xref: /openbmc/linux/arch/s390/mm/fault.c (revision 79f08d9e)
1 /*
2  *  S390 version
3  *    Copyright IBM Corp. 1999
4  *    Author(s): Hartmut Penner (hp@de.ibm.com)
5  *               Ulrich Weigand (uweigand@de.ibm.com)
6  *
7  *  Derived from "arch/i386/mm/fault.c"
8  *    Copyright (C) 1995  Linus Torvalds
9  */
10 
11 #include <linux/kernel_stat.h>
12 #include <linux/perf_event.h>
13 #include <linux/signal.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/mman.h>
21 #include <linux/mm.h>
22 #include <linux/compat.h>
23 #include <linux/smp.h>
24 #include <linux/kdebug.h>
25 #include <linux/init.h>
26 #include <linux/console.h>
27 #include <linux/module.h>
28 #include <linux/hardirq.h>
29 #include <linux/kprobes.h>
30 #include <linux/uaccess.h>
31 #include <linux/hugetlb.h>
32 #include <asm/asm-offsets.h>
33 #include <asm/pgtable.h>
34 #include <asm/irq.h>
35 #include <asm/mmu_context.h>
36 #include <asm/facility.h>
37 #include "../kernel/entry.h"
38 
39 #ifndef CONFIG_64BIT
40 #define __FAIL_ADDR_MASK 0x7ffff000
41 #define __SUBCODE_MASK 0x0200
42 #define __PF_RES_FIELD 0ULL
43 #else /* CONFIG_64BIT */
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #endif /* CONFIG_64BIT */
48 
49 #define VM_FAULT_BADCONTEXT	0x010000
50 #define VM_FAULT_BADMAP		0x020000
51 #define VM_FAULT_BADACCESS	0x040000
52 #define VM_FAULT_SIGNAL		0x080000
53 
54 static unsigned long store_indication __read_mostly;
55 
56 #ifdef CONFIG_64BIT
57 static int __init fault_init(void)
58 {
59 	if (test_facility(75))
60 		store_indication = 0xc00;
61 	return 0;
62 }
63 early_initcall(fault_init);
64 #endif
65 
66 static inline int notify_page_fault(struct pt_regs *regs)
67 {
68 	int ret = 0;
69 
70 	/* kprobe_running() needs smp_processor_id() */
71 	if (kprobes_built_in() && !user_mode(regs)) {
72 		preempt_disable();
73 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
74 			ret = 1;
75 		preempt_enable();
76 	}
77 	return ret;
78 }
79 
80 
81 /*
82  * Unlock any spinlocks which will prevent us from getting the
83  * message out.
84  */
85 void bust_spinlocks(int yes)
86 {
87 	if (yes) {
88 		oops_in_progress = 1;
89 	} else {
90 		int loglevel_save = console_loglevel;
91 		console_unblank();
92 		oops_in_progress = 0;
93 		/*
94 		 * OK, the message is on the console.  Now we call printk()
95 		 * without oops_in_progress set so that printk will give klogd
96 		 * a poke.  Hold onto your hats...
97 		 */
98 		console_loglevel = 15;
99 		printk(" ");
100 		console_loglevel = loglevel_save;
101 	}
102 }
103 
104 /*
105  * Returns the address space associated with the fault.
106  * Returns 0 for kernel space and 1 for user space.
107  */
108 static inline int user_space_fault(unsigned long trans_exc_code)
109 {
110 	/*
111 	 * The lowest two bits of the translation exception
112 	 * identification indicate which paging table was used.
113 	 */
114 	trans_exc_code &= 3;
115 	if (trans_exc_code == 2)
116 		/* Access via secondary space, set_fs setting decides */
117 		return current->thread.mm_segment.ar4;
118 	/*
119 	 * Access via primary space or access register is from user space
120 	 * and access via home space is from the kernel.
121 	 */
122 	return trans_exc_code != 3;
123 }
124 
125 static inline void report_user_fault(struct pt_regs *regs, long signr)
126 {
127 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
128 		return;
129 	if (!unhandled_signal(current, signr))
130 		return;
131 	if (!printk_ratelimit())
132 		return;
133 	printk(KERN_ALERT "User process fault: interruption code 0x%X ",
134 	       regs->int_code);
135 	print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
136 	printk(KERN_CONT "\n");
137 	printk(KERN_ALERT "failing address: %lX\n",
138 	       regs->int_parm_long & __FAIL_ADDR_MASK);
139 	show_regs(regs);
140 }
141 
142 /*
143  * Send SIGSEGV to task.  This is an external routine
144  * to keep the stack usage of do_page_fault small.
145  */
146 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
147 {
148 	struct siginfo si;
149 
150 	report_user_fault(regs, SIGSEGV);
151 	si.si_signo = SIGSEGV;
152 	si.si_code = si_code;
153 	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
154 	force_sig_info(SIGSEGV, &si, current);
155 }
156 
157 static noinline void do_no_context(struct pt_regs *regs)
158 {
159 	const struct exception_table_entry *fixup;
160 	unsigned long address;
161 
162 	/* Are we prepared to handle this kernel fault?  */
163 	fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
164 	if (fixup) {
165 		regs->psw.addr = extable_fixup(fixup) | PSW_ADDR_AMODE;
166 		return;
167 	}
168 
169 	/*
170 	 * Oops. The kernel tried to access some bad page. We'll have to
171 	 * terminate things with extreme prejudice.
172 	 */
173 	address = regs->int_parm_long & __FAIL_ADDR_MASK;
174 	if (!user_space_fault(regs->int_parm_long))
175 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
176 		       " at virtual kernel address %p\n", (void *)address);
177 	else
178 		printk(KERN_ALERT "Unable to handle kernel paging request"
179 		       " at virtual user address %p\n", (void *)address);
180 
181 	die(regs, "Oops");
182 	do_exit(SIGKILL);
183 }
184 
185 static noinline void do_low_address(struct pt_regs *regs)
186 {
187 	/* Low-address protection hit in kernel mode means
188 	   NULL pointer write access in kernel mode.  */
189 	if (regs->psw.mask & PSW_MASK_PSTATE) {
190 		/* Low-address protection hit in user mode 'cannot happen'. */
191 		die (regs, "Low-address protection");
192 		do_exit(SIGKILL);
193 	}
194 
195 	do_no_context(regs);
196 }
197 
198 static noinline void do_sigbus(struct pt_regs *regs)
199 {
200 	struct task_struct *tsk = current;
201 	struct siginfo si;
202 
203 	/*
204 	 * Send a sigbus, regardless of whether we were in kernel
205 	 * or user mode.
206 	 */
207 	si.si_signo = SIGBUS;
208 	si.si_errno = 0;
209 	si.si_code = BUS_ADRERR;
210 	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
211 	force_sig_info(SIGBUS, &si, tsk);
212 }
213 
214 static noinline void do_fault_error(struct pt_regs *regs, int fault)
215 {
216 	int si_code;
217 
218 	switch (fault) {
219 	case VM_FAULT_BADACCESS:
220 	case VM_FAULT_BADMAP:
221 		/* Bad memory access. Check if it is kernel or user space. */
222 		if (user_mode(regs)) {
223 			/* User mode accesses just cause a SIGSEGV */
224 			si_code = (fault == VM_FAULT_BADMAP) ?
225 				SEGV_MAPERR : SEGV_ACCERR;
226 			do_sigsegv(regs, si_code);
227 			return;
228 		}
229 	case VM_FAULT_BADCONTEXT:
230 		do_no_context(regs);
231 		break;
232 	case VM_FAULT_SIGNAL:
233 		if (!user_mode(regs))
234 			do_no_context(regs);
235 		break;
236 	default: /* fault & VM_FAULT_ERROR */
237 		if (fault & VM_FAULT_OOM) {
238 			if (!user_mode(regs))
239 				do_no_context(regs);
240 			else
241 				pagefault_out_of_memory();
242 		} else if (fault & VM_FAULT_SIGBUS) {
243 			/* Kernel mode? Handle exceptions or die */
244 			if (!user_mode(regs))
245 				do_no_context(regs);
246 			else
247 				do_sigbus(regs);
248 		} else
249 			BUG();
250 		break;
251 	}
252 }
253 
254 /*
255  * This routine handles page faults.  It determines the address,
256  * and the problem, and then passes it off to one of the appropriate
257  * routines.
258  *
259  * interruption code (int_code):
260  *   04       Protection           ->  Write-Protection  (suprression)
261  *   10       Segment translation  ->  Not present       (nullification)
262  *   11       Page translation     ->  Not present       (nullification)
263  *   3b       Region third trans.  ->  Not present       (nullification)
264  */
265 static inline int do_exception(struct pt_regs *regs, int access)
266 {
267 	struct task_struct *tsk;
268 	struct mm_struct *mm;
269 	struct vm_area_struct *vma;
270 	unsigned long trans_exc_code;
271 	unsigned long address;
272 	unsigned int flags;
273 	int fault;
274 
275 	tsk = current;
276 	/*
277 	 * The instruction that caused the program check has
278 	 * been nullified. Don't signal single step via SIGTRAP.
279 	 */
280 	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
281 
282 	if (notify_page_fault(regs))
283 		return 0;
284 
285 	mm = tsk->mm;
286 	trans_exc_code = regs->int_parm_long;
287 
288 	/*
289 	 * Verify that the fault happened in user space, that
290 	 * we are not in an interrupt and that there is a
291 	 * user context.
292 	 */
293 	fault = VM_FAULT_BADCONTEXT;
294 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
295 		goto out;
296 
297 	address = trans_exc_code & __FAIL_ADDR_MASK;
298 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
299 	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
300 	if (user_mode(regs))
301 		flags |= FAULT_FLAG_USER;
302 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
303 		flags |= FAULT_FLAG_WRITE;
304 	down_read(&mm->mmap_sem);
305 
306 #ifdef CONFIG_PGSTE
307 	if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
308 		address = __gmap_fault(address,
309 				     (struct gmap *) S390_lowcore.gmap);
310 		if (address == -EFAULT) {
311 			fault = VM_FAULT_BADMAP;
312 			goto out_up;
313 		}
314 		if (address == -ENOMEM) {
315 			fault = VM_FAULT_OOM;
316 			goto out_up;
317 		}
318 	}
319 #endif
320 
321 retry:
322 	fault = VM_FAULT_BADMAP;
323 	vma = find_vma(mm, address);
324 	if (!vma)
325 		goto out_up;
326 
327 	if (unlikely(vma->vm_start > address)) {
328 		if (!(vma->vm_flags & VM_GROWSDOWN))
329 			goto out_up;
330 		if (expand_stack(vma, address))
331 			goto out_up;
332 	}
333 
334 	/*
335 	 * Ok, we have a good vm_area for this memory access, so
336 	 * we can handle it..
337 	 */
338 	fault = VM_FAULT_BADACCESS;
339 	if (unlikely(!(vma->vm_flags & access)))
340 		goto out_up;
341 
342 	if (is_vm_hugetlb_page(vma))
343 		address &= HPAGE_MASK;
344 	/*
345 	 * If for any reason at all we couldn't handle the fault,
346 	 * make sure we exit gracefully rather than endlessly redo
347 	 * the fault.
348 	 */
349 	fault = handle_mm_fault(mm, vma, address, flags);
350 	/* No reason to continue if interrupted by SIGKILL. */
351 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
352 		fault = VM_FAULT_SIGNAL;
353 		goto out;
354 	}
355 	if (unlikely(fault & VM_FAULT_ERROR))
356 		goto out_up;
357 
358 	/*
359 	 * Major/minor page fault accounting is only done on the
360 	 * initial attempt. If we go through a retry, it is extremely
361 	 * likely that the page will be found in page cache at that point.
362 	 */
363 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
364 		if (fault & VM_FAULT_MAJOR) {
365 			tsk->maj_flt++;
366 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
367 				      regs, address);
368 		} else {
369 			tsk->min_flt++;
370 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
371 				      regs, address);
372 		}
373 		if (fault & VM_FAULT_RETRY) {
374 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
375 			 * of starvation. */
376 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
377 			flags |= FAULT_FLAG_TRIED;
378 			down_read(&mm->mmap_sem);
379 			goto retry;
380 		}
381 	}
382 	fault = 0;
383 out_up:
384 	up_read(&mm->mmap_sem);
385 out:
386 	return fault;
387 }
388 
389 void __kprobes do_protection_exception(struct pt_regs *regs)
390 {
391 	unsigned long trans_exc_code;
392 	int fault;
393 
394 	trans_exc_code = regs->int_parm_long;
395 	/*
396 	 * Protection exceptions are suppressing, decrement psw address.
397 	 * The exception to this rule are aborted transactions, for these
398 	 * the PSW already points to the correct location.
399 	 */
400 	if (!(regs->int_code & 0x200))
401 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
402 	/*
403 	 * Check for low-address protection.  This needs to be treated
404 	 * as a special case because the translation exception code
405 	 * field is not guaranteed to contain valid data in this case.
406 	 */
407 	if (unlikely(!(trans_exc_code & 4))) {
408 		do_low_address(regs);
409 		return;
410 	}
411 	fault = do_exception(regs, VM_WRITE);
412 	if (unlikely(fault))
413 		do_fault_error(regs, fault);
414 }
415 
416 void __kprobes do_dat_exception(struct pt_regs *regs)
417 {
418 	int access, fault;
419 
420 	access = VM_READ | VM_EXEC | VM_WRITE;
421 	fault = do_exception(regs, access);
422 	if (unlikely(fault))
423 		do_fault_error(regs, fault);
424 }
425 
426 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
427 {
428 	struct pt_regs regs;
429 	int access, fault;
430 
431 	/* Emulate a uaccess fault from kernel mode. */
432 	regs.psw.mask = PSW_KERNEL_BITS | PSW_MASK_DAT | PSW_MASK_MCHECK;
433 	if (!irqs_disabled())
434 		regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
435 	regs.psw.addr = (unsigned long) __builtin_return_address(0);
436 	regs.psw.addr |= PSW_ADDR_AMODE;
437 	regs.int_code = pgm_int_code;
438 	regs.int_parm_long = (uaddr & PAGE_MASK) | 2;
439 	access = write ? VM_WRITE : VM_READ;
440 	fault = do_exception(&regs, access);
441 	/*
442 	 * Since the fault happened in kernel mode while performing a uaccess
443 	 * all we need to do now is emulating a fixup in case "fault" is not
444 	 * zero.
445 	 * For the calling uaccess functions this results always in -EFAULT.
446 	 */
447 	return fault ? -EFAULT : 0;
448 }
449 
450 #ifdef CONFIG_PFAULT
451 /*
452  * 'pfault' pseudo page faults routines.
453  */
454 static int pfault_disable;
455 
456 static int __init nopfault(char *str)
457 {
458 	pfault_disable = 1;
459 	return 1;
460 }
461 
462 __setup("nopfault", nopfault);
463 
464 struct pfault_refbk {
465 	u16 refdiagc;
466 	u16 reffcode;
467 	u16 refdwlen;
468 	u16 refversn;
469 	u64 refgaddr;
470 	u64 refselmk;
471 	u64 refcmpmk;
472 	u64 reserved;
473 } __attribute__ ((packed, aligned(8)));
474 
475 int pfault_init(void)
476 {
477 	struct pfault_refbk refbk = {
478 		.refdiagc = 0x258,
479 		.reffcode = 0,
480 		.refdwlen = 5,
481 		.refversn = 2,
482 		.refgaddr = __LC_CURRENT_PID,
483 		.refselmk = 1ULL << 48,
484 		.refcmpmk = 1ULL << 48,
485 		.reserved = __PF_RES_FIELD };
486         int rc;
487 
488 	if (pfault_disable)
489 		return -1;
490 	asm volatile(
491 		"	diag	%1,%0,0x258\n"
492 		"0:	j	2f\n"
493 		"1:	la	%0,8\n"
494 		"2:\n"
495 		EX_TABLE(0b,1b)
496 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
497         return rc;
498 }
499 
500 void pfault_fini(void)
501 {
502 	struct pfault_refbk refbk = {
503 		.refdiagc = 0x258,
504 		.reffcode = 1,
505 		.refdwlen = 5,
506 		.refversn = 2,
507 	};
508 
509 	if (pfault_disable)
510 		return;
511 	asm volatile(
512 		"	diag	%0,0,0x258\n"
513 		"0:\n"
514 		EX_TABLE(0b,0b)
515 		: : "a" (&refbk), "m" (refbk) : "cc");
516 }
517 
518 static DEFINE_SPINLOCK(pfault_lock);
519 static LIST_HEAD(pfault_list);
520 
521 static void pfault_interrupt(struct ext_code ext_code,
522 			     unsigned int param32, unsigned long param64)
523 {
524 	struct task_struct *tsk;
525 	__u16 subcode;
526 	pid_t pid;
527 
528 	/*
529 	 * Get the external interruption subcode & pfault
530 	 * initial/completion signal bit. VM stores this
531 	 * in the 'cpu address' field associated with the
532          * external interrupt.
533 	 */
534 	subcode = ext_code.subcode;
535 	if ((subcode & 0xff00) != __SUBCODE_MASK)
536 		return;
537 	inc_irq_stat(IRQEXT_PFL);
538 	/* Get the token (= pid of the affected task). */
539 	pid = sizeof(void *) == 4 ? param32 : param64;
540 	rcu_read_lock();
541 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
542 	if (tsk)
543 		get_task_struct(tsk);
544 	rcu_read_unlock();
545 	if (!tsk)
546 		return;
547 	spin_lock(&pfault_lock);
548 	if (subcode & 0x0080) {
549 		/* signal bit is set -> a page has been swapped in by VM */
550 		if (tsk->thread.pfault_wait == 1) {
551 			/* Initial interrupt was faster than the completion
552 			 * interrupt. pfault_wait is valid. Set pfault_wait
553 			 * back to zero and wake up the process. This can
554 			 * safely be done because the task is still sleeping
555 			 * and can't produce new pfaults. */
556 			tsk->thread.pfault_wait = 0;
557 			list_del(&tsk->thread.list);
558 			wake_up_process(tsk);
559 			put_task_struct(tsk);
560 		} else {
561 			/* Completion interrupt was faster than initial
562 			 * interrupt. Set pfault_wait to -1 so the initial
563 			 * interrupt doesn't put the task to sleep.
564 			 * If the task is not running, ignore the completion
565 			 * interrupt since it must be a leftover of a PFAULT
566 			 * CANCEL operation which didn't remove all pending
567 			 * completion interrupts. */
568 			if (tsk->state == TASK_RUNNING)
569 				tsk->thread.pfault_wait = -1;
570 		}
571 	} else {
572 		/* signal bit not set -> a real page is missing. */
573 		if (WARN_ON_ONCE(tsk != current))
574 			goto out;
575 		if (tsk->thread.pfault_wait == 1) {
576 			/* Already on the list with a reference: put to sleep */
577 			__set_task_state(tsk, TASK_UNINTERRUPTIBLE);
578 			set_tsk_need_resched(tsk);
579 		} else if (tsk->thread.pfault_wait == -1) {
580 			/* Completion interrupt was faster than the initial
581 			 * interrupt (pfault_wait == -1). Set pfault_wait
582 			 * back to zero and exit. */
583 			tsk->thread.pfault_wait = 0;
584 		} else {
585 			/* Initial interrupt arrived before completion
586 			 * interrupt. Let the task sleep.
587 			 * An extra task reference is needed since a different
588 			 * cpu may set the task state to TASK_RUNNING again
589 			 * before the scheduler is reached. */
590 			get_task_struct(tsk);
591 			tsk->thread.pfault_wait = 1;
592 			list_add(&tsk->thread.list, &pfault_list);
593 			__set_task_state(tsk, TASK_UNINTERRUPTIBLE);
594 			set_tsk_need_resched(tsk);
595 		}
596 	}
597 out:
598 	spin_unlock(&pfault_lock);
599 	put_task_struct(tsk);
600 }
601 
602 static int pfault_cpu_notify(struct notifier_block *self, unsigned long action,
603 			     void *hcpu)
604 {
605 	struct thread_struct *thread, *next;
606 	struct task_struct *tsk;
607 
608 	switch (action & ~CPU_TASKS_FROZEN) {
609 	case CPU_DEAD:
610 		spin_lock_irq(&pfault_lock);
611 		list_for_each_entry_safe(thread, next, &pfault_list, list) {
612 			thread->pfault_wait = 0;
613 			list_del(&thread->list);
614 			tsk = container_of(thread, struct task_struct, thread);
615 			wake_up_process(tsk);
616 			put_task_struct(tsk);
617 		}
618 		spin_unlock_irq(&pfault_lock);
619 		break;
620 	default:
621 		break;
622 	}
623 	return NOTIFY_OK;
624 }
625 
626 static int __init pfault_irq_init(void)
627 {
628 	int rc;
629 
630 	rc = register_external_interrupt(0x2603, pfault_interrupt);
631 	if (rc)
632 		goto out_extint;
633 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
634 	if (rc)
635 		goto out_pfault;
636 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
637 	hotcpu_notifier(pfault_cpu_notify, 0);
638 	return 0;
639 
640 out_pfault:
641 	unregister_external_interrupt(0x2603, pfault_interrupt);
642 out_extint:
643 	pfault_disable = 1;
644 	return rc;
645 }
646 early_initcall(pfault_irq_init);
647 
648 #endif /* CONFIG_PFAULT */
649