1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Based on arch/arm/mm/fault.c
4 *
5 * Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1995-2004 Russell King
7 * Copyright (C) 2012 ARM Ltd.
8 */
9
10 #include <linux/acpi.h>
11 #include <linux/bitfield.h>
12 #include <linux/extable.h>
13 #include <linux/signal.h>
14 #include <linux/mm.h>
15 #include <linux/hardirq.h>
16 #include <linux/init.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/page-flags.h>
20 #include <linux/sched/signal.h>
21 #include <linux/sched/debug.h>
22 #include <linux/highmem.h>
23 #include <linux/perf_event.h>
24 #include <linux/preempt.h>
25 #include <linux/hugetlb.h>
26
27 #include <asm/acpi.h>
28 #include <asm/bug.h>
29 #include <asm/cmpxchg.h>
30 #include <asm/cpufeature.h>
31 #include <asm/exception.h>
32 #include <asm/daifflags.h>
33 #include <asm/debug-monitors.h>
34 #include <asm/esr.h>
35 #include <asm/kasan.h>
36 #include <asm/sysreg.h>
37 #include <asm/system_misc.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <asm/traps.h>
41
42 struct fault_info {
43 int (*fn)(unsigned long addr, unsigned int esr,
44 struct pt_regs *regs);
45 int sig;
46 int code;
47 const char *name;
48 };
49
50 static const struct fault_info fault_info[];
51 static struct fault_info debug_fault_info[];
52
esr_to_fault_info(unsigned int esr)53 static inline const struct fault_info *esr_to_fault_info(unsigned int esr)
54 {
55 return fault_info + (esr & ESR_ELx_FSC);
56 }
57
esr_to_debug_fault_info(unsigned int esr)58 static inline const struct fault_info *esr_to_debug_fault_info(unsigned int esr)
59 {
60 return debug_fault_info + DBG_ESR_EVT(esr);
61 }
62
data_abort_decode(unsigned int esr)63 static void data_abort_decode(unsigned int esr)
64 {
65 pr_alert("Data abort info:\n");
66
67 if (esr & ESR_ELx_ISV) {
68 pr_alert(" Access size = %u byte(s)\n",
69 1U << ((esr & ESR_ELx_SAS) >> ESR_ELx_SAS_SHIFT));
70 pr_alert(" SSE = %lu, SRT = %lu\n",
71 (esr & ESR_ELx_SSE) >> ESR_ELx_SSE_SHIFT,
72 (esr & ESR_ELx_SRT_MASK) >> ESR_ELx_SRT_SHIFT);
73 pr_alert(" SF = %lu, AR = %lu\n",
74 (esr & ESR_ELx_SF) >> ESR_ELx_SF_SHIFT,
75 (esr & ESR_ELx_AR) >> ESR_ELx_AR_SHIFT);
76 } else {
77 pr_alert(" ISV = 0, ISS = 0x%08lx\n", esr & ESR_ELx_ISS_MASK);
78 }
79
80 pr_alert(" CM = %lu, WnR = %lu\n",
81 (esr & ESR_ELx_CM) >> ESR_ELx_CM_SHIFT,
82 (esr & ESR_ELx_WNR) >> ESR_ELx_WNR_SHIFT);
83 }
84
mem_abort_decode(unsigned int esr)85 static void mem_abort_decode(unsigned int esr)
86 {
87 pr_alert("Mem abort info:\n");
88
89 pr_alert(" ESR = 0x%08x\n", esr);
90 pr_alert(" EC = 0x%02lx: %s, IL = %u bits\n",
91 ESR_ELx_EC(esr), esr_get_class_string(esr),
92 (esr & ESR_ELx_IL) ? 32 : 16);
93 pr_alert(" SET = %lu, FnV = %lu\n",
94 (esr & ESR_ELx_SET_MASK) >> ESR_ELx_SET_SHIFT,
95 (esr & ESR_ELx_FnV) >> ESR_ELx_FnV_SHIFT);
96 pr_alert(" EA = %lu, S1PTW = %lu\n",
97 (esr & ESR_ELx_EA) >> ESR_ELx_EA_SHIFT,
98 (esr & ESR_ELx_S1PTW) >> ESR_ELx_S1PTW_SHIFT);
99
100 if (esr_is_data_abort(esr))
101 data_abort_decode(esr);
102 }
103
is_ttbr0_addr(unsigned long addr)104 static inline bool is_ttbr0_addr(unsigned long addr)
105 {
106 /* entry assembly clears tags for TTBR0 addrs */
107 return addr < TASK_SIZE;
108 }
109
is_ttbr1_addr(unsigned long addr)110 static inline bool is_ttbr1_addr(unsigned long addr)
111 {
112 /* TTBR1 addresses may have a tag if KASAN_SW_TAGS is in use */
113 return arch_kasan_reset_tag(addr) >= PAGE_OFFSET;
114 }
115
mm_to_pgd_phys(struct mm_struct * mm)116 static inline unsigned long mm_to_pgd_phys(struct mm_struct *mm)
117 {
118 /* Either init_pg_dir or swapper_pg_dir */
119 if (mm == &init_mm)
120 return __pa_symbol(mm->pgd);
121
122 return (unsigned long)virt_to_phys(mm->pgd);
123 }
124
125 /*
126 * Dump out the page tables associated with 'addr' in the currently active mm.
127 */
show_pte(unsigned long addr)128 static void show_pte(unsigned long addr)
129 {
130 struct mm_struct *mm;
131 pgd_t *pgdp;
132 pgd_t pgd;
133
134 if (is_ttbr0_addr(addr)) {
135 /* TTBR0 */
136 mm = current->active_mm;
137 if (mm == &init_mm) {
138 pr_alert("[%016lx] user address but active_mm is swapper\n",
139 addr);
140 return;
141 }
142 } else if (is_ttbr1_addr(addr)) {
143 /* TTBR1 */
144 mm = &init_mm;
145 } else {
146 pr_alert("[%016lx] address between user and kernel address ranges\n",
147 addr);
148 return;
149 }
150
151 pr_alert("%s pgtable: %luk pages, %llu-bit VAs, pgdp=%016lx\n",
152 mm == &init_mm ? "swapper" : "user", PAGE_SIZE / SZ_1K,
153 vabits_actual, mm_to_pgd_phys(mm));
154 pgdp = pgd_offset(mm, addr);
155 pgd = READ_ONCE(*pgdp);
156 pr_alert("[%016lx] pgd=%016llx", addr, pgd_val(pgd));
157
158 do {
159 pud_t *pudp, pud;
160 pmd_t *pmdp, pmd;
161 pte_t *ptep, pte;
162
163 if (pgd_none(pgd) || pgd_bad(pgd))
164 break;
165
166 pudp = pud_offset(pgdp, addr);
167 pud = READ_ONCE(*pudp);
168 pr_cont(", pud=%016llx", pud_val(pud));
169 if (pud_none(pud) || pud_bad(pud))
170 break;
171
172 pmdp = pmd_offset(pudp, addr);
173 pmd = READ_ONCE(*pmdp);
174 pr_cont(", pmd=%016llx", pmd_val(pmd));
175 if (pmd_none(pmd) || pmd_bad(pmd))
176 break;
177
178 ptep = pte_offset_map(pmdp, addr);
179 pte = READ_ONCE(*ptep);
180 pr_cont(", pte=%016llx", pte_val(pte));
181 pte_unmap(ptep);
182 } while(0);
183
184 pr_cont("\n");
185 }
186
187 /*
188 * This function sets the access flags (dirty, accessed), as well as write
189 * permission, and only to a more permissive setting.
190 *
191 * It needs to cope with hardware update of the accessed/dirty state by other
192 * agents in the system and can safely skip the __sync_icache_dcache() call as,
193 * like set_pte_at(), the PTE is never changed from no-exec to exec here.
194 *
195 * Returns whether or not the PTE actually changed.
196 */
ptep_set_access_flags(struct vm_area_struct * vma,unsigned long address,pte_t * ptep,pte_t entry,int dirty)197 int ptep_set_access_flags(struct vm_area_struct *vma,
198 unsigned long address, pte_t *ptep,
199 pte_t entry, int dirty)
200 {
201 pteval_t old_pteval, pteval;
202 pte_t pte = READ_ONCE(*ptep);
203
204 if (pte_same(pte, entry))
205 return 0;
206
207 /* only preserve the access flags and write permission */
208 pte_val(entry) &= PTE_RDONLY | PTE_AF | PTE_WRITE | PTE_DIRTY;
209
210 /*
211 * Setting the flags must be done atomically to avoid racing with the
212 * hardware update of the access/dirty state. The PTE_RDONLY bit must
213 * be set to the most permissive (lowest value) of *ptep and entry
214 * (calculated as: a & b == ~(~a | ~b)).
215 */
216 pte_val(entry) ^= PTE_RDONLY;
217 pteval = pte_val(pte);
218 do {
219 old_pteval = pteval;
220 pteval ^= PTE_RDONLY;
221 pteval |= pte_val(entry);
222 pteval ^= PTE_RDONLY;
223 pteval = cmpxchg_relaxed(&pte_val(*ptep), old_pteval, pteval);
224 } while (pteval != old_pteval);
225
226 flush_tlb_fix_spurious_fault(vma, address);
227 return 1;
228 }
229
is_el1_instruction_abort(unsigned int esr)230 static bool is_el1_instruction_abort(unsigned int esr)
231 {
232 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
233 }
234
is_el1_permission_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)235 static inline bool is_el1_permission_fault(unsigned long addr, unsigned int esr,
236 struct pt_regs *regs)
237 {
238 unsigned int ec = ESR_ELx_EC(esr);
239 unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
240
241 if (ec != ESR_ELx_EC_DABT_CUR && ec != ESR_ELx_EC_IABT_CUR)
242 return false;
243
244 if (fsc_type == ESR_ELx_FSC_PERM)
245 return true;
246
247 if (is_ttbr0_addr(addr) && system_uses_ttbr0_pan())
248 return fsc_type == ESR_ELx_FSC_FAULT &&
249 (regs->pstate & PSR_PAN_BIT);
250
251 return false;
252 }
253
is_spurious_el1_translation_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)254 static bool __kprobes is_spurious_el1_translation_fault(unsigned long addr,
255 unsigned int esr,
256 struct pt_regs *regs)
257 {
258 unsigned long flags;
259 u64 par, dfsc;
260
261 if (ESR_ELx_EC(esr) != ESR_ELx_EC_DABT_CUR ||
262 (esr & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT)
263 return false;
264
265 local_irq_save(flags);
266 asm volatile("at s1e1r, %0" :: "r" (addr));
267 isb();
268 par = read_sysreg(par_el1);
269 local_irq_restore(flags);
270
271 /*
272 * If we now have a valid translation, treat the translation fault as
273 * spurious.
274 */
275 if (!(par & SYS_PAR_EL1_F))
276 return true;
277
278 /*
279 * If we got a different type of fault from the AT instruction,
280 * treat the translation fault as spurious.
281 */
282 dfsc = FIELD_GET(SYS_PAR_EL1_FST, par);
283 return (dfsc & ESR_ELx_FSC_TYPE) != ESR_ELx_FSC_FAULT;
284 }
285
die_kernel_fault(const char * msg,unsigned long addr,unsigned int esr,struct pt_regs * regs)286 static void die_kernel_fault(const char *msg, unsigned long addr,
287 unsigned int esr, struct pt_regs *regs)
288 {
289 bust_spinlocks(1);
290
291 pr_alert("Unable to handle kernel %s at virtual address %016lx\n", msg,
292 addr);
293
294 mem_abort_decode(esr);
295
296 show_pte(addr);
297 die("Oops", regs, esr);
298 bust_spinlocks(0);
299 make_task_dead(SIGKILL);
300 }
301
__do_kernel_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)302 static void __do_kernel_fault(unsigned long addr, unsigned int esr,
303 struct pt_regs *regs)
304 {
305 const char *msg;
306
307 /*
308 * Are we prepared to handle this kernel fault?
309 * We are almost certainly not prepared to handle instruction faults.
310 */
311 if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
312 return;
313
314 if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
315 "Ignoring spurious kernel translation fault at virtual address %016lx\n", addr))
316 return;
317
318 if (is_el1_permission_fault(addr, esr, regs)) {
319 if (esr & ESR_ELx_WNR)
320 msg = "write to read-only memory";
321 else
322 msg = "read from unreadable memory";
323 } else if (addr < PAGE_SIZE) {
324 msg = "NULL pointer dereference";
325 } else {
326 msg = "paging request";
327 }
328
329 die_kernel_fault(msg, addr, esr, regs);
330 }
331
set_thread_esr(unsigned long address,unsigned int esr)332 static void set_thread_esr(unsigned long address, unsigned int esr)
333 {
334 current->thread.fault_address = address;
335
336 /*
337 * If the faulting address is in the kernel, we must sanitize the ESR.
338 * From userspace's point of view, kernel-only mappings don't exist
339 * at all, so we report them as level 0 translation faults.
340 * (This is not quite the way that "no mapping there at all" behaves:
341 * an alignment fault not caused by the memory type would take
342 * precedence over translation fault for a real access to empty
343 * space. Unfortunately we can't easily distinguish "alignment fault
344 * not caused by memory type" from "alignment fault caused by memory
345 * type", so we ignore this wrinkle and just return the translation
346 * fault.)
347 */
348 if (!is_ttbr0_addr(current->thread.fault_address)) {
349 switch (ESR_ELx_EC(esr)) {
350 case ESR_ELx_EC_DABT_LOW:
351 /*
352 * These bits provide only information about the
353 * faulting instruction, which userspace knows already.
354 * We explicitly clear bits which are architecturally
355 * RES0 in case they are given meanings in future.
356 * We always report the ESR as if the fault was taken
357 * to EL1 and so ISV and the bits in ISS[23:14] are
358 * clear. (In fact it always will be a fault to EL1.)
359 */
360 esr &= ESR_ELx_EC_MASK | ESR_ELx_IL |
361 ESR_ELx_CM | ESR_ELx_WNR;
362 esr |= ESR_ELx_FSC_FAULT;
363 break;
364 case ESR_ELx_EC_IABT_LOW:
365 /*
366 * Claim a level 0 translation fault.
367 * All other bits are architecturally RES0 for faults
368 * reported with that DFSC value, so we clear them.
369 */
370 esr &= ESR_ELx_EC_MASK | ESR_ELx_IL;
371 esr |= ESR_ELx_FSC_FAULT;
372 break;
373 default:
374 /*
375 * This should never happen (entry.S only brings us
376 * into this code for insn and data aborts from a lower
377 * exception level). Fail safe by not providing an ESR
378 * context record at all.
379 */
380 WARN(1, "ESR 0x%x is not DABT or IABT from EL0\n", esr);
381 esr = 0;
382 break;
383 }
384 }
385
386 current->thread.fault_code = esr;
387 }
388
do_bad_area(unsigned long addr,unsigned int esr,struct pt_regs * regs)389 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
390 {
391 /*
392 * If we are in kernel mode at this point, we have no context to
393 * handle this fault with.
394 */
395 if (user_mode(regs)) {
396 const struct fault_info *inf = esr_to_fault_info(esr);
397
398 set_thread_esr(addr, esr);
399 arm64_force_sig_fault(inf->sig, inf->code, (void __user *)addr,
400 inf->name);
401 } else {
402 __do_kernel_fault(addr, esr, regs);
403 }
404 }
405
406 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x010000)
407 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x020000)
408
__do_page_fault(struct mm_struct * mm,unsigned long addr,unsigned int mm_flags,unsigned long vm_flags)409 static vm_fault_t __do_page_fault(struct mm_struct *mm, unsigned long addr,
410 unsigned int mm_flags, unsigned long vm_flags)
411 {
412 struct vm_area_struct *vma = find_vma(mm, addr);
413
414 if (unlikely(!vma))
415 return VM_FAULT_BADMAP;
416
417 /*
418 * Ok, we have a good vm_area for this memory access, so we can handle
419 * it.
420 */
421 if (unlikely(vma->vm_start > addr)) {
422 if (!(vma->vm_flags & VM_GROWSDOWN))
423 return VM_FAULT_BADMAP;
424 if (expand_stack(vma, addr))
425 return VM_FAULT_BADMAP;
426 }
427
428 /*
429 * Check that the permissions on the VMA allow for the fault which
430 * occurred.
431 */
432 if (!(vma->vm_flags & vm_flags))
433 return VM_FAULT_BADACCESS;
434 return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
435 }
436
is_el0_instruction_abort(unsigned int esr)437 static bool is_el0_instruction_abort(unsigned int esr)
438 {
439 return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
440 }
441
442 /*
443 * Note: not valid for EL1 DC IVAC, but we never use that such that it
444 * should fault. EL0 cannot issue DC IVAC (undef).
445 */
is_write_abort(unsigned int esr)446 static bool is_write_abort(unsigned int esr)
447 {
448 return (esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM);
449 }
450
do_page_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)451 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
452 struct pt_regs *regs)
453 {
454 const struct fault_info *inf;
455 struct mm_struct *mm = current->mm;
456 vm_fault_t fault, major = 0;
457 unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
458 unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
459
460 if (kprobe_page_fault(regs, esr))
461 return 0;
462
463 /*
464 * If we're in an interrupt or have no user context, we must not take
465 * the fault.
466 */
467 if (faulthandler_disabled() || !mm)
468 goto no_context;
469
470 if (user_mode(regs))
471 mm_flags |= FAULT_FLAG_USER;
472
473 if (is_el0_instruction_abort(esr)) {
474 vm_flags = VM_EXEC;
475 mm_flags |= FAULT_FLAG_INSTRUCTION;
476 } else if (is_write_abort(esr)) {
477 vm_flags = VM_WRITE;
478 mm_flags |= FAULT_FLAG_WRITE;
479 }
480
481 if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
482 /* regs->orig_addr_limit may be 0 if we entered from EL0 */
483 if (regs->orig_addr_limit == KERNEL_DS)
484 die_kernel_fault("access to user memory with fs=KERNEL_DS",
485 addr, esr, regs);
486
487 if (is_el1_instruction_abort(esr))
488 die_kernel_fault("execution of user memory",
489 addr, esr, regs);
490
491 if (!search_exception_tables(regs->pc))
492 die_kernel_fault("access to user memory outside uaccess routines",
493 addr, esr, regs);
494 }
495
496 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
497
498 /*
499 * As per x86, we may deadlock here. However, since the kernel only
500 * validly references user space from well defined areas of the code,
501 * we can bug out early if this is from code which shouldn't.
502 */
503 if (!down_read_trylock(&mm->mmap_sem)) {
504 if (!user_mode(regs) && !search_exception_tables(regs->pc))
505 goto no_context;
506 retry:
507 down_read(&mm->mmap_sem);
508 } else {
509 /*
510 * The above down_read_trylock() might have succeeded in which
511 * case, we'll have missed the might_sleep() from down_read().
512 */
513 might_sleep();
514 #ifdef CONFIG_DEBUG_VM
515 if (!user_mode(regs) && !search_exception_tables(regs->pc)) {
516 up_read(&mm->mmap_sem);
517 goto no_context;
518 }
519 #endif
520 }
521
522 fault = __do_page_fault(mm, addr, mm_flags, vm_flags);
523 major |= fault & VM_FAULT_MAJOR;
524
525 if (fault & VM_FAULT_RETRY) {
526 /*
527 * If we need to retry but a fatal signal is pending,
528 * handle the signal first. We do not need to release
529 * the mmap_sem because it would already be released
530 * in __lock_page_or_retry in mm/filemap.c.
531 */
532 if (fatal_signal_pending(current)) {
533 if (!user_mode(regs))
534 goto no_context;
535 return 0;
536 }
537
538 /*
539 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
540 * starvation.
541 */
542 if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
543 mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
544 mm_flags |= FAULT_FLAG_TRIED;
545 goto retry;
546 }
547 }
548 up_read(&mm->mmap_sem);
549
550 /*
551 * Handle the "normal" (no error) case first.
552 */
553 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
554 VM_FAULT_BADACCESS)))) {
555 /*
556 * Major/minor page fault accounting is only done
557 * once. If we go through a retry, it is extremely
558 * likely that the page will be found in page cache at
559 * that point.
560 */
561 if (major) {
562 current->maj_flt++;
563 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
564 addr);
565 } else {
566 current->min_flt++;
567 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
568 addr);
569 }
570
571 return 0;
572 }
573
574 /*
575 * If we are in kernel mode at this point, we have no context to
576 * handle this fault with.
577 */
578 if (!user_mode(regs))
579 goto no_context;
580
581 if (fault & VM_FAULT_OOM) {
582 /*
583 * We ran out of memory, call the OOM killer, and return to
584 * userspace (which will retry the fault, or kill us if we got
585 * oom-killed).
586 */
587 pagefault_out_of_memory();
588 return 0;
589 }
590
591 inf = esr_to_fault_info(esr);
592 set_thread_esr(addr, esr);
593 if (fault & VM_FAULT_SIGBUS) {
594 /*
595 * We had some memory, but were unable to successfully fix up
596 * this page fault.
597 */
598 arm64_force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)addr,
599 inf->name);
600 } else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
601 unsigned int lsb;
602
603 lsb = PAGE_SHIFT;
604 if (fault & VM_FAULT_HWPOISON_LARGE)
605 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
606
607 arm64_force_sig_mceerr(BUS_MCEERR_AR, (void __user *)addr, lsb,
608 inf->name);
609 } else {
610 /*
611 * Something tried to access memory that isn't in our memory
612 * map.
613 */
614 arm64_force_sig_fault(SIGSEGV,
615 fault == VM_FAULT_BADACCESS ? SEGV_ACCERR : SEGV_MAPERR,
616 (void __user *)addr,
617 inf->name);
618 }
619
620 return 0;
621
622 no_context:
623 __do_kernel_fault(addr, esr, regs);
624 return 0;
625 }
626
do_translation_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)627 static int __kprobes do_translation_fault(unsigned long addr,
628 unsigned int esr,
629 struct pt_regs *regs)
630 {
631 if (is_ttbr0_addr(addr))
632 return do_page_fault(addr, esr, regs);
633
634 do_bad_area(addr, esr, regs);
635 return 0;
636 }
637
do_alignment_fault(unsigned long addr,unsigned int esr,struct pt_regs * regs)638 static int do_alignment_fault(unsigned long addr, unsigned int esr,
639 struct pt_regs *regs)
640 {
641 do_bad_area(addr, esr, regs);
642 return 0;
643 }
644
do_bad(unsigned long addr,unsigned int esr,struct pt_regs * regs)645 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
646 {
647 return 1; /* "fault" */
648 }
649
do_sea(unsigned long addr,unsigned int esr,struct pt_regs * regs)650 static int do_sea(unsigned long addr, unsigned int esr, struct pt_regs *regs)
651 {
652 const struct fault_info *inf;
653 void __user *siaddr;
654
655 inf = esr_to_fault_info(esr);
656
657 if (user_mode(regs) && apei_claim_sea(regs) == 0) {
658 /*
659 * APEI claimed this as a firmware-first notification.
660 * Some processing deferred to task_work before ret_to_user().
661 */
662 return 0;
663 }
664
665 if (esr & ESR_ELx_FnV)
666 siaddr = NULL;
667 else
668 siaddr = (void __user *)addr;
669 arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
670
671 return 0;
672 }
673
674 static const struct fault_info fault_info[] = {
675 { do_bad, SIGKILL, SI_KERNEL, "ttbr address size fault" },
676 { do_bad, SIGKILL, SI_KERNEL, "level 1 address size fault" },
677 { do_bad, SIGKILL, SI_KERNEL, "level 2 address size fault" },
678 { do_bad, SIGKILL, SI_KERNEL, "level 3 address size fault" },
679 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 0 translation fault" },
680 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 1 translation fault" },
681 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 2 translation fault" },
682 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "level 3 translation fault" },
683 { do_bad, SIGKILL, SI_KERNEL, "unknown 8" },
684 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 access flag fault" },
685 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 access flag fault" },
686 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 access flag fault" },
687 { do_bad, SIGKILL, SI_KERNEL, "unknown 12" },
688 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 1 permission fault" },
689 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 2 permission fault" },
690 { do_page_fault, SIGSEGV, SEGV_ACCERR, "level 3 permission fault" },
691 { do_sea, SIGBUS, BUS_OBJERR, "synchronous external abort" },
692 { do_bad, SIGKILL, SI_KERNEL, "unknown 17" },
693 { do_bad, SIGKILL, SI_KERNEL, "unknown 18" },
694 { do_bad, SIGKILL, SI_KERNEL, "unknown 19" },
695 { do_sea, SIGKILL, SI_KERNEL, "level 0 (translation table walk)" },
696 { do_sea, SIGKILL, SI_KERNEL, "level 1 (translation table walk)" },
697 { do_sea, SIGKILL, SI_KERNEL, "level 2 (translation table walk)" },
698 { do_sea, SIGKILL, SI_KERNEL, "level 3 (translation table walk)" },
699 { do_sea, SIGBUS, BUS_OBJERR, "synchronous parity or ECC error" }, // Reserved when RAS is implemented
700 { do_bad, SIGKILL, SI_KERNEL, "unknown 25" },
701 { do_bad, SIGKILL, SI_KERNEL, "unknown 26" },
702 { do_bad, SIGKILL, SI_KERNEL, "unknown 27" },
703 { do_sea, SIGKILL, SI_KERNEL, "level 0 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
704 { do_sea, SIGKILL, SI_KERNEL, "level 1 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
705 { do_sea, SIGKILL, SI_KERNEL, "level 2 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
706 { do_sea, SIGKILL, SI_KERNEL, "level 3 synchronous parity error (translation table walk)" }, // Reserved when RAS is implemented
707 { do_bad, SIGKILL, SI_KERNEL, "unknown 32" },
708 { do_alignment_fault, SIGBUS, BUS_ADRALN, "alignment fault" },
709 { do_bad, SIGKILL, SI_KERNEL, "unknown 34" },
710 { do_bad, SIGKILL, SI_KERNEL, "unknown 35" },
711 { do_bad, SIGKILL, SI_KERNEL, "unknown 36" },
712 { do_bad, SIGKILL, SI_KERNEL, "unknown 37" },
713 { do_bad, SIGKILL, SI_KERNEL, "unknown 38" },
714 { do_bad, SIGKILL, SI_KERNEL, "unknown 39" },
715 { do_bad, SIGKILL, SI_KERNEL, "unknown 40" },
716 { do_bad, SIGKILL, SI_KERNEL, "unknown 41" },
717 { do_bad, SIGKILL, SI_KERNEL, "unknown 42" },
718 { do_bad, SIGKILL, SI_KERNEL, "unknown 43" },
719 { do_bad, SIGKILL, SI_KERNEL, "unknown 44" },
720 { do_bad, SIGKILL, SI_KERNEL, "unknown 45" },
721 { do_bad, SIGKILL, SI_KERNEL, "unknown 46" },
722 { do_bad, SIGKILL, SI_KERNEL, "unknown 47" },
723 { do_bad, SIGKILL, SI_KERNEL, "TLB conflict abort" },
724 { do_bad, SIGKILL, SI_KERNEL, "Unsupported atomic hardware update fault" },
725 { do_bad, SIGKILL, SI_KERNEL, "unknown 50" },
726 { do_bad, SIGKILL, SI_KERNEL, "unknown 51" },
727 { do_bad, SIGKILL, SI_KERNEL, "implementation fault (lockdown abort)" },
728 { do_bad, SIGBUS, BUS_OBJERR, "implementation fault (unsupported exclusive)" },
729 { do_bad, SIGKILL, SI_KERNEL, "unknown 54" },
730 { do_bad, SIGKILL, SI_KERNEL, "unknown 55" },
731 { do_bad, SIGKILL, SI_KERNEL, "unknown 56" },
732 { do_bad, SIGKILL, SI_KERNEL, "unknown 57" },
733 { do_bad, SIGKILL, SI_KERNEL, "unknown 58" },
734 { do_bad, SIGKILL, SI_KERNEL, "unknown 59" },
735 { do_bad, SIGKILL, SI_KERNEL, "unknown 60" },
736 { do_bad, SIGKILL, SI_KERNEL, "section domain fault" },
737 { do_bad, SIGKILL, SI_KERNEL, "page domain fault" },
738 { do_bad, SIGKILL, SI_KERNEL, "unknown 63" },
739 };
740
do_mem_abort(unsigned long addr,unsigned int esr,struct pt_regs * regs)741 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
742 struct pt_regs *regs)
743 {
744 const struct fault_info *inf = esr_to_fault_info(esr);
745
746 if (!inf->fn(addr, esr, regs))
747 return;
748
749 if (!user_mode(regs)) {
750 pr_alert("Unhandled fault at 0x%016lx\n", addr);
751 mem_abort_decode(esr);
752 show_pte(addr);
753 }
754
755 arm64_notify_die(inf->name, regs,
756 inf->sig, inf->code, (void __user *)addr, esr);
757 }
758
do_el0_irq_bp_hardening(void)759 asmlinkage void __exception do_el0_irq_bp_hardening(void)
760 {
761 /* PC has already been checked in entry.S */
762 arm64_apply_bp_hardening();
763 }
764
do_el0_ia_bp_hardening(unsigned long addr,unsigned int esr,struct pt_regs * regs)765 asmlinkage void __exception do_el0_ia_bp_hardening(unsigned long addr,
766 unsigned int esr,
767 struct pt_regs *regs)
768 {
769 /*
770 * We've taken an instruction abort from userspace and not yet
771 * re-enabled IRQs. If the address is a kernel address, apply
772 * BP hardening prior to enabling IRQs and pre-emption.
773 */
774 if (!is_ttbr0_addr(addr))
775 arm64_apply_bp_hardening();
776
777 local_daif_restore(DAIF_PROCCTX);
778 do_mem_abort(addr, esr, regs);
779 }
780
781
do_sp_pc_abort(unsigned long addr,unsigned int esr,struct pt_regs * regs)782 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
783 unsigned int esr,
784 struct pt_regs *regs)
785 {
786 if (user_mode(regs)) {
787 if (!is_ttbr0_addr(instruction_pointer(regs)))
788 arm64_apply_bp_hardening();
789 local_daif_restore(DAIF_PROCCTX);
790 }
791
792 arm64_notify_die("SP/PC alignment exception", regs,
793 SIGBUS, BUS_ADRALN, (void __user *)addr, esr);
794 }
795
796 int __init early_brk64(unsigned long addr, unsigned int esr,
797 struct pt_regs *regs);
798
799 /*
800 * __refdata because early_brk64 is __init, but the reference to it is
801 * clobbered at arch_initcall time.
802 * See traps.c and debug-monitors.c:debug_traps_init().
803 */
804 static struct fault_info __refdata debug_fault_info[] = {
805 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware breakpoint" },
806 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware single-step" },
807 { do_bad, SIGTRAP, TRAP_HWBKPT, "hardware watchpoint" },
808 { do_bad, SIGKILL, SI_KERNEL, "unknown 3" },
809 { do_bad, SIGTRAP, TRAP_BRKPT, "aarch32 BKPT" },
810 { do_bad, SIGKILL, SI_KERNEL, "aarch32 vector catch" },
811 { early_brk64, SIGTRAP, TRAP_BRKPT, "aarch64 BRK" },
812 { do_bad, SIGKILL, SI_KERNEL, "unknown 7" },
813 };
814
hook_debug_fault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)815 void __init hook_debug_fault_code(int nr,
816 int (*fn)(unsigned long, unsigned int, struct pt_regs *),
817 int sig, int code, const char *name)
818 {
819 BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
820
821 debug_fault_info[nr].fn = fn;
822 debug_fault_info[nr].sig = sig;
823 debug_fault_info[nr].code = code;
824 debug_fault_info[nr].name = name;
825 }
826
827 /*
828 * In debug exception context, we explicitly disable preemption despite
829 * having interrupts disabled.
830 * This serves two purposes: it makes it much less likely that we would
831 * accidentally schedule in exception context and it will force a warning
832 * if we somehow manage to schedule by accident.
833 */
debug_exception_enter(struct pt_regs * regs)834 static void debug_exception_enter(struct pt_regs *regs)
835 {
836 /*
837 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
838 * already disabled to preserve the last enabled/disabled addresses.
839 */
840 if (interrupts_enabled(regs))
841 trace_hardirqs_off();
842
843 if (user_mode(regs)) {
844 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
845 } else {
846 /*
847 * We might have interrupted pretty much anything. In
848 * fact, if we're a debug exception, we can even interrupt
849 * NMI processing. We don't want this code makes in_nmi()
850 * to return true, but we need to notify RCU.
851 */
852 rcu_nmi_enter();
853 }
854
855 preempt_disable();
856
857 /* This code is a bit fragile. Test it. */
858 RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
859 }
860 NOKPROBE_SYMBOL(debug_exception_enter);
861
debug_exception_exit(struct pt_regs * regs)862 static void debug_exception_exit(struct pt_regs *regs)
863 {
864 preempt_enable_no_resched();
865
866 if (!user_mode(regs))
867 rcu_nmi_exit();
868
869 if (interrupts_enabled(regs))
870 trace_hardirqs_on();
871 }
872 NOKPROBE_SYMBOL(debug_exception_exit);
873
874 #ifdef CONFIG_ARM64_ERRATUM_1463225
875 DECLARE_PER_CPU(int, __in_cortex_a76_erratum_1463225_wa);
876
877 static int __exception
cortex_a76_erratum_1463225_debug_handler(struct pt_regs * regs)878 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
879 {
880 if (user_mode(regs))
881 return 0;
882
883 if (!__this_cpu_read(__in_cortex_a76_erratum_1463225_wa))
884 return 0;
885
886 /*
887 * We've taken a dummy step exception from the kernel to ensure
888 * that interrupts are re-enabled on the syscall path. Return back
889 * to cortex_a76_erratum_1463225_svc_handler() with debug exceptions
890 * masked so that we can safely restore the mdscr and get on with
891 * handling the syscall.
892 */
893 regs->pstate |= PSR_D_BIT;
894 return 1;
895 }
896 #else
897 static int __exception
cortex_a76_erratum_1463225_debug_handler(struct pt_regs * regs)898 cortex_a76_erratum_1463225_debug_handler(struct pt_regs *regs)
899 {
900 return 0;
901 }
902 #endif /* CONFIG_ARM64_ERRATUM_1463225 */
903
do_debug_exception(unsigned long addr_if_watchpoint,unsigned int esr,struct pt_regs * regs)904 asmlinkage void __exception do_debug_exception(unsigned long addr_if_watchpoint,
905 unsigned int esr,
906 struct pt_regs *regs)
907 {
908 const struct fault_info *inf = esr_to_debug_fault_info(esr);
909 unsigned long pc = instruction_pointer(regs);
910
911 if (cortex_a76_erratum_1463225_debug_handler(regs))
912 return;
913
914 debug_exception_enter(regs);
915
916 if (user_mode(regs) && !is_ttbr0_addr(pc))
917 arm64_apply_bp_hardening();
918
919 if (inf->fn(addr_if_watchpoint, esr, regs)) {
920 arm64_notify_die(inf->name, regs,
921 inf->sig, inf->code, (void __user *)pc, esr);
922 }
923
924 debug_exception_exit(regs);
925 }
926 NOKPROBE_SYMBOL(do_debug_exception);
927