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/module.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.h>
30 #include <linux/highmem.h>
31 #include <linux/perf_event.h>
32 #include <linux/preempt.h>
35 #include <asm/cpufeature.h>
36 #include <asm/exception.h>
37 #include <asm/debug-monitors.h>
39 #include <asm/sysreg.h>
40 #include <asm/system_misc.h>
41 #include <asm/pgtable.h>
42 #include <asm/tlbflush.h>
44 static const char *fault_name(unsigned int esr);
47 * Dump out the page tables associated with 'addr' in mm 'mm'.
49 void show_pte(struct mm_struct *mm, unsigned long addr)
56 pr_alert("pgd = %p\n", mm->pgd);
57 pgd = pgd_offset(mm, addr);
58 pr_alert("[%08lx] *pgd=%016llx", addr, pgd_val(*pgd));
65 if (pgd_none(*pgd) || pgd_bad(*pgd))
68 pud = pud_offset(pgd, addr);
69 pr_cont(", *pud=%016llx", pud_val(*pud));
70 if (pud_none(*pud) || pud_bad(*pud))
73 pmd = pmd_offset(pud, addr);
74 pr_cont(", *pmd=%016llx", pmd_val(*pmd));
75 if (pmd_none(*pmd) || pmd_bad(*pmd))
78 pte = pte_offset_map(pmd, addr);
79 pr_cont(", *pte=%016llx", pte_val(*pte));
86 #ifdef CONFIG_ARM64_HW_AFDBM
88 * This function sets the access flags (dirty, accessed), as well as write
89 * permission, and only to a more permissive setting.
91 * It needs to cope with hardware update of the accessed/dirty state by other
92 * agents in the system and can safely skip the __sync_icache_dcache() call as,
93 * like set_pte_at(), the PTE is never changed from no-exec to exec here.
95 * Returns whether or not the PTE actually changed.
97 int ptep_set_access_flags(struct vm_area_struct *vma,
98 unsigned long address, pte_t *ptep,
99 pte_t entry, int dirty)
104 if (pte_same(*ptep, entry))
107 /* only preserve the access flags and write permission */
108 pte_val(entry) &= PTE_AF | PTE_WRITE | PTE_DIRTY;
110 /* set PTE_RDONLY if actual read-only or clean PTE */
111 if (!pte_write(entry) || !pte_sw_dirty(entry))
112 pte_val(entry) |= PTE_RDONLY;
115 * Setting the flags must be done atomically to avoid racing with the
116 * hardware update of the access/dirty state. The PTE_RDONLY bit must
117 * be set to the most permissive (lowest value) of *ptep and entry
118 * (calculated as: a & b == ~(~a | ~b)).
120 pte_val(entry) ^= PTE_RDONLY;
121 asm volatile("// ptep_set_access_flags\n"
122 " prfm pstl1strm, %2\n"
124 " eor %0, %0, %3 // negate PTE_RDONLY in *ptep\n"
125 " orr %0, %0, %4 // set flags\n"
126 " eor %0, %0, %3 // negate final PTE_RDONLY\n"
127 " stxr %w1, %0, %2\n"
129 : "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
130 : "L" (PTE_RDONLY), "r" (pte_val(entry)));
132 flush_tlb_fix_spurious_fault(vma, address);
137 static bool is_el1_instruction_abort(unsigned int esr)
139 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
143 * The kernel tried to access some page that wasn't present.
145 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
146 unsigned int esr, struct pt_regs *regs)
149 * Are we prepared to handle this kernel fault?
150 * We are almost certainly not prepared to handle instruction faults.
152 if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
156 * No handler, we'll have to terminate things with extreme prejudice.
159 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
160 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
161 "paging request", addr);
164 die("Oops", regs, esr);
170 * Something tried to access memory that isn't in our memory map. User mode
171 * accesses just cause a SIGSEGV
173 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
174 unsigned int esr, unsigned int sig, int code,
175 struct pt_regs *regs)
179 if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
180 pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
181 tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,
183 show_pte(tsk->mm, addr);
187 tsk->thread.fault_address = addr;
188 tsk->thread.fault_code = esr;
192 si.si_addr = (void __user *)addr;
193 force_sig_info(sig, &si, tsk);
196 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
198 struct task_struct *tsk = current;
199 struct mm_struct *mm = tsk->active_mm;
202 * If we are in kernel mode at this point, we have no context to
203 * handle this fault with.
206 __do_user_fault(tsk, addr, esr, SIGSEGV, SEGV_MAPERR, regs);
208 __do_kernel_fault(mm, addr, esr, regs);
211 #define VM_FAULT_BADMAP 0x010000
212 #define VM_FAULT_BADACCESS 0x020000
214 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
215 unsigned int mm_flags, unsigned long vm_flags,
216 struct task_struct *tsk)
218 struct vm_area_struct *vma;
221 vma = find_vma(mm, addr);
222 fault = VM_FAULT_BADMAP;
225 if (unlikely(vma->vm_start > addr))
229 * Ok, we have a good vm_area for this memory access, so we can handle
234 * Check that the permissions on the VMA allow for the fault which
235 * occurred. If we encountered a write or exec fault, we must have
236 * appropriate permissions, otherwise we allow any permission.
238 if (!(vma->vm_flags & vm_flags)) {
239 fault = VM_FAULT_BADACCESS;
243 return handle_mm_fault(mm, vma, addr & PAGE_MASK, mm_flags);
246 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
252 static inline bool is_permission_fault(unsigned int esr, struct pt_regs *regs)
254 unsigned int ec = ESR_ELx_EC(esr);
255 unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
257 if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
260 if (system_uses_ttbr0_pan())
261 return fsc_type == ESR_ELx_FSC_FAULT &&
262 (regs->pstate & PSR_PAN_BIT);
264 return fsc_type == ESR_ELx_FSC_PERM;
267 static bool is_el0_instruction_abort(unsigned int esr)
269 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
272 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
273 struct pt_regs *regs)
275 struct task_struct *tsk;
276 struct mm_struct *mm;
277 int fault, sig, code;
278 unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
279 unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
284 /* Enable interrupts if they were enabled in the parent context. */
285 if (interrupts_enabled(regs))
289 * If we're in an interrupt or have no user context, we must not take
292 if (faulthandler_disabled() || !mm)
296 mm_flags |= FAULT_FLAG_USER;
298 if (is_el0_instruction_abort(esr)) {
300 } else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
302 mm_flags |= FAULT_FLAG_WRITE;
305 if (addr < USER_DS && is_permission_fault(esr, regs)) {
306 /* regs->orig_addr_limit may be 0 if we entered from EL0 */
307 if (regs->orig_addr_limit == KERNEL_DS)
308 die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
310 if (is_el1_instruction_abort(esr))
311 die("Attempting to execute userspace memory", regs, esr);
313 if (!search_exception_tables(regs->pc))
314 die("Accessing user space memory outside uaccess.h routines", regs, esr);
318 * As per x86, we may deadlock here. However, since the kernel only
319 * validly references user space from well defined areas of the code,
320 * we can bug out early if this is from code which shouldn't.
322 if (!down_read_trylock(&mm->mmap_sem)) {
323 if (!user_mode(regs) && !search_exception_tables(regs->pc))
326 down_read(&mm->mmap_sem);
329 * The above down_read_trylock() might have succeeded in which
330 * case, we'll have missed the might_sleep() from down_read().
333 #ifdef CONFIG_DEBUG_VM
334 if (!user_mode(regs) && !search_exception_tables(regs->pc))
339 fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
342 * If we need to retry but a fatal signal is pending, handle the
343 * signal first. We do not need to release the mmap_sem because it
344 * would already be released in __lock_page_or_retry in mm/filemap.c.
346 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
347 if (!user_mode(regs))
353 * Major/minor page fault accounting is only done on the initial
354 * attempt. If we go through a retry, it is extremely likely that the
355 * page will be found in page cache at that point.
358 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
359 if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
360 if (fault & VM_FAULT_MAJOR) {
362 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
366 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
369 if (fault & VM_FAULT_RETRY) {
371 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
374 mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
375 mm_flags |= FAULT_FLAG_TRIED;
380 up_read(&mm->mmap_sem);
383 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
385 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
386 VM_FAULT_BADACCESS))))
390 * If we are in kernel mode at this point, we have no context to
391 * handle this fault with.
393 if (!user_mode(regs))
396 if (fault & VM_FAULT_OOM) {
398 * We ran out of memory, call the OOM killer, and return to
399 * userspace (which will retry the fault, or kill us if we got
402 pagefault_out_of_memory();
406 if (fault & VM_FAULT_SIGBUS) {
408 * We had some memory, but were unable to successfully fix up
415 * Something tried to access memory that isn't in our memory
419 code = fault == VM_FAULT_BADACCESS ?
420 SEGV_ACCERR : SEGV_MAPERR;
423 __do_user_fault(tsk, addr, esr, sig, code, regs);
427 __do_kernel_fault(mm, addr, esr, regs);
432 * First Level Translation Fault Handler
434 * We enter here because the first level page table doesn't contain a valid
435 * entry for the address.
437 * If the address is in kernel space (>= TASK_SIZE), then we are probably
438 * faulting in the vmalloc() area.
440 * If the init_task's first level page tables contains the relevant entry, we
441 * copy the it to this task. If not, we send the process a signal, fixup the
442 * exception, or oops the kernel.
444 * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
445 * or a critical region, and should only copy the information from the master
446 * page table, nothing more.
448 static int __kprobes do_translation_fault(unsigned long addr,
450 struct pt_regs *regs)
452 if (addr < TASK_SIZE)
453 return do_page_fault(addr, esr, regs);
455 do_bad_area(addr, esr, regs);
460 * This abort handler always returns "fault".
462 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
467 static const struct fault_info {
468 int (*fn)(unsigned long addr, unsigned int esr, struct pt_regs *regs);
473 { do_bad, SIGBUS, 0, "ttbr address size fault" },
474 { do_bad, SIGBUS, 0, "level 1 address size fault" },
475 { do_bad, SIGBUS, 0, "level 2 address size fault" },
476 { do_bad, SIGBUS, 0, "level 3 address size fault" },
477 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 0 translation fault" },
478 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 1 translation fault" },
479 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 2 translation fault" },
480 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 3 translation fault" },
481 { do_bad, SIGBUS, 0, "unknown 8" },
482 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 access flag fault" },
483 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 access flag fault" },
484 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 access flag fault" },
485 { do_bad, SIGBUS, 0, "unknown 12" },
486 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 permission fault" },
487 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 permission fault" },
488 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 permission fault" },
489 { do_bad, SIGBUS, 0, "synchronous external abort" },
490 { do_bad, SIGBUS, 0, "unknown 17" },
491 { do_bad, SIGBUS, 0, "unknown 18" },
492 { do_bad, SIGBUS, 0, "unknown 19" },
493 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
494 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
495 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
496 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
497 { do_bad, SIGBUS, 0, "synchronous parity error" },
498 { do_bad, SIGBUS, 0, "unknown 25" },
499 { do_bad, SIGBUS, 0, "unknown 26" },
500 { do_bad, SIGBUS, 0, "unknown 27" },
501 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
502 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
503 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
504 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
505 { do_bad, SIGBUS, 0, "unknown 32" },
506 { do_bad, SIGBUS, BUS_ADRALN, "alignment fault" },
507 { do_bad, SIGBUS, 0, "unknown 34" },
508 { do_bad, SIGBUS, 0, "unknown 35" },
509 { do_bad, SIGBUS, 0, "unknown 36" },
510 { do_bad, SIGBUS, 0, "unknown 37" },
511 { do_bad, SIGBUS, 0, "unknown 38" },
512 { do_bad, SIGBUS, 0, "unknown 39" },
513 { do_bad, SIGBUS, 0, "unknown 40" },
514 { do_bad, SIGBUS, 0, "unknown 41" },
515 { do_bad, SIGBUS, 0, "unknown 42" },
516 { do_bad, SIGBUS, 0, "unknown 43" },
517 { do_bad, SIGBUS, 0, "unknown 44" },
518 { do_bad, SIGBUS, 0, "unknown 45" },
519 { do_bad, SIGBUS, 0, "unknown 46" },
520 { do_bad, SIGBUS, 0, "unknown 47" },
521 { do_bad, SIGBUS, 0, "TLB conflict abort" },
522 { do_bad, SIGBUS, 0, "unknown 49" },
523 { do_bad, SIGBUS, 0, "unknown 50" },
524 { do_bad, SIGBUS, 0, "unknown 51" },
525 { do_bad, SIGBUS, 0, "implementation fault (lockdown abort)" },
526 { do_bad, SIGBUS, 0, "implementation fault (unsupported exclusive)" },
527 { do_bad, SIGBUS, 0, "unknown 54" },
528 { do_bad, SIGBUS, 0, "unknown 55" },
529 { do_bad, SIGBUS, 0, "unknown 56" },
530 { do_bad, SIGBUS, 0, "unknown 57" },
531 { do_bad, SIGBUS, 0, "unknown 58" },
532 { do_bad, SIGBUS, 0, "unknown 59" },
533 { do_bad, SIGBUS, 0, "unknown 60" },
534 { do_bad, SIGBUS, 0, "section domain fault" },
535 { do_bad, SIGBUS, 0, "page domain fault" },
536 { do_bad, SIGBUS, 0, "unknown 63" },
539 static const char *fault_name(unsigned int esr)
541 const struct fault_info *inf = fault_info + (esr & 63);
546 * Dispatch a data abort to the relevant handler.
548 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
549 struct pt_regs *regs)
551 const struct fault_info *inf = fault_info + (esr & 63);
554 if (!inf->fn(addr, esr, regs))
557 pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
558 inf->name, esr, addr);
560 info.si_signo = inf->sig;
562 info.si_code = inf->code;
563 info.si_addr = (void __user *)addr;
564 arm64_notify_die("", regs, &info, esr);
568 * Handle stack alignment exceptions.
570 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
572 struct pt_regs *regs)
575 struct task_struct *tsk = current;
577 if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
578 pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
579 tsk->comm, task_pid_nr(tsk),
580 esr_get_class_string(esr), (void *)regs->pc,
583 info.si_signo = SIGBUS;
585 info.si_code = BUS_ADRALN;
586 info.si_addr = (void __user *)addr;
587 arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
590 int __init early_brk64(unsigned long addr, unsigned int esr,
591 struct pt_regs *regs);
594 * __refdata because early_brk64 is __init, but the reference to it is
595 * clobbered at arch_initcall time.
596 * See traps.c and debug-monitors.c:debug_traps_init().
598 static struct fault_info __refdata debug_fault_info[] = {
599 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware breakpoint" },
600 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware single-step" },
601 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware watchpoint" },
602 { do_bad, SIGBUS, 0, "unknown 3" },
603 { do_bad, SIGTRAP, TRAP_BRKPT, "aarch32 BKPT" },
604 { do_bad, SIGTRAP, 0, "aarch32 vector catch" },
605 { early_brk64, SIGTRAP, TRAP_BRKPT, "aarch64 BRK" },
606 { do_bad, SIGBUS, 0, "unknown 7" },
609 void __init hook_debug_fault_code(int nr,
610 int (*fn)(unsigned long, unsigned int, struct pt_regs *),
611 int sig, int code, const char *name)
613 BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
615 debug_fault_info[nr].fn = fn;
616 debug_fault_info[nr].sig = sig;
617 debug_fault_info[nr].code = code;
618 debug_fault_info[nr].name = name;
621 asmlinkage int __exception do_debug_exception(unsigned long addr,
623 struct pt_regs *regs)
625 const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
628 if (!inf->fn(addr, esr, regs))
631 pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
632 inf->name, esr, addr);
634 info.si_signo = inf->sig;
636 info.si_code = inf->code;
637 info.si_addr = (void __user *)addr;
638 arm64_notify_die("", regs, &info, 0);
643 #ifdef CONFIG_ARM64_PAN
644 int cpu_enable_pan(void *__unused)
647 * We modify PSTATE. This won't work from irq context as the PSTATE
648 * is discarded once we return from the exception.
650 WARN_ON_ONCE(in_interrupt());
652 config_sctlr_el1(SCTLR_EL1_SPAN, 0);
653 asm(SET_PSTATE_PAN(1));
656 #endif /* CONFIG_ARM64_PAN */
658 #ifdef CONFIG_ARM64_UAO
660 * Kernel threads have fs=KERNEL_DS by default, and don't need to call
661 * set_fs(), devtmpfs in particular relies on this behaviour.
662 * We need to enable the feature at runtime (instead of adding it to
663 * PSR_MODE_EL1h) as the feature may not be implemented by the cpu.
665 int cpu_enable_uao(void *__unused)
667 asm(SET_PSTATE_UAO(1));
670 #endif /* CONFIG_ARM64_UAO */