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// 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
30bool 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 */
41void 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 */
109void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
110{ }
111#endif /* CONFIG_MMU */
112
113static inline bool is_write_fault(unsigned int fsr)
114{
115 return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
116}
117
118static 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
131static 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 */
150static 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 */
181static 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
213void 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
229static 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
243static 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
257static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
258{
259 return true;
260}
261#endif
262
263static int __kprobes
264do_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 }
346lock_mmap:
347
348retry:
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);
391done:
392
393 /* Handle the "normal" case first */
394 if (likely(!(fault & VM_FAULT_ERROR)))
395 return 0;
396
397 code = SEGV_MAPERR;
398bad_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
434no_context:
435 __do_kernel_fault(mm, addr, fsr, regs);
436 return 0;
437}
438#else /* CONFIG_MMU */
439static int
440do_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
464static int __kprobes
465do_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
526bad_area:
527 do_bad_area(addr, fsr, regs);
528 return 0;
529}
530#else /* CONFIG_MMU */
531static int
532do_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
544static int
545do_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 */
555static int
556do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
557{
558 return 1;
559}
560
561struct 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
575void __init
576hook_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 */
591asmlinkage void
592do_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
608void __init
609hook_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
621asmlinkage void
622do_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 */
642static 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
652void __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
660static 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
681arch_initcall(exceptions_init);
682#endif