xref: /openbmc/linux/arch/arm64/mm/fault.c (revision eb3fcf00)
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/module.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 
33 #include <asm/cpufeature.h>
34 #include <asm/exception.h>
35 #include <asm/debug-monitors.h>
36 #include <asm/esr.h>
37 #include <asm/sysreg.h>
38 #include <asm/system_misc.h>
39 #include <asm/pgtable.h>
40 #include <asm/tlbflush.h>
41 
42 static const char *fault_name(unsigned int esr);
43 
44 /*
45  * Dump out the page tables associated with 'addr' in mm 'mm'.
46  */
47 void show_pte(struct mm_struct *mm, unsigned long addr)
48 {
49 	pgd_t *pgd;
50 
51 	if (!mm)
52 		mm = &init_mm;
53 
54 	pr_alert("pgd = %p\n", mm->pgd);
55 	pgd = pgd_offset(mm, addr);
56 	pr_alert("[%08lx] *pgd=%016llx", addr, pgd_val(*pgd));
57 
58 	do {
59 		pud_t *pud;
60 		pmd_t *pmd;
61 		pte_t *pte;
62 
63 		if (pgd_none(*pgd) || pgd_bad(*pgd))
64 			break;
65 
66 		pud = pud_offset(pgd, addr);
67 		printk(", *pud=%016llx", pud_val(*pud));
68 		if (pud_none(*pud) || pud_bad(*pud))
69 			break;
70 
71 		pmd = pmd_offset(pud, addr);
72 		printk(", *pmd=%016llx", pmd_val(*pmd));
73 		if (pmd_none(*pmd) || pmd_bad(*pmd))
74 			break;
75 
76 		pte = pte_offset_map(pmd, addr);
77 		printk(", *pte=%016llx", pte_val(*pte));
78 		pte_unmap(pte);
79 	} while(0);
80 
81 	printk("\n");
82 }
83 
84 /*
85  * The kernel tried to access some page that wasn't present.
86  */
87 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
88 			      unsigned int esr, struct pt_regs *regs)
89 {
90 	/*
91 	 * Are we prepared to handle this kernel fault?
92 	 */
93 	if (fixup_exception(regs))
94 		return;
95 
96 	/*
97 	 * No handler, we'll have to terminate things with extreme prejudice.
98 	 */
99 	bust_spinlocks(1);
100 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
101 		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
102 		 "paging request", addr);
103 
104 	show_pte(mm, addr);
105 	die("Oops", regs, esr);
106 	bust_spinlocks(0);
107 	do_exit(SIGKILL);
108 }
109 
110 /*
111  * Something tried to access memory that isn't in our memory map. User mode
112  * accesses just cause a SIGSEGV
113  */
114 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
115 			    unsigned int esr, unsigned int sig, int code,
116 			    struct pt_regs *regs)
117 {
118 	struct siginfo si;
119 
120 	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
121 		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
122 			tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,
123 			addr, esr);
124 		show_pte(tsk->mm, addr);
125 		show_regs(regs);
126 	}
127 
128 	tsk->thread.fault_address = addr;
129 	tsk->thread.fault_code = esr;
130 	si.si_signo = sig;
131 	si.si_errno = 0;
132 	si.si_code = code;
133 	si.si_addr = (void __user *)addr;
134 	force_sig_info(sig, &si, tsk);
135 }
136 
137 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
138 {
139 	struct task_struct *tsk = current;
140 	struct mm_struct *mm = tsk->active_mm;
141 
142 	/*
143 	 * If we are in kernel mode at this point, we have no context to
144 	 * handle this fault with.
145 	 */
146 	if (user_mode(regs))
147 		__do_user_fault(tsk, addr, esr, SIGSEGV, SEGV_MAPERR, regs);
148 	else
149 		__do_kernel_fault(mm, addr, esr, regs);
150 }
151 
152 #define VM_FAULT_BADMAP		0x010000
153 #define VM_FAULT_BADACCESS	0x020000
154 
155 #define ESR_LNX_EXEC		(1 << 24)
156 
157 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
158 			   unsigned int mm_flags, unsigned long vm_flags,
159 			   struct task_struct *tsk)
160 {
161 	struct vm_area_struct *vma;
162 	int fault;
163 
164 	vma = find_vma(mm, addr);
165 	fault = VM_FAULT_BADMAP;
166 	if (unlikely(!vma))
167 		goto out;
168 	if (unlikely(vma->vm_start > addr))
169 		goto check_stack;
170 
171 	/*
172 	 * Ok, we have a good vm_area for this memory access, so we can handle
173 	 * it.
174 	 */
175 good_area:
176 	/*
177 	 * Check that the permissions on the VMA allow for the fault which
178 	 * occurred. If we encountered a write or exec fault, we must have
179 	 * appropriate permissions, otherwise we allow any permission.
180 	 */
181 	if (!(vma->vm_flags & vm_flags)) {
182 		fault = VM_FAULT_BADACCESS;
183 		goto out;
184 	}
185 
186 	return handle_mm_fault(mm, vma, addr & PAGE_MASK, mm_flags);
187 
188 check_stack:
189 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
190 		goto good_area;
191 out:
192 	return fault;
193 }
194 
195 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
196 				   struct pt_regs *regs)
197 {
198 	struct task_struct *tsk;
199 	struct mm_struct *mm;
200 	int fault, sig, code;
201 	unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
202 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
203 
204 	tsk = current;
205 	mm  = tsk->mm;
206 
207 	/* Enable interrupts if they were enabled in the parent context. */
208 	if (interrupts_enabled(regs))
209 		local_irq_enable();
210 
211 	/*
212 	 * If we're in an interrupt or have no user context, we must not take
213 	 * the fault.
214 	 */
215 	if (faulthandler_disabled() || !mm)
216 		goto no_context;
217 
218 	if (user_mode(regs))
219 		mm_flags |= FAULT_FLAG_USER;
220 
221 	if (esr & ESR_LNX_EXEC) {
222 		vm_flags = VM_EXEC;
223 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
224 		vm_flags = VM_WRITE;
225 		mm_flags |= FAULT_FLAG_WRITE;
226 	}
227 
228 	/*
229 	 * PAN bit set implies the fault happened in kernel space, but not
230 	 * in the arch's user access functions.
231 	 */
232 	if (IS_ENABLED(CONFIG_ARM64_PAN) && (regs->pstate & PSR_PAN_BIT))
233 		goto no_context;
234 
235 	/*
236 	 * As per x86, we may deadlock here. However, since the kernel only
237 	 * validly references user space from well defined areas of the code,
238 	 * we can bug out early if this is from code which shouldn't.
239 	 */
240 	if (!down_read_trylock(&mm->mmap_sem)) {
241 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
242 			goto no_context;
243 retry:
244 		down_read(&mm->mmap_sem);
245 	} else {
246 		/*
247 		 * The above down_read_trylock() might have succeeded in which
248 		 * case, we'll have missed the might_sleep() from down_read().
249 		 */
250 		might_sleep();
251 #ifdef CONFIG_DEBUG_VM
252 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
253 			goto no_context;
254 #endif
255 	}
256 
257 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
258 
259 	/*
260 	 * If we need to retry but a fatal signal is pending, handle the
261 	 * signal first. We do not need to release the mmap_sem because it
262 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
263 	 */
264 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
265 		return 0;
266 
267 	/*
268 	 * Major/minor page fault accounting is only done on the initial
269 	 * attempt. If we go through a retry, it is extremely likely that the
270 	 * page will be found in page cache at that point.
271 	 */
272 
273 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
274 	if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
275 		if (fault & VM_FAULT_MAJOR) {
276 			tsk->maj_flt++;
277 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
278 				      addr);
279 		} else {
280 			tsk->min_flt++;
281 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
282 				      addr);
283 		}
284 		if (fault & VM_FAULT_RETRY) {
285 			/*
286 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
287 			 * starvation.
288 			 */
289 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
290 			goto retry;
291 		}
292 	}
293 
294 	up_read(&mm->mmap_sem);
295 
296 	/*
297 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
298 	 */
299 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
300 			      VM_FAULT_BADACCESS))))
301 		return 0;
302 
303 	/*
304 	 * If we are in kernel mode at this point, we have no context to
305 	 * handle this fault with.
306 	 */
307 	if (!user_mode(regs))
308 		goto no_context;
309 
310 	if (fault & VM_FAULT_OOM) {
311 		/*
312 		 * We ran out of memory, call the OOM killer, and return to
313 		 * userspace (which will retry the fault, or kill us if we got
314 		 * oom-killed).
315 		 */
316 		pagefault_out_of_memory();
317 		return 0;
318 	}
319 
320 	if (fault & VM_FAULT_SIGBUS) {
321 		/*
322 		 * We had some memory, but were unable to successfully fix up
323 		 * this page fault.
324 		 */
325 		sig = SIGBUS;
326 		code = BUS_ADRERR;
327 	} else {
328 		/*
329 		 * Something tried to access memory that isn't in our memory
330 		 * map.
331 		 */
332 		sig = SIGSEGV;
333 		code = fault == VM_FAULT_BADACCESS ?
334 			SEGV_ACCERR : SEGV_MAPERR;
335 	}
336 
337 	__do_user_fault(tsk, addr, esr, sig, code, regs);
338 	return 0;
339 
340 no_context:
341 	__do_kernel_fault(mm, addr, esr, regs);
342 	return 0;
343 }
344 
345 /*
346  * First Level Translation Fault Handler
347  *
348  * We enter here because the first level page table doesn't contain a valid
349  * entry for the address.
350  *
351  * If the address is in kernel space (>= TASK_SIZE), then we are probably
352  * faulting in the vmalloc() area.
353  *
354  * If the init_task's first level page tables contains the relevant entry, we
355  * copy the it to this task.  If not, we send the process a signal, fixup the
356  * exception, or oops the kernel.
357  *
358  * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
359  * or a critical region, and should only copy the information from the master
360  * page table, nothing more.
361  */
362 static int __kprobes do_translation_fault(unsigned long addr,
363 					  unsigned int esr,
364 					  struct pt_regs *regs)
365 {
366 	if (addr < TASK_SIZE)
367 		return do_page_fault(addr, esr, regs);
368 
369 	do_bad_area(addr, esr, regs);
370 	return 0;
371 }
372 
373 /*
374  * This abort handler always returns "fault".
375  */
376 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
377 {
378 	return 1;
379 }
380 
381 static struct fault_info {
382 	int	(*fn)(unsigned long addr, unsigned int esr, struct pt_regs *regs);
383 	int	sig;
384 	int	code;
385 	const char *name;
386 } fault_info[] = {
387 	{ do_bad,		SIGBUS,  0,		"ttbr address size fault"	},
388 	{ do_bad,		SIGBUS,  0,		"level 1 address size fault"	},
389 	{ do_bad,		SIGBUS,  0,		"level 2 address size fault"	},
390 	{ do_bad,		SIGBUS,  0,		"level 3 address size fault"	},
391 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
392 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
393 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
394 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
395 	{ do_bad,		SIGBUS,  0,		"reserved access flag fault"	},
396 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
397 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
398 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
399 	{ do_bad,		SIGBUS,  0,		"reserved permission fault"	},
400 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
401 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
402 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
403 	{ do_bad,		SIGBUS,  0,		"synchronous external abort"	},
404 	{ do_bad,		SIGBUS,  0,		"asynchronous external abort"	},
405 	{ do_bad,		SIGBUS,  0,		"unknown 18"			},
406 	{ do_bad,		SIGBUS,  0,		"unknown 19"			},
407 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
408 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
409 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
410 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
411 	{ do_bad,		SIGBUS,  0,		"synchronous parity error"	},
412 	{ do_bad,		SIGBUS,  0,		"asynchronous parity error"	},
413 	{ do_bad,		SIGBUS,  0,		"unknown 26"			},
414 	{ do_bad,		SIGBUS,  0,		"unknown 27"			},
415 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk" },
416 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk" },
417 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk" },
418 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk" },
419 	{ do_bad,		SIGBUS,  0,		"unknown 32"			},
420 	{ do_bad,		SIGBUS,  BUS_ADRALN,	"alignment fault"		},
421 	{ do_bad,		SIGBUS,  0,		"debug event"			},
422 	{ do_bad,		SIGBUS,  0,		"unknown 35"			},
423 	{ do_bad,		SIGBUS,  0,		"unknown 36"			},
424 	{ do_bad,		SIGBUS,  0,		"unknown 37"			},
425 	{ do_bad,		SIGBUS,  0,		"unknown 38"			},
426 	{ do_bad,		SIGBUS,  0,		"unknown 39"			},
427 	{ do_bad,		SIGBUS,  0,		"unknown 40"			},
428 	{ do_bad,		SIGBUS,  0,		"unknown 41"			},
429 	{ do_bad,		SIGBUS,  0,		"unknown 42"			},
430 	{ do_bad,		SIGBUS,  0,		"unknown 43"			},
431 	{ do_bad,		SIGBUS,  0,		"unknown 44"			},
432 	{ do_bad,		SIGBUS,  0,		"unknown 45"			},
433 	{ do_bad,		SIGBUS,  0,		"unknown 46"			},
434 	{ do_bad,		SIGBUS,  0,		"unknown 47"			},
435 	{ do_bad,		SIGBUS,  0,		"unknown 48"			},
436 	{ do_bad,		SIGBUS,  0,		"unknown 49"			},
437 	{ do_bad,		SIGBUS,  0,		"unknown 50"			},
438 	{ do_bad,		SIGBUS,  0,		"unknown 51"			},
439 	{ do_bad,		SIGBUS,  0,		"implementation fault (lockdown abort)" },
440 	{ do_bad,		SIGBUS,  0,		"unknown 53"			},
441 	{ do_bad,		SIGBUS,  0,		"unknown 54"			},
442 	{ do_bad,		SIGBUS,  0,		"unknown 55"			},
443 	{ do_bad,		SIGBUS,  0,		"unknown 56"			},
444 	{ do_bad,		SIGBUS,  0,		"unknown 57"			},
445 	{ do_bad,		SIGBUS,  0,		"implementation fault (coprocessor abort)" },
446 	{ do_bad,		SIGBUS,  0,		"unknown 59"			},
447 	{ do_bad,		SIGBUS,  0,		"unknown 60"			},
448 	{ do_bad,		SIGBUS,  0,		"unknown 61"			},
449 	{ do_bad,		SIGBUS,  0,		"unknown 62"			},
450 	{ do_bad,		SIGBUS,  0,		"unknown 63"			},
451 };
452 
453 static const char *fault_name(unsigned int esr)
454 {
455 	const struct fault_info *inf = fault_info + (esr & 63);
456 	return inf->name;
457 }
458 
459 /*
460  * Dispatch a data abort to the relevant handler.
461  */
462 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
463 					 struct pt_regs *regs)
464 {
465 	const struct fault_info *inf = fault_info + (esr & 63);
466 	struct siginfo info;
467 
468 	if (!inf->fn(addr, esr, regs))
469 		return;
470 
471 	pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
472 		 inf->name, esr, addr);
473 
474 	info.si_signo = inf->sig;
475 	info.si_errno = 0;
476 	info.si_code  = inf->code;
477 	info.si_addr  = (void __user *)addr;
478 	arm64_notify_die("", regs, &info, esr);
479 }
480 
481 /*
482  * Handle stack alignment exceptions.
483  */
484 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
485 					   unsigned int esr,
486 					   struct pt_regs *regs)
487 {
488 	struct siginfo info;
489 	struct task_struct *tsk = current;
490 
491 	if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
492 		pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
493 				    tsk->comm, task_pid_nr(tsk),
494 				    esr_get_class_string(esr), (void *)regs->pc,
495 				    (void *)regs->sp);
496 
497 	info.si_signo = SIGBUS;
498 	info.si_errno = 0;
499 	info.si_code  = BUS_ADRALN;
500 	info.si_addr  = (void __user *)addr;
501 	arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
502 }
503 
504 int __init early_brk64(unsigned long addr, unsigned int esr,
505 		       struct pt_regs *regs);
506 
507 /*
508  * __refdata because early_brk64 is __init, but the reference to it is
509  * clobbered at arch_initcall time.
510  * See traps.c and debug-monitors.c:debug_traps_init().
511  */
512 static struct fault_info __refdata debug_fault_info[] = {
513 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
514 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
515 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
516 	{ do_bad,	SIGBUS,		0,		"unknown 3"		},
517 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
518 	{ do_bad,	SIGTRAP,	0,		"aarch32 vector catch"	},
519 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
520 	{ do_bad,	SIGBUS,		0,		"unknown 7"		},
521 };
522 
523 void __init hook_debug_fault_code(int nr,
524 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
525 				  int sig, int code, const char *name)
526 {
527 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
528 
529 	debug_fault_info[nr].fn		= fn;
530 	debug_fault_info[nr].sig	= sig;
531 	debug_fault_info[nr].code	= code;
532 	debug_fault_info[nr].name	= name;
533 }
534 
535 asmlinkage int __exception do_debug_exception(unsigned long addr,
536 					      unsigned int esr,
537 					      struct pt_regs *regs)
538 {
539 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
540 	struct siginfo info;
541 
542 	if (!inf->fn(addr, esr, regs))
543 		return 1;
544 
545 	pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
546 		 inf->name, esr, addr);
547 
548 	info.si_signo = inf->sig;
549 	info.si_errno = 0;
550 	info.si_code  = inf->code;
551 	info.si_addr  = (void __user *)addr;
552 	arm64_notify_die("", regs, &info, 0);
553 
554 	return 0;
555 }
556 
557 #ifdef CONFIG_ARM64_PAN
558 void cpu_enable_pan(void)
559 {
560 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
561 }
562 #endif /* CONFIG_ARM64_PAN */
563