warning[] = |
ad361c98 JP |
515 | KERN_ERR |
516 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
517 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
518 | "******* Please consider a BIOS update.\n" | |
519 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 520 | #endif |
f2f13a85 IM |
521 | |
522 | /* | |
523 | * No vm86 mode in 64-bit mode: | |
524 | */ | |
525 | static inline void | |
526 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
527 | struct task_struct *tsk) | |
528 | { | |
529 | } | |
530 | ||
531 | static int bad_address(void *p) | |
532 | { | |
533 | unsigned long dummy; | |
534 | ||
535 | return probe_kernel_address((unsigned long *)p, dummy); | |
536 | } | |
537 | ||
538 | static void dump_pagetable(unsigned long address) | |
539 | { | |
6c690ee1 | 540 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 541 | pgd_t *pgd = base + pgd_index(address); |
e0c4f675 | 542 | p4d_t *p4d; |
1da177e4 LT |
543 | pud_t *pud; |
544 | pmd_t *pmd; | |
545 | pte_t *pte; | |
546 | ||
2d4a7167 IM |
547 | if (bad_address(pgd)) |
548 | goto bad; | |
549 | ||
39e48d9b | 550 | pr_info("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
551 | |
552 | if (!pgd_present(*pgd)) | |
553 | goto out; | |
1da177e4 | 554 | |
e0c4f675 KS |
555 | p4d = p4d_offset(pgd, address); |
556 | if (bad_address(p4d)) | |
557 | goto bad; | |
558 | ||
39e48d9b | 559 | pr_cont("P4D %lx ", p4d_val(*p4d)); |
e0c4f675 KS |
560 | if (!p4d_present(*p4d) || p4d_large(*p4d)) |
561 | goto out; | |
562 | ||
563 | pud = pud_offset(p4d, address); | |
2d4a7167 IM |
564 | if (bad_address(pud)) |
565 | goto bad; | |
566 | ||
39e48d9b | 567 | pr_cont("PUD %lx ", pud_val(*pud)); |
b5360222 | 568 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 569 | goto out; |
1da177e4 LT |
570 | |
571 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
572 | if (bad_address(pmd)) |
573 | goto bad; | |
574 | ||
39e48d9b | 575 | pr_cont("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
576 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
577 | goto out; | |
1da177e4 LT |
578 | |
579 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
580 | if (bad_address(pte)) |
581 | goto bad; | |
582 | ||
39e48d9b | 583 | pr_cont("PTE %lx", pte_val(*pte)); |
2d4a7167 | 584 | out: |
39e48d9b | 585 | pr_cont("\n"); |
1da177e4 LT |
586 | return; |
587 | bad: | |
39e48d9b | 588 | pr_info("BAD\n"); |
8c938f9f IM |
589 | } |
590 | ||
f2f13a85 | 591 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 592 | |
2d4a7167 IM |
593 | /* |
594 | * Workaround for K8 erratum #93 & buggy BIOS. | |
595 | * | |
596 | * BIOS SMM functions are required to use a specific workaround | |
597 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
598 | * | |
599 | * A lot of BIOS that didn't get tested properly miss this. | |
600 | * | |
601 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
602 | * Try to work around it here. | |
603 | * | |
604 | * Note we only handle faults in kernel here. | |
605 | * Does nothing on 32-bit. | |
fdfe8aa8 | 606 | */ |
33cb5243 | 607 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 608 | { |
e05139f2 JB |
609 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
610 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
611 | || boot_cpu_data.x86 != 0xf) | |
612 | return 0; | |
613 | ||
65ea5b03 | 614 | if (address != regs->ip) |
1da177e4 | 615 | return 0; |
2d4a7167 | 616 | |
33cb5243 | 617 | if ((address >> 32) != 0) |
1da177e4 | 618 | return 0; |
2d4a7167 | 619 | |
1da177e4 | 620 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
621 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
622 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 623 | printk_once(errata93_warning); |
65ea5b03 | 624 | regs->ip = address; |
1da177e4 LT |
625 | return 1; |
626 | } | |
fdfe8aa8 | 627 | #endif |
1da177e4 | 628 | return 0; |
33cb5243 | 629 | } |
1da177e4 | 630 | |
35f3266f | 631 | /* |
2d4a7167 IM |
632 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
633 | * to illegal addresses >4GB. | |
634 | * | |
635 | * We catch this in the page fault handler because these addresses | |
636 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
637 | * segment in LDT is compatibility mode. |
638 | */ | |
639 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
640 | { | |
641 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 642 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
643 | return 1; |
644 | #endif | |
645 | return 0; | |
646 | } | |
647 | ||
29caf2f9 HH |
648 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
649 | { | |
650 | #ifdef CONFIG_X86_F00F_BUG | |
651 | unsigned long nr; | |
2d4a7167 | 652 | |
29caf2f9 | 653 | /* |
2d4a7167 | 654 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 655 | */ |
e2604b49 | 656 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
657 | nr = (address - idt_descr.address) >> 3; |
658 | ||
659 | if (nr == 6) { | |
660 | do_invalid_op(regs, 0); | |
661 | return 1; | |
662 | } | |
663 | } | |
664 | #endif | |
665 | return 0; | |
666 | } | |
667 | ||
8f766149 IM |
668 | static const char nx_warning[] = KERN_CRIT |
669 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
670 | static const char smep_warning[] = KERN_CRIT |
671 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 672 | |
2d4a7167 IM |
673 | static void |
674 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
675 | unsigned long address) | |
b3279c7f | 676 | { |
1156e098 HH |
677 | if (!oops_may_print()) |
678 | return; | |
679 | ||
1067f030 | 680 | if (error_code & X86_PF_INSTR) { |
93809be8 | 681 | unsigned int level; |
426e34cc MF |
682 | pgd_t *pgd; |
683 | pte_t *pte; | |
2d4a7167 | 684 | |
6c690ee1 | 685 | pgd = __va(read_cr3_pa()); |
426e34cc MF |
686 | pgd += pgd_index(address); |
687 | ||
688 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 689 | |
8f766149 | 690 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 691 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
692 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
693 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 694 | (__read_cr4() & X86_CR4_SMEP)) |
eff50c34 | 695 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 696 | } |
1156e098 | 697 | |
19f0dda9 | 698 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 699 | if (address < PAGE_SIZE) |
19f0dda9 | 700 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 701 | else |
19f0dda9 | 702 | printk(KERN_CONT "paging request"); |
2d4a7167 | 703 | |
328b4ed9 | 704 | printk(KERN_CONT " at %px\n", (void *) address); |
bb5e5ce5 | 705 | printk(KERN_ALERT "IP: %pS\n", (void *)regs->ip); |
2d4a7167 | 706 | |
b3279c7f HH |
707 | dump_pagetable(address); |
708 | } | |
709 | ||
2d4a7167 IM |
710 | static noinline void |
711 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
712 | unsigned long address) | |
1da177e4 | 713 | { |
2d4a7167 IM |
714 | struct task_struct *tsk; |
715 | unsigned long flags; | |
716 | int sig; | |
717 | ||
718 | flags = oops_begin(); | |
719 | tsk = current; | |
720 | sig = SIGKILL; | |
1209140c | 721 | |
1da177e4 | 722 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 723 | tsk->comm, address); |
1da177e4 | 724 | dump_pagetable(address); |
2d4a7167 IM |
725 | |
726 | tsk->thread.cr2 = address; | |
51e7dc70 | 727 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
728 | tsk->thread.error_code = error_code; |
729 | ||
22f5991c | 730 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 731 | sig = 0; |
2d4a7167 | 732 | |
874d93d1 | 733 | oops_end(flags, regs, sig); |
1da177e4 LT |
734 | } |
735 | ||
2d4a7167 IM |
736 | static noinline void |
737 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 738 | unsigned long address, int signal, int si_code) |
92181f19 NP |
739 | { |
740 | struct task_struct *tsk = current; | |
92181f19 NP |
741 | unsigned long flags; |
742 | int sig; | |
92181f19 | 743 | |
2d4a7167 | 744 | /* Are we prepared to handle this kernel fault? */ |
548acf19 | 745 | if (fixup_exception(regs, X86_TRAP_PF)) { |
c026b359 PZ |
746 | /* |
747 | * Any interrupt that takes a fault gets the fixup. This makes | |
748 | * the below recursive fault logic only apply to a faults from | |
749 | * task context. | |
750 | */ | |
751 | if (in_interrupt()) | |
752 | return; | |
753 | ||
754 | /* | |
755 | * Per the above we're !in_interrupt(), aka. task context. | |
756 | * | |
757 | * In this case we need to make sure we're not recursively | |
758 | * faulting through the emulate_vsyscall() logic. | |
759 | */ | |
2a53ccbc | 760 | if (current->thread.sig_on_uaccess_err && signal) { |
51e7dc70 | 761 | tsk->thread.trap_nr = X86_TRAP_PF; |
1067f030 | 762 | tsk->thread.error_code = error_code | X86_PF_USER; |
4fc34901 AL |
763 | tsk->thread.cr2 = address; |
764 | ||
765 | /* XXX: hwpoison faults will set the wrong code. */ | |
7b2d0dba | 766 | force_sig_info_fault(signal, si_code, address, |
a3c4fb7c | 767 | tsk, NULL, 0); |
4fc34901 | 768 | } |
c026b359 PZ |
769 | |
770 | /* | |
771 | * Barring that, we can do the fixup and be happy. | |
772 | */ | |
92181f19 | 773 | return; |
4fc34901 | 774 | } |
92181f19 | 775 | |
6271cfdf AL |
776 | #ifdef CONFIG_VMAP_STACK |
777 | /* | |
778 | * Stack overflow? During boot, we can fault near the initial | |
779 | * stack in the direct map, but that's not an overflow -- check | |
780 | * that we're in vmalloc space to avoid this. | |
781 | */ | |
782 | if (is_vmalloc_addr((void *)address) && | |
783 | (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || | |
784 | address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { | |
6271cfdf AL |
785 | unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *); |
786 | /* | |
787 | * We're likely to be running with very little stack space | |
788 | * left. It's plausible that we'd hit this condition but | |
789 | * double-fault even before we get this far, in which case | |
790 | * we're fine: the double-fault handler will deal with it. | |
791 | * | |
792 | * We don't want to make it all the way into the oops code | |
793 | * and then double-fault, though, because we're likely to | |
794 | * break the console driver and lose most of the stack dump. | |
795 | */ | |
796 | asm volatile ("movq %[stack], %%rsp\n\t" | |
797 | "call handle_stack_overflow\n\t" | |
798 | "1: jmp 1b" | |
f5caf621 | 799 | : ASM_CALL_CONSTRAINT |
6271cfdf AL |
800 | : "D" ("kernel stack overflow (page fault)"), |
801 | "S" (regs), "d" (address), | |
802 | [stack] "rm" (stack)); | |
803 | unreachable(); | |
804 | } | |
805 | #endif | |
806 | ||
92181f19 | 807 | /* |
2d4a7167 IM |
808 | * 32-bit: |
809 | * | |
810 | * Valid to do another page fault here, because if this fault | |
811 | * had been triggered by is_prefetch fixup_exception would have | |
812 | * handled it. | |
813 | * | |
814 | * 64-bit: | |
92181f19 | 815 | * |
2d4a7167 | 816 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
817 | */ |
818 | if (is_prefetch(regs, error_code, address)) | |
819 | return; | |
820 | ||
821 | if (is_errata93(regs, address)) | |
822 | return; | |
823 | ||
824 | /* | |
825 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 826 | * terminate things with extreme prejudice: |
92181f19 | 827 | */ |
92181f19 | 828 | flags = oops_begin(); |
92181f19 NP |
829 | |
830 | show_fault_oops(regs, error_code, address); | |
831 | ||
a70857e4 | 832 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 833 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 834 | |
1cc99544 | 835 | tsk->thread.cr2 = address; |
51e7dc70 | 836 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 837 | tsk->thread.error_code = error_code; |
92181f19 | 838 | |
92181f19 NP |
839 | sig = SIGKILL; |
840 | if (__die("Oops", regs, error_code)) | |
841 | sig = 0; | |
2d4a7167 | 842 | |
92181f19 | 843 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 844 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 845 | |
92181f19 | 846 | oops_end(flags, regs, sig); |
92181f19 NP |
847 | } |
848 | ||
2d4a7167 IM |
849 | /* |
850 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
851 | * sysctl is set: | |
852 | */ | |
853 | static inline void | |
854 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
855 | unsigned long address, struct task_struct *tsk) | |
856 | { | |
857 | if (!unhandled_signal(tsk, SIGSEGV)) | |
858 | return; | |
859 | ||
860 | if (!printk_ratelimit()) | |
861 | return; | |
862 | ||
10a7e9d8 | 863 | printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", |
2d4a7167 IM |
864 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
865 | tsk->comm, task_pid_nr(tsk), address, | |
866 | (void *)regs->ip, (void *)regs->sp, error_code); | |
867 | ||
868 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
869 | ||
870 | printk(KERN_CONT "\n"); | |
871 | } | |
872 | ||
873 | static void | |
874 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
a3c4fb7c | 875 | unsigned long address, u32 *pkey, int si_code) |
92181f19 NP |
876 | { |
877 | struct task_struct *tsk = current; | |
878 | ||
879 | /* User mode accesses just cause a SIGSEGV */ | |
1067f030 | 880 | if (error_code & X86_PF_USER) { |
92181f19 | 881 | /* |
2d4a7167 | 882 | * It's possible to have interrupts off here: |
92181f19 NP |
883 | */ |
884 | local_irq_enable(); | |
885 | ||
886 | /* | |
887 | * Valid to do another page fault here because this one came | |
2d4a7167 | 888 | * from user space: |
92181f19 NP |
889 | */ |
890 | if (is_prefetch(regs, error_code, address)) | |
891 | return; | |
892 | ||
893 | if (is_errata100(regs, address)) | |
894 | return; | |
895 | ||
3ae36655 AL |
896 | #ifdef CONFIG_X86_64 |
897 | /* | |
898 | * Instruction fetch faults in the vsyscall page might need | |
899 | * emulation. | |
900 | */ | |
1067f030 | 901 | if (unlikely((error_code & X86_PF_INSTR) && |
f40c3300 | 902 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
903 | if (emulate_vsyscall(regs, address)) |
904 | return; | |
905 | } | |
906 | #endif | |
dc4fac84 AL |
907 | |
908 | /* | |
909 | * To avoid leaking information about the kernel page table | |
910 | * layout, pretend that user-mode accesses to kernel addresses | |
911 | * are always protection faults. | |
912 | */ | |
913 | if (address >= TASK_SIZE_MAX) | |
1067f030 | 914 | error_code |= X86_PF_PROT; |
3ae36655 | 915 | |
e575a86f | 916 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
917 | show_signal_msg(regs, error_code, address, tsk); |
918 | ||
2d4a7167 | 919 | tsk->thread.cr2 = address; |
e575a86f | 920 | tsk->thread.error_code = error_code; |
51e7dc70 | 921 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 922 | |
a3c4fb7c | 923 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0); |
2d4a7167 | 924 | |
92181f19 NP |
925 | return; |
926 | } | |
927 | ||
928 | if (is_f00f_bug(regs, address)) | |
929 | return; | |
930 | ||
4fc34901 | 931 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
932 | } |
933 | ||
2d4a7167 IM |
934 | static noinline void |
935 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
a3c4fb7c | 936 | unsigned long address, u32 *pkey) |
92181f19 | 937 | { |
a3c4fb7c | 938 | __bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR); |
92181f19 NP |
939 | } |
940 | ||
2d4a7167 IM |
941 | static void |
942 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 943 | unsigned long address, struct vm_area_struct *vma, int si_code) |
92181f19 NP |
944 | { |
945 | struct mm_struct *mm = current->mm; | |
a3c4fb7c LD |
946 | u32 pkey; |
947 | ||
948 | if (vma) | |
949 | pkey = vma_pkey(vma); | |
92181f19 NP |
950 | |
951 | /* | |
952 | * Something tried to access memory that isn't in our memory map.. | |
953 | * Fix it, but check if it's kernel or user first.. | |
954 | */ | |
955 | up_read(&mm->mmap_sem); | |
956 | ||
a3c4fb7c LD |
957 | __bad_area_nosemaphore(regs, error_code, address, |
958 | (vma) ? &pkey : NULL, si_code); | |
92181f19 NP |
959 | } |
960 | ||
2d4a7167 IM |
961 | static noinline void |
962 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 | 963 | { |
7b2d0dba | 964 | __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); |
92181f19 NP |
965 | } |
966 | ||
33a709b2 DH |
967 | static inline bool bad_area_access_from_pkeys(unsigned long error_code, |
968 | struct vm_area_struct *vma) | |
969 | { | |
07f146f5 DH |
970 | /* This code is always called on the current mm */ |
971 | bool foreign = false; | |
972 | ||
33a709b2 DH |
973 | if (!boot_cpu_has(X86_FEATURE_OSPKE)) |
974 | return false; | |
1067f030 | 975 | if (error_code & X86_PF_PK) |
33a709b2 | 976 | return true; |
07f146f5 | 977 | /* this checks permission keys on the VMA: */ |
1067f030 RN |
978 | if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), |
979 | (error_code & X86_PF_INSTR), foreign)) | |
07f146f5 | 980 | return true; |
33a709b2 | 981 | return false; |
92181f19 NP |
982 | } |
983 | ||
2d4a7167 IM |
984 | static noinline void |
985 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 986 | unsigned long address, struct vm_area_struct *vma) |
92181f19 | 987 | { |
019132ff DH |
988 | /* |
989 | * This OSPKE check is not strictly necessary at runtime. | |
990 | * But, doing it this way allows compiler optimizations | |
991 | * if pkeys are compiled out. | |
992 | */ | |
33a709b2 | 993 | if (bad_area_access_from_pkeys(error_code, vma)) |
019132ff DH |
994 | __bad_area(regs, error_code, address, vma, SEGV_PKUERR); |
995 | else | |
996 | __bad_area(regs, error_code, address, vma, SEGV_ACCERR); | |
92181f19 NP |
997 | } |
998 | ||
2d4a7167 | 999 | static void |
a6e04aa9 | 1000 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
a3c4fb7c | 1001 | u32 *pkey, unsigned int fault) |
92181f19 NP |
1002 | { |
1003 | struct task_struct *tsk = current; | |
a6e04aa9 | 1004 | int code = BUS_ADRERR; |
92181f19 | 1005 | |
2d4a7167 | 1006 | /* Kernel mode? Handle exceptions or die: */ |
1067f030 | 1007 | if (!(error_code & X86_PF_USER)) { |
4fc34901 | 1008 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
1009 | return; |
1010 | } | |
2d4a7167 | 1011 | |
cd1b68f0 | 1012 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
1013 | if (is_prefetch(regs, error_code, address)) |
1014 | return; | |
2d4a7167 IM |
1015 | |
1016 | tsk->thread.cr2 = address; | |
1017 | tsk->thread.error_code = error_code; | |
51e7dc70 | 1018 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 1019 | |
a6e04aa9 | 1020 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 1021 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
1022 | printk(KERN_ERR |
1023 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
1024 | tsk->comm, tsk->pid, address); | |
1025 | code = BUS_MCEERR_AR; | |
1026 | } | |
1027 | #endif | |
a3c4fb7c | 1028 | force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault); |
92181f19 NP |
1029 | } |
1030 | ||
3a13c4d7 | 1031 | static noinline void |
2d4a7167 | 1032 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
a3c4fb7c | 1033 | unsigned long address, u32 *pkey, unsigned int fault) |
92181f19 | 1034 | { |
1067f030 | 1035 | if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { |
3a13c4d7 JW |
1036 | no_context(regs, error_code, address, 0, 0); |
1037 | return; | |
b80ef10e | 1038 | } |
b80ef10e | 1039 | |
2d4a7167 | 1040 | if (fault & VM_FAULT_OOM) { |
f8626854 | 1041 | /* Kernel mode? Handle exceptions or die: */ |
1067f030 | 1042 | if (!(error_code & X86_PF_USER)) { |
4fc34901 AL |
1043 | no_context(regs, error_code, address, |
1044 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 1045 | return; |
f8626854 AV |
1046 | } |
1047 | ||
c2d23f91 DR |
1048 | /* |
1049 | * We ran out of memory, call the OOM killer, and return the | |
1050 | * userspace (which will retry the fault, or kill us if we got | |
1051 | * oom-killed): | |
1052 | */ | |
1053 | pagefault_out_of_memory(); | |
2d4a7167 | 1054 | } else { |
f672b49b AK |
1055 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
1056 | VM_FAULT_HWPOISON_LARGE)) | |
a3c4fb7c | 1057 | do_sigbus(regs, error_code, address, pkey, fault); |
33692f27 | 1058 | else if (fault & VM_FAULT_SIGSEGV) |
a3c4fb7c | 1059 | bad_area_nosemaphore(regs, error_code, address, pkey); |
2d4a7167 IM |
1060 | else |
1061 | BUG(); | |
1062 | } | |
92181f19 NP |
1063 | } |
1064 | ||
d8b57bb7 TG |
1065 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
1066 | { | |
1067f030 | 1067 | if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) |
d8b57bb7 | 1068 | return 0; |
2d4a7167 | 1069 | |
1067f030 | 1070 | if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) |
d8b57bb7 | 1071 | return 0; |
b3ecd515 DH |
1072 | /* |
1073 | * Note: We do not do lazy flushing on protection key | |
1067f030 | 1074 | * changes, so no spurious fault will ever set X86_PF_PK. |
b3ecd515 | 1075 | */ |
1067f030 | 1076 | if ((error_code & X86_PF_PK)) |
b3ecd515 | 1077 | return 1; |
d8b57bb7 TG |
1078 | |
1079 | return 1; | |
1080 | } | |
1081 | ||
5b727a3b | 1082 | /* |
2d4a7167 IM |
1083 | * Handle a spurious fault caused by a stale TLB entry. |
1084 | * | |
1085 | * This allows us to lazily refresh the TLB when increasing the | |
1086 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
1087 | * eagerly is very expensive since that implies doing a full | |
1088 | * cross-processor TLB flush, even if no stale TLB entries exist | |
1089 | * on other processors. | |
1090 | * | |
31668511 DV |
1091 | * Spurious faults may only occur if the TLB contains an entry with |
1092 | * fewer permission than the page table entry. Non-present (P = 0) | |
1093 | * and reserved bit (R = 1) faults are never spurious. | |
1094 | * | |
5b727a3b JF |
1095 | * There are no security implications to leaving a stale TLB when |
1096 | * increasing the permissions on a page. | |
31668511 DV |
1097 | * |
1098 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
1099 | * | |
1100 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
1101 | * (Optional Invalidation). | |
5b727a3b | 1102 | */ |
9326638c | 1103 | static noinline int |
2d4a7167 | 1104 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
1105 | { |
1106 | pgd_t *pgd; | |
e0c4f675 | 1107 | p4d_t *p4d; |
5b727a3b JF |
1108 | pud_t *pud; |
1109 | pmd_t *pmd; | |
1110 | pte_t *pte; | |
3c3e5694 | 1111 | int ret; |
5b727a3b | 1112 | |
31668511 DV |
1113 | /* |
1114 | * Only writes to RO or instruction fetches from NX may cause | |
1115 | * spurious faults. | |
1116 | * | |
1117 | * These could be from user or supervisor accesses but the TLB | |
1118 | * is only lazily flushed after a kernel mapping protection | |
1119 | * change, so user accesses are not expected to cause spurious | |
1120 | * faults. | |
1121 | */ | |
1067f030 RN |
1122 | if (error_code != (X86_PF_WRITE | X86_PF_PROT) && |
1123 | error_code != (X86_PF_INSTR | X86_PF_PROT)) | |
5b727a3b JF |
1124 | return 0; |
1125 | ||
1126 | pgd = init_mm.pgd + pgd_index(address); | |
1127 | if (!pgd_present(*pgd)) | |
1128 | return 0; | |
1129 | ||
e0c4f675 KS |
1130 | p4d = p4d_offset(pgd, address); |
1131 | if (!p4d_present(*p4d)) | |
1132 | return 0; | |
1133 | ||
1134 | if (p4d_large(*p4d)) | |
1135 | return spurious_fault_check(error_code, (pte_t *) p4d); | |
1136 | ||
1137 | pud = pud_offset(p4d, address); | |
5b727a3b JF |
1138 | if (!pud_present(*pud)) |
1139 | return 0; | |
1140 | ||
d8b57bb7 TG |
1141 | if (pud_large(*pud)) |
1142 | return spurious_fault_check(error_code, (pte_t *) pud); | |
1143 | ||
5b727a3b JF |
1144 | pmd = pmd_offset(pud, address); |
1145 | if (!pmd_present(*pmd)) | |
1146 | return 0; | |
1147 | ||
d8b57bb7 TG |
1148 | if (pmd_large(*pmd)) |
1149 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
1150 | ||
5b727a3b | 1151 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 1152 | if (!pte_present(*pte)) |
5b727a3b JF |
1153 | return 0; |
1154 | ||
3c3e5694 SR |
1155 | ret = spurious_fault_check(error_code, pte); |
1156 | if (!ret) | |
1157 | return 0; | |
1158 | ||
1159 | /* | |
2d4a7167 IM |
1160 | * Make sure we have permissions in PMD. |
1161 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
1162 | */ |
1163 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
1164 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 1165 | |
3c3e5694 | 1166 | return ret; |
5b727a3b | 1167 | } |
9326638c | 1168 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 1169 | |
abd4f750 | 1170 | int show_unhandled_signals = 1; |
1da177e4 | 1171 | |
2d4a7167 | 1172 | static inline int |
68da336a | 1173 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1174 | { |
07f146f5 DH |
1175 | /* This is only called for the current mm, so: */ |
1176 | bool foreign = false; | |
e8c6226d DH |
1177 | |
1178 | /* | |
1179 | * Read or write was blocked by protection keys. This is | |
1180 | * always an unconditional error and can never result in | |
1181 | * a follow-up action to resolve the fault, like a COW. | |
1182 | */ | |
1067f030 | 1183 | if (error_code & X86_PF_PK) |
e8c6226d DH |
1184 | return 1; |
1185 | ||
07f146f5 DH |
1186 | /* |
1187 | * Make sure to check the VMA so that we do not perform | |
1067f030 | 1188 | * faults just to hit a X86_PF_PK as soon as we fill in a |
07f146f5 DH |
1189 | * page. |
1190 | */ | |
1067f030 RN |
1191 | if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), |
1192 | (error_code & X86_PF_INSTR), foreign)) | |
07f146f5 | 1193 | return 1; |
33a709b2 | 1194 | |
1067f030 | 1195 | if (error_code & X86_PF_WRITE) { |
2d4a7167 | 1196 | /* write, present and write, not present: */ |
92181f19 NP |
1197 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1198 | return 1; | |
2d4a7167 | 1199 | return 0; |
92181f19 NP |
1200 | } |
1201 | ||
2d4a7167 | 1202 | /* read, present: */ |
1067f030 | 1203 | if (unlikely(error_code & X86_PF_PROT)) |
2d4a7167 IM |
1204 | return 1; |
1205 | ||
1206 | /* read, not present: */ | |
1207 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1208 | return 1; | |
1209 | ||
92181f19 NP |
1210 | return 0; |
1211 | } | |
1212 | ||
0973a06c HS |
1213 | static int fault_in_kernel_space(unsigned long address) |
1214 | { | |
d9517346 | 1215 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1216 | } |
1217 | ||
40d3cd66 PA |
1218 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1219 | { | |
4640c7ee PA |
1220 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1221 | return false; | |
1222 | ||
1223 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1224 | return false; | |
1225 | ||
1067f030 | 1226 | if (error_code & X86_PF_USER) |
40d3cd66 PA |
1227 | return false; |
1228 | ||
f39b6f0e | 1229 | if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) |
40d3cd66 PA |
1230 | return false; |
1231 | ||
1232 | return true; | |
1233 | } | |
1234 | ||
1da177e4 LT |
1235 | /* |
1236 | * This routine handles page faults. It determines the address, | |
1237 | * and the problem, and then passes it off to one of the appropriate | |
1238 | * routines. | |
1da177e4 | 1239 | */ |
9326638c | 1240 | static noinline void |
0ac09f9f JO |
1241 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1242 | unsigned long address) | |
1da177e4 | 1243 | { |
2d4a7167 | 1244 | struct vm_area_struct *vma; |
1da177e4 LT |
1245 | struct task_struct *tsk; |
1246 | struct mm_struct *mm; | |
26178ec1 | 1247 | int fault, major = 0; |
759496ba | 1248 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
a3c4fb7c | 1249 | u32 pkey; |
1da177e4 | 1250 | |
a9ba9a3b AV |
1251 | tsk = current; |
1252 | mm = tsk->mm; | |
2d4a7167 | 1253 | |
f8561296 VN |
1254 | /* |
1255 | * Detect and handle instructions that would cause a page fault for | |
1256 | * both a tracked kernel page and a userspace page. | |
1257 | */ | |
5dfaf90f | 1258 | prefetchw(&mm->mmap_sem); |
f8561296 | 1259 | |
0fd0e3da | 1260 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1261 | return; |
1da177e4 LT |
1262 | |
1263 | /* | |
1264 | * We fault-in kernel-space virtual memory on-demand. The | |
1265 | * 'reference' page table is init_mm.pgd. | |
1266 | * | |
1267 | * NOTE! We MUST NOT take any locks for this case. We may | |
1268 | * be in an interrupt or a critical region, and should | |
1269 | * only copy the information from the master page table, | |
1270 | * nothing more. | |
1271 | * | |
1272 | * This verifies that the fault happens in kernel space | |
1273 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1274 | * protection error (error_code & 9) == 0. |
1da177e4 | 1275 | */ |
0973a06c | 1276 | if (unlikely(fault_in_kernel_space(address))) { |
1067f030 | 1277 | if (!(error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { |
f8561296 VN |
1278 | if (vmalloc_fault(address) >= 0) |
1279 | return; | |
f8561296 | 1280 | } |
5b727a3b | 1281 | |
2d4a7167 | 1282 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1283 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1284 | return; |
1285 | ||
2d4a7167 | 1286 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1287 | if (kprobes_fault(regs)) |
9be260a6 | 1288 | return; |
f8c2ee22 HH |
1289 | /* |
1290 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1291 | * fault we could otherwise deadlock: |
f8c2ee22 | 1292 | */ |
7b2d0dba | 1293 | bad_area_nosemaphore(regs, error_code, address, NULL); |
2d4a7167 | 1294 | |
92181f19 | 1295 | return; |
f8c2ee22 HH |
1296 | } |
1297 | ||
2d4a7167 | 1298 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1299 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1300 | return; |
8c914cb7 | 1301 | |
1067f030 | 1302 | if (unlikely(error_code & X86_PF_RSVD)) |
92181f19 | 1303 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1304 | |
4640c7ee | 1305 | if (unlikely(smap_violation(error_code, regs))) { |
7b2d0dba | 1306 | bad_area_nosemaphore(regs, error_code, address, NULL); |
4640c7ee | 1307 | return; |
40d3cd66 PA |
1308 | } |
1309 | ||
1da177e4 | 1310 | /* |
2d4a7167 | 1311 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1312 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1313 | */ |
70ffdb93 | 1314 | if (unlikely(faulthandler_disabled() || !mm)) { |
7b2d0dba | 1315 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1316 | return; |
1317 | } | |
1da177e4 | 1318 | |
e00b12e6 PZ |
1319 | /* |
1320 | * It's safe to allow irq's after cr2 has been saved and the | |
1321 | * vmalloc fault has been handled. | |
1322 | * | |
1323 | * User-mode registers count as a user access even for any | |
1324 | * potential system fault or CPU buglet: | |
1325 | */ | |
f39b6f0e | 1326 | if (user_mode(regs)) { |
e00b12e6 | 1327 | local_irq_enable(); |
1067f030 | 1328 | error_code |= X86_PF_USER; |
e00b12e6 PZ |
1329 | flags |= FAULT_FLAG_USER; |
1330 | } else { | |
1331 | if (regs->flags & X86_EFLAGS_IF) | |
1332 | local_irq_enable(); | |
1333 | } | |
1334 | ||
1335 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1336 | ||
1067f030 | 1337 | if (error_code & X86_PF_WRITE) |
759496ba | 1338 | flags |= FAULT_FLAG_WRITE; |
1067f030 | 1339 | if (error_code & X86_PF_INSTR) |
d61172b4 | 1340 | flags |= FAULT_FLAG_INSTRUCTION; |
759496ba | 1341 | |
3a1dfe6e IM |
1342 | /* |
1343 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1344 | * addresses in user space. All other faults represent errors in |
1345 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1346 | * case of an erroneous fault occurring in a code path which already | |
1347 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1348 | * against the address space. Luckily the kernel only validly | |
1349 | * references user space from well defined areas of code, which are | |
1350 | * listed in the exceptions table. | |
1da177e4 LT |
1351 | * |
1352 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1353 | * the source reference check when there is a possibility of a |
1354 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1355 | * validate the source. If this is invalid we can skip the address | |
1356 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1357 | */ |
92181f19 | 1358 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
1067f030 | 1359 | if (!(error_code & X86_PF_USER) && |
92181f19 | 1360 | !search_exception_tables(regs->ip)) { |
7b2d0dba | 1361 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1362 | return; |
1363 | } | |
d065bd81 | 1364 | retry: |
1da177e4 | 1365 | down_read(&mm->mmap_sem); |
01006074 PZ |
1366 | } else { |
1367 | /* | |
2d4a7167 IM |
1368 | * The above down_read_trylock() might have succeeded in |
1369 | * which case we'll have missed the might_sleep() from | |
1370 | * down_read(): | |
01006074 PZ |
1371 | */ |
1372 | might_sleep(); | |
1da177e4 LT |
1373 | } |
1374 | ||
1375 | vma = find_vma(mm, address); | |
92181f19 NP |
1376 | if (unlikely(!vma)) { |
1377 | bad_area(regs, error_code, address); | |
1378 | return; | |
1379 | } | |
1380 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1381 | goto good_area; |
92181f19 NP |
1382 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1383 | bad_area(regs, error_code, address); | |
1384 | return; | |
1385 | } | |
1067f030 | 1386 | if (error_code & X86_PF_USER) { |
6f4d368e HH |
1387 | /* |
1388 | * Accessing the stack below %sp is always a bug. | |
1389 | * The large cushion allows instructions like enter | |
2d4a7167 | 1390 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1391 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1392 | */ |
92181f19 NP |
1393 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1394 | bad_area(regs, error_code, address); | |
1395 | return; | |
1396 | } | |
1da177e4 | 1397 | } |
92181f19 NP |
1398 | if (unlikely(expand_stack(vma, address))) { |
1399 | bad_area(regs, error_code, address); | |
1400 | return; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Ok, we have a good vm_area for this memory access, so | |
1405 | * we can handle it.. | |
1406 | */ | |
1da177e4 | 1407 | good_area: |
68da336a | 1408 | if (unlikely(access_error(error_code, vma))) { |
7b2d0dba | 1409 | bad_area_access_error(regs, error_code, address, vma); |
92181f19 | 1410 | return; |
1da177e4 LT |
1411 | } |
1412 | ||
1413 | /* | |
1414 | * If for any reason at all we couldn't handle the fault, | |
1415 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1416 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1417 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
cb0631fd VB |
1418 | * |
1419 | * Note that handle_userfault() may also release and reacquire mmap_sem | |
1420 | * (and not return with VM_FAULT_RETRY), when returning to userland to | |
1421 | * repeat the page fault later with a VM_FAULT_NOPAGE retval | |
1422 | * (potentially after handling any pending signal during the return to | |
1423 | * userland). The return to userland is identified whenever | |
1424 | * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. | |
1425 | * Thus we have to be careful about not touching vma after handling the | |
1426 | * fault, so we read the pkey beforehand. | |
1da177e4 | 1427 | */ |
cb0631fd | 1428 | pkey = vma_pkey(vma); |
dcddffd4 | 1429 | fault = handle_mm_fault(vma, address, flags); |
26178ec1 | 1430 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1431 | |
3a13c4d7 | 1432 | /* |
26178ec1 LT |
1433 | * If we need to retry the mmap_sem has already been released, |
1434 | * and if there is a fatal signal pending there is no guarantee | |
1435 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1436 | */ |
26178ec1 LT |
1437 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1438 | /* Retry at most once */ | |
1439 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1440 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1441 | flags |= FAULT_FLAG_TRIED; | |
1442 | if (!fatal_signal_pending(tsk)) | |
1443 | goto retry; | |
1444 | } | |
1445 | ||
1446 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1447 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1448 | return; |
1449 | ||
1450 | /* Not returning to user mode? Handle exceptions or die: */ | |
1451 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1452 | return; |
26178ec1 | 1453 | } |
3a13c4d7 | 1454 | |
26178ec1 | 1455 | up_read(&mm->mmap_sem); |
3a13c4d7 | 1456 | if (unlikely(fault & VM_FAULT_ERROR)) { |
a3c4fb7c | 1457 | mm_fault_error(regs, error_code, address, &pkey, fault); |
3a13c4d7 | 1458 | return; |
37b23e05 KM |
1459 | } |
1460 | ||
d065bd81 | 1461 | /* |
26178ec1 LT |
1462 | * Major/minor page fault accounting. If any of the events |
1463 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1464 | */ |
26178ec1 LT |
1465 | if (major) { |
1466 | tsk->maj_flt++; | |
1467 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1468 | } else { | |
1469 | tsk->min_flt++; | |
1470 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1471 | } |
d729ab35 | 1472 | |
8c938f9f | 1473 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1474 | } |
9326638c | 1475 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1476 | |
9326638c MH |
1477 | static nokprobe_inline void |
1478 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1479 | unsigned long error_code) | |
d34603b0 SA |
1480 | { |
1481 | if (user_mode(regs)) | |
d4078e23 | 1482 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1483 | else |
d4078e23 | 1484 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1485 | } |
1486 | ||
11a7ffb0 TG |
1487 | /* |
1488 | * We must have this function blacklisted from kprobes, tagged with notrace | |
1489 | * and call read_cr2() before calling anything else. To avoid calling any | |
1490 | * kind of tracing machinery before we've observed the CR2 value. | |
1491 | * | |
1492 | * exception_{enter,exit}() contains all sorts of tracepoints. | |
1493 | */ | |
9326638c | 1494 | dotraplinkage void notrace |
11a7ffb0 | 1495 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
25c74b10 | 1496 | { |
11a7ffb0 | 1497 | unsigned long address = read_cr2(); /* Get the faulting address */ |
d4078e23 | 1498 | enum ctx_state prev_state; |
25c74b10 SA |
1499 | |
1500 | prev_state = exception_enter(); | |
80954747 | 1501 | if (trace_pagefault_enabled()) |
11a7ffb0 TG |
1502 | trace_page_fault_entries(address, regs, error_code); |
1503 | ||
0ac09f9f | 1504 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1505 | exception_exit(prev_state); |
1506 | } | |
11a7ffb0 | 1507 | NOKPROBE_SYMBOL(do_page_fault); |