]>
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/magic.h> /* STACK_END_MAGIC */ |
7 | #include <linux/sched.h> /* test_thread_flag(), ... */ | |
8 | #include <linux/kdebug.h> /* oops_begin/end, ... */ | |
9 | #include <linux/module.h> /* search_exception_table */ | |
10 | #include <linux/bootmem.h> /* max_low_pfn */ | |
11 | #include <linux/kprobes.h> /* __kprobes, ... */ | |
12 | #include <linux/mmiotrace.h> /* kmmio_handler, ... */ | |
cdd6c482 | 13 | #include <linux/perf_event.h> /* perf_sw_event */ |
2d4a7167 | 14 | |
a2bcd473 IM |
15 | #include <asm/traps.h> /* dotraplinkage, ... */ |
16 | #include <asm/pgalloc.h> /* pgd_*(), ... */ | |
f8561296 | 17 | #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ |
1da177e4 | 18 | |
33cb5243 | 19 | /* |
2d4a7167 IM |
20 | * Page fault error code bits: |
21 | * | |
22 | * bit 0 == 0: no page found 1: protection fault | |
23 | * bit 1 == 0: read access 1: write access | |
24 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
25 | * bit 3 == 1: use of reserved bit detected | |
26 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 27 | */ |
2d4a7167 IM |
28 | enum x86_pf_error_code { |
29 | ||
30 | PF_PROT = 1 << 0, | |
31 | PF_WRITE = 1 << 1, | |
32 | PF_USER = 1 << 2, | |
33 | PF_RSVD = 1 << 3, | |
34 | PF_INSTR = 1 << 4, | |
35 | }; | |
66c58156 | 36 | |
b814d41f | 37 | /* |
b319eed0 IM |
38 | * Returns 0 if mmiotrace is disabled, or if the fault is not |
39 | * handled by mmiotrace: | |
b814d41f | 40 | */ |
62c9295f MH |
41 | static inline int __kprobes |
42 | kmmio_fault(struct pt_regs *regs, unsigned long addr) | |
86069782 | 43 | { |
0fd0e3da PP |
44 | if (unlikely(is_kmmio_active())) |
45 | if (kmmio_handler(regs, addr) == 1) | |
46 | return -1; | |
0fd0e3da | 47 | return 0; |
86069782 PP |
48 | } |
49 | ||
62c9295f | 50 | static inline int __kprobes notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 51 | { |
74a0b576 CH |
52 | int ret = 0; |
53 | ||
54 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 55 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
56 | preempt_disable(); |
57 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
58 | ret = 1; | |
59 | preempt_enable(); | |
60 | } | |
1bd858a5 | 61 | |
74a0b576 | 62 | return ret; |
33cb5243 | 63 | } |
1bd858a5 | 64 | |
1dc85be0 | 65 | /* |
2d4a7167 IM |
66 | * Prefetch quirks: |
67 | * | |
68 | * 32-bit mode: | |
69 | * | |
70 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
71 | * Check that here and ignore it. | |
1dc85be0 | 72 | * |
2d4a7167 | 73 | * 64-bit mode: |
1dc85be0 | 74 | * |
2d4a7167 IM |
75 | * Sometimes the CPU reports invalid exceptions on prefetch. |
76 | * Check that here and ignore it. | |
77 | * | |
78 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 79 | */ |
107a0367 IM |
80 | static inline int |
81 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
82 | unsigned char opcode, int *prefetch) | |
83 | { | |
84 | unsigned char instr_hi = opcode & 0xf0; | |
85 | unsigned char instr_lo = opcode & 0x0f; | |
86 | ||
87 | switch (instr_hi) { | |
88 | case 0x20: | |
89 | case 0x30: | |
90 | /* | |
91 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
92 | * In X86_64 long mode, the CPU will signal invalid | |
93 | * opcode if some of these prefixes are present so | |
94 | * X86_64 will never get here anyway | |
95 | */ | |
96 | return ((instr_lo & 7) == 0x6); | |
97 | #ifdef CONFIG_X86_64 | |
98 | case 0x40: | |
99 | /* | |
100 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
101 | * Need to figure out under what instruction mode the | |
102 | * instruction was issued. Could check the LDT for lm, | |
103 | * but for now it's good enough to assume that long | |
104 | * mode only uses well known segments or kernel. | |
105 | */ | |
106 | return (!user_mode(regs)) || (regs->cs == __USER_CS); | |
107 | #endif | |
108 | case 0x60: | |
109 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
110 | return (instr_lo & 0xC) == 0x4; | |
111 | case 0xF0: | |
112 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
113 | return !instr_lo || (instr_lo>>1) == 1; | |
114 | case 0x00: | |
115 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
116 | if (probe_kernel_address(instr, opcode)) | |
117 | return 0; | |
118 | ||
119 | *prefetch = (instr_lo == 0xF) && | |
120 | (opcode == 0x0D || opcode == 0x18); | |
121 | return 0; | |
122 | default: | |
123 | return 0; | |
124 | } | |
125 | } | |
126 | ||
2d4a7167 IM |
127 | static int |
128 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 129 | { |
2d4a7167 | 130 | unsigned char *max_instr; |
ab2bf0c1 | 131 | unsigned char *instr; |
33cb5243 | 132 | int prefetch = 0; |
1da177e4 | 133 | |
3085354d IM |
134 | /* |
135 | * If it was a exec (instruction fetch) fault on NX page, then | |
136 | * do not ignore the fault: | |
137 | */ | |
66c58156 | 138 | if (error_code & PF_INSTR) |
1da177e4 | 139 | return 0; |
1dc85be0 | 140 | |
107a0367 | 141 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 142 | max_instr = instr + 15; |
1da177e4 | 143 | |
76381fee | 144 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
145 | return 0; |
146 | ||
107a0367 | 147 | while (instr < max_instr) { |
2d4a7167 | 148 | unsigned char opcode; |
1da177e4 | 149 | |
ab2bf0c1 | 150 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 151 | break; |
1da177e4 | 152 | |
1da177e4 LT |
153 | instr++; |
154 | ||
107a0367 | 155 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 156 | break; |
1da177e4 LT |
157 | } |
158 | return prefetch; | |
159 | } | |
160 | ||
2d4a7167 IM |
161 | static void |
162 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
163 | struct task_struct *tsk) | |
c4aba4a8 HH |
164 | { |
165 | siginfo_t info; | |
166 | ||
2d4a7167 IM |
167 | info.si_signo = si_signo; |
168 | info.si_errno = 0; | |
169 | info.si_code = si_code; | |
170 | info.si_addr = (void __user *)address; | |
a6e04aa9 | 171 | info.si_addr_lsb = si_code == BUS_MCEERR_AR ? PAGE_SHIFT : 0; |
2d4a7167 | 172 | |
c4aba4a8 HH |
173 | force_sig_info(si_signo, &info, tsk); |
174 | } | |
175 | ||
f2f13a85 IM |
176 | DEFINE_SPINLOCK(pgd_lock); |
177 | LIST_HEAD(pgd_list); | |
178 | ||
179 | #ifdef CONFIG_X86_32 | |
180 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
33cb5243 | 181 | { |
f2f13a85 IM |
182 | unsigned index = pgd_index(address); |
183 | pgd_t *pgd_k; | |
184 | pud_t *pud, *pud_k; | |
185 | pmd_t *pmd, *pmd_k; | |
2d4a7167 | 186 | |
f2f13a85 IM |
187 | pgd += index; |
188 | pgd_k = init_mm.pgd + index; | |
189 | ||
190 | if (!pgd_present(*pgd_k)) | |
191 | return NULL; | |
192 | ||
193 | /* | |
194 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
195 | * and redundant with the set_pmd() on non-PAE. As would | |
196 | * set_pud. | |
197 | */ | |
198 | pud = pud_offset(pgd, address); | |
199 | pud_k = pud_offset(pgd_k, address); | |
200 | if (!pud_present(*pud_k)) | |
201 | return NULL; | |
202 | ||
203 | pmd = pmd_offset(pud, address); | |
204 | pmd_k = pmd_offset(pud_k, address); | |
205 | if (!pmd_present(*pmd_k)) | |
206 | return NULL; | |
207 | ||
b8bcfe99 | 208 | if (!pmd_present(*pmd)) |
f2f13a85 | 209 | set_pmd(pmd, *pmd_k); |
b8bcfe99 | 210 | else |
f2f13a85 | 211 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
f2f13a85 IM |
212 | |
213 | return pmd_k; | |
214 | } | |
215 | ||
216 | void vmalloc_sync_all(void) | |
217 | { | |
218 | unsigned long address; | |
219 | ||
220 | if (SHARED_KERNEL_PMD) | |
221 | return; | |
222 | ||
223 | for (address = VMALLOC_START & PMD_MASK; | |
224 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
225 | address += PMD_SIZE) { | |
226 | ||
227 | unsigned long flags; | |
228 | struct page *page; | |
229 | ||
230 | spin_lock_irqsave(&pgd_lock, flags); | |
231 | list_for_each_entry(page, &pgd_list, lru) { | |
617d34d9 JF |
232 | spinlock_t *pgt_lock; |
233 | int ret; | |
234 | ||
235 | pgt_lock = &pgd_page_get_mm(page)->page_table_lock; | |
236 | ||
237 | spin_lock(pgt_lock); | |
238 | ret = vmalloc_sync_one(page_address(page), address); | |
239 | spin_unlock(pgt_lock); | |
240 | ||
241 | if (!ret) | |
f2f13a85 IM |
242 | break; |
243 | } | |
244 | spin_unlock_irqrestore(&pgd_lock, flags); | |
245 | } | |
246 | } | |
247 | ||
248 | /* | |
249 | * 32-bit: | |
250 | * | |
251 | * Handle a fault on the vmalloc or module mapping area | |
252 | */ | |
62c9295f | 253 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
254 | { |
255 | unsigned long pgd_paddr; | |
256 | pmd_t *pmd_k; | |
257 | pte_t *pte_k; | |
258 | ||
259 | /* Make sure we are in vmalloc area: */ | |
260 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
261 | return -1; | |
262 | ||
263 | /* | |
264 | * Synchronize this task's top level page-table | |
265 | * with the 'reference' page table. | |
266 | * | |
267 | * Do _not_ use "current" here. We might be inside | |
268 | * an interrupt in the middle of a task switch.. | |
269 | */ | |
270 | pgd_paddr = read_cr3(); | |
271 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
272 | if (!pmd_k) | |
273 | return -1; | |
274 | ||
275 | pte_k = pte_offset_kernel(pmd_k, address); | |
276 | if (!pte_present(*pte_k)) | |
277 | return -1; | |
278 | ||
279 | return 0; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Did it hit the DOS screen memory VA from vm86 mode? | |
284 | */ | |
285 | static inline void | |
286 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
287 | struct task_struct *tsk) | |
288 | { | |
289 | unsigned long bit; | |
290 | ||
291 | if (!v8086_mode(regs)) | |
292 | return; | |
293 | ||
294 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
295 | if (bit < 32) | |
296 | tsk->thread.screen_bitmap |= 1 << bit; | |
33cb5243 | 297 | } |
1da177e4 | 298 | |
087975b0 | 299 | static bool low_pfn(unsigned long pfn) |
1da177e4 | 300 | { |
087975b0 AM |
301 | return pfn < max_low_pfn; |
302 | } | |
1156e098 | 303 | |
087975b0 AM |
304 | static void dump_pagetable(unsigned long address) |
305 | { | |
306 | pgd_t *base = __va(read_cr3()); | |
307 | pgd_t *pgd = &base[pgd_index(address)]; | |
308 | pmd_t *pmd; | |
309 | pte_t *pte; | |
2d4a7167 | 310 | |
1156e098 | 311 | #ifdef CONFIG_X86_PAE |
087975b0 AM |
312 | printk("*pdpt = %016Lx ", pgd_val(*pgd)); |
313 | if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) | |
314 | goto out; | |
1156e098 | 315 | #endif |
087975b0 AM |
316 | pmd = pmd_offset(pud_offset(pgd, address), address); |
317 | printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); | |
1156e098 HH |
318 | |
319 | /* | |
320 | * We must not directly access the pte in the highpte | |
321 | * case if the page table is located in highmem. | |
322 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 323 | * it's allocated already: |
1156e098 | 324 | */ |
087975b0 AM |
325 | if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) |
326 | goto out; | |
1156e098 | 327 | |
087975b0 AM |
328 | pte = pte_offset_kernel(pmd, address); |
329 | printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); | |
330 | out: | |
1156e098 | 331 | printk("\n"); |
f2f13a85 IM |
332 | } |
333 | ||
334 | #else /* CONFIG_X86_64: */ | |
335 | ||
336 | void vmalloc_sync_all(void) | |
337 | { | |
6afb5157 | 338 | sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); |
f2f13a85 IM |
339 | } |
340 | ||
341 | /* | |
342 | * 64-bit: | |
343 | * | |
344 | * Handle a fault on the vmalloc area | |
345 | * | |
346 | * This assumes no large pages in there. | |
347 | */ | |
62c9295f | 348 | static noinline __kprobes int vmalloc_fault(unsigned long address) |
f2f13a85 IM |
349 | { |
350 | pgd_t *pgd, *pgd_ref; | |
351 | pud_t *pud, *pud_ref; | |
352 | pmd_t *pmd, *pmd_ref; | |
353 | pte_t *pte, *pte_ref; | |
354 | ||
355 | /* Make sure we are in vmalloc area: */ | |
356 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) | |
357 | return -1; | |
358 | ||
359 | /* | |
360 | * Copy kernel mappings over when needed. This can also | |
361 | * happen within a race in page table update. In the later | |
362 | * case just flush: | |
363 | */ | |
364 | pgd = pgd_offset(current->active_mm, address); | |
365 | pgd_ref = pgd_offset_k(address); | |
366 | if (pgd_none(*pgd_ref)) | |
367 | return -1; | |
368 | ||
369 | if (pgd_none(*pgd)) | |
370 | set_pgd(pgd, *pgd_ref); | |
371 | else | |
372 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
373 | ||
374 | /* | |
375 | * Below here mismatches are bugs because these lower tables | |
376 | * are shared: | |
377 | */ | |
378 | ||
379 | pud = pud_offset(pgd, address); | |
380 | pud_ref = pud_offset(pgd_ref, address); | |
381 | if (pud_none(*pud_ref)) | |
382 | return -1; | |
383 | ||
384 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) | |
385 | BUG(); | |
386 | ||
387 | pmd = pmd_offset(pud, address); | |
388 | pmd_ref = pmd_offset(pud_ref, address); | |
389 | if (pmd_none(*pmd_ref)) | |
390 | return -1; | |
391 | ||
392 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
393 | BUG(); | |
394 | ||
395 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
396 | if (!pte_present(*pte_ref)) | |
397 | return -1; | |
398 | ||
399 | pte = pte_offset_kernel(pmd, address); | |
400 | ||
401 | /* | |
402 | * Don't use pte_page here, because the mappings can point | |
403 | * outside mem_map, and the NUMA hash lookup cannot handle | |
404 | * that: | |
405 | */ | |
406 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
407 | BUG(); | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
412 | static const char errata93_warning[] = | |
ad361c98 JP |
413 | KERN_ERR |
414 | "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
415 | "******* Working around it, but it may cause SEGVs or burn power.\n" | |
416 | "******* Please consider a BIOS update.\n" | |
417 | "******* Disabling USB legacy in the BIOS may also help.\n"; | |
f2f13a85 IM |
418 | |
419 | /* | |
420 | * No vm86 mode in 64-bit mode: | |
421 | */ | |
422 | static inline void | |
423 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
424 | struct task_struct *tsk) | |
425 | { | |
426 | } | |
427 | ||
428 | static int bad_address(void *p) | |
429 | { | |
430 | unsigned long dummy; | |
431 | ||
432 | return probe_kernel_address((unsigned long *)p, dummy); | |
433 | } | |
434 | ||
435 | static void dump_pagetable(unsigned long address) | |
436 | { | |
087975b0 AM |
437 | pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); |
438 | pgd_t *pgd = base + pgd_index(address); | |
1da177e4 LT |
439 | pud_t *pud; |
440 | pmd_t *pmd; | |
441 | pte_t *pte; | |
442 | ||
2d4a7167 IM |
443 | if (bad_address(pgd)) |
444 | goto bad; | |
445 | ||
d646bce4 | 446 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
447 | |
448 | if (!pgd_present(*pgd)) | |
449 | goto out; | |
1da177e4 | 450 | |
d2ae5b5f | 451 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
452 | if (bad_address(pud)) |
453 | goto bad; | |
454 | ||
1da177e4 | 455 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 456 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 457 | goto out; |
1da177e4 LT |
458 | |
459 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
460 | if (bad_address(pmd)) |
461 | goto bad; | |
462 | ||
1da177e4 | 463 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
464 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
465 | goto out; | |
1da177e4 LT |
466 | |
467 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
468 | if (bad_address(pte)) |
469 | goto bad; | |
470 | ||
33cb5243 | 471 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 472 | out: |
1da177e4 LT |
473 | printk("\n"); |
474 | return; | |
475 | bad: | |
476 | printk("BAD\n"); | |
8c938f9f IM |
477 | } |
478 | ||
f2f13a85 | 479 | #endif /* CONFIG_X86_64 */ |
1da177e4 | 480 | |
2d4a7167 IM |
481 | /* |
482 | * Workaround for K8 erratum #93 & buggy BIOS. | |
483 | * | |
484 | * BIOS SMM functions are required to use a specific workaround | |
485 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
486 | * | |
487 | * A lot of BIOS that didn't get tested properly miss this. | |
488 | * | |
489 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
490 | * Try to work around it here. | |
491 | * | |
492 | * Note we only handle faults in kernel here. | |
493 | * Does nothing on 32-bit. | |
fdfe8aa8 | 494 | */ |
33cb5243 | 495 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 496 | { |
fdfe8aa8 | 497 | #ifdef CONFIG_X86_64 |
65ea5b03 | 498 | if (address != regs->ip) |
1da177e4 | 499 | return 0; |
2d4a7167 | 500 | |
33cb5243 | 501 | if ((address >> 32) != 0) |
1da177e4 | 502 | return 0; |
2d4a7167 | 503 | |
1da177e4 | 504 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
505 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
506 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
a454ab31 | 507 | printk_once(errata93_warning); |
65ea5b03 | 508 | regs->ip = address; |
1da177e4 LT |
509 | return 1; |
510 | } | |
fdfe8aa8 | 511 | #endif |
1da177e4 | 512 | return 0; |
33cb5243 | 513 | } |
1da177e4 | 514 | |
35f3266f | 515 | /* |
2d4a7167 IM |
516 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
517 | * to illegal addresses >4GB. | |
518 | * | |
519 | * We catch this in the page fault handler because these addresses | |
520 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
521 | * segment in LDT is compatibility mode. |
522 | */ | |
523 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
524 | { | |
525 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 526 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
527 | return 1; |
528 | #endif | |
529 | return 0; | |
530 | } | |
531 | ||
29caf2f9 HH |
532 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
533 | { | |
534 | #ifdef CONFIG_X86_F00F_BUG | |
535 | unsigned long nr; | |
2d4a7167 | 536 | |
29caf2f9 | 537 | /* |
2d4a7167 | 538 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 HH |
539 | */ |
540 | if (boot_cpu_data.f00f_bug) { | |
541 | nr = (address - idt_descr.address) >> 3; | |
542 | ||
543 | if (nr == 6) { | |
544 | do_invalid_op(regs, 0); | |
545 | return 1; | |
546 | } | |
547 | } | |
548 | #endif | |
549 | return 0; | |
550 | } | |
551 | ||
8f766149 IM |
552 | static const char nx_warning[] = KERN_CRIT |
553 | "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; | |
554 | ||
2d4a7167 IM |
555 | static void |
556 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
557 | unsigned long address) | |
b3279c7f | 558 | { |
1156e098 HH |
559 | if (!oops_may_print()) |
560 | return; | |
561 | ||
1156e098 | 562 | if (error_code & PF_INSTR) { |
93809be8 | 563 | unsigned int level; |
2d4a7167 | 564 | |
1156e098 HH |
565 | pte_t *pte = lookup_address(address, &level); |
566 | ||
8f766149 IM |
567 | if (pte && pte_present(*pte) && !pte_exec(*pte)) |
568 | printk(nx_warning, current_uid()); | |
1156e098 | 569 | } |
1156e098 | 570 | |
19f0dda9 | 571 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 572 | if (address < PAGE_SIZE) |
19f0dda9 | 573 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 574 | else |
19f0dda9 | 575 | printk(KERN_CONT "paging request"); |
2d4a7167 | 576 | |
f294a8ce | 577 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 578 | printk(KERN_ALERT "IP:"); |
b3279c7f | 579 | printk_address(regs->ip, 1); |
2d4a7167 | 580 | |
b3279c7f HH |
581 | dump_pagetable(address); |
582 | } | |
583 | ||
2d4a7167 IM |
584 | static noinline void |
585 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
586 | unsigned long address) | |
1da177e4 | 587 | { |
2d4a7167 IM |
588 | struct task_struct *tsk; |
589 | unsigned long flags; | |
590 | int sig; | |
591 | ||
592 | flags = oops_begin(); | |
593 | tsk = current; | |
594 | sig = SIGKILL; | |
1209140c | 595 | |
1da177e4 | 596 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 597 | tsk->comm, address); |
1da177e4 | 598 | dump_pagetable(address); |
2d4a7167 IM |
599 | |
600 | tsk->thread.cr2 = address; | |
601 | tsk->thread.trap_no = 14; | |
602 | tsk->thread.error_code = error_code; | |
603 | ||
22f5991c | 604 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 605 | sig = 0; |
2d4a7167 | 606 | |
874d93d1 | 607 | oops_end(flags, regs, sig); |
1da177e4 LT |
608 | } |
609 | ||
2d4a7167 IM |
610 | static noinline void |
611 | no_context(struct pt_regs *regs, unsigned long error_code, | |
612 | unsigned long address) | |
92181f19 NP |
613 | { |
614 | struct task_struct *tsk = current; | |
19803078 | 615 | unsigned long *stackend; |
92181f19 NP |
616 | unsigned long flags; |
617 | int sig; | |
92181f19 | 618 | |
2d4a7167 | 619 | /* Are we prepared to handle this kernel fault? */ |
92181f19 NP |
620 | if (fixup_exception(regs)) |
621 | return; | |
622 | ||
623 | /* | |
2d4a7167 IM |
624 | * 32-bit: |
625 | * | |
626 | * Valid to do another page fault here, because if this fault | |
627 | * had been triggered by is_prefetch fixup_exception would have | |
628 | * handled it. | |
629 | * | |
630 | * 64-bit: | |
92181f19 | 631 | * |
2d4a7167 | 632 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
633 | */ |
634 | if (is_prefetch(regs, error_code, address)) | |
635 | return; | |
636 | ||
637 | if (is_errata93(regs, address)) | |
638 | return; | |
639 | ||
640 | /* | |
641 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 642 | * terminate things with extreme prejudice: |
92181f19 | 643 | */ |
92181f19 | 644 | flags = oops_begin(); |
92181f19 NP |
645 | |
646 | show_fault_oops(regs, error_code, address); | |
647 | ||
2d4a7167 | 648 | stackend = end_of_stack(tsk); |
0e7810be | 649 | if (tsk != &init_task && *stackend != STACK_END_MAGIC) |
19803078 IM |
650 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); |
651 | ||
1cc99544 IM |
652 | tsk->thread.cr2 = address; |
653 | tsk->thread.trap_no = 14; | |
654 | tsk->thread.error_code = error_code; | |
92181f19 | 655 | |
92181f19 NP |
656 | sig = SIGKILL; |
657 | if (__die("Oops", regs, error_code)) | |
658 | sig = 0; | |
2d4a7167 | 659 | |
92181f19 NP |
660 | /* Executive summary in case the body of the oops scrolled away */ |
661 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
2d4a7167 | 662 | |
92181f19 | 663 | oops_end(flags, regs, sig); |
92181f19 NP |
664 | } |
665 | ||
2d4a7167 IM |
666 | /* |
667 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
668 | * sysctl is set: | |
669 | */ | |
670 | static inline void | |
671 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
672 | unsigned long address, struct task_struct *tsk) | |
673 | { | |
674 | if (!unhandled_signal(tsk, SIGSEGV)) | |
675 | return; | |
676 | ||
677 | if (!printk_ratelimit()) | |
678 | return; | |
679 | ||
a1a08d1c | 680 | printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", |
2d4a7167 IM |
681 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, |
682 | tsk->comm, task_pid_nr(tsk), address, | |
683 | (void *)regs->ip, (void *)regs->sp, error_code); | |
684 | ||
685 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
686 | ||
687 | printk(KERN_CONT "\n"); | |
688 | } | |
689 | ||
690 | static void | |
691 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
692 | unsigned long address, int si_code) | |
92181f19 NP |
693 | { |
694 | struct task_struct *tsk = current; | |
695 | ||
696 | /* User mode accesses just cause a SIGSEGV */ | |
697 | if (error_code & PF_USER) { | |
698 | /* | |
2d4a7167 | 699 | * It's possible to have interrupts off here: |
92181f19 NP |
700 | */ |
701 | local_irq_enable(); | |
702 | ||
703 | /* | |
704 | * Valid to do another page fault here because this one came | |
2d4a7167 | 705 | * from user space: |
92181f19 NP |
706 | */ |
707 | if (is_prefetch(regs, error_code, address)) | |
708 | return; | |
709 | ||
710 | if (is_errata100(regs, address)) | |
711 | return; | |
712 | ||
2d4a7167 IM |
713 | if (unlikely(show_unhandled_signals)) |
714 | show_signal_msg(regs, error_code, address, tsk); | |
715 | ||
716 | /* Kernel addresses are always protection faults: */ | |
717 | tsk->thread.cr2 = address; | |
718 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
719 | tsk->thread.trap_no = 14; | |
92181f19 | 720 | |
92181f19 | 721 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
2d4a7167 | 722 | |
92181f19 NP |
723 | return; |
724 | } | |
725 | ||
726 | if (is_f00f_bug(regs, address)) | |
727 | return; | |
728 | ||
729 | no_context(regs, error_code, address); | |
730 | } | |
731 | ||
2d4a7167 IM |
732 | static noinline void |
733 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
734 | unsigned long address) | |
92181f19 NP |
735 | { |
736 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
737 | } | |
738 | ||
2d4a7167 IM |
739 | static void |
740 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
741 | unsigned long address, int si_code) | |
92181f19 NP |
742 | { |
743 | struct mm_struct *mm = current->mm; | |
744 | ||
745 | /* | |
746 | * Something tried to access memory that isn't in our memory map.. | |
747 | * Fix it, but check if it's kernel or user first.. | |
748 | */ | |
749 | up_read(&mm->mmap_sem); | |
750 | ||
751 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
752 | } | |
753 | ||
2d4a7167 IM |
754 | static noinline void |
755 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
756 | { |
757 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
758 | } | |
759 | ||
2d4a7167 IM |
760 | static noinline void |
761 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
762 | unsigned long address) | |
92181f19 NP |
763 | { |
764 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
765 | } | |
766 | ||
767 | /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */ | |
2d4a7167 IM |
768 | static void |
769 | out_of_memory(struct pt_regs *regs, unsigned long error_code, | |
770 | unsigned long address) | |
92181f19 NP |
771 | { |
772 | /* | |
773 | * We ran out of memory, call the OOM killer, and return the userspace | |
2d4a7167 | 774 | * (which will retry the fault, or kill us if we got oom-killed): |
92181f19 NP |
775 | */ |
776 | up_read(¤t->mm->mmap_sem); | |
2d4a7167 | 777 | |
92181f19 NP |
778 | pagefault_out_of_memory(); |
779 | } | |
780 | ||
2d4a7167 | 781 | static void |
a6e04aa9 AK |
782 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, |
783 | unsigned int fault) | |
92181f19 NP |
784 | { |
785 | struct task_struct *tsk = current; | |
786 | struct mm_struct *mm = tsk->mm; | |
a6e04aa9 | 787 | int code = BUS_ADRERR; |
92181f19 NP |
788 | |
789 | up_read(&mm->mmap_sem); | |
790 | ||
2d4a7167 | 791 | /* Kernel mode? Handle exceptions or die: */ |
96054569 | 792 | if (!(error_code & PF_USER)) { |
92181f19 | 793 | no_context(regs, error_code, address); |
96054569 LT |
794 | return; |
795 | } | |
2d4a7167 | 796 | |
cd1b68f0 | 797 | /* User-space => ok to do another page fault: */ |
92181f19 NP |
798 | if (is_prefetch(regs, error_code, address)) |
799 | return; | |
2d4a7167 IM |
800 | |
801 | tsk->thread.cr2 = address; | |
802 | tsk->thread.error_code = error_code; | |
803 | tsk->thread.trap_no = 14; | |
804 | ||
a6e04aa9 AK |
805 | #ifdef CONFIG_MEMORY_FAILURE |
806 | if (fault & VM_FAULT_HWPOISON) { | |
807 | printk(KERN_ERR | |
808 | "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", | |
809 | tsk->comm, tsk->pid, address); | |
810 | code = BUS_MCEERR_AR; | |
811 | } | |
812 | #endif | |
813 | force_sig_info_fault(SIGBUS, code, address, tsk); | |
92181f19 NP |
814 | } |
815 | ||
2d4a7167 IM |
816 | static noinline void |
817 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, | |
818 | unsigned long address, unsigned int fault) | |
92181f19 | 819 | { |
2d4a7167 | 820 | if (fault & VM_FAULT_OOM) { |
92181f19 | 821 | out_of_memory(regs, error_code, address); |
2d4a7167 | 822 | } else { |
a6e04aa9 AK |
823 | if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON)) |
824 | do_sigbus(regs, error_code, address, fault); | |
2d4a7167 IM |
825 | else |
826 | BUG(); | |
827 | } | |
92181f19 NP |
828 | } |
829 | ||
d8b57bb7 TG |
830 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
831 | { | |
832 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
833 | return 0; | |
2d4a7167 | 834 | |
d8b57bb7 TG |
835 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
836 | return 0; | |
837 | ||
838 | return 1; | |
839 | } | |
840 | ||
5b727a3b | 841 | /* |
2d4a7167 IM |
842 | * Handle a spurious fault caused by a stale TLB entry. |
843 | * | |
844 | * This allows us to lazily refresh the TLB when increasing the | |
845 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
846 | * eagerly is very expensive since that implies doing a full | |
847 | * cross-processor TLB flush, even if no stale TLB entries exist | |
848 | * on other processors. | |
849 | * | |
5b727a3b JF |
850 | * There are no security implications to leaving a stale TLB when |
851 | * increasing the permissions on a page. | |
852 | */ | |
62c9295f | 853 | static noinline __kprobes int |
2d4a7167 | 854 | spurious_fault(unsigned long error_code, unsigned long address) |
5b727a3b JF |
855 | { |
856 | pgd_t *pgd; | |
857 | pud_t *pud; | |
858 | pmd_t *pmd; | |
859 | pte_t *pte; | |
3c3e5694 | 860 | int ret; |
5b727a3b JF |
861 | |
862 | /* Reserved-bit violation or user access to kernel space? */ | |
863 | if (error_code & (PF_USER | PF_RSVD)) | |
864 | return 0; | |
865 | ||
866 | pgd = init_mm.pgd + pgd_index(address); | |
867 | if (!pgd_present(*pgd)) | |
868 | return 0; | |
869 | ||
870 | pud = pud_offset(pgd, address); | |
871 | if (!pud_present(*pud)) | |
872 | return 0; | |
873 | ||
d8b57bb7 TG |
874 | if (pud_large(*pud)) |
875 | return spurious_fault_check(error_code, (pte_t *) pud); | |
876 | ||
5b727a3b JF |
877 | pmd = pmd_offset(pud, address); |
878 | if (!pmd_present(*pmd)) | |
879 | return 0; | |
880 | ||
d8b57bb7 TG |
881 | if (pmd_large(*pmd)) |
882 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
883 | ||
660a293e SL |
884 | /* |
885 | * Note: don't use pte_present() here, since it returns true | |
886 | * if the _PAGE_PROTNONE bit is set. However, this aliases the | |
887 | * _PAGE_GLOBAL bit, which for kernel pages give false positives | |
888 | * when CONFIG_DEBUG_PAGEALLOC is used. | |
889 | */ | |
5b727a3b | 890 | pte = pte_offset_kernel(pmd, address); |
660a293e | 891 | if (!(pte_flags(*pte) & _PAGE_PRESENT)) |
5b727a3b JF |
892 | return 0; |
893 | ||
3c3e5694 SR |
894 | ret = spurious_fault_check(error_code, pte); |
895 | if (!ret) | |
896 | return 0; | |
897 | ||
898 | /* | |
2d4a7167 IM |
899 | * Make sure we have permissions in PMD. |
900 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
901 | */ |
902 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
903 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 904 | |
3c3e5694 | 905 | return ret; |
5b727a3b JF |
906 | } |
907 | ||
abd4f750 | 908 | int show_unhandled_signals = 1; |
1da177e4 | 909 | |
2d4a7167 IM |
910 | static inline int |
911 | access_error(unsigned long error_code, int write, struct vm_area_struct *vma) | |
92181f19 NP |
912 | { |
913 | if (write) { | |
2d4a7167 | 914 | /* write, present and write, not present: */ |
92181f19 NP |
915 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
916 | return 1; | |
2d4a7167 | 917 | return 0; |
92181f19 NP |
918 | } |
919 | ||
2d4a7167 IM |
920 | /* read, present: */ |
921 | if (unlikely(error_code & PF_PROT)) | |
922 | return 1; | |
923 | ||
924 | /* read, not present: */ | |
925 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
926 | return 1; | |
927 | ||
92181f19 NP |
928 | return 0; |
929 | } | |
930 | ||
0973a06c HS |
931 | static int fault_in_kernel_space(unsigned long address) |
932 | { | |
d9517346 | 933 | return address >= TASK_SIZE_MAX; |
0973a06c HS |
934 | } |
935 | ||
1da177e4 LT |
936 | /* |
937 | * This routine handles page faults. It determines the address, | |
938 | * and the problem, and then passes it off to one of the appropriate | |
939 | * routines. | |
1da177e4 | 940 | */ |
c3731c68 IM |
941 | dotraplinkage void __kprobes |
942 | do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 | 943 | { |
2d4a7167 | 944 | struct vm_area_struct *vma; |
1da177e4 | 945 | struct task_struct *tsk; |
2d4a7167 | 946 | unsigned long address; |
1da177e4 | 947 | struct mm_struct *mm; |
92181f19 | 948 | int write; |
f8c2ee22 | 949 | int fault; |
1da177e4 | 950 | |
a9ba9a3b AV |
951 | tsk = current; |
952 | mm = tsk->mm; | |
2d4a7167 | 953 | |
2d4a7167 | 954 | /* Get the faulting address: */ |
f51c9452 | 955 | address = read_cr2(); |
1da177e4 | 956 | |
f8561296 VN |
957 | /* |
958 | * Detect and handle instructions that would cause a page fault for | |
959 | * both a tracked kernel page and a userspace page. | |
960 | */ | |
961 | if (kmemcheck_active(regs)) | |
962 | kmemcheck_hide(regs); | |
5dfaf90f | 963 | prefetchw(&mm->mmap_sem); |
f8561296 | 964 | |
0fd0e3da | 965 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 966 | return; |
1da177e4 LT |
967 | |
968 | /* | |
969 | * We fault-in kernel-space virtual memory on-demand. The | |
970 | * 'reference' page table is init_mm.pgd. | |
971 | * | |
972 | * NOTE! We MUST NOT take any locks for this case. We may | |
973 | * be in an interrupt or a critical region, and should | |
974 | * only copy the information from the master page table, | |
975 | * nothing more. | |
976 | * | |
977 | * This verifies that the fault happens in kernel space | |
978 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 979 | * protection error (error_code & 9) == 0. |
1da177e4 | 980 | */ |
0973a06c | 981 | if (unlikely(fault_in_kernel_space(address))) { |
f8561296 VN |
982 | if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { |
983 | if (vmalloc_fault(address) >= 0) | |
984 | return; | |
985 | ||
986 | if (kmemcheck_fault(regs, address, error_code)) | |
987 | return; | |
988 | } | |
5b727a3b | 989 | |
2d4a7167 | 990 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 991 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
992 | return; |
993 | ||
2d4a7167 | 994 | /* kprobes don't want to hook the spurious faults: */ |
9be260a6 MH |
995 | if (notify_page_fault(regs)) |
996 | return; | |
f8c2ee22 HH |
997 | /* |
998 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 999 | * fault we could otherwise deadlock: |
f8c2ee22 | 1000 | */ |
92181f19 | 1001 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 1002 | |
92181f19 | 1003 | return; |
f8c2ee22 HH |
1004 | } |
1005 | ||
2d4a7167 | 1006 | /* kprobes don't want to hook the spurious faults: */ |
f8a6b2b9 | 1007 | if (unlikely(notify_page_fault(regs))) |
9be260a6 | 1008 | return; |
f8c2ee22 | 1009 | /* |
891cffbd LT |
1010 | * It's safe to allow irq's after cr2 has been saved and the |
1011 | * vmalloc fault has been handled. | |
1012 | * | |
1013 | * User-mode registers count as a user access even for any | |
2d4a7167 | 1014 | * potential system fault or CPU buglet: |
f8c2ee22 | 1015 | */ |
891cffbd LT |
1016 | if (user_mode_vm(regs)) { |
1017 | local_irq_enable(); | |
1018 | error_code |= PF_USER; | |
2d4a7167 IM |
1019 | } else { |
1020 | if (regs->flags & X86_EFLAGS_IF) | |
1021 | local_irq_enable(); | |
1022 | } | |
8c914cb7 | 1023 | |
66c58156 | 1024 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1025 | pgtable_bad(regs, error_code, address); |
1da177e4 | 1026 | |
cdd6c482 | 1027 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); |
7dd1fcc2 | 1028 | |
1da177e4 | 1029 | /* |
2d4a7167 IM |
1030 | * If we're in an interrupt, have no user context or are running |
1031 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1032 | */ |
92181f19 NP |
1033 | if (unlikely(in_atomic() || !mm)) { |
1034 | bad_area_nosemaphore(regs, error_code, address); | |
1035 | return; | |
1036 | } | |
1da177e4 | 1037 | |
3a1dfe6e IM |
1038 | /* |
1039 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1040 | * addresses in user space. All other faults represent errors in |
1041 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1042 | * case of an erroneous fault occurring in a code path which already | |
1043 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1044 | * against the address space. Luckily the kernel only validly | |
1045 | * references user space from well defined areas of code, which are | |
1046 | * listed in the exceptions table. | |
1da177e4 LT |
1047 | * |
1048 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1049 | * the source reference check when there is a possibility of a |
1050 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1051 | * validate the source. If this is invalid we can skip the address | |
1052 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1053 | */ |
92181f19 | 1054 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1055 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1056 | !search_exception_tables(regs->ip)) { |
1057 | bad_area_nosemaphore(regs, error_code, address); | |
1058 | return; | |
1059 | } | |
1da177e4 | 1060 | down_read(&mm->mmap_sem); |
01006074 PZ |
1061 | } else { |
1062 | /* | |
2d4a7167 IM |
1063 | * The above down_read_trylock() might have succeeded in |
1064 | * which case we'll have missed the might_sleep() from | |
1065 | * down_read(): | |
01006074 PZ |
1066 | */ |
1067 | might_sleep(); | |
1da177e4 LT |
1068 | } |
1069 | ||
1070 | vma = find_vma(mm, address); | |
92181f19 NP |
1071 | if (unlikely(!vma)) { |
1072 | bad_area(regs, error_code, address); | |
1073 | return; | |
1074 | } | |
1075 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1076 | goto good_area; |
92181f19 NP |
1077 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1078 | bad_area(regs, error_code, address); | |
1079 | return; | |
1080 | } | |
33cb5243 | 1081 | if (error_code & PF_USER) { |
6f4d368e HH |
1082 | /* |
1083 | * Accessing the stack below %sp is always a bug. | |
1084 | * The large cushion allows instructions like enter | |
2d4a7167 | 1085 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1086 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1087 | */ |
92181f19 NP |
1088 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1089 | bad_area(regs, error_code, address); | |
1090 | return; | |
1091 | } | |
1da177e4 | 1092 | } |
92181f19 NP |
1093 | if (unlikely(expand_stack(vma, address))) { |
1094 | bad_area(regs, error_code, address); | |
1095 | return; | |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * Ok, we have a good vm_area for this memory access, so | |
1100 | * we can handle it.. | |
1101 | */ | |
1da177e4 | 1102 | good_area: |
92181f19 | 1103 | write = error_code & PF_WRITE; |
2d4a7167 | 1104 | |
92181f19 NP |
1105 | if (unlikely(access_error(error_code, write, vma))) { |
1106 | bad_area_access_error(regs, error_code, address); | |
1107 | return; | |
1da177e4 LT |
1108 | } |
1109 | ||
1110 | /* | |
1111 | * If for any reason at all we couldn't handle the fault, | |
1112 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1113 | * the fault: |
1da177e4 | 1114 | */ |
d06063cc | 1115 | fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0); |
2d4a7167 | 1116 | |
83c54070 | 1117 | if (unlikely(fault & VM_FAULT_ERROR)) { |
92181f19 NP |
1118 | mm_fault_error(regs, error_code, address, fault); |
1119 | return; | |
1da177e4 | 1120 | } |
2d4a7167 | 1121 | |
ac17dc8e | 1122 | if (fault & VM_FAULT_MAJOR) { |
83c54070 | 1123 | tsk->maj_flt++; |
cdd6c482 | 1124 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, |
78f13e95 | 1125 | regs, address); |
ac17dc8e | 1126 | } else { |
83c54070 | 1127 | tsk->min_flt++; |
cdd6c482 | 1128 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, |
78f13e95 | 1129 | regs, address); |
ac17dc8e | 1130 | } |
d729ab35 | 1131 | |
8c938f9f IM |
1132 | check_v8086_mode(regs, address, tsk); |
1133 | ||
1da177e4 | 1134 | up_read(&mm->mmap_sem); |
1da177e4 | 1135 | } |