Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * Copyright (C) 1995 Linus Torvalds |
2d4a7167 | 3 | * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. |
f8eeb2e6 | 4 | * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar |
1da177e4 | 5 | */ |
a2bcd473 IM |
6 | #include <linux/sched.h> /* test_thread_flag(), ... */ |
7 | #include <linux/kdebug.h> /* oops_begin/end, ... */ | |
8 | #include <linux/module.h> /* search_exception_table */ | |
9 | #include <linux/bootmem.h> /* max_low_pfn */ | |
9326638c | 10 | #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ |
a2bcd473 | 11 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ |
cdd6c482 | 12 | #include <linux/perf_event.h> /* perf_sw_event */ |
f672b49b | 13 | #include <linux/hugetlb.h> /* hstate_index_to_shift */ |
268bb0ce | 14 | #include <linux/prefetch.h> /* prefetchw */ |
56dd9470 | 15 | #include <linux/context_tracking.h> /* exception_enter(), ... */ |
70ffdb93 | 16 | #include <linux/uaccess.h> /* faulthandler_disabled() */ |
2d4a7167 | 17 | |
a2bcd473 IM |
18 | #include <asm/traps.h> /* dotraplinkage, ... */ |
19 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f8561296 | 20 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
f40c3300 AL |
21 | #include <asm/fixmap.h> /* VSYSCALL_ADDR */ |
22 | #include <asm/vsyscall.h> /* emulate_vsyscall */ | |
ba3e127e | 23 | #include <asm/vm86.h> /* struct vm86 */ |
1da177e4 | 24 | |
d34603b0 SA |
25 | #define CREATE_TRACE_POINTS |
26 | #include <asm/trace/exceptions.h> | |
27 | ||
33cb5243 | 28 | /* |
2d4a7167 IM |
29 | * Page fault error code bits: |
30 | * | |
31 | * bit 0 == 0: no page found 1: protection fault | |
32 | * bit 1 == 0: read access 1: write access | |
33 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
34 | * bit 3 == 1: use of reserved bit detected | |
35 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 36 | */ |
2d4a7167 IM |
37 | enum x86_pf_error_code { |
38 | ||
39 | PF_PROT = 1 << 0, | |
40 | PF_WRITE = 1 << 1, | |
41 | PF_USER = 1 << 2, | |
42 | PF_RSVD = 1 << 3, | |
43 | PF_INSTR = 1 << 4, | |
44 | }; | |
66c58156 | 45 | |
b814d41f | 46 | /* |
b319eed0 IM |
47 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
48 | * handled by mmiotrace: | |
b814d41f | 49 | */ |
9326638c | 50 | static nokprobe_inline int |
62c9295f | 51 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 52 | { |
0fd0e3da PP |
53 | if (unlikely(is_kmmio_active())) |
54 | if (kmmio_handler(regs, addr) == 1) | |
55 | return -1; | |
0fd0e3da | 56 | return 0; |
86069782 PP |
57 | } |
58 | ||
9326638c | 59 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 60 | { |
74a0b576 CH |
61 | int ret = 0; |
62 | ||
63 | /* kprobe_running() needs smp_processor_id() */ | |
f39b6f0e | 64 | if (kprobes_built_in() && !user_mode(regs)) { |
74a0b576 CH |
65 | preempt_disable(); |
66 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
67 | ret = 1; | |
68 | preempt_enable(); | |
69 | } | |
1bd858a5 | 70 | |
74a0b576 | 71 | return ret; |
33cb5243 | 72 | } |
1bd858a5 | 73 | |
1dc85be0 | 74 | /* |
2d4a7167 IM |
75 | * Prefetch quirks: |
76 | * | |
77 | * 32-bit mode: | |
78 | * | |
79 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
80 | * Check that here and ignore it. | |
1dc85be0 | 81 | * |
2d4a7167 | 82 | * 64-bit mode: |
1dc85be0 | 83 | * |
2d4a7167 IM |
84 | * Sometimes the CPU reports invalid exceptions on prefetch. |
85 | * Check that here and ignore it. | |
86 | * | |
87 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 88 | */ |
107a0367 IM |
89 | static inline int |
90 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
91 | unsigned char opcode, int *prefetch) | |
92 | { | |
93 | unsigned char instr_hi = opcode & 0xf0; | |
94 | unsigned char instr_lo = opcode & 0x0f; | |
95 | ||
96 | switch (instr_hi) { | |
97 | case 0x20: | |
98 | case 0x30: | |
99 | /* | |
100 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
101 | * In X86_64 long mode, the CPU will signal invalid | |
102 | * opcode if some of these prefixes are present so | |
103 | * X86_64 will never get here anyway | |
104 | */ | |
105 | return ((instr_lo & 7) == 0x6); | |
106 | #ifdef CONFIG_X86_64 | |
107 | case 0x40: | |
108 | /* | |
109 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
110 | * Need to figure out under what instruction mode the | |
111 | * instruction was issued. Could check the LDT for lm, | |
112 | * but for now it's good enough to assume that long | |
113 | * mode only uses well known segments or kernel. | |
114 | */ | |
318f5a2a | 115 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
116 | #endif |
117 | case 0x60: | |
118 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
119 | return (instr_lo & 0xC) == 0x4; | |
120 | case 0xF0: | |
121 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
122 | return !instr_lo || (instr_lo>>1) == 1; | |
123 | case 0x00: | |
124 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
125 | if (probe_kernel_address(instr, opcode)) | |
126 | return 0; | |
127 | ||
128 | *prefetch = (instr_lo == 0xF) && | |
129 | (opcode == 0x0D || opcode == 0x18); | |
130 | return 0; | |
131 | default: | |
132 | return 0; | |
133 | } | |
134 | } | |
135 | ||
2d4a7167 IM |
136 | static int |
137 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 138 | { |
2d4a7167 | 139 | unsigned char *max_instr; |
ab2bf0c1 | 140 | unsigned char *instr; |
33cb5243 | 141 | int prefetch = 0; |
1da177e4 | 142 | |
3085354d IM |
143 | /* |
144 | * If it was a exec (instruction fetch) fault on NX page, then | |
145 | * do not ignore the fault: | |
146 | */ | |
66c58156 | 147 | if (error_code & PF_INSTR) |
1da177e4 | 148 | return 0; |
1dc85be0 | 149 | |
107a0367 | 150 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 151 | max_instr = instr + 15; |
1da177e4 | 152 | |
d31bf07f | 153 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
154 | return 0; |
155 | ||
107a0367 | 156 | while (instr < max_instr) { |
2d4a7167 | 157 | unsigned char opcode; |
1da177e4 | 158 | |
ab2bf0c1 | 159 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 160 | break; |
1da177e4 | 161 | |
1da177e4 LT |
162 | instr++; |
163 | ||
107a0367 | 164 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 165 | break; |
1da177e4 LT |
166 | } |
167 | return prefetch; | |
168 | } | |
169 | ||
2d4a7167 IM |
170 | static void |
171 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
f672b49b | 172 | struct task_struct *tsk, int fault) |
c4aba4a8 | 173 | { |
f672b49b | 174 | unsigned lsb = 0; |
c4aba4a8 HH |
175 | siginfo_t info; |
176 | ||
2d4a7167 IM |
177 | info.si_signo = si_signo; |
178 | info.si_errno = 0; | |
179 | info.si_code = si_code; | |
180 | info.si_addr = (void __user *)address; | |
f672b49b AK |
181 | if (fault & VM_FAULT_HWPOISON_LARGE) |
182 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
183 | if (fault & VM_FAULT_HWPOISON) | |
184 | lsb = PAGE_SHIFT; | |
185 | info.si_addr_lsb = lsb; | |
2d4a7167 | 186 | |
c4aba4a8 HH |
187 | force_sig_info(si_signo, &info, tsk); |
188 | } | |
189 | ||
f2f13a85 IM |
190 | DEFINE_SPINLOCK(pgd_lock); |
191 | LIST_HEAD(pgd_list); | |
192 | ||
193 | #ifdef CONFIG_X86_32 | |
194 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 195 | { |
f2f13a85 IM |
196 | unsigned index = pgd_index(address); |
197 | pgd_t *pgd_k; | |
198 | pud_t *pud, *pud_k; | |
199 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 200 | |
f2f13a85 IM |
201 | pgd += index; |
202 | pgd_k = init_mm.pgd + index; | |
203 | ||
204 | if (!pgd_present(*pgd_k)) | |
205 | return NULL; | |
206 | ||
207 | /* | |
208 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
209 | * and redundant with the set_pmd() on non-PAE. As would | |
210 | * set_pud. | |
211 | */ | |
212 | pud = pud_offset(pgd, address); | |
213 | pud_k = pud_offset(pgd_k, address); | |
214 | if (!pud_present(*pud_k)) | |
215 | return NULL; | |
216 | ||
217 | pmd = pmd_offset(pud, address); | |
218 | pmd_k = pmd_offset(pud_k, address); | |
219 | if (!pmd_present(*pmd_k)) | |
220 | return NULL; | |
221 | ||
b8bcfe99 | 222 | if (!pmd_present(*pmd)) |
f2f13a85 | 223 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 224 | else |
f2f13a85 | 225 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
226 | |
227 | return pmd_k; | |
228 | } | |
229 | ||
230 | void vmalloc_sync_all(void) | |
231 | { | |
232 | unsigned long address; | |
233 | ||
234 | if (SHARED_KERNEL_PMD) | |
235 | return; | |
236 | ||
237 | for (address = VMALLOC_START & PMD_MASK; | |
238 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
239 | address += PMD_SIZE) { | |
f2f13a85 IM |
240 | struct page *page; |
241 | ||
a79e53d8 | 242 | spin_lock(&pgd_lock); |
f2f13a85 | 243 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 244 | spinlock_t *pgt_lock; |
f01f7c56 | 245 | pmd_t *ret; |
617d34d9 | 246 | |
a79e53d8 | 247 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
248 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
249 | ||
250 | spin_lock(pgt_lock); | |
251 | ret = vmalloc_sync_one(page_address(page), address); | |
252 | spin_unlock(pgt_lock); | |
253 | ||
254 | if (!ret) | |
f2f13a85 IM |
255 | break; |
256 | } | |
a79e53d8 | 257 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
258 | } |
259 | } | |
260 | ||
261 | /* | |
262 | * 32-bit: | |
263 | * | |
264 | * Handle a fault on the vmalloc or module mapping area | |
265 | */ | |
9326638c | 266 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
267 | { |
268 | unsigned long pgd_paddr; | |
269 | pmd_t *pmd_k; | |
270 | pte_t *pte_k; | |
271 | ||
272 | /* Make sure we are in vmalloc area: */ | |
273 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
274 | return -1; | |
275 | ||
ebc8827f FW |
276 | WARN_ON_ONCE(in_nmi()); |
277 | ||
f2f13a85 IM |
278 | /* |
279 | * Synchronize this task's top level page-table | |
280 | * with the 'reference' page table. | |
281 | * | |
282 | * Do _not_ use "current" here. We might be inside | |
283 | * an interrupt in the middle of a task switch.. | |
284 | */ | |
285 | pgd_paddr = read_cr3(); | |
286 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
287 | if (!pmd_k) | |
288 | return -1; | |
289 | ||
290 | pte_k = pte_offset_kernel(pmd_k, address); | |
291 | if (!pte_present(*pte_k)) | |
292 | return -1; | |
293 | ||
294 | return 0; | |
295 | } | |
9326638c | 296 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
297 | |
298 | /* | |
299 | * Did it hit the DOS screen memory VA from vm86 mode? | |
300 | */ | |
301 | static inline void | |
302 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
303 | struct task_struct *tsk) | |
304 | { | |
9fda6a06 | 305 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
306 | unsigned long bit; |
307 | ||
9fda6a06 | 308 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
309 | return; |
310 | ||
311 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
312 | if (bit < 32) | |
9fda6a06 BG |
313 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
314 | #endif | |
33cb5243 | 315 | } |
1da177e4 | 316 | |
087975b0 | 317 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 318 | { |
087975b0 AM |
319 | return pfn < max_low_pfn; |
320 | } | |
1156e098 | 321 | |
087975b0 AM |
322 | static void dump_pagetable(unsigned long address) |
323 | { | |
324 | pgd_t *base = __va(read_cr3()); | |
325 | pgd_t *pgd = &base[pgd_index(address)]; | |
326 | pmd_t *pmd; | |
327 | pte_t *pte; | |
2d4a7167 | 328 | |
1156e098 | 329 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
330 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
331 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
332 | goto out; | |
1156e098 | 333 | #endif |
087975b0 AM |
334 | pmd = pmd_offset(pud_offset(pgd, address), address); |
335 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
336 | |
337 | /* | |
338 | * We must not directly access the pte in the highpte | |
339 | * case if the page table is located in highmem. | |
340 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 341 | * it's allocated already: |
1156e098 | 342 | */ |
087975b0 AM |
343 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
344 | goto out; | |
1156e098 | 345 | |
087975b0 AM |
346 | pte = pte_offset_kernel(pmd, address); |
347 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
348 | out: | |
1156e098 | 349 | printk("\n"); |
f2f13a85 IM |
350 | } |
351 | ||
352 | #else /* CONFIG_X86_64: */ | |
353 | ||
354 | void vmalloc_sync_all(void) | |
355 | { | |
9661d5bc | 356 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); |
f2f13a85 IM |
357 | } |
358 | ||
359 | /* | |
360 | * 64-bit: | |
361 | * | |
362 | * Handle a fault on the vmalloc area | |
363 | * | |
364 | * This assumes no large pages in there. | |
365 | */ | |
9326638c | 366 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
367 | { |
368 | pgd_t *pgd, *pgd_ref; | |
369 | pud_t *pud, *pud_ref; | |
370 | pmd_t *pmd, *pmd_ref; | |
371 | pte_t *pte, *pte_ref; | |
372 | ||
373 | /* Make sure we are in vmalloc area: */ | |
374 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
375 | return -1; | |
376 | ||
ebc8827f FW |
377 | WARN_ON_ONCE(in_nmi()); |
378 | ||
f2f13a85 IM |
379 | /* |
380 | * Copy kernel mappings over when needed. This can also | |
381 | * happen within a race in page table update. In the later | |
382 | * case just flush: | |
383 | */ | |
384 | pgd = pgd_offset(current->active_mm, address); | |
385 | pgd_ref = pgd_offset_k(address); | |
386 | if (pgd_none(*pgd_ref)) | |
387 | return -1; | |
388 | ||
1160c277 | 389 | if (pgd_none(*pgd)) { |
f2f13a85 | 390 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
391 | arch_flush_lazy_mmu_mode(); |
392 | } else { | |
f2f13a85 | 393 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 394 | } |
f2f13a85 IM |
395 | |
396 | /* | |
397 | * Below here mismatches are bugs because these lower tables | |
398 | * are shared: | |
399 | */ | |
400 | ||
401 | pud = pud_offset(pgd, address); | |
402 | pud_ref = pud_offset(pgd_ref, address); | |
403 | if (pud_none(*pud_ref)) | |
404 | return -1; | |
405 | ||
406 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
407 | BUG(); | |
408 | ||
409 | pmd = pmd_offset(pud, address); | |
410 | pmd_ref = pmd_offset(pud_ref, address); | |
411 | if (pmd_none(*pmd_ref)) | |
412 | return -1; | |
413 | ||
414 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
415 | BUG(); | |
416 | ||
417 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
418 | if (!pte_present(*pte_ref)) | |
419 | return -1; | |
420 | ||
421 | pte = pte_offset_kernel(pmd, address); | |
422 | ||
423 | /* | |
424 | * Don't use pte_page here, because the mappings can point | |
425 | * outside mem_map, and the NUMA hash lookup cannot handle | |
426 | * that: | |
427 | */ | |
428 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
429 | BUG(); | |
430 | ||
431 | return 0; | |
432 | } | |
9326638c | 433 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 434 | |
e05139f2 | 435 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 436 | static const char errata93_warning[] = |
ad361c98 JP |
437 | KERN_ERR |
438 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
439 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
440 | "******* Please consider a BIOS update.\n" | |
441 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 442 | #endif |
f2f13a85 IM |
443 | |
444 | /* | |
445 | * No vm86 mode in 64-bit mode: | |
446 | */ | |
447 | static inline void | |
448 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
449 | struct task_struct *tsk) | |
450 | { | |
451 | } | |
452 | ||
453 | static int bad_address(void *p) | |
454 | { | |
455 | unsigned long dummy; | |
456 | ||
457 | return probe_kernel_address((unsigned long *)p, dummy); | |
458 | } | |
459 | ||
460 | static void dump_pagetable(unsigned long address) | |
461 | { | |
087975b0 AM |
462 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
463 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
464 | pud_t *pud; |
465 | pmd_t *pmd; | |
466 | pte_t *pte; | |
467 | ||
2d4a7167 IM |
468 | if (bad_address(pgd)) |
469 | goto bad; | |
470 | ||
d646bce4 | 471 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
472 | |
473 | if (!pgd_present(*pgd)) | |
474 | goto out; | |
1da177e4 | 475 | |
d2ae5b5f | 476 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
477 | if (bad_address(pud)) |
478 | goto bad; | |
479 | ||
1da177e4 | 480 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 481 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 482 | goto out; |
1da177e4 LT |
483 | |
484 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
485 | if (bad_address(pmd)) |
486 | goto bad; | |
487 | ||
1da177e4 | 488 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
489 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
490 | goto out; | |
1da177e4 LT |
491 | |
492 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
493 | if (bad_address(pte)) |
494 | goto bad; | |
495 | ||
33cb5243 | 496 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 497 | out: |
1da177e4 LT |
498 | printk("\n"); |
499 | return; | |
500 | bad: | |
501 | printk("BAD\n"); | |
8c938f9f IM |
502 | } |
503 | ||
f2f13a85 | 504 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 505 | |
2d4a7167 IM |
506 | /* |
507 | * Workaround for K8 erratum #93 & buggy BIOS. | |
508 | * | |
509 | * BIOS SMM functions are required to use a specific workaround | |
510 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
511 | * | |
512 | * A lot of BIOS that didn't get tested properly miss this. | |
513 | * | |
514 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
515 | * Try to work around it here. | |
516 | * | |
517 | * Note we only handle faults in kernel here. | |
518 | * Does nothing on 32-bit. | |
fdfe8aa8 | 519 | */ |
33cb5243 | 520 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 521 | { |
e05139f2 JB |
522 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
523 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
524 | || boot_cpu_data.x86 != 0xf) | |
525 | return 0; | |
526 | ||
65ea5b03 | 527 | if (address != regs->ip) |
1da177e4 | 528 | return 0; |
2d4a7167 | 529 | |
33cb5243 | 530 | if ((address >> 32) != 0) |
1da177e4 | 531 | return 0; |
2d4a7167 | 532 | |
1da177e4 | 533 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
534 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
535 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 536 | printk_once(errata93_warning); |
65ea5b03 | 537 | regs->ip = address; |
1da177e4 LT |
538 | return 1; |
539 | } | |
fdfe8aa8 | 540 | #endif |
1da177e4 | 541 | return 0; |
33cb5243 | 542 | } |
1da177e4 | 543 | |
35f3266f | 544 | /* |
2d4a7167 IM |
545 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
546 | * to illegal addresses >4GB. | |
547 | * | |
548 | * We catch this in the page fault handler because these addresses | |
549 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
550 | * segment in LDT is compatibility mode. |
551 | */ | |
552 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
553 | { | |
554 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 555 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
556 | return 1; |
557 | #endif | |
558 | return 0; | |
559 | } | |
560 | ||
29caf2f9 HH |
561 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
562 | { | |
563 | #ifdef CONFIG_X86_F00F_BUG | |
564 | unsigned long nr; | |
2d4a7167 | 565 | |
29caf2f9 | 566 | /* |
2d4a7167 | 567 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 568 | */ |
e2604b49 | 569 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
570 | nr = (address - idt_descr.address) >> 3; |
571 | ||
572 | if (nr == 6) { | |
573 | do_invalid_op(regs, 0); | |
574 | return 1; | |
575 | } | |
576 | } | |
577 | #endif | |
578 | return 0; | |
579 | } | |
580 | ||
8f766149 IM |
581 | static const char nx_warning[] = KERN_CRIT |
582 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
583 | static const char smep_warning[] = KERN_CRIT |
584 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 585 | |
2d4a7167 IM |
586 | static void |
587 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
588 | unsigned long address) | |
b3279c7f | 589 | { |
1156e098 HH |
590 | if (!oops_may_print()) |
591 | return; | |
592 | ||
1156e098 | 593 | if (error_code & PF_INSTR) { |
93809be8 | 594 | unsigned int level; |
426e34cc MF |
595 | pgd_t *pgd; |
596 | pte_t *pte; | |
2d4a7167 | 597 | |
426e34cc MF |
598 | pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
599 | pgd += pgd_index(address); | |
600 | ||
601 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 602 | |
8f766149 | 603 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 604 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
605 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
606 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 607 | (__read_cr4() & X86_CR4_SMEP)) |
eff50c34 | 608 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 609 | } |
1156e098 | 610 | |
19f0dda9 | 611 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 612 | if (address < PAGE_SIZE) |
19f0dda9 | 613 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 614 | else |
19f0dda9 | 615 | printk(KERN_CONT "paging request"); |
2d4a7167 | 616 | |
f294a8ce | 617 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 618 | printk(KERN_ALERT "IP:"); |
5f01c988 | 619 | printk_address(regs->ip); |
2d4a7167 | 620 | |
b3279c7f HH |
621 | dump_pagetable(address); |
622 | } | |
623 | ||
2d4a7167 IM |
624 | static noinline void |
625 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
626 | unsigned long address) | |
1da177e4 | 627 | { |
2d4a7167 IM |
628 | struct task_struct *tsk; |
629 | unsigned long flags; | |
630 | int sig; | |
631 | ||
632 | flags = oops_begin(); | |
633 | tsk = current; | |
634 | sig = SIGKILL; | |
1209140c | 635 | |
1da177e4 | 636 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 637 | tsk->comm, address); |
1da177e4 | 638 | dump_pagetable(address); |
2d4a7167 IM |
639 | |
640 | tsk->thread.cr2 = address; | |
51e7dc70 | 641 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
642 | tsk->thread.error_code = error_code; |
643 | ||
22f5991c | 644 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 645 | sig = 0; |
2d4a7167 | 646 | |
874d93d1 | 647 | oops_end(flags, regs, sig); |
1da177e4 LT |
648 | } |
649 | ||
2d4a7167 IM |
650 | static noinline void |
651 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 652 | unsigned long address, int signal, int si_code) |
92181f19 NP |
653 | { |
654 | struct task_struct *tsk = current; | |
92181f19 NP |
655 | unsigned long flags; |
656 | int sig; | |
92181f19 | 657 | |
2d4a7167 | 658 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 659 | if (fixup_exception(regs)) { |
c026b359 PZ |
660 | /* |
661 | * Any interrupt that takes a fault gets the fixup. This makes | |
662 | * the below recursive fault logic only apply to a faults from | |
663 | * task context. | |
664 | */ | |
665 | if (in_interrupt()) | |
666 | return; | |
667 | ||
668 | /* | |
669 | * Per the above we're !in_interrupt(), aka. task context. | |
670 | * | |
671 | * In this case we need to make sure we're not recursively | |
672 | * faulting through the emulate_vsyscall() logic. | |
673 | */ | |
4fc34901 | 674 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 675 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
676 | tsk->thread.error_code = error_code | PF_USER; |
677 | tsk->thread.cr2 = address; | |
678 | ||
679 | /* XXX: hwpoison faults will set the wrong code. */ | |
680 | force_sig_info_fault(signal, si_code, address, tsk, 0); | |
681 | } | |
c026b359 PZ |
682 | |
683 | /* | |
684 | * Barring that, we can do the fixup and be happy. | |
685 | */ | |
92181f19 | 686 | return; |
4fc34901 | 687 | } |
92181f19 NP |
688 | |
689 | /* | |
2d4a7167 IM |
690 | * 32-bit: |
691 | * | |
692 | * Valid to do another page fault here, because if this fault | |
693 | * had been triggered by is_prefetch fixup_exception would have | |
694 | * handled it. | |
695 | * | |
696 | * 64-bit: | |
92181f19 | 697 | * |
2d4a7167 | 698 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
699 | */ |
700 | if (is_prefetch(regs, error_code, address)) | |
701 | return; | |
702 | ||
703 | if (is_errata93(regs, address)) | |
704 | return; | |
705 | ||
706 | /* | |
707 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 708 | * terminate things with extreme prejudice: |
92181f19 | 709 | */ |
92181f19 | 710 | flags = oops_begin(); |
92181f19 NP |
711 | |
712 | show_fault_oops(regs, error_code, address); | |
713 | ||
a70857e4 | 714 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 715 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 716 | |
1cc99544 | 717 | tsk->thread.cr2 = address; |
51e7dc70 | 718 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 719 | tsk->thread.error_code = error_code; |
92181f19 | 720 | |
92181f19 NP |
721 | sig = SIGKILL; |
722 | if (__die("Oops", regs, error_code)) | |
723 | sig = 0; | |
2d4a7167 | 724 | |
92181f19 | 725 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 726 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 727 | |
92181f19 | 728 | oops_end(flags, regs, sig); |
92181f19 NP |
729 | } |
730 | ||
2d4a7167 IM |
731 | /* |
732 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
733 | * sysctl is set: | |
734 | */ | |
735 | static inline void | |
736 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
737 | unsigned long address, struct task_struct *tsk) | |
738 | { | |
739 | if (!unhandled_signal(tsk, SIGSEGV)) | |
740 | return; | |
741 | ||
742 | if (!printk_ratelimit()) | |
743 | return; | |
744 | ||
a1a08d1c | 745 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
746 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
747 | tsk->comm, task_pid_nr(tsk), address, | |
748 | (void *)regs->ip, (void *)regs->sp, error_code); | |
749 | ||
750 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
751 | ||
752 | printk(KERN_CONT "\n"); | |
753 | } | |
754 | ||
755 | static void | |
756 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
757 | unsigned long address, int si_code) | |
92181f19 NP |
758 | { |
759 | struct task_struct *tsk = current; | |
760 | ||
761 | /* User mode accesses just cause a SIGSEGV */ | |
762 | if (error_code & PF_USER) { | |
763 | /* | |
2d4a7167 | 764 | * It's possible to have interrupts off here: |
92181f19 NP |
765 | */ |
766 | local_irq_enable(); | |
767 | ||
768 | /* | |
769 | * Valid to do another page fault here because this one came | |
2d4a7167 | 770 | * from user space: |
92181f19 NP |
771 | */ |
772 | if (is_prefetch(regs, error_code, address)) | |
773 | return; | |
774 | ||
775 | if (is_errata100(regs, address)) | |
776 | return; | |
777 | ||
3ae36655 AL |
778 | #ifdef CONFIG_X86_64 |
779 | /* | |
780 | * Instruction fetch faults in the vsyscall page might need | |
781 | * emulation. | |
782 | */ | |
783 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 784 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
785 | if (emulate_vsyscall(regs, address)) |
786 | return; | |
787 | } | |
788 | #endif | |
e575a86f KC |
789 | /* Kernel addresses are always protection faults: */ |
790 | if (address >= TASK_SIZE) | |
791 | error_code |= PF_PROT; | |
3ae36655 | 792 | |
e575a86f | 793 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
794 | show_signal_msg(regs, error_code, address, tsk); |
795 | ||
2d4a7167 | 796 | tsk->thread.cr2 = address; |
e575a86f | 797 | tsk->thread.error_code = error_code; |
51e7dc70 | 798 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 799 | |
f672b49b | 800 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); |
2d4a7167 | 801 | |
92181f19 NP |
802 | return; |
803 | } | |
804 | ||
805 | if (is_f00f_bug(regs, address)) | |
806 | return; | |
807 | ||
4fc34901 | 808 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
809 | } |
810 | ||
2d4a7167 IM |
811 | static noinline void |
812 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
813 | unsigned long address) | |
92181f19 NP |
814 | { |
815 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
816 | } | |
817 | ||
2d4a7167 IM |
818 | static void |
819 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
820 | unsigned long address, int si_code) | |
92181f19 NP |
821 | { |
822 | struct mm_struct *mm = current->mm; | |
823 | ||
824 | /* | |
825 | * Something tried to access memory that isn't in our memory map.. | |
826 | * Fix it, but check if it's kernel or user first.. | |
827 | */ | |
828 | up_read(&mm->mmap_sem); | |
829 | ||
830 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
831 | } | |
832 | ||
2d4a7167 IM |
833 | static noinline void |
834 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
835 | { |
836 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
837 | } | |
838 | ||
2d4a7167 IM |
839 | static noinline void |
840 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
841 | unsigned long address) | |
92181f19 NP |
842 | { |
843 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
844 | } | |
845 | ||
2d4a7167 | 846 | static void |
a6e04aa9 AK |
847 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
848 | unsigned int fault) | |
92181f19 NP |
849 | { |
850 | struct task_struct *tsk = current; | |
a6e04aa9 | 851 | int code = BUS_ADRERR; |
92181f19 | 852 | |
2d4a7167 | 853 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 854 | if (!(error_code & PF_USER)) { |
4fc34901 | 855 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
856 | return; |
857 | } | |
2d4a7167 | 858 | |
cd1b68f0 | 859 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
860 | if (is_prefetch(regs, error_code, address)) |
861 | return; | |
2d4a7167 IM |
862 | |
863 | tsk->thread.cr2 = address; | |
864 | tsk->thread.error_code = error_code; | |
51e7dc70 | 865 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 866 | |
a6e04aa9 | 867 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 868 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
869 | printk(KERN_ERR |
870 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
871 | tsk->comm, tsk->pid, address); | |
872 | code = BUS_MCEERR_AR; | |
873 | } | |
874 | #endif | |
f672b49b | 875 | force_sig_info_fault(SIGBUS, code, address, tsk, fault); |
92181f19 NP |
876 | } |
877 | ||
3a13c4d7 | 878 | static noinline void |
2d4a7167 IM |
879 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
880 | unsigned long address, unsigned int fault) | |
92181f19 | 881 | { |
3a13c4d7 | 882 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
3a13c4d7 JW |
883 | no_context(regs, error_code, address, 0, 0); |
884 | return; | |
b80ef10e | 885 | } |
b80ef10e | 886 | |
2d4a7167 | 887 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
888 | /* Kernel mode? Handle exceptions or die: */ |
889 | if (!(error_code & PF_USER)) { | |
4fc34901 AL |
890 | no_context(regs, error_code, address, |
891 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 892 | return; |
f8626854 AV |
893 | } |
894 | ||
c2d23f91 DR |
895 | /* |
896 | * We ran out of memory, call the OOM killer, and return the | |
897 | * userspace (which will retry the fault, or kill us if we got | |
898 | * oom-killed): | |
899 | */ | |
900 | pagefault_out_of_memory(); | |
2d4a7167 | 901 | } else { |
f672b49b AK |
902 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
903 | VM_FAULT_HWPOISON_LARGE)) | |
a6e04aa9 | 904 | do_sigbus(regs, error_code, address, fault); |
33692f27 LT |
905 | else if (fault & VM_FAULT_SIGSEGV) |
906 | bad_area_nosemaphore(regs, error_code, address); | |
2d4a7167 IM |
907 | else |
908 | BUG(); | |
909 | } | |
92181f19 NP |
910 | } |
911 | ||
d8b57bb7 TG |
912 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
913 | { | |
914 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
915 | return 0; | |
2d4a7167 | 916 | |
d8b57bb7 TG |
917 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
918 | return 0; | |
919 | ||
920 | return 1; | |
921 | } | |
922 | ||
5b727a3b | 923 | /* |
2d4a7167 IM |
924 | * Handle a spurious fault caused by a stale TLB entry. |
925 | * | |
926 | * This allows us to lazily refresh the TLB when increasing the | |
927 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
928 | * eagerly is very expensive since that implies doing a full | |
929 | * cross-processor TLB flush, even if no stale TLB entries exist | |
930 | * on other processors. | |
931 | * | |
31668511 DV |
932 | * Spurious faults may only occur if the TLB contains an entry with |
933 | * fewer permission than the page table entry. Non-present (P = 0) | |
934 | * and reserved bit (R = 1) faults are never spurious. | |
935 | * | |
5b727a3b JF |
936 | * There are no security implications to leaving a stale TLB when |
937 | * increasing the permissions on a page. | |
31668511 DV |
938 | * |
939 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
940 | * | |
941 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
942 | * (Optional Invalidation). | |
5b727a3b | 943 | */ |
9326638c | 944 | static noinline int |
2d4a7167 | 945 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
946 | { |
947 | pgd_t *pgd; | |
948 | pud_t *pud; | |
949 | pmd_t *pmd; | |
950 | pte_t *pte; | |
3c3e5694 | 951 | int ret; |
5b727a3b | 952 | |
31668511 DV |
953 | /* |
954 | * Only writes to RO or instruction fetches from NX may cause | |
955 | * spurious faults. | |
956 | * | |
957 | * These could be from user or supervisor accesses but the TLB | |
958 | * is only lazily flushed after a kernel mapping protection | |
959 | * change, so user accesses are not expected to cause spurious | |
960 | * faults. | |
961 | */ | |
962 | if (error_code != (PF_WRITE | PF_PROT) | |
963 | && error_code != (PF_INSTR | PF_PROT)) | |
5b727a3b JF |
964 | return 0; |
965 | ||
966 | pgd = init_mm.pgd + pgd_index(address); | |
967 | if (!pgd_present(*pgd)) | |
968 | return 0; | |
969 | ||
970 | pud = pud_offset(pgd, address); | |
971 | if (!pud_present(*pud)) | |
972 | return 0; | |
973 | ||
d8b57bb7 TG |
974 | if (pud_large(*pud)) |
975 | return spurious_fault_check(error_code, (pte_t *) pud); | |
976 | ||
5b727a3b JF |
977 | pmd = pmd_offset(pud, address); |
978 | if (!pmd_present(*pmd)) | |
979 | return 0; | |
980 | ||
d8b57bb7 TG |
981 | if (pmd_large(*pmd)) |
982 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
983 | ||
5b727a3b | 984 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 985 | if (!pte_present(*pte)) |
5b727a3b JF |
986 | return 0; |
987 | ||
3c3e5694 SR |
988 | ret = spurious_fault_check(error_code, pte); |
989 | if (!ret) | |
990 | return 0; | |
991 | ||
992 | /* | |
2d4a7167 IM |
993 | * Make sure we have permissions in PMD. |
994 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
995 | */ |
996 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
997 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 998 | |
3c3e5694 | 999 | return ret; |
5b727a3b | 1000 | } |
9326638c | 1001 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 1002 | |
abd4f750 | 1003 | int show_unhandled_signals = 1; |
1da177e4 | 1004 | |
2d4a7167 | 1005 | static inline int |
68da336a | 1006 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1007 | { |
68da336a | 1008 | if (error_code & PF_WRITE) { |
2d4a7167 | 1009 | /* write, present and write, not present: */ |
92181f19 NP |
1010 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1011 | return 1; | |
2d4a7167 | 1012 | return 0; |
92181f19 NP |
1013 | } |
1014 | ||
2d4a7167 IM |
1015 | /* read, present: */ |
1016 | if (unlikely(error_code & PF_PROT)) | |
1017 | return 1; | |
1018 | ||
1019 | /* read, not present: */ | |
1020 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1021 | return 1; | |
1022 | ||
92181f19 NP |
1023 | return 0; |
1024 | } | |
1025 | ||
0973a06c HS |
1026 | static int fault_in_kernel_space(unsigned long address) |
1027 | { | |
d9517346 | 1028 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1029 | } |
1030 | ||
40d3cd66 PA |
1031 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1032 | { | |
4640c7ee PA |
1033 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1034 | return false; | |
1035 | ||
1036 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1037 | return false; | |
1038 | ||
40d3cd66 PA |
1039 | if (error_code & PF_USER) |
1040 | return false; | |
1041 | ||
f39b6f0e | 1042 | if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) |
40d3cd66 PA |
1043 | return false; |
1044 | ||
1045 | return true; | |
1046 | } | |
1047 | ||
1da177e4 LT |
1048 | /* |
1049 | * This routine handles page faults. It determines the address, | |
1050 | * and the problem, and then passes it off to one of the appropriate | |
1051 | * routines. | |
d4078e23 PZ |
1052 | * |
1053 | * This function must have noinline because both callers | |
1054 | * {,trace_}do_page_fault() have notrace on. Having this an actual function | |
1055 | * guarantees there's a function trace entry. | |
1da177e4 | 1056 | */ |
9326638c | 1057 | static noinline void |
0ac09f9f JO |
1058 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1059 | unsigned long address) | |
1da177e4 | 1060 | { |
2d4a7167 | 1061 | struct vm_area_struct *vma; |
1da177e4 LT |
1062 | struct task_struct *tsk; |
1063 | struct mm_struct *mm; | |
26178ec1 | 1064 | int fault, major = 0; |
759496ba | 1065 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1066 | |
a9ba9a3b AV |
1067 | tsk = current; |
1068 | mm = tsk->mm; | |
2d4a7167 | 1069 | |
f8561296 VN |
1070 | /* |
1071 | * Detect and handle instructions that would cause a page fault for | |
1072 | * both a tracked kernel page and a userspace page. | |
1073 | */ | |
1074 | if (kmemcheck_active(regs)) | |
1075 | kmemcheck_hide(regs); | |
5dfaf90f | 1076 | prefetchw(&mm->mmap_sem); |
f8561296 | 1077 | |
0fd0e3da | 1078 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1079 | return; |
1da177e4 LT |
1080 | |
1081 | /* | |
1082 | * We fault-in kernel-space virtual memory on-demand. The | |
1083 | * 'reference' page table is init_mm.pgd. | |
1084 | * | |
1085 | * NOTE! We MUST NOT take any locks for this case. We may | |
1086 | * be in an interrupt or a critical region, and should | |
1087 | * only copy the information from the master page table, | |
1088 | * nothing more. | |
1089 | * | |
1090 | * This verifies that the fault happens in kernel space | |
1091 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1092 | * protection error (error_code & 9) == 0. |
1da177e4 | 1093 | */ |
0973a06c | 1094 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1095 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1096 | if (vmalloc_fault(address) >= 0) | |
1097 | return; | |
1098 | ||
1099 | if (kmemcheck_fault(regs, address, error_code)) | |
1100 | return; | |
1101 | } | |
5b727a3b | 1102 | |
2d4a7167 | 1103 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1104 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1105 | return; |
1106 | ||
2d4a7167 | 1107 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1108 | if (kprobes_fault(regs)) |
9be260a6 | 1109 | return; |
f8c2ee22 HH |
1110 | /* |
1111 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1112 | * fault we could otherwise deadlock: |
f8c2ee22 | 1113 | */ |
92181f19 | 1114 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1115 | |
92181f19 | 1116 | return; |
f8c2ee22 HH |
1117 | } |
1118 | ||
2d4a7167 | 1119 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1120 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1121 | return; |
8c914cb7 | 1122 | |
66c58156 | 1123 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1124 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1125 | |
4640c7ee PA |
1126 | if (unlikely(smap_violation(error_code, regs))) { |
1127 | bad_area_nosemaphore(regs, error_code, address); | |
1128 | return; | |
40d3cd66 PA |
1129 | } |
1130 | ||
1da177e4 | 1131 | /* |
2d4a7167 | 1132 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1133 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1134 | */ |
70ffdb93 | 1135 | if (unlikely(faulthandler_disabled() || !mm)) { |
92181f19 NP |
1136 | bad_area_nosemaphore(regs, error_code, address); |
1137 | return; | |
1138 | } | |
1da177e4 | 1139 | |
e00b12e6 PZ |
1140 | /* |
1141 | * It's safe to allow irq's after cr2 has been saved and the | |
1142 | * vmalloc fault has been handled. | |
1143 | * | |
1144 | * User-mode registers count as a user access even for any | |
1145 | * potential system fault or CPU buglet: | |
1146 | */ | |
f39b6f0e | 1147 | if (user_mode(regs)) { |
e00b12e6 PZ |
1148 | local_irq_enable(); |
1149 | error_code |= PF_USER; | |
1150 | flags |= FAULT_FLAG_USER; | |
1151 | } else { | |
1152 | if (regs->flags & X86_EFLAGS_IF) | |
1153 | local_irq_enable(); | |
1154 | } | |
1155 | ||
1156 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1157 | ||
759496ba JW |
1158 | if (error_code & PF_WRITE) |
1159 | flags |= FAULT_FLAG_WRITE; | |
1160 | ||
3a1dfe6e IM |
1161 | /* |
1162 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1163 | * addresses in user space. All other faults represent errors in |
1164 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1165 | * case of an erroneous fault occurring in a code path which already | |
1166 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1167 | * against the address space. Luckily the kernel only validly | |
1168 | * references user space from well defined areas of code, which are | |
1169 | * listed in the exceptions table. | |
1da177e4 LT |
1170 | * |
1171 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1172 | * the source reference check when there is a possibility of a |
1173 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1174 | * validate the source. If this is invalid we can skip the address | |
1175 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1176 | */ |
92181f19 | 1177 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1178 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1179 | !search_exception_tables(regs->ip)) { |
1180 | bad_area_nosemaphore(regs, error_code, address); | |
1181 | return; | |
1182 | } | |
d065bd81 | 1183 | retry: |
1da177e4 | 1184 | down_read(&mm->mmap_sem); |
01006074 PZ |
1185 | } else { |
1186 | /* | |
2d4a7167 IM |
1187 | * The above down_read_trylock() might have succeeded in |
1188 | * which case we'll have missed the might_sleep() from | |
1189 | * down_read(): | |
01006074 PZ |
1190 | */ |
1191 | might_sleep(); | |
1da177e4 LT |
1192 | } |
1193 | ||
1194 | vma = find_vma(mm, address); | |
92181f19 NP |
1195 | if (unlikely(!vma)) { |
1196 | bad_area(regs, error_code, address); | |
1197 | return; | |
1198 | } | |
1199 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1200 | goto good_area; |
92181f19 NP |
1201 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1202 | bad_area(regs, error_code, address); | |
1203 | return; | |
1204 | } | |
33cb5243 | 1205 | if (error_code & PF_USER) { |
6f4d368e HH |
1206 | /* |
1207 | * Accessing the stack below %sp is always a bug. | |
1208 | * The large cushion allows instructions like enter | |
2d4a7167 | 1209 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1210 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1211 | */ |
92181f19 NP |
1212 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1213 | bad_area(regs, error_code, address); | |
1214 | return; | |
1215 | } | |
1da177e4 | 1216 | } |
92181f19 NP |
1217 | if (unlikely(expand_stack(vma, address))) { |
1218 | bad_area(regs, error_code, address); | |
1219 | return; | |
1220 | } | |
1221 | ||
1222 | /* | |
1223 | * Ok, we have a good vm_area for this memory access, so | |
1224 | * we can handle it.. | |
1225 | */ | |
1da177e4 | 1226 | good_area: |
68da336a | 1227 | if (unlikely(access_error(error_code, vma))) { |
92181f19 NP |
1228 | bad_area_access_error(regs, error_code, address); |
1229 | return; | |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | /* | |
1233 | * If for any reason at all we couldn't handle the fault, | |
1234 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1235 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1236 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
1da177e4 | 1237 | */ |
d065bd81 | 1238 | fault = handle_mm_fault(mm, vma, address, flags); |
26178ec1 | 1239 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1240 | |
3a13c4d7 | 1241 | /* |
26178ec1 LT |
1242 | * If we need to retry the mmap_sem has already been released, |
1243 | * and if there is a fatal signal pending there is no guarantee | |
1244 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1245 | */ |
26178ec1 LT |
1246 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1247 | /* Retry at most once */ | |
1248 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1249 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1250 | flags |= FAULT_FLAG_TRIED; | |
1251 | if (!fatal_signal_pending(tsk)) | |
1252 | goto retry; | |
1253 | } | |
1254 | ||
1255 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1256 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1257 | return; |
1258 | ||
1259 | /* Not returning to user mode? Handle exceptions or die: */ | |
1260 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1261 | return; |
26178ec1 | 1262 | } |
3a13c4d7 | 1263 | |
26178ec1 | 1264 | up_read(&mm->mmap_sem); |
3a13c4d7 JW |
1265 | if (unlikely(fault & VM_FAULT_ERROR)) { |
1266 | mm_fault_error(regs, error_code, address, fault); | |
1267 | return; | |
37b23e05 KM |
1268 | } |
1269 | ||
d065bd81 | 1270 | /* |
26178ec1 LT |
1271 | * Major/minor page fault accounting. If any of the events |
1272 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1273 | */ |
26178ec1 LT |
1274 | if (major) { |
1275 | tsk->maj_flt++; | |
1276 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1277 | } else { | |
1278 | tsk->min_flt++; | |
1279 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1280 | } |
d729ab35 | 1281 | |
8c938f9f | 1282 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1283 | } |
9326638c | 1284 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1285 | |
9326638c | 1286 | dotraplinkage void notrace |
6ba3c97a FW |
1287 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1288 | { | |
d4078e23 | 1289 | unsigned long address = read_cr2(); /* Get the faulting address */ |
6c1e0256 | 1290 | enum ctx_state prev_state; |
d4078e23 PZ |
1291 | |
1292 | /* | |
1293 | * We must have this function tagged with __kprobes, notrace and call | |
1294 | * read_cr2() before calling anything else. To avoid calling any kind | |
1295 | * of tracing machinery before we've observed the CR2 value. | |
1296 | * | |
1297 | * exception_{enter,exit}() contain all sorts of tracepoints. | |
1298 | */ | |
6c1e0256 FW |
1299 | |
1300 | prev_state = exception_enter(); | |
0ac09f9f | 1301 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1302 | exception_exit(prev_state); |
6ba3c97a | 1303 | } |
9326638c | 1304 | NOKPROBE_SYMBOL(do_page_fault); |
25c74b10 | 1305 | |
d4078e23 | 1306 | #ifdef CONFIG_TRACING |
9326638c MH |
1307 | static nokprobe_inline void |
1308 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1309 | unsigned long error_code) | |
d34603b0 SA |
1310 | { |
1311 | if (user_mode(regs)) | |
d4078e23 | 1312 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1313 | else |
d4078e23 | 1314 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1315 | } |
1316 | ||
9326638c | 1317 | dotraplinkage void notrace |
25c74b10 SA |
1318 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1319 | { | |
0ac09f9f JO |
1320 | /* |
1321 | * The exception_enter and tracepoint processing could | |
1322 | * trigger another page faults (user space callchain | |
1323 | * reading) and destroy the original cr2 value, so read | |
1324 | * the faulting address now. | |
1325 | */ | |
1326 | unsigned long address = read_cr2(); | |
d4078e23 | 1327 | enum ctx_state prev_state; |
25c74b10 SA |
1328 | |
1329 | prev_state = exception_enter(); | |
d4078e23 | 1330 | trace_page_fault_entries(address, regs, error_code); |
0ac09f9f | 1331 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1332 | exception_exit(prev_state); |
1333 | } | |
9326638c | 1334 | NOKPROBE_SYMBOL(trace_do_page_fault); |
d4078e23 | 1335 | #endif /* CONFIG_TRACING */ |