1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/arch/arm/mm/fault.c
4 *
5 * Copyright (C) 1995 Linus Torvalds
6 * Modifications for ARM processor (c) 1995-2004 Russell King
7 */
8 #include <linux/extable.h>
9 #include <linux/signal.h>
10 #include <linux/mm.h>
11 #include <linux/hardirq.h>
12 #include <linux/init.h>
13 #include <linux/kprobes.h>
14 #include <linux/uaccess.h>
15 #include <linux/page-flags.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/debug.h>
18 #include <linux/highmem.h>
19 #include <linux/perf_event.h>
20 #include <linux/kfence.h>
21
22 #include <asm/system_misc.h>
23 #include <asm/system_info.h>
24 #include <asm/tlbflush.h>
25
26 #include "fault.h"
27
28 #ifdef CONFIG_MMU
29
30 bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
31 {
32 unsigned long addr = (unsigned long)unsafe_src;
33
34 return addr >= TASK_SIZE && ULONG_MAX - addr >= size;
35 }
36
37 /*
38 * This is useful to dump out the page tables associated with
39 * 'addr' in mm 'mm'.
40 */
41 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
42 {
43 pgd_t *pgd;
44
45 if (!mm)
46 mm = &init_mm;
47
48 pgd = pgd_offset(mm, addr);
49 printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
50
51 do {
52 p4d_t *p4d;
53 pud_t *pud;
54 pmd_t *pmd;
55 pte_t *pte;
56
57 p4d = p4d_offset(pgd, addr);
58 if (p4d_none(*p4d))
59 break;
60
61 if (p4d_bad(*p4d)) {
62 pr_cont("(bad)");
63 break;
64 }
65
66 pud = pud_offset(p4d, addr);
67 if (PTRS_PER_PUD != 1)
68 pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
69
70 if (pud_none(*pud))
71 break;
72
73 if (pud_bad(*pud)) {
74 pr_cont("(bad)");
75 break;
76 }
77
78 pmd = pmd_offset(pud, addr);
79 if (PTRS_PER_PMD != 1)
80 pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
81
82 if (pmd_none(*pmd))
83 break;
84
85 if (pmd_bad(*pmd)) {
86 pr_cont("(bad)");
87 break;
88 }
89
90 /* We must not map this if we have highmem enabled */
91 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
92 break;
93
94 pte = pte_offset_map(pmd, addr);
95 if (!pte)
96 break;
97
98 pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
99 #ifndef CONFIG_ARM_LPAE
100 pr_cont(", *ppte=%08llx",
101 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
102 #endif
103 pte_unmap(pte);
104 } while(0);
105
106 pr_cont("\n");
107 }
108 #else /* CONFIG_MMU */
109 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
110 { }
111 #endif /* CONFIG_MMU */
112
113 static inline bool is_write_fault(unsigned int fsr)
114 {
115 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
116 }
117
118 static inline bool is_translation_fault(unsigned int fsr)
119 {
120 int fs = fsr_fs(fsr);
121 #ifdef CONFIG_ARM_LPAE
122 if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
123 return true;
124 #else
125 if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
126 return true;
127 #endif
128 return false;
129 }
130
131 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
132 unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
134 {
135 bust_spinlocks(1);
136 pr_alert("8<--- cut here ---\n");
137 pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
138 msg, addr, fsr & FSR_LNX_PF ? "execute" :
139 fsr & FSR_WRITE ? "write" : "read");
140
141 show_pte(KERN_ALERT, mm, addr);
142 die("Oops", regs, fsr);
143 bust_spinlocks(0);
144 make_task_dead(SIGKILL);
145 }
146
147 /*
148 * Oops. The kernel tried to access some page that wasn't present.
149 */
150 static void
151 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
152 struct pt_regs *regs)
153 {
154 const char *msg;
155 /*
156 * Are we prepared to handle this kernel fault?
157 */
158 if (fixup_exception(regs))
159 return;
160
161 /*
162 * No handler, we'll have to terminate things with extreme prejudice.
163 */
164 if (addr < PAGE_SIZE) {
165 msg = "NULL pointer dereference";
166 } else {
167 if (is_translation_fault(fsr) &&
168 kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
169 return;
170
171 msg = "paging request";
172 }
173
174 die_kernel_fault(msg, mm, addr, fsr, regs);
175 }
176
177 /*
178 * Something tried to access memory that isn't in our memory map..
179 * User mode accesses just cause a SIGSEGV
180 */
181 static void
182 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
183 int code, struct pt_regs *regs)
184 {
185 struct task_struct *tsk = current;
186
187 if (addr > TASK_SIZE)
188 harden_branch_predictor();
189
190 #ifdef CONFIG_DEBUG_USER
191 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
192 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
193 pr_err("8<--- cut here ---\n");
194 pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
195 tsk->comm, sig, addr, fsr);
196 show_pte(KERN_ERR, tsk->mm, addr);
197 show_regs(regs);
198 }
199 #endif
200 #ifndef CONFIG_KUSER_HELPERS
201 if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
202 printk_ratelimited(KERN_DEBUG
203 "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
204 tsk->comm, addr);
205 #endif
206
207 tsk->thread.address = addr;
208 tsk->thread.error_code = fsr;
209 tsk->thread.trap_no = 14;
210 force_sig_fault(sig, code, (void __user *)addr);
211 }
212
213 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
214 {
215 struct task_struct *tsk = current;
216 struct mm_struct *mm = tsk->active_mm;
217
218 /*
219 * If we are in kernel mode at this point, we
220 * have no context to handle this fault with.
221 */
222 if (user_mode(regs))
223 __do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
224 else
225 __do_kernel_fault(mm, addr, fsr, regs);
226 }
227
228 #ifdef CONFIG_MMU
229 static inline bool is_permission_fault(unsigned int fsr)
230 {
231 int fs = fsr_fs(fsr);
232 #ifdef CONFIG_ARM_LPAE
233 if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
234 return true;
235 #else
236 if (fs == FS_L1_PERM || fs == FS_L2_PERM)
237 return true;
238 #endif
239 return false;
240 }
241
242 #ifdef CONFIG_CPU_TTBR0_PAN
243 static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
244 {
245 struct svc_pt_regs *svcregs;
246
247 /* If we are in user mode: permission granted */
248 if (user_mode(regs))
249 return true;
250
251 /* uaccess state saved above pt_regs on SVC exception entry */
252 svcregs = to_svc_pt_regs(regs);
253
254 return !(svcregs->ttbcr & TTBCR_EPD0);
255 }
256 #else
257 static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
258 {
259 return true;
260 }
261 #endif
262
263 static int __kprobes
264 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
265 {
266 struct mm_struct *mm = current->mm;
267 struct vm_area_struct *vma;
268 int sig, code;
269 vm_fault_t fault;
270 unsigned int flags = FAULT_FLAG_DEFAULT;
271 unsigned long vm_flags = VM_ACCESS_FLAGS;
272
273 if (kprobe_page_fault(regs, fsr))
274 return 0;
275
276
277 /* Enable interrupts if they were enabled in the parent context. */
278 if (interrupts_enabled(regs))
279 local_irq_enable();
280
281 /*
282 * If we're in an interrupt or have no user
283 * context, we must not take the fault..
284 */
285 if (faulthandler_disabled() || !mm)
286 goto no_context;
287
288 if (user_mode(regs))
289 flags |= FAULT_FLAG_USER;
290
291 if (is_write_fault(fsr)) {
292 flags |= FAULT_FLAG_WRITE;
293 vm_flags = VM_WRITE;
294 }
295
296 if (fsr & FSR_LNX_PF) {
297 vm_flags = VM_EXEC;
298
299 if (is_permission_fault(fsr) && !user_mode(regs))
300 die_kernel_fault("execution of memory",
301 mm, addr, fsr, regs);
302 }
303
304 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
305
306 /*
307 * Privileged access aborts with CONFIG_CPU_TTBR0_PAN enabled are
308 * routed via the translation fault mechanism. Check whether uaccess
309 * is disabled while in kernel mode.
310 */
311 if (!ttbr0_usermode_access_allowed(regs))
312 goto no_context;
313
314 if (!(flags & FAULT_FLAG_USER))
315 goto lock_mmap;
316
317 vma = lock_vma_under_rcu(mm, addr);
318 if (!vma)
319 goto lock_mmap;
320
321 if (!(vma->vm_flags & vm_flags)) {
322 vma_end_read(vma);
323 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
324 fault = 0;
325 code = SEGV_ACCERR;
326 goto bad_area;
327 }
328 fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
329 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
330 vma_end_read(vma);
331
332 if (!(fault & VM_FAULT_RETRY)) {
333 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
334 goto done;
335 }
336 count_vm_vma_lock_event(VMA_LOCK_RETRY);
337 if (fault & VM_FAULT_MAJOR)
338 flags |= FAULT_FLAG_TRIED;
339
340 /* Quick path to respond to signals */
341 if (fault_signal_pending(fault, regs)) {
342 if (!user_mode(regs))
343 goto no_context;
344 return 0;
345 }
346 lock_mmap:
347
348 retry:
349 vma = lock_mm_and_find_vma(mm, addr, regs);
350 if (unlikely(!vma)) {
351 fault = 0;
352 code = SEGV_MAPERR;
353 goto bad_area;
354 }
355
356 /*
357 * ok, we have a good vm_area for this memory access, check the
358 * permissions on the VMA allow for the fault which occurred.
359 */
360 if (!(vma->vm_flags & vm_flags)) {
361 mmap_read_unlock(mm);
362 fault = 0;
363 code = SEGV_ACCERR;
364 goto bad_area;
365 }
366
367 fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
368
369 /* If we need to retry but a fatal signal is pending, handle the
370 * signal first. We do not need to release the mmap_lock because
371 * it would already be released in __lock_page_or_retry in
372 * mm/filemap.c. */
373 if (fault_signal_pending(fault, regs)) {
374 if (!user_mode(regs))
375 goto no_context;
376 return 0;
377 }
378
379 /* The fault is fully completed (including releasing mmap lock) */
380 if (fault & VM_FAULT_COMPLETED)
381 return 0;
382
383 if (!(fault & VM_FAULT_ERROR)) {
384 if (fault & VM_FAULT_RETRY) {
385 flags |= FAULT_FLAG_TRIED;
386 goto retry;
387 }
388 }
389
390 mmap_read_unlock(mm);
391 done:
392
393 /* Handle the "normal" case first */
394 if (likely(!(fault & VM_FAULT_ERROR)))
395 return 0;
396
397 code = SEGV_MAPERR;
398 bad_area:
399 /*
400 * If we are in kernel mode at this point, we
401 * have no context to handle this fault with.
402 */
403 if (!user_mode(regs))
404 goto no_context;
405
406 if (fault & VM_FAULT_OOM) {
407 /*
408 * We ran out of memory, call the OOM killer, and return to
409 * userspace (which will retry the fault, or kill us if we
410 * got oom-killed)
411 */
412 pagefault_out_of_memory();
413 return 0;
414 }
415
416 if (fault & VM_FAULT_SIGBUS) {
417 /*
418 * We had some memory, but were unable to
419 * successfully fix up this page fault.
420 */
421 sig = SIGBUS;
422 code = BUS_ADRERR;
423 } else {
424 /*
425 * Something tried to access memory that
426 * isn't in our memory map..
427 */
428 sig = SIGSEGV;
429 }
430
431 __do_user_fault(addr, fsr, sig, code, regs);
432 return 0;
433
434 no_context:
435 __do_kernel_fault(mm, addr, fsr, regs);
436 return 0;
437 }
438 #else /* CONFIG_MMU */
439 static int
440 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
441 {
442 return 0;
443 }
444 #endif /* CONFIG_MMU */
445
446 /*
447 * First Level Translation Fault Handler
448 *
449 * We enter here because the first level page table doesn't contain
450 * a valid entry for the address.
451 *
452 * If the address is in kernel space (>= TASK_SIZE), then we are
453 * probably faulting in the vmalloc() area.
454 *
455 * If the init_task's first level page tables contains the relevant
456 * entry, we copy the it to this task. If not, we send the process
457 * a signal, fixup the exception, or oops the kernel.
458 *
459 * NOTE! We MUST NOT take any locks for this case. We may be in an
460 * interrupt or a critical region, and should only copy the information
461 * from the master page table, nothing more.
462 */
463 #ifdef CONFIG_MMU
464 static int __kprobes
465 do_translation_fault(unsigned long addr, unsigned int fsr,
466 struct pt_regs *regs)
467 {
468 unsigned int index;
469 pgd_t *pgd, *pgd_k;
470 p4d_t *p4d, *p4d_k;
471 pud_t *pud, *pud_k;
472 pmd_t *pmd, *pmd_k;
473
474 if (addr < TASK_SIZE)
475 return do_page_fault(addr, fsr, regs);
476
477 if (user_mode(regs))
478 goto bad_area;
479
480 index = pgd_index(addr);
481
482 pgd = cpu_get_pgd() + index;
483 pgd_k = init_mm.pgd + index;
484
485 p4d = p4d_offset(pgd, addr);
486 p4d_k = p4d_offset(pgd_k, addr);
487
488 if (p4d_none(*p4d_k))
489 goto bad_area;
490 if (!p4d_present(*p4d))
491 set_p4d(p4d, *p4d_k);
492
493 pud = pud_offset(p4d, addr);
494 pud_k = pud_offset(p4d_k, addr);
495
496 if (pud_none(*pud_k))
497 goto bad_area;
498 if (!pud_present(*pud))
499 set_pud(pud, *pud_k);
500
501 pmd = pmd_offset(pud, addr);
502 pmd_k = pmd_offset(pud_k, addr);
503
504 #ifdef CONFIG_ARM_LPAE
505 /*
506 * Only one hardware entry per PMD with LPAE.
507 */
508 index = 0;
509 #else
510 /*
511 * On ARM one Linux PGD entry contains two hardware entries (see page
512 * tables layout in pgtable.h). We normally guarantee that we always
513 * fill both L1 entries. But create_mapping() doesn't follow the rule.
514 * It can create inidividual L1 entries, so here we have to call
515 * pmd_none() check for the entry really corresponded to address, not
516 * for the first of pair.
517 */
518 index = (addr >> SECTION_SHIFT) & 1;
519 #endif
520 if (pmd_none(pmd_k[index]))
521 goto bad_area;
522
523 copy_pmd(pmd, pmd_k);
524 return 0;
525
526 bad_area:
527 do_bad_area(addr, fsr, regs);
528 return 0;
529 }
530 #else /* CONFIG_MMU */
531 static int
532 do_translation_fault(unsigned long addr, unsigned int fsr,
533 struct pt_regs *regs)
534 {
535 return 0;
536 }
537 #endif /* CONFIG_MMU */
538
539 /*
540 * Some section permission faults need to be handled gracefully.
541 * They can happen due to a __{get,put}_user during an oops.
542 */
543 #ifndef CONFIG_ARM_LPAE
544 static int
545 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
546 {
547 do_bad_area(addr, fsr, regs);
548 return 0;
549 }
550 #endif /* CONFIG_ARM_LPAE */
551
552 /*
553 * This abort handler always returns "fault".
554 */
555 static int
556 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
557 {
558 return 1;
559 }
560
561 struct fsr_info {
562 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
563 int sig;
564 int code;
565 const char *name;
566 };
567
568 /* FSR definition */
569 #ifdef CONFIG_ARM_LPAE
570 #include "fsr-3level.c"
571 #else
572 #include "fsr-2level.c"
573 #endif
574
575 void __init
576 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
577 int sig, int code, const char *name)
578 {
579 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
580 BUG();
581
582 fsr_info[nr].fn = fn;
583 fsr_info[nr].sig = sig;
584 fsr_info[nr].code = code;
585 fsr_info[nr].name = name;
586 }
587
588 /*
589 * Dispatch a data abort to the relevant handler.
590 */
591 asmlinkage void
592 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
593 {
594 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
595
596 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
597 return;
598
599 pr_alert("8<--- cut here ---\n");
600 pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
601 inf->name, fsr, addr);
602 show_pte(KERN_ALERT, current->mm, addr);
603
604 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
605 fsr, 0);
606 }
607
608 void __init
609 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
610 int sig, int code, const char *name)
611 {
612 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
613 BUG();
614
615 ifsr_info[nr].fn = fn;
616 ifsr_info[nr].sig = sig;
617 ifsr_info[nr].code = code;
618 ifsr_info[nr].name = name;
619 }
620
621 asmlinkage void
622 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
623 {
624 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
625
626 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
627 return;
628
629 pr_alert("8<--- cut here ---\n");
630 pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
631 inf->name, ifsr, addr);
632
633 arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
634 ifsr, 0);
635 }
636
637 /*
638 * Abort handler to be used only during first unmasking of asynchronous aborts
639 * on the boot CPU. This makes sure that the machine will not die if the
640 * firmware/bootloader left an imprecise abort pending for us to trip over.
641 */
642 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
643 struct pt_regs *regs)
644 {
645 pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
646 "first unmask, this is most likely caused by a "
647 "firmware/bootloader bug.\n", fsr);
648
649 return 0;
650 }
651
652 void __init early_abt_enable(void)
653 {
654 fsr_info[FSR_FS_AEA].fn = early_abort_handler;
655 local_abt_enable();
656 fsr_info[FSR_FS_AEA].fn = do_bad;
657 }
658
659 #ifndef CONFIG_ARM_LPAE
660 static int __init exceptions_init(void)
661 {
662 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
663 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
664 "I-cache maintenance fault");
665 }
666
667 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
668 /*
669 * TODO: Access flag faults introduced in ARMv6K.
670 * Runtime check for 'K' extension is needed
671 */
672 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
673 "section access flag fault");
674 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
675 "section access flag fault");
676 }
677
678 return 0;
679 }
680
681 arch_initcall(exceptions_init);
682 #endif
683
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.