Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86-64/mm/fault.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs. | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/signal.h> |
9 | #include <linux/sched.h> | |
10 | #include <linux/kernel.h> | |
11 | #include <linux/errno.h> | |
12 | #include <linux/string.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/ptrace.h> | |
15 | #include <linux/mman.h> | |
16 | #include <linux/mm.h> | |
17 | #include <linux/smp.h> | |
18 | #include <linux/smp_lock.h> | |
19 | #include <linux/interrupt.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/tty.h> | |
22 | #include <linux/vt_kern.h> /* For unblank_screen() */ | |
23 | #include <linux/compiler.h> | |
24 | #include <linux/module.h> | |
0f2fbdcb | 25 | #include <linux/kprobes.h> |
1da177e4 LT |
26 | |
27 | #include <asm/system.h> | |
28 | #include <asm/uaccess.h> | |
29 | #include <asm/pgalloc.h> | |
30 | #include <asm/smp.h> | |
31 | #include <asm/tlbflush.h> | |
32 | #include <asm/proto.h> | |
33 | #include <asm/kdebug.h> | |
34 | #include <asm-generic/sections.h> | |
1da177e4 | 35 | |
66c58156 AK |
36 | /* Page fault error code bits */ |
37 | #define PF_PROT (1<<0) /* or no page found */ | |
38 | #define PF_WRITE (1<<1) | |
39 | #define PF_USER (1<<2) | |
40 | #define PF_RSVD (1<<3) | |
41 | #define PF_INSTR (1<<4) | |
42 | ||
1bd858a5 AK |
43 | #ifdef CONFIG_KPROBES |
44 | ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain); | |
45 | ||
46 | /* Hook to register for page fault notifications */ | |
47 | int register_page_fault_notifier(struct notifier_block *nb) | |
48 | { | |
49 | vmalloc_sync_all(); | |
50 | return atomic_notifier_chain_register(¬ify_page_fault_chain, nb); | |
51 | } | |
52 | ||
53 | int unregister_page_fault_notifier(struct notifier_block *nb) | |
54 | { | |
55 | return atomic_notifier_chain_unregister(¬ify_page_fault_chain, nb); | |
56 | } | |
57 | ||
58 | static inline int notify_page_fault(enum die_val val, const char *str, | |
59 | struct pt_regs *regs, long err, int trap, int sig) | |
60 | { | |
61 | struct die_args args = { | |
62 | .regs = regs, | |
63 | .str = str, | |
64 | .err = err, | |
65 | .trapnr = trap, | |
66 | .signr = sig | |
67 | }; | |
68 | return atomic_notifier_call_chain(¬ify_page_fault_chain, val, &args); | |
69 | } | |
70 | #else | |
71 | static inline int notify_page_fault(enum die_val val, const char *str, | |
72 | struct pt_regs *regs, long err, int trap, int sig) | |
73 | { | |
74 | return NOTIFY_DONE; | |
75 | } | |
76 | #endif | |
77 | ||
1da177e4 LT |
78 | void bust_spinlocks(int yes) |
79 | { | |
80 | int loglevel_save = console_loglevel; | |
81 | if (yes) { | |
82 | oops_in_progress = 1; | |
83 | } else { | |
84 | #ifdef CONFIG_VT | |
85 | unblank_screen(); | |
86 | #endif | |
87 | oops_in_progress = 0; | |
88 | /* | |
89 | * OK, the message is on the console. Now we call printk() | |
90 | * without oops_in_progress set so that printk will give klogd | |
91 | * a poke. Hold onto your hats... | |
92 | */ | |
93 | console_loglevel = 15; /* NMI oopser may have shut the console up */ | |
94 | printk(" "); | |
95 | console_loglevel = loglevel_save; | |
96 | } | |
97 | } | |
98 | ||
99 | /* Sometimes the CPU reports invalid exceptions on prefetch. | |
100 | Check that here and ignore. | |
101 | Opcode checker based on code by Richard Brunner */ | |
102 | static noinline int is_prefetch(struct pt_regs *regs, unsigned long addr, | |
103 | unsigned long error_code) | |
104 | { | |
f1290ec9 | 105 | unsigned char *instr; |
1da177e4 LT |
106 | int scan_more = 1; |
107 | int prefetch = 0; | |
f1290ec9 | 108 | unsigned char *max_instr; |
1da177e4 LT |
109 | |
110 | /* If it was a exec fault ignore */ | |
66c58156 | 111 | if (error_code & PF_INSTR) |
1da177e4 LT |
112 | return 0; |
113 | ||
f1290ec9 AK |
114 | instr = (unsigned char *)convert_rip_to_linear(current, regs); |
115 | max_instr = instr + 15; | |
1da177e4 | 116 | |
76381fee | 117 | if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) |
1da177e4 LT |
118 | return 0; |
119 | ||
120 | while (scan_more && instr < max_instr) { | |
121 | unsigned char opcode; | |
122 | unsigned char instr_hi; | |
123 | unsigned char instr_lo; | |
124 | ||
125 | if (__get_user(opcode, instr)) | |
126 | break; | |
127 | ||
128 | instr_hi = opcode & 0xf0; | |
129 | instr_lo = opcode & 0x0f; | |
130 | instr++; | |
131 | ||
132 | switch (instr_hi) { | |
133 | case 0x20: | |
134 | case 0x30: | |
135 | /* Values 0x26,0x2E,0x36,0x3E are valid x86 | |
136 | prefixes. In long mode, the CPU will signal | |
137 | invalid opcode if some of these prefixes are | |
138 | present so we will never get here anyway */ | |
139 | scan_more = ((instr_lo & 7) == 0x6); | |
140 | break; | |
141 | ||
142 | case 0x40: | |
143 | /* In AMD64 long mode, 0x40 to 0x4F are valid REX prefixes | |
144 | Need to figure out under what instruction mode the | |
145 | instruction was issued ... */ | |
146 | /* Could check the LDT for lm, but for now it's good | |
147 | enough to assume that long mode only uses well known | |
148 | segments or kernel. */ | |
76381fee | 149 | scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS); |
1da177e4 LT |
150 | break; |
151 | ||
152 | case 0x60: | |
153 | /* 0x64 thru 0x67 are valid prefixes in all modes. */ | |
154 | scan_more = (instr_lo & 0xC) == 0x4; | |
155 | break; | |
156 | case 0xF0: | |
157 | /* 0xF0, 0xF2, and 0xF3 are valid prefixes in all modes. */ | |
158 | scan_more = !instr_lo || (instr_lo>>1) == 1; | |
159 | break; | |
160 | case 0x00: | |
161 | /* Prefetch instruction is 0x0F0D or 0x0F18 */ | |
162 | scan_more = 0; | |
163 | if (__get_user(opcode, instr)) | |
164 | break; | |
165 | prefetch = (instr_lo == 0xF) && | |
166 | (opcode == 0x0D || opcode == 0x18); | |
167 | break; | |
168 | default: | |
169 | scan_more = 0; | |
170 | break; | |
171 | } | |
172 | } | |
173 | return prefetch; | |
174 | } | |
175 | ||
176 | static int bad_address(void *p) | |
177 | { | |
178 | unsigned long dummy; | |
179 | return __get_user(dummy, (unsigned long *)p); | |
180 | } | |
181 | ||
182 | void dump_pagetable(unsigned long address) | |
183 | { | |
184 | pgd_t *pgd; | |
185 | pud_t *pud; | |
186 | pmd_t *pmd; | |
187 | pte_t *pte; | |
188 | ||
189 | asm("movq %%cr3,%0" : "=r" (pgd)); | |
190 | ||
191 | pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK); | |
192 | pgd += pgd_index(address); | |
1da177e4 | 193 | if (bad_address(pgd)) goto bad; |
d646bce4 | 194 | printk("PGD %lx ", pgd_val(*pgd)); |
1da177e4 LT |
195 | if (!pgd_present(*pgd)) goto ret; |
196 | ||
d2ae5b5f | 197 | pud = pud_offset(pgd, address); |
1da177e4 LT |
198 | if (bad_address(pud)) goto bad; |
199 | printk("PUD %lx ", pud_val(*pud)); | |
200 | if (!pud_present(*pud)) goto ret; | |
201 | ||
202 | pmd = pmd_offset(pud, address); | |
203 | if (bad_address(pmd)) goto bad; | |
204 | printk("PMD %lx ", pmd_val(*pmd)); | |
205 | if (!pmd_present(*pmd)) goto ret; | |
206 | ||
207 | pte = pte_offset_kernel(pmd, address); | |
208 | if (bad_address(pte)) goto bad; | |
209 | printk("PTE %lx", pte_val(*pte)); | |
210 | ret: | |
211 | printk("\n"); | |
212 | return; | |
213 | bad: | |
214 | printk("BAD\n"); | |
215 | } | |
216 | ||
217 | static const char errata93_warning[] = | |
218 | KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n" | |
219 | KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n" | |
220 | KERN_ERR "******* Please consider a BIOS update.\n" | |
221 | KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n"; | |
222 | ||
223 | /* Workaround for K8 erratum #93 & buggy BIOS. | |
224 | BIOS SMM functions are required to use a specific workaround | |
225 | to avoid corruption of the 64bit RIP register on C stepping K8. | |
226 | A lot of BIOS that didn't get tested properly miss this. | |
227 | The OS sees this as a page fault with the upper 32bits of RIP cleared. | |
228 | Try to work around it here. | |
229 | Note we only handle faults in kernel here. */ | |
230 | ||
231 | static int is_errata93(struct pt_regs *regs, unsigned long address) | |
232 | { | |
233 | static int warned; | |
234 | if (address != regs->rip) | |
235 | return 0; | |
236 | if ((address >> 32) != 0) | |
237 | return 0; | |
238 | address |= 0xffffffffUL << 32; | |
239 | if ((address >= (u64)_stext && address <= (u64)_etext) || | |
240 | (address >= MODULES_VADDR && address <= MODULES_END)) { | |
241 | if (!warned) { | |
242 | printk(errata93_warning); | |
243 | warned = 1; | |
244 | } | |
245 | regs->rip = address; | |
246 | return 1; | |
247 | } | |
248 | return 0; | |
249 | } | |
250 | ||
251 | int unhandled_signal(struct task_struct *tsk, int sig) | |
252 | { | |
253 | if (tsk->pid == 1) | |
254 | return 1; | |
5e5ec104 | 255 | if (tsk->ptrace & PT_PTRACED) |
1da177e4 LT |
256 | return 0; |
257 | return (tsk->sighand->action[sig-1].sa.sa_handler == SIG_IGN) || | |
258 | (tsk->sighand->action[sig-1].sa.sa_handler == SIG_DFL); | |
259 | } | |
260 | ||
261 | static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs, | |
262 | unsigned long error_code) | |
263 | { | |
1209140c | 264 | unsigned long flags = oops_begin(); |
6e3f3617 | 265 | struct task_struct *tsk; |
1209140c | 266 | |
1da177e4 LT |
267 | printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", |
268 | current->comm, address); | |
269 | dump_pagetable(address); | |
6e3f3617 JB |
270 | tsk = current; |
271 | tsk->thread.cr2 = address; | |
272 | tsk->thread.trap_no = 14; | |
273 | tsk->thread.error_code = error_code; | |
1da177e4 | 274 | __die("Bad pagetable", regs, error_code); |
1209140c | 275 | oops_end(flags); |
1da177e4 LT |
276 | do_exit(SIGKILL); |
277 | } | |
278 | ||
279 | /* | |
f95190b2 | 280 | * Handle a fault on the vmalloc area |
3b9ba4d5 AK |
281 | * |
282 | * This assumes no large pages in there. | |
1da177e4 LT |
283 | */ |
284 | static int vmalloc_fault(unsigned long address) | |
285 | { | |
286 | pgd_t *pgd, *pgd_ref; | |
287 | pud_t *pud, *pud_ref; | |
288 | pmd_t *pmd, *pmd_ref; | |
289 | pte_t *pte, *pte_ref; | |
290 | ||
291 | /* Copy kernel mappings over when needed. This can also | |
292 | happen within a race in page table update. In the later | |
293 | case just flush. */ | |
294 | ||
295 | pgd = pgd_offset(current->mm ?: &init_mm, address); | |
296 | pgd_ref = pgd_offset_k(address); | |
297 | if (pgd_none(*pgd_ref)) | |
298 | return -1; | |
299 | if (pgd_none(*pgd)) | |
300 | set_pgd(pgd, *pgd_ref); | |
8c914cb7 | 301 | else |
46a82b2d | 302 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
1da177e4 LT |
303 | |
304 | /* Below here mismatches are bugs because these lower tables | |
305 | are shared */ | |
306 | ||
307 | pud = pud_offset(pgd, address); | |
308 | pud_ref = pud_offset(pgd_ref, address); | |
309 | if (pud_none(*pud_ref)) | |
310 | return -1; | |
46a82b2d | 311 | if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) |
1da177e4 LT |
312 | BUG(); |
313 | pmd = pmd_offset(pud, address); | |
314 | pmd_ref = pmd_offset(pud_ref, address); | |
315 | if (pmd_none(*pmd_ref)) | |
316 | return -1; | |
317 | if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) | |
318 | BUG(); | |
319 | pte_ref = pte_offset_kernel(pmd_ref, address); | |
320 | if (!pte_present(*pte_ref)) | |
321 | return -1; | |
322 | pte = pte_offset_kernel(pmd, address); | |
3b9ba4d5 AK |
323 | /* Don't use pte_page here, because the mappings can point |
324 | outside mem_map, and the NUMA hash lookup cannot handle | |
325 | that. */ | |
326 | if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) | |
1da177e4 | 327 | BUG(); |
1da177e4 LT |
328 | return 0; |
329 | } | |
330 | ||
331 | int page_fault_trace = 0; | |
332 | int exception_trace = 1; | |
333 | ||
334 | /* | |
335 | * This routine handles page faults. It determines the address, | |
336 | * and the problem, and then passes it off to one of the appropriate | |
337 | * routines. | |
1da177e4 | 338 | */ |
0f2fbdcb PP |
339 | asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, |
340 | unsigned long error_code) | |
1da177e4 LT |
341 | { |
342 | struct task_struct *tsk; | |
343 | struct mm_struct *mm; | |
344 | struct vm_area_struct * vma; | |
345 | unsigned long address; | |
346 | const struct exception_table_entry *fixup; | |
347 | int write; | |
1209140c | 348 | unsigned long flags; |
1da177e4 LT |
349 | siginfo_t info; |
350 | ||
a9ba9a3b AV |
351 | tsk = current; |
352 | mm = tsk->mm; | |
353 | prefetchw(&mm->mmap_sem); | |
354 | ||
1da177e4 LT |
355 | /* get the address */ |
356 | __asm__("movq %%cr2,%0":"=r" (address)); | |
1da177e4 | 357 | |
1da177e4 LT |
358 | info.si_code = SEGV_MAPERR; |
359 | ||
360 | ||
361 | /* | |
362 | * We fault-in kernel-space virtual memory on-demand. The | |
363 | * 'reference' page table is init_mm.pgd. | |
364 | * | |
365 | * NOTE! We MUST NOT take any locks for this case. We may | |
366 | * be in an interrupt or a critical region, and should | |
367 | * only copy the information from the master page table, | |
368 | * nothing more. | |
369 | * | |
370 | * This verifies that the fault happens in kernel space | |
371 | * (error_code & 4) == 0, and that the fault was not a | |
8b1bde93 | 372 | * protection error (error_code & 9) == 0. |
1da177e4 | 373 | */ |
84929801 | 374 | if (unlikely(address >= TASK_SIZE64)) { |
f95190b2 AK |
375 | /* |
376 | * Don't check for the module range here: its PML4 | |
377 | * is always initialized because it's shared with the main | |
378 | * kernel text. Only vmalloc may need PML4 syncups. | |
379 | */ | |
66c58156 | 380 | if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) && |
f95190b2 | 381 | ((address >= VMALLOC_START && address < VMALLOC_END))) { |
8c914cb7 JB |
382 | if (vmalloc_fault(address) >= 0) |
383 | return; | |
1da177e4 | 384 | } |
1bd858a5 | 385 | if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14, |
8c914cb7 JB |
386 | SIGSEGV) == NOTIFY_STOP) |
387 | return; | |
1da177e4 LT |
388 | /* |
389 | * Don't take the mm semaphore here. If we fixup a prefetch | |
390 | * fault we could otherwise deadlock. | |
391 | */ | |
392 | goto bad_area_nosemaphore; | |
393 | } | |
394 | ||
1bd858a5 | 395 | if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14, |
8c914cb7 JB |
396 | SIGSEGV) == NOTIFY_STOP) |
397 | return; | |
398 | ||
399 | if (likely(regs->eflags & X86_EFLAGS_IF)) | |
400 | local_irq_enable(); | |
401 | ||
402 | if (unlikely(page_fault_trace)) | |
403 | printk("pagefault rip:%lx rsp:%lx cs:%lu ss:%lu address %lx error %lx\n", | |
404 | regs->rip,regs->rsp,regs->cs,regs->ss,address,error_code); | |
405 | ||
66c58156 | 406 | if (unlikely(error_code & PF_RSVD)) |
1da177e4 LT |
407 | pgtable_bad(address, regs, error_code); |
408 | ||
409 | /* | |
410 | * If we're in an interrupt or have no user | |
411 | * context, we must not take the fault.. | |
412 | */ | |
413 | if (unlikely(in_atomic() || !mm)) | |
414 | goto bad_area_nosemaphore; | |
415 | ||
416 | again: | |
417 | /* When running in the kernel we expect faults to occur only to | |
418 | * addresses in user space. All other faults represent errors in the | |
419 | * kernel and should generate an OOPS. Unfortunatly, in the case of an | |
80f7228b | 420 | * erroneous fault occurring in a code path which already holds mmap_sem |
1da177e4 LT |
421 | * we will deadlock attempting to validate the fault against the |
422 | * address space. Luckily the kernel only validly references user | |
423 | * space from well defined areas of code, which are listed in the | |
424 | * exceptions table. | |
425 | * | |
426 | * As the vast majority of faults will be valid we will only perform | |
427 | * the source reference check when there is a possibilty of a deadlock. | |
428 | * Attempt to lock the address space, if we cannot we then validate the | |
429 | * source. If this is invalid we can skip the address space check, | |
430 | * thus avoiding the deadlock. | |
431 | */ | |
432 | if (!down_read_trylock(&mm->mmap_sem)) { | |
66c58156 | 433 | if ((error_code & PF_USER) == 0 && |
1da177e4 LT |
434 | !search_exception_tables(regs->rip)) |
435 | goto bad_area_nosemaphore; | |
436 | down_read(&mm->mmap_sem); | |
437 | } | |
438 | ||
439 | vma = find_vma(mm, address); | |
440 | if (!vma) | |
441 | goto bad_area; | |
442 | if (likely(vma->vm_start <= address)) | |
443 | goto good_area; | |
444 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
445 | goto bad_area; | |
446 | if (error_code & 4) { | |
03fdc2c2 CE |
447 | /* Allow userspace just enough access below the stack pointer |
448 | * to let the 'enter' instruction work. | |
449 | */ | |
450 | if (address + 65536 + 32 * sizeof(unsigned long) < regs->rsp) | |
1da177e4 LT |
451 | goto bad_area; |
452 | } | |
453 | if (expand_stack(vma, address)) | |
454 | goto bad_area; | |
455 | /* | |
456 | * Ok, we have a good vm_area for this memory access, so | |
457 | * we can handle it.. | |
458 | */ | |
459 | good_area: | |
460 | info.si_code = SEGV_ACCERR; | |
461 | write = 0; | |
66c58156 | 462 | switch (error_code & (PF_PROT|PF_WRITE)) { |
1da177e4 LT |
463 | default: /* 3: write, present */ |
464 | /* fall through */ | |
66c58156 | 465 | case PF_WRITE: /* write, not present */ |
1da177e4 LT |
466 | if (!(vma->vm_flags & VM_WRITE)) |
467 | goto bad_area; | |
468 | write++; | |
469 | break; | |
66c58156 | 470 | case PF_PROT: /* read, present */ |
1da177e4 | 471 | goto bad_area; |
66c58156 | 472 | case 0: /* read, not present */ |
1da177e4 LT |
473 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) |
474 | goto bad_area; | |
475 | } | |
476 | ||
477 | /* | |
478 | * If for any reason at all we couldn't handle the fault, | |
479 | * make sure we exit gracefully rather than endlessly redo | |
480 | * the fault. | |
481 | */ | |
482 | switch (handle_mm_fault(mm, vma, address, write)) { | |
96800216 | 483 | case VM_FAULT_MINOR: |
1da177e4 LT |
484 | tsk->min_flt++; |
485 | break; | |
96800216 | 486 | case VM_FAULT_MAJOR: |
1da177e4 LT |
487 | tsk->maj_flt++; |
488 | break; | |
96800216 | 489 | case VM_FAULT_SIGBUS: |
1da177e4 LT |
490 | goto do_sigbus; |
491 | default: | |
492 | goto out_of_memory; | |
493 | } | |
494 | ||
495 | up_read(&mm->mmap_sem); | |
496 | return; | |
497 | ||
498 | /* | |
499 | * Something tried to access memory that isn't in our memory map.. | |
500 | * Fix it, but check if it's kernel or user first.. | |
501 | */ | |
502 | bad_area: | |
503 | up_read(&mm->mmap_sem); | |
504 | ||
505 | bad_area_nosemaphore: | |
1da177e4 | 506 | /* User mode accesses just cause a SIGSEGV */ |
66c58156 | 507 | if (error_code & PF_USER) { |
1da177e4 LT |
508 | if (is_prefetch(regs, address, error_code)) |
509 | return; | |
510 | ||
511 | /* Work around K8 erratum #100 K8 in compat mode | |
512 | occasionally jumps to illegal addresses >4GB. We | |
513 | catch this here in the page fault handler because | |
514 | these addresses are not reachable. Just detect this | |
515 | case and return. Any code segment in LDT is | |
516 | compatibility mode. */ | |
517 | if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && | |
518 | (address >> 32)) | |
519 | return; | |
520 | ||
521 | if (exception_trace && unhandled_signal(tsk, SIGSEGV)) { | |
522 | printk( | |
523 | "%s%s[%d]: segfault at %016lx rip %016lx rsp %016lx error %lx\n", | |
524 | tsk->pid > 1 ? KERN_INFO : KERN_EMERG, | |
525 | tsk->comm, tsk->pid, address, regs->rip, | |
526 | regs->rsp, error_code); | |
527 | } | |
528 | ||
529 | tsk->thread.cr2 = address; | |
530 | /* Kernel addresses are always protection faults */ | |
531 | tsk->thread.error_code = error_code | (address >= TASK_SIZE); | |
532 | tsk->thread.trap_no = 14; | |
533 | info.si_signo = SIGSEGV; | |
534 | info.si_errno = 0; | |
535 | /* info.si_code has been set above */ | |
536 | info.si_addr = (void __user *)address; | |
537 | force_sig_info(SIGSEGV, &info, tsk); | |
538 | return; | |
539 | } | |
540 | ||
541 | no_context: | |
542 | ||
543 | /* Are we prepared to handle this kernel fault? */ | |
544 | fixup = search_exception_tables(regs->rip); | |
545 | if (fixup) { | |
546 | regs->rip = fixup->fixup; | |
547 | return; | |
548 | } | |
549 | ||
550 | /* | |
551 | * Hall of shame of CPU/BIOS bugs. | |
552 | */ | |
553 | ||
554 | if (is_prefetch(regs, address, error_code)) | |
555 | return; | |
556 | ||
557 | if (is_errata93(regs, address)) | |
558 | return; | |
559 | ||
560 | /* | |
561 | * Oops. The kernel tried to access some bad page. We'll have to | |
562 | * terminate things with extreme prejudice. | |
563 | */ | |
564 | ||
1209140c | 565 | flags = oops_begin(); |
1da177e4 LT |
566 | |
567 | if (address < PAGE_SIZE) | |
568 | printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference"); | |
569 | else | |
570 | printk(KERN_ALERT "Unable to handle kernel paging request"); | |
571 | printk(" at %016lx RIP: \n" KERN_ALERT,address); | |
572 | printk_address(regs->rip); | |
1da177e4 | 573 | dump_pagetable(address); |
6e3f3617 JB |
574 | tsk->thread.cr2 = address; |
575 | tsk->thread.trap_no = 14; | |
576 | tsk->thread.error_code = error_code; | |
1da177e4 LT |
577 | __die("Oops", regs, error_code); |
578 | /* Executive summary in case the body of the oops scrolled away */ | |
579 | printk(KERN_EMERG "CR2: %016lx\n", address); | |
1209140c | 580 | oops_end(flags); |
1da177e4 LT |
581 | do_exit(SIGKILL); |
582 | ||
583 | /* | |
584 | * We ran out of memory, or some other thing happened to us that made | |
585 | * us unable to handle the page fault gracefully. | |
586 | */ | |
587 | out_of_memory: | |
588 | up_read(&mm->mmap_sem); | |
1da177e4 LT |
589 | if (current->pid == 1) { |
590 | yield(); | |
591 | goto again; | |
592 | } | |
593 | printk("VM: killing process %s\n", tsk->comm); | |
594 | if (error_code & 4) | |
595 | do_exit(SIGKILL); | |
596 | goto no_context; | |
597 | ||
598 | do_sigbus: | |
599 | up_read(&mm->mmap_sem); | |
600 | ||
601 | /* Kernel mode? Handle exceptions or die */ | |
66c58156 | 602 | if (!(error_code & PF_USER)) |
1da177e4 LT |
603 | goto no_context; |
604 | ||
605 | tsk->thread.cr2 = address; | |
606 | tsk->thread.error_code = error_code; | |
607 | tsk->thread.trap_no = 14; | |
608 | info.si_signo = SIGBUS; | |
609 | info.si_errno = 0; | |
610 | info.si_code = BUS_ADRERR; | |
611 | info.si_addr = (void __user *)address; | |
612 | force_sig_info(SIGBUS, &info, tsk); | |
613 | return; | |
614 | } | |
9e43e1b7 | 615 | |
8c914cb7 JB |
616 | DEFINE_SPINLOCK(pgd_lock); |
617 | struct page *pgd_list; | |
618 | ||
619 | void vmalloc_sync_all(void) | |
620 | { | |
621 | /* Note that races in the updates of insync and start aren't | |
622 | problematic: | |
623 | insync can only get set bits added, and updates to start are only | |
624 | improving performance (without affecting correctness if undone). */ | |
625 | static DECLARE_BITMAP(insync, PTRS_PER_PGD); | |
626 | static unsigned long start = VMALLOC_START & PGDIR_MASK; | |
627 | unsigned long address; | |
628 | ||
629 | for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) { | |
630 | if (!test_bit(pgd_index(address), insync)) { | |
631 | const pgd_t *pgd_ref = pgd_offset_k(address); | |
632 | struct page *page; | |
633 | ||
634 | if (pgd_none(*pgd_ref)) | |
635 | continue; | |
636 | spin_lock(&pgd_lock); | |
637 | for (page = pgd_list; page; | |
638 | page = (struct page *)page->index) { | |
639 | pgd_t *pgd; | |
640 | pgd = (pgd_t *)page_address(page) + pgd_index(address); | |
641 | if (pgd_none(*pgd)) | |
642 | set_pgd(pgd, *pgd_ref); | |
643 | else | |
46a82b2d | 644 | BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); |
8c914cb7 JB |
645 | } |
646 | spin_unlock(&pgd_lock); | |
647 | set_bit(pgd_index(address), insync); | |
648 | } | |
649 | if (address == start) | |
650 | start = address + PGDIR_SIZE; | |
651 | } | |
652 | /* Check that there is no need to do the same for the modules area. */ | |
653 | BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL)); | |
654 | BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) == | |
655 | (__START_KERNEL & PGDIR_MASK))); | |
656 | } | |
657 | ||
9e43e1b7 AK |
658 | static int __init enable_pagefaulttrace(char *str) |
659 | { | |
660 | page_fault_trace = 1; | |
9b41046c | 661 | return 1; |
9e43e1b7 AK |
662 | } |
663 | __setup("pagefaulttrace", enable_pagefaulttrace); |