1 /*
2 * Based on arch/arm/mm/fault.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 1995-2004 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 *
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.
11 *
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.
16 *
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/>.
19 */
20
21 #include <linux/extable.h>
22 #include <linux/signal.h>
23 #include <linux/mm.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>
33
34 #include <asm/bug.h>
35 #include <asm/cpufeature.h>
36 #include <asm/exception.h>
37 #include <asm/debug-monitors.h>
38 #include <asm/esr.h>
39 #include <asm/sysreg.h>
40 #include <asm/system_misc.h>
41 #include <asm/pgtable.h>
42 #include <asm/tlbflush.h>
43
44 struct fault_info {
45 int (*fn)(unsigned long addr, unsigned int esr,
46 struct pt_regs *regs);
47 int sig;
48 int code;
49 const char *name;
50 };
51
52 static const struct fault_info fault_info[];
53
esr_to_fault_info(unsigned int esr)54 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
55 {
56 return fault_info + (esr & 63);
57 }
58
59 #ifdef CONFIG_KPROBES
notify_page_fault(struct pt_regs * regs,unsigned int esr)60 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
61 {
62 int ret = 0;
63
64 /* kprobe_running() needs smp_processor_id() */
65 if (!user_mode(regs)) {
66 preempt_disable();
67 if (kprobe_running() && kprobe_fault_handler(regs, esr))
68 ret = 1;
69 preempt_enable();
70 }
71
72 return ret;
73 }
74 #else
notify_page_fault(struct pt_regs * regs,unsigned int esr)75 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
76 {
77 return 0;
78 }
79 #endif
80
81 /*
82 * Dump out the page tables associated with 'addr' in mm 'mm'.
83 */
show_pte(struct mm_struct * mm,unsigned long addr)84 void show_pte(struct mm_struct *mm, unsigned long addr)
85 {
86 pgd_t *pgd;
87
88 if (!mm)
89 mm = &init_mm;
90
91 pr_alert("pgd = %p\n", mm->pgd);
92 pgd = pgd_offset(mm, addr);
93 pr_alert("[%08lx] *pgd=%016llx", addr, pgd_val(*pgd));
94
95 do {
96 pud_t *pud;
97 pmd_t *pmd;
98 pte_t *pte;
99
100 if (pgd_none(*pgd) || pgd_bad(*pgd))
101 break;
102
103 pud = pud_offset(pgd, addr);
104 pr_cont(", *pud=%016llx", pud_val(*pud));
105 if (pud_none(*pud) || pud_bad(*pud))
106 break;
107
108 pmd = pmd_offset(pud, addr);
109 pr_cont(", *pmd=%016llx", pmd_val(*pmd));
110 if (pmd_none(*pmd) || pmd_bad(*pmd))
111 break;
112
113 pte = pte_offset_map(pmd, addr);
114 pr_cont(", *pte=%016llx", pte_val(*pte));
115 pte_unmap(pte);
116 } while(0);
117
118 pr_cont("\n");
119 }
120
121 #ifdef CONFIG_ARM64_HW_AFDBM
122 /*
123 * This function sets the access flags (dirty, accessed), as well as write
124 * permission, and only to a more permissive setting.
125 *
126 * It needs to cope with hardware update of the accessed/dirty state by other
127 * agents in the system and can safely skip the __sync_icache_dcache() call as,
128 * like set_pte_at(), the PTE is never changed from no-exec to exec here.
129 *
130 * Returns whether or not the PTE actually changed.
131 */
ptep_set_access_flags(struct vm_area_struct * vma,unsigned long address,pte_t * ptep,pte_t entry,int dirty)132 int ptep_set_access_flags(struct vm_area_struct *vma,
133 unsigned long address, pte_t *ptep,
134 pte_t entry, int dirty)
135 {
136 pteval_t old_pteval;
137 unsigned int tmp;
138
139 if (pte_same(*ptep, entry))
140 return 0;
141
142 /* only preserve the access flags and write permission */
143 pte_val(entry) &= PTE_AF | PTE_WRITE | PTE_DIRTY;
144
145 /*
146 * PTE_RDONLY is cleared by default in the asm below, so set it in
147 * back if necessary (read-only or clean PTE).
148 */
149 if (!pte_write(entry) || !pte_sw_dirty(entry))
150 pte_val(entry) |= PTE_RDONLY;
151
152 /*
153 * Setting the flags must be done atomically to avoid racing with the
154 * hardware update of the access/dirty state.
155 */
156 asm volatile("// ptep_set_access_flags\n"
157 " prfm pstl1strm, %2\n"
158 "1: ldxr %0, %2\n"
159 " and %0, %0, %3 // clear PTE_RDONLY\n"
160 " orr %0, %0, %4 // set flags\n"
161 " stxr %w1, %0, %2\n"
162 " cbnz %w1, 1b\n"
163 : "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
164 : "L" (~PTE_RDONLY), "r" (pte_val(entry)));
165
166 flush_tlb_fix_spurious_fault(vma, address);
167 return 1;
168 }
169 #endif
170
is_el1_instruction_abort(unsigned int esr)171 static bool is_el1_instruction_abort(unsigned int esr)
172 {
173 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
174 }
175
176 /*
177 * The kernel tried to access some page that wasn't present.
178 */
__do_kernel_fault(struct mm_struct * mm,unsigned long addr,unsigned int esr,struct pt_regs * regs)179 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
180 unsigned int esr, struct pt_regs *regs)
181 {
182 /*
183 * Are we prepared to handle this kernel fault?
184 * We are almost certainly not prepared to handle instruction faults.
185 */
186 if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
187 return;
188
189 /*
190 * No handler, we'll have to terminate things with extreme prejudice.
191 */
192 bust_spinlocks(1);
193 pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
194 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
195 "paging request", addr);
196
197 show_pte(mm, addr);
198 die("Oops", regs, esr);
199 bust_spinlocks(0);
200 do_exit(SIGKILL);
201 }
202
203 /*
204 * Something tried to access memory that isn't in our memory map. User mode
205 * accesses just cause a SIGSEGV
206 */
__do_user_fault(struct task_struct * tsk,unsigned long addr,unsigned int esr,unsigned int sig,int code,struct pt_regs * regs)207 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
208 unsigned int esr, unsigned int sig, int code,
209 struct pt_regs *regs)
210 {
211 struct siginfo si;
212 const struct fault_info *inf;
213
214 if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
215 inf = esr_to_fault_info(esr);
216 pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
217 tsk->comm, task_pid_nr(tsk), inf->name, sig,
218 addr, esr);
219 show_pte(tsk->mm, addr);
220 show_regs(regs);
221 }
222
223 tsk->thread.fault_address = addr;
224 tsk->thread.fault_code = esr;
225 si.si_signo = sig;
226 si.si_errno = 0;
227 si.si_code = code;
228 si.si_addr = (void __user *)addr;
229 force_sig_info(sig, &si, tsk);
230 }
231
do_bad_area(unsigned long addr,unsigned int esr,struct pt_regs * regs)232 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
233 {
234 struct task_struct *tsk = current;
235 struct mm_struct *mm = tsk->active_mm;
236 const struct fault_info *inf;
237
238 /*
239 * If we are in kernel mode at this point, we have no context to
240 * handle this fault with.
241 */
242 if (user_mode(regs)) {
243 inf = esr_to_fault_info(esr);
244 __do_user_fault(tsk, addr, esr, inf->sig, inf->code, regs);
245 } else
246 __do_kernel_fault(mm, addr, esr, regs);
247 }
248
249 #define VM_FAULT_BADMAP 0x010000
250 #define VM_FAULT_BADACCESS 0x020000
251
__do_page_fault(struct mm_struct * mm,unsigned long addr,unsigned int mm_flags,unsigned long vm_flags,struct task_struct * tsk)252 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
253 unsigned int mm_flags, unsigned long vm_flags,
254 struct task_struct *tsk)
255 {
256 struct vm_area_struct *vma;
257 int fault;
258
259 vma = find_vma(mm, addr);
260 fault = VM_FAULT_BADMAP;
261 if (unlikely(!vma))
262 goto out;
263 if (unlikely(vma->vm_start > addr))
264 goto check_stack;
265
266 /*
267 * Ok, we have a good vm_area for this memory access, so we can handle
268 * it.
269 */
270 good_area:
271 /*
272 * Check that the permissions on the VMA allow for the fault which
273 * occurred.
274 */
275 if (!(vma->vm_flags & vm_flags)) {
276 fault = VM_FAULT_BADACCESS;
277 goto out;
278 }
279
280 return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
281
282 check_stack:
283 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
284 goto good_area;
285 out:
286 return fault;
287 }
288
is_permission_fault(unsigned int esr,struct pt_regs * regs)289 static inline bool is_permission_fault(unsigned int esr, struct pt_regs *regs)
290 {
291 unsigned int ec = ESR_ELx_EC(esr);
292 unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
293
294 if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
295 return false;
296
297 if (system_uses_ttbr0_pan())
298 return fsc_type == ESR_ELx_FSC_FAULT &&
299 (regs->pstate & PSR_PAN_BIT);
300 else
301 return fsc_type == ESR_ELx_FSC_PERM;
302 }
303
is_el0_instruction_abort(unsigned int esr)304 static bool is_el0_instruction_abort(unsigned int esr)
305 {
306 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
307 }
308
do_page_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)309 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
310 struct pt_regs *regs)
311 {
312 struct task_struct *tsk;
313 struct mm_struct *mm;
314 int fault, sig, code;
315 unsigned long vm_flags = VM_READ | VM_WRITE;
316 unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
317
318 if (notify_page_fault(regs, esr))
319 return 0;
320
321 tsk = current;
322 mm = tsk->mm;
323
324 /*
325 * If we're in an interrupt or have no user context, we must not take
326 * the fault.
327 */
328 if (faulthandler_disabled() || !mm)
329 goto no_context;
330
331 if (user_mode(regs))
332 mm_flags |= FAULT_FLAG_USER;
333
334 if (is_el0_instruction_abort(esr)) {
335 vm_flags = VM_EXEC;
336 } else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
337 vm_flags = VM_WRITE;
338 mm_flags |= FAULT_FLAG_WRITE;
339 }
340
341 if (addr < TASK_SIZE && is_permission_fault(esr, regs)) {
342 /* regs->orig_addr_limit may be 0 if we entered from EL0 */
343 if (regs->orig_addr_limit == KERNEL_DS)
344 die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
345
346 if (is_el1_instruction_abort(esr))
347 die("Attempting to execute userspace memory", regs, esr);
348
349 if (!search_exception_tables(regs->pc))
350 die("Accessing user space memory outside uaccess.h routines", regs, esr);
351 }
352
353 /*
354 * As per x86, we may deadlock here. However, since the kernel only
355 * validly references user space from well defined areas of the code,
356 * we can bug out early if this is from code which shouldn't.
357 */
358 if (!down_read_trylock(&mm->mmap_sem)) {
359 if (!user_mode(regs) && !search_exception_tables(regs->pc))
360 goto no_context;
361 retry:
362 down_read(&mm->mmap_sem);
363 } else {
364 /*
365 * The above down_read_trylock() might have succeeded in which
366 * case, we'll have missed the might_sleep() from down_read().
367 */
368 might_sleep();
369 #ifdef CONFIG_DEBUG_VM
370 if (!user_mode(regs) && !search_exception_tables(regs->pc))
371 goto no_context;
372 #endif
373 }
374
375 fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
376
377 /*
378 * If we need to retry but a fatal signal is pending, handle the
379 * signal first. We do not need to release the mmap_sem because it
380 * would already be released in __lock_page_or_retry in mm/filemap.c.
381 */
382 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
383 if (!user_mode(regs))
384 goto no_context;
385 return 0;
386 }
387
388 /*
389 * Major/minor page fault accounting is only done on the initial
390 * attempt. If we go through a retry, it is extremely likely that the
391 * page will be found in page cache at that point.
392 */
393
394 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
395 if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
396 if (fault & VM_FAULT_MAJOR) {
397 tsk->maj_flt++;
398 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
399 addr);
400 } else {
401 tsk->min_flt++;
402 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
403 addr);
404 }
405 if (fault & VM_FAULT_RETRY) {
406 /*
407 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
408 * starvation.
409 */
410 mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
411 mm_flags |= FAULT_FLAG_TRIED;
412 goto retry;
413 }
414 }
415
416 up_read(&mm->mmap_sem);
417
418 /*
419 * Handle the "normal" case first - VM_FAULT_MAJOR
420 */
421 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
422 VM_FAULT_BADACCESS))))
423 return 0;
424
425 /*
426 * If we are in kernel mode at this point, we have no context to
427 * handle this fault with.
428 */
429 if (!user_mode(regs))
430 goto no_context;
431
432 if (fault & VM_FAULT_OOM) {
433 /*
434 * We ran out of memory, call the OOM killer, and return to
435 * userspace (which will retry the fault, or kill us if we got
436 * oom-killed).
437 */
438 pagefault_out_of_memory();
439 return 0;
440 }
441
442 if (fault & VM_FAULT_SIGBUS) {
443 /*
444 * We had some memory, but were unable to successfully fix up
445 * this page fault.
446 */
447 sig = SIGBUS;
448 code = BUS_ADRERR;
449 } else {
450 /*
451 * Something tried to access memory that isn't in our memory
452 * map.
453 */
454 sig = SIGSEGV;
455 code = fault == VM_FAULT_BADACCESS ?
456 SEGV_ACCERR : SEGV_MAPERR;
457 }
458
459 __do_user_fault(tsk, addr, esr, sig, code, regs);
460 return 0;
461
462 no_context:
463 __do_kernel_fault(mm, addr, esr, regs);
464 return 0;
465 }
466
467 /*
468 * First Level Translation Fault Handler
469 *
470 * We enter here because the first level page table doesn't contain a valid
471 * entry for the address.
472 *
473 * If the address is in kernel space (>= TASK_SIZE), then we are probably
474 * faulting in the vmalloc() area.
475 *
476 * If the init_task's first level page tables contains the relevant entry, we
477 * copy the it to this task. If not, we send the process a signal, fixup the
478 * exception, or oops the kernel.
479 *
480 * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
481 * or a critical region, and should only copy the information from the master
482 * page table, nothing more.
483 */
do_translation_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)484 static int __kprobes do_translation_fault(unsigned long addr,
485 unsigned int esr,
486 struct pt_regs *regs)
487 {
488 if (addr < TASK_SIZE)
489 return do_page_fault(addr, esr, regs);
490
491 do_bad_area(addr, esr, regs);
492 return 0;
493 }
494
do_alignment_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)495 static int do_alignment_fault(unsigned long addr, unsigned int esr,
496 struct pt_regs *regs)
497 {
498 do_bad_area(addr, esr, regs);
499 return 0;
500 }
501
502 /*
503 * This abort handler always returns "fault".
504 */
do_bad(unsigned long addr,unsigned int esr,struct pt_regs * regs)505 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
506 {
507 return 1;
508 }
509
510 static const struct fault_info fault_info[] = {
511 { do_bad, SIGBUS, 0, "ttbr address size fault" },
512 { do_bad, SIGBUS, 0, "level 1 address size fault" },
513 { do_bad, SIGBUS, 0, "level 2 address size fault" },
514 { do_bad, SIGBUS, 0, "level 3 address size fault" },
515 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 0 translation fault" },
516 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 1 translation fault" },
517 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 2 translation fault" },
518 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 3 translation fault" },
519 { do_bad, SIGBUS, 0, "unknown 8" },
520 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 access flag fault" },
521 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 access flag fault" },
522 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 access flag fault" },
523 { do_bad, SIGBUS, 0, "unknown 12" },
524 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 permission fault" },
525 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 permission fault" },
526 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 permission fault" },
527 { do_bad, SIGBUS, 0, "synchronous external abort" },
528 { do_bad, SIGBUS, 0, "unknown 17" },
529 { do_bad, SIGBUS, 0, "unknown 18" },
530 { do_bad, SIGBUS, 0, "unknown 19" },
531 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
532 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
533 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
534 { do_bad, SIGBUS, 0, "synchronous abort (translation table walk)" },
535 { do_bad, SIGBUS, 0, "synchronous parity error" },
536 { do_bad, SIGBUS, 0, "unknown 25" },
537 { do_bad, SIGBUS, 0, "unknown 26" },
538 { do_bad, SIGBUS, 0, "unknown 27" },
539 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
540 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
541 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
542 { do_bad, SIGBUS, 0, "synchronous parity error (translation table walk)" },
543 { do_bad, SIGBUS, 0, "unknown 32" },
544 { do_alignment_fault, SIGBUS, BUS_ADRALN, "alignment fault" },
545 { do_bad, SIGBUS, 0, "unknown 34" },
546 { do_bad, SIGBUS, 0, "unknown 35" },
547 { do_bad, SIGBUS, 0, "unknown 36" },
548 { do_bad, SIGBUS, 0, "unknown 37" },
549 { do_bad, SIGBUS, 0, "unknown 38" },
550 { do_bad, SIGBUS, 0, "unknown 39" },
551 { do_bad, SIGBUS, 0, "unknown 40" },
552 { do_bad, SIGBUS, 0, "unknown 41" },
553 { do_bad, SIGBUS, 0, "unknown 42" },
554 { do_bad, SIGBUS, 0, "unknown 43" },
555 { do_bad, SIGBUS, 0, "unknown 44" },
556 { do_bad, SIGBUS, 0, "unknown 45" },
557 { do_bad, SIGBUS, 0, "unknown 46" },
558 { do_bad, SIGBUS, 0, "unknown 47" },
559 { do_bad, SIGBUS, 0, "TLB conflict abort" },
560 { do_bad, SIGBUS, 0, "unknown 49" },
561 { do_bad, SIGBUS, 0, "unknown 50" },
562 { do_bad, SIGBUS, 0, "unknown 51" },
563 { do_bad, SIGBUS, 0, "implementation fault (lockdown abort)" },
564 { do_bad, SIGBUS, 0, "implementation fault (unsupported exclusive)" },
565 { do_bad, SIGBUS, 0, "unknown 54" },
566 { do_bad, SIGBUS, 0, "unknown 55" },
567 { do_bad, SIGBUS, 0, "unknown 56" },
568 { do_bad, SIGBUS, 0, "unknown 57" },
569 { do_bad, SIGBUS, 0, "unknown 58" },
570 { do_bad, SIGBUS, 0, "unknown 59" },
571 { do_bad, SIGBUS, 0, "unknown 60" },
572 { do_bad, SIGBUS, 0, "section domain fault" },
573 { do_bad, SIGBUS, 0, "page domain fault" },
574 { do_bad, SIGBUS, 0, "unknown 63" },
575 };
576
577 /*
578 * Dispatch a data abort to the relevant handler.
579 */
do_mem_abort(unsigned long addr,unsigned int esr,struct pt_regs * regs)580 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
581 struct pt_regs *regs)
582 {
583 const struct fault_info *inf = esr_to_fault_info(esr);
584 struct siginfo info;
585
586 if (!inf->fn(addr, esr, regs))
587 return;
588
589 pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
590 inf->name, esr, addr);
591
592 info.si_signo = inf->sig;
593 info.si_errno = 0;
594 info.si_code = inf->code;
595 info.si_addr = (void __user *)addr;
596 arm64_notify_die("", regs, &info, esr);
597 }
598
do_el0_irq_bp_hardening(void)599 asmlinkage void __exception do_el0_irq_bp_hardening(void)
600 {
601 /* PC has already been checked in entry.S */
602 arm64_apply_bp_hardening();
603 }
604
do_el0_ia_bp_hardening(unsigned long addr,unsigned int esr,struct pt_regs * regs)605 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
606 unsigned int esr,
607 struct pt_regs *regs)
608 {
609 /*
610 * We've taken an instruction abort from userspace and not yet
611 * re-enabled IRQs. If the address is a kernel address, apply
612 * BP hardening prior to enabling IRQs and pre-emption.
613 */
614 if (addr > TASK_SIZE)
615 arm64_apply_bp_hardening();
616
617 local_irq_enable();
618 do_mem_abort(addr, esr, regs);
619 }
620
621
622 /*
623 * Handle stack alignment exceptions.
624 */
do_sp_pc_abort(unsigned long addr,unsigned int esr,struct pt_regs * regs)625 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
626 unsigned int esr,
627 struct pt_regs *regs)
628 {
629 struct siginfo info;
630 struct task_struct *tsk = current;
631
632 if (user_mode(regs)) {
633 if (instruction_pointer(regs) > TASK_SIZE)
634 arm64_apply_bp_hardening();
635 local_irq_enable();
636 }
637
638 if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
639 pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
640 tsk->comm, task_pid_nr(tsk),
641 esr_get_class_string(esr), (void *)regs->pc,
642 (void *)regs->sp);
643
644 info.si_signo = SIGBUS;
645 info.si_errno = 0;
646 info.si_code = BUS_ADRALN;
647 info.si_addr = (void __user *)addr;
648 arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
649 }
650
651 int __init early_brk64(unsigned long addr, unsigned int esr,
652 struct pt_regs *regs);
653
654 /*
655 * __refdata because early_brk64 is __init, but the reference to it is
656 * clobbered at arch_initcall time.
657 * See traps.c and debug-monitors.c:debug_traps_init().
658 */
659 static struct fault_info __refdata debug_fault_info[] = {
660 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware breakpoint" },
661 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware single-step" },
662 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware watchpoint" },
663 { do_bad, SIGBUS, 0, "unknown 3" },
664 { do_bad, SIGTRAP, TRAP_BRKPT, "aarch32 BKPT" },
665 { do_bad, SIGTRAP, 0, "aarch32 vector catch" },
666 { early_brk64, SIGTRAP, TRAP_BRKPT, "aarch64 BRK" },
667 { do_bad, SIGBUS, 0, "unknown 7" },
668 };
669
hook_debug_fault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)670 void __init hook_debug_fault_code(int nr,
671 int (*fn)(unsigned long, unsigned int, struct pt_regs *),
672 int sig, int code, const char *name)
673 {
674 BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
675
676 debug_fault_info[nr].fn = fn;
677 debug_fault_info[nr].sig = sig;
678 debug_fault_info[nr].code = code;
679 debug_fault_info[nr].name = name;
680 }
681
do_debug_exception(unsigned long addr,unsigned int esr,struct pt_regs * regs)682 asmlinkage int __exception do_debug_exception(unsigned long addr,
683 unsigned int esr,
684 struct pt_regs *regs)
685 {
686 const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
687 struct siginfo info;
688 int rv;
689
690 /*
691 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
692 * already disabled to preserve the last enabled/disabled addresses.
693 */
694 if (interrupts_enabled(regs))
695 trace_hardirqs_off();
696
697 if (user_mode(regs) && instruction_pointer(regs) > TASK_SIZE)
698 arm64_apply_bp_hardening();
699
700 if (!inf->fn(addr, esr, regs)) {
701 rv = 1;
702 } else {
703 pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
704 inf->name, esr, addr);
705
706 info.si_signo = inf->sig;
707 info.si_errno = 0;
708 info.si_code = inf->code;
709 info.si_addr = (void __user *)addr;
710 arm64_notify_die("", regs, &info, 0);
711 rv = 0;
712 }
713
714 if (interrupts_enabled(regs))
715 trace_hardirqs_on();
716
717 return rv;
718 }
719 NOKPROBE_SYMBOL(do_debug_exception);
720
721 #ifdef CONFIG_ARM64_PAN
cpu_enable_pan(void * __unused)722 int cpu_enable_pan(void *__unused)
723 {
724 /*
725 * We modify PSTATE. This won't work from irq context as the PSTATE
726 * is discarded once we return from the exception.
727 */
728 WARN_ON_ONCE(in_interrupt());
729
730 config_sctlr_el1(SCTLR_EL1_SPAN, 0);
731 asm(SET_PSTATE_PAN(1));
732 return 0;
733 }
734 #endif /* CONFIG_ARM64_PAN */
735
736 #ifdef CONFIG_ARM64_UAO
737 /*
738 * Kernel threads have fs=KERNEL_DS by default, and don't need to call
739 * set_fs(), devtmpfs in particular relies on this behaviour.
740 * We need to enable the feature at runtime (instead of adding it to
741 * PSR_MODE_EL1h) as the feature may not be implemented by the cpu.
742 */
cpu_enable_uao(void * __unused)743 int cpu_enable_uao(void *__unused)
744 {
745 asm(SET_PSTATE_UAO(1));
746 return 0;
747 }
748 #endif /* CONFIG_ARM64_UAO */
749