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> |
412bb0a6 | 19 | #include <linux/sched.h> |
1da177e4 LT |
20 | |
21 | #include <asm/system.h> | |
22 | #include <asm/pgtable.h> | |
23 | #include <asm/tlbflush.h> | |
1da177e4 LT |
24 | |
25 | #include "fault.h" | |
26 | ||
09529f7a | 27 | #ifdef CONFIG_MMU |
25ce1dd7 NP |
28 | |
29 | #ifdef CONFIG_KPROBES | |
30 | static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) | |
31 | { | |
32 | int ret = 0; | |
33 | ||
34 | if (!user_mode(regs)) { | |
35 | /* kprobe_running() needs smp_processor_id() */ | |
36 | preempt_disable(); | |
37 | if (kprobe_running() && kprobe_fault_handler(regs, fsr)) | |
38 | ret = 1; | |
39 | preempt_enable(); | |
40 | } | |
41 | ||
42 | return ret; | |
43 | } | |
44 | #else | |
45 | static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr) | |
46 | { | |
47 | return 0; | |
48 | } | |
49 | #endif | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * This is useful to dump out the page tables associated with | |
53 | * 'addr' in mm 'mm'. | |
54 | */ | |
55 | void show_pte(struct mm_struct *mm, unsigned long addr) | |
56 | { | |
57 | pgd_t *pgd; | |
58 | ||
59 | if (!mm) | |
60 | mm = &init_mm; | |
61 | ||
62 | printk(KERN_ALERT "pgd = %p\n", mm->pgd); | |
63 | pgd = pgd_offset(mm, addr); | |
64 | printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd)); | |
65 | ||
66 | do { | |
67 | pmd_t *pmd; | |
68 | pte_t *pte; | |
69 | ||
70 | if (pgd_none(*pgd)) | |
71 | break; | |
72 | ||
73 | if (pgd_bad(*pgd)) { | |
74 | printk("(bad)"); | |
75 | break; | |
76 | } | |
77 | ||
78 | pmd = pmd_offset(pgd, addr); | |
da46c79a NP |
79 | if (PTRS_PER_PMD != 1) |
80 | printk(", *pmd=%08lx", pmd_val(*pmd)); | |
1da177e4 LT |
81 | |
82 | if (pmd_none(*pmd)) | |
83 | break; | |
84 | ||
85 | if (pmd_bad(*pmd)) { | |
86 | printk("(bad)"); | |
87 | break; | |
88 | } | |
89 | ||
1da177e4 | 90 | /* We must not map this if we have highmem enabled */ |
252d4c27 NP |
91 | if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT))) |
92 | break; | |
93 | ||
1da177e4 LT |
94 | pte = pte_offset_map(pmd, addr); |
95 | printk(", *pte=%08lx", pte_val(*pte)); | |
96 | printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE])); | |
97 | pte_unmap(pte); | |
1da177e4 LT |
98 | } while(0); |
99 | ||
100 | printk("\n"); | |
101 | } | |
09529f7a CM |
102 | #else /* CONFIG_MMU */ |
103 | void show_pte(struct mm_struct *mm, unsigned long addr) | |
104 | { } | |
105 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
106 | |
107 | /* | |
108 | * Oops. The kernel tried to access some page that wasn't present. | |
109 | */ | |
110 | static void | |
111 | __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, | |
112 | struct pt_regs *regs) | |
113 | { | |
114 | /* | |
115 | * Are we prepared to handle this kernel fault? | |
116 | */ | |
117 | if (fixup_exception(regs)) | |
118 | return; | |
119 | ||
120 | /* | |
121 | * No handler, we'll have to terminate things with extreme prejudice. | |
122 | */ | |
123 | bust_spinlocks(1); | |
124 | printk(KERN_ALERT | |
125 | "Unable to handle kernel %s at virtual address %08lx\n", | |
126 | (addr < PAGE_SIZE) ? "NULL pointer dereference" : | |
127 | "paging request", addr); | |
128 | ||
129 | show_pte(mm, addr); | |
130 | die("Oops", regs, fsr); | |
131 | bust_spinlocks(0); | |
132 | do_exit(SIGKILL); | |
133 | } | |
134 | ||
135 | /* | |
136 | * Something tried to access memory that isn't in our memory map.. | |
137 | * User mode accesses just cause a SIGSEGV | |
138 | */ | |
139 | static void | |
140 | __do_user_fault(struct task_struct *tsk, unsigned long addr, | |
2d137c24 | 141 | unsigned int fsr, unsigned int sig, int code, |
142 | struct pt_regs *regs) | |
1da177e4 LT |
143 | { |
144 | struct siginfo si; | |
145 | ||
146 | #ifdef CONFIG_DEBUG_USER | |
147 | if (user_debug & UDBG_SEGV) { | |
2d137c24 | 148 | printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n", |
149 | tsk->comm, sig, addr, fsr); | |
1da177e4 LT |
150 | show_pte(tsk->mm, addr); |
151 | show_regs(regs); | |
152 | } | |
153 | #endif | |
154 | ||
155 | tsk->thread.address = addr; | |
156 | tsk->thread.error_code = fsr; | |
157 | tsk->thread.trap_no = 14; | |
2d137c24 | 158 | si.si_signo = sig; |
1da177e4 LT |
159 | si.si_errno = 0; |
160 | si.si_code = code; | |
161 | si.si_addr = (void __user *)addr; | |
2d137c24 | 162 | force_sig_info(sig, &si, tsk); |
1da177e4 LT |
163 | } |
164 | ||
e5beac37 | 165 | void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
1da177e4 | 166 | { |
e5beac37 RK |
167 | struct task_struct *tsk = current; |
168 | struct mm_struct *mm = tsk->active_mm; | |
169 | ||
1da177e4 LT |
170 | /* |
171 | * If we are in kernel mode at this point, we | |
172 | * have no context to handle this fault with. | |
173 | */ | |
174 | if (user_mode(regs)) | |
2d137c24 | 175 | __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs); |
1da177e4 LT |
176 | else |
177 | __do_kernel_fault(mm, addr, fsr, regs); | |
178 | } | |
179 | ||
09529f7a | 180 | #ifdef CONFIG_MMU |
5c72fc5c NP |
181 | #define VM_FAULT_BADMAP 0x010000 |
182 | #define VM_FAULT_BADACCESS 0x020000 | |
1da177e4 LT |
183 | |
184 | static int | |
185 | __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr, | |
186 | struct task_struct *tsk) | |
187 | { | |
188 | struct vm_area_struct *vma; | |
189 | int fault, mask; | |
190 | ||
191 | vma = find_vma(mm, addr); | |
192 | fault = VM_FAULT_BADMAP; | |
193 | if (!vma) | |
194 | goto out; | |
195 | if (vma->vm_start > addr) | |
196 | goto check_stack; | |
197 | ||
198 | /* | |
199 | * Ok, we have a good vm_area for this | |
200 | * memory access, so we can handle it. | |
201 | */ | |
202 | good_area: | |
203 | if (fsr & (1 << 11)) /* write? */ | |
204 | mask = VM_WRITE; | |
205 | else | |
df67b3da | 206 | mask = VM_READ|VM_EXEC|VM_WRITE; |
1da177e4 LT |
207 | |
208 | fault = VM_FAULT_BADACCESS; | |
209 | if (!(vma->vm_flags & mask)) | |
210 | goto out; | |
211 | ||
212 | /* | |
213 | * If for any reason at all we couldn't handle | |
214 | * the fault, make sure we exit gracefully rather | |
215 | * than endlessly redo the fault. | |
216 | */ | |
217 | survive: | |
d06063cc | 218 | fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0); |
83c54070 NP |
219 | if (unlikely(fault & VM_FAULT_ERROR)) { |
220 | if (fault & VM_FAULT_OOM) | |
221 | goto out_of_memory; | |
222 | else if (fault & VM_FAULT_SIGBUS) | |
223 | return fault; | |
224 | BUG(); | |
225 | } | |
226 | if (fault & VM_FAULT_MAJOR) | |
1da177e4 | 227 | tsk->maj_flt++; |
83c54070 | 228 | else |
1da177e4 | 229 | tsk->min_flt++; |
83c54070 | 230 | return fault; |
1da177e4 | 231 | |
83c54070 | 232 | out_of_memory: |
b460cbc5 | 233 | if (!is_global_init(tsk)) |
1da177e4 LT |
234 | goto out; |
235 | ||
236 | /* | |
2d137c24 | 237 | * If we are out of memory for pid1, sleep for a while and retry |
1da177e4 | 238 | */ |
2d137c24 | 239 | up_read(&mm->mmap_sem); |
1da177e4 | 240 | yield(); |
2d137c24 | 241 | down_read(&mm->mmap_sem); |
1da177e4 LT |
242 | goto survive; |
243 | ||
244 | check_stack: | |
245 | if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr)) | |
246 | goto good_area; | |
247 | out: | |
248 | return fault; | |
249 | } | |
250 | ||
785d3cd2 | 251 | static int __kprobes |
1da177e4 LT |
252 | do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
253 | { | |
254 | struct task_struct *tsk; | |
255 | struct mm_struct *mm; | |
2d137c24 | 256 | int fault, sig, code; |
1da177e4 | 257 | |
25ce1dd7 NP |
258 | if (notify_page_fault(regs, fsr)) |
259 | return 0; | |
260 | ||
1da177e4 LT |
261 | tsk = current; |
262 | mm = tsk->mm; | |
263 | ||
264 | /* | |
265 | * If we're in an interrupt or have no user | |
266 | * context, we must not take the fault.. | |
267 | */ | |
6edaf68a | 268 | if (in_atomic() || !mm) |
1da177e4 LT |
269 | goto no_context; |
270 | ||
840ff6a4 RK |
271 | /* |
272 | * As per x86, we may deadlock here. However, since the kernel only | |
273 | * validly references user space from well defined areas of the code, | |
274 | * we can bug out early if this is from code which shouldn't. | |
275 | */ | |
276 | if (!down_read_trylock(&mm->mmap_sem)) { | |
277 | if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc)) | |
278 | goto no_context; | |
279 | down_read(&mm->mmap_sem); | |
280 | } | |
281 | ||
1da177e4 LT |
282 | fault = __do_page_fault(mm, addr, fsr, tsk); |
283 | up_read(&mm->mmap_sem); | |
284 | ||
285 | /* | |
ff2afb9d | 286 | * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR |
1da177e4 | 287 | */ |
5c72fc5c | 288 | if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS)))) |
1da177e4 LT |
289 | return 0; |
290 | ||
1da177e4 LT |
291 | /* |
292 | * If we are in kernel mode at this point, we | |
293 | * have no context to handle this fault with. | |
294 | */ | |
295 | if (!user_mode(regs)) | |
296 | goto no_context; | |
297 | ||
83c54070 | 298 | if (fault & VM_FAULT_OOM) { |
1da177e4 | 299 | /* |
2d137c24 | 300 | * We ran out of memory, or some other thing |
301 | * happened to us that made us unable to handle | |
302 | * the page fault gracefully. | |
1da177e4 LT |
303 | */ |
304 | printk("VM: killing process %s\n", tsk->comm); | |
dcca2bde | 305 | do_group_exit(SIGKILL); |
2d137c24 | 306 | return 0; |
83c54070 NP |
307 | } |
308 | if (fault & VM_FAULT_SIGBUS) { | |
2d137c24 | 309 | /* |
310 | * We had some memory, but were unable to | |
311 | * successfully fix up this page fault. | |
312 | */ | |
313 | sig = SIGBUS; | |
314 | code = BUS_ADRERR; | |
83c54070 | 315 | } else { |
2d137c24 | 316 | /* |
317 | * Something tried to access memory that | |
318 | * isn't in our memory map.. | |
319 | */ | |
320 | sig = SIGSEGV; | |
321 | code = fault == VM_FAULT_BADACCESS ? | |
322 | SEGV_ACCERR : SEGV_MAPERR; | |
1da177e4 | 323 | } |
1da177e4 | 324 | |
2d137c24 | 325 | __do_user_fault(tsk, addr, fsr, sig, code, regs); |
326 | return 0; | |
1da177e4 LT |
327 | |
328 | no_context: | |
329 | __do_kernel_fault(mm, addr, fsr, regs); | |
330 | return 0; | |
331 | } | |
09529f7a CM |
332 | #else /* CONFIG_MMU */ |
333 | static int | |
334 | do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
335 | { | |
336 | return 0; | |
337 | } | |
338 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
339 | |
340 | /* | |
341 | * First Level Translation Fault Handler | |
342 | * | |
343 | * We enter here because the first level page table doesn't contain | |
344 | * a valid entry for the address. | |
345 | * | |
346 | * If the address is in kernel space (>= TASK_SIZE), then we are | |
347 | * probably faulting in the vmalloc() area. | |
348 | * | |
349 | * If the init_task's first level page tables contains the relevant | |
350 | * entry, we copy the it to this task. If not, we send the process | |
351 | * a signal, fixup the exception, or oops the kernel. | |
352 | * | |
353 | * NOTE! We MUST NOT take any locks for this case. We may be in an | |
354 | * interrupt or a critical region, and should only copy the information | |
355 | * from the master page table, nothing more. | |
356 | */ | |
09529f7a | 357 | #ifdef CONFIG_MMU |
785d3cd2 | 358 | static int __kprobes |
1da177e4 LT |
359 | do_translation_fault(unsigned long addr, unsigned int fsr, |
360 | struct pt_regs *regs) | |
361 | { | |
1da177e4 LT |
362 | unsigned int index; |
363 | pgd_t *pgd, *pgd_k; | |
364 | pmd_t *pmd, *pmd_k; | |
365 | ||
366 | if (addr < TASK_SIZE) | |
367 | return do_page_fault(addr, fsr, regs); | |
368 | ||
369 | index = pgd_index(addr); | |
370 | ||
371 | /* | |
372 | * FIXME: CP15 C1 is write only on ARMv3 architectures. | |
373 | */ | |
374 | pgd = cpu_get_pgd() + index; | |
375 | pgd_k = init_mm.pgd + index; | |
376 | ||
377 | if (pgd_none(*pgd_k)) | |
378 | goto bad_area; | |
379 | ||
380 | if (!pgd_present(*pgd)) | |
381 | set_pgd(pgd, *pgd_k); | |
382 | ||
383 | pmd_k = pmd_offset(pgd_k, addr); | |
384 | pmd = pmd_offset(pgd, addr); | |
385 | ||
386 | if (pmd_none(*pmd_k)) | |
387 | goto bad_area; | |
388 | ||
389 | copy_pmd(pmd, pmd_k); | |
390 | return 0; | |
391 | ||
392 | bad_area: | |
e5beac37 | 393 | do_bad_area(addr, fsr, regs); |
1da177e4 LT |
394 | return 0; |
395 | } | |
09529f7a CM |
396 | #else /* CONFIG_MMU */ |
397 | static int | |
398 | do_translation_fault(unsigned long addr, unsigned int fsr, | |
399 | struct pt_regs *regs) | |
400 | { | |
401 | return 0; | |
402 | } | |
403 | #endif /* CONFIG_MMU */ | |
1da177e4 LT |
404 | |
405 | /* | |
406 | * Some section permission faults need to be handled gracefully. | |
407 | * They can happen due to a __{get,put}_user during an oops. | |
408 | */ | |
409 | static int | |
410 | do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
411 | { | |
e5beac37 | 412 | do_bad_area(addr, fsr, regs); |
1da177e4 LT |
413 | return 0; |
414 | } | |
415 | ||
416 | /* | |
417 | * This abort handler always returns "fault". | |
418 | */ | |
419 | static int | |
420 | do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs) | |
421 | { | |
422 | return 1; | |
423 | } | |
424 | ||
425 | static struct fsr_info { | |
426 | int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs); | |
427 | int sig; | |
cfb0810e | 428 | int code; |
1da177e4 LT |
429 | const char *name; |
430 | } fsr_info[] = { | |
431 | /* | |
432 | * The following are the standard ARMv3 and ARMv4 aborts. ARMv5 | |
433 | * defines these to be "precise" aborts. | |
434 | */ | |
cfb0810e RK |
435 | { do_bad, SIGSEGV, 0, "vector exception" }, |
436 | { do_bad, SIGILL, BUS_ADRALN, "alignment exception" }, | |
437 | { do_bad, SIGKILL, 0, "terminal exception" }, | |
438 | { do_bad, SIGILL, BUS_ADRALN, "alignment exception" }, | |
439 | { do_bad, SIGBUS, 0, "external abort on linefetch" }, | |
440 | { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" }, | |
441 | { do_bad, SIGBUS, 0, "external abort on linefetch" }, | |
442 | { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" }, | |
443 | { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, | |
444 | { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" }, | |
445 | { do_bad, SIGBUS, 0, "external abort on non-linefetch" }, | |
446 | { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" }, | |
447 | { do_bad, SIGBUS, 0, "external abort on translation" }, | |
448 | { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" }, | |
449 | { do_bad, SIGBUS, 0, "external abort on translation" }, | |
450 | { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" }, | |
1da177e4 LT |
451 | /* |
452 | * The following are "imprecise" aborts, which are signalled by bit | |
453 | * 10 of the FSR, and may not be recoverable. These are only | |
454 | * supported if the CPU abort handler supports bit 10. | |
455 | */ | |
cfb0810e RK |
456 | { do_bad, SIGBUS, 0, "unknown 16" }, |
457 | { do_bad, SIGBUS, 0, "unknown 17" }, | |
458 | { do_bad, SIGBUS, 0, "unknown 18" }, | |
459 | { do_bad, SIGBUS, 0, "unknown 19" }, | |
460 | { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */ | |
461 | { do_bad, SIGBUS, 0, "unknown 21" }, | |
462 | { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */ | |
463 | { do_bad, SIGBUS, 0, "unknown 23" }, | |
464 | { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */ | |
465 | { do_bad, SIGBUS, 0, "unknown 25" }, | |
466 | { do_bad, SIGBUS, 0, "unknown 26" }, | |
467 | { do_bad, SIGBUS, 0, "unknown 27" }, | |
468 | { do_bad, SIGBUS, 0, "unknown 28" }, | |
469 | { do_bad, SIGBUS, 0, "unknown 29" }, | |
470 | { do_bad, SIGBUS, 0, "unknown 30" }, | |
471 | { do_bad, SIGBUS, 0, "unknown 31" } | |
1da177e4 LT |
472 | }; |
473 | ||
474 | void __init | |
475 | hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *), | |
476 | int sig, const char *name) | |
477 | { | |
478 | if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) { | |
479 | fsr_info[nr].fn = fn; | |
480 | fsr_info[nr].sig = sig; | |
481 | fsr_info[nr].name = name; | |
482 | } | |
483 | } | |
484 | ||
485 | /* | |
486 | * Dispatch a data abort to the relevant handler. | |
487 | */ | |
7ab3f8d5 | 488 | asmlinkage void __exception |
1da177e4 LT |
489 | do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs) |
490 | { | |
491 | const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6); | |
cfb0810e | 492 | struct siginfo info; |
1da177e4 LT |
493 | |
494 | if (!inf->fn(addr, fsr, regs)) | |
495 | return; | |
496 | ||
497 | printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n", | |
498 | inf->name, fsr, addr); | |
cfb0810e RK |
499 | |
500 | info.si_signo = inf->sig; | |
501 | info.si_errno = 0; | |
502 | info.si_code = inf->code; | |
503 | info.si_addr = (void __user *)addr; | |
1eeb66a1 | 504 | arm_notify_die("", regs, &info, fsr, 0); |
1da177e4 LT |
505 | } |
506 | ||
7ab3f8d5 | 507 | asmlinkage void __exception |
1da177e4 LT |
508 | do_PrefetchAbort(unsigned long addr, struct pt_regs *regs) |
509 | { | |
510 | do_translation_fault(addr, 0, regs); | |
511 | } | |
512 |