2 * linux/arch/arm/mm/fault.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Modifications for ARM processor (c) 1995-2004 Russell King
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.
11 #include <linux/module.h>
12 #include <linux/signal.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>
23 #include <asm/system.h>
24 #include <asm/pgtable.h>
25 #include <asm/tlbflush.h>
30 * Fault status register encodings. We steal bit 31 for our own purposes.
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)
37 static inline int fsr_fs(unsigned int fsr
)
39 return (fsr
& FSR_FS3_0
) | (fsr
& FSR_FS4
) >> 6;
45 static inline int notify_page_fault(struct pt_regs
*regs
, unsigned int fsr
)
49 if (!user_mode(regs
)) {
50 /* kprobe_running() needs smp_processor_id() */
52 if (kprobe_running() && kprobe_fault_handler(regs
, fsr
))
60 static inline int notify_page_fault(struct pt_regs
*regs
, unsigned int fsr
)
67 * This is useful to dump out the page tables associated with
70 void show_pte(struct mm_struct
*mm
, unsigned long addr
)
77 printk(KERN_ALERT
"pgd = %p\n", mm
->pgd
);
78 pgd
= pgd_offset(mm
, addr
);
79 printk(KERN_ALERT
"[%08lx] *pgd=%08lx", addr
, pgd_val(*pgd
));
93 pmd
= pmd_offset(pgd
, addr
);
94 if (PTRS_PER_PMD
!= 1)
95 printk(", *pmd=%08lx", pmd_val(*pmd
));
105 /* We must not map this if we have highmem enabled */
106 if (PageHighMem(pfn_to_page(pmd_val(*pmd
) >> PAGE_SHIFT
)))
109 pte
= pte_offset_map(pmd
, addr
);
110 printk(", *pte=%08lx", pte_val(*pte
));
111 printk(", *ppte=%08lx", pte_val(pte
[-PTRS_PER_PTE
]));
117 #else /* CONFIG_MMU */
118 void show_pte(struct mm_struct
*mm
, unsigned long addr
)
120 #endif /* CONFIG_MMU */
123 * Oops. The kernel tried to access some page that wasn't present.
126 __do_kernel_fault(struct mm_struct
*mm
, unsigned long addr
, unsigned int fsr
,
127 struct pt_regs
*regs
)
130 * Are we prepared to handle this kernel fault?
132 if (fixup_exception(regs
))
136 * No handler, we'll have to terminate things with extreme prejudice.
140 "Unable to handle kernel %s at virtual address %08lx\n",
141 (addr
< PAGE_SIZE
) ? "NULL pointer dereference" :
142 "paging request", addr
);
145 die("Oops", regs
, fsr
);
151 * Something tried to access memory that isn't in our memory map..
152 * User mode accesses just cause a SIGSEGV
155 __do_user_fault(struct task_struct
*tsk
, unsigned long addr
,
156 unsigned int fsr
, unsigned int sig
, int code
,
157 struct pt_regs
*regs
)
161 #ifdef CONFIG_DEBUG_USER
162 if (user_debug
& UDBG_SEGV
) {
163 printk(KERN_DEBUG
"%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
164 tsk
->comm
, sig
, addr
, fsr
);
165 show_pte(tsk
->mm
, addr
);
170 tsk
->thread
.address
= addr
;
171 tsk
->thread
.error_code
= fsr
;
172 tsk
->thread
.trap_no
= 14;
176 si
.si_addr
= (void __user
*)addr
;
177 force_sig_info(sig
, &si
, tsk
);
180 void do_bad_area(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
182 struct task_struct
*tsk
= current
;
183 struct mm_struct
*mm
= tsk
->active_mm
;
186 * If we are in kernel mode at this point, we
187 * have no context to handle this fault with.
190 __do_user_fault(tsk
, addr
, fsr
, SIGSEGV
, SEGV_MAPERR
, regs
);
192 __do_kernel_fault(mm
, addr
, fsr
, regs
);
196 #define VM_FAULT_BADMAP 0x010000
197 #define VM_FAULT_BADACCESS 0x020000
200 * Check that the permissions on the VMA allow for the fault which occurred.
201 * If we encountered a write fault, we must have write permission, otherwise
202 * we allow any permission.
204 static inline bool access_error(unsigned int fsr
, struct vm_area_struct
*vma
)
206 unsigned int mask
= VM_READ
| VM_WRITE
| VM_EXEC
;
210 if (fsr
& FSR_LNX_PF
)
213 return vma
->vm_flags
& mask
? false : true;
217 __do_page_fault(struct mm_struct
*mm
, unsigned long addr
, unsigned int fsr
,
218 struct task_struct
*tsk
)
220 struct vm_area_struct
*vma
;
223 vma
= find_vma(mm
, addr
);
224 fault
= VM_FAULT_BADMAP
;
227 if (unlikely(vma
->vm_start
> addr
))
231 * Ok, we have a good vm_area for this
232 * memory access, so we can handle it.
235 if (access_error(fsr
, vma
)) {
236 fault
= VM_FAULT_BADACCESS
;
241 * If for any reason at all we couldn't handle the fault, make
242 * sure we exit gracefully rather than endlessly redo the fault.
244 fault
= handle_mm_fault(mm
, vma
, addr
& PAGE_MASK
, (fsr
& FSR_WRITE
) ? FAULT_FLAG_WRITE
: 0);
245 if (unlikely(fault
& VM_FAULT_ERROR
))
247 if (fault
& VM_FAULT_MAJOR
)
254 if (vma
->vm_flags
& VM_GROWSDOWN
&& !expand_stack(vma
, addr
))
261 do_page_fault(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
263 struct task_struct
*tsk
;
264 struct mm_struct
*mm
;
265 int fault
, sig
, code
;
267 if (notify_page_fault(regs
, fsr
))
274 * If we're in an interrupt or have no user
275 * context, we must not take the fault..
277 if (in_atomic() || !mm
)
281 * As per x86, we may deadlock here. However, since the kernel only
282 * validly references user space from well defined areas of the code,
283 * we can bug out early if this is from code which shouldn't.
285 if (!down_read_trylock(&mm
->mmap_sem
)) {
286 if (!user_mode(regs
) && !search_exception_tables(regs
->ARM_pc
))
288 down_read(&mm
->mmap_sem
);
291 * The above down_read_trylock() might have succeeded in
292 * which case, we'll have missed the might_sleep() from
296 #ifdef CONFIG_DEBUG_VM
297 if (!user_mode(regs
) &&
298 !search_exception_tables(regs
->ARM_pc
))
303 fault
= __do_page_fault(mm
, addr
, fsr
, tsk
);
304 up_read(&mm
->mmap_sem
);
306 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, addr
);
307 if (fault
& VM_FAULT_MAJOR
)
308 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0, regs
, addr
);
309 else if (fault
& VM_FAULT_MINOR
)
310 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0, regs
, addr
);
313 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
315 if (likely(!(fault
& (VM_FAULT_ERROR
| VM_FAULT_BADMAP
| VM_FAULT_BADACCESS
))))
318 if (fault
& VM_FAULT_OOM
) {
320 * We ran out of memory, call the OOM killer, and return to
321 * userspace (which will retry the fault, or kill us if we
324 pagefault_out_of_memory();
329 * If we are in kernel mode at this point, we
330 * have no context to handle this fault with.
332 if (!user_mode(regs
))
335 if (fault
& VM_FAULT_SIGBUS
) {
337 * We had some memory, but were unable to
338 * successfully fix up this page fault.
344 * Something tried to access memory that
345 * isn't in our memory map..
348 code
= fault
== VM_FAULT_BADACCESS
?
349 SEGV_ACCERR
: SEGV_MAPERR
;
352 __do_user_fault(tsk
, addr
, fsr
, sig
, code
, regs
);
356 __do_kernel_fault(mm
, addr
, fsr
, regs
);
359 #else /* CONFIG_MMU */
361 do_page_fault(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
365 #endif /* CONFIG_MMU */
368 * First Level Translation Fault Handler
370 * We enter here because the first level page table doesn't contain
371 * a valid entry for the address.
373 * If the address is in kernel space (>= TASK_SIZE), then we are
374 * probably faulting in the vmalloc() area.
376 * If the init_task's first level page tables contains the relevant
377 * entry, we copy the it to this task. If not, we send the process
378 * a signal, fixup the exception, or oops the kernel.
380 * NOTE! We MUST NOT take any locks for this case. We may be in an
381 * interrupt or a critical region, and should only copy the information
382 * from the master page table, nothing more.
386 do_translation_fault(unsigned long addr
, unsigned int fsr
,
387 struct pt_regs
*regs
)
393 if (addr
< TASK_SIZE
)
394 return do_page_fault(addr
, fsr
, regs
);
396 index
= pgd_index(addr
);
399 * FIXME: CP15 C1 is write only on ARMv3 architectures.
401 pgd
= cpu_get_pgd() + index
;
402 pgd_k
= init_mm
.pgd
+ index
;
404 if (pgd_none(*pgd_k
))
407 if (!pgd_present(*pgd
))
408 set_pgd(pgd
, *pgd_k
);
410 pmd_k
= pmd_offset(pgd_k
, addr
);
411 pmd
= pmd_offset(pgd
, addr
);
413 if (pmd_none(*pmd_k
))
416 copy_pmd(pmd
, pmd_k
);
420 do_bad_area(addr
, fsr
, regs
);
423 #else /* CONFIG_MMU */
425 do_translation_fault(unsigned long addr
, unsigned int fsr
,
426 struct pt_regs
*regs
)
430 #endif /* CONFIG_MMU */
433 * Some section permission faults need to be handled gracefully.
434 * They can happen due to a __{get,put}_user during an oops.
437 do_sect_fault(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
439 do_bad_area(addr
, fsr
, regs
);
444 * This abort handler always returns "fault".
447 do_bad(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
452 static struct fsr_info
{
453 int (*fn
)(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
);
459 * The following are the standard ARMv3 and ARMv4 aborts. ARMv5
460 * defines these to be "precise" aborts.
462 { do_bad
, SIGSEGV
, 0, "vector exception" },
463 { do_bad
, SIGILL
, BUS_ADRALN
, "alignment exception" },
464 { do_bad
, SIGKILL
, 0, "terminal exception" },
465 { do_bad
, SIGILL
, BUS_ADRALN
, "alignment exception" },
466 /* Do we need runtime check ? */
467 #if __LINUX_ARM_ARCH__ < 6
468 { do_bad
, SIGBUS
, 0, "external abort on linefetch" },
470 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "I-cache maintenance fault" },
472 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "section translation fault" },
473 { do_bad
, SIGBUS
, 0, "external abort on linefetch" },
474 { do_page_fault
, SIGSEGV
, SEGV_MAPERR
, "page translation fault" },
475 { do_bad
, SIGBUS
, 0, "external abort on non-linefetch" },
476 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "section domain fault" },
477 { do_bad
, SIGBUS
, 0, "external abort on non-linefetch" },
478 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "page domain fault" },
479 { do_bad
, SIGBUS
, 0, "external abort on translation" },
480 { do_sect_fault
, SIGSEGV
, SEGV_ACCERR
, "section permission fault" },
481 { do_bad
, SIGBUS
, 0, "external abort on translation" },
482 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "page permission fault" },
484 * The following are "imprecise" aborts, which are signalled by bit
485 * 10 of the FSR, and may not be recoverable. These are only
486 * supported if the CPU abort handler supports bit 10.
488 { do_bad
, SIGBUS
, 0, "unknown 16" },
489 { do_bad
, SIGBUS
, 0, "unknown 17" },
490 { do_bad
, SIGBUS
, 0, "unknown 18" },
491 { do_bad
, SIGBUS
, 0, "unknown 19" },
492 { do_bad
, SIGBUS
, 0, "lock abort" }, /* xscale */
493 { do_bad
, SIGBUS
, 0, "unknown 21" },
494 { do_bad
, SIGBUS
, BUS_OBJERR
, "imprecise external abort" }, /* xscale */
495 { do_bad
, SIGBUS
, 0, "unknown 23" },
496 { do_bad
, SIGBUS
, 0, "dcache parity error" }, /* xscale */
497 { do_bad
, SIGBUS
, 0, "unknown 25" },
498 { do_bad
, SIGBUS
, 0, "unknown 26" },
499 { do_bad
, SIGBUS
, 0, "unknown 27" },
500 { do_bad
, SIGBUS
, 0, "unknown 28" },
501 { do_bad
, SIGBUS
, 0, "unknown 29" },
502 { do_bad
, SIGBUS
, 0, "unknown 30" },
503 { do_bad
, SIGBUS
, 0, "unknown 31" }
507 hook_fault_code(int nr
, int (*fn
)(unsigned long, unsigned int, struct pt_regs
*),
508 int sig
, const char *name
)
510 if (nr
>= 0 && nr
< ARRAY_SIZE(fsr_info
)) {
511 fsr_info
[nr
].fn
= fn
;
512 fsr_info
[nr
].sig
= sig
;
513 fsr_info
[nr
].name
= name
;
518 * Dispatch a data abort to the relevant handler.
520 asmlinkage
void __exception
521 do_DataAbort(unsigned long addr
, unsigned int fsr
, struct pt_regs
*regs
)
523 const struct fsr_info
*inf
= fsr_info
+ fsr_fs(fsr
);
526 if (!inf
->fn(addr
, fsr
& ~FSR_LNX_PF
, regs
))
529 printk(KERN_ALERT
"Unhandled fault: %s (0x%03x) at 0x%08lx\n",
530 inf
->name
, fsr
, addr
);
532 info
.si_signo
= inf
->sig
;
534 info
.si_code
= inf
->code
;
535 info
.si_addr
= (void __user
*)addr
;
536 arm_notify_die("", regs
, &info
, fsr
, 0);
540 static struct fsr_info ifsr_info
[] = {
541 { do_bad
, SIGBUS
, 0, "unknown 0" },
542 { do_bad
, SIGBUS
, 0, "unknown 1" },
543 { do_bad
, SIGBUS
, 0, "debug event" },
544 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "section access flag fault" },
545 { do_bad
, SIGBUS
, 0, "unknown 4" },
546 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "section translation fault" },
547 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "page access flag fault" },
548 { do_page_fault
, SIGSEGV
, SEGV_MAPERR
, "page translation fault" },
549 { do_bad
, SIGBUS
, 0, "external abort on non-linefetch" },
550 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "section domain fault" },
551 { do_bad
, SIGBUS
, 0, "unknown 10" },
552 { do_bad
, SIGSEGV
, SEGV_ACCERR
, "page domain fault" },
553 { do_bad
, SIGBUS
, 0, "external abort on translation" },
554 { do_sect_fault
, SIGSEGV
, SEGV_ACCERR
, "section permission fault" },
555 { do_bad
, SIGBUS
, 0, "external abort on translation" },
556 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "page permission fault" },
557 { do_bad
, SIGBUS
, 0, "unknown 16" },
558 { do_bad
, SIGBUS
, 0, "unknown 17" },
559 { do_bad
, SIGBUS
, 0, "unknown 18" },
560 { do_bad
, SIGBUS
, 0, "unknown 19" },
561 { do_bad
, SIGBUS
, 0, "unknown 20" },
562 { do_bad
, SIGBUS
, 0, "unknown 21" },
563 { do_bad
, SIGBUS
, 0, "unknown 22" },
564 { do_bad
, SIGBUS
, 0, "unknown 23" },
565 { do_bad
, SIGBUS
, 0, "unknown 24" },
566 { do_bad
, SIGBUS
, 0, "unknown 25" },
567 { do_bad
, SIGBUS
, 0, "unknown 26" },
568 { do_bad
, SIGBUS
, 0, "unknown 27" },
569 { do_bad
, SIGBUS
, 0, "unknown 28" },
570 { do_bad
, SIGBUS
, 0, "unknown 29" },
571 { do_bad
, SIGBUS
, 0, "unknown 30" },
572 { do_bad
, SIGBUS
, 0, "unknown 31" },
575 asmlinkage
void __exception
576 do_PrefetchAbort(unsigned long addr
, unsigned int ifsr
, struct pt_regs
*regs
)
578 const struct fsr_info
*inf
= ifsr_info
+ fsr_fs(ifsr
);
581 if (!inf
->fn(addr
, ifsr
| FSR_LNX_PF
, regs
))
584 printk(KERN_ALERT
"Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
585 inf
->name
, ifsr
, addr
);
587 info
.si_signo
= inf
->sig
;
589 info
.si_code
= inf
->code
;
590 info
.si_addr
= (void __user
*)addr
;
591 arm_notify_die("", regs
, &info
, ifsr
, 0);