Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * Copyright (C) 1995 Linus Torvalds |
2d4a7167 | 3 | * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. |
1da177e4 | 4 | */ |
1da177e4 | 5 | #include <linux/interrupt.h> |
2d4a7167 IM |
6 | #include <linux/mmiotrace.h> |
7 | #include <linux/bootmem.h> | |
1da177e4 | 8 | #include <linux/compiler.h> |
c61e211d | 9 | #include <linux/highmem.h> |
0f2fbdcb | 10 | #include <linux/kprobes.h> |
ab2bf0c1 | 11 | #include <linux/uaccess.h> |
2d4a7167 IM |
12 | #include <linux/vmalloc.h> |
13 | #include <linux/vt_kern.h> | |
14 | #include <linux/signal.h> | |
15 | #include <linux/kernel.h> | |
16 | #include <linux/ptrace.h> | |
17 | #include <linux/string.h> | |
18 | #include <linux/module.h> | |
1eeb66a1 | 19 | #include <linux/kdebug.h> |
2d4a7167 | 20 | #include <linux/errno.h> |
7c9f8861 | 21 | #include <linux/magic.h> |
2d4a7167 IM |
22 | #include <linux/sched.h> |
23 | #include <linux/types.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/mman.h> | |
26 | #include <linux/tty.h> | |
27 | #include <linux/smp.h> | |
28 | #include <linux/mm.h> | |
29 | ||
30 | #include <asm-generic/sections.h> | |
1da177e4 | 31 | |
1da177e4 | 32 | #include <asm/tlbflush.h> |
2d4a7167 IM |
33 | #include <asm/pgalloc.h> |
34 | #include <asm/segment.h> | |
35 | #include <asm/system.h> | |
1da177e4 | 36 | #include <asm/proto.h> |
70ef5641 | 37 | #include <asm/traps.h> |
2d4a7167 | 38 | #include <asm/desc.h> |
1da177e4 | 39 | |
33cb5243 | 40 | /* |
2d4a7167 IM |
41 | * Page fault error code bits: |
42 | * | |
43 | * bit 0 == 0: no page found 1: protection fault | |
44 | * bit 1 == 0: read access 1: write access | |
45 | * bit 2 == 0: kernel-mode access 1: user-mode access | |
46 | * bit 3 == 1: use of reserved bit detected | |
47 | * bit 4 == 1: fault was an instruction fetch | |
33cb5243 | 48 | */ |
2d4a7167 IM |
49 | enum x86_pf_error_code { |
50 | ||
51 | PF_PROT = 1 << 0, | |
52 | PF_WRITE = 1 << 1, | |
53 | PF_USER = 1 << 2, | |
54 | PF_RSVD = 1 << 3, | |
55 | PF_INSTR = 1 << 4, | |
56 | }; | |
66c58156 | 57 | |
b814d41f IM |
58 | /* |
59 | * (returns 0 if mmiotrace is disabled) | |
60 | */ | |
0fd0e3da | 61 | static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 62 | { |
0fd0e3da PP |
63 | if (unlikely(is_kmmio_active())) |
64 | if (kmmio_handler(regs, addr) == 1) | |
65 | return -1; | |
0fd0e3da | 66 | return 0; |
86069782 PP |
67 | } |
68 | ||
74a0b576 | 69 | static inline int notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 70 | { |
74a0b576 CH |
71 | int ret = 0; |
72 | ||
73 | /* kprobe_running() needs smp_processor_id() */ | |
b1801812 | 74 | if (kprobes_built_in() && !user_mode_vm(regs)) { |
74a0b576 CH |
75 | preempt_disable(); |
76 | if (kprobe_running() && kprobe_fault_handler(regs, 14)) | |
77 | ret = 1; | |
78 | preempt_enable(); | |
79 | } | |
1bd858a5 | 80 | |
74a0b576 | 81 | return ret; |
33cb5243 | 82 | } |
1bd858a5 | 83 | |
1dc85be0 | 84 | /* |
2d4a7167 IM |
85 | * Prefetch quirks: |
86 | * | |
87 | * 32-bit mode: | |
88 | * | |
89 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
90 | * Check that here and ignore it. | |
1dc85be0 | 91 | * |
2d4a7167 | 92 | * 64-bit mode: |
1dc85be0 | 93 | * |
2d4a7167 IM |
94 | * Sometimes the CPU reports invalid exceptions on prefetch. |
95 | * Check that here and ignore it. | |
96 | * | |
97 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 98 | */ |
107a0367 IM |
99 | static inline int |
100 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
101 | unsigned char opcode, int *prefetch) | |
102 | { | |
103 | unsigned char instr_hi = opcode & 0xf0; | |
104 | unsigned char instr_lo = opcode & 0x0f; | |
105 | ||
106 | switch (instr_hi) { | |
107 | case 0x20: | |
108 | case 0x30: | |
109 | /* | |
110 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
111 | * In X86_64 long mode, the CPU will signal invalid | |
112 | * opcode if some of these prefixes are present so | |
113 | * X86_64 will never get here anyway | |
114 | */ | |
115 | return ((instr_lo & 7) == 0x6); | |
116 | #ifdef CONFIG_X86_64 | |
117 | case 0x40: | |
118 | /* | |
119 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
120 | * Need to figure out under what instruction mode the | |
121 | * instruction was issued. Could check the LDT for lm, | |
122 | * but for now it's good enough to assume that long | |
123 | * mode only uses well known segments or kernel. | |
124 | */ | |
125 | return (!user_mode(regs)) || (regs->cs == __USER_CS); | |
126 | #endif | |
127 | case 0x60: | |
128 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
129 | return (instr_lo & 0xC) == 0x4; | |
130 | case 0xF0: | |
131 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
132 | return !instr_lo || (instr_lo>>1) == 1; | |
133 | case 0x00: | |
134 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
135 | if (probe_kernel_address(instr, opcode)) | |
136 | return 0; | |
137 | ||
138 | *prefetch = (instr_lo == 0xF) && | |
139 | (opcode == 0x0D || opcode == 0x18); | |
140 | return 0; | |
141 | default: | |
142 | return 0; | |
143 | } | |
144 | } | |
145 | ||
2d4a7167 IM |
146 | static int |
147 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 148 | { |
2d4a7167 | 149 | unsigned char *max_instr; |
ab2bf0c1 | 150 | unsigned char *instr; |
33cb5243 | 151 | int prefetch = 0; |
1da177e4 | 152 | |
3085354d IM |
153 | /* |
154 | * If it was a exec (instruction fetch) fault on NX page, then | |
155 | * do not ignore the fault: | |
156 | */ | |
66c58156 | 157 | if (error_code & PF_INSTR) |
1da177e4 | 158 | return 0; |
1dc85be0 | 159 | |
107a0367 | 160 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 161 | max_instr = instr + 15; |
1da177e4 | 162 | |
76381fee | 163 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
164 | return 0; |
165 | ||
107a0367 | 166 | while (instr < max_instr) { |
2d4a7167 | 167 | unsigned char opcode; |
1da177e4 | 168 | |
ab2bf0c1 | 169 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 170 | break; |
1da177e4 | 171 | |
1da177e4 LT |
172 | instr++; |
173 | ||
107a0367 | 174 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 175 | break; |
1da177e4 LT |
176 | } |
177 | return prefetch; | |
178 | } | |
179 | ||
2d4a7167 IM |
180 | static void |
181 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
182 | struct task_struct *tsk) | |
c4aba4a8 HH |
183 | { |
184 | siginfo_t info; | |
185 | ||
2d4a7167 IM |
186 | info.si_signo = si_signo; |
187 | info.si_errno = 0; | |
188 | info.si_code = si_code; | |
189 | info.si_addr = (void __user *)address; | |
190 | ||
c4aba4a8 HH |
191 | force_sig_info(si_signo, &info, tsk); |
192 | } | |
193 | ||
1156e098 | 194 | #ifdef CONFIG_X86_64 |
33cb5243 HH |
195 | static int bad_address(void *p) |
196 | { | |
1da177e4 | 197 | unsigned long dummy; |
2d4a7167 | 198 | |
ab2bf0c1 | 199 | return probe_kernel_address((unsigned long *)p, dummy); |
33cb5243 | 200 | } |
1156e098 | 201 | #endif |
1da177e4 | 202 | |
cae30f82 | 203 | static void dump_pagetable(unsigned long address) |
1da177e4 | 204 | { |
1156e098 HH |
205 | #ifdef CONFIG_X86_32 |
206 | __typeof__(pte_val(__pte(0))) page; | |
207 | ||
208 | page = read_cr3(); | |
209 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | |
2d4a7167 | 210 | |
1156e098 HH |
211 | #ifdef CONFIG_X86_PAE |
212 | printk("*pdpt = %016Lx ", page); | |
213 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
214 | && page & _PAGE_PRESENT) { | |
215 | page &= PAGE_MASK; | |
216 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | |
2d4a7167 | 217 | & (PTRS_PER_PMD - 1)]; |
1156e098 HH |
218 | printk(KERN_CONT "*pde = %016Lx ", page); |
219 | page &= ~_PAGE_NX; | |
220 | } | |
221 | #else | |
222 | printk("*pde = %08lx ", page); | |
223 | #endif | |
224 | ||
225 | /* | |
226 | * We must not directly access the pte in the highpte | |
227 | * case if the page table is located in highmem. | |
228 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 229 | * it's allocated already: |
1156e098 HH |
230 | */ |
231 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
232 | && (page & _PAGE_PRESENT) | |
233 | && !(page & _PAGE_PSE)) { | |
2d4a7167 | 234 | |
1156e098 HH |
235 | page &= PAGE_MASK; |
236 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | |
2d4a7167 | 237 | & (PTRS_PER_PTE - 1)]; |
1156e098 HH |
238 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); |
239 | } | |
240 | ||
241 | printk("\n"); | |
242 | #else /* CONFIG_X86_64 */ | |
1da177e4 LT |
243 | pgd_t *pgd; |
244 | pud_t *pud; | |
245 | pmd_t *pmd; | |
246 | pte_t *pte; | |
247 | ||
f51c9452 | 248 | pgd = (pgd_t *)read_cr3(); |
1da177e4 | 249 | |
33cb5243 | 250 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); |
2d4a7167 | 251 | |
1da177e4 | 252 | pgd += pgd_index(address); |
2d4a7167 IM |
253 | if (bad_address(pgd)) |
254 | goto bad; | |
255 | ||
d646bce4 | 256 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
257 | |
258 | if (!pgd_present(*pgd)) | |
259 | goto out; | |
1da177e4 | 260 | |
d2ae5b5f | 261 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
262 | if (bad_address(pud)) |
263 | goto bad; | |
264 | ||
1da177e4 | 265 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 266 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 267 | goto out; |
1da177e4 LT |
268 | |
269 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
270 | if (bad_address(pmd)) |
271 | goto bad; | |
272 | ||
1da177e4 | 273 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
274 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
275 | goto out; | |
1da177e4 LT |
276 | |
277 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
278 | if (bad_address(pte)) |
279 | goto bad; | |
280 | ||
33cb5243 | 281 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 282 | out: |
1da177e4 LT |
283 | printk("\n"); |
284 | return; | |
285 | bad: | |
286 | printk("BAD\n"); | |
1156e098 HH |
287 | #endif |
288 | } | |
289 | ||
290 | #ifdef CONFIG_X86_32 | |
291 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
292 | { | |
293 | unsigned index = pgd_index(address); | |
294 | pgd_t *pgd_k; | |
295 | pud_t *pud, *pud_k; | |
296 | pmd_t *pmd, *pmd_k; | |
297 | ||
298 | pgd += index; | |
299 | pgd_k = init_mm.pgd + index; | |
300 | ||
301 | if (!pgd_present(*pgd_k)) | |
302 | return NULL; | |
303 | ||
304 | /* | |
305 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
306 | * and redundant with the set_pmd() on non-PAE. As would | |
307 | * set_pud. | |
308 | */ | |
1156e098 HH |
309 | pud = pud_offset(pgd, address); |
310 | pud_k = pud_offset(pgd_k, address); | |
311 | if (!pud_present(*pud_k)) | |
312 | return NULL; | |
313 | ||
314 | pmd = pmd_offset(pud, address); | |
315 | pmd_k = pmd_offset(pud_k, address); | |
316 | if (!pmd_present(*pmd_k)) | |
317 | return NULL; | |
2d4a7167 | 318 | |
1156e098 HH |
319 | if (!pmd_present(*pmd)) { |
320 | set_pmd(pmd, *pmd_k); | |
321 | arch_flush_lazy_mmu_mode(); | |
2d4a7167 | 322 | } else { |
1156e098 | 323 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
2d4a7167 IM |
324 | } |
325 | ||
1156e098 | 326 | return pmd_k; |
1da177e4 LT |
327 | } |
328 | ||
8c938f9f IM |
329 | /* |
330 | * Did it hit the DOS screen memory VA from vm86 mode? | |
331 | */ | |
332 | static inline void | |
333 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
334 | struct task_struct *tsk) | |
335 | { | |
336 | unsigned long bit; | |
337 | ||
338 | if (!v8086_mode(regs)) | |
339 | return; | |
340 | ||
341 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
342 | if (bit < 32) | |
343 | tsk->thread.screen_bitmap |= 1 << bit; | |
344 | } | |
345 | ||
346 | #else /* CONFIG_X86_64: */ | |
347 | ||
33cb5243 | 348 | static const char errata93_warning[] = |
1da177e4 LT |
349 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" |
350 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
351 | KERN_ERR "******* Please consider a BIOS update.\n" | |
352 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
8c938f9f IM |
353 | |
354 | /* | |
355 | * No vm86 mode in 64-bit mode: | |
356 | */ | |
357 | static inline void | |
358 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
359 | struct task_struct *tsk) | |
360 | { | |
361 | } | |
362 | ||
fdfe8aa8 | 363 | #endif |
1da177e4 | 364 | |
2d4a7167 IM |
365 | /* |
366 | * Workaround for K8 erratum #93 & buggy BIOS. | |
367 | * | |
368 | * BIOS SMM functions are required to use a specific workaround | |
369 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
370 | * | |
371 | * A lot of BIOS that didn't get tested properly miss this. | |
372 | * | |
373 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
374 | * Try to work around it here. | |
375 | * | |
376 | * Note we only handle faults in kernel here. | |
377 | * Does nothing on 32-bit. | |
fdfe8aa8 | 378 | */ |
33cb5243 | 379 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 380 | { |
fdfe8aa8 | 381 | #ifdef CONFIG_X86_64 |
2d4a7167 IM |
382 | static int once; |
383 | ||
65ea5b03 | 384 | if (address != regs->ip) |
1da177e4 | 385 | return 0; |
2d4a7167 | 386 | |
33cb5243 | 387 | if ((address >> 32) != 0) |
1da177e4 | 388 | return 0; |
2d4a7167 | 389 | |
1da177e4 | 390 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
391 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
392 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
2d4a7167 | 393 | if (!once) { |
33cb5243 | 394 | printk(errata93_warning); |
2d4a7167 | 395 | once = 1; |
1da177e4 | 396 | } |
65ea5b03 | 397 | regs->ip = address; |
1da177e4 LT |
398 | return 1; |
399 | } | |
fdfe8aa8 | 400 | #endif |
1da177e4 | 401 | return 0; |
33cb5243 | 402 | } |
1da177e4 | 403 | |
35f3266f | 404 | /* |
2d4a7167 IM |
405 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
406 | * to illegal addresses >4GB. | |
407 | * | |
408 | * We catch this in the page fault handler because these addresses | |
409 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
410 | * segment in LDT is compatibility mode. |
411 | */ | |
412 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
413 | { | |
414 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 415 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
416 | return 1; |
417 | #endif | |
418 | return 0; | |
419 | } | |
420 | ||
29caf2f9 HH |
421 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
422 | { | |
423 | #ifdef CONFIG_X86_F00F_BUG | |
424 | unsigned long nr; | |
2d4a7167 | 425 | |
29caf2f9 | 426 | /* |
2d4a7167 | 427 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 HH |
428 | */ |
429 | if (boot_cpu_data.f00f_bug) { | |
430 | nr = (address - idt_descr.address) >> 3; | |
431 | ||
432 | if (nr == 6) { | |
433 | do_invalid_op(regs, 0); | |
434 | return 1; | |
435 | } | |
436 | } | |
437 | #endif | |
438 | return 0; | |
439 | } | |
440 | ||
2d4a7167 IM |
441 | static void |
442 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
443 | unsigned long address) | |
b3279c7f | 444 | { |
1156e098 HH |
445 | #ifdef CONFIG_X86_32 |
446 | if (!oops_may_print()) | |
447 | return; | |
fd40d6e3 | 448 | #endif |
1156e098 HH |
449 | |
450 | #ifdef CONFIG_X86_PAE | |
451 | if (error_code & PF_INSTR) { | |
93809be8 | 452 | unsigned int level; |
2d4a7167 | 453 | |
1156e098 HH |
454 | pte_t *pte = lookup_address(address, &level); |
455 | ||
2d4a7167 | 456 | if (pte && pte_present(*pte) && !pte_exec(*pte)) { |
1156e098 HH |
457 | printk(KERN_CRIT "kernel tried to execute " |
458 | "NX-protected page - exploit attempt? " | |
350b4da7 | 459 | "(uid: %d)\n", current_uid()); |
2d4a7167 | 460 | } |
1156e098 HH |
461 | } |
462 | #endif | |
1156e098 | 463 | |
19f0dda9 | 464 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 465 | if (address < PAGE_SIZE) |
19f0dda9 | 466 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 467 | else |
19f0dda9 | 468 | printk(KERN_CONT "paging request"); |
2d4a7167 | 469 | |
f294a8ce | 470 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 471 | printk(KERN_ALERT "IP:"); |
b3279c7f | 472 | printk_address(regs->ip, 1); |
2d4a7167 | 473 | |
b3279c7f HH |
474 | dump_pagetable(address); |
475 | } | |
476 | ||
2d4a7167 IM |
477 | static noinline void |
478 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
479 | unsigned long address) | |
1da177e4 | 480 | { |
2d4a7167 IM |
481 | struct task_struct *tsk; |
482 | unsigned long flags; | |
483 | int sig; | |
484 | ||
485 | flags = oops_begin(); | |
486 | tsk = current; | |
487 | sig = SIGKILL; | |
1209140c | 488 | |
1da177e4 | 489 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 490 | tsk->comm, address); |
1da177e4 | 491 | dump_pagetable(address); |
2d4a7167 IM |
492 | |
493 | tsk->thread.cr2 = address; | |
494 | tsk->thread.trap_no = 14; | |
495 | tsk->thread.error_code = error_code; | |
496 | ||
22f5991c | 497 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 498 | sig = 0; |
2d4a7167 | 499 | |
874d93d1 | 500 | oops_end(flags, regs, sig); |
1da177e4 LT |
501 | } |
502 | ||
2d4a7167 IM |
503 | static noinline void |
504 | no_context(struct pt_regs *regs, unsigned long error_code, | |
505 | unsigned long address) | |
92181f19 NP |
506 | { |
507 | struct task_struct *tsk = current; | |
19803078 IM |
508 | unsigned long *stackend; |
509 | ||
92181f19 NP |
510 | #ifdef CONFIG_X86_64 |
511 | unsigned long flags; | |
512 | int sig; | |
513 | #endif | |
514 | ||
2d4a7167 | 515 | /* Are we prepared to handle this kernel fault? */ |
92181f19 NP |
516 | if (fixup_exception(regs)) |
517 | return; | |
518 | ||
519 | /* | |
2d4a7167 IM |
520 | * 32-bit: |
521 | * | |
522 | * Valid to do another page fault here, because if this fault | |
523 | * had been triggered by is_prefetch fixup_exception would have | |
524 | * handled it. | |
525 | * | |
526 | * 64-bit: | |
92181f19 | 527 | * |
2d4a7167 | 528 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
529 | */ |
530 | if (is_prefetch(regs, error_code, address)) | |
531 | return; | |
532 | ||
533 | if (is_errata93(regs, address)) | |
534 | return; | |
535 | ||
536 | /* | |
537 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 538 | * terminate things with extreme prejudice: |
92181f19 NP |
539 | */ |
540 | #ifdef CONFIG_X86_32 | |
541 | bust_spinlocks(1); | |
542 | #else | |
543 | flags = oops_begin(); | |
544 | #endif | |
545 | ||
546 | show_fault_oops(regs, error_code, address); | |
547 | ||
2d4a7167 | 548 | stackend = end_of_stack(tsk); |
19803078 IM |
549 | if (*stackend != STACK_END_MAGIC) |
550 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); | |
551 | ||
92181f19 NP |
552 | tsk->thread.cr2 = address; |
553 | tsk->thread.trap_no = 14; | |
554 | tsk->thread.error_code = error_code; | |
555 | ||
556 | #ifdef CONFIG_X86_32 | |
557 | die("Oops", regs, error_code); | |
558 | bust_spinlocks(0); | |
559 | do_exit(SIGKILL); | |
560 | #else | |
561 | sig = SIGKILL; | |
562 | if (__die("Oops", regs, error_code)) | |
563 | sig = 0; | |
2d4a7167 | 564 | |
92181f19 NP |
565 | /* Executive summary in case the body of the oops scrolled away */ |
566 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
2d4a7167 | 567 | |
92181f19 NP |
568 | oops_end(flags, regs, sig); |
569 | #endif | |
570 | } | |
571 | ||
2d4a7167 IM |
572 | /* |
573 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
574 | * sysctl is set: | |
575 | */ | |
576 | static inline void | |
577 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
578 | unsigned long address, struct task_struct *tsk) | |
579 | { | |
580 | if (!unhandled_signal(tsk, SIGSEGV)) | |
581 | return; | |
582 | ||
583 | if (!printk_ratelimit()) | |
584 | return; | |
585 | ||
586 | printk(KERN_CONT "%s%s[%d]: segfault at %lx ip %p sp %p error %lx", | |
587 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
588 | tsk->comm, task_pid_nr(tsk), address, | |
589 | (void *)regs->ip, (void *)regs->sp, error_code); | |
590 | ||
591 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
592 | ||
593 | printk(KERN_CONT "\n"); | |
594 | } | |
595 | ||
596 | static void | |
597 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
598 | unsigned long address, int si_code) | |
92181f19 NP |
599 | { |
600 | struct task_struct *tsk = current; | |
601 | ||
602 | /* User mode accesses just cause a SIGSEGV */ | |
603 | if (error_code & PF_USER) { | |
604 | /* | |
2d4a7167 | 605 | * It's possible to have interrupts off here: |
92181f19 NP |
606 | */ |
607 | local_irq_enable(); | |
608 | ||
609 | /* | |
610 | * Valid to do another page fault here because this one came | |
2d4a7167 | 611 | * from user space: |
92181f19 NP |
612 | */ |
613 | if (is_prefetch(regs, error_code, address)) | |
614 | return; | |
615 | ||
616 | if (is_errata100(regs, address)) | |
617 | return; | |
618 | ||
2d4a7167 IM |
619 | if (unlikely(show_unhandled_signals)) |
620 | show_signal_msg(regs, error_code, address, tsk); | |
621 | ||
622 | /* Kernel addresses are always protection faults: */ | |
623 | tsk->thread.cr2 = address; | |
624 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
625 | tsk->thread.trap_no = 14; | |
92181f19 | 626 | |
92181f19 | 627 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
2d4a7167 | 628 | |
92181f19 NP |
629 | return; |
630 | } | |
631 | ||
632 | if (is_f00f_bug(regs, address)) | |
633 | return; | |
634 | ||
635 | no_context(regs, error_code, address); | |
636 | } | |
637 | ||
2d4a7167 IM |
638 | static noinline void |
639 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
640 | unsigned long address) | |
92181f19 NP |
641 | { |
642 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
643 | } | |
644 | ||
2d4a7167 IM |
645 | static void |
646 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
647 | unsigned long address, int si_code) | |
92181f19 NP |
648 | { |
649 | struct mm_struct *mm = current->mm; | |
650 | ||
651 | /* | |
652 | * Something tried to access memory that isn't in our memory map.. | |
653 | * Fix it, but check if it's kernel or user first.. | |
654 | */ | |
655 | up_read(&mm->mmap_sem); | |
656 | ||
657 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
658 | } | |
659 | ||
2d4a7167 IM |
660 | static noinline void |
661 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
662 | { |
663 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
664 | } | |
665 | ||
2d4a7167 IM |
666 | static noinline void |
667 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
668 | unsigned long address) | |
92181f19 NP |
669 | { |
670 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
671 | } | |
672 | ||
673 | /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */ | |
2d4a7167 IM |
674 | static void |
675 | out_of_memory(struct pt_regs *regs, unsigned long error_code, | |
676 | unsigned long address) | |
92181f19 NP |
677 | { |
678 | /* | |
679 | * We ran out of memory, call the OOM killer, and return the userspace | |
2d4a7167 | 680 | * (which will retry the fault, or kill us if we got oom-killed): |
92181f19 NP |
681 | */ |
682 | up_read(¤t->mm->mmap_sem); | |
2d4a7167 | 683 | |
92181f19 NP |
684 | pagefault_out_of_memory(); |
685 | } | |
686 | ||
2d4a7167 IM |
687 | static void |
688 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
689 | { |
690 | struct task_struct *tsk = current; | |
691 | struct mm_struct *mm = tsk->mm; | |
692 | ||
693 | up_read(&mm->mmap_sem); | |
694 | ||
2d4a7167 | 695 | /* Kernel mode? Handle exceptions or die: */ |
92181f19 NP |
696 | if (!(error_code & PF_USER)) |
697 | no_context(regs, error_code, address); | |
2d4a7167 | 698 | |
92181f19 | 699 | #ifdef CONFIG_X86_32 |
2d4a7167 | 700 | /* User space => ok to do another page fault: */ |
92181f19 NP |
701 | if (is_prefetch(regs, error_code, address)) |
702 | return; | |
703 | #endif | |
2d4a7167 IM |
704 | |
705 | tsk->thread.cr2 = address; | |
706 | tsk->thread.error_code = error_code; | |
707 | tsk->thread.trap_no = 14; | |
708 | ||
92181f19 NP |
709 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
710 | } | |
711 | ||
2d4a7167 IM |
712 | static noinline void |
713 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, | |
714 | unsigned long address, unsigned int fault) | |
92181f19 | 715 | { |
2d4a7167 | 716 | if (fault & VM_FAULT_OOM) { |
92181f19 | 717 | out_of_memory(regs, error_code, address); |
2d4a7167 IM |
718 | } else { |
719 | if (fault & VM_FAULT_SIGBUS) | |
720 | do_sigbus(regs, error_code, address); | |
721 | else | |
722 | BUG(); | |
723 | } | |
92181f19 NP |
724 | } |
725 | ||
d8b57bb7 TG |
726 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
727 | { | |
728 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
729 | return 0; | |
2d4a7167 | 730 | |
d8b57bb7 TG |
731 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
732 | return 0; | |
733 | ||
734 | return 1; | |
735 | } | |
736 | ||
5b727a3b | 737 | /* |
2d4a7167 IM |
738 | * Handle a spurious fault caused by a stale TLB entry. |
739 | * | |
740 | * This allows us to lazily refresh the TLB when increasing the | |
741 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
742 | * eagerly is very expensive since that implies doing a full | |
743 | * cross-processor TLB flush, even if no stale TLB entries exist | |
744 | * on other processors. | |
745 | * | |
5b727a3b JF |
746 | * There are no security implications to leaving a stale TLB when |
747 | * increasing the permissions on a page. | |
748 | */ | |
2d4a7167 IM |
749 | static noinline int |
750 | spurious_fault(unsigned long error_code, unsigned long address) | |
5b727a3b JF |
751 | { |
752 | pgd_t *pgd; | |
753 | pud_t *pud; | |
754 | pmd_t *pmd; | |
755 | pte_t *pte; | |
3c3e5694 | 756 | int ret; |
5b727a3b JF |
757 | |
758 | /* Reserved-bit violation or user access to kernel space? */ | |
759 | if (error_code & (PF_USER | PF_RSVD)) | |
760 | return 0; | |
761 | ||
762 | pgd = init_mm.pgd + pgd_index(address); | |
763 | if (!pgd_present(*pgd)) | |
764 | return 0; | |
765 | ||
766 | pud = pud_offset(pgd, address); | |
767 | if (!pud_present(*pud)) | |
768 | return 0; | |
769 | ||
d8b57bb7 TG |
770 | if (pud_large(*pud)) |
771 | return spurious_fault_check(error_code, (pte_t *) pud); | |
772 | ||
5b727a3b JF |
773 | pmd = pmd_offset(pud, address); |
774 | if (!pmd_present(*pmd)) | |
775 | return 0; | |
776 | ||
d8b57bb7 TG |
777 | if (pmd_large(*pmd)) |
778 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
779 | ||
5b727a3b JF |
780 | pte = pte_offset_kernel(pmd, address); |
781 | if (!pte_present(*pte)) | |
782 | return 0; | |
783 | ||
3c3e5694 SR |
784 | ret = spurious_fault_check(error_code, pte); |
785 | if (!ret) | |
786 | return 0; | |
787 | ||
788 | /* | |
2d4a7167 IM |
789 | * Make sure we have permissions in PMD. |
790 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
791 | */ |
792 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
793 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 794 | |
3c3e5694 | 795 | return ret; |
5b727a3b JF |
796 | } |
797 | ||
1da177e4 | 798 | /* |
2d4a7167 IM |
799 | * 32-bit: |
800 | * | |
801 | * Handle a fault on the vmalloc or module mapping area | |
f8c2ee22 | 802 | * |
2d4a7167 IM |
803 | * 64-bit: |
804 | * | |
805 | * Handle a fault on the vmalloc area | |
3b9ba4d5 AK |
806 | * |
807 | * This assumes no large pages in there. | |
1da177e4 | 808 | */ |
92181f19 | 809 | static noinline int vmalloc_fault(unsigned long address) |
1da177e4 | 810 | { |
fdfe8aa8 HH |
811 | #ifdef CONFIG_X86_32 |
812 | unsigned long pgd_paddr; | |
813 | pmd_t *pmd_k; | |
814 | pte_t *pte_k; | |
b29c701d | 815 | |
2d4a7167 | 816 | /* Make sure we are in vmalloc area: */ |
b29c701d HN |
817 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) |
818 | return -1; | |
819 | ||
fdfe8aa8 HH |
820 | /* |
821 | * Synchronize this task's top level page-table | |
822 | * with the 'reference' page table. | |
823 | * | |
824 | * Do _not_ use "current" here. We might be inside | |
825 | * an interrupt in the middle of a task switch.. | |
826 | */ | |
827 | pgd_paddr = read_cr3(); | |
828 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
829 | if (!pmd_k) | |
830 | return -1; | |
2d4a7167 | 831 | |
fdfe8aa8 HH |
832 | pte_k = pte_offset_kernel(pmd_k, address); |
833 | if (!pte_present(*pte_k)) | |
834 | return -1; | |
2d4a7167 | 835 | |
fdfe8aa8 HH |
836 | return 0; |
837 | #else | |
1da177e4 LT |
838 | pgd_t *pgd, *pgd_ref; |
839 | pud_t *pud, *pud_ref; | |
840 | pmd_t *pmd, *pmd_ref; | |
841 | pte_t *pte, *pte_ref; | |
842 | ||
2d4a7167 | 843 | /* Make sure we are in vmalloc area: */ |
cf89ec92 HH |
844 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) |
845 | return -1; | |
846 | ||
2d4a7167 IM |
847 | /* |
848 | * Copy kernel mappings over when needed. This can also | |
849 | * happen within a race in page table update. In the later | |
850 | * case just flush: | |
851 | */ | |
f313e123 | 852 | pgd = pgd_offset(current->active_mm, address); |
1da177e4 LT |
853 | pgd_ref = pgd_offset_k(address); |
854 | if (pgd_none(*pgd_ref)) | |
855 | return -1; | |
2d4a7167 | 856 | |
1da177e4 LT |
857 | if (pgd_none(*pgd)) |
858 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 859 | else |
46a82b2d | 860 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 | 861 | |
2d4a7167 IM |
862 | /* |
863 | * Below here mismatches are bugs because these lower tables | |
864 | * are shared: | |
865 | */ | |
1da177e4 LT |
866 | |
867 | pud = pud_offset(pgd, address); | |
868 | pud_ref = pud_offset(pgd_ref, address); | |
869 | if (pud_none(*pud_ref)) | |
870 | return -1; | |
2d4a7167 | 871 | |
46a82b2d | 872 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 | 873 | BUG(); |
2d4a7167 | 874 | |
1da177e4 LT |
875 | pmd = pmd_offset(pud, address); |
876 | pmd_ref = pmd_offset(pud_ref, address); | |
877 | if (pmd_none(*pmd_ref)) | |
878 | return -1; | |
2d4a7167 | 879 | |
1da177e4 LT |
880 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) |
881 | BUG(); | |
2d4a7167 | 882 | |
1da177e4 LT |
883 | pte_ref = pte_offset_kernel(pmd_ref, address); |
884 | if (!pte_present(*pte_ref)) | |
885 | return -1; | |
2d4a7167 | 886 | |
1da177e4 | 887 | pte = pte_offset_kernel(pmd, address); |
2d4a7167 IM |
888 | |
889 | /* | |
890 | * Don't use pte_page here, because the mappings can point | |
891 | * outside mem_map, and the NUMA hash lookup cannot handle | |
892 | * that: | |
893 | */ | |
3b9ba4d5 | 894 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) |
1da177e4 | 895 | BUG(); |
2d4a7167 | 896 | |
1da177e4 | 897 | return 0; |
fdfe8aa8 | 898 | #endif |
1da177e4 LT |
899 | } |
900 | ||
abd4f750 | 901 | int show_unhandled_signals = 1; |
1da177e4 | 902 | |
2d4a7167 IM |
903 | static inline int |
904 | access_error(unsigned long error_code, int write, struct vm_area_struct *vma) | |
92181f19 NP |
905 | { |
906 | if (write) { | |
2d4a7167 | 907 | /* write, present and write, not present: */ |
92181f19 NP |
908 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
909 | return 1; | |
2d4a7167 | 910 | return 0; |
92181f19 NP |
911 | } |
912 | ||
2d4a7167 IM |
913 | /* read, present: */ |
914 | if (unlikely(error_code & PF_PROT)) | |
915 | return 1; | |
916 | ||
917 | /* read, not present: */ | |
918 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
919 | return 1; | |
920 | ||
92181f19 NP |
921 | return 0; |
922 | } | |
923 | ||
0973a06c HS |
924 | static int fault_in_kernel_space(unsigned long address) |
925 | { | |
926 | #ifdef CONFIG_X86_32 | |
927 | return address >= TASK_SIZE; | |
2d4a7167 | 928 | #else |
0973a06c | 929 | return address >= TASK_SIZE64; |
2d4a7167 | 930 | #endif |
0973a06c HS |
931 | } |
932 | ||
1da177e4 LT |
933 | /* |
934 | * This routine handles page faults. It determines the address, | |
935 | * and the problem, and then passes it off to one of the appropriate | |
936 | * routines. | |
1da177e4 | 937 | */ |
f8c2ee22 HH |
938 | #ifdef CONFIG_X86_64 |
939 | asmlinkage | |
940 | #endif | |
941 | void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 | 942 | { |
2d4a7167 | 943 | struct vm_area_struct *vma; |
1da177e4 | 944 | struct task_struct *tsk; |
2d4a7167 | 945 | unsigned long address; |
1da177e4 | 946 | struct mm_struct *mm; |
92181f19 | 947 | int write; |
f8c2ee22 | 948 | int fault; |
1da177e4 | 949 | |
a9ba9a3b AV |
950 | tsk = current; |
951 | mm = tsk->mm; | |
2d4a7167 | 952 | |
a9ba9a3b AV |
953 | prefetchw(&mm->mmap_sem); |
954 | ||
2d4a7167 | 955 | /* Get the faulting address: */ |
f51c9452 | 956 | address = read_cr2(); |
1da177e4 | 957 | |
0fd0e3da | 958 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 959 | return; |
1da177e4 LT |
960 | |
961 | /* | |
962 | * We fault-in kernel-space virtual memory on-demand. The | |
963 | * 'reference' page table is init_mm.pgd. | |
964 | * | |
965 | * NOTE! We MUST NOT take any locks for this case. We may | |
966 | * be in an interrupt or a critical region, and should | |
967 | * only copy the information from the master page table, | |
968 | * nothing more. | |
969 | * | |
970 | * This verifies that the fault happens in kernel space | |
971 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 972 | * protection error (error_code & 9) == 0. |
1da177e4 | 973 | */ |
0973a06c | 974 | if (unlikely(fault_in_kernel_space(address))) { |
f8c2ee22 HH |
975 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
976 | vmalloc_fault(address) >= 0) | |
977 | return; | |
5b727a3b | 978 | |
2d4a7167 | 979 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 980 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
981 | return; |
982 | ||
2d4a7167 | 983 | /* kprobes don't want to hook the spurious faults: */ |
9be260a6 MH |
984 | if (notify_page_fault(regs)) |
985 | return; | |
f8c2ee22 HH |
986 | /* |
987 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 988 | * fault we could otherwise deadlock: |
f8c2ee22 | 989 | */ |
92181f19 | 990 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 991 | |
92181f19 | 992 | return; |
f8c2ee22 HH |
993 | } |
994 | ||
2d4a7167 | 995 | /* kprobes don't want to hook the spurious faults: */ |
f8a6b2b9 | 996 | if (unlikely(notify_page_fault(regs))) |
9be260a6 | 997 | return; |
f8c2ee22 | 998 | /* |
891cffbd LT |
999 | * It's safe to allow irq's after cr2 has been saved and the |
1000 | * vmalloc fault has been handled. | |
1001 | * | |
1002 | * User-mode registers count as a user access even for any | |
2d4a7167 | 1003 | * potential system fault or CPU buglet: |
f8c2ee22 | 1004 | */ |
891cffbd LT |
1005 | if (user_mode_vm(regs)) { |
1006 | local_irq_enable(); | |
1007 | error_code |= PF_USER; | |
2d4a7167 IM |
1008 | } else { |
1009 | if (regs->flags & X86_EFLAGS_IF) | |
1010 | local_irq_enable(); | |
1011 | } | |
8c914cb7 | 1012 | |
66c58156 | 1013 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1014 | pgtable_bad(regs, error_code, address); |
1da177e4 LT |
1015 | |
1016 | /* | |
2d4a7167 IM |
1017 | * If we're in an interrupt, have no user context or are running |
1018 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1019 | */ |
92181f19 NP |
1020 | if (unlikely(in_atomic() || !mm)) { |
1021 | bad_area_nosemaphore(regs, error_code, address); | |
1022 | return; | |
1023 | } | |
1da177e4 | 1024 | |
3a1dfe6e IM |
1025 | /* |
1026 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1027 | * addresses in user space. All other faults represent errors in |
1028 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1029 | * case of an erroneous fault occurring in a code path which already | |
1030 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1031 | * against the address space. Luckily the kernel only validly | |
1032 | * references user space from well defined areas of code, which are | |
1033 | * listed in the exceptions table. | |
1da177e4 LT |
1034 | * |
1035 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1036 | * the source reference check when there is a possibility of a |
1037 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1038 | * validate the source. If this is invalid we can skip the address | |
1039 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1040 | */ |
92181f19 | 1041 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1042 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1043 | !search_exception_tables(regs->ip)) { |
1044 | bad_area_nosemaphore(regs, error_code, address); | |
1045 | return; | |
1046 | } | |
1da177e4 | 1047 | down_read(&mm->mmap_sem); |
01006074 PZ |
1048 | } else { |
1049 | /* | |
2d4a7167 IM |
1050 | * The above down_read_trylock() might have succeeded in |
1051 | * which case we'll have missed the might_sleep() from | |
1052 | * down_read(): | |
01006074 PZ |
1053 | */ |
1054 | might_sleep(); | |
1da177e4 LT |
1055 | } |
1056 | ||
1057 | vma = find_vma(mm, address); | |
92181f19 NP |
1058 | if (unlikely(!vma)) { |
1059 | bad_area(regs, error_code, address); | |
1060 | return; | |
1061 | } | |
1062 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1063 | goto good_area; |
92181f19 NP |
1064 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1065 | bad_area(regs, error_code, address); | |
1066 | return; | |
1067 | } | |
33cb5243 | 1068 | if (error_code & PF_USER) { |
6f4d368e HH |
1069 | /* |
1070 | * Accessing the stack below %sp is always a bug. | |
1071 | * The large cushion allows instructions like enter | |
2d4a7167 | 1072 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1073 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1074 | */ |
92181f19 NP |
1075 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1076 | bad_area(regs, error_code, address); | |
1077 | return; | |
1078 | } | |
1da177e4 | 1079 | } |
92181f19 NP |
1080 | if (unlikely(expand_stack(vma, address))) { |
1081 | bad_area(regs, error_code, address); | |
1082 | return; | |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * Ok, we have a good vm_area for this memory access, so | |
1087 | * we can handle it.. | |
1088 | */ | |
1da177e4 | 1089 | good_area: |
92181f19 | 1090 | write = error_code & PF_WRITE; |
2d4a7167 | 1091 | |
92181f19 NP |
1092 | if (unlikely(access_error(error_code, write, vma))) { |
1093 | bad_area_access_error(regs, error_code, address); | |
1094 | return; | |
1da177e4 LT |
1095 | } |
1096 | ||
1097 | /* | |
1098 | * If for any reason at all we couldn't handle the fault, | |
1099 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1100 | * the fault: |
1da177e4 | 1101 | */ |
83c54070 | 1102 | fault = handle_mm_fault(mm, vma, address, write); |
2d4a7167 | 1103 | |
83c54070 | 1104 | if (unlikely(fault & VM_FAULT_ERROR)) { |
92181f19 NP |
1105 | mm_fault_error(regs, error_code, address, fault); |
1106 | return; | |
1da177e4 | 1107 | } |
2d4a7167 | 1108 | |
83c54070 NP |
1109 | if (fault & VM_FAULT_MAJOR) |
1110 | tsk->maj_flt++; | |
1111 | else | |
1112 | tsk->min_flt++; | |
d729ab35 | 1113 | |
8c938f9f IM |
1114 | check_v8086_mode(regs, address, tsk); |
1115 | ||
1da177e4 | 1116 | up_read(&mm->mmap_sem); |
1da177e4 | 1117 | } |
9e43e1b7 | 1118 | |
8c914cb7 | 1119 | DEFINE_SPINLOCK(pgd_lock); |
2bff7383 | 1120 | LIST_HEAD(pgd_list); |
8c914cb7 JB |
1121 | |
1122 | void vmalloc_sync_all(void) | |
1123 | { | |
1156e098 HH |
1124 | unsigned long address; |
1125 | ||
cc643d46 | 1126 | #ifdef CONFIG_X86_32 |
1156e098 HH |
1127 | if (SHARED_KERNEL_PMD) |
1128 | return; | |
1129 | ||
cc643d46 JB |
1130 | for (address = VMALLOC_START & PMD_MASK; |
1131 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
1132 | address += PMD_SIZE) { | |
2d4a7167 | 1133 | |
67350a5c JF |
1134 | unsigned long flags; |
1135 | struct page *page; | |
1136 | ||
1137 | spin_lock_irqsave(&pgd_lock, flags); | |
1138 | list_for_each_entry(page, &pgd_list, lru) { | |
2d4a7167 | 1139 | if (!vmalloc_sync_one(page_address(page), address)) |
67350a5c | 1140 | break; |
1156e098 | 1141 | } |
67350a5c | 1142 | spin_unlock_irqrestore(&pgd_lock, flags); |
1156e098 HH |
1143 | } |
1144 | #else /* CONFIG_X86_64 */ | |
cc643d46 JB |
1145 | for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END; |
1146 | address += PGDIR_SIZE) { | |
2d4a7167 | 1147 | |
67350a5c JF |
1148 | const pgd_t *pgd_ref = pgd_offset_k(address); |
1149 | unsigned long flags; | |
1150 | struct page *page; | |
1151 | ||
1152 | if (pgd_none(*pgd_ref)) | |
1153 | continue; | |
2d4a7167 | 1154 | |
67350a5c JF |
1155 | spin_lock_irqsave(&pgd_lock, flags); |
1156 | list_for_each_entry(page, &pgd_list, lru) { | |
1157 | pgd_t *pgd; | |
1158 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
1159 | if (pgd_none(*pgd)) | |
1160 | set_pgd(pgd, *pgd_ref); | |
1161 | else | |
1162 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
8c914cb7 | 1163 | } |
67350a5c | 1164 | spin_unlock_irqrestore(&pgd_lock, flags); |
8c914cb7 | 1165 | } |
1156e098 | 1166 | #endif |
8c914cb7 | 1167 | } |