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