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 | |
b3ecd515 | 36 | * bit 5 == 1: protection keys block access |
33cb5243 | 37 | */ |
2d4a7167 IM |
38 | enum x86_pf_error_code { |
39 | ||
40 | PF_PROT = 1 << 0, | |
41 | PF_WRITE = 1 << 1, | |
42 | PF_USER = 1 << 2, | |
43 | PF_RSVD = 1 << 3, | |
44 | PF_INSTR = 1 << 4, | |
b3ecd515 | 45 | PF_PK = 1 << 5, |
2d4a7167 | 46 | }; |
66c58156 | 47 | |
b814d41f | 48 | /* |
b319eed0 IM |
49 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
50 | * handled by mmiotrace: | |
b814d41f | 51 | */ |
9326638c | 52 | static nokprobe_inline int |
62c9295f | 53 | kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 54 | { |
0fd0e3da PP |
55 | if (unlikely(is_kmmio_active())) |
56 | if (kmmio_handler(regs, addr) == 1) | |
57 | return -1; | |
0fd0e3da | 58 | return 0; |
86069782 PP |
59 | } |
60 | ||
9326638c | 61 | static nokprobe_inline int kprobes_fault(struct pt_regs *regs) |
1bd858a5 | 62 | { |
74a0b576 CH |
63 | int ret = 0; |
64 | ||
65 | /* kprobe_running() needs smp_processor_id() */ | |
f39b6f0e | 66 | if (kprobes_built_in() && !user_mode(regs)) { |
74a0b576 CH |
67 | preempt_disable(); |
68 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
69 | ret = 1; | |
70 | preempt_enable(); | |
71 | } | |
1bd858a5 | 72 | |
74a0b576 | 73 | return ret; |
33cb5243 | 74 | } |
1bd858a5 | 75 | |
1dc85be0 | 76 | /* |
2d4a7167 IM |
77 | * Prefetch quirks: |
78 | * | |
79 | * 32-bit mode: | |
80 | * | |
81 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
82 | * Check that here and ignore it. | |
1dc85be0 | 83 | * |
2d4a7167 | 84 | * 64-bit mode: |
1dc85be0 | 85 | * |
2d4a7167 IM |
86 | * Sometimes the CPU reports invalid exceptions on prefetch. |
87 | * Check that here and ignore it. | |
88 | * | |
89 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 90 | */ |
107a0367 IM |
91 | static inline int |
92 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
93 | unsigned char opcode, int *prefetch) | |
94 | { | |
95 | unsigned char instr_hi = opcode & 0xf0; | |
96 | unsigned char instr_lo = opcode & 0x0f; | |
97 | ||
98 | switch (instr_hi) { | |
99 | case 0x20: | |
100 | case 0x30: | |
101 | /* | |
102 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
103 | * In X86_64 long mode, the CPU will signal invalid | |
104 | * opcode if some of these prefixes are present so | |
105 | * X86_64 will never get here anyway | |
106 | */ | |
107 | return ((instr_lo & 7) == 0x6); | |
108 | #ifdef CONFIG_X86_64 | |
109 | case 0x40: | |
110 | /* | |
111 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
112 | * Need to figure out under what instruction mode the | |
113 | * instruction was issued. Could check the LDT for lm, | |
114 | * but for now it's good enough to assume that long | |
115 | * mode only uses well known segments or kernel. | |
116 | */ | |
318f5a2a | 117 | return (!user_mode(regs) || user_64bit_mode(regs)); |
107a0367 IM |
118 | #endif |
119 | case 0x60: | |
120 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
121 | return (instr_lo & 0xC) == 0x4; | |
122 | case 0xF0: | |
123 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
124 | return !instr_lo || (instr_lo>>1) == 1; | |
125 | case 0x00: | |
126 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
127 | if (probe_kernel_address(instr, opcode)) | |
128 | return 0; | |
129 | ||
130 | *prefetch = (instr_lo == 0xF) && | |
131 | (opcode == 0x0D || opcode == 0x18); | |
132 | return 0; | |
133 | default: | |
134 | return 0; | |
135 | } | |
136 | } | |
137 | ||
2d4a7167 IM |
138 | static int |
139 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 140 | { |
2d4a7167 | 141 | unsigned char *max_instr; |
ab2bf0c1 | 142 | unsigned char *instr; |
33cb5243 | 143 | int prefetch = 0; |
1da177e4 | 144 | |
3085354d IM |
145 | /* |
146 | * If it was a exec (instruction fetch) fault on NX page, then | |
147 | * do not ignore the fault: | |
148 | */ | |
66c58156 | 149 | if (error_code & PF_INSTR) |
1da177e4 | 150 | return 0; |
1dc85be0 | 151 | |
107a0367 | 152 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 153 | max_instr = instr + 15; |
1da177e4 | 154 | |
d31bf07f | 155 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) |
1da177e4 LT |
156 | return 0; |
157 | ||
107a0367 | 158 | while (instr < max_instr) { |
2d4a7167 | 159 | unsigned char opcode; |
1da177e4 | 160 | |
ab2bf0c1 | 161 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 162 | break; |
1da177e4 | 163 | |
1da177e4 LT |
164 | instr++; |
165 | ||
107a0367 | 166 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 167 | break; |
1da177e4 LT |
168 | } |
169 | return prefetch; | |
170 | } | |
171 | ||
2d4a7167 IM |
172 | static void |
173 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
7b2d0dba DH |
174 | struct task_struct *tsk, struct vm_area_struct *vma, |
175 | int fault) | |
c4aba4a8 | 176 | { |
f672b49b | 177 | unsigned lsb = 0; |
c4aba4a8 HH |
178 | siginfo_t info; |
179 | ||
2d4a7167 IM |
180 | info.si_signo = si_signo; |
181 | info.si_errno = 0; | |
182 | info.si_code = si_code; | |
183 | info.si_addr = (void __user *)address; | |
f672b49b AK |
184 | if (fault & VM_FAULT_HWPOISON_LARGE) |
185 | lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); | |
186 | if (fault & VM_FAULT_HWPOISON) | |
187 | lsb = PAGE_SHIFT; | |
188 | info.si_addr_lsb = lsb; | |
2d4a7167 | 189 | |
c4aba4a8 HH |
190 | force_sig_info(si_signo, &info, tsk); |
191 | } | |
192 | ||
f2f13a85 IM |
193 | DEFINE_SPINLOCK(pgd_lock); |
194 | LIST_HEAD(pgd_list); | |
195 | ||
196 | #ifdef CONFIG_X86_32 | |
197 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 198 | { |
f2f13a85 IM |
199 | unsigned index = pgd_index(address); |
200 | pgd_t *pgd_k; | |
201 | pud_t *pud, *pud_k; | |
202 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 203 | |
f2f13a85 IM |
204 | pgd += index; |
205 | pgd_k = init_mm.pgd + index; | |
206 | ||
207 | if (!pgd_present(*pgd_k)) | |
208 | return NULL; | |
209 | ||
210 | /* | |
211 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
212 | * and redundant with the set_pmd() on non-PAE. As would | |
213 | * set_pud. | |
214 | */ | |
215 | pud = pud_offset(pgd, address); | |
216 | pud_k = pud_offset(pgd_k, address); | |
217 | if (!pud_present(*pud_k)) | |
218 | return NULL; | |
219 | ||
220 | pmd = pmd_offset(pud, address); | |
221 | pmd_k = pmd_offset(pud_k, address); | |
222 | if (!pmd_present(*pmd_k)) | |
223 | return NULL; | |
224 | ||
b8bcfe99 | 225 | if (!pmd_present(*pmd)) |
f2f13a85 | 226 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 227 | else |
f2f13a85 | 228 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
229 | |
230 | return pmd_k; | |
231 | } | |
232 | ||
233 | void vmalloc_sync_all(void) | |
234 | { | |
235 | unsigned long address; | |
236 | ||
237 | if (SHARED_KERNEL_PMD) | |
238 | return; | |
239 | ||
240 | for (address = VMALLOC_START & PMD_MASK; | |
241 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
242 | address += PMD_SIZE) { | |
f2f13a85 IM |
243 | struct page *page; |
244 | ||
a79e53d8 | 245 | spin_lock(&pgd_lock); |
f2f13a85 | 246 | list_for_each_entry(page, &pgd_list, lru) { |
617d34d9 | 247 | spinlock_t *pgt_lock; |
f01f7c56 | 248 | pmd_t *ret; |
617d34d9 | 249 | |
a79e53d8 | 250 | /* the pgt_lock only for Xen */ |
617d34d9 JF |
251 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; |
252 | ||
253 | spin_lock(pgt_lock); | |
254 | ret = vmalloc_sync_one(page_address(page), address); | |
255 | spin_unlock(pgt_lock); | |
256 | ||
257 | if (!ret) | |
f2f13a85 IM |
258 | break; |
259 | } | |
a79e53d8 | 260 | spin_unlock(&pgd_lock); |
f2f13a85 IM |
261 | } |
262 | } | |
263 | ||
264 | /* | |
265 | * 32-bit: | |
266 | * | |
267 | * Handle a fault on the vmalloc or module mapping area | |
268 | */ | |
9326638c | 269 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
270 | { |
271 | unsigned long pgd_paddr; | |
272 | pmd_t *pmd_k; | |
273 | pte_t *pte_k; | |
274 | ||
275 | /* Make sure we are in vmalloc area: */ | |
276 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
277 | return -1; | |
278 | ||
ebc8827f FW |
279 | WARN_ON_ONCE(in_nmi()); |
280 | ||
f2f13a85 IM |
281 | /* |
282 | * Synchronize this task's top level page-table | |
283 | * with the 'reference' page table. | |
284 | * | |
285 | * Do _not_ use "current" here. We might be inside | |
286 | * an interrupt in the middle of a task switch.. | |
287 | */ | |
288 | pgd_paddr = read_cr3(); | |
289 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
290 | if (!pmd_k) | |
291 | return -1; | |
292 | ||
293 | pte_k = pte_offset_kernel(pmd_k, address); | |
294 | if (!pte_present(*pte_k)) | |
295 | return -1; | |
296 | ||
297 | return 0; | |
298 | } | |
9326638c | 299 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 IM |
300 | |
301 | /* | |
302 | * Did it hit the DOS screen memory VA from vm86 mode? | |
303 | */ | |
304 | static inline void | |
305 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
306 | struct task_struct *tsk) | |
307 | { | |
9fda6a06 | 308 | #ifdef CONFIG_VM86 |
f2f13a85 IM |
309 | unsigned long bit; |
310 | ||
9fda6a06 | 311 | if (!v8086_mode(regs) || !tsk->thread.vm86) |
f2f13a85 IM |
312 | return; |
313 | ||
314 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
315 | if (bit < 32) | |
9fda6a06 BG |
316 | tsk->thread.vm86->screen_bitmap |= 1 << bit; |
317 | #endif | |
33cb5243 | 318 | } |
1da177e4 | 319 | |
087975b0 | 320 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 321 | { |
087975b0 AM |
322 | return pfn < max_low_pfn; |
323 | } | |
1156e098 | 324 | |
087975b0 AM |
325 | static void dump_pagetable(unsigned long address) |
326 | { | |
327 | pgd_t *base = __va(read_cr3()); | |
328 | pgd_t *pgd = &base[pgd_index(address)]; | |
329 | pmd_t *pmd; | |
330 | pte_t *pte; | |
2d4a7167 | 331 | |
1156e098 | 332 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
333 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
334 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
335 | goto out; | |
1156e098 | 336 | #endif |
087975b0 AM |
337 | pmd = pmd_offset(pud_offset(pgd, address), address); |
338 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
339 | |
340 | /* | |
341 | * We must not directly access the pte in the highpte | |
342 | * case if the page table is located in highmem. | |
343 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 344 | * it's allocated already: |
1156e098 | 345 | */ |
087975b0 AM |
346 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
347 | goto out; | |
1156e098 | 348 | |
087975b0 AM |
349 | pte = pte_offset_kernel(pmd, address); |
350 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
351 | out: | |
1156e098 | 352 | printk("\n"); |
f2f13a85 IM |
353 | } |
354 | ||
355 | #else /* CONFIG_X86_64: */ | |
356 | ||
357 | void vmalloc_sync_all(void) | |
358 | { | |
9661d5bc | 359 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); |
f2f13a85 IM |
360 | } |
361 | ||
362 | /* | |
363 | * 64-bit: | |
364 | * | |
365 | * Handle a fault on the vmalloc area | |
366 | * | |
367 | * This assumes no large pages in there. | |
368 | */ | |
9326638c | 369 | static noinline int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
370 | { |
371 | pgd_t *pgd, *pgd_ref; | |
372 | pud_t *pud, *pud_ref; | |
373 | pmd_t *pmd, *pmd_ref; | |
374 | pte_t *pte, *pte_ref; | |
375 | ||
376 | /* Make sure we are in vmalloc area: */ | |
377 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
378 | return -1; | |
379 | ||
ebc8827f FW |
380 | WARN_ON_ONCE(in_nmi()); |
381 | ||
f2f13a85 IM |
382 | /* |
383 | * Copy kernel mappings over when needed. This can also | |
384 | * happen within a race in page table update. In the later | |
385 | * case just flush: | |
386 | */ | |
387 | pgd = pgd_offset(current->active_mm, address); | |
388 | pgd_ref = pgd_offset_k(address); | |
389 | if (pgd_none(*pgd_ref)) | |
390 | return -1; | |
391 | ||
1160c277 | 392 | if (pgd_none(*pgd)) { |
f2f13a85 | 393 | set_pgd(pgd, *pgd_ref); |
1160c277 SK |
394 | arch_flush_lazy_mmu_mode(); |
395 | } else { | |
f2f13a85 | 396 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1160c277 | 397 | } |
f2f13a85 IM |
398 | |
399 | /* | |
400 | * Below here mismatches are bugs because these lower tables | |
401 | * are shared: | |
402 | */ | |
403 | ||
404 | pud = pud_offset(pgd, address); | |
405 | pud_ref = pud_offset(pgd_ref, address); | |
406 | if (pud_none(*pud_ref)) | |
407 | return -1; | |
408 | ||
409 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
410 | BUG(); | |
411 | ||
412 | pmd = pmd_offset(pud, address); | |
413 | pmd_ref = pmd_offset(pud_ref, address); | |
414 | if (pmd_none(*pmd_ref)) | |
415 | return -1; | |
416 | ||
417 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
418 | BUG(); | |
419 | ||
420 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
421 | if (!pte_present(*pte_ref)) | |
422 | return -1; | |
423 | ||
424 | pte = pte_offset_kernel(pmd, address); | |
425 | ||
426 | /* | |
427 | * Don't use pte_page here, because the mappings can point | |
428 | * outside mem_map, and the NUMA hash lookup cannot handle | |
429 | * that: | |
430 | */ | |
431 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
432 | BUG(); | |
433 | ||
434 | return 0; | |
435 | } | |
9326638c | 436 | NOKPROBE_SYMBOL(vmalloc_fault); |
f2f13a85 | 437 | |
e05139f2 | 438 | #ifdef CONFIG_CPU_SUP_AMD |
f2f13a85 | 439 | static const char errata93_warning[] = |
ad361c98 JP |
440 | KERN_ERR |
441 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
442 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
443 | "******* Please consider a BIOS update.\n" | |
444 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
e05139f2 | 445 | #endif |
f2f13a85 IM |
446 | |
447 | /* | |
448 | * No vm86 mode in 64-bit mode: | |
449 | */ | |
450 | static inline void | |
451 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
452 | struct task_struct *tsk) | |
453 | { | |
454 | } | |
455 | ||
456 | static int bad_address(void *p) | |
457 | { | |
458 | unsigned long dummy; | |
459 | ||
460 | return probe_kernel_address((unsigned long *)p, dummy); | |
461 | } | |
462 | ||
463 | static void dump_pagetable(unsigned long address) | |
464 | { | |
087975b0 AM |
465 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
466 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
467 | pud_t *pud; |
468 | pmd_t *pmd; | |
469 | pte_t *pte; | |
470 | ||
2d4a7167 IM |
471 | if (bad_address(pgd)) |
472 | goto bad; | |
473 | ||
d646bce4 | 474 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
475 | |
476 | if (!pgd_present(*pgd)) | |
477 | goto out; | |
1da177e4 | 478 | |
d2ae5b5f | 479 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
480 | if (bad_address(pud)) |
481 | goto bad; | |
482 | ||
1da177e4 | 483 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 484 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 485 | goto out; |
1da177e4 LT |
486 | |
487 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
488 | if (bad_address(pmd)) |
489 | goto bad; | |
490 | ||
1da177e4 | 491 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
492 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
493 | goto out; | |
1da177e4 LT |
494 | |
495 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
496 | if (bad_address(pte)) |
497 | goto bad; | |
498 | ||
33cb5243 | 499 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 500 | out: |
1da177e4 LT |
501 | printk("\n"); |
502 | return; | |
503 | bad: | |
504 | printk("BAD\n"); | |
8c938f9f IM |
505 | } |
506 | ||
f2f13a85 | 507 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 508 | |
2d4a7167 IM |
509 | /* |
510 | * Workaround for K8 erratum #93 & buggy BIOS. | |
511 | * | |
512 | * BIOS SMM functions are required to use a specific workaround | |
513 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
514 | * | |
515 | * A lot of BIOS that didn't get tested properly miss this. | |
516 | * | |
517 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
518 | * Try to work around it here. | |
519 | * | |
520 | * Note we only handle faults in kernel here. | |
521 | * Does nothing on 32-bit. | |
fdfe8aa8 | 522 | */ |
33cb5243 | 523 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 524 | { |
e05139f2 JB |
525 | #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) |
526 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD | |
527 | || boot_cpu_data.x86 != 0xf) | |
528 | return 0; | |
529 | ||
65ea5b03 | 530 | if (address != regs->ip) |
1da177e4 | 531 | return 0; |
2d4a7167 | 532 | |
33cb5243 | 533 | if ((address >> 32) != 0) |
1da177e4 | 534 | return 0; |
2d4a7167 | 535 | |
1da177e4 | 536 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
537 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
538 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 539 | printk_once(errata93_warning); |
65ea5b03 | 540 | regs->ip = address; |
1da177e4 LT |
541 | return 1; |
542 | } | |
fdfe8aa8 | 543 | #endif |
1da177e4 | 544 | return 0; |
33cb5243 | 545 | } |
1da177e4 | 546 | |
35f3266f | 547 | /* |
2d4a7167 IM |
548 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
549 | * to illegal addresses >4GB. | |
550 | * | |
551 | * We catch this in the page fault handler because these addresses | |
552 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
553 | * segment in LDT is compatibility mode. |
554 | */ | |
555 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
556 | { | |
557 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 558 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
559 | return 1; |
560 | #endif | |
561 | return 0; | |
562 | } | |
563 | ||
29caf2f9 HH |
564 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
565 | { | |
566 | #ifdef CONFIG_X86_F00F_BUG | |
567 | unsigned long nr; | |
2d4a7167 | 568 | |
29caf2f9 | 569 | /* |
2d4a7167 | 570 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 | 571 | */ |
e2604b49 | 572 | if (boot_cpu_has_bug(X86_BUG_F00F)) { |
29caf2f9 HH |
573 | nr = (address - idt_descr.address) >> 3; |
574 | ||
575 | if (nr == 6) { | |
576 | do_invalid_op(regs, 0); | |
577 | return 1; | |
578 | } | |
579 | } | |
580 | #endif | |
581 | return 0; | |
582 | } | |
583 | ||
8f766149 IM |
584 | static const char nx_warning[] = KERN_CRIT |
585 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
eff50c34 JK |
586 | static const char smep_warning[] = KERN_CRIT |
587 | "unable to execute userspace code (SMEP?) (uid: %d)\n"; | |
8f766149 | 588 | |
2d4a7167 IM |
589 | static void |
590 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
591 | unsigned long address) | |
b3279c7f | 592 | { |
1156e098 HH |
593 | if (!oops_may_print()) |
594 | return; | |
595 | ||
1156e098 | 596 | if (error_code & PF_INSTR) { |
93809be8 | 597 | unsigned int level; |
426e34cc MF |
598 | pgd_t *pgd; |
599 | pte_t *pte; | |
2d4a7167 | 600 | |
426e34cc MF |
601 | pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
602 | pgd += pgd_index(address); | |
603 | ||
604 | pte = lookup_address_in_pgd(pgd, address, &level); | |
1156e098 | 605 | |
8f766149 | 606 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
078de5f7 | 607 | printk(nx_warning, from_kuid(&init_user_ns, current_uid())); |
eff50c34 JK |
608 | if (pte && pte_present(*pte) && pte_exec(*pte) && |
609 | (pgd_flags(*pgd) & _PAGE_USER) && | |
1e02ce4c | 610 | (__read_cr4() & X86_CR4_SMEP)) |
eff50c34 | 611 | printk(smep_warning, from_kuid(&init_user_ns, current_uid())); |
1156e098 | 612 | } |
1156e098 | 613 | |
19f0dda9 | 614 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 615 | if (address < PAGE_SIZE) |
19f0dda9 | 616 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 617 | else |
19f0dda9 | 618 | printk(KERN_CONT "paging request"); |
2d4a7167 | 619 | |
f294a8ce | 620 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 621 | printk(KERN_ALERT "IP:"); |
5f01c988 | 622 | printk_address(regs->ip); |
2d4a7167 | 623 | |
b3279c7f HH |
624 | dump_pagetable(address); |
625 | } | |
626 | ||
2d4a7167 IM |
627 | static noinline void |
628 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
629 | unsigned long address) | |
1da177e4 | 630 | { |
2d4a7167 IM |
631 | struct task_struct *tsk; |
632 | unsigned long flags; | |
633 | int sig; | |
634 | ||
635 | flags = oops_begin(); | |
636 | tsk = current; | |
637 | sig = SIGKILL; | |
1209140c | 638 | |
1da177e4 | 639 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 640 | tsk->comm, address); |
1da177e4 | 641 | dump_pagetable(address); |
2d4a7167 IM |
642 | |
643 | tsk->thread.cr2 = address; | |
51e7dc70 | 644 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 IM |
645 | tsk->thread.error_code = error_code; |
646 | ||
22f5991c | 647 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 648 | sig = 0; |
2d4a7167 | 649 | |
874d93d1 | 650 | oops_end(flags, regs, sig); |
1da177e4 LT |
651 | } |
652 | ||
2d4a7167 IM |
653 | static noinline void |
654 | no_context(struct pt_regs *regs, unsigned long error_code, | |
4fc34901 | 655 | unsigned long address, int signal, int si_code) |
92181f19 NP |
656 | { |
657 | struct task_struct *tsk = current; | |
92181f19 NP |
658 | unsigned long flags; |
659 | int sig; | |
7b2d0dba DH |
660 | /* No context means no VMA to pass down */ |
661 | struct vm_area_struct *vma = NULL; | |
92181f19 | 662 | |
2d4a7167 | 663 | /* Are we prepared to handle this kernel fault? */ |
4fc34901 | 664 | if (fixup_exception(regs)) { |
c026b359 PZ |
665 | /* |
666 | * Any interrupt that takes a fault gets the fixup. This makes | |
667 | * the below recursive fault logic only apply to a faults from | |
668 | * task context. | |
669 | */ | |
670 | if (in_interrupt()) | |
671 | return; | |
672 | ||
673 | /* | |
674 | * Per the above we're !in_interrupt(), aka. task context. | |
675 | * | |
676 | * In this case we need to make sure we're not recursively | |
677 | * faulting through the emulate_vsyscall() logic. | |
678 | */ | |
4fc34901 | 679 | if (current_thread_info()->sig_on_uaccess_error && signal) { |
51e7dc70 | 680 | tsk->thread.trap_nr = X86_TRAP_PF; |
4fc34901 AL |
681 | tsk->thread.error_code = error_code | PF_USER; |
682 | tsk->thread.cr2 = address; | |
683 | ||
684 | /* XXX: hwpoison faults will set the wrong code. */ | |
7b2d0dba DH |
685 | force_sig_info_fault(signal, si_code, address, |
686 | tsk, vma, 0); | |
4fc34901 | 687 | } |
c026b359 PZ |
688 | |
689 | /* | |
690 | * Barring that, we can do the fixup and be happy. | |
691 | */ | |
92181f19 | 692 | return; |
4fc34901 | 693 | } |
92181f19 NP |
694 | |
695 | /* | |
2d4a7167 IM |
696 | * 32-bit: |
697 | * | |
698 | * Valid to do another page fault here, because if this fault | |
699 | * had been triggered by is_prefetch fixup_exception would have | |
700 | * handled it. | |
701 | * | |
702 | * 64-bit: | |
92181f19 | 703 | * |
2d4a7167 | 704 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
705 | */ |
706 | if (is_prefetch(regs, error_code, address)) | |
707 | return; | |
708 | ||
709 | if (is_errata93(regs, address)) | |
710 | return; | |
711 | ||
712 | /* | |
713 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 714 | * terminate things with extreme prejudice: |
92181f19 | 715 | */ |
92181f19 | 716 | flags = oops_begin(); |
92181f19 NP |
717 | |
718 | show_fault_oops(regs, error_code, address); | |
719 | ||
a70857e4 | 720 | if (task_stack_end_corrupted(tsk)) |
b0f4c4b3 | 721 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
19803078 | 722 | |
1cc99544 | 723 | tsk->thread.cr2 = address; |
51e7dc70 | 724 | tsk->thread.trap_nr = X86_TRAP_PF; |
1cc99544 | 725 | tsk->thread.error_code = error_code; |
92181f19 | 726 | |
92181f19 NP |
727 | sig = SIGKILL; |
728 | if (__die("Oops", regs, error_code)) | |
729 | sig = 0; | |
2d4a7167 | 730 | |
92181f19 | 731 | /* Executive summary in case the body of the oops scrolled away */ |
b0f4c4b3 | 732 | printk(KERN_DEFAULT "CR2: %016lx\n", address); |
2d4a7167 | 733 | |
92181f19 | 734 | oops_end(flags, regs, sig); |
92181f19 NP |
735 | } |
736 | ||
2d4a7167 IM |
737 | /* |
738 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
739 | * sysctl is set: | |
740 | */ | |
741 | static inline void | |
742 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
743 | unsigned long address, struct task_struct *tsk) | |
744 | { | |
745 | if (!unhandled_signal(tsk, SIGSEGV)) | |
746 | return; | |
747 | ||
748 | if (!printk_ratelimit()) | |
749 | return; | |
750 | ||
a1a08d1c | 751 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
752 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
753 | tsk->comm, task_pid_nr(tsk), address, | |
754 | (void *)regs->ip, (void *)regs->sp, error_code); | |
755 | ||
756 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
757 | ||
758 | printk(KERN_CONT "\n"); | |
759 | } | |
760 | ||
761 | static void | |
762 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba DH |
763 | unsigned long address, struct vm_area_struct *vma, |
764 | int si_code) | |
92181f19 NP |
765 | { |
766 | struct task_struct *tsk = current; | |
767 | ||
768 | /* User mode accesses just cause a SIGSEGV */ | |
769 | if (error_code & PF_USER) { | |
770 | /* | |
2d4a7167 | 771 | * It's possible to have interrupts off here: |
92181f19 NP |
772 | */ |
773 | local_irq_enable(); | |
774 | ||
775 | /* | |
776 | * Valid to do another page fault here because this one came | |
2d4a7167 | 777 | * from user space: |
92181f19 NP |
778 | */ |
779 | if (is_prefetch(regs, error_code, address)) | |
780 | return; | |
781 | ||
782 | if (is_errata100(regs, address)) | |
783 | return; | |
784 | ||
3ae36655 AL |
785 | #ifdef CONFIG_X86_64 |
786 | /* | |
787 | * Instruction fetch faults in the vsyscall page might need | |
788 | * emulation. | |
789 | */ | |
790 | if (unlikely((error_code & PF_INSTR) && | |
f40c3300 | 791 | ((address & ~0xfff) == VSYSCALL_ADDR))) { |
3ae36655 AL |
792 | if (emulate_vsyscall(regs, address)) |
793 | return; | |
794 | } | |
795 | #endif | |
e575a86f KC |
796 | /* Kernel addresses are always protection faults: */ |
797 | if (address >= TASK_SIZE) | |
798 | error_code |= PF_PROT; | |
3ae36655 | 799 | |
e575a86f | 800 | if (likely(show_unhandled_signals)) |
2d4a7167 IM |
801 | show_signal_msg(regs, error_code, address, tsk); |
802 | ||
2d4a7167 | 803 | tsk->thread.cr2 = address; |
e575a86f | 804 | tsk->thread.error_code = error_code; |
51e7dc70 | 805 | tsk->thread.trap_nr = X86_TRAP_PF; |
92181f19 | 806 | |
7b2d0dba | 807 | force_sig_info_fault(SIGSEGV, si_code, address, tsk, vma, 0); |
2d4a7167 | 808 | |
92181f19 NP |
809 | return; |
810 | } | |
811 | ||
812 | if (is_f00f_bug(regs, address)) | |
813 | return; | |
814 | ||
4fc34901 | 815 | no_context(regs, error_code, address, SIGSEGV, si_code); |
92181f19 NP |
816 | } |
817 | ||
2d4a7167 IM |
818 | static noinline void |
819 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 820 | unsigned long address, struct vm_area_struct *vma) |
92181f19 | 821 | { |
7b2d0dba | 822 | __bad_area_nosemaphore(regs, error_code, address, vma, SEGV_MAPERR); |
92181f19 NP |
823 | } |
824 | ||
2d4a7167 IM |
825 | static void |
826 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 827 | unsigned long address, struct vm_area_struct *vma, int si_code) |
92181f19 NP |
828 | { |
829 | struct mm_struct *mm = current->mm; | |
830 | ||
831 | /* | |
832 | * Something tried to access memory that isn't in our memory map.. | |
833 | * Fix it, but check if it's kernel or user first.. | |
834 | */ | |
835 | up_read(&mm->mmap_sem); | |
836 | ||
7b2d0dba | 837 | __bad_area_nosemaphore(regs, error_code, address, vma, si_code); |
92181f19 NP |
838 | } |
839 | ||
2d4a7167 IM |
840 | static noinline void |
841 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 | 842 | { |
7b2d0dba | 843 | __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); |
92181f19 NP |
844 | } |
845 | ||
2d4a7167 IM |
846 | static noinline void |
847 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
7b2d0dba | 848 | unsigned long address, struct vm_area_struct *vma) |
92181f19 | 849 | { |
7b2d0dba | 850 | __bad_area(regs, error_code, address, vma, SEGV_ACCERR); |
92181f19 NP |
851 | } |
852 | ||
2d4a7167 | 853 | static void |
a6e04aa9 | 854 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
7b2d0dba | 855 | struct vm_area_struct *vma, unsigned int fault) |
92181f19 NP |
856 | { |
857 | struct task_struct *tsk = current; | |
a6e04aa9 | 858 | int code = BUS_ADRERR; |
92181f19 | 859 | |
2d4a7167 | 860 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 861 | if (!(error_code & PF_USER)) { |
4fc34901 | 862 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); |
96054569 LT |
863 | return; |
864 | } | |
2d4a7167 | 865 | |
cd1b68f0 | 866 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
867 | if (is_prefetch(regs, error_code, address)) |
868 | return; | |
2d4a7167 IM |
869 | |
870 | tsk->thread.cr2 = address; | |
871 | tsk->thread.error_code = error_code; | |
51e7dc70 | 872 | tsk->thread.trap_nr = X86_TRAP_PF; |
2d4a7167 | 873 | |
a6e04aa9 | 874 | #ifdef CONFIG_MEMORY_FAILURE |
f672b49b | 875 | if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { |
a6e04aa9 AK |
876 | printk(KERN_ERR |
877 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
878 | tsk->comm, tsk->pid, address); | |
879 | code = BUS_MCEERR_AR; | |
880 | } | |
881 | #endif | |
7b2d0dba | 882 | force_sig_info_fault(SIGBUS, code, address, tsk, vma, fault); |
92181f19 NP |
883 | } |
884 | ||
3a13c4d7 | 885 | static noinline void |
2d4a7167 | 886 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
7b2d0dba DH |
887 | unsigned long address, struct vm_area_struct *vma, |
888 | unsigned int fault) | |
92181f19 | 889 | { |
3a13c4d7 | 890 | if (fatal_signal_pending(current) && !(error_code & PF_USER)) { |
3a13c4d7 JW |
891 | no_context(regs, error_code, address, 0, 0); |
892 | return; | |
b80ef10e | 893 | } |
b80ef10e | 894 | |
2d4a7167 | 895 | if (fault & VM_FAULT_OOM) { |
f8626854 AV |
896 | /* Kernel mode? Handle exceptions or die: */ |
897 | if (!(error_code & PF_USER)) { | |
4fc34901 AL |
898 | no_context(regs, error_code, address, |
899 | SIGSEGV, SEGV_MAPERR); | |
3a13c4d7 | 900 | return; |
f8626854 AV |
901 | } |
902 | ||
c2d23f91 DR |
903 | /* |
904 | * We ran out of memory, call the OOM killer, and return the | |
905 | * userspace (which will retry the fault, or kill us if we got | |
906 | * oom-killed): | |
907 | */ | |
908 | pagefault_out_of_memory(); | |
2d4a7167 | 909 | } else { |
f672b49b AK |
910 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| |
911 | VM_FAULT_HWPOISON_LARGE)) | |
7b2d0dba | 912 | do_sigbus(regs, error_code, address, vma, fault); |
33692f27 | 913 | else if (fault & VM_FAULT_SIGSEGV) |
7b2d0dba | 914 | bad_area_nosemaphore(regs, error_code, address, vma); |
2d4a7167 IM |
915 | else |
916 | BUG(); | |
917 | } | |
92181f19 NP |
918 | } |
919 | ||
d8b57bb7 TG |
920 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
921 | { | |
922 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
923 | return 0; | |
2d4a7167 | 924 | |
d8b57bb7 TG |
925 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
926 | return 0; | |
b3ecd515 DH |
927 | /* |
928 | * Note: We do not do lazy flushing on protection key | |
929 | * changes, so no spurious fault will ever set PF_PK. | |
930 | */ | |
931 | if ((error_code & PF_PK)) | |
932 | return 1; | |
d8b57bb7 TG |
933 | |
934 | return 1; | |
935 | } | |
936 | ||
5b727a3b | 937 | /* |
2d4a7167 IM |
938 | * Handle a spurious fault caused by a stale TLB entry. |
939 | * | |
940 | * This allows us to lazily refresh the TLB when increasing the | |
941 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
942 | * eagerly is very expensive since that implies doing a full | |
943 | * cross-processor TLB flush, even if no stale TLB entries exist | |
944 | * on other processors. | |
945 | * | |
31668511 DV |
946 | * Spurious faults may only occur if the TLB contains an entry with |
947 | * fewer permission than the page table entry. Non-present (P = 0) | |
948 | * and reserved bit (R = 1) faults are never spurious. | |
949 | * | |
5b727a3b JF |
950 | * There are no security implications to leaving a stale TLB when |
951 | * increasing the permissions on a page. | |
31668511 DV |
952 | * |
953 | * Returns non-zero if a spurious fault was handled, zero otherwise. | |
954 | * | |
955 | * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 | |
956 | * (Optional Invalidation). | |
5b727a3b | 957 | */ |
9326638c | 958 | static noinline int |
2d4a7167 | 959 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
960 | { |
961 | pgd_t *pgd; | |
962 | pud_t *pud; | |
963 | pmd_t *pmd; | |
964 | pte_t *pte; | |
3c3e5694 | 965 | int ret; |
5b727a3b | 966 | |
31668511 DV |
967 | /* |
968 | * Only writes to RO or instruction fetches from NX may cause | |
969 | * spurious faults. | |
970 | * | |
971 | * These could be from user or supervisor accesses but the TLB | |
972 | * is only lazily flushed after a kernel mapping protection | |
973 | * change, so user accesses are not expected to cause spurious | |
974 | * faults. | |
975 | */ | |
976 | if (error_code != (PF_WRITE | PF_PROT) | |
977 | && error_code != (PF_INSTR | PF_PROT)) | |
5b727a3b JF |
978 | return 0; |
979 | ||
980 | pgd = init_mm.pgd + pgd_index(address); | |
981 | if (!pgd_present(*pgd)) | |
982 | return 0; | |
983 | ||
984 | pud = pud_offset(pgd, address); | |
985 | if (!pud_present(*pud)) | |
986 | return 0; | |
987 | ||
d8b57bb7 TG |
988 | if (pud_large(*pud)) |
989 | return spurious_fault_check(error_code, (pte_t *) pud); | |
990 | ||
5b727a3b JF |
991 | pmd = pmd_offset(pud, address); |
992 | if (!pmd_present(*pmd)) | |
993 | return 0; | |
994 | ||
d8b57bb7 TG |
995 | if (pmd_large(*pmd)) |
996 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
997 | ||
5b727a3b | 998 | pte = pte_offset_kernel(pmd, address); |
954f8571 | 999 | if (!pte_present(*pte)) |
5b727a3b JF |
1000 | return 0; |
1001 | ||
3c3e5694 SR |
1002 | ret = spurious_fault_check(error_code, pte); |
1003 | if (!ret) | |
1004 | return 0; | |
1005 | ||
1006 | /* | |
2d4a7167 IM |
1007 | * Make sure we have permissions in PMD. |
1008 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
1009 | */ |
1010 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
1011 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 1012 | |
3c3e5694 | 1013 | return ret; |
5b727a3b | 1014 | } |
9326638c | 1015 | NOKPROBE_SYMBOL(spurious_fault); |
5b727a3b | 1016 | |
abd4f750 | 1017 | int show_unhandled_signals = 1; |
1da177e4 | 1018 | |
2d4a7167 | 1019 | static inline int |
68da336a | 1020 | access_error(unsigned long error_code, struct vm_area_struct *vma) |
92181f19 | 1021 | { |
68da336a | 1022 | if (error_code & PF_WRITE) { |
2d4a7167 | 1023 | /* write, present and write, not present: */ |
92181f19 NP |
1024 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
1025 | return 1; | |
2d4a7167 | 1026 | return 0; |
92181f19 NP |
1027 | } |
1028 | ||
2d4a7167 IM |
1029 | /* read, present: */ |
1030 | if (unlikely(error_code & PF_PROT)) | |
1031 | return 1; | |
1032 | ||
1033 | /* read, not present: */ | |
1034 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
1035 | return 1; | |
1036 | ||
92181f19 NP |
1037 | return 0; |
1038 | } | |
1039 | ||
0973a06c HS |
1040 | static int fault_in_kernel_space(unsigned long address) |
1041 | { | |
d9517346 | 1042 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
1043 | } |
1044 | ||
40d3cd66 PA |
1045 | static inline bool smap_violation(int error_code, struct pt_regs *regs) |
1046 | { | |
4640c7ee PA |
1047 | if (!IS_ENABLED(CONFIG_X86_SMAP)) |
1048 | return false; | |
1049 | ||
1050 | if (!static_cpu_has(X86_FEATURE_SMAP)) | |
1051 | return false; | |
1052 | ||
40d3cd66 PA |
1053 | if (error_code & PF_USER) |
1054 | return false; | |
1055 | ||
f39b6f0e | 1056 | if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) |
40d3cd66 PA |
1057 | return false; |
1058 | ||
1059 | return true; | |
1060 | } | |
1061 | ||
1da177e4 LT |
1062 | /* |
1063 | * This routine handles page faults. It determines the address, | |
1064 | * and the problem, and then passes it off to one of the appropriate | |
1065 | * routines. | |
d4078e23 PZ |
1066 | * |
1067 | * This function must have noinline because both callers | |
1068 | * {,trace_}do_page_fault() have notrace on. Having this an actual function | |
1069 | * guarantees there's a function trace entry. | |
1da177e4 | 1070 | */ |
9326638c | 1071 | static noinline void |
0ac09f9f JO |
1072 | __do_page_fault(struct pt_regs *regs, unsigned long error_code, |
1073 | unsigned long address) | |
1da177e4 | 1074 | { |
2d4a7167 | 1075 | struct vm_area_struct *vma; |
1da177e4 LT |
1076 | struct task_struct *tsk; |
1077 | struct mm_struct *mm; | |
26178ec1 | 1078 | int fault, major = 0; |
759496ba | 1079 | unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
1da177e4 | 1080 | |
a9ba9a3b AV |
1081 | tsk = current; |
1082 | mm = tsk->mm; | |
2d4a7167 | 1083 | |
f8561296 VN |
1084 | /* |
1085 | * Detect and handle instructions that would cause a page fault for | |
1086 | * both a tracked kernel page and a userspace page. | |
1087 | */ | |
1088 | if (kmemcheck_active(regs)) | |
1089 | kmemcheck_hide(regs); | |
5dfaf90f | 1090 | prefetchw(&mm->mmap_sem); |
f8561296 | 1091 | |
0fd0e3da | 1092 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 1093 | return; |
1da177e4 LT |
1094 | |
1095 | /* | |
1096 | * We fault-in kernel-space virtual memory on-demand. The | |
1097 | * 'reference' page table is init_mm.pgd. | |
1098 | * | |
1099 | * NOTE! We MUST NOT take any locks for this case. We may | |
1100 | * be in an interrupt or a critical region, and should | |
1101 | * only copy the information from the master page table, | |
1102 | * nothing more. | |
1103 | * | |
1104 | * This verifies that the fault happens in kernel space | |
1105 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 1106 | * protection error (error_code & 9) == 0. |
1da177e4 | 1107 | */ |
0973a06c | 1108 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
1109 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
1110 | if (vmalloc_fault(address) >= 0) | |
1111 | return; | |
1112 | ||
1113 | if (kmemcheck_fault(regs, address, error_code)) | |
1114 | return; | |
1115 | } | |
5b727a3b | 1116 | |
2d4a7167 | 1117 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 1118 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
1119 | return; |
1120 | ||
2d4a7167 | 1121 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1122 | if (kprobes_fault(regs)) |
9be260a6 | 1123 | return; |
f8c2ee22 HH |
1124 | /* |
1125 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 1126 | * fault we could otherwise deadlock: |
f8c2ee22 | 1127 | */ |
7b2d0dba | 1128 | bad_area_nosemaphore(regs, error_code, address, NULL); |
2d4a7167 | 1129 | |
92181f19 | 1130 | return; |
f8c2ee22 HH |
1131 | } |
1132 | ||
2d4a7167 | 1133 | /* kprobes don't want to hook the spurious faults: */ |
e00b12e6 | 1134 | if (unlikely(kprobes_fault(regs))) |
9be260a6 | 1135 | return; |
8c914cb7 | 1136 | |
66c58156 | 1137 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1138 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1139 | |
4640c7ee | 1140 | if (unlikely(smap_violation(error_code, regs))) { |
7b2d0dba | 1141 | bad_area_nosemaphore(regs, error_code, address, NULL); |
4640c7ee | 1142 | return; |
40d3cd66 PA |
1143 | } |
1144 | ||
1da177e4 | 1145 | /* |
2d4a7167 | 1146 | * If we're in an interrupt, have no user context or are running |
70ffdb93 | 1147 | * in a region with pagefaults disabled then we must not take the fault |
1da177e4 | 1148 | */ |
70ffdb93 | 1149 | if (unlikely(faulthandler_disabled() || !mm)) { |
7b2d0dba | 1150 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1151 | return; |
1152 | } | |
1da177e4 | 1153 | |
e00b12e6 PZ |
1154 | /* |
1155 | * It's safe to allow irq's after cr2 has been saved and the | |
1156 | * vmalloc fault has been handled. | |
1157 | * | |
1158 | * User-mode registers count as a user access even for any | |
1159 | * potential system fault or CPU buglet: | |
1160 | */ | |
f39b6f0e | 1161 | if (user_mode(regs)) { |
e00b12e6 PZ |
1162 | local_irq_enable(); |
1163 | error_code |= PF_USER; | |
1164 | flags |= FAULT_FLAG_USER; | |
1165 | } else { | |
1166 | if (regs->flags & X86_EFLAGS_IF) | |
1167 | local_irq_enable(); | |
1168 | } | |
1169 | ||
1170 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); | |
1171 | ||
759496ba JW |
1172 | if (error_code & PF_WRITE) |
1173 | flags |= FAULT_FLAG_WRITE; | |
1174 | ||
3a1dfe6e IM |
1175 | /* |
1176 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1177 | * addresses in user space. All other faults represent errors in |
1178 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1179 | * case of an erroneous fault occurring in a code path which already | |
1180 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1181 | * against the address space. Luckily the kernel only validly | |
1182 | * references user space from well defined areas of code, which are | |
1183 | * listed in the exceptions table. | |
1da177e4 LT |
1184 | * |
1185 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1186 | * the source reference check when there is a possibility of a |
1187 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1188 | * validate the source. If this is invalid we can skip the address | |
1189 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1190 | */ |
92181f19 | 1191 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1192 | if ((error_code & PF_USER) == 0 && |
92181f19 | 1193 | !search_exception_tables(regs->ip)) { |
7b2d0dba | 1194 | bad_area_nosemaphore(regs, error_code, address, NULL); |
92181f19 NP |
1195 | return; |
1196 | } | |
d065bd81 | 1197 | retry: |
1da177e4 | 1198 | down_read(&mm->mmap_sem); |
01006074 PZ |
1199 | } else { |
1200 | /* | |
2d4a7167 IM |
1201 | * The above down_read_trylock() might have succeeded in |
1202 | * which case we'll have missed the might_sleep() from | |
1203 | * down_read(): | |
01006074 PZ |
1204 | */ |
1205 | might_sleep(); | |
1da177e4 LT |
1206 | } |
1207 | ||
1208 | vma = find_vma(mm, address); | |
92181f19 NP |
1209 | if (unlikely(!vma)) { |
1210 | bad_area(regs, error_code, address); | |
1211 | return; | |
1212 | } | |
1213 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1214 | goto good_area; |
92181f19 NP |
1215 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1216 | bad_area(regs, error_code, address); | |
1217 | return; | |
1218 | } | |
33cb5243 | 1219 | if (error_code & PF_USER) { |
6f4d368e HH |
1220 | /* |
1221 | * Accessing the stack below %sp is always a bug. | |
1222 | * The large cushion allows instructions like enter | |
2d4a7167 | 1223 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1224 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1225 | */ |
92181f19 NP |
1226 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1227 | bad_area(regs, error_code, address); | |
1228 | return; | |
1229 | } | |
1da177e4 | 1230 | } |
92181f19 NP |
1231 | if (unlikely(expand_stack(vma, address))) { |
1232 | bad_area(regs, error_code, address); | |
1233 | return; | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * Ok, we have a good vm_area for this memory access, so | |
1238 | * we can handle it.. | |
1239 | */ | |
1da177e4 | 1240 | good_area: |
68da336a | 1241 | if (unlikely(access_error(error_code, vma))) { |
7b2d0dba | 1242 | bad_area_access_error(regs, error_code, address, vma); |
92181f19 | 1243 | return; |
1da177e4 LT |
1244 | } |
1245 | ||
1246 | /* | |
1247 | * If for any reason at all we couldn't handle the fault, | |
1248 | * make sure we exit gracefully rather than endlessly redo | |
9a95f3cf PC |
1249 | * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if |
1250 | * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. | |
1da177e4 | 1251 | */ |
d065bd81 | 1252 | fault = handle_mm_fault(mm, vma, address, flags); |
26178ec1 | 1253 | major |= fault & VM_FAULT_MAJOR; |
2d4a7167 | 1254 | |
3a13c4d7 | 1255 | /* |
26178ec1 LT |
1256 | * If we need to retry the mmap_sem has already been released, |
1257 | * and if there is a fatal signal pending there is no guarantee | |
1258 | * that we made any progress. Handle this case first. | |
3a13c4d7 | 1259 | */ |
26178ec1 LT |
1260 | if (unlikely(fault & VM_FAULT_RETRY)) { |
1261 | /* Retry at most once */ | |
1262 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
1263 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
1264 | flags |= FAULT_FLAG_TRIED; | |
1265 | if (!fatal_signal_pending(tsk)) | |
1266 | goto retry; | |
1267 | } | |
1268 | ||
1269 | /* User mode? Just return to handle the fatal exception */ | |
cf3c0a15 | 1270 | if (flags & FAULT_FLAG_USER) |
26178ec1 LT |
1271 | return; |
1272 | ||
1273 | /* Not returning to user mode? Handle exceptions or die: */ | |
1274 | no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); | |
3a13c4d7 | 1275 | return; |
26178ec1 | 1276 | } |
3a13c4d7 | 1277 | |
26178ec1 | 1278 | up_read(&mm->mmap_sem); |
3a13c4d7 | 1279 | if (unlikely(fault & VM_FAULT_ERROR)) { |
7b2d0dba | 1280 | mm_fault_error(regs, error_code, address, vma, fault); |
3a13c4d7 | 1281 | return; |
37b23e05 KM |
1282 | } |
1283 | ||
d065bd81 | 1284 | /* |
26178ec1 LT |
1285 | * Major/minor page fault accounting. If any of the events |
1286 | * returned VM_FAULT_MAJOR, we account it as a major fault. | |
d065bd81 | 1287 | */ |
26178ec1 LT |
1288 | if (major) { |
1289 | tsk->maj_flt++; | |
1290 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); | |
1291 | } else { | |
1292 | tsk->min_flt++; | |
1293 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); | |
ac17dc8e | 1294 | } |
d729ab35 | 1295 | |
8c938f9f | 1296 | check_v8086_mode(regs, address, tsk); |
1da177e4 | 1297 | } |
9326638c | 1298 | NOKPROBE_SYMBOL(__do_page_fault); |
6ba3c97a | 1299 | |
9326638c | 1300 | dotraplinkage void notrace |
6ba3c97a FW |
1301 | do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1302 | { | |
d4078e23 | 1303 | unsigned long address = read_cr2(); /* Get the faulting address */ |
6c1e0256 | 1304 | enum ctx_state prev_state; |
d4078e23 PZ |
1305 | |
1306 | /* | |
1307 | * We must have this function tagged with __kprobes, notrace and call | |
1308 | * read_cr2() before calling anything else. To avoid calling any kind | |
1309 | * of tracing machinery before we've observed the CR2 value. | |
1310 | * | |
1311 | * exception_{enter,exit}() contain all sorts of tracepoints. | |
1312 | */ | |
6c1e0256 FW |
1313 | |
1314 | prev_state = exception_enter(); | |
0ac09f9f | 1315 | __do_page_fault(regs, error_code, address); |
6c1e0256 | 1316 | exception_exit(prev_state); |
6ba3c97a | 1317 | } |
9326638c | 1318 | NOKPROBE_SYMBOL(do_page_fault); |
25c74b10 | 1319 | |
d4078e23 | 1320 | #ifdef CONFIG_TRACING |
9326638c MH |
1321 | static nokprobe_inline void |
1322 | trace_page_fault_entries(unsigned long address, struct pt_regs *regs, | |
1323 | unsigned long error_code) | |
d34603b0 SA |
1324 | { |
1325 | if (user_mode(regs)) | |
d4078e23 | 1326 | trace_page_fault_user(address, regs, error_code); |
d34603b0 | 1327 | else |
d4078e23 | 1328 | trace_page_fault_kernel(address, regs, error_code); |
d34603b0 SA |
1329 | } |
1330 | ||
9326638c | 1331 | dotraplinkage void notrace |
25c74b10 SA |
1332 | trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) |
1333 | { | |
0ac09f9f JO |
1334 | /* |
1335 | * The exception_enter and tracepoint processing could | |
1336 | * trigger another page faults (user space callchain | |
1337 | * reading) and destroy the original cr2 value, so read | |
1338 | * the faulting address now. | |
1339 | */ | |
1340 | unsigned long address = read_cr2(); | |
d4078e23 | 1341 | enum ctx_state prev_state; |
25c74b10 SA |
1342 | |
1343 | prev_state = exception_enter(); | |
d4078e23 | 1344 | trace_page_fault_entries(address, regs, error_code); |
0ac09f9f | 1345 | __do_page_fault(regs, error_code, address); |
25c74b10 SA |
1346 | exception_exit(prev_state); |
1347 | } | |
9326638c | 1348 | NOKPROBE_SYMBOL(trace_do_page_fault); |
d4078e23 | 1349 | #endif /* CONFIG_TRACING */ |