]>
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_*(), ... */ | |
f8561296 | 23 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
f40c3300 AL |
24 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
25 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
ba3e127e | 26 | #include <asm/vm86.h> /* struct vm86 */ |
019132ff | 27 | #include <asm/mmu_context.h> /* vma_pkey() */ |
1da177e4 | 28 | |
d34603b0 SA |
29 | #define CREATE_TRACE_POINTS |
30 | #include <asm/trace/exceptions.h> | |
31 | ||
33cb5243 | 32 | /* |
2d4a7167 IM |
33 | * Page fault error code bits: |
34 | * | |
35 | * bit 0 == 0: no page found 1: protection fault | |
36 | * bit 1 == 0: read access 1: write access | |
37 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
38 | * bit 3 == 1: use of reserved bit detected | |
39 | * bit 4 == 1: fault was an instruction fetch | |
b3ecd515 | 40 | * bit 5 == 1: protection keys block access |
33cb5243 | 41 | */ |
2d4a7167 IM |
42 | enum x86_pf_error_code { |
43 | ||
44 | PF_PROT = 1 << 0, | |
45 | PF_WRITE = 1 << 1, | |
46 | PF_USER = 1 << 2, | |
47 | PF_RSVD = 1 << 3, | |
48 | PF_INSTR = 1 << 4, | |
b3ecd515 | 49 | PF_PK = 1 << 5, |
2d4a7167 | 50 | }; |
66c58156 | 51 | |
b814d41f | 52 | /* |
b319eed0 IM |
53 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
54 | * handled by mmiotrace: | |
b814d41f | 55 | */ |
9326638c | 56 | static nokprobe_inline int |
62c9295f | 57 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 58 | { |
0fd0e3da PP |
59 | if (unlikely(is_kmmio_active())) |
60 | if (kmmio_handler(regs, addr) == 1) | |
61 | return -1; | |
0fd0e3da | 62 | return 0; |
86069782 PP |
63 | } |
64 | ||
9326638c | 65 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 66 | { |
74a0b576 CH |
67 | int ret = 0; |
68 | ||
69 | /* kprobe_running() needs smp_processor_id() */ | |
f39b6f0e | 70 | if (kprobes_built_in() && !user_mode(regs)) { |
74a0b576 CH |
71 | preempt_disable(); |
72 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
73 | ret = 1; | |
74 | preempt_enable(); | |
75 | } | |
1bd858a5 | 76 | |
74a0b576 | 77 | return ret; |
33cb5243 | 78 | } |
1bd858a5 | 79 | |
1dc85be0 | 80 | /* |
2d4a7167 IM |
81 | * Prefetch quirks: |
82 | * | |
83 | * 32-bit mode: | |
84 | * | |
85 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
86 | * Check that here and ignore it. | |
1dc85be0 | 87 | * |
2d4a7167 | 88 | * 64-bit mode: |
1dc85be0 | 89 | * |
2d4a7167 IM |
90 | * Sometimes the CPU reports invalid exceptions on prefetch. |
91 | * Check that here and ignore it. | |
92 | * | |
93 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 94 | */ |
107a0367 IM |
95 | static inline int |
96 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
97 | unsigned char opcode, int *prefetch) | |
98 | { | |
99 | unsigned char instr_hi = opcode & 0xf0; | |
100 | unsigned char instr_lo = opcode & 0x0f; | |
101 | ||
102 | switch (instr_hi) { | |
103 | case 0x20: | |
104 | case 0x30: | |
105 | /* | |
106 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
107 | * In X86_64 long mode, the CPU will signal invalid | |
108 | * opcode if some of these prefixes are present so | |
109 | * X86_64 will never get here anyway | |
110 | */ | |
111 | return ((instr_lo & 7) == 0x6); | |
112 | #ifdef CONFIG_X86_64 | |
113 | case 0x40: | |
114 | /* | |
115 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
116 | * Need to figure out under what instruction mode the | |
117 | * instruction was issued. Could check the LDT for lm, | |
118 | * but for now it's good enough to assume that long | |
119 | * mode only uses well known segments or kernel. | |
120 | */ | |
318f5a2a | 121 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
122 | #endif |
123 | case 0x60: | |
124 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
125 | return (instr_lo & 0xC) == 0x4; | |
126 | case 0xF0: | |
127 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
128 | return !instr_lo || (instr_lo>>1) == 1; | |
129 | case 0x00: | |
130 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
131 | if (probe_kernel_address(instr, opcode)) | |
132 | return 0; | |
133 | ||
134 | *prefetch = (instr_lo == 0xF) && | |
135 | (opcode == 0x0D || opcode == 0x18); | |
136 | return 0; | |
137 | default: | |
138 | return 0; | |
139 | } | |
140 | } | |
141 | ||
2d4a7167 IM |
142 | static int |
143 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 144 | { |
2d4a7167 | 145 | unsigned char *max_instr; |
ab2bf0c1 | 146 | unsigned char *instr; |
33cb5243 | 147 | int prefetch = 0; |
1da177e4 | 148 | |
3085354d IM |
149 | /* |
150 | * If it was a exec (instruction fetch) fault on NX page, then | |
151 | * do not ignore the fault: | |
152 | */ | |
66c58156 | 153 | if (error_code & PF_INSTR) |
1da177e4 | 154 | return 0; |
1dc85be0 | 155 | |
107a0367 | 156 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 157 | max_instr = instr + 15; |
1da177e4 | 158 | |
d31bf07f | 159 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
160 | return 0; |
161 | ||
107a0367 | 162 | while (instr < max_instr) { |
2d4a7167 | 163 | unsigned char opcode; |
1da177e4 | 164 | |
ab2bf0c1 | 165 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 166 | break; |
1da177e4 | 167 | |
1da177e4 LT |
168 | instr++; |
169 | ||
107a0367 | 170 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 171 | break; |
1da177e4 LT |
172 | } |
173 | return prefetch; | |
174 | } | |
175 | ||
019132ff DH |
176 | /* |
177 | * A protection key fault means that the PKRU value did not allow | |
178 | * access to some PTE. Userspace can figure out what PKRU was | |
179 | * from the XSAVE state, and this function fills out a field in | |
180 | * siginfo so userspace can discover which protection key was set | |
181 | * on the PTE. | |
182 | * | |
183 | * If we get here, we know that the hardware signaled a PF_PK | |
184 | * fault and that there was a VMA once we got in the fault | |
185 | * handler. It does *not* guarantee that the VMA we find here | |
186 | * was the one that we faulted on. | |
187 | * | |
188 | * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); | |
189 | * 2. T1 : set PKRU to deny access to pkey=4, touches page | |
190 | * 3. T1 : faults... | |
191 | * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); | |
192 | * 5. T1 : enters fault handler, takes mmap_sem, etc... | |
193 | * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really | |
194 | * faulted on a pte with its pkey=4. | |
195 | */ | |
a3c4fb7c | 196 | static void fill_sig_info_pkey(int si_code, siginfo_t *info, u32 *pkey) |
019132ff DH |
197 | { |
198 | /* This is effectively an #ifdef */ | |
199 | if (!boot_cpu_has(X86_FEATURE_OSPKE)) | |
200 | return; | |
201 | ||
202 | /* Fault not from Protection Keys: nothing to do */ | |
203 | if (si_code != SEGV_PKUERR) | |
204 | return; | |
205 | /* | |
206 | * force_sig_info_fault() is called from a number of | |
207 | * contexts, some of which have a VMA and some of which | |
208 | * do not. The PF_PK handing happens after we have a | |
209 | * valid VMA, so we should never reach this without a | |
210 | * valid VMA. | |
211 | */ | |
a3c4fb7c | 212 | if (!pkey) { |
019132ff DH |
213 | WARN_ONCE(1, "PKU fault with no VMA passed in"); |
214 | info->si_pkey = 0; | |
215 | return; | |
216 | } | |
217 | /* | |
218 | * si_pkey should be thought of as a strong hint, but not | |
219 | * absolutely guranteed to be 100% accurate because of | |
220 | * the race explained above. | |
221 | */ | |
a3c4fb7c | 222 | info->si_pkey = *pkey; |
019132ff DH |
223 | } |
224 | ||
2d4a7167 IM |
225 | static void |
226 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
a3c4fb7c | 227 | struct task_struct *tsk, u32 *pkey, int fault) |
c4aba4a8 | 228 | { |
f672b49b | 229 | unsigned lsb = 0; |
c4aba4a8 HH |
230 | siginfo_t info; |
231 | ||
2d4a7167 IM |
232 | info.si_signo = si_signo; |
233 | info.si_errno = 0; | |
234 | info.si_code = si_code; | |
235 | info.si_addr = (void __user *)address; | |
f672b49b AK |
236 | if (fault & VM_FAULT_HWPOISON_LARGE) |
237 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
238 | if (fault & VM_FAULT_HWPOISON) | |
239 | lsb = PAGE_SHIFT; | |
240 | info.si_addr_lsb = lsb; | |
2d4a7167 | 241 | |
a3c4fb7c | 242 | fill_sig_info_pkey(si_code, &info, pkey); |
019132ff | 243 | |
c4aba4a8 HH |
244 | force_sig_info(si_signo, &info, tsk); |
245 | } | |
246 | ||
f2f13a85 IM |
247 | DEFINE_SPINLOCK(pgd_lock); |
248 | LIST_HEAD(pgd_list); | |
249 | ||
250 | #ifdef CONFIG_X86_32 | |
251 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 252 | { |
f2f13a85 IM |
253 | unsigned index = pgd_index(address); |
254 | pgd_t *pgd_k; | |
e0c4f675 | 255 | p4d_t *p4d, *p4d_k; |
f2f13a85 IM |
256 | pud_t *pud, *pud_k; |
257 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 258 | |
f2f13a85 IM |
259 | pgd += index; |
260 | pgd_k = init_mm.pgd + index; | |
261 | ||
262 | if (!pgd_present(*pgd_k)) | |
263 | return NULL; | |
264 | ||
265 | /* | |
266 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
267 | * and redundant with the set_pmd() on non-PAE. As would | |
e0c4f675 | 268 | * set_p4d/set_pud. |
f2f13a85 | 269 | */ |
e0c4f675 KS |
270 | p4d = p4d_offset(pgd, address); |
271 | p4d_k = p4d_offset(pgd_k, address); | |
272 | if (!p4d_present(*p4d_k)) | |
273 | return NULL; | |
274 | ||
275 | pud = pud_offset(p4d, address); | |
276 | pud_k = pud_offset(p4d_k, address); | |
f2f13a85 IM |
277 | if (!pud_present(*pud_k)) |
278 | return NULL; | |
279 | ||
280 | pmd = pmd_offset(pud, address); | |
281 | pmd_k = pmd_offset(pud_k, address); | |
282 | if (!pmd_present(*pmd_k)) | |
283 | return NULL; | |
284 | ||
b8bcfe99 | 285 | if (!pmd_present(*pmd)) |
f2f13a85 | 286 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 287 | else |
f2f13a85 | 288 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
289 | |
290 | return pmd_k; | |
291 | } | |
292 | ||
293 | void vmalloc_sync_all(void) | |
294 | { | |
295 | unsigned long address; | |
296 | ||
297 | if (SHARED_KERNEL_PMD) | |
298 | return; | |
299 | ||
300 | for (address = VMALLOC_START & PMD_MASK; | |
dc4fac84 | 301 | address >= TASK_SIZE_MAX && address < FIXADDR_TOP; |
f2f13a85 | 302 | address += PMD_SIZE) { |
f2f13a85 IM |
303 | struct page *page; |
304 | ||
a79e53d8 | 305 | spin_lock(&pgd_lock); |
f2f13a85 | 306 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 307 | spinlock_t *pgt_lock; |
f01f7c56 | 308 | pmd_t *ret; |
617d34d9 | 309 | |
a79e53d8 | 310 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
311 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
312 | ||
313 | spin_lock(pgt_lock); | |
314 | ret = vmalloc_sync_one(page_address(page), address); | |
315 | spin_unlock(pgt_lock); | |
316 | ||
317 | if (!ret) | |
f2f13a85 IM |
318 | break; |
319 | } | |
a79e53d8 | 320 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
321 | } |
322 | } | |
323 | ||
324 | /* | |
325 | * 32-bit: | |
326 | * | |
327 | * Handle a fault on the vmalloc or module mapping area | |
328 | */ | |
9326638c | 329 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
330 | { |
331 | unsigned long pgd_paddr; | |
332 | pmd_t *pmd_k; | |
333 | pte_t *pte_k; | |
334 | ||
335 | /* Make sure we are in vmalloc area: */ | |
336 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
337 | return -1; | |
338 | ||
ebc8827f FW |
339 | WARN_ON_ONCE(in_nmi()); |
340 | ||
f2f13a85 IM |
341 | /* |
342 | * Synchronize this task's top level page-table | |
343 | * with the 'reference' page table. | |
344 | * | |
345 | * Do _not_ use "current" here. We might be inside | |
346 | * an interrupt in the middle of a task switch.. | |
347 | */ | |
6c690ee1 | 348 | pgd_paddr = read_cr3_pa(); |
f2f13a85 IM |
349 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); |
350 | if (!pmd_k) | |
351 | return -1; | |
352 | ||
f4eafd8b TK |
353 | if (pmd_huge(*pmd_k)) |
354 | return 0; | |
355 | ||
f2f13a85 IM |
356 | pte_k = pte_offset_kernel(pmd_k, address); |
357 | if (!pte_present(*pte_k)) | |
358 | return -1; | |
359 | ||
360 | return 0; | |
361 | } | |
9326638c | 362 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
363 | |
364 | /* | |
365 | * Did it hit the DOS screen memory VA from vm86 mode? | |
366 | */ | |
367 | static inline void | |
368 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
369 | struct task_struct *tsk) | |
370 | { | |
9fda6a06 | 371 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
372 | unsigned long bit; |
373 | ||
9fda6a06 | 374 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
375 | return; |
376 | ||
377 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
378 | if (bit < 32) | |
9fda6a06 BG |
379 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
380 | #endif | |
33cb5243 | 381 | } |
1da177e4 | 382 | |
087975b0 | 383 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 384 | { |
087975b0 AM |
385 | return pfn < max_low_pfn; |
386 | } | |
1156e098 | 387 | |
087975b0 AM |
388 | static void dump_pagetable(unsigned long address) |
389 | { | |
6c690ee1 | 390 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 391 | pgd_t *pgd = &base[pgd_index(address)]; |
e0c4f675 KS |
392 | p4d_t *p4d; |
393 | pud_t *pud; | |
087975b0 AM |
394 | pmd_t *pmd; |
395 | pte_t *pte; | |
2d4a7167 | 396 | |
1156e098 | 397 | #ifdef CONFIG_X86_PAE |
39e48d9b | 398 | pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); |
087975b0 AM |
399 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) |
400 | goto out; | |
39e48d9b JB |
401 | #define pr_pde pr_cont |
402 | #else | |
403 | #define pr_pde pr_info | |
1156e098 | 404 | #endif |
e0c4f675 KS |
405 | p4d = p4d_offset(pgd, address); |
406 | pud = pud_offset(p4d, address); | |
407 | pmd = pmd_offset(pud, address); | |
39e48d9b JB |
408 | pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); |
409 | #undef pr_pde | |
1156e098 HH |
410 | |
411 | /* | |
412 | * We must not directly access the pte in the highpte | |
413 | * case if the page table is located in highmem. | |
414 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 415 | * it's allocated already: |
1156e098 | 416 | */ |
087975b0 AM |
417 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
418 | goto out; | |
1156e098 | 419 | |
087975b0 | 420 | pte = pte_offset_kernel(pmd, address); |
39e48d9b | 421 | pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); |
087975b0 | 422 | out: |
39e48d9b | 423 | pr_cont("\n"); |
f2f13a85 IM |
424 | } |
425 | ||
426 | #else /* CONFIG_X86_64: */ | |
427 | ||
428 | void vmalloc_sync_all(void) | |
429 | { | |
5372e155 | 430 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
431 | } |
432 | ||
433 | /* | |
434 | * 64-bit: | |
435 | * | |
436 | * Handle a fault on the vmalloc area | |
f2f13a85 | 437 | */ |
9326638c | 438 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
439 | { |
440 | pgd_t *pgd, *pgd_ref; | |
b50858ce | 441 | p4d_t *p4d, *p4d_ref; |
f2f13a85 IM |
442 | pud_t *pud, *pud_ref; |
443 | pmd_t *pmd, *pmd_ref; | |
444 | pte_t *pte, *pte_ref; | |
445 | ||
446 | /* Make sure we are in vmalloc area: */ | |
447 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
448 | return -1; | |
449 | ||
ebc8827f FW |
450 | WARN_ON_ONCE(in_nmi()); |
451 | ||
f2f13a85 IM |
452 | /* |
453 | * Copy kernel mappings over when needed. This can also | |
454 | * happen within a race in page table update. In the later | |
455 | * case just flush: | |
456 | */ | |
6c690ee1 | 457 | pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); |
f2f13a85 IM |
458 | pgd_ref = pgd_offset_k(address); |
459 | if (pgd_none(*pgd_ref)) | |
460 | return -1; | |
461 | ||
1160c277 | 462 | if (pgd_none(*pgd)) { |
f2f13a85 | 463 | set_pgd(pgd, *pgd_ref); |
1160c277 | 464 | arch_flush_lazy_mmu_mode(); |
b50858ce KS |
465 | } else if (CONFIG_PGTABLE_LEVELS > 4) { |
466 | /* | |
467 | * With folded p4d, pgd_none() is always false, so the pgd may | |
468 | * point to an empty page table entry and pgd_page_vaddr() | |
469 | * will return garbage. | |
470 | * | |
471 | * We will do the correct sanity check on the p4d level. | |
472 | */ | |
f2f13a85 | 473 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 474 | } |
f2f13a85 | 475 | |
b50858ce KS |
476 | /* With 4-level paging, copying happens on the p4d level. */ |
477 | p4d = p4d_offset(pgd, address); | |
478 | p4d_ref = p4d_offset(pgd_ref, address); | |
479 | if (p4d_none(*p4d_ref)) | |
480 | return -1; | |
481 | ||
482 | if (p4d_none(*p4d)) { | |
483 | set_p4d(p4d, *p4d_ref); | |
484 | arch_flush_lazy_mmu_mode(); | |
485 | } else { | |
486 | BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref)); | |
487 | } | |
488 | ||
f2f13a85 IM |
489 | /* |
490 | * Below here mismatches are bugs because these lower tables | |
491 | * are shared: | |
492 | */ | |
493 | ||
b50858ce KS |
494 | pud = pud_offset(p4d, address); |
495 | pud_ref = pud_offset(p4d_ref, address); | |
f2f13a85 IM |
496 | if (pud_none(*pud_ref)) |
497 | return -1; | |
498 | ||
f4eafd8b | 499 | if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref)) |
f2f13a85 IM |
500 | BUG(); |
501 | ||
f4eafd8b TK |
502 | if (pud_huge(*pud)) |
503 | return 0; | |
504 | ||
f2f13a85 IM |
505 | pmd = pmd_offset(pud, address); |
506 | pmd_ref = pmd_offset(pud_ref, address); | |
507 | if (pmd_none(*pmd_ref)) | |
508 | return -1; | |
509 | ||
f4eafd8b | 510 | if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref)) |
f2f13a85 IM |
511 | BUG(); |
512 | ||
f4eafd8b TK |
513 | if (pmd_huge(*pmd)) |
514 | return 0; | |
515 | ||
f2f13a85 IM |
516 | pte_ref = pte_offset_kernel(pmd_ref, address); |
517 | if (!pte_present(*pte_ref)) | |
518 | return -1; | |
519 | ||
520 | pte = pte_offset_kernel(pmd, address); | |
521 | ||
522 | /* | |
523 | * Don't use pte_page here, because the mappings can point | |
524 | * outside mem_map, and the NUMA hash lookup cannot handle | |
525 | * that: | |
526 | */ | |
527 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
528 | BUG(); | |
529 | ||
530 | return 0; | |
531 | } | |
9326638c | 532 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 533 | |
e05139f2 | 534 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 535 | static const char errata93_warning[] = |
ad361c98 JP |
536 | KERN_ERR |
537 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
538 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
539 | "******* Please consider a BIOS update.\n" | |
540 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 541 | #endif |
f2f13a85 IM |
542 | |
543 | /* | |
544 | * No vm86 mode in 64-bit mode: | |
545 | */ | |
546 | static inline void | |
547 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
548 | struct task_struct *tsk) | |
549 | { | |
550 | } | |
551 | ||
552 | static int bad_address(void *p) | |
553 | { | |
554 | unsigned long dummy; | |
555 | ||
556 | return probe_kernel_address((unsigned long *)p, dummy); | |
557 | } | |
558 | ||
559 | static void dump_pagetable(unsigned long address) | |
560 | { | |
6c690ee1 | 561 | pgd_t *base = __va(read_cr3_pa()); |
087975b0 | 562 | pgd_t *pgd = base + pgd_index(address); |
e0c4f675 | 563 | p4d_t *p4d; |
1da177e4 LT |
564 | pud_t *pud; |
565 | pmd_t *pmd; | |
566 | pte_t *pte; | |
567 | ||
2d4a7167 IM |
568 | if (bad_address(pgd)) |
569 | goto bad; | |
570 | ||
39e48d9b | 571 | pr_info("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
572 | |
573 | if (!pgd_present(*pgd)) | |
574 | goto out; | |
1da177e4 | 575 | |
e0c4f675 KS |
576 | p4d = p4d_offset(pgd, address); |
577 | if (bad_address(p4d)) | |
578 | goto bad; | |
579 | ||
39e48d9b | 580 | pr_cont("P4D %lx ", p4d_val(*p4d)); |
e0c4f675 KS |
581 | if (!p4d_present(*p4d) || p4d_large(*p4d)) |
582 | goto out; | |
583 | ||
584 | pud = pud_offset(p4d, address); | |
2d4a7167 IM |
585 | if (bad_address(pud)) |
586 | goto bad; | |
587 | ||
39e48d9b | 588 | pr_cont("PUD %lx ", pud_val(*pud)); |
b5360222 | 589 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 590 | goto out; |
1da177e4 LT |
591 | |
592 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
593 | if (bad_address(pmd)) |
594 | goto bad; | |
595 | ||
39e48d9b | 596 | pr_cont("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
597 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
598 | goto out; | |
1da177e4 LT |
599 | |
600 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
601 | if (bad_address(pte)) |
602 | goto bad; | |
603 | ||
39e48d9b | 604 | pr_cont("PTE %lx", pte_val(*pte)); |
2d4a7167 | 605 | out: |
39e48d9b | 606 | pr_cont("\n"); |
1da177e4 LT |
607 | return; |
608 | bad: | |
39e48d9b | 609 | pr_info("BAD\n"); |
8c938f9f IM |
610 | } |
611 | ||
f2f13a85 | 612 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 613 | |
2d4a7167 IM |
614 | /* |
615 | * Workaround for K8 erratum #93 & buggy BIOS. | |
616 | * | |
617 | * BIOS SMM functions are required to use a specific workaround | |
618 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
619 | * | |
620 | * A lot of BIOS that didn't get tested properly miss this. | |
621 | * | |
622 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
623 | * Try to work around it here. | |
624 | * | |
625 | * Note we only handle faults in kernel here. | |
626 | * Does nothing on 32-bit. | |
fdfe8aa8 | 627 | */ |
33cb5243 | 628 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 629 | { |
e05139f2 JB |
630 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
631 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
632 | || boot_cpu_data.x86 != 0xf) | |
633 | return 0; | |
634 | ||
65ea5b03 | 635 | if (address != regs->ip) |
1da177e4 | 636 | return 0; |
2d4a7167 | 637 | |
33cb5243 | 638 | if ((address >> 32) != 0) |
1da177e4 | 639 | return 0; |
2d4a7167 | 640 | |
1da177e4 | 641 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
642 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
643 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 644 | printk_once(errata93_warning); |
65ea5b03 | 645 | regs->ip = address; |
1da177e4 LT |
646 | return 1; |
647 | } | |
fdfe8aa8 | 648 | #endif |
1da177e4 | 649 | return 0; |
33cb5243 | 650 | } |
1da177e4 | 651 | |
35f3266f | 652 | /* |
2d4a7167 IM |
653 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
654 | * to illegal addresses >4GB. | |
655 | * | |
656 | * We catch this in the page fault handler because these addresses | |
657 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
658 | * segment in LDT is compatibility mode. |
659 | */ | |
660 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
661 | { | |
662 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 663 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
664 | return 1; |
665 | #endif | |
666 | return 0; | |
667 | } | |
668 | ||
29caf2f9 HH |
669 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
670 | { | |
671 | #ifdef CONFIG_X86_F00F_BUG | |
672 | unsigned long nr; | |
2d4a7167 | 673 | |
29caf2f9 | 674 | /* |
2d4a7167 | 675 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 676 | */ |
e2604b49 | 677 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
678 | nr = (address - idt_descr.address) >> 3; |
679 | ||
680 | if (nr == 6) { | |
681 | do_invalid_op(regs, 0); | |
682 | return 1; | |
683 | } | |
684 | } | |
685 | #endif | |
686 | return 0; | |
687 | } | |
688 | ||
8f766149 IM |
689 | static const char nx_warning[] = KERN_CRIT |
690 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
691 | static const char smep_warning[] = KERN_CRIT |
692 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 693 | |
2d4a7167 IM |
694 | static void |
695 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
696 | unsigned long address) | |
b3279c7f | 697 | { |
1156e098 HH |
698 | if (!oops_may_print()) |
699 | return; | |
700 | ||
1156e098 | 701 | if (error_code & PF_INSTR) { |
93809be8 | 702 | unsigned int level; |
426e34cc MF |
703 | pgd_t *pgd; |
704 | pte_t *pte; | |
2d4a7167 | 705 | |
6c690ee1 | 706 | pgd = __va(read_cr3_pa()); |
426e34cc MF |
707 | pgd += pgd_index(address); |
708 | ||
709 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 710 | |
8f766149 | 711 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 712 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
713 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
714 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 715 | (__read_cr4() & X86_CR4_SMEP)) |
eff50c34 | 716 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 717 | } |
1156e098 | 718 | |
19f0dda9 | 719 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 720 | if (address < PAGE_SIZE) |
19f0dda9 | 721 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 722 | else |
19f0dda9 | 723 | printk(KERN_CONT "paging request"); |
2d4a7167 | 724 | |
f294a8ce | 725 | printk(KERN_CONT " at %p\n", (void *) address); |
bb5e5ce5 | 726 | printk(KERN_ALERT "IP: %pS\n", (void *)regs->ip); |
2d4a7167 | 727 | |
b3279c7f HH |
728 | dump_pagetable(address); |
729 | } | |
730 | ||
2d4a7167 IM |
731 | static noinline void |
732 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
733 | unsigned long address) | |
1da177e4 | 734 | { |
2d4a7167 IM |
735 | struct task_struct *tsk; |
736 | unsigned long flags; | |
737 | int sig; | |
738 | ||
739 | flags = oops_begin(); | |
740 | tsk = current; | |
741 | sig = SIGKILL; | |
1209140c | 742 | |
1da177e4 | 743 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 744 | tsk->comm, address); |
1da177e4 | 745 | dump_pagetable(address); |
2d4a7167 IM |
746 | |
747 | tsk->thread.cr2 = address; | |
51e7dc70 | 748 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
749 | tsk->thread.error_code = error_code; |
750 | ||
22f5991c | 751 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 752 | sig = 0; |
2d4a7167 | 753 | |
874d93d1 | 754 | oops_end(flags, regs, sig); |
1da177e4 LT |
755 | } |
756 | ||
2d4a7167 IM |
757 | static noinline void |
758 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 759 | unsigned long address, int signal, int si_code) |
92181f19 NP |
760 | { |
761 | struct task_struct *tsk = current; | |
92181f19 NP |
762 | unsigned long flags; |
763 | int sig; | |
92181f19 | 764 | |
2d4a7167 | 765 | /* Are we prepared to handle this kernel fault? */ |
548acf19 | 766 | if (fixup_exception(regs, X86_TRAP_PF)) { |
c026b359 PZ |
767 | /* |
768 | * Any interrupt that takes a fault gets the fixup. This makes | |
769 | * the below recursive fault logic only apply to a faults from | |
770 | * task context. | |
771 | */ | |
772 | if (in_interrupt()) | |
773 | return; | |
774 | ||
775 | /* | |
776 | * Per the above we're !in_interrupt(), aka. task context. | |
777 | * | |
778 | * In this case we need to make sure we're not recursively | |
779 | * faulting through the emulate_vsyscall() logic. | |
780 | */ | |
2a53ccbc | 781 | if (current->thread.sig_on_uaccess_err && signal) { |
51e7dc70 | 782 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
783 | tsk->thread.error_code = error_code | PF_USER; |
784 | tsk->thread.cr2 = address; | |
785 | ||
786 | /* XXX: hwpoison faults will set the wrong code. */ | |
7b2d0dba | 787 | force_sig_info_fault(signal, si_code, address, |
a3c4fb7c | 788 | tsk, NULL, 0); |
4fc34901 | 789 | } |
c026b359 PZ |
790 | |
791 | /* | |
792 | * Barring that, we can do the fixup and be happy. | |
793 | */ | |
92181f19 | 794 | return; |
4fc34901 | 795 | } |
92181f19 | 796 | |
6271cfdf AL |
797 | #ifdef CONFIG_VMAP_STACK |
798 | /* | |
799 | * Stack overflow? During boot, we can fault near the initial | |
800 | * stack in the direct map, but that's not an overflow -- check | |
801 | * that we're in vmalloc space to avoid this. | |
802 | */ | |
803 | if (is_vmalloc_addr((void *)address) && | |
804 | (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || | |
805 | address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { | |
6271cfdf AL |
806 | unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *); |
807 | /* | |
808 | * We're likely to be running with very little stack space | |
809 | * left. It's plausible that we'd hit this condition but | |
810 | * double-fault even before we get this far, in which case | |
811 | * we're fine: the double-fault handler will deal with it. | |
812 | * | |
813 | * We don't want to make it all the way into the oops code | |
814 | * and then double-fault, though, because we're likely to | |
815 | * break the console driver and lose most of the stack dump. | |
816 | */ | |
817 | asm volatile ("movq %[stack], %%rsp\n\t" | |
818 | "call handle_stack_overflow\n\t" | |
819 | "1: jmp 1b" | |
f5caf621 | 820 | : ASM_CALL_CONSTRAINT |
6271cfdf AL |
821 | : "D" ("kernel stack overflow (page fault)"), |
822 | "S" (regs), "d" (address), | |
823 | [stack] "rm" (stack)); | |
824 | unreachable(); | |
825 | } | |
826 | #endif | |
827 | ||
92181f19 | 828 | /* |
2d4a7167 IM |
829 | * 32-bit: |
830 | * | |
831 | * Valid to do another page fault here, because if this fault | |
832 | * had been triggered by is_prefetch fixup_exception would have | |
833 | * handled it. | |
834 | * | |
835 | * 64-bit: | |
92181f19 | 836 | * |
2d4a7167 | 837 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
838 | */ |
839 | if (is_prefetch(regs, error_code, address)) | |
840 | return; | |
841 | ||
842 | if (is_errata93(regs, address)) | |
843 | return; | |
844 | ||
845 | /* | |
846 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 847 | * terminate things with extreme prejudice: |
92181f19 | 848 | */ |
92181f19 | 849 | flags = oops_begin(); |
92181f19 NP |
850 | |
851 | show_fault_oops(regs, error_code, address); | |
852 | ||
a70857e4 | 853 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 854 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 855 | |
1cc99544 | 856 | tsk->thread.cr2 = address; |
51e7dc70 | 857 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 858 | tsk->thread.error_code = error_code; |
92181f19 | 859 | |
92181f19 NP |
860 | sig = SIGKILL; |
861 | if (__die("Oops", regs, error_code)) | |
862 | sig = 0; | |
2d4a7167 | 863 | |
92181f19 | 864 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 865 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 866 | |
92181f19 | 867 | oops_end(flags, regs, sig); |
92181f19 NP |
868 | } |
869 | ||
2d4a7167 IM |
870 | /* |
871 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
872 | * sysctl is set: | |
873 | */ | |
874 | static inline void | |
875 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
876 | unsigned long address, struct task_struct *tsk) | |
877 | { | |
878 | if (!unhandled_signal(tsk, SIGSEGV)) | |
879 | return; | |
880 | ||
881 | if (!printk_ratelimit()) | |
882 | return; | |
883 | ||
a1a08d1c | 884 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
885 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
886 | tsk->comm, task_pid_nr(tsk), address, | |
887 | (void *)regs->ip, (void *)regs->sp, error_code); | |
888 | ||
889 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
890 | ||
891 | printk(KERN_CONT "\n"); | |
892 | } | |
893 | ||
894 | static void | |
895 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
a3c4fb7c | 896 | unsigned long address, u32 *pkey, int si_code) |
92181f19 NP |
897 | { |
898 | struct task_struct *tsk = current; | |
899 | ||
900 | /* User mode accesses just cause a SIGSEGV */ | |
901 | if (error_code & PF_USER) { | |
902 | /* | |
2d4a7167 | 903 | * It's possible to have interrupts off here: |
92181f19 NP |
904 | */ |
905 | local_irq_enable(); | |
906 | ||
907 | /* | |
908 | * Valid to do another page fault here because this one came | |
2d4a7167 | 909 | * from user space: |
92181f19 NP |
910 | */ |
911 | if (is_prefetch(regs, error_code, address)) | |
912 | return; | |
913 | ||
914 | if (is_errata100(regs, address)) | |
915 | return; | |
916 | ||
3ae36655 AL |
917 | #ifdef CONFIG_X86_64 |
918 | /* | |
919 | * Instruction fetch faults in the vsyscall page might need | |
920 | * emulation. | |
921 | */ | |
922 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 923 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
924 | if (emulate_vsyscall(regs, address)) |
925 | return; | |
926 | } | |
927 | #endif | |
dc4fac84 AL |
928 | |
929 | /* | |
930 | * To avoid leaking information about the kernel page table | |
931 | * layout, pretend that user-mode accesses to kernel addresses | |
932 | * are always protection faults. | |
933 | */ | |
934 | if (address >= TASK_SIZE_MAX) | |
e575a86f | 935 | error_code |= PF_PROT; |
3ae36655 | 936 | |
e575a86f | 937 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
938 | show_signal_msg(regs, error_code, address, tsk); |
939 | ||
2d4a7167 | 940 | tsk->thread.cr2 = address; |
e575a86f | 941 | tsk->thread.error_code = error_code; |
51e7dc70 | 942 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 943 | |
a3c4fb7c | 944 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0); |
2d4a7167 | 945 | |
92181f19 NP |
946 | return; |
947 | } | |
948 | ||
949 | if (is_f00f_bug(regs, address)) | |
950 | return; | |
951 | ||
4fc34901 | 952 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
953 | } |
954 | ||
2d4a7167 IM |
955 | static noinline void |
956 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
a3c4fb7c | 957 | unsigned long address, u32 *pkey) |
92181f19 | 958 | { |
a3c4fb7c | 959 | __bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR); |
92181f19 NP |
960 | } |
961 | ||
2d4a7167 IM |
962 | static void |
963 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 964 | unsigned long address, struct vm_area_struct *vma, int si_code) |
92181f19 NP |
965 | { |
966 | struct mm_struct *mm = current->mm; | |
a3c4fb7c LD |
967 | u32 pkey; |
968 | ||
969 | if (vma) | |
970 | pkey = vma_pkey(vma); | |
92181f19 NP |
971 | |
972 | /* | |
973 | * Something tried to access memory that isn't in our memory map.. | |
974 | * Fix it, but check if it's kernel or user first.. | |
975 | */ | |
976 | up_read(&mm->mmap_sem); | |
977 | ||
a3c4fb7c LD |
978 | __bad_area_nosemaphore(regs, error_code, address, |
979 | (vma) ? &pkey : NULL, si_code); | |
92181f19 NP |
980 | } |
981 | ||
2d4a7167 IM |
982 | static noinline void |
983 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 | 984 | { |
7b2d0dba | 985 | __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); |
92181f19 NP |
986 | } |
987 | ||
33a709b2 DH |
988 | static inline bool bad_area_access_from_pkeys(unsigned long error_code, |
989 | struct vm_area_struct *vma) | |
990 | { | |
07f146f5 DH |
991 | /* This code is always called on the current mm */ |
992 | bool foreign = false; | |
993 | ||
33a709b2 DH |
994 | if (!boot_cpu_has(X86_FEATURE_OSPKE)) |
995 | return false; | |
996 | if (error_code & PF_PK) | |
997 | return true; | |
07f146f5 | 998 | /* this checks permission keys on the VMA: */ |
d61172b4 DH |
999 | if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE), |
1000 | (error_code & PF_INSTR), foreign)) | |
07f146f5 | 1001 | return true; |
33a709b2 | 1002 | return false; |
92181f19 NP |
1003 | } |
1004 | ||
2d4a7167 IM |
1005 | static noinline void |
1006 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 1007 | unsigned long address, struct vm_area_struct *vma) |
92181f19 | 1008 | { |
019132ff DH |
1009 | /* |
1010 | * This OSPKE check is not strictly necessary at runtime. | |
1011 | * But, doing it this way allows compiler optimizations | |
1012 | * if pkeys are compiled out. | |
1013 | */ | |
33a709b2 | 1014 | if (bad_area_access_from_pkeys(error_code, vma)) |
019132ff DH |
1015 | __bad_area(regs, error_code, address, vma, SEGV_PKUERR); |
1016 | else | |
1017 | __bad_area(regs, error_code, address, vma, SEGV_ACCERR); | |
92181f19 NP |
1018 | } |
1019 | ||
2d4a7167 | 1020 | static void |
a6e04aa9 | 1021 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
a3c4fb7c | 1022 | u32 *pkey, unsigned int fault) |
92181f19 NP |
1023 | { |
1024 | struct task_struct *tsk = current; | |
a6e04aa9 | 1025 | int code = BUS_ADRERR; |
92181f19 | 1026 | |
2d4a7167 | 1027 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 1028 | if (!(error_code & PF_USER)) { |
4fc34901 | 1029 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
1030 | return; |
1031 | } | |
2d4a7167 | 1032 | |
cd1b68f0 | 1033 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
1034 | if (is_prefetch(regs, error_code, address)) |
1035 | return; | |
2d4a7167 IM |
1036 | |
1037 | tsk->thread.cr2 = address; | |
1038 | tsk->thread.error_code = error_code; | |
51e7dc70 | 1039 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 1040 | |
a6e04aa9 | 1041 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 1042 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
1043 | printk(KERN_ERR |
1044 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
1045 | tsk->comm, tsk->pid, address); | |
1046 | code = BUS_MCEERR_AR; | |
1047 | } | |
1048 | #endif | |
a3c4fb7c | 1049 | force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault); |
92181f19 NP |
1050 | } |
1051 | ||
3a13c4d7 | 1052 | static noinline void |
2d4a7167 | 1053 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
a3c4fb7c | 1054 | unsigned long address, u32 *pkey, unsigned int fault) |
92181f19 | 1055 | { |
3a13c4d7 | 1056 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
3a13c4d7 JW |
1057 | no_context(regs, error_code, address, 0, 0); |
1058 | return; | |
b80ef10e | 1059 | } |
b80ef10e | 1060 | |
2d4a7167 | 1061 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
1062 | /* Kernel mode? Handle exceptions or die: */ |
1063 | if (!(error_code & PF_USER)) { | |
4fc34901 AL |
1064 | no_context(regs, error_code, address, |
1065 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 1066 | return; |
f8626854 AV |
1067 | } |
1068 | ||
c2d23f91 DR |
1069 | /* |
1070 | * We ran out of memory, call the OOM killer, and return the | |
1071 | * userspace (which will retry the fault, or kill us if we got | |
1072 | * oom-killed): | |
1073 | */ | |
1074 | pagefault_out_of_memory(); | |
2d4a7167 | 1075 | } else { |
f672b49b AK |
1076 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
1077 | VM_FAULT_HWPOISON_LARGE)) | |
a3c4fb7c | 1078 | do_sigbus(regs, error_code, address, pkey, fault); |
33692f27 | 1079 | else if (fault & VM_FAULT_SIGSEGV) |
a3c4fb7c | 1080 | bad_area_nosemaphore(regs, error_code, address, pkey); |
2d4a7167 IM |
1081 | else |
1082 | BUG(); | |
1083 | } | |
92181f19 NP |
1084 | } |
1085 | ||
d8b57bb7 TG |
1086 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
1087 | { | |
1088 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
1089 | return 0; | |
2d4a7167 | 1090 | |
d8b57bb7 TG |
1091 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
1092 | return 0; | |
b3ecd515 DH |
1093 | /* |
1094 | * Note: We do not do lazy flushing on protection key | |
1095 | * changes, so no spurious fault will ever set PF_PK. | |
1096 | */ | |
1097 | if ((error_code & PF_PK)) | |
1098 | return 1; | |
d8b57bb7 TG |
1099 | |
1100 | return 1; | |
1101 | } | |
1102 | ||
5b727a3b | 1103 | /* |
2d4a7167 IM |
1104 | * Handle a spurious fault caused by a stale TLB entry. |
1105 | * | |
1106 | * This allows us to lazily refresh the TLB when increasing the | |
1107 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
1108 | * eagerly is very expensive since that implies doing a full | |
1109 | * cross-processor TLB flush, even if no stale TLB entries exist | |
1110 | * on other processors. | |
1111 | * | |
31668511 DV |
1112 | * Spurious faults may only occur if the TLB contains an entry with |
1113 | * fewer permission than the page table entry. Non-present (P = 0) | |
1114 | * and reserved bit (R = 1) faults are never spurious. | |
1115 | * | |
5b727a3b JF |
1116 | * There are no security implications to leaving a stale TLB when |
1117 | * increasing the permissions on a page. | |
31668511 DV |
1118 | * |
1119 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
1120 | * | |
1121 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
1122 | * (Optional Invalidation). | |
5b727a3b | 1123 | */ |
9326638c | 1124 | static noinline int |
2d4a7167 | 1125 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
1126 | { |
1127 | pgd_t *pgd; | |
e0c4f675 | 1128 | p4d_t *p4d; |
5b727a3b JF |
1129 | pud_t *pud; |
1130 | pmd_t *pmd; | |
1131 | pte_t *pte; | |
3c3e5694 | 1132 | int ret; |
5b727a3b | 1133 | |
31668511 DV |
1134 | /* |
1135 | * Only writes to RO or instruction fetches from NX may cause | |
1136 | * spurious faults. | |
1137 | * | |
1138 | * These could be from user or supervisor accesses but the TLB | |
1139 | * is only lazily flushed after a kernel mapping protection | |
1140 | * change, so user accesses are not expected to cause spurious | |
1141 | * faults. | |
1142 | */ | |
1143 | if (error_code != (PF_WRITE | PF_PROT) | |
1144 | && error_code != (PF_INSTR | PF_PROT)) | |
5b727a3b JF |
1145 | return 0; |
1146 | ||
1147 | pgd = init_mm.pgd + pgd_index(address); | |
1148 | if (!pgd_present(*pgd)) | |
1149 | return 0; | |
1150 | ||
e0c4f675 KS |
1151 | p4d = p4d_offset(pgd, address); |
1152 | if (!p4d_present(*p4d)) | |
1153 | return 0; | |
1154 | ||
1155 | if (p4d_large(*p4d)) | |
1156 | return spurious_fault_check(error_code, (pte_t *) p4d); | |
1157 | ||
1158 | pud = pud_offset(p4d, address); | |
5b727a3b JF |
1159 | if (!pud_present(*pud)) |
1160 | return 0; | |
1161 | ||
d8b57bb7 TG |
1162 | if (pud_large(*pud)) |
1163 | return spurious_fault_check(error_code, (pte_t *) pud); | |
1164 | ||
5b727a3b JF |
1165 | pmd = pmd_offset(pud, address); |
1166 | if (!pmd_present(*pmd)) | |
1167 | return 0; | |
1168 | ||
d8b57bb7 TG |
1169 | if (pmd_large(*pmd)) |
1170 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
1171 | ||
5b727a3b | 1172 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 1173 | if (!pte_present(*pte)) |
5b727a3b JF |
1174 | return 0; |
1175 | ||
3c3e5694 SR |
1176 | ret = spurious_fault_check(error_code, pte); |
1177 | if (!ret) | |
1178 | return 0; | |
1179 | ||
1180 | /* | |
2d4a7167 IM |
1181 | * Make sure we have permissions in PMD. |
1182 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
1183 | */ |
1184 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
1185 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 1186 | |
3c3e5694 | 1187 | return ret; |
5b727a3b | 1188 | } |
9326638c | 1189 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 1190 | |
abd4f750 | 1191 | int show_unhandled_signals = 1; |
1da177e4 | 1192 | |
2d4a7167 | 1193 | static inline int |
68da336a | 1194 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1195 | { |
07f146f5 DH |
1196 | /* This is only called for the current mm, so: */ |
1197 | bool foreign = false; | |
e8c6226d DH |
1198 | |
1199 | /* | |
1200 | * Read or write was blocked by protection keys. This is | |
1201 | * always an unconditional error and can never result in | |
1202 | * a follow-up action to resolve the fault, like a COW. | |
1203 | */ | |
1204 | if (error_code & PF_PK) | |
1205 | return 1; | |
1206 | ||
07f146f5 DH |
1207 | /* |
1208 | * Make sure to check the VMA so that we do not perform | |
1209 | * faults just to hit a PF_PK as soon as we fill in a | |
1210 | * page. | |
1211 | */ | |
d61172b4 DH |
1212 | if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE), |
1213 | (error_code & PF_INSTR), foreign)) | |
07f146f5 | 1214 | return 1; |
33a709b2 | 1215 | |
68da336a | 1216 | if (error_code & PF_WRITE) { |
2d4a7167 | 1217 | /* write, present and write, not present: */ |
92181f19 NP |
1218 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1219 | return 1; | |
2d4a7167 | 1220 | return 0; |
92181f19 NP |
1221 | } |
1222 | ||
2d4a7167 IM |
1223 | /* read, present: */ |
1224 | if (unlikely(error_code & PF_PROT)) | |
1225 | return 1; | |
1226 | ||
1227 | /* read, not present: */ | |
1228 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1229 | return 1; | |
1230 | ||
92181f19 NP |
1231 | return 0; |
1232 | } | |
1233 | ||
0973a06c HS |
1234 | static int fault_in_kernel_space(unsigned long address) |
1235 | { | |
d9517346 | 1236 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1237 | } |
1238 | ||
40d3cd66 PA |
1239 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1240 | { | |
4640c7ee PA |
1241 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1242 | return false; | |
1243 | ||
1244 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1245 | return false; | |
1246 | ||
40d3cd66 PA |
1247 | if (error_code & PF_USER) |
1248 | return false; | |
1249 | ||
f39b6f0e | 1250 | if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) |
40d3cd66 PA |
1251 | return false; |
1252 | ||
1253 | return true; | |
1254 | } | |
1255 | ||
1da177e4 LT |
1256 | /* |
1257 | * This routine handles page faults. It determines the address, | |
1258 | * and the problem, and then passes it off to one of the appropriate | |
1259 | * routines. | |
1da177e4 | 1260 | */ |
9326638c | 1261 | static noinline void |
0ac09f9f JO |
1262 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1263 | unsigned long address) | |
1da177e4 | 1264 | { |
2d4a7167 | 1265 | struct vm_area_struct *vma; |
1da177e4 LT |
1266 | struct task_struct *tsk; |
1267 | struct mm_struct *mm; | |
26178ec1 | 1268 | int fault, major = 0; |
759496ba | 1269 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
a3c4fb7c | 1270 | u32 pkey; |
1da177e4 | 1271 | |
a9ba9a3b AV |
1272 | tsk = current; |
1273 | mm = tsk->mm; | |
2d4a7167 | 1274 | |
f8561296 VN |
1275 | /* |
1276 | * Detect and handle instructions that would cause a page fault for | |
1277 | * both a tracked kernel page and a userspace page. | |
1278 | */ | |
1279 | if (kmemcheck_active(regs)) | |
1280 | kmemcheck_hide(regs); | |
5dfaf90f | 1281 | prefetchw(&mm->mmap_sem); |
f8561296 | 1282 | |
0fd0e3da | 1283 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1284 | return; |
1da177e4 LT |
1285 | |
1286 | /* | |
1287 | * We fault-in kernel-space virtual memory on-demand. The | |
1288 | * 'reference' page table is init_mm.pgd. | |
1289 | * | |
1290 | * NOTE! We MUST NOT take any locks for this case. We may | |
1291 | * be in an interrupt or a critical region, and should | |
1292 | * only copy the information from the master page table, | |
1293 | * nothing more. | |
1294 | * | |
1295 | * This verifies that the fault happens in kernel space | |
1296 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1297 | * protection error (error_code & 9) == 0. |
1da177e4 | 1298 | */ |
0973a06c | 1299 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1300 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1301 | if (vmalloc_fault(address) >= 0) | |
1302 | return; | |
1303 | ||
1304 | if (kmemcheck_fault(regs, address, error_code)) | |
1305 | return; | |
1306 | } | |
5b727a3b | 1307 | |
2d4a7167 | 1308 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1309 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1310 | return; |
1311 | ||
2d4a7167 | 1312 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1313 | if (kprobes_fault(regs)) |
9be260a6 | 1314 | return; |
f8c2ee22 HH |
1315 | /* |
1316 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1317 | * fault we could otherwise deadlock: |
f8c2ee22 | 1318 | */ |
7b2d0dba | 1319 | bad_area_nosemaphore(regs, error_code, address, NULL); |
2d4a7167 | 1320 | |
92181f19 | 1321 | return; |
f8c2ee22 HH |
1322 | } |
1323 | ||
2d4a7167 | 1324 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1325 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1326 | return; |
8c914cb7 | 1327 | |
66c58156 | 1328 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1329 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1330 | |
4640c7ee | 1331 | if (unlikely(smap_violation(error_code, regs))) { |
7b2d0dba | 1332 | bad_area_nosemaphore(regs, error_code, address, NULL); |
4640c7ee | 1333 | return; |
40d3cd66 PA |
1334 | } |
1335 | ||
1da177e4 | 1336 | /* |
2d4a7167 | 1337 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1338 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1339 | */ |
70ffdb93 | 1340 | if (unlikely(faulthandler_disabled() || !mm)) { |
7b2d0dba | 1341 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1342 | return; |
1343 | } | |
1da177e4 | 1344 | |
e00b12e6 PZ |
1345 | /* |
1346 | * It's safe to allow irq's after cr2 has been saved and the | |
1347 | * vmalloc fault has been handled. | |
1348 | * | |
1349 | * User-mode registers count as a user access even for any | |
1350 | * potential system fault or CPU buglet: | |
1351 | */ | |
f39b6f0e | 1352 | if (user_mode(regs)) { |
e00b12e6 PZ |
1353 | local_irq_enable(); |
1354 | error_code |= PF_USER; | |
1355 | flags |= FAULT_FLAG_USER; | |
1356 | } else { | |
1357 | if (regs->flags & X86_EFLAGS_IF) | |
1358 | local_irq_enable(); | |
1359 | } | |
1360 | ||
1361 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1362 | ||
759496ba JW |
1363 | if (error_code & PF_WRITE) |
1364 | flags |= FAULT_FLAG_WRITE; | |
d61172b4 DH |
1365 | if (error_code & PF_INSTR) |
1366 | flags |= FAULT_FLAG_INSTRUCTION; | |
759496ba | 1367 | |
3a1dfe6e IM |
1368 | /* |
1369 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1370 | * addresses in user space. All other faults represent errors in |
1371 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1372 | * case of an erroneous fault occurring in a code path which already | |
1373 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1374 | * against the address space. Luckily the kernel only validly | |
1375 | * references user space from well defined areas of code, which are | |
1376 | * listed in the exceptions table. | |
1da177e4 LT |
1377 | * |
1378 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1379 | * the source reference check when there is a possibility of a |
1380 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1381 | * validate the source. If this is invalid we can skip the address | |
1382 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1383 | */ |
92181f19 | 1384 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1385 | if ((error_code & PF_USER) == 0 && |
92181f19 | 1386 | !search_exception_tables(regs->ip)) { |
7b2d0dba | 1387 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1388 | return; |
1389 | } | |
d065bd81 | 1390 | retry: |
1da177e4 | 1391 | down_read(&mm->mmap_sem); |
01006074 PZ |
1392 | } else { |
1393 | /* | |
2d4a7167 IM |
1394 | * The above down_read_trylock() might have succeeded in |
1395 | * which case we'll have missed the might_sleep() from | |
1396 | * down_read(): | |
01006074 PZ |
1397 | */ |
1398 | might_sleep(); | |
1da177e4 LT |
1399 | } |
1400 | ||
1401 | vma = find_vma(mm, address); | |
92181f19 NP |
1402 | if (unlikely(!vma)) { |
1403 | bad_area(regs, error_code, address); | |
1404 | return; | |
1405 | } | |
1406 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1407 | goto good_area; |
92181f19 NP |
1408 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1409 | bad_area(regs, error_code, address); | |
1410 | return; | |
1411 | } | |
33cb5243 | 1412 | if (error_code & PF_USER) { |
6f4d368e HH |
1413 | /* |
1414 | * Accessing the stack below %sp is always a bug. | |
1415 | * The large cushion allows instructions like enter | |
2d4a7167 | 1416 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1417 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1418 | */ |
92181f19 NP |
1419 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1420 | bad_area(regs, error_code, address); | |
1421 | return; | |
1422 | } | |
1da177e4 | 1423 | } |
92181f19 NP |
1424 | if (unlikely(expand_stack(vma, address))) { |
1425 | bad_area(regs, error_code, address); | |
1426 | return; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * Ok, we have a good vm_area for this memory access, so | |
1431 | * we can handle it.. | |
1432 | */ | |
1da177e4 | 1433 | good_area: |
68da336a | 1434 | if (unlikely(access_error(error_code, vma))) { |
7b2d0dba | 1435 | bad_area_access_error(regs, error_code, address, vma); |
92181f19 | 1436 | return; |
1da177e4 LT |
1437 | } |
1438 | ||
1439 | /* | |
1440 | * If for any reason at all we couldn't handle the fault, | |
1441 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1442 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1443 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
cb0631fd VB |
1444 | * |
1445 | * Note that handle_userfault() may also release and reacquire mmap_sem | |
1446 | * (and not return with VM_FAULT_RETRY), when returning to userland to | |
1447 | * repeat the page fault later with a VM_FAULT_NOPAGE retval | |
1448 | * (potentially after handling any pending signal during the return to | |
1449 | * userland). The return to userland is identified whenever | |
1450 | * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. | |
1451 | * Thus we have to be careful about not touching vma after handling the | |
1452 | * fault, so we read the pkey beforehand. | |
1da177e4 | 1453 | */ |
cb0631fd | 1454 | pkey = vma_pkey(vma); |
dcddffd4 | 1455 | fault = handle_mm_fault(vma, address, flags); |
26178ec1 | 1456 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1457 | |
3a13c4d7 | 1458 | /* |
26178ec1 LT |
1459 | * If we need to retry the mmap_sem has already been released, |
1460 | * and if there is a fatal signal pending there is no guarantee | |
1461 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1462 | */ |
26178ec1 LT |
1463 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1464 | /* Retry at most once */ | |
1465 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1466 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1467 | flags |= FAULT_FLAG_TRIED; | |
1468 | if (!fatal_signal_pending(tsk)) | |
1469 | goto retry; | |
1470 | } | |
1471 | ||
1472 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1473 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1474 | return; |
1475 | ||
1476 | /* Not returning to user mode? Handle exceptions or die: */ | |
1477 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1478 | return; |
26178ec1 | 1479 | } |
3a13c4d7 | 1480 | |
26178ec1 | 1481 | up_read(&mm->mmap_sem); |
3a13c4d7 | 1482 | if (unlikely(fault & VM_FAULT_ERROR)) { |
a3c4fb7c | 1483 | mm_fault_error(regs, error_code, address, &pkey, fault); |
3a13c4d7 | 1484 | return; |
37b23e05 KM |
1485 | } |
1486 | ||
d065bd81 | 1487 | /* |
26178ec1 LT |
1488 | * Major/minor page fault accounting. If any of the events |
1489 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1490 | */ |
26178ec1 LT |
1491 | if (major) { |
1492 | tsk->maj_flt++; | |
1493 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1494 | } else { | |
1495 | tsk->min_flt++; | |
1496 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1497 | } |
d729ab35 | 1498 | |
8c938f9f | 1499 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1500 | } |
9326638c | 1501 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1502 | |
9326638c MH |
1503 | static nokprobe_inline void |
1504 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1505 | unsigned long error_code) | |
d34603b0 SA |
1506 | { |
1507 | if (user_mode(regs)) | |
d4078e23 | 1508 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1509 | else |
d4078e23 | 1510 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1511 | } |
1512 | ||
11a7ffb0 TG |
1513 | /* |
1514 | * We must have this function blacklisted from kprobes, tagged with notrace | |
1515 | * and call read_cr2() before calling anything else. To avoid calling any | |
1516 | * kind of tracing machinery before we've observed the CR2 value. | |
1517 | * | |
1518 | * exception_{enter,exit}() contains all sorts of tracepoints. | |
1519 | */ | |
9326638c | 1520 | dotraplinkage void notrace |
11a7ffb0 | 1521 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
25c74b10 | 1522 | { |
11a7ffb0 | 1523 | unsigned long address = read_cr2(); /* Get the faulting address */ |
d4078e23 | 1524 | enum ctx_state prev_state; |
25c74b10 SA |
1525 | |
1526 | prev_state = exception_enter(); | |
80954747 | 1527 | if (trace_pagefault_enabled()) |
11a7ffb0 TG |
1528 | trace_page_fault_entries(address, regs, error_code); |
1529 | ||
0ac09f9f | 1530 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1531 | exception_exit(prev_state); |
1532 | } | |
11a7ffb0 | 1533 | NOKPROBE_SYMBOL(do_page_fault); |