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