2 * Based on arch/arm/mm/fault.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 1995-2004 Russell King
6 * Copyright (C) 2012 ARM Ltd.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program. If not, see <http://www.gnu.org/licenses/>.
21 #include <linux/extable.h>
22 #include <linux/signal.h>
24 #include <linux/hardirq.h>
25 #include <linux/init.h>
26 #include <linux/kprobes.h>
27 #include <linux/uaccess.h>
28 #include <linux/page-flags.h>
29 #include <linux/sched/signal.h>
30 #include <linux/sched/debug.h>
31 #include <linux/highmem.h>
32 #include <linux/perf_event.h>
33 #include <linux/preempt.h>
36 #include <asm/cpufeature.h>
37 #include <asm/exception.h>
38 #include <asm/debug-monitors.h>
40 #include <asm/sysreg.h>
41 #include <asm/system_misc.h>
42 #include <asm/pgtable.h>
43 #include <asm/tlbflush.h>
45 static const char *fault_name(unsigned int esr
);
48 static inline int notify_page_fault(struct pt_regs
*regs
, unsigned int esr
)
52 /* kprobe_running() needs smp_processor_id() */
53 if (!user_mode(regs
)) {
55 if (kprobe_running() && kprobe_fault_handler(regs
, esr
))
63 static inline int notify_page_fault(struct pt_regs
*regs
, unsigned int esr
)
70 * Dump out the page tables associated with 'addr' in mm 'mm'.
72 void show_pte(struct mm_struct
*mm
, unsigned long addr
)
79 pr_alert("pgd = %p\n", mm
->pgd
);
80 pgd
= pgd_offset(mm
, addr
);
81 pr_alert("[%08lx] *pgd=%016llx", addr
, pgd_val(*pgd
));
88 if (pgd_none(*pgd
) || pgd_bad(*pgd
))
91 pud
= pud_offset(pgd
, addr
);
92 pr_cont(", *pud=%016llx", pud_val(*pud
));
93 if (pud_none(*pud
) || pud_bad(*pud
))
96 pmd
= pmd_offset(pud
, addr
);
97 pr_cont(", *pmd=%016llx", pmd_val(*pmd
));
98 if (pmd_none(*pmd
) || pmd_bad(*pmd
))
101 pte
= pte_offset_map(pmd
, addr
);
102 pr_cont(", *pte=%016llx", pte_val(*pte
));
109 #ifdef CONFIG_ARM64_HW_AFDBM
111 * This function sets the access flags (dirty, accessed), as well as write
112 * permission, and only to a more permissive setting.
114 * It needs to cope with hardware update of the accessed/dirty state by other
115 * agents in the system and can safely skip the __sync_icache_dcache() call as,
116 * like set_pte_at(), the PTE is never changed from no-exec to exec here.
118 * Returns whether or not the PTE actually changed.
120 int ptep_set_access_flags(struct vm_area_struct
*vma
,
121 unsigned long address
, pte_t
*ptep
,
122 pte_t entry
, int dirty
)
127 if (pte_same(*ptep
, entry
))
130 /* only preserve the access flags and write permission */
131 pte_val(entry
) &= PTE_AF
| PTE_WRITE
| PTE_DIRTY
;
134 * PTE_RDONLY is cleared by default in the asm below, so set it in
135 * back if necessary (read-only or clean PTE).
137 if (!pte_write(entry
) || !pte_sw_dirty(entry
))
138 pte_val(entry
) |= PTE_RDONLY
;
141 * Setting the flags must be done atomically to avoid racing with the
142 * hardware update of the access/dirty state.
144 asm volatile("// ptep_set_access_flags\n"
145 " prfm pstl1strm, %2\n"
147 " and %0, %0, %3 // clear PTE_RDONLY\n"
148 " orr %0, %0, %4 // set flags\n"
149 " stxr %w1, %0, %2\n"
151 : "=&r" (old_pteval
), "=&r" (tmp
), "+Q" (pte_val(*ptep
))
152 : "L" (~PTE_RDONLY
), "r" (pte_val(entry
)));
154 flush_tlb_fix_spurious_fault(vma
, address
);
159 static bool is_el1_instruction_abort(unsigned int esr
)
161 return ESR_ELx_EC(esr
) == ESR_ELx_EC_IABT_CUR
;
165 * The kernel tried to access some page that wasn't present.
167 static void __do_kernel_fault(struct mm_struct
*mm
, unsigned long addr
,
168 unsigned int esr
, struct pt_regs
*regs
)
171 * Are we prepared to handle this kernel fault?
172 * We are almost certainly not prepared to handle instruction faults.
174 if (!is_el1_instruction_abort(esr
) && fixup_exception(regs
))
178 * No handler, we'll have to terminate things with extreme prejudice.
181 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
182 (addr
< PAGE_SIZE
) ? "NULL pointer dereference" :
183 "paging request", addr
);
186 die("Oops", regs
, esr
);
192 * Something tried to access memory that isn't in our memory map. User mode
193 * accesses just cause a SIGSEGV
195 static void __do_user_fault(struct task_struct
*tsk
, unsigned long addr
,
196 unsigned int esr
, unsigned int sig
, int code
,
197 struct pt_regs
*regs
)
201 if (unhandled_signal(tsk
, sig
) && show_unhandled_signals_ratelimited()) {
202 pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
203 tsk
->comm
, task_pid_nr(tsk
), fault_name(esr
), sig
,
205 show_pte(tsk
->mm
, addr
);
209 tsk
->thread
.fault_address
= addr
;
210 tsk
->thread
.fault_code
= esr
;
214 si
.si_addr
= (void __user
*)addr
;
215 force_sig_info(sig
, &si
, tsk
);
218 static void do_bad_area(unsigned long addr
, unsigned int esr
, struct pt_regs
*regs
)
220 struct task_struct
*tsk
= current
;
221 struct mm_struct
*mm
= tsk
->active_mm
;
224 * If we are in kernel mode at this point, we have no context to
225 * handle this fault with.
228 __do_user_fault(tsk
, addr
, esr
, SIGSEGV
, SEGV_MAPERR
, regs
);
230 __do_kernel_fault(mm
, addr
, esr
, regs
);
233 #define VM_FAULT_BADMAP 0x010000
234 #define VM_FAULT_BADACCESS 0x020000
236 static int __do_page_fault(struct mm_struct
*mm
, unsigned long addr
,
237 unsigned int mm_flags
, unsigned long vm_flags
,
238 struct task_struct
*tsk
)
240 struct vm_area_struct
*vma
;
243 vma
= find_vma(mm
, addr
);
244 fault
= VM_FAULT_BADMAP
;
247 if (unlikely(vma
->vm_start
> addr
))
251 * Ok, we have a good vm_area for this memory access, so we can handle
256 * Check that the permissions on the VMA allow for the fault which
259 if (!(vma
->vm_flags
& vm_flags
)) {
260 fault
= VM_FAULT_BADACCESS
;
264 return handle_mm_fault(vma
, addr
& PAGE_MASK
, mm_flags
);
267 if (vma
->vm_flags
& VM_GROWSDOWN
&& !expand_stack(vma
, addr
))
273 static inline bool is_permission_fault(unsigned int esr
, struct pt_regs
*regs
)
275 unsigned int ec
= ESR_ELx_EC(esr
);
276 unsigned int fsc_type
= esr
& ESR_ELx_FSC_TYPE
;
278 if (ec
!= ESR_ELx_EC_DABT_CUR
&& ec
!= ESR_ELx_EC_IABT_CUR
)
281 if (system_uses_ttbr0_pan())
282 return fsc_type
== ESR_ELx_FSC_FAULT
&&
283 (regs
->pstate
& PSR_PAN_BIT
);
285 return fsc_type
== ESR_ELx_FSC_PERM
;
288 static bool is_el0_instruction_abort(unsigned int esr
)
290 return ESR_ELx_EC(esr
) == ESR_ELx_EC_IABT_LOW
;
293 static int __kprobes
do_page_fault(unsigned long addr
, unsigned int esr
,
294 struct pt_regs
*regs
)
296 struct task_struct
*tsk
;
297 struct mm_struct
*mm
;
298 int fault
, sig
, code
;
299 unsigned long vm_flags
= VM_READ
| VM_WRITE
;
300 unsigned int mm_flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
302 if (notify_page_fault(regs
, esr
))
309 * If we're in an interrupt or have no user context, we must not take
312 if (faulthandler_disabled() || !mm
)
316 mm_flags
|= FAULT_FLAG_USER
;
318 if (is_el0_instruction_abort(esr
)) {
320 } else if ((esr
& ESR_ELx_WNR
) && !(esr
& ESR_ELx_CM
)) {
322 mm_flags
|= FAULT_FLAG_WRITE
;
325 if (addr
< USER_DS
&& is_permission_fault(esr
, regs
)) {
326 /* regs->orig_addr_limit may be 0 if we entered from EL0 */
327 if (regs
->orig_addr_limit
== KERNEL_DS
)
328 die("Accessing user space memory with fs=KERNEL_DS", regs
, esr
);
330 if (is_el1_instruction_abort(esr
))
331 die("Attempting to execute userspace memory", regs
, esr
);
333 if (!search_exception_tables(regs
->pc
))
334 die("Accessing user space memory outside uaccess.h routines", regs
, esr
);
338 * As per x86, we may deadlock here. However, since the kernel only
339 * validly references user space from well defined areas of the code,
340 * we can bug out early if this is from code which shouldn't.
342 if (!down_read_trylock(&mm
->mmap_sem
)) {
343 if (!user_mode(regs
) && !search_exception_tables(regs
->pc
))
346 down_read(&mm
->mmap_sem
);
349 * The above down_read_trylock() might have succeeded in which
350 * case, we'll have missed the might_sleep() from down_read().
353 #ifdef CONFIG_DEBUG_VM
354 if (!user_mode(regs
) && !search_exception_tables(regs
->pc
))
359 fault
= __do_page_fault(mm
, addr
, mm_flags
, vm_flags
, tsk
);
362 * If we need to retry but a fatal signal is pending, handle the
363 * signal first. We do not need to release the mmap_sem because it
364 * would already be released in __lock_page_or_retry in mm/filemap.c.
366 if ((fault
& VM_FAULT_RETRY
) && fatal_signal_pending(current
))
370 * Major/minor page fault accounting is only done on the initial
371 * attempt. If we go through a retry, it is extremely likely that the
372 * page will be found in page cache at that point.
375 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, addr
);
376 if (mm_flags
& FAULT_FLAG_ALLOW_RETRY
) {
377 if (fault
& VM_FAULT_MAJOR
) {
379 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, regs
,
383 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, regs
,
386 if (fault
& VM_FAULT_RETRY
) {
388 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
391 mm_flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
392 mm_flags
|= FAULT_FLAG_TRIED
;
397 up_read(&mm
->mmap_sem
);
400 * Handle the "normal" case first - VM_FAULT_MAJOR
402 if (likely(!(fault
& (VM_FAULT_ERROR
| VM_FAULT_BADMAP
|
403 VM_FAULT_BADACCESS
))))
407 * If we are in kernel mode at this point, we have no context to
408 * handle this fault with.
410 if (!user_mode(regs
))
413 if (fault
& VM_FAULT_OOM
) {
415 * We ran out of memory, call the OOM killer, and return to
416 * userspace (which will retry the fault, or kill us if we got
419 pagefault_out_of_memory();
423 if (fault
& VM_FAULT_SIGBUS
) {
425 * We had some memory, but were unable to successfully fix up
432 * Something tried to access memory that isn't in our memory
436 code
= fault
== VM_FAULT_BADACCESS
?
437 SEGV_ACCERR
: SEGV_MAPERR
;
440 __do_user_fault(tsk
, addr
, esr
, sig
, code
, regs
);
444 __do_kernel_fault(mm
, addr
, esr
, regs
);
449 * First Level Translation Fault Handler
451 * We enter here because the first level page table doesn't contain a valid
452 * entry for the address.
454 * If the address is in kernel space (>= TASK_SIZE), then we are probably
455 * faulting in the vmalloc() area.
457 * If the init_task's first level page tables contains the relevant entry, we
458 * copy the it to this task. If not, we send the process a signal, fixup the
459 * exception, or oops the kernel.
461 * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
462 * or a critical region, and should only copy the information from the master
463 * page table, nothing more.
465 static int __kprobes
do_translation_fault(unsigned long addr
,
467 struct pt_regs
*regs
)
469 if (addr
< TASK_SIZE
)
470 return do_page_fault(addr
, esr
, regs
);
472 do_bad_area(addr
, esr
, regs
);
476 static int do_alignment_fault(unsigned long addr
, unsigned int esr
,
477 struct pt_regs
*regs
)
479 do_bad_area(addr
, esr
, regs
);
484 * This abort handler always returns "fault".
486 static int do_bad(unsigned long addr
, unsigned int esr
, struct pt_regs
*regs
)
491 static const struct fault_info
{
492 int (*fn
)(unsigned long addr
, unsigned int esr
, struct pt_regs
*regs
);
497 { do_bad
, SIGBUS
, 0, "ttbr address size fault" },
498 { do_bad
, SIGBUS
, 0, "level 1 address size fault" },
499 { do_bad
, SIGBUS
, 0, "level 2 address size fault" },
500 { do_bad
, SIGBUS
, 0, "level 3 address size fault" },
501 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "level 0 translation fault" },
502 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "level 1 translation fault" },
503 { do_translation_fault
, SIGSEGV
, SEGV_MAPERR
, "level 2 translation fault" },
504 { do_page_fault
, SIGSEGV
, SEGV_MAPERR
, "level 3 translation fault" },
505 { do_bad
, SIGBUS
, 0, "unknown 8" },
506 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 1 access flag fault" },
507 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 2 access flag fault" },
508 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 3 access flag fault" },
509 { do_bad
, SIGBUS
, 0, "unknown 12" },
510 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 1 permission fault" },
511 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 2 permission fault" },
512 { do_page_fault
, SIGSEGV
, SEGV_ACCERR
, "level 3 permission fault" },
513 { do_bad
, SIGBUS
, 0, "synchronous external abort" },
514 { do_bad
, SIGBUS
, 0, "unknown 17" },
515 { do_bad
, SIGBUS
, 0, "unknown 18" },
516 { do_bad
, SIGBUS
, 0, "unknown 19" },
517 { do_bad
, SIGBUS
, 0, "synchronous external abort (translation table walk)" },
518 { do_bad
, SIGBUS
, 0, "synchronous external abort (translation table walk)" },
519 { do_bad
, SIGBUS
, 0, "synchronous external abort (translation table walk)" },
520 { do_bad
, SIGBUS
, 0, "synchronous external abort (translation table walk)" },
521 { do_bad
, SIGBUS
, 0, "synchronous parity error" },
522 { do_bad
, SIGBUS
, 0, "unknown 25" },
523 { do_bad
, SIGBUS
, 0, "unknown 26" },
524 { do_bad
, SIGBUS
, 0, "unknown 27" },
525 { do_bad
, SIGBUS
, 0, "synchronous parity error (translation table walk)" },
526 { do_bad
, SIGBUS
, 0, "synchronous parity error (translation table walk)" },
527 { do_bad
, SIGBUS
, 0, "synchronous parity error (translation table walk)" },
528 { do_bad
, SIGBUS
, 0, "synchronous parity error (translation table walk)" },
529 { do_bad
, SIGBUS
, 0, "unknown 32" },
530 { do_alignment_fault
, SIGBUS
, BUS_ADRALN
, "alignment fault" },
531 { do_bad
, SIGBUS
, 0, "unknown 34" },
532 { do_bad
, SIGBUS
, 0, "unknown 35" },
533 { do_bad
, SIGBUS
, 0, "unknown 36" },
534 { do_bad
, SIGBUS
, 0, "unknown 37" },
535 { do_bad
, SIGBUS
, 0, "unknown 38" },
536 { do_bad
, SIGBUS
, 0, "unknown 39" },
537 { do_bad
, SIGBUS
, 0, "unknown 40" },
538 { do_bad
, SIGBUS
, 0, "unknown 41" },
539 { do_bad
, SIGBUS
, 0, "unknown 42" },
540 { do_bad
, SIGBUS
, 0, "unknown 43" },
541 { do_bad
, SIGBUS
, 0, "unknown 44" },
542 { do_bad
, SIGBUS
, 0, "unknown 45" },
543 { do_bad
, SIGBUS
, 0, "unknown 46" },
544 { do_bad
, SIGBUS
, 0, "unknown 47" },
545 { do_bad
, SIGBUS
, 0, "TLB conflict abort" },
546 { do_bad
, SIGBUS
, 0, "unknown 49" },
547 { do_bad
, SIGBUS
, 0, "unknown 50" },
548 { do_bad
, SIGBUS
, 0, "unknown 51" },
549 { do_bad
, SIGBUS
, 0, "implementation fault (lockdown abort)" },
550 { do_bad
, SIGBUS
, 0, "implementation fault (unsupported exclusive)" },
551 { do_bad
, SIGBUS
, 0, "unknown 54" },
552 { do_bad
, SIGBUS
, 0, "unknown 55" },
553 { do_bad
, SIGBUS
, 0, "unknown 56" },
554 { do_bad
, SIGBUS
, 0, "unknown 57" },
555 { do_bad
, SIGBUS
, 0, "unknown 58" },
556 { do_bad
, SIGBUS
, 0, "unknown 59" },
557 { do_bad
, SIGBUS
, 0, "unknown 60" },
558 { do_bad
, SIGBUS
, 0, "section domain fault" },
559 { do_bad
, SIGBUS
, 0, "page domain fault" },
560 { do_bad
, SIGBUS
, 0, "unknown 63" },
563 static const char *fault_name(unsigned int esr
)
565 const struct fault_info
*inf
= fault_info
+ (esr
& 63);
570 * Dispatch a data abort to the relevant handler.
572 asmlinkage
void __exception
do_mem_abort(unsigned long addr
, unsigned int esr
,
573 struct pt_regs
*regs
)
575 const struct fault_info
*inf
= fault_info
+ (esr
& 63);
578 if (!inf
->fn(addr
, esr
, regs
))
581 pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
582 inf
->name
, esr
, addr
);
584 info
.si_signo
= inf
->sig
;
586 info
.si_code
= inf
->code
;
587 info
.si_addr
= (void __user
*)addr
;
588 arm64_notify_die("", regs
, &info
, esr
);
592 * Handle stack alignment exceptions.
594 asmlinkage
void __exception
do_sp_pc_abort(unsigned long addr
,
596 struct pt_regs
*regs
)
599 struct task_struct
*tsk
= current
;
601 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGBUS
))
602 pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
603 tsk
->comm
, task_pid_nr(tsk
),
604 esr_get_class_string(esr
), (void *)regs
->pc
,
607 info
.si_signo
= SIGBUS
;
609 info
.si_code
= BUS_ADRALN
;
610 info
.si_addr
= (void __user
*)addr
;
611 arm64_notify_die("Oops - SP/PC alignment exception", regs
, &info
, esr
);
614 int __init
early_brk64(unsigned long addr
, unsigned int esr
,
615 struct pt_regs
*regs
);
618 * __refdata because early_brk64 is __init, but the reference to it is
619 * clobbered at arch_initcall time.
620 * See traps.c and debug-monitors.c:debug_traps_init().
622 static struct fault_info __refdata debug_fault_info
[] = {
623 { do_bad
, SIGTRAP
, TRAP_HWBKPT
, "hardware breakpoint" },
624 { do_bad
, SIGTRAP
, TRAP_HWBKPT
, "hardware single-step" },
625 { do_bad
, SIGTRAP
, TRAP_HWBKPT
, "hardware watchpoint" },
626 { do_bad
, SIGBUS
, 0, "unknown 3" },
627 { do_bad
, SIGTRAP
, TRAP_BRKPT
, "aarch32 BKPT" },
628 { do_bad
, SIGTRAP
, 0, "aarch32 vector catch" },
629 { early_brk64
, SIGTRAP
, TRAP_BRKPT
, "aarch64 BRK" },
630 { do_bad
, SIGBUS
, 0, "unknown 7" },
633 void __init
hook_debug_fault_code(int nr
,
634 int (*fn
)(unsigned long, unsigned int, struct pt_regs
*),
635 int sig
, int code
, const char *name
)
637 BUG_ON(nr
< 0 || nr
>= ARRAY_SIZE(debug_fault_info
));
639 debug_fault_info
[nr
].fn
= fn
;
640 debug_fault_info
[nr
].sig
= sig
;
641 debug_fault_info
[nr
].code
= code
;
642 debug_fault_info
[nr
].name
= name
;
645 asmlinkage
int __exception
do_debug_exception(unsigned long addr
,
647 struct pt_regs
*regs
)
649 const struct fault_info
*inf
= debug_fault_info
+ DBG_ESR_EVT(esr
);
654 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
655 * already disabled to preserve the last enabled/disabled addresses.
657 if (interrupts_enabled(regs
))
658 trace_hardirqs_off();
660 if (!inf
->fn(addr
, esr
, regs
)) {
663 pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
664 inf
->name
, esr
, addr
);
666 info
.si_signo
= inf
->sig
;
668 info
.si_code
= inf
->code
;
669 info
.si_addr
= (void __user
*)addr
;
670 arm64_notify_die("", regs
, &info
, 0);
674 if (interrupts_enabled(regs
))
679 NOKPROBE_SYMBOL(do_debug_exception
);
681 #ifdef CONFIG_ARM64_PAN
682 int cpu_enable_pan(void *__unused
)
685 * We modify PSTATE. This won't work from irq context as the PSTATE
686 * is discarded once we return from the exception.
688 WARN_ON_ONCE(in_interrupt());
690 config_sctlr_el1(SCTLR_EL1_SPAN
, 0);
691 asm(SET_PSTATE_PAN(1));
694 #endif /* CONFIG_ARM64_PAN */