Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/arm/mm/fault.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * Modifications for ARM processor (c) 1995-2004 Russell King | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License version 2 as | |
9 | * published by the Free Software Foundation. | |
10 | */ | |
1da177e4 LT |
11 | #include <linux/module.h> |
12 | #include <linux/signal.h> | |
1da177e4 | 13 | #include <linux/mm.h> |
67306da6 | 14 | #include <linux/hardirq.h> |
1da177e4 | 15 | #include <linux/init.h> |
25ce1dd7 | 16 | #include <linux/kprobes.h> |
33fa9b13 | 17 | #include <linux/uaccess.h> |
252d4c27 | 18 | #include <linux/page-flags.h> |
1da177e4 LT |
19 | |
20 | #include <asm/system.h> | |
21 | #include <asm/pgtable.h> | |
22 | #include <asm/tlbflush.h> | |
1da177e4 LT |
23 | |
24 | #include "fault.h" | |
25 | ||
09529f7a | 26 | #ifdef CONFIG_MMU |
25ce1dd7 NP |
27 | |
28 | #ifdef CONFIG_KPROBES | |
29 | static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) | |
30 | { | |
31 | int ret = 0; | |
32 | ||
33 | if (!user_mode(regs)) { | |
34 | /* kprobe_running() needs smp_processor_id() */ | |
35 | preempt_disable(); | |
36 | if (kprobe_running() && kprobe_fault_handler(regs, fsr)) | |
37 | ret = 1; | |
38 | preempt_enable(); | |
39 | } | |
40 | ||
41 | return ret; | |
42 | } | |
43 | #else | |
44 | static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) | |
45 | { | |
46 | return 0; | |
47 | } | |
48 | #endif | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * This is useful to dump out the page tables associated with | |
52 | * 'addr' in mm 'mm'. | |
53 | */ | |
54 | void show_pte(struct mm_struct *mm, unsigned long addr) | |
55 | { | |
56 | pgd_t *pgd; | |
57 | ||
58 | if (!mm) | |
59 | mm = &init_mm; | |
60 | ||
61 | printk(KERN_ALERT "pgd = %p\n", mm->pgd); | |
62 | pgd = pgd_offset(mm, addr); | |
63 | printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd)); | |
64 | ||
65 | do { | |
66 | pmd_t *pmd; | |
67 | pte_t *pte; | |
68 | ||
69 | if (pgd_none(*pgd)) | |
70 | break; | |
71 | ||
72 | if (pgd_bad(*pgd)) { | |
73 | printk("(bad)"); | |
74 | break; | |
75 | } | |
76 | ||
77 | pmd = pmd_offset(pgd, addr); | |
da46c79a NP |
78 | if (PTRS_PER_PMD != 1) |
79 | printk(", *pmd=%08lx", pmd_val(*pmd)); | |
1da177e4 LT |
80 | |
81 | if (pmd_none(*pmd)) | |
82 | break; | |
83 | ||
84 | if (pmd_bad(*pmd)) { | |
85 | printk("(bad)"); | |
86 | break; | |
87 | } | |
88 | ||
1da177e4 | 89 | /* We must not map this if we have highmem enabled */ |
252d4c27 NP |
90 | if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT))) |
91 | break; | |
92 | ||
1da177e4 LT |
93 | pte = pte_offset_map(pmd, addr); |
94 | printk(", *pte=%08lx", pte_val(*pte)); | |
95 | printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE])); | |
96 | pte_unmap(pte); | |
1da177e4 LT |
97 | } while(0); |
98 | ||
99 | printk("\n"); | |
100 | } | |
09529f7a CM |
101 | #else /* CONFIG_MMU */ |
102 | void show_pte(struct mm_struct *mm, unsigned long addr) | |
103 | { } | |
104 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
105 | |
106 | /* | |
107 | * Oops. The kernel tried to access some page that wasn't present. | |
108 | */ | |
109 | static void | |
110 | __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, | |
111 | struct pt_regs *regs) | |
112 | { | |
113 | /* | |
114 | * Are we prepared to handle this kernel fault? | |
115 | */ | |
116 | if (fixup_exception(regs)) | |
117 | return; | |
118 | ||
119 | /* | |
120 | * No handler, we'll have to terminate things with extreme prejudice. | |
121 | */ | |
122 | bust_spinlocks(1); | |
123 | printk(KERN_ALERT | |
124 | "Unable to handle kernel %s at virtual address %08lx\n", | |
125 | (addr < PAGE_SIZE) ? "NULL pointer dereference" : | |
126 | "paging request", addr); | |
127 | ||
128 | show_pte(mm, addr); | |
129 | die("Oops", regs, fsr); | |
130 | bust_spinlocks(0); | |
131 | do_exit(SIGKILL); | |
132 | } | |
133 | ||
134 | /* | |
135 | * Something tried to access memory that isn't in our memory map.. | |
136 | * User mode accesses just cause a SIGSEGV | |
137 | */ | |
138 | static void | |
139 | __do_user_fault(struct task_struct *tsk, unsigned long addr, | |
2d137c24 AM |
140 | unsigned int fsr, unsigned int sig, int code, |
141 | struct pt_regs *regs) | |
1da177e4 LT |
142 | { |
143 | struct siginfo si; | |
144 | ||
145 | #ifdef CONFIG_DEBUG_USER | |
146 | if (user_debug & UDBG_SEGV) { | |
2d137c24 AM |
147 | printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n", |
148 | tsk->comm, sig, addr, fsr); | |
1da177e4 LT |
149 | show_pte(tsk->mm, addr); |
150 | show_regs(regs); | |
151 | } | |
152 | #endif | |
153 | ||
154 | tsk->thread.address = addr; | |
155 | tsk->thread.error_code = fsr; | |
156 | tsk->thread.trap_no = 14; | |
2d137c24 | 157 | si.si_signo = sig; |
1da177e4 LT |
158 | si.si_errno = 0; |
159 | si.si_code = code; | |
160 | si.si_addr = (void __user *)addr; | |
2d137c24 | 161 | force_sig_info(sig, &si, tsk); |
1da177e4 LT |
162 | } |
163 | ||
e5beac37 | 164 | void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
1da177e4 | 165 | { |
e5beac37 RK |
166 | struct task_struct *tsk = current; |
167 | struct mm_struct *mm = tsk->active_mm; | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * If we are in kernel mode at this point, we | |
171 | * have no context to handle this fault with. | |
172 | */ | |
173 | if (user_mode(regs)) | |
2d137c24 | 174 | __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs); |
1da177e4 LT |
175 | else |
176 | __do_kernel_fault(mm, addr, fsr, regs); | |
177 | } | |
178 | ||
09529f7a | 179 | #ifdef CONFIG_MMU |
5c72fc5c NP |
180 | #define VM_FAULT_BADMAP 0x010000 |
181 | #define VM_FAULT_BADACCESS 0x020000 | |
1da177e4 LT |
182 | |
183 | static int | |
184 | __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, | |
185 | struct task_struct *tsk) | |
186 | { | |
187 | struct vm_area_struct *vma; | |
188 | int fault, mask; | |
189 | ||
190 | vma = find_vma(mm, addr); | |
191 | fault = VM_FAULT_BADMAP; | |
192 | if (!vma) | |
193 | goto out; | |
194 | if (vma->vm_start > addr) | |
195 | goto check_stack; | |
196 | ||
197 | /* | |
198 | * Ok, we have a good vm_area for this | |
199 | * memory access, so we can handle it. | |
200 | */ | |
201 | good_area: | |
202 | if (fsr & (1 << 11)) /* write? */ | |
203 | mask = VM_WRITE; | |
204 | else | |
df67b3da | 205 | mask = VM_READ|VM_EXEC|VM_WRITE; |
1da177e4 LT |
206 | |
207 | fault = VM_FAULT_BADACCESS; | |
208 | if (!(vma->vm_flags & mask)) | |
209 | goto out; | |
210 | ||
211 | /* | |
212 | * If for any reason at all we couldn't handle | |
213 | * the fault, make sure we exit gracefully rather | |
214 | * than endlessly redo the fault. | |
215 | */ | |
216 | survive: | |
d06063cc | 217 | fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0); |
83c54070 NP |
218 | if (unlikely(fault & VM_FAULT_ERROR)) { |
219 | if (fault & VM_FAULT_OOM) | |
220 | goto out_of_memory; | |
221 | else if (fault & VM_FAULT_SIGBUS) | |
222 | return fault; | |
223 | BUG(); | |
224 | } | |
225 | if (fault & VM_FAULT_MAJOR) | |
1da177e4 | 226 | tsk->maj_flt++; |
83c54070 | 227 | else |
1da177e4 | 228 | tsk->min_flt++; |
83c54070 | 229 | return fault; |
1da177e4 | 230 | |
83c54070 | 231 | out_of_memory: |
b460cbc5 | 232 | if (!is_global_init(tsk)) |
1da177e4 LT |
233 | goto out; |
234 | ||
235 | /* | |
2d137c24 | 236 | * If we are out of memory for pid1, sleep for a while and retry |
1da177e4 | 237 | */ |
2d137c24 | 238 | up_read(&mm->mmap_sem); |
1da177e4 | 239 | yield(); |
2d137c24 | 240 | down_read(&mm->mmap_sem); |
1da177e4 LT |
241 | goto survive; |
242 | ||
243 | check_stack: | |
244 | if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr)) | |
245 | goto good_area; | |
246 | out: | |
247 | return fault; | |
248 | } | |
249 | ||
785d3cd2 | 250 | static int __kprobes |
1da177e4 LT |
251 | do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
252 | { | |
253 | struct task_struct *tsk; | |
254 | struct mm_struct *mm; | |
2d137c24 | 255 | int fault, sig, code; |
1da177e4 | 256 | |
25ce1dd7 NP |
257 | if (notify_page_fault(regs, fsr)) |
258 | return 0; | |
259 | ||
1da177e4 LT |
260 | tsk = current; |
261 | mm = tsk->mm; | |
262 | ||
263 | /* | |
264 | * If we're in an interrupt or have no user | |
265 | * context, we must not take the fault.. | |
266 | */ | |
6edaf68a | 267 | if (in_atomic() || !mm) |
1da177e4 LT |
268 | goto no_context; |
269 | ||
840ff6a4 RK |
270 | /* |
271 | * As per x86, we may deadlock here. However, since the kernel only | |
272 | * validly references user space from well defined areas of the code, | |
273 | * we can bug out early if this is from code which shouldn't. | |
274 | */ | |
275 | if (!down_read_trylock(&mm->mmap_sem)) { | |
276 | if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc)) | |
277 | goto no_context; | |
278 | down_read(&mm->mmap_sem); | |
279 | } | |
280 | ||
1da177e4 LT |
281 | fault = __do_page_fault(mm, addr, fsr, tsk); |
282 | up_read(&mm->mmap_sem); | |
283 | ||
284 | /* | |
ff2afb9d | 285 | * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR |
1da177e4 | 286 | */ |
5c72fc5c | 287 | if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS)))) |
1da177e4 LT |
288 | return 0; |
289 | ||
1da177e4 LT |
290 | /* |
291 | * If we are in kernel mode at this point, we | |
292 | * have no context to handle this fault with. | |
293 | */ | |
294 | if (!user_mode(regs)) | |
295 | goto no_context; | |
296 | ||
83c54070 | 297 | if (fault & VM_FAULT_OOM) { |
1da177e4 | 298 | /* |
2d137c24 AM |
299 | * We ran out of memory, or some other thing |
300 | * happened to us that made us unable to handle | |
301 | * the page fault gracefully. | |
1da177e4 LT |
302 | */ |
303 | printk("VM: killing process %s\n", tsk->comm); | |
dcca2bde | 304 | do_group_exit(SIGKILL); |
2d137c24 | 305 | return 0; |
83c54070 NP |
306 | } |
307 | if (fault & VM_FAULT_SIGBUS) { | |
2d137c24 AM |
308 | /* |
309 | * We had some memory, but were unable to | |
310 | * successfully fix up this page fault. | |
311 | */ | |
312 | sig = SIGBUS; | |
313 | code = BUS_ADRERR; | |
83c54070 | 314 | } else { |
2d137c24 AM |
315 | /* |
316 | * Something tried to access memory that | |
317 | * isn't in our memory map.. | |
318 | */ | |
319 | sig = SIGSEGV; | |
320 | code = fault == VM_FAULT_BADACCESS ? | |
321 | SEGV_ACCERR : SEGV_MAPERR; | |
1da177e4 | 322 | } |
1da177e4 | 323 | |
2d137c24 AM |
324 | __do_user_fault(tsk, addr, fsr, sig, code, regs); |
325 | return 0; | |
1da177e4 LT |
326 | |
327 | no_context: | |
328 | __do_kernel_fault(mm, addr, fsr, regs); | |
329 | return 0; | |
330 | } | |
09529f7a CM |
331 | #else /* CONFIG_MMU */ |
332 | static int | |
333 | do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
334 | { | |
335 | return 0; | |
336 | } | |
337 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
338 | |
339 | /* | |
340 | * First Level Translation Fault Handler | |
341 | * | |
342 | * We enter here because the first level page table doesn't contain | |
343 | * a valid entry for the address. | |
344 | * | |
345 | * If the address is in kernel space (>= TASK_SIZE), then we are | |
346 | * probably faulting in the vmalloc() area. | |
347 | * | |
348 | * If the init_task's first level page tables contains the relevant | |
349 | * entry, we copy the it to this task. If not, we send the process | |
350 | * a signal, fixup the exception, or oops the kernel. | |
351 | * | |
352 | * NOTE! We MUST NOT take any locks for this case. We may be in an | |
353 | * interrupt or a critical region, and should only copy the information | |
354 | * from the master page table, nothing more. | |
355 | */ | |
09529f7a | 356 | #ifdef CONFIG_MMU |
785d3cd2 | 357 | static int __kprobes |
1da177e4 LT |
358 | do_translation_fault(unsigned long addr, unsigned int fsr, |
359 | struct pt_regs *regs) | |
360 | { | |
1da177e4 LT |
361 | unsigned int index; |
362 | pgd_t *pgd, *pgd_k; | |
363 | pmd_t *pmd, *pmd_k; | |
364 | ||
365 | if (addr < TASK_SIZE) | |
366 | return do_page_fault(addr, fsr, regs); | |
367 | ||
368 | index = pgd_index(addr); | |
369 | ||
370 | /* | |
371 | * FIXME: CP15 C1 is write only on ARMv3 architectures. | |
372 | */ | |
373 | pgd = cpu_get_pgd() + index; | |
374 | pgd_k = init_mm.pgd + index; | |
375 | ||
376 | if (pgd_none(*pgd_k)) | |
377 | goto bad_area; | |
378 | ||
379 | if (!pgd_present(*pgd)) | |
380 | set_pgd(pgd, *pgd_k); | |
381 | ||
382 | pmd_k = pmd_offset(pgd_k, addr); | |
383 | pmd = pmd_offset(pgd, addr); | |
384 | ||
385 | if (pmd_none(*pmd_k)) | |
386 | goto bad_area; | |
387 | ||
388 | copy_pmd(pmd, pmd_k); | |
389 | return 0; | |
390 | ||
391 | bad_area: | |
e5beac37 | 392 | do_bad_area(addr, fsr, regs); |
1da177e4 LT |
393 | return 0; |
394 | } | |
09529f7a CM |
395 | #else /* CONFIG_MMU */ |
396 | static int | |
397 | do_translation_fault(unsigned long addr, unsigned int fsr, | |
398 | struct pt_regs *regs) | |
399 | { | |
400 | return 0; | |
401 | } | |
402 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
403 | |
404 | /* | |
405 | * Some section permission faults need to be handled gracefully. | |
406 | * They can happen due to a __{get,put}_user during an oops. | |
407 | */ | |
408 | static int | |
409 | do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
410 | { | |
e5beac37 | 411 | do_bad_area(addr, fsr, regs); |
1da177e4 LT |
412 | return 0; |
413 | } | |
414 | ||
415 | /* | |
416 | * This abort handler always returns "fault". | |
417 | */ | |
418 | static int | |
419 | do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
420 | { | |
421 | return 1; | |
422 | } | |
423 | ||
424 | static struct fsr_info { | |
425 | int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs); | |
426 | int sig; | |
cfb0810e | 427 | int code; |
1da177e4 LT |
428 | const char *name; |
429 | } fsr_info[] = { | |
430 | /* | |
431 | * The following are the standard ARMv3 and ARMv4 aborts. ARMv5 | |
432 | * defines these to be "precise" aborts. | |
433 | */ | |
cfb0810e RK |
434 | { do_bad, SIGSEGV, 0, "vector exception" }, |
435 | { do_bad, SIGILL, BUS_ADRALN, "alignment exception" }, | |
436 | { do_bad, SIGKILL, 0, "terminal exception" }, | |
437 | { do_bad, SIGILL, BUS_ADRALN, "alignment exception" }, | |
438 | { do_bad, SIGBUS, 0, "external abort on linefetch" }, | |
439 | { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" }, | |
440 | { do_bad, SIGBUS, 0, "external abort on linefetch" }, | |
441 | { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" }, | |
442 | { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, | |
443 | { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" }, | |
444 | { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, | |
445 | { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" }, | |
446 | { do_bad, SIGBUS, 0, "external abort on translation" }, | |
447 | { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" }, | |
448 | { do_bad, SIGBUS, 0, "external abort on translation" }, | |
449 | { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" }, | |
1da177e4 LT |
450 | /* |
451 | * The following are "imprecise" aborts, which are signalled by bit | |
452 | * 10 of the FSR, and may not be recoverable. These are only | |
453 | * supported if the CPU abort handler supports bit 10. | |
454 | */ | |
cfb0810e RK |
455 | { do_bad, SIGBUS, 0, "unknown 16" }, |
456 | { do_bad, SIGBUS, 0, "unknown 17" }, | |
457 | { do_bad, SIGBUS, 0, "unknown 18" }, | |
458 | { do_bad, SIGBUS, 0, "unknown 19" }, | |
459 | { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */ | |
460 | { do_bad, SIGBUS, 0, "unknown 21" }, | |
461 | { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */ | |
462 | { do_bad, SIGBUS, 0, "unknown 23" }, | |
463 | { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */ | |
464 | { do_bad, SIGBUS, 0, "unknown 25" }, | |
465 | { do_bad, SIGBUS, 0, "unknown 26" }, | |
466 | { do_bad, SIGBUS, 0, "unknown 27" }, | |
467 | { do_bad, SIGBUS, 0, "unknown 28" }, | |
468 | { do_bad, SIGBUS, 0, "unknown 29" }, | |
469 | { do_bad, SIGBUS, 0, "unknown 30" }, | |
470 | { do_bad, SIGBUS, 0, "unknown 31" } | |
1da177e4 LT |
471 | }; |
472 | ||
473 | void __init | |
474 | hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), | |
475 | int sig, const char *name) | |
476 | { | |
477 | if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) { | |
478 | fsr_info[nr].fn = fn; | |
479 | fsr_info[nr].sig = sig; | |
480 | fsr_info[nr].name = name; | |
481 | } | |
482 | } | |
483 | ||
484 | /* | |
485 | * Dispatch a data abort to the relevant handler. | |
486 | */ | |
7ab3f8d5 | 487 | asmlinkage void __exception |
1da177e4 LT |
488 | do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
489 | { | |
490 | const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6); | |
cfb0810e | 491 | struct siginfo info; |
1da177e4 LT |
492 | |
493 | if (!inf->fn(addr, fsr, regs)) | |
494 | return; | |
495 | ||
496 | printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n", | |
497 | inf->name, fsr, addr); | |
cfb0810e RK |
498 | |
499 | info.si_signo = inf->sig; | |
500 | info.si_errno = 0; | |
501 | info.si_code = inf->code; | |
502 | info.si_addr = (void __user *)addr; | |
1eeb66a1 | 503 | arm_notify_die("", regs, &info, fsr, 0); |
1da177e4 LT |
504 | } |
505 | ||
7ab3f8d5 | 506 | asmlinkage void __exception |
1da177e4 LT |
507 | do_PrefetchAbort(unsigned long addr, struct pt_regs *regs) |
508 | { | |
509 | do_translation_fault(addr, 0, regs); | |
510 | } | |
511 |