Commit | Line | Data |
---|---|---|
26ff6c11 PM |
1 | /* |
2 | * Page fault handler for SH with an MMU. | |
1da177e4 | 3 | * |
1da177e4 | 4 | * Copyright (C) 1999 Niibe Yutaka |
dbdb4e9f | 5 | * Copyright (C) 2003 - 2012 Paul Mundt |
1da177e4 LT |
6 | * |
7 | * Based on linux/arch/i386/mm/fault.c: | |
8 | * Copyright (C) 1995 Linus Torvalds | |
26ff6c11 PM |
9 | * |
10 | * This file is subject to the terms and conditions of the GNU General Public | |
11 | * License. See the file "COPYING" in the main directory of this archive | |
12 | * for more details. | |
1da177e4 | 13 | */ |
1da177e4 | 14 | #include <linux/kernel.h> |
1da177e4 | 15 | #include <linux/mm.h> |
0f08f338 PM |
16 | #include <linux/hardirq.h> |
17 | #include <linux/kprobes.h> | |
cdd6c482 | 18 | #include <linux/perf_event.h> |
dbdb4e9f | 19 | #include <linux/kdebug.h> |
e7cc9a73 | 20 | #include <asm/io_trapped.h> |
1da177e4 | 21 | #include <asm/mmu_context.h> |
db2e1fa3 | 22 | #include <asm/tlbflush.h> |
e839ca52 | 23 | #include <asm/traps.h> |
1da177e4 | 24 | |
7433ab77 PM |
25 | static inline int notify_page_fault(struct pt_regs *regs, int trap) |
26 | { | |
27 | int ret = 0; | |
28 | ||
c63c3105 | 29 | if (kprobes_built_in() && !user_mode(regs)) { |
7433ab77 PM |
30 | preempt_disable(); |
31 | if (kprobe_running() && kprobe_fault_handler(regs, trap)) | |
32 | ret = 1; | |
33 | preempt_enable(); | |
34 | } | |
7433ab77 PM |
35 | |
36 | return ret; | |
37 | } | |
38 | ||
dbdb4e9f PM |
39 | static void |
40 | force_sig_info_fault(int si_signo, int si_code, unsigned long address, | |
41 | struct task_struct *tsk) | |
42 | { | |
43 | siginfo_t info; | |
44 | ||
45 | info.si_signo = si_signo; | |
46 | info.si_errno = 0; | |
47 | info.si_code = si_code; | |
48 | info.si_addr = (void __user *)address; | |
49 | ||
50 | force_sig_info(si_signo, &info, tsk); | |
51 | } | |
52 | ||
45c0e0e2 SM |
53 | /* |
54 | * This is useful to dump out the page tables associated with | |
55 | * 'addr' in mm 'mm'. | |
56 | */ | |
57 | static void show_pte(struct mm_struct *mm, unsigned long addr) | |
58 | { | |
59 | pgd_t *pgd; | |
60 | ||
61 | if (mm) | |
62 | pgd = mm->pgd; | |
63 | else | |
64 | pgd = get_TTB(); | |
65 | ||
66 | printk(KERN_ALERT "pgd = %p\n", pgd); | |
67 | pgd += pgd_index(addr); | |
68 | printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr, | |
28080329 | 69 | (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd)); |
45c0e0e2 SM |
70 | |
71 | do { | |
72 | pud_t *pud; | |
73 | pmd_t *pmd; | |
74 | pte_t *pte; | |
75 | ||
76 | if (pgd_none(*pgd)) | |
77 | break; | |
78 | ||
79 | if (pgd_bad(*pgd)) { | |
80 | printk("(bad)"); | |
81 | break; | |
82 | } | |
83 | ||
84 | pud = pud_offset(pgd, addr); | |
85 | if (PTRS_PER_PUD != 1) | |
28080329 | 86 | printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2), |
45c0e0e2 SM |
87 | (u64)pud_val(*pud)); |
88 | ||
89 | if (pud_none(*pud)) | |
90 | break; | |
91 | ||
92 | if (pud_bad(*pud)) { | |
93 | printk("(bad)"); | |
94 | break; | |
95 | } | |
96 | ||
97 | pmd = pmd_offset(pud, addr); | |
98 | if (PTRS_PER_PMD != 1) | |
28080329 | 99 | printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2), |
45c0e0e2 SM |
100 | (u64)pmd_val(*pmd)); |
101 | ||
102 | if (pmd_none(*pmd)) | |
103 | break; | |
104 | ||
105 | if (pmd_bad(*pmd)) { | |
106 | printk("(bad)"); | |
107 | break; | |
108 | } | |
109 | ||
110 | /* We must not map this if we have highmem enabled */ | |
111 | if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT))) | |
112 | break; | |
113 | ||
114 | pte = pte_offset_kernel(pmd, addr); | |
28080329 PM |
115 | printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2), |
116 | (u64)pte_val(*pte)); | |
45c0e0e2 SM |
117 | } while (0); |
118 | ||
119 | printk("\n"); | |
120 | } | |
121 | ||
0f60bb25 PM |
122 | static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) |
123 | { | |
124 | unsigned index = pgd_index(address); | |
125 | pgd_t *pgd_k; | |
126 | pud_t *pud, *pud_k; | |
127 | pmd_t *pmd, *pmd_k; | |
128 | ||
129 | pgd += index; | |
130 | pgd_k = init_mm.pgd + index; | |
131 | ||
132 | if (!pgd_present(*pgd_k)) | |
133 | return NULL; | |
134 | ||
135 | pud = pud_offset(pgd, address); | |
136 | pud_k = pud_offset(pgd_k, address); | |
137 | if (!pud_present(*pud_k)) | |
138 | return NULL; | |
139 | ||
5d9b4b19 MF |
140 | if (!pud_present(*pud)) |
141 | set_pud(pud, *pud_k); | |
142 | ||
0f60bb25 PM |
143 | pmd = pmd_offset(pud, address); |
144 | pmd_k = pmd_offset(pud_k, address); | |
145 | if (!pmd_present(*pmd_k)) | |
146 | return NULL; | |
147 | ||
148 | if (!pmd_present(*pmd)) | |
149 | set_pmd(pmd, *pmd_k); | |
05dd2cd3 MF |
150 | else { |
151 | /* | |
152 | * The page tables are fully synchronised so there must | |
153 | * be another reason for the fault. Return NULL here to | |
154 | * signal that we have not taken care of the fault. | |
155 | */ | |
0f60bb25 | 156 | BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); |
05dd2cd3 MF |
157 | return NULL; |
158 | } | |
0f60bb25 PM |
159 | |
160 | return pmd_k; | |
161 | } | |
162 | ||
163 | /* | |
164 | * Handle a fault on the vmalloc or module mapping area | |
165 | */ | |
166 | static noinline int vmalloc_fault(unsigned long address) | |
167 | { | |
168 | pgd_t *pgd_k; | |
169 | pmd_t *pmd_k; | |
170 | pte_t *pte_k; | |
171 | ||
c3e0af98 PM |
172 | /* Make sure we are in vmalloc/module/P3 area: */ |
173 | if (!(address >= P3SEG && address < P3_ADDR_MAX)) | |
0f60bb25 PM |
174 | return -1; |
175 | ||
176 | /* | |
177 | * Synchronize this task's top level page-table | |
178 | * with the 'reference' page table. | |
179 | * | |
180 | * Do _not_ use "current" here. We might be inside | |
181 | * an interrupt in the middle of a task switch.. | |
182 | */ | |
183 | pgd_k = get_TTB(); | |
05dd2cd3 | 184 | pmd_k = vmalloc_sync_one(pgd_k, address); |
0f60bb25 PM |
185 | if (!pmd_k) |
186 | return -1; | |
187 | ||
188 | pte_k = pte_offset_kernel(pmd_k, address); | |
189 | if (!pte_present(*pte_k)) | |
190 | return -1; | |
191 | ||
192 | return 0; | |
193 | } | |
194 | ||
dbdb4e9f PM |
195 | static void |
196 | show_fault_oops(struct pt_regs *regs, unsigned long address) | |
197 | { | |
198 | if (!oops_may_print()) | |
199 | return; | |
200 | ||
201 | printk(KERN_ALERT "BUG: unable to handle kernel "); | |
202 | if (address < PAGE_SIZE) | |
203 | printk(KERN_CONT "NULL pointer dereference"); | |
204 | else | |
205 | printk(KERN_CONT "paging request"); | |
206 | ||
207 | printk(KERN_CONT " at %08lx\n", address); | |
208 | printk(KERN_ALERT "PC:"); | |
209 | printk_address(regs->pc, 1); | |
210 | ||
211 | show_pte(NULL, address); | |
212 | } | |
213 | ||
214 | static noinline void | |
5a1dc78a | 215 | no_context(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
216 | unsigned long address) |
217 | { | |
218 | /* Are we prepared to handle this kernel fault? */ | |
219 | if (fixup_exception(regs)) | |
220 | return; | |
221 | ||
222 | if (handle_trapped_io(regs, address)) | |
223 | return; | |
224 | ||
225 | /* | |
226 | * Oops. The kernel tried to access some bad page. We'll have to | |
227 | * terminate things with extreme prejudice. | |
228 | */ | |
229 | bust_spinlocks(1); | |
230 | ||
231 | show_fault_oops(regs, address); | |
232 | ||
5a1dc78a | 233 | die("Oops", regs, error_code); |
dbdb4e9f PM |
234 | bust_spinlocks(0); |
235 | do_exit(SIGKILL); | |
236 | } | |
237 | ||
238 | static void | |
5a1dc78a | 239 | __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
240 | unsigned long address, int si_code) |
241 | { | |
242 | struct task_struct *tsk = current; | |
243 | ||
244 | /* User mode accesses just cause a SIGSEGV */ | |
245 | if (user_mode(regs)) { | |
246 | /* | |
247 | * It's possible to have interrupts off here: | |
248 | */ | |
249 | local_irq_enable(); | |
250 | ||
251 | force_sig_info_fault(SIGSEGV, si_code, address, tsk); | |
252 | ||
253 | return; | |
254 | } | |
255 | ||
5a1dc78a | 256 | no_context(regs, error_code, address); |
dbdb4e9f PM |
257 | } |
258 | ||
259 | static noinline void | |
5a1dc78a | 260 | bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
261 | unsigned long address) |
262 | { | |
5a1dc78a | 263 | __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); |
dbdb4e9f PM |
264 | } |
265 | ||
266 | static void | |
5a1dc78a | 267 | __bad_area(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
268 | unsigned long address, int si_code) |
269 | { | |
270 | struct mm_struct *mm = current->mm; | |
271 | ||
272 | /* | |
273 | * Something tried to access memory that isn't in our memory map.. | |
274 | * Fix it, but check if it's kernel or user first.. | |
275 | */ | |
276 | up_read(&mm->mmap_sem); | |
277 | ||
5a1dc78a | 278 | __bad_area_nosemaphore(regs, error_code, address, si_code); |
dbdb4e9f PM |
279 | } |
280 | ||
281 | static noinline void | |
5a1dc78a | 282 | bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) |
dbdb4e9f | 283 | { |
5a1dc78a | 284 | __bad_area(regs, error_code, address, SEGV_MAPERR); |
dbdb4e9f PM |
285 | } |
286 | ||
287 | static noinline void | |
5a1dc78a | 288 | bad_area_access_error(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
289 | unsigned long address) |
290 | { | |
5a1dc78a | 291 | __bad_area(regs, error_code, address, SEGV_ACCERR); |
dbdb4e9f PM |
292 | } |
293 | ||
294 | static void out_of_memory(void) | |
295 | { | |
296 | /* | |
297 | * We ran out of memory, call the OOM killer, and return the userspace | |
298 | * (which will retry the fault, or kill us if we got oom-killed): | |
299 | */ | |
300 | up_read(¤t->mm->mmap_sem); | |
301 | ||
302 | pagefault_out_of_memory(); | |
303 | } | |
304 | ||
305 | static void | |
5a1dc78a | 306 | do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address) |
dbdb4e9f PM |
307 | { |
308 | struct task_struct *tsk = current; | |
309 | struct mm_struct *mm = tsk->mm; | |
310 | ||
311 | up_read(&mm->mmap_sem); | |
312 | ||
313 | /* Kernel mode? Handle exceptions or die: */ | |
314 | if (!user_mode(regs)) | |
5a1dc78a | 315 | no_context(regs, error_code, address); |
dbdb4e9f PM |
316 | |
317 | force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk); | |
318 | } | |
319 | ||
320 | static noinline int | |
5a1dc78a | 321 | mm_fault_error(struct pt_regs *regs, unsigned long error_code, |
dbdb4e9f PM |
322 | unsigned long address, unsigned int fault) |
323 | { | |
324 | /* | |
325 | * Pagefault was interrupted by SIGKILL. We have no reason to | |
326 | * continue pagefault. | |
327 | */ | |
328 | if (fatal_signal_pending(current)) { | |
329 | if (!(fault & VM_FAULT_RETRY)) | |
330 | up_read(¤t->mm->mmap_sem); | |
331 | if (!user_mode(regs)) | |
5a1dc78a | 332 | no_context(regs, error_code, address); |
dbdb4e9f PM |
333 | return 1; |
334 | } | |
335 | ||
336 | if (!(fault & VM_FAULT_ERROR)) | |
337 | return 0; | |
338 | ||
339 | if (fault & VM_FAULT_OOM) { | |
340 | /* Kernel mode? Handle exceptions or die: */ | |
341 | if (!user_mode(regs)) { | |
342 | up_read(¤t->mm->mmap_sem); | |
5a1dc78a | 343 | no_context(regs, error_code, address); |
dbdb4e9f PM |
344 | return 1; |
345 | } | |
346 | ||
347 | out_of_memory(); | |
348 | } else { | |
349 | if (fault & VM_FAULT_SIGBUS) | |
5a1dc78a | 350 | do_sigbus(regs, error_code, address); |
dbdb4e9f PM |
351 | else |
352 | BUG(); | |
353 | } | |
354 | ||
355 | return 1; | |
356 | } | |
357 | ||
28080329 | 358 | static inline int access_error(int error_code, struct vm_area_struct *vma) |
dbdb4e9f | 359 | { |
28080329 | 360 | if (error_code & FAULT_CODE_WRITE) { |
dbdb4e9f PM |
361 | /* write, present and write, not present: */ |
362 | if (unlikely(!(vma->vm_flags & VM_WRITE))) | |
363 | return 1; | |
364 | return 0; | |
365 | } | |
366 | ||
28080329 PM |
367 | /* ITLB miss on NX page */ |
368 | if (unlikely((error_code & FAULT_CODE_ITLB) && | |
369 | !(vma->vm_flags & VM_EXEC))) | |
370 | return 1; | |
371 | ||
dbdb4e9f PM |
372 | /* read, not present: */ |
373 | if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) | |
374 | return 1; | |
375 | ||
376 | return 0; | |
377 | } | |
378 | ||
0f60bb25 PM |
379 | static int fault_in_kernel_space(unsigned long address) |
380 | { | |
381 | return address >= TASK_SIZE; | |
382 | } | |
383 | ||
1da177e4 LT |
384 | /* |
385 | * This routine handles page faults. It determines the address, | |
386 | * and the problem, and then passes it off to one of the appropriate | |
387 | * routines. | |
388 | */ | |
b5a1bcbe | 389 | asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, |
5a1dc78a | 390 | unsigned long error_code, |
b5a1bcbe | 391 | unsigned long address) |
1da177e4 | 392 | { |
0f60bb25 | 393 | unsigned long vec; |
1da177e4 LT |
394 | struct task_struct *tsk; |
395 | struct mm_struct *mm; | |
396 | struct vm_area_struct * vma; | |
83c54070 | 397 | int fault; |
5a1dc78a | 398 | int write = error_code & FAULT_CODE_WRITE; |
11fd9824 | 399 | unsigned int flags = (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE | |
5a1dc78a | 400 | (write ? FAULT_FLAG_WRITE : 0)); |
1da177e4 | 401 | |
1da177e4 | 402 | tsk = current; |
0f60bb25 | 403 | mm = tsk->mm; |
0f60bb25 | 404 | vec = lookup_exception_vector(); |
1da177e4 | 405 | |
0f60bb25 PM |
406 | /* |
407 | * We fault-in kernel-space virtual memory on-demand. The | |
408 | * 'reference' page table is init_mm.pgd. | |
409 | * | |
410 | * NOTE! We MUST NOT take any locks for this case. We may | |
411 | * be in an interrupt or a critical region, and should | |
412 | * only copy the information from the master page table, | |
413 | * nothing more. | |
414 | */ | |
415 | if (unlikely(fault_in_kernel_space(address))) { | |
416 | if (vmalloc_fault(address) >= 0) | |
99a596f9 | 417 | return; |
0f60bb25 | 418 | if (notify_page_fault(regs, vec)) |
96e14e54 | 419 | return; |
99a596f9 | 420 | |
5a1dc78a | 421 | bad_area_nosemaphore(regs, error_code, address); |
dbdb4e9f | 422 | return; |
99a596f9 SM |
423 | } |
424 | ||
0f60bb25 | 425 | if (unlikely(notify_page_fault(regs, vec))) |
7433ab77 PM |
426 | return; |
427 | ||
f2fb4e4f | 428 | /* Only enable interrupts if they were on before the fault */ |
7433ab77 | 429 | if ((regs->sr & SR_IMASK) != SR_IMASK) |
f2fb4e4f | 430 | local_irq_enable(); |
f2fb4e4f | 431 | |
a8b0ca17 | 432 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); |
f2fb4e4f | 433 | |
1da177e4 | 434 | /* |
0f60bb25 PM |
435 | * If we're in an interrupt, have no user context or are running |
436 | * in an atomic region then we must not take the fault: | |
1da177e4 | 437 | */ |
dbdb4e9f | 438 | if (unlikely(in_atomic() || !mm)) { |
5a1dc78a | 439 | bad_area_nosemaphore(regs, error_code, address); |
dbdb4e9f PM |
440 | return; |
441 | } | |
1da177e4 | 442 | |
11fd9824 | 443 | retry: |
1da177e4 LT |
444 | down_read(&mm->mmap_sem); |
445 | ||
446 | vma = find_vma(mm, address); | |
dbdb4e9f | 447 | if (unlikely(!vma)) { |
5a1dc78a | 448 | bad_area(regs, error_code, address); |
dbdb4e9f PM |
449 | return; |
450 | } | |
451 | if (likely(vma->vm_start <= address)) | |
1da177e4 | 452 | goto good_area; |
dbdb4e9f | 453 | if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { |
5a1dc78a | 454 | bad_area(regs, error_code, address); |
dbdb4e9f PM |
455 | return; |
456 | } | |
457 | if (unlikely(expand_stack(vma, address))) { | |
5a1dc78a | 458 | bad_area(regs, error_code, address); |
dbdb4e9f PM |
459 | return; |
460 | } | |
0f60bb25 PM |
461 | |
462 | /* | |
463 | * Ok, we have a good vm_area for this memory access, so | |
464 | * we can handle it.. | |
465 | */ | |
1da177e4 | 466 | good_area: |
5a1dc78a PM |
467 | if (unlikely(access_error(error_code, vma))) { |
468 | bad_area_access_error(regs, error_code, address); | |
dbdb4e9f | 469 | return; |
1da177e4 LT |
470 | } |
471 | ||
5a1dc78a PM |
472 | set_thread_fault_code(error_code); |
473 | ||
1da177e4 LT |
474 | /* |
475 | * If for any reason at all we couldn't handle the fault, | |
476 | * make sure we exit gracefully rather than endlessly redo | |
477 | * the fault. | |
478 | */ | |
11fd9824 KC |
479 | fault = handle_mm_fault(mm, vma, address, flags); |
480 | ||
dbdb4e9f | 481 | if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR))) |
5a1dc78a | 482 | if (mm_fault_error(regs, error_code, address, fault)) |
dbdb4e9f | 483 | return; |
11fd9824 KC |
484 | |
485 | if (flags & FAULT_FLAG_ALLOW_RETRY) { | |
486 | if (fault & VM_FAULT_MAJOR) { | |
487 | tsk->maj_flt++; | |
488 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, | |
489 | regs, address); | |
490 | } else { | |
491 | tsk->min_flt++; | |
492 | perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, | |
493 | regs, address); | |
494 | } | |
495 | if (fault & VM_FAULT_RETRY) { | |
496 | flags &= ~FAULT_FLAG_ALLOW_RETRY; | |
497 | ||
498 | /* | |
499 | * No need to up_read(&mm->mmap_sem) as we would | |
500 | * have already released it in __lock_page_or_retry | |
501 | * in mm/filemap.c. | |
502 | */ | |
503 | goto retry; | |
504 | } | |
7433ab77 | 505 | } |
1da177e4 LT |
506 | |
507 | up_read(&mm->mmap_sem); | |
1da177e4 | 508 | } |