Loading...
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 */
11#include <linux/module.h>
12#include <linux/signal.h>
13#include <linux/mm.h>
14#include <linux/hardirq.h>
15#include <linux/init.h>
16#include <linux/kprobes.h>
17#include <linux/uaccess.h>
18#include <linux/page-flags.h>
19#include <linux/sched.h>
20#include <linux/highmem.h>
21#include <linux/perf_event.h>
22
23#include <asm/system.h>
24#include <asm/pgtable.h>
25#include <asm/tlbflush.h>
26
27#include "fault.h"
28
29/*
30 * Fault status register encodings. We steal bit 31 for our own purposes.
31 */
32#define FSR_LNX_PF (1 << 31)
33#define FSR_WRITE (1 << 11)
34#define FSR_FS4 (1 << 10)
35#define FSR_FS3_0 (15)
36
37static inline int fsr_fs(unsigned int fsr)
38{
39 return (fsr & FSR_FS3_0) | (fsr & FSR_FS4) >> 6;
40}
41
42#ifdef CONFIG_MMU
43
44#ifdef CONFIG_KPROBES
45static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
46{
47 int ret = 0;
48
49 if (!user_mode(regs)) {
50 /* kprobe_running() needs smp_processor_id() */
51 preempt_disable();
52 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
53 ret = 1;
54 preempt_enable();
55 }
56
57 return ret;
58}
59#else
60static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
61{
62 return 0;
63}
64#endif
65
66/*
67 * This is useful to dump out the page tables associated with
68 * 'addr' in mm 'mm'.
69 */
70void show_pte(struct mm_struct *mm, unsigned long addr)
71{
72 pgd_t *pgd;
73
74 if (!mm)
75 mm = &init_mm;
76
77 printk(KERN_ALERT "pgd = %p\n", mm->pgd);
78 pgd = pgd_offset(mm, addr);
79 printk(KERN_ALERT "[%08lx] *pgd=%08llx",
80 addr, (long long)pgd_val(*pgd));
81
82 do {
83 pud_t *pud;
84 pmd_t *pmd;
85 pte_t *pte;
86
87 if (pgd_none(*pgd))
88 break;
89
90 if (pgd_bad(*pgd)) {
91 printk("(bad)");
92 break;
93 }
94
95 pud = pud_offset(pgd, addr);
96 if (PTRS_PER_PUD != 1)
97 printk(", *pud=%08llx", (long long)pud_val(*pud));
98
99 if (pud_none(*pud))
100 break;
101
102 if (pud_bad(*pud)) {
103 printk("(bad)");
104 break;
105 }
106
107 pmd = pmd_offset(pud, addr);
108 if (PTRS_PER_PMD != 1)
109 printk(", *pmd=%08llx", (long long)pmd_val(*pmd));
110
111 if (pmd_none(*pmd))
112 break;
113
114 if (pmd_bad(*pmd)) {
115 printk("(bad)");
116 break;
117 }
118
119 /* We must not map this if we have highmem enabled */
120 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
121 break;
122
123 pte = pte_offset_map(pmd, addr);
124 printk(", *pte=%08llx", (long long)pte_val(*pte));
125 printk(", *ppte=%08llx",
126 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
127 pte_unmap(pte);
128 } while(0);
129
130 printk("\n");
131}
132#else /* CONFIG_MMU */
133void show_pte(struct mm_struct *mm, unsigned long addr)
134{ }
135#endif /* CONFIG_MMU */
136
137/*
138 * Oops. The kernel tried to access some page that wasn't present.
139 */
140static void
141__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
142 struct pt_regs *regs)
143{
144 /*
145 * Are we prepared to handle this kernel fault?
146 */
147 if (fixup_exception(regs))
148 return;
149
150 /*
151 * No handler, we'll have to terminate things with extreme prejudice.
152 */
153 bust_spinlocks(1);
154 printk(KERN_ALERT
155 "Unable to handle kernel %s at virtual address %08lx\n",
156 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
157 "paging request", addr);
158
159 show_pte(mm, addr);
160 die("Oops", regs, fsr);
161 bust_spinlocks(0);
162 do_exit(SIGKILL);
163}
164
165/*
166 * Something tried to access memory that isn't in our memory map..
167 * User mode accesses just cause a SIGSEGV
168 */
169static void
170__do_user_fault(struct task_struct *tsk, unsigned long addr,
171 unsigned int fsr, unsigned int sig, int code,
172 struct pt_regs *regs)
173{
174 struct siginfo si;
175
176#ifdef CONFIG_DEBUG_USER
177 if (user_debug & UDBG_SEGV) {
178 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
179 tsk->comm, sig, addr, fsr);
180 show_pte(tsk->mm, addr);
181 show_regs(regs);
182 }
183#endif
184
185 tsk->thread.address = addr;
186 tsk->thread.error_code = fsr;
187 tsk->thread.trap_no = 14;
188 si.si_signo = sig;
189 si.si_errno = 0;
190 si.si_code = code;
191 si.si_addr = (void __user *)addr;
192 force_sig_info(sig, &si, tsk);
193}
194
195void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
196{
197 struct task_struct *tsk = current;
198 struct mm_struct *mm = tsk->active_mm;
199
200 /*
201 * If we are in kernel mode at this point, we
202 * have no context to handle this fault with.
203 */
204 if (user_mode(regs))
205 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
206 else
207 __do_kernel_fault(mm, addr, fsr, regs);
208}
209
210#ifdef CONFIG_MMU
211#define VM_FAULT_BADMAP 0x010000
212#define VM_FAULT_BADACCESS 0x020000
213
214/*
215 * Check that the permissions on the VMA allow for the fault which occurred.
216 * If we encountered a write fault, we must have write permission, otherwise
217 * we allow any permission.
218 */
219static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
220{
221 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
222
223 if (fsr & FSR_WRITE)
224 mask = VM_WRITE;
225 if (fsr & FSR_LNX_PF)
226 mask = VM_EXEC;
227
228 return vma->vm_flags & mask ? false : true;
229}
230
231static int __kprobes
232__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
233 struct task_struct *tsk)
234{
235 struct vm_area_struct *vma;
236 int fault;
237
238 vma = find_vma(mm, addr);
239 fault = VM_FAULT_BADMAP;
240 if (unlikely(!vma))
241 goto out;
242 if (unlikely(vma->vm_start > addr))
243 goto check_stack;
244
245 /*
246 * Ok, we have a good vm_area for this
247 * memory access, so we can handle it.
248 */
249good_area:
250 if (access_error(fsr, vma)) {
251 fault = VM_FAULT_BADACCESS;
252 goto out;
253 }
254
255 /*
256 * If for any reason at all we couldn't handle the fault, make
257 * sure we exit gracefully rather than endlessly redo the fault.
258 */
259 fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0);
260 if (unlikely(fault & VM_FAULT_ERROR))
261 return fault;
262 if (fault & VM_FAULT_MAJOR)
263 tsk->maj_flt++;
264 else
265 tsk->min_flt++;
266 return fault;
267
268check_stack:
269 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
270 goto good_area;
271out:
272 return fault;
273}
274
275static int __kprobes
276do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
277{
278 struct task_struct *tsk;
279 struct mm_struct *mm;
280 int fault, sig, code;
281
282 if (notify_page_fault(regs, fsr))
283 return 0;
284
285 tsk = current;
286 mm = tsk->mm;
287
288 /* Enable interrupts if they were enabled in the parent context. */
289 if (interrupts_enabled(regs))
290 local_irq_enable();
291
292 /*
293 * If we're in an interrupt or have no user
294 * context, we must not take the fault..
295 */
296 if (in_atomic() || !mm)
297 goto no_context;
298
299 /*
300 * As per x86, we may deadlock here. However, since the kernel only
301 * validly references user space from well defined areas of the code,
302 * we can bug out early if this is from code which shouldn't.
303 */
304 if (!down_read_trylock(&mm->mmap_sem)) {
305 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
306 goto no_context;
307 down_read(&mm->mmap_sem);
308 } else {
309 /*
310 * The above down_read_trylock() might have succeeded in
311 * which case, we'll have missed the might_sleep() from
312 * down_read()
313 */
314 might_sleep();
315#ifdef CONFIG_DEBUG_VM
316 if (!user_mode(regs) &&
317 !search_exception_tables(regs->ARM_pc))
318 goto no_context;
319#endif
320 }
321
322 fault = __do_page_fault(mm, addr, fsr, tsk);
323 up_read(&mm->mmap_sem);
324
325 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
326 if (fault & VM_FAULT_MAJOR)
327 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, addr);
328 else if (fault & VM_FAULT_MINOR)
329 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, addr);
330
331 /*
332 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
333 */
334 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
335 return 0;
336
337 if (fault & VM_FAULT_OOM) {
338 /*
339 * We ran out of memory, call the OOM killer, and return to
340 * userspace (which will retry the fault, or kill us if we
341 * got oom-killed)
342 */
343 pagefault_out_of_memory();
344 return 0;
345 }
346
347 /*
348 * If we are in kernel mode at this point, we
349 * have no context to handle this fault with.
350 */
351 if (!user_mode(regs))
352 goto no_context;
353
354 if (fault & VM_FAULT_SIGBUS) {
355 /*
356 * We had some memory, but were unable to
357 * successfully fix up this page fault.
358 */
359 sig = SIGBUS;
360 code = BUS_ADRERR;
361 } else {
362 /*
363 * Something tried to access memory that
364 * isn't in our memory map..
365 */
366 sig = SIGSEGV;
367 code = fault == VM_FAULT_BADACCESS ?
368 SEGV_ACCERR : SEGV_MAPERR;
369 }
370
371 __do_user_fault(tsk, addr, fsr, sig, code, regs);
372 return 0;
373
374no_context:
375 __do_kernel_fault(mm, addr, fsr, regs);
376 return 0;
377}
378#else /* CONFIG_MMU */
379static int
380do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
381{
382 return 0;
383}
384#endif /* CONFIG_MMU */
385
386/*
387 * First Level Translation Fault Handler
388 *
389 * We enter here because the first level page table doesn't contain
390 * a valid entry for the address.
391 *
392 * If the address is in kernel space (>= TASK_SIZE), then we are
393 * probably faulting in the vmalloc() area.
394 *
395 * If the init_task's first level page tables contains the relevant
396 * entry, we copy the it to this task. If not, we send the process
397 * a signal, fixup the exception, or oops the kernel.
398 *
399 * NOTE! We MUST NOT take any locks for this case. We may be in an
400 * interrupt or a critical region, and should only copy the information
401 * from the master page table, nothing more.
402 */
403#ifdef CONFIG_MMU
404static int __kprobes
405do_translation_fault(unsigned long addr, unsigned int fsr,
406 struct pt_regs *regs)
407{
408 unsigned int index;
409 pgd_t *pgd, *pgd_k;
410 pud_t *pud, *pud_k;
411 pmd_t *pmd, *pmd_k;
412
413 if (addr < TASK_SIZE)
414 return do_page_fault(addr, fsr, regs);
415
416 if (user_mode(regs))
417 goto bad_area;
418
419 index = pgd_index(addr);
420
421 /*
422 * FIXME: CP15 C1 is write only on ARMv3 architectures.
423 */
424 pgd = cpu_get_pgd() + index;
425 pgd_k = init_mm.pgd + index;
426
427 if (pgd_none(*pgd_k))
428 goto bad_area;
429 if (!pgd_present(*pgd))
430 set_pgd(pgd, *pgd_k);
431
432 pud = pud_offset(pgd, addr);
433 pud_k = pud_offset(pgd_k, addr);
434
435 if (pud_none(*pud_k))
436 goto bad_area;
437 if (!pud_present(*pud))
438 set_pud(pud, *pud_k);
439
440 pmd = pmd_offset(pud, addr);
441 pmd_k = pmd_offset(pud_k, addr);
442
443 /*
444 * On ARM one Linux PGD entry contains two hardware entries (see page
445 * tables layout in pgtable.h). We normally guarantee that we always
446 * fill both L1 entries. But create_mapping() doesn't follow the rule.
447 * It can create inidividual L1 entries, so here we have to call
448 * pmd_none() check for the entry really corresponded to address, not
449 * for the first of pair.
450 */
451 index = (addr >> SECTION_SHIFT) & 1;
452 if (pmd_none(pmd_k[index]))
453 goto bad_area;
454
455 copy_pmd(pmd, pmd_k);
456 return 0;
457
458bad_area:
459 do_bad_area(addr, fsr, regs);
460 return 0;
461}
462#else /* CONFIG_MMU */
463static int
464do_translation_fault(unsigned long addr, unsigned int fsr,
465 struct pt_regs *regs)
466{
467 return 0;
468}
469#endif /* CONFIG_MMU */
470
471/*
472 * Some section permission faults need to be handled gracefully.
473 * They can happen due to a __{get,put}_user during an oops.
474 */
475static int
476do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
477{
478 do_bad_area(addr, fsr, regs);
479 return 0;
480}
481
482/*
483 * This abort handler always returns "fault".
484 */
485static int
486do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
487{
488 return 1;
489}
490
491static struct fsr_info {
492 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
493 int sig;
494 int code;
495 const char *name;
496} fsr_info[] = {
497 /*
498 * The following are the standard ARMv3 and ARMv4 aborts. ARMv5
499 * defines these to be "precise" aborts.
500 */
501 { do_bad, SIGSEGV, 0, "vector exception" },
502 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" },
503 { do_bad, SIGKILL, 0, "terminal exception" },
504 { do_bad, SIGBUS, BUS_ADRALN, "alignment exception" },
505 { do_bad, SIGBUS, 0, "external abort on linefetch" },
506 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" },
507 { do_bad, SIGBUS, 0, "external abort on linefetch" },
508 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" },
509 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
510 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" },
511 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
512 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" },
513 { do_bad, SIGBUS, 0, "external abort on translation" },
514 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" },
515 { do_bad, SIGBUS, 0, "external abort on translation" },
516 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" },
517 /*
518 * The following are "imprecise" aborts, which are signalled by bit
519 * 10 of the FSR, and may not be recoverable. These are only
520 * supported if the CPU abort handler supports bit 10.
521 */
522 { do_bad, SIGBUS, 0, "unknown 16" },
523 { do_bad, SIGBUS, 0, "unknown 17" },
524 { do_bad, SIGBUS, 0, "unknown 18" },
525 { do_bad, SIGBUS, 0, "unknown 19" },
526 { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */
527 { do_bad, SIGBUS, 0, "unknown 21" },
528 { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */
529 { do_bad, SIGBUS, 0, "unknown 23" },
530 { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */
531 { do_bad, SIGBUS, 0, "unknown 25" },
532 { do_bad, SIGBUS, 0, "unknown 26" },
533 { do_bad, SIGBUS, 0, "unknown 27" },
534 { do_bad, SIGBUS, 0, "unknown 28" },
535 { do_bad, SIGBUS, 0, "unknown 29" },
536 { do_bad, SIGBUS, 0, "unknown 30" },
537 { do_bad, SIGBUS, 0, "unknown 31" }
538};
539
540void __init
541hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
542 int sig, int code, const char *name)
543{
544 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
545 BUG();
546
547 fsr_info[nr].fn = fn;
548 fsr_info[nr].sig = sig;
549 fsr_info[nr].code = code;
550 fsr_info[nr].name = name;
551}
552
553/*
554 * Dispatch a data abort to the relevant handler.
555 */
556asmlinkage void __exception
557do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
558{
559 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
560 struct siginfo info;
561
562 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
563 return;
564
565 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
566 inf->name, fsr, addr);
567
568 info.si_signo = inf->sig;
569 info.si_errno = 0;
570 info.si_code = inf->code;
571 info.si_addr = (void __user *)addr;
572 arm_notify_die("", regs, &info, fsr, 0);
573}
574
575
576static struct fsr_info ifsr_info[] = {
577 { do_bad, SIGBUS, 0, "unknown 0" },
578 { do_bad, SIGBUS, 0, "unknown 1" },
579 { do_bad, SIGBUS, 0, "debug event" },
580 { do_bad, SIGSEGV, SEGV_ACCERR, "section access flag fault" },
581 { do_bad, SIGBUS, 0, "unknown 4" },
582 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" },
583 { do_bad, SIGSEGV, SEGV_ACCERR, "page access flag fault" },
584 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" },
585 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
586 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" },
587 { do_bad, SIGBUS, 0, "unknown 10" },
588 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" },
589 { do_bad, SIGBUS, 0, "external abort on translation" },
590 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" },
591 { do_bad, SIGBUS, 0, "external abort on translation" },
592 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" },
593 { do_bad, SIGBUS, 0, "unknown 16" },
594 { do_bad, SIGBUS, 0, "unknown 17" },
595 { do_bad, SIGBUS, 0, "unknown 18" },
596 { do_bad, SIGBUS, 0, "unknown 19" },
597 { do_bad, SIGBUS, 0, "unknown 20" },
598 { do_bad, SIGBUS, 0, "unknown 21" },
599 { do_bad, SIGBUS, 0, "unknown 22" },
600 { do_bad, SIGBUS, 0, "unknown 23" },
601 { do_bad, SIGBUS, 0, "unknown 24" },
602 { do_bad, SIGBUS, 0, "unknown 25" },
603 { do_bad, SIGBUS, 0, "unknown 26" },
604 { do_bad, SIGBUS, 0, "unknown 27" },
605 { do_bad, SIGBUS, 0, "unknown 28" },
606 { do_bad, SIGBUS, 0, "unknown 29" },
607 { do_bad, SIGBUS, 0, "unknown 30" },
608 { do_bad, SIGBUS, 0, "unknown 31" },
609};
610
611void __init
612hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
613 int sig, int code, const char *name)
614{
615 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
616 BUG();
617
618 ifsr_info[nr].fn = fn;
619 ifsr_info[nr].sig = sig;
620 ifsr_info[nr].code = code;
621 ifsr_info[nr].name = name;
622}
623
624asmlinkage void __exception
625do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
626{
627 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
628 struct siginfo info;
629
630 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
631 return;
632
633 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
634 inf->name, ifsr, addr);
635
636 info.si_signo = inf->sig;
637 info.si_errno = 0;
638 info.si_code = inf->code;
639 info.si_addr = (void __user *)addr;
640 arm_notify_die("", regs, &info, ifsr, 0);
641}
642
643static int __init exceptions_init(void)
644{
645 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
646 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
647 "I-cache maintenance fault");
648 }
649
650 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
651 /*
652 * TODO: Access flag faults introduced in ARMv6K.
653 * Runtime check for 'K' extension is needed
654 */
655 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
656 "section access flag fault");
657 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
658 "section access flag fault");
659 }
660
661 return 0;
662}
663
664arch_initcall(exceptions_init);
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 */
11#include <linux/module.h>
12#include <linux/signal.h>
13#include <linux/mm.h>
14#include <linux/hardirq.h>
15#include <linux/init.h>
16#include <linux/kprobes.h>
17#include <linux/uaccess.h>
18#include <linux/page-flags.h>
19#include <linux/sched.h>
20#include <linux/highmem.h>
21#include <linux/perf_event.h>
22
23#include <asm/exception.h>
24#include <asm/pgtable.h>
25#include <asm/system_misc.h>
26#include <asm/system_info.h>
27#include <asm/tlbflush.h>
28
29#include "fault.h"
30
31#ifdef CONFIG_MMU
32
33#ifdef CONFIG_KPROBES
34static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
35{
36 int ret = 0;
37
38 if (!user_mode(regs)) {
39 /* kprobe_running() needs smp_processor_id() */
40 preempt_disable();
41 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
42 ret = 1;
43 preempt_enable();
44 }
45
46 return ret;
47}
48#else
49static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
50{
51 return 0;
52}
53#endif
54
55/*
56 * This is useful to dump out the page tables associated with
57 * 'addr' in mm 'mm'.
58 */
59void show_pte(struct mm_struct *mm, unsigned long addr)
60{
61 pgd_t *pgd;
62
63 if (!mm)
64 mm = &init_mm;
65
66 printk(KERN_ALERT "pgd = %p\n", mm->pgd);
67 pgd = pgd_offset(mm, addr);
68 printk(KERN_ALERT "[%08lx] *pgd=%08llx",
69 addr, (long long)pgd_val(*pgd));
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)
86 printk(", *pud=%08llx", (long long)pud_val(*pud));
87
88 if (pud_none(*pud))
89 break;
90
91 if (pud_bad(*pud)) {
92 printk("(bad)");
93 break;
94 }
95
96 pmd = pmd_offset(pud, addr);
97 if (PTRS_PER_PMD != 1)
98 printk(", *pmd=%08llx", (long long)pmd_val(*pmd));
99
100 if (pmd_none(*pmd))
101 break;
102
103 if (pmd_bad(*pmd)) {
104 printk("(bad)");
105 break;
106 }
107
108 /* We must not map this if we have highmem enabled */
109 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
110 break;
111
112 pte = pte_offset_map(pmd, addr);
113 printk(", *pte=%08llx", (long long)pte_val(*pte));
114#ifndef CONFIG_ARM_LPAE
115 printk(", *ppte=%08llx",
116 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
117#endif
118 pte_unmap(pte);
119 } while(0);
120
121 printk("\n");
122}
123#else /* CONFIG_MMU */
124void show_pte(struct mm_struct *mm, unsigned long addr)
125{ }
126#endif /* CONFIG_MMU */
127
128/*
129 * Oops. The kernel tried to access some page that wasn't present.
130 */
131static void
132__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
134{
135 /*
136 * Are we prepared to handle this kernel fault?
137 */
138 if (fixup_exception(regs))
139 return;
140
141 /*
142 * No handler, we'll have to terminate things with extreme prejudice.
143 */
144 bust_spinlocks(1);
145 printk(KERN_ALERT
146 "Unable to handle kernel %s at virtual address %08lx\n",
147 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
148 "paging request", addr);
149
150 show_pte(mm, addr);
151 die("Oops", regs, fsr);
152 bust_spinlocks(0);
153 do_exit(SIGKILL);
154}
155
156/*
157 * Something tried to access memory that isn't in our memory map..
158 * User mode accesses just cause a SIGSEGV
159 */
160static void
161__do_user_fault(struct task_struct *tsk, unsigned long addr,
162 unsigned int fsr, unsigned int sig, int code,
163 struct pt_regs *regs)
164{
165 struct siginfo si;
166
167#ifdef CONFIG_DEBUG_USER
168 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
169 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
170 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
171 tsk->comm, sig, addr, fsr);
172 show_pte(tsk->mm, addr);
173 show_regs(regs);
174 }
175#endif
176
177 tsk->thread.address = addr;
178 tsk->thread.error_code = fsr;
179 tsk->thread.trap_no = 14;
180 si.si_signo = sig;
181 si.si_errno = 0;
182 si.si_code = code;
183 si.si_addr = (void __user *)addr;
184 force_sig_info(sig, &si, tsk);
185}
186
187void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
188{
189 struct task_struct *tsk = current;
190 struct mm_struct *mm = tsk->active_mm;
191
192 /*
193 * If we are in kernel mode at this point, we
194 * have no context to handle this fault with.
195 */
196 if (user_mode(regs))
197 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
198 else
199 __do_kernel_fault(mm, addr, fsr, regs);
200}
201
202#ifdef CONFIG_MMU
203#define VM_FAULT_BADMAP 0x010000
204#define VM_FAULT_BADACCESS 0x020000
205
206/*
207 * Check that the permissions on the VMA allow for the fault which occurred.
208 * If we encountered a write fault, we must have write permission, otherwise
209 * we allow any permission.
210 */
211static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
212{
213 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
214
215 if (fsr & FSR_WRITE)
216 mask = VM_WRITE;
217 if (fsr & FSR_LNX_PF)
218 mask = VM_EXEC;
219
220 return vma->vm_flags & mask ? false : true;
221}
222
223static int __kprobes
224__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
225 unsigned int flags, struct task_struct *tsk)
226{
227 struct vm_area_struct *vma;
228 int fault;
229
230 vma = find_vma(mm, addr);
231 fault = VM_FAULT_BADMAP;
232 if (unlikely(!vma))
233 goto out;
234 if (unlikely(vma->vm_start > addr))
235 goto check_stack;
236
237 /*
238 * Ok, we have a good vm_area for this
239 * memory access, so we can handle it.
240 */
241good_area:
242 if (access_error(fsr, vma)) {
243 fault = VM_FAULT_BADACCESS;
244 goto out;
245 }
246
247 return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
248
249check_stack:
250 /* Don't allow expansion below FIRST_USER_ADDRESS */
251 if (vma->vm_flags & VM_GROWSDOWN &&
252 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
253 goto good_area;
254out:
255 return fault;
256}
257
258static int __kprobes
259do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
260{
261 struct task_struct *tsk;
262 struct mm_struct *mm;
263 int fault, sig, code;
264 int write = fsr & FSR_WRITE;
265 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
266 (write ? FAULT_FLAG_WRITE : 0);
267
268 if (notify_page_fault(regs, fsr))
269 return 0;
270
271 tsk = current;
272 mm = tsk->mm;
273
274 /* Enable interrupts if they were enabled in the parent context. */
275 if (interrupts_enabled(regs))
276 local_irq_enable();
277
278 /*
279 * If we're in an interrupt or have no user
280 * context, we must not take the fault..
281 */
282 if (in_atomic() || !mm)
283 goto no_context;
284
285 /*
286 * As per x86, we may deadlock here. However, since the kernel only
287 * validly references user space from well defined areas of the code,
288 * we can bug out early if this is from code which shouldn't.
289 */
290 if (!down_read_trylock(&mm->mmap_sem)) {
291 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
292 goto no_context;
293retry:
294 down_read(&mm->mmap_sem);
295 } else {
296 /*
297 * The above down_read_trylock() might have succeeded in
298 * which case, we'll have missed the might_sleep() from
299 * down_read()
300 */
301 might_sleep();
302#ifdef CONFIG_DEBUG_VM
303 if (!user_mode(regs) &&
304 !search_exception_tables(regs->ARM_pc))
305 goto no_context;
306#endif
307 }
308
309 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
310
311 /* If we need to retry but a fatal signal is pending, handle the
312 * signal first. We do not need to release the mmap_sem because
313 * it would already be released in __lock_page_or_retry in
314 * mm/filemap.c. */
315 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
316 return 0;
317
318 /*
319 * Major/minor page fault accounting is only done on the
320 * initial attempt. If we go through a retry, it is extremely
321 * likely that the page will be found in page cache at that point.
322 */
323
324 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
325 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
326 if (fault & VM_FAULT_MAJOR) {
327 tsk->maj_flt++;
328 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
329 regs, addr);
330 } else {
331 tsk->min_flt++;
332 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
333 regs, addr);
334 }
335 if (fault & VM_FAULT_RETRY) {
336 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
337 * of starvation. */
338 flags &= ~FAULT_FLAG_ALLOW_RETRY;
339 goto retry;
340 }
341 }
342
343 up_read(&mm->mmap_sem);
344
345 /*
346 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
347 */
348 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
349 return 0;
350
351 if (fault & VM_FAULT_OOM) {
352 /*
353 * We ran out of memory, call the OOM killer, and return to
354 * userspace (which will retry the fault, or kill us if we
355 * got oom-killed)
356 */
357 pagefault_out_of_memory();
358 return 0;
359 }
360
361 /*
362 * If we are in kernel mode at this point, we
363 * have no context to handle this fault with.
364 */
365 if (!user_mode(regs))
366 goto no_context;
367
368 if (fault & VM_FAULT_SIGBUS) {
369 /*
370 * We had some memory, but were unable to
371 * successfully fix up this page fault.
372 */
373 sig = SIGBUS;
374 code = BUS_ADRERR;
375 } else {
376 /*
377 * Something tried to access memory that
378 * isn't in our memory map..
379 */
380 sig = SIGSEGV;
381 code = fault == VM_FAULT_BADACCESS ?
382 SEGV_ACCERR : SEGV_MAPERR;
383 }
384
385 __do_user_fault(tsk, addr, fsr, sig, code, regs);
386 return 0;
387
388no_context:
389 __do_kernel_fault(mm, addr, fsr, regs);
390 return 0;
391}
392#else /* CONFIG_MMU */
393static int
394do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
395{
396 return 0;
397}
398#endif /* CONFIG_MMU */
399
400/*
401 * First Level Translation Fault Handler
402 *
403 * We enter here because the first level page table doesn't contain
404 * a valid entry for the address.
405 *
406 * If the address is in kernel space (>= TASK_SIZE), then we are
407 * probably faulting in the vmalloc() area.
408 *
409 * If the init_task's first level page tables contains the relevant
410 * entry, we copy the it to this task. If not, we send the process
411 * a signal, fixup the exception, or oops the kernel.
412 *
413 * NOTE! We MUST NOT take any locks for this case. We may be in an
414 * interrupt or a critical region, and should only copy the information
415 * from the master page table, nothing more.
416 */
417#ifdef CONFIG_MMU
418static int __kprobes
419do_translation_fault(unsigned long addr, unsigned int fsr,
420 struct pt_regs *regs)
421{
422 unsigned int index;
423 pgd_t *pgd, *pgd_k;
424 pud_t *pud, *pud_k;
425 pmd_t *pmd, *pmd_k;
426
427 if (addr < TASK_SIZE)
428 return do_page_fault(addr, fsr, regs);
429
430 if (user_mode(regs))
431 goto bad_area;
432
433 index = pgd_index(addr);
434
435 pgd = cpu_get_pgd() + index;
436 pgd_k = init_mm.pgd + index;
437
438 if (pgd_none(*pgd_k))
439 goto bad_area;
440 if (!pgd_present(*pgd))
441 set_pgd(pgd, *pgd_k);
442
443 pud = pud_offset(pgd, addr);
444 pud_k = pud_offset(pgd_k, addr);
445
446 if (pud_none(*pud_k))
447 goto bad_area;
448 if (!pud_present(*pud))
449 set_pud(pud, *pud_k);
450
451 pmd = pmd_offset(pud, addr);
452 pmd_k = pmd_offset(pud_k, addr);
453
454#ifdef CONFIG_ARM_LPAE
455 /*
456 * Only one hardware entry per PMD with LPAE.
457 */
458 index = 0;
459#else
460 /*
461 * On ARM one Linux PGD entry contains two hardware entries (see page
462 * tables layout in pgtable.h). We normally guarantee that we always
463 * fill both L1 entries. But create_mapping() doesn't follow the rule.
464 * It can create inidividual L1 entries, so here we have to call
465 * pmd_none() check for the entry really corresponded to address, not
466 * for the first of pair.
467 */
468 index = (addr >> SECTION_SHIFT) & 1;
469#endif
470 if (pmd_none(pmd_k[index]))
471 goto bad_area;
472
473 copy_pmd(pmd, pmd_k);
474 return 0;
475
476bad_area:
477 do_bad_area(addr, fsr, regs);
478 return 0;
479}
480#else /* CONFIG_MMU */
481static int
482do_translation_fault(unsigned long addr, unsigned int fsr,
483 struct pt_regs *regs)
484{
485 return 0;
486}
487#endif /* CONFIG_MMU */
488
489/*
490 * Some section permission faults need to be handled gracefully.
491 * They can happen due to a __{get,put}_user during an oops.
492 */
493static int
494do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
495{
496 do_bad_area(addr, fsr, regs);
497 return 0;
498}
499
500/*
501 * This abort handler always returns "fault".
502 */
503static int
504do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
505{
506 return 1;
507}
508
509struct fsr_info {
510 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
511 int sig;
512 int code;
513 const char *name;
514};
515
516/* FSR definition */
517#ifdef CONFIG_ARM_LPAE
518#include "fsr-3level.c"
519#else
520#include "fsr-2level.c"
521#endif
522
523void __init
524hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
525 int sig, int code, const char *name)
526{
527 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
528 BUG();
529
530 fsr_info[nr].fn = fn;
531 fsr_info[nr].sig = sig;
532 fsr_info[nr].code = code;
533 fsr_info[nr].name = name;
534}
535
536/*
537 * Dispatch a data abort to the relevant handler.
538 */
539asmlinkage void __exception
540do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
541{
542 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
543 struct siginfo info;
544
545 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
546 return;
547
548 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
549 inf->name, fsr, addr);
550
551 info.si_signo = inf->sig;
552 info.si_errno = 0;
553 info.si_code = inf->code;
554 info.si_addr = (void __user *)addr;
555 arm_notify_die("", regs, &info, fsr, 0);
556}
557
558void __init
559hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
560 int sig, int code, const char *name)
561{
562 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
563 BUG();
564
565 ifsr_info[nr].fn = fn;
566 ifsr_info[nr].sig = sig;
567 ifsr_info[nr].code = code;
568 ifsr_info[nr].name = name;
569}
570
571asmlinkage void __exception
572do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
573{
574 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
575 struct siginfo info;
576
577 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
578 return;
579
580 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
581 inf->name, ifsr, addr);
582
583 info.si_signo = inf->sig;
584 info.si_errno = 0;
585 info.si_code = inf->code;
586 info.si_addr = (void __user *)addr;
587 arm_notify_die("", regs, &info, ifsr, 0);
588}
589
590#ifndef CONFIG_ARM_LPAE
591static int __init exceptions_init(void)
592{
593 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
594 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
595 "I-cache maintenance fault");
596 }
597
598 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
599 /*
600 * TODO: Access flag faults introduced in ARMv6K.
601 * Runtime check for 'K' extension is needed
602 */
603 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
604 "section access flag fault");
605 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
606 "section access flag fault");
607 }
608
609 return 0;
610}
611
612arch_initcall(exceptions_init);
613#endif