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 | |
0fd0e3da | 58 | static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr) |
86069782 | 59 | { |
fd3fdf11 | 60 | #ifdef CONFIG_MMIOTRACE |
0fd0e3da PP |
61 | if (unlikely(is_kmmio_active())) |
62 | if (kmmio_handler(regs, addr) == 1) | |
63 | return -1; | |
86069782 | 64 | #endif |
0fd0e3da | 65 | return 0; |
86069782 PP |
66 | } |
67 | ||
74a0b576 | 68 | static inline int notify_page_fault(struct pt_regs *regs) |
1bd858a5 | 69 | { |
33cb5243 | 70 | #ifdef CONFIG_KPROBES |
74a0b576 CH |
71 | int ret = 0; |
72 | ||
73 | /* kprobe_running() needs smp_processor_id() */ | |
f8c2ee22 | 74 | if (!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; |
74a0b576 | 82 | #else |
74a0b576 | 83 | return 0; |
74a0b576 | 84 | #endif |
33cb5243 | 85 | } |
1bd858a5 | 86 | |
1dc85be0 | 87 | /* |
2d4a7167 IM |
88 | * Prefetch quirks: |
89 | * | |
90 | * 32-bit mode: | |
91 | * | |
92 | * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. | |
93 | * Check that here and ignore it. | |
1dc85be0 | 94 | * |
2d4a7167 | 95 | * 64-bit mode: |
1dc85be0 | 96 | * |
2d4a7167 IM |
97 | * Sometimes the CPU reports invalid exceptions on prefetch. |
98 | * Check that here and ignore it. | |
99 | * | |
100 | * Opcode checker based on code by Richard Brunner. | |
1dc85be0 | 101 | */ |
107a0367 IM |
102 | static inline int |
103 | check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, | |
104 | unsigned char opcode, int *prefetch) | |
105 | { | |
106 | unsigned char instr_hi = opcode & 0xf0; | |
107 | unsigned char instr_lo = opcode & 0x0f; | |
108 | ||
109 | switch (instr_hi) { | |
110 | case 0x20: | |
111 | case 0x30: | |
112 | /* | |
113 | * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. | |
114 | * In X86_64 long mode, the CPU will signal invalid | |
115 | * opcode if some of these prefixes are present so | |
116 | * X86_64 will never get here anyway | |
117 | */ | |
118 | return ((instr_lo & 7) == 0x6); | |
119 | #ifdef CONFIG_X86_64 | |
120 | case 0x40: | |
121 | /* | |
122 | * In AMD64 long mode 0x40..0x4F are valid REX prefixes | |
123 | * Need to figure out under what instruction mode the | |
124 | * instruction was issued. Could check the LDT for lm, | |
125 | * but for now it's good enough to assume that long | |
126 | * mode only uses well known segments or kernel. | |
127 | */ | |
128 | return (!user_mode(regs)) || (regs->cs == __USER_CS); | |
129 | #endif | |
130 | case 0x60: | |
131 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
132 | return (instr_lo & 0xC) == 0x4; | |
133 | case 0xF0: | |
134 | /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ | |
135 | return !instr_lo || (instr_lo>>1) == 1; | |
136 | case 0x00: | |
137 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
138 | if (probe_kernel_address(instr, opcode)) | |
139 | return 0; | |
140 | ||
141 | *prefetch = (instr_lo == 0xF) && | |
142 | (opcode == 0x0D || opcode == 0x18); | |
143 | return 0; | |
144 | default: | |
145 | return 0; | |
146 | } | |
147 | } | |
148 | ||
2d4a7167 IM |
149 | static int |
150 | is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) | |
33cb5243 | 151 | { |
2d4a7167 | 152 | unsigned char *max_instr; |
ab2bf0c1 | 153 | unsigned char *instr; |
33cb5243 | 154 | int prefetch = 0; |
1da177e4 | 155 | |
3085354d IM |
156 | /* |
157 | * If it was a exec (instruction fetch) fault on NX page, then | |
158 | * do not ignore the fault: | |
159 | */ | |
66c58156 | 160 | if (error_code & PF_INSTR) |
1da177e4 | 161 | return 0; |
1dc85be0 | 162 | |
107a0367 | 163 | instr = (void *)convert_ip_to_linear(current, regs); |
f1290ec9 | 164 | max_instr = instr + 15; |
1da177e4 | 165 | |
76381fee | 166 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
167 | return 0; |
168 | ||
107a0367 | 169 | while (instr < max_instr) { |
2d4a7167 | 170 | unsigned char opcode; |
1da177e4 | 171 | |
ab2bf0c1 | 172 | if (probe_kernel_address(instr, opcode)) |
33cb5243 | 173 | break; |
1da177e4 | 174 | |
1da177e4 LT |
175 | instr++; |
176 | ||
107a0367 | 177 | if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) |
1da177e4 | 178 | break; |
1da177e4 LT |
179 | } |
180 | return prefetch; | |
181 | } | |
182 | ||
2d4a7167 IM |
183 | static void |
184 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
185 | struct task_struct *tsk) | |
c4aba4a8 HH |
186 | { |
187 | siginfo_t info; | |
188 | ||
2d4a7167 IM |
189 | info.si_signo = si_signo; |
190 | info.si_errno = 0; | |
191 | info.si_code = si_code; | |
192 | info.si_addr = (void __user *)address; | |
193 | ||
c4aba4a8 HH |
194 | force_sig_info(si_signo, &info, tsk); |
195 | } | |
196 | ||
1156e098 | 197 | #ifdef CONFIG_X86_64 |
33cb5243 HH |
198 | static int bad_address(void *p) |
199 | { | |
1da177e4 | 200 | unsigned long dummy; |
2d4a7167 | 201 | |
ab2bf0c1 | 202 | return probe_kernel_address((unsigned long *)p, dummy); |
33cb5243 | 203 | } |
1156e098 | 204 | #endif |
1da177e4 | 205 | |
cae30f82 | 206 | static void dump_pagetable(unsigned long address) |
1da177e4 | 207 | { |
1156e098 HH |
208 | #ifdef CONFIG_X86_32 |
209 | __typeof__(pte_val(__pte(0))) page; | |
210 | ||
211 | page = read_cr3(); | |
212 | page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT]; | |
2d4a7167 | 213 | |
1156e098 HH |
214 | #ifdef CONFIG_X86_PAE |
215 | printk("*pdpt = %016Lx ", page); | |
216 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
217 | && page & _PAGE_PRESENT) { | |
218 | page &= PAGE_MASK; | |
219 | page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT) | |
2d4a7167 | 220 | & (PTRS_PER_PMD - 1)]; |
1156e098 HH |
221 | printk(KERN_CONT "*pde = %016Lx ", page); |
222 | page &= ~_PAGE_NX; | |
223 | } | |
224 | #else | |
225 | printk("*pde = %08lx ", page); | |
226 | #endif | |
227 | ||
228 | /* | |
229 | * We must not directly access the pte in the highpte | |
230 | * case if the page table is located in highmem. | |
231 | * And let's rather not kmap-atomic the pte, just in case | |
2d4a7167 | 232 | * it's allocated already: |
1156e098 HH |
233 | */ |
234 | if ((page >> PAGE_SHIFT) < max_low_pfn | |
235 | && (page & _PAGE_PRESENT) | |
236 | && !(page & _PAGE_PSE)) { | |
2d4a7167 | 237 | |
1156e098 HH |
238 | page &= PAGE_MASK; |
239 | page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT) | |
2d4a7167 | 240 | & (PTRS_PER_PTE - 1)]; |
1156e098 HH |
241 | printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page); |
242 | } | |
243 | ||
244 | printk("\n"); | |
245 | #else /* CONFIG_X86_64 */ | |
1da177e4 LT |
246 | pgd_t *pgd; |
247 | pud_t *pud; | |
248 | pmd_t *pmd; | |
249 | pte_t *pte; | |
250 | ||
f51c9452 | 251 | pgd = (pgd_t *)read_cr3(); |
1da177e4 | 252 | |
33cb5243 | 253 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); |
2d4a7167 | 254 | |
1da177e4 | 255 | pgd += pgd_index(address); |
2d4a7167 IM |
256 | if (bad_address(pgd)) |
257 | goto bad; | |
258 | ||
d646bce4 | 259 | printk("PGD %lx ", pgd_val(*pgd)); |
2d4a7167 IM |
260 | |
261 | if (!pgd_present(*pgd)) | |
262 | goto out; | |
1da177e4 | 263 | |
d2ae5b5f | 264 | pud = pud_offset(pgd, address); |
2d4a7167 IM |
265 | if (bad_address(pud)) |
266 | goto bad; | |
267 | ||
1da177e4 | 268 | printk("PUD %lx ", pud_val(*pud)); |
b5360222 | 269 | if (!pud_present(*pud) || pud_large(*pud)) |
2d4a7167 | 270 | goto out; |
1da177e4 LT |
271 | |
272 | pmd = pmd_offset(pud, address); | |
2d4a7167 IM |
273 | if (bad_address(pmd)) |
274 | goto bad; | |
275 | ||
1da177e4 | 276 | printk("PMD %lx ", pmd_val(*pmd)); |
2d4a7167 IM |
277 | if (!pmd_present(*pmd) || pmd_large(*pmd)) |
278 | goto out; | |
1da177e4 LT |
279 | |
280 | pte = pte_offset_kernel(pmd, address); | |
2d4a7167 IM |
281 | if (bad_address(pte)) |
282 | goto bad; | |
283 | ||
33cb5243 | 284 | printk("PTE %lx", pte_val(*pte)); |
2d4a7167 | 285 | out: |
1da177e4 LT |
286 | printk("\n"); |
287 | return; | |
288 | bad: | |
289 | printk("BAD\n"); | |
1156e098 HH |
290 | #endif |
291 | } | |
292 | ||
293 | #ifdef CONFIG_X86_32 | |
294 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) | |
295 | { | |
296 | unsigned index = pgd_index(address); | |
297 | pgd_t *pgd_k; | |
298 | pud_t *pud, *pud_k; | |
299 | pmd_t *pmd, *pmd_k; | |
300 | ||
301 | pgd += index; | |
302 | pgd_k = init_mm.pgd + index; | |
303 | ||
304 | if (!pgd_present(*pgd_k)) | |
305 | return NULL; | |
306 | ||
307 | /* | |
308 | * set_pgd(pgd, *pgd_k); here would be useless on PAE | |
309 | * and redundant with the set_pmd() on non-PAE. As would | |
310 | * set_pud. | |
311 | */ | |
1156e098 HH |
312 | pud = pud_offset(pgd, address); |
313 | pud_k = pud_offset(pgd_k, address); | |
314 | if (!pud_present(*pud_k)) | |
315 | return NULL; | |
316 | ||
317 | pmd = pmd_offset(pud, address); | |
318 | pmd_k = pmd_offset(pud_k, address); | |
319 | if (!pmd_present(*pmd_k)) | |
320 | return NULL; | |
2d4a7167 | 321 | |
1156e098 HH |
322 | if (!pmd_present(*pmd)) { |
323 | set_pmd(pmd, *pmd_k); | |
324 | arch_flush_lazy_mmu_mode(); | |
2d4a7167 | 325 | } else { |
1156e098 | 326 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
2d4a7167 IM |
327 | } |
328 | ||
1156e098 | 329 | return pmd_k; |
1da177e4 LT |
330 | } |
331 | ||
8c938f9f IM |
332 | /* |
333 | * Did it hit the DOS screen memory VA from vm86 mode? | |
334 | */ | |
335 | static inline void | |
336 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
337 | struct task_struct *tsk) | |
338 | { | |
339 | unsigned long bit; | |
340 | ||
341 | if (!v8086_mode(regs)) | |
342 | return; | |
343 | ||
344 | bit = (address - 0xA0000) >> PAGE_SHIFT; | |
345 | if (bit < 32) | |
346 | tsk->thread.screen_bitmap |= 1 << bit; | |
347 | } | |
348 | ||
349 | #else /* CONFIG_X86_64: */ | |
350 | ||
33cb5243 | 351 | static const char errata93_warning[] = |
1da177e4 LT |
352 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" |
353 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
354 | KERN_ERR "******* Please consider a BIOS update.\n" | |
355 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
8c938f9f IM |
356 | |
357 | /* | |
358 | * No vm86 mode in 64-bit mode: | |
359 | */ | |
360 | static inline void | |
361 | check_v8086_mode(struct pt_regs *regs, unsigned long address, | |
362 | struct task_struct *tsk) | |
363 | { | |
364 | } | |
365 | ||
fdfe8aa8 | 366 | #endif |
1da177e4 | 367 | |
2d4a7167 IM |
368 | /* |
369 | * Workaround for K8 erratum #93 & buggy BIOS. | |
370 | * | |
371 | * BIOS SMM functions are required to use a specific workaround | |
372 | * to avoid corruption of the 64bit RIP register on C stepping K8. | |
373 | * | |
374 | * A lot of BIOS that didn't get tested properly miss this. | |
375 | * | |
376 | * The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
377 | * Try to work around it here. | |
378 | * | |
379 | * Note we only handle faults in kernel here. | |
380 | * Does nothing on 32-bit. | |
fdfe8aa8 | 381 | */ |
33cb5243 | 382 | static int is_errata93(struct pt_regs *regs, unsigned long address) |
1da177e4 | 383 | { |
fdfe8aa8 | 384 | #ifdef CONFIG_X86_64 |
2d4a7167 IM |
385 | static int once; |
386 | ||
65ea5b03 | 387 | if (address != regs->ip) |
1da177e4 | 388 | return 0; |
2d4a7167 | 389 | |
33cb5243 | 390 | if ((address >> 32) != 0) |
1da177e4 | 391 | return 0; |
2d4a7167 | 392 | |
1da177e4 | 393 | address |= 0xffffffffUL << 32; |
33cb5243 HH |
394 | if ((address >= (u64)_stext && address <= (u64)_etext) || |
395 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
2d4a7167 | 396 | if (!once) { |
33cb5243 | 397 | printk(errata93_warning); |
2d4a7167 | 398 | once = 1; |
1da177e4 | 399 | } |
65ea5b03 | 400 | regs->ip = address; |
1da177e4 LT |
401 | return 1; |
402 | } | |
fdfe8aa8 | 403 | #endif |
1da177e4 | 404 | return 0; |
33cb5243 | 405 | } |
1da177e4 | 406 | |
35f3266f | 407 | /* |
2d4a7167 IM |
408 | * Work around K8 erratum #100 K8 in compat mode occasionally jumps |
409 | * to illegal addresses >4GB. | |
410 | * | |
411 | * We catch this in the page fault handler because these addresses | |
412 | * are not reachable. Just detect this case and return. Any code | |
35f3266f HH |
413 | * segment in LDT is compatibility mode. |
414 | */ | |
415 | static int is_errata100(struct pt_regs *regs, unsigned long address) | |
416 | { | |
417 | #ifdef CONFIG_X86_64 | |
2d4a7167 | 418 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) |
35f3266f HH |
419 | return 1; |
420 | #endif | |
421 | return 0; | |
422 | } | |
423 | ||
29caf2f9 HH |
424 | static int is_f00f_bug(struct pt_regs *regs, unsigned long address) |
425 | { | |
426 | #ifdef CONFIG_X86_F00F_BUG | |
427 | unsigned long nr; | |
2d4a7167 | 428 | |
29caf2f9 | 429 | /* |
2d4a7167 | 430 | * Pentium F0 0F C7 C8 bug workaround: |
29caf2f9 HH |
431 | */ |
432 | if (boot_cpu_data.f00f_bug) { | |
433 | nr = (address - idt_descr.address) >> 3; | |
434 | ||
435 | if (nr == 6) { | |
436 | do_invalid_op(regs, 0); | |
437 | return 1; | |
438 | } | |
439 | } | |
440 | #endif | |
441 | return 0; | |
442 | } | |
443 | ||
2d4a7167 IM |
444 | static void |
445 | show_fault_oops(struct pt_regs *regs, unsigned long error_code, | |
446 | unsigned long address) | |
b3279c7f | 447 | { |
1156e098 HH |
448 | #ifdef CONFIG_X86_32 |
449 | if (!oops_may_print()) | |
450 | return; | |
fd40d6e3 | 451 | #endif |
1156e098 HH |
452 | |
453 | #ifdef CONFIG_X86_PAE | |
454 | if (error_code & PF_INSTR) { | |
93809be8 | 455 | unsigned int level; |
2d4a7167 | 456 | |
1156e098 HH |
457 | pte_t *pte = lookup_address(address, &level); |
458 | ||
2d4a7167 | 459 | if (pte && pte_present(*pte) && !pte_exec(*pte)) { |
1156e098 HH |
460 | printk(KERN_CRIT "kernel tried to execute " |
461 | "NX-protected page - exploit attempt? " | |
350b4da7 | 462 | "(uid: %d)\n", current_uid()); |
2d4a7167 | 463 | } |
1156e098 HH |
464 | } |
465 | #endif | |
1156e098 | 466 | |
19f0dda9 | 467 | printk(KERN_ALERT "BUG: unable to handle kernel "); |
b3279c7f | 468 | if (address < PAGE_SIZE) |
19f0dda9 | 469 | printk(KERN_CONT "NULL pointer dereference"); |
b3279c7f | 470 | else |
19f0dda9 | 471 | printk(KERN_CONT "paging request"); |
2d4a7167 | 472 | |
f294a8ce | 473 | printk(KERN_CONT " at %p\n", (void *) address); |
19f0dda9 | 474 | printk(KERN_ALERT "IP:"); |
b3279c7f | 475 | printk_address(regs->ip, 1); |
2d4a7167 | 476 | |
b3279c7f HH |
477 | dump_pagetable(address); |
478 | } | |
479 | ||
2d4a7167 IM |
480 | static noinline void |
481 | pgtable_bad(struct pt_regs *regs, unsigned long error_code, | |
482 | unsigned long address) | |
1da177e4 | 483 | { |
2d4a7167 IM |
484 | struct task_struct *tsk; |
485 | unsigned long flags; | |
486 | int sig; | |
487 | ||
488 | flags = oops_begin(); | |
489 | tsk = current; | |
490 | sig = SIGKILL; | |
1209140c | 491 | |
1da177e4 | 492 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
92181f19 | 493 | tsk->comm, address); |
1da177e4 | 494 | dump_pagetable(address); |
2d4a7167 IM |
495 | |
496 | tsk->thread.cr2 = address; | |
497 | tsk->thread.trap_no = 14; | |
498 | tsk->thread.error_code = error_code; | |
499 | ||
22f5991c | 500 | if (__die("Bad pagetable", regs, error_code)) |
874d93d1 | 501 | sig = 0; |
2d4a7167 | 502 | |
874d93d1 | 503 | oops_end(flags, regs, sig); |
1da177e4 LT |
504 | } |
505 | ||
2d4a7167 IM |
506 | static noinline void |
507 | no_context(struct pt_regs *regs, unsigned long error_code, | |
508 | unsigned long address) | |
92181f19 NP |
509 | { |
510 | struct task_struct *tsk = current; | |
19803078 IM |
511 | unsigned long *stackend; |
512 | ||
92181f19 NP |
513 | #ifdef CONFIG_X86_64 |
514 | unsigned long flags; | |
515 | int sig; | |
516 | #endif | |
517 | ||
2d4a7167 | 518 | /* Are we prepared to handle this kernel fault? */ |
92181f19 NP |
519 | if (fixup_exception(regs)) |
520 | return; | |
521 | ||
522 | /* | |
2d4a7167 IM |
523 | * 32-bit: |
524 | * | |
525 | * Valid to do another page fault here, because if this fault | |
526 | * had been triggered by is_prefetch fixup_exception would have | |
527 | * handled it. | |
528 | * | |
529 | * 64-bit: | |
92181f19 | 530 | * |
2d4a7167 | 531 | * Hall of shame of CPU/BIOS bugs. |
92181f19 NP |
532 | */ |
533 | if (is_prefetch(regs, error_code, address)) | |
534 | return; | |
535 | ||
536 | if (is_errata93(regs, address)) | |
537 | return; | |
538 | ||
539 | /* | |
540 | * Oops. The kernel tried to access some bad page. We'll have to | |
2d4a7167 | 541 | * terminate things with extreme prejudice: |
92181f19 NP |
542 | */ |
543 | #ifdef CONFIG_X86_32 | |
544 | bust_spinlocks(1); | |
545 | #else | |
546 | flags = oops_begin(); | |
547 | #endif | |
548 | ||
549 | show_fault_oops(regs, error_code, address); | |
550 | ||
2d4a7167 | 551 | stackend = end_of_stack(tsk); |
19803078 IM |
552 | if (*stackend != STACK_END_MAGIC) |
553 | printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); | |
554 | ||
92181f19 NP |
555 | tsk->thread.cr2 = address; |
556 | tsk->thread.trap_no = 14; | |
557 | tsk->thread.error_code = error_code; | |
558 | ||
559 | #ifdef CONFIG_X86_32 | |
560 | die("Oops", regs, error_code); | |
561 | bust_spinlocks(0); | |
562 | do_exit(SIGKILL); | |
563 | #else | |
564 | sig = SIGKILL; | |
565 | if (__die("Oops", regs, error_code)) | |
566 | sig = 0; | |
2d4a7167 | 567 | |
92181f19 NP |
568 | /* Executive summary in case the body of the oops scrolled away */ |
569 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
2d4a7167 | 570 | |
92181f19 NP |
571 | oops_end(flags, regs, sig); |
572 | #endif | |
573 | } | |
574 | ||
2d4a7167 IM |
575 | /* |
576 | * Print out info about fatal segfaults, if the show_unhandled_signals | |
577 | * sysctl is set: | |
578 | */ | |
579 | static inline void | |
580 | show_signal_msg(struct pt_regs *regs, unsigned long error_code, | |
581 | unsigned long address, struct task_struct *tsk) | |
582 | { | |
583 | if (!unhandled_signal(tsk, SIGSEGV)) | |
584 | return; | |
585 | ||
586 | if (!printk_ratelimit()) | |
587 | return; | |
588 | ||
589 | printk(KERN_CONT "%s%s[%d]: segfault at %lx ip %p sp %p error %lx", | |
590 | task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, | |
591 | tsk->comm, task_pid_nr(tsk), address, | |
592 | (void *)regs->ip, (void *)regs->sp, error_code); | |
593 | ||
594 | print_vma_addr(KERN_CONT " in ", regs->ip); | |
595 | ||
596 | printk(KERN_CONT "\n"); | |
597 | } | |
598 | ||
599 | static void | |
600 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
601 | unsigned long address, int si_code) | |
92181f19 NP |
602 | { |
603 | struct task_struct *tsk = current; | |
604 | ||
605 | /* User mode accesses just cause a SIGSEGV */ | |
606 | if (error_code & PF_USER) { | |
607 | /* | |
2d4a7167 | 608 | * It's possible to have interrupts off here: |
92181f19 NP |
609 | */ |
610 | local_irq_enable(); | |
611 | ||
612 | /* | |
613 | * Valid to do another page fault here because this one came | |
2d4a7167 | 614 | * from user space: |
92181f19 NP |
615 | */ |
616 | if (is_prefetch(regs, error_code, address)) | |
617 | return; | |
618 | ||
619 | if (is_errata100(regs, address)) | |
620 | return; | |
621 | ||
2d4a7167 IM |
622 | if (unlikely(show_unhandled_signals)) |
623 | show_signal_msg(regs, error_code, address, tsk); | |
624 | ||
625 | /* Kernel addresses are always protection faults: */ | |
626 | tsk->thread.cr2 = address; | |
627 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
628 | tsk->thread.trap_no = 14; | |
92181f19 | 629 | |
92181f19 | 630 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); |
2d4a7167 | 631 | |
92181f19 NP |
632 | return; |
633 | } | |
634 | ||
635 | if (is_f00f_bug(regs, address)) | |
636 | return; | |
637 | ||
638 | no_context(regs, error_code, address); | |
639 | } | |
640 | ||
2d4a7167 IM |
641 | static noinline void |
642 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, | |
643 | unsigned long address) | |
92181f19 NP |
644 | { |
645 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); | |
646 | } | |
647 | ||
2d4a7167 IM |
648 | static void |
649 | __bad_area(struct pt_regs *regs, unsigned long error_code, | |
650 | unsigned long address, int si_code) | |
92181f19 NP |
651 | { |
652 | struct mm_struct *mm = current->mm; | |
653 | ||
654 | /* | |
655 | * Something tried to access memory that isn't in our memory map.. | |
656 | * Fix it, but check if it's kernel or user first.. | |
657 | */ | |
658 | up_read(&mm->mmap_sem); | |
659 | ||
660 | __bad_area_nosemaphore(regs, error_code, address, si_code); | |
661 | } | |
662 | ||
2d4a7167 IM |
663 | static noinline void |
664 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
665 | { |
666 | __bad_area(regs, error_code, address, SEGV_MAPERR); | |
667 | } | |
668 | ||
2d4a7167 IM |
669 | static noinline void |
670 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, | |
671 | unsigned long address) | |
92181f19 NP |
672 | { |
673 | __bad_area(regs, error_code, address, SEGV_ACCERR); | |
674 | } | |
675 | ||
676 | /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */ | |
2d4a7167 IM |
677 | static void |
678 | out_of_memory(struct pt_regs *regs, unsigned long error_code, | |
679 | unsigned long address) | |
92181f19 NP |
680 | { |
681 | /* | |
682 | * We ran out of memory, call the OOM killer, and return the userspace | |
2d4a7167 | 683 | * (which will retry the fault, or kill us if we got oom-killed): |
92181f19 NP |
684 | */ |
685 | up_read(¤t->mm->mmap_sem); | |
2d4a7167 | 686 | |
92181f19 NP |
687 | pagefault_out_of_memory(); |
688 | } | |
689 | ||
2d4a7167 IM |
690 | static void |
691 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address) | |
92181f19 NP |
692 | { |
693 | struct task_struct *tsk = current; | |
694 | struct mm_struct *mm = tsk->mm; | |
695 | ||
696 | up_read(&mm->mmap_sem); | |
697 | ||
2d4a7167 | 698 | /* Kernel mode? Handle exceptions or die: */ |
92181f19 NP |
699 | if (!(error_code & PF_USER)) |
700 | no_context(regs, error_code, address); | |
2d4a7167 | 701 | |
92181f19 | 702 | #ifdef CONFIG_X86_32 |
2d4a7167 | 703 | /* User space => ok to do another page fault: */ |
92181f19 NP |
704 | if (is_prefetch(regs, error_code, address)) |
705 | return; | |
706 | #endif | |
2d4a7167 IM |
707 | |
708 | tsk->thread.cr2 = address; | |
709 | tsk->thread.error_code = error_code; | |
710 | tsk->thread.trap_no = 14; | |
711 | ||
92181f19 NP |
712 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); |
713 | } | |
714 | ||
2d4a7167 IM |
715 | static noinline void |
716 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, | |
717 | unsigned long address, unsigned int fault) | |
92181f19 | 718 | { |
2d4a7167 | 719 | if (fault & VM_FAULT_OOM) { |
92181f19 | 720 | out_of_memory(regs, error_code, address); |
2d4a7167 IM |
721 | } else { |
722 | if (fault & VM_FAULT_SIGBUS) | |
723 | do_sigbus(regs, error_code, address); | |
724 | else | |
725 | BUG(); | |
726 | } | |
92181f19 NP |
727 | } |
728 | ||
d8b57bb7 TG |
729 | static int spurious_fault_check(unsigned long error_code, pte_t *pte) |
730 | { | |
731 | if ((error_code & PF_WRITE) && !pte_write(*pte)) | |
732 | return 0; | |
2d4a7167 | 733 | |
d8b57bb7 TG |
734 | if ((error_code & PF_INSTR) && !pte_exec(*pte)) |
735 | return 0; | |
736 | ||
737 | return 1; | |
738 | } | |
739 | ||
5b727a3b | 740 | /* |
2d4a7167 IM |
741 | * Handle a spurious fault caused by a stale TLB entry. |
742 | * | |
743 | * This allows us to lazily refresh the TLB when increasing the | |
744 | * permissions of a kernel page (RO -> RW or NX -> X). Doing it | |
745 | * eagerly is very expensive since that implies doing a full | |
746 | * cross-processor TLB flush, even if no stale TLB entries exist | |
747 | * on other processors. | |
748 | * | |
5b727a3b JF |
749 | * There are no security implications to leaving a stale TLB when |
750 | * increasing the permissions on a page. | |
751 | */ | |
2d4a7167 IM |
752 | static noinline int |
753 | spurious_fault(unsigned long error_code, unsigned long address) | |
5b727a3b JF |
754 | { |
755 | pgd_t *pgd; | |
756 | pud_t *pud; | |
757 | pmd_t *pmd; | |
758 | pte_t *pte; | |
3c3e5694 | 759 | int ret; |
5b727a3b JF |
760 | |
761 | /* Reserved-bit violation or user access to kernel space? */ | |
762 | if (error_code & (PF_USER | PF_RSVD)) | |
763 | return 0; | |
764 | ||
765 | pgd = init_mm.pgd + pgd_index(address); | |
766 | if (!pgd_present(*pgd)) | |
767 | return 0; | |
768 | ||
769 | pud = pud_offset(pgd, address); | |
770 | if (!pud_present(*pud)) | |
771 | return 0; | |
772 | ||
d8b57bb7 TG |
773 | if (pud_large(*pud)) |
774 | return spurious_fault_check(error_code, (pte_t *) pud); | |
775 | ||
5b727a3b JF |
776 | pmd = pmd_offset(pud, address); |
777 | if (!pmd_present(*pmd)) | |
778 | return 0; | |
779 | ||
d8b57bb7 TG |
780 | if (pmd_large(*pmd)) |
781 | return spurious_fault_check(error_code, (pte_t *) pmd); | |
782 | ||
5b727a3b JF |
783 | pte = pte_offset_kernel(pmd, address); |
784 | if (!pte_present(*pte)) | |
785 | return 0; | |
786 | ||
3c3e5694 SR |
787 | ret = spurious_fault_check(error_code, pte); |
788 | if (!ret) | |
789 | return 0; | |
790 | ||
791 | /* | |
2d4a7167 IM |
792 | * Make sure we have permissions in PMD. |
793 | * If not, then there's a bug in the page tables: | |
3c3e5694 SR |
794 | */ |
795 | ret = spurious_fault_check(error_code, (pte_t *) pmd); | |
796 | WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); | |
2d4a7167 | 797 | |
3c3e5694 | 798 | return ret; |
5b727a3b JF |
799 | } |
800 | ||
1da177e4 | 801 | /* |
2d4a7167 IM |
802 | * 32-bit: |
803 | * | |
804 | * Handle a fault on the vmalloc or module mapping area | |
f8c2ee22 | 805 | * |
2d4a7167 IM |
806 | * 64-bit: |
807 | * | |
808 | * Handle a fault on the vmalloc area | |
3b9ba4d5 AK |
809 | * |
810 | * This assumes no large pages in there. | |
1da177e4 | 811 | */ |
92181f19 | 812 | static noinline int vmalloc_fault(unsigned long address) |
1da177e4 | 813 | { |
fdfe8aa8 HH |
814 | #ifdef CONFIG_X86_32 |
815 | unsigned long pgd_paddr; | |
816 | pmd_t *pmd_k; | |
817 | pte_t *pte_k; | |
b29c701d | 818 | |
2d4a7167 | 819 | /* Make sure we are in vmalloc area: */ |
b29c701d HN |
820 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) |
821 | return -1; | |
822 | ||
fdfe8aa8 HH |
823 | /* |
824 | * Synchronize this task's top level page-table | |
825 | * with the 'reference' page table. | |
826 | * | |
827 | * Do _not_ use "current" here. We might be inside | |
828 | * an interrupt in the middle of a task switch.. | |
829 | */ | |
830 | pgd_paddr = read_cr3(); | |
831 | pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); | |
832 | if (!pmd_k) | |
833 | return -1; | |
2d4a7167 | 834 | |
fdfe8aa8 HH |
835 | pte_k = pte_offset_kernel(pmd_k, address); |
836 | if (!pte_present(*pte_k)) | |
837 | return -1; | |
2d4a7167 | 838 | |
fdfe8aa8 HH |
839 | return 0; |
840 | #else | |
1da177e4 LT |
841 | pgd_t *pgd, *pgd_ref; |
842 | pud_t *pud, *pud_ref; | |
843 | pmd_t *pmd, *pmd_ref; | |
844 | pte_t *pte, *pte_ref; | |
845 | ||
2d4a7167 | 846 | /* Make sure we are in vmalloc area: */ |
cf89ec92 HH |
847 | if (!(address >= VMALLOC_START && address < VMALLOC_END)) |
848 | return -1; | |
849 | ||
2d4a7167 IM |
850 | /* |
851 | * Copy kernel mappings over when needed. This can also | |
852 | * happen within a race in page table update. In the later | |
853 | * case just flush: | |
854 | */ | |
f313e123 | 855 | pgd = pgd_offset(current->active_mm, address); |
1da177e4 LT |
856 | pgd_ref = pgd_offset_k(address); |
857 | if (pgd_none(*pgd_ref)) | |
858 | return -1; | |
2d4a7167 | 859 | |
1da177e4 LT |
860 | if (pgd_none(*pgd)) |
861 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 862 | else |
46a82b2d | 863 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 | 864 | |
2d4a7167 IM |
865 | /* |
866 | * Below here mismatches are bugs because these lower tables | |
867 | * are shared: | |
868 | */ | |
1da177e4 LT |
869 | |
870 | pud = pud_offset(pgd, address); | |
871 | pud_ref = pud_offset(pgd_ref, address); | |
872 | if (pud_none(*pud_ref)) | |
873 | return -1; | |
2d4a7167 | 874 | |
46a82b2d | 875 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 | 876 | BUG(); |
2d4a7167 | 877 | |
1da177e4 LT |
878 | pmd = pmd_offset(pud, address); |
879 | pmd_ref = pmd_offset(pud_ref, address); | |
880 | if (pmd_none(*pmd_ref)) | |
881 | return -1; | |
2d4a7167 | 882 | |
1da177e4 LT |
883 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) |
884 | BUG(); | |
2d4a7167 | 885 | |
1da177e4 LT |
886 | pte_ref = pte_offset_kernel(pmd_ref, address); |
887 | if (!pte_present(*pte_ref)) | |
888 | return -1; | |
2d4a7167 | 889 | |
1da177e4 | 890 | pte = pte_offset_kernel(pmd, address); |
2d4a7167 IM |
891 | |
892 | /* | |
893 | * Don't use pte_page here, because the mappings can point | |
894 | * outside mem_map, and the NUMA hash lookup cannot handle | |
895 | * that: | |
896 | */ | |
3b9ba4d5 | 897 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) |
1da177e4 | 898 | BUG(); |
2d4a7167 | 899 | |
1da177e4 | 900 | return 0; |
fdfe8aa8 | 901 | #endif |
1da177e4 LT |
902 | } |
903 | ||
abd4f750 | 904 | int show_unhandled_signals = 1; |
1da177e4 | 905 | |
2d4a7167 IM |
906 | static inline int |
907 | access_error(unsigned long error_code, int write, struct vm_area_struct *vma) | |
92181f19 NP |
908 | { |
909 | if (write) { | |
2d4a7167 | 910 | /* write, present and write, not present: */ |
92181f19 NP |
911 | if (unlikely(!(vma->vm_flags & VM_WRITE))) |
912 | return 1; | |
2d4a7167 | 913 | return 0; |
92181f19 NP |
914 | } |
915 | ||
2d4a7167 IM |
916 | /* read, present: */ |
917 | if (unlikely(error_code & PF_PROT)) | |
918 | return 1; | |
919 | ||
920 | /* read, not present: */ | |
921 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
922 | return 1; | |
923 | ||
92181f19 NP |
924 | return 0; |
925 | } | |
926 | ||
0973a06c HS |
927 | static int fault_in_kernel_space(unsigned long address) |
928 | { | |
929 | #ifdef CONFIG_X86_32 | |
930 | return address >= TASK_SIZE; | |
2d4a7167 | 931 | #else |
0973a06c | 932 | return address >= TASK_SIZE64; |
2d4a7167 | 933 | #endif |
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 | */ |
f8c2ee22 HH |
941 | #ifdef CONFIG_X86_64 |
942 | asmlinkage | |
943 | #endif | |
944 | void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code) | |
1da177e4 | 945 | { |
2d4a7167 | 946 | struct vm_area_struct *vma; |
1da177e4 | 947 | struct task_struct *tsk; |
2d4a7167 | 948 | unsigned long address; |
1da177e4 | 949 | struct mm_struct *mm; |
92181f19 | 950 | int write; |
f8c2ee22 | 951 | int fault; |
1da177e4 | 952 | |
a9ba9a3b AV |
953 | tsk = current; |
954 | mm = tsk->mm; | |
2d4a7167 | 955 | |
a9ba9a3b AV |
956 | prefetchw(&mm->mmap_sem); |
957 | ||
2d4a7167 | 958 | /* Get the faulting address: */ |
f51c9452 | 959 | address = read_cr2(); |
1da177e4 | 960 | |
0fd0e3da | 961 | if (unlikely(kmmio_fault(regs, address))) |
86069782 | 962 | return; |
1da177e4 LT |
963 | |
964 | /* | |
965 | * We fault-in kernel-space virtual memory on-demand. The | |
966 | * 'reference' page table is init_mm.pgd. | |
967 | * | |
968 | * NOTE! We MUST NOT take any locks for this case. We may | |
969 | * be in an interrupt or a critical region, and should | |
970 | * only copy the information from the master page table, | |
971 | * nothing more. | |
972 | * | |
973 | * This verifies that the fault happens in kernel space | |
974 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 975 | * protection error (error_code & 9) == 0. |
1da177e4 | 976 | */ |
0973a06c | 977 | if (unlikely(fault_in_kernel_space(address))) { |
f8c2ee22 HH |
978 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
979 | vmalloc_fault(address) >= 0) | |
980 | return; | |
5b727a3b | 981 | |
2d4a7167 | 982 | /* Can handle a stale RO->RW TLB: */ |
92181f19 | 983 | if (spurious_fault(error_code, address)) |
5b727a3b JF |
984 | return; |
985 | ||
2d4a7167 | 986 | /* kprobes don't want to hook the spurious faults: */ |
9be260a6 MH |
987 | if (notify_page_fault(regs)) |
988 | return; | |
f8c2ee22 HH |
989 | /* |
990 | * Don't take the mm semaphore here. If we fixup a prefetch | |
2d4a7167 | 991 | * fault we could otherwise deadlock: |
f8c2ee22 | 992 | */ |
92181f19 | 993 | bad_area_nosemaphore(regs, error_code, address); |
2d4a7167 | 994 | |
92181f19 | 995 | return; |
f8c2ee22 HH |
996 | } |
997 | ||
2d4a7167 | 998 | /* kprobes don't want to hook the spurious faults: */ |
f8a6b2b9 | 999 | if (unlikely(notify_page_fault(regs))) |
9be260a6 | 1000 | return; |
f8c2ee22 | 1001 | /* |
891cffbd LT |
1002 | * It's safe to allow irq's after cr2 has been saved and the |
1003 | * vmalloc fault has been handled. | |
1004 | * | |
1005 | * User-mode registers count as a user access even for any | |
2d4a7167 | 1006 | * potential system fault or CPU buglet: |
f8c2ee22 | 1007 | */ |
891cffbd LT |
1008 | if (user_mode_vm(regs)) { |
1009 | local_irq_enable(); | |
1010 | error_code |= PF_USER; | |
2d4a7167 IM |
1011 | } else { |
1012 | if (regs->flags & X86_EFLAGS_IF) | |
1013 | local_irq_enable(); | |
1014 | } | |
8c914cb7 | 1015 | |
66c58156 | 1016 | if (unlikely(error_code & PF_RSVD)) |
92181f19 | 1017 | pgtable_bad(regs, error_code, address); |
1da177e4 LT |
1018 | |
1019 | /* | |
2d4a7167 IM |
1020 | * If we're in an interrupt, have no user context or are running |
1021 | * in an atomic region then we must not take the fault: | |
1da177e4 | 1022 | */ |
92181f19 NP |
1023 | if (unlikely(in_atomic() || !mm)) { |
1024 | bad_area_nosemaphore(regs, error_code, address); | |
1025 | return; | |
1026 | } | |
1da177e4 | 1027 | |
3a1dfe6e IM |
1028 | /* |
1029 | * When running in the kernel we expect faults to occur only to | |
2d4a7167 IM |
1030 | * addresses in user space. All other faults represent errors in |
1031 | * the kernel and should generate an OOPS. Unfortunately, in the | |
1032 | * case of an erroneous fault occurring in a code path which already | |
1033 | * holds mmap_sem we will deadlock attempting to validate the fault | |
1034 | * against the address space. Luckily the kernel only validly | |
1035 | * references user space from well defined areas of code, which are | |
1036 | * listed in the exceptions table. | |
1da177e4 LT |
1037 | * |
1038 | * As the vast majority of faults will be valid we will only perform | |
2d4a7167 IM |
1039 | * the source reference check when there is a possibility of a |
1040 | * deadlock. Attempt to lock the address space, if we cannot we then | |
1041 | * validate the source. If this is invalid we can skip the address | |
1042 | * space check, thus avoiding the deadlock: | |
1da177e4 | 1043 | */ |
92181f19 | 1044 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
66c58156 | 1045 | if ((error_code & PF_USER) == 0 && |
92181f19 NP |
1046 | !search_exception_tables(regs->ip)) { |
1047 | bad_area_nosemaphore(regs, error_code, address); | |
1048 | return; | |
1049 | } | |
1da177e4 | 1050 | down_read(&mm->mmap_sem); |
01006074 PZ |
1051 | } else { |
1052 | /* | |
2d4a7167 IM |
1053 | * The above down_read_trylock() might have succeeded in |
1054 | * which case we'll have missed the might_sleep() from | |
1055 | * down_read(): | |
01006074 PZ |
1056 | */ |
1057 | might_sleep(); | |
1da177e4 LT |
1058 | } |
1059 | ||
1060 | vma = find_vma(mm, address); | |
92181f19 NP |
1061 | if (unlikely(!vma)) { |
1062 | bad_area(regs, error_code, address); | |
1063 | return; | |
1064 | } | |
1065 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 1066 | goto good_area; |
92181f19 NP |
1067 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
1068 | bad_area(regs, error_code, address); | |
1069 | return; | |
1070 | } | |
33cb5243 | 1071 | if (error_code & PF_USER) { |
6f4d368e HH |
1072 | /* |
1073 | * Accessing the stack below %sp is always a bug. | |
1074 | * The large cushion allows instructions like enter | |
2d4a7167 | 1075 | * and pusha to work. ("enter $65535, $31" pushes |
6f4d368e | 1076 | * 32 pointers and then decrements %sp by 65535.) |
03fdc2c2 | 1077 | */ |
92181f19 NP |
1078 | if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { |
1079 | bad_area(regs, error_code, address); | |
1080 | return; | |
1081 | } | |
1da177e4 | 1082 | } |
92181f19 NP |
1083 | if (unlikely(expand_stack(vma, address))) { |
1084 | bad_area(regs, error_code, address); | |
1085 | return; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * Ok, we have a good vm_area for this memory access, so | |
1090 | * we can handle it.. | |
1091 | */ | |
1da177e4 | 1092 | good_area: |
92181f19 | 1093 | write = error_code & PF_WRITE; |
2d4a7167 | 1094 | |
92181f19 NP |
1095 | if (unlikely(access_error(error_code, write, vma))) { |
1096 | bad_area_access_error(regs, error_code, address); | |
1097 | return; | |
1da177e4 LT |
1098 | } |
1099 | ||
1100 | /* | |
1101 | * If for any reason at all we couldn't handle the fault, | |
1102 | * make sure we exit gracefully rather than endlessly redo | |
2d4a7167 | 1103 | * the fault: |
1da177e4 | 1104 | */ |
83c54070 | 1105 | fault = handle_mm_fault(mm, vma, address, write); |
2d4a7167 | 1106 | |
83c54070 | 1107 | if (unlikely(fault & VM_FAULT_ERROR)) { |
92181f19 NP |
1108 | mm_fault_error(regs, error_code, address, fault); |
1109 | return; | |
1da177e4 | 1110 | } |
2d4a7167 | 1111 | |
83c54070 NP |
1112 | if (fault & VM_FAULT_MAJOR) |
1113 | tsk->maj_flt++; | |
1114 | else | |
1115 | tsk->min_flt++; | |
d729ab35 | 1116 | |
8c938f9f IM |
1117 | check_v8086_mode(regs, address, tsk); |
1118 | ||
1da177e4 | 1119 | up_read(&mm->mmap_sem); |
1da177e4 | 1120 | } |
9e43e1b7 | 1121 | |
8c914cb7 | 1122 | DEFINE_SPINLOCK(pgd_lock); |
2bff7383 | 1123 | LIST_HEAD(pgd_list); |
8c914cb7 JB |
1124 | |
1125 | void vmalloc_sync_all(void) | |
1126 | { | |
1156e098 HH |
1127 | unsigned long address; |
1128 | ||
cc643d46 | 1129 | #ifdef CONFIG_X86_32 |
1156e098 HH |
1130 | if (SHARED_KERNEL_PMD) |
1131 | return; | |
1132 | ||
cc643d46 JB |
1133 | for (address = VMALLOC_START & PMD_MASK; |
1134 | address >= TASK_SIZE && address < FIXADDR_TOP; | |
1135 | address += PMD_SIZE) { | |
2d4a7167 | 1136 | |
67350a5c JF |
1137 | unsigned long flags; |
1138 | struct page *page; | |
1139 | ||
1140 | spin_lock_irqsave(&pgd_lock, flags); | |
1141 | list_for_each_entry(page, &pgd_list, lru) { | |
2d4a7167 | 1142 | if (!vmalloc_sync_one(page_address(page), address)) |
67350a5c | 1143 | break; |
1156e098 | 1144 | } |
67350a5c | 1145 | spin_unlock_irqrestore(&pgd_lock, flags); |
1156e098 HH |
1146 | } |
1147 | #else /* CONFIG_X86_64 */ | |
cc643d46 JB |
1148 | for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END; |
1149 | address += PGDIR_SIZE) { | |
2d4a7167 | 1150 | |
67350a5c JF |
1151 | const pgd_t *pgd_ref = pgd_offset_k(address); |
1152 | unsigned long flags; | |
1153 | struct page *page; | |
1154 | ||
1155 | if (pgd_none(*pgd_ref)) | |
1156 | continue; | |
2d4a7167 | 1157 | |
67350a5c JF |
1158 | spin_lock_irqsave(&pgd_lock, flags); |
1159 | list_for_each_entry(page, &pgd_list, lru) { | |
1160 | pgd_t *pgd; | |
1161 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
1162 | if (pgd_none(*pgd)) | |
1163 | set_pgd(pgd, *pgd_ref); | |
1164 | else | |
1165 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); | |
8c914cb7 | 1166 | } |
67350a5c | 1167 | spin_unlock_irqrestore(&pgd_lock, flags); |
8c914cb7 | 1168 | } |
1156e098 | 1169 | #endif |
8c914cb7 | 1170 | } |