xref: /openbmc/linux/arch/s390/mm/fault.c (revision 81d67439)
1 /*
2  *  arch/s390/mm/fault.c
3  *
4  *  S390 version
5  *    Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6  *    Author(s): Hartmut Penner (hp@de.ibm.com)
7  *               Ulrich Weigand (uweigand@de.ibm.com)
8  *
9  *  Derived from "arch/i386/mm/fault.c"
10  *    Copyright (C) 1995  Linus Torvalds
11  */
12 
13 #include <linux/kernel_stat.h>
14 #include <linux/perf_event.h>
15 #include <linux/signal.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
23 #include <linux/mm.h>
24 #include <linux/compat.h>
25 #include <linux/smp.h>
26 #include <linux/kdebug.h>
27 #include <linux/init.h>
28 #include <linux/console.h>
29 #include <linux/module.h>
30 #include <linux/hardirq.h>
31 #include <linux/kprobes.h>
32 #include <linux/uaccess.h>
33 #include <linux/hugetlb.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/system.h>
36 #include <asm/pgtable.h>
37 #include <asm/irq.h>
38 #include <asm/mmu_context.h>
39 #include <asm/compat.h>
40 #include "../kernel/entry.h"
41 
42 #ifndef CONFIG_64BIT
43 #define __FAIL_ADDR_MASK 0x7ffff000
44 #define __SUBCODE_MASK 0x0200
45 #define __PF_RES_FIELD 0ULL
46 #else /* CONFIG_64BIT */
47 #define __FAIL_ADDR_MASK -4096L
48 #define __SUBCODE_MASK 0x0600
49 #define __PF_RES_FIELD 0x8000000000000000ULL
50 #endif /* CONFIG_64BIT */
51 
52 #define VM_FAULT_BADCONTEXT	0x010000
53 #define VM_FAULT_BADMAP		0x020000
54 #define VM_FAULT_BADACCESS	0x040000
55 
56 static unsigned long store_indication;
57 
58 void fault_init(void)
59 {
60 	if (test_facility(2) && test_facility(75))
61 		store_indication = 0xc00;
62 }
63 
64 static inline int notify_page_fault(struct pt_regs *regs)
65 {
66 	int ret = 0;
67 
68 	/* kprobe_running() needs smp_processor_id() */
69 	if (kprobes_built_in() && !user_mode(regs)) {
70 		preempt_disable();
71 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
72 			ret = 1;
73 		preempt_enable();
74 	}
75 	return ret;
76 }
77 
78 
79 /*
80  * Unlock any spinlocks which will prevent us from getting the
81  * message out.
82  */
83 void bust_spinlocks(int yes)
84 {
85 	if (yes) {
86 		oops_in_progress = 1;
87 	} else {
88 		int loglevel_save = console_loglevel;
89 		console_unblank();
90 		oops_in_progress = 0;
91 		/*
92 		 * OK, the message is on the console.  Now we call printk()
93 		 * without oops_in_progress set so that printk will give klogd
94 		 * a poke.  Hold onto your hats...
95 		 */
96 		console_loglevel = 15;
97 		printk(" ");
98 		console_loglevel = loglevel_save;
99 	}
100 }
101 
102 /*
103  * Returns the address space associated with the fault.
104  * Returns 0 for kernel space and 1 for user space.
105  */
106 static inline int user_space_fault(unsigned long trans_exc_code)
107 {
108 	/*
109 	 * The lowest two bits of the translation exception
110 	 * identification indicate which paging table was used.
111 	 */
112 	trans_exc_code &= 3;
113 	if (trans_exc_code == 2)
114 		/* Access via secondary space, set_fs setting decides */
115 		return current->thread.mm_segment.ar4;
116 	if (user_mode == HOME_SPACE_MODE)
117 		/* User space if the access has been done via home space. */
118 		return trans_exc_code == 3;
119 	/*
120 	 * If the user space is not the home space the kernel runs in home
121 	 * space. Access via secondary space has already been covered,
122 	 * access via primary space or access register is from user space
123 	 * and access via home space is from the kernel.
124 	 */
125 	return trans_exc_code != 3;
126 }
127 
128 static inline void report_user_fault(struct pt_regs *regs, long int_code,
129 				     int signr, unsigned long address)
130 {
131 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
132 		return;
133 	if (!unhandled_signal(current, signr))
134 		return;
135 	if (!printk_ratelimit())
136 		return;
137 	printk("User process fault: interruption code 0x%lX ", int_code);
138 	print_vma_addr(KERN_CONT "in ", regs->psw.addr & PSW_ADDR_INSN);
139 	printk("\n");
140 	printk("failing address: %lX\n", address);
141 	show_regs(regs);
142 }
143 
144 /*
145  * Send SIGSEGV to task.  This is an external routine
146  * to keep the stack usage of do_page_fault small.
147  */
148 static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
149 				int si_code, unsigned long trans_exc_code)
150 {
151 	struct siginfo si;
152 	unsigned long address;
153 
154 	address = trans_exc_code & __FAIL_ADDR_MASK;
155 	current->thread.prot_addr = address;
156 	current->thread.trap_no = int_code;
157 	report_user_fault(regs, int_code, SIGSEGV, address);
158 	si.si_signo = SIGSEGV;
159 	si.si_code = si_code;
160 	si.si_addr = (void __user *) address;
161 	force_sig_info(SIGSEGV, &si, current);
162 }
163 
164 static noinline void do_no_context(struct pt_regs *regs, long int_code,
165 				   unsigned long trans_exc_code)
166 {
167 	const struct exception_table_entry *fixup;
168 	unsigned long address;
169 
170 	/* Are we prepared to handle this kernel fault?  */
171 	fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
172 	if (fixup) {
173 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
174 		return;
175 	}
176 
177 	/*
178 	 * Oops. The kernel tried to access some bad page. We'll have to
179 	 * terminate things with extreme prejudice.
180 	 */
181 	address = trans_exc_code & __FAIL_ADDR_MASK;
182 	if (!user_space_fault(trans_exc_code))
183 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
184 		       " at virtual kernel address %p\n", (void *)address);
185 	else
186 		printk(KERN_ALERT "Unable to handle kernel paging request"
187 		       " at virtual user address %p\n", (void *)address);
188 
189 	die("Oops", regs, int_code);
190 	do_exit(SIGKILL);
191 }
192 
193 static noinline void do_low_address(struct pt_regs *regs, long int_code,
194 				    unsigned long trans_exc_code)
195 {
196 	/* Low-address protection hit in kernel mode means
197 	   NULL pointer write access in kernel mode.  */
198 	if (regs->psw.mask & PSW_MASK_PSTATE) {
199 		/* Low-address protection hit in user mode 'cannot happen'. */
200 		die ("Low-address protection", regs, int_code);
201 		do_exit(SIGKILL);
202 	}
203 
204 	do_no_context(regs, int_code, trans_exc_code);
205 }
206 
207 static noinline void do_sigbus(struct pt_regs *regs, long int_code,
208 			       unsigned long trans_exc_code)
209 {
210 	struct task_struct *tsk = current;
211 	unsigned long address;
212 	struct siginfo si;
213 
214 	/*
215 	 * Send a sigbus, regardless of whether we were in kernel
216 	 * or user mode.
217 	 */
218 	address = trans_exc_code & __FAIL_ADDR_MASK;
219 	tsk->thread.prot_addr = address;
220 	tsk->thread.trap_no = int_code;
221 	si.si_signo = SIGBUS;
222 	si.si_errno = 0;
223 	si.si_code = BUS_ADRERR;
224 	si.si_addr = (void __user *) address;
225 	force_sig_info(SIGBUS, &si, tsk);
226 }
227 
228 static noinline void do_fault_error(struct pt_regs *regs, long int_code,
229 				    unsigned long trans_exc_code, int fault)
230 {
231 	int si_code;
232 
233 	switch (fault) {
234 	case VM_FAULT_BADACCESS:
235 	case VM_FAULT_BADMAP:
236 		/* Bad memory access. Check if it is kernel or user space. */
237 		if (regs->psw.mask & PSW_MASK_PSTATE) {
238 			/* User mode accesses just cause a SIGSEGV */
239 			si_code = (fault == VM_FAULT_BADMAP) ?
240 				SEGV_MAPERR : SEGV_ACCERR;
241 			do_sigsegv(regs, int_code, si_code, trans_exc_code);
242 			return;
243 		}
244 	case VM_FAULT_BADCONTEXT:
245 		do_no_context(regs, int_code, trans_exc_code);
246 		break;
247 	default: /* fault & VM_FAULT_ERROR */
248 		if (fault & VM_FAULT_OOM) {
249 			if (!(regs->psw.mask & PSW_MASK_PSTATE))
250 				do_no_context(regs, int_code, trans_exc_code);
251 			else
252 				pagefault_out_of_memory();
253 		} else if (fault & VM_FAULT_SIGBUS) {
254 			/* Kernel mode? Handle exceptions or die */
255 			if (!(regs->psw.mask & PSW_MASK_PSTATE))
256 				do_no_context(regs, int_code, trans_exc_code);
257 			else
258 				do_sigbus(regs, int_code, trans_exc_code);
259 		} else
260 			BUG();
261 		break;
262 	}
263 }
264 
265 /*
266  * This routine handles page faults.  It determines the address,
267  * and the problem, and then passes it off to one of the appropriate
268  * routines.
269  *
270  * interruption code (int_code):
271  *   04       Protection           ->  Write-Protection  (suprression)
272  *   10       Segment translation  ->  Not present       (nullification)
273  *   11       Page translation     ->  Not present       (nullification)
274  *   3b       Region third trans.  ->  Not present       (nullification)
275  */
276 static inline int do_exception(struct pt_regs *regs, int access,
277 			       unsigned long trans_exc_code)
278 {
279 	struct task_struct *tsk;
280 	struct mm_struct *mm;
281 	struct vm_area_struct *vma;
282 	unsigned long address;
283 	unsigned int flags;
284 	int fault;
285 
286 	if (notify_page_fault(regs))
287 		return 0;
288 
289 	tsk = current;
290 	mm = tsk->mm;
291 
292 	/*
293 	 * Verify that the fault happened in user space, that
294 	 * we are not in an interrupt and that there is a
295 	 * user context.
296 	 */
297 	fault = VM_FAULT_BADCONTEXT;
298 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
299 		goto out;
300 
301 	address = trans_exc_code & __FAIL_ADDR_MASK;
302 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
303 	flags = FAULT_FLAG_ALLOW_RETRY;
304 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
305 		flags |= FAULT_FLAG_WRITE;
306 	down_read(&mm->mmap_sem);
307 
308 #ifdef CONFIG_PGSTE
309 	if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
310 		address = gmap_fault(address,
311 				     (struct gmap *) S390_lowcore.gmap);
312 		if (address == -EFAULT) {
313 			fault = VM_FAULT_BADMAP;
314 			goto out_up;
315 		}
316 		if (address == -ENOMEM) {
317 			fault = VM_FAULT_OOM;
318 			goto out_up;
319 		}
320 	}
321 #endif
322 
323 retry:
324 	fault = VM_FAULT_BADMAP;
325 	vma = find_vma(mm, address);
326 	if (!vma)
327 		goto out_up;
328 
329 	if (unlikely(vma->vm_start > address)) {
330 		if (!(vma->vm_flags & VM_GROWSDOWN))
331 			goto out_up;
332 		if (expand_stack(vma, address))
333 			goto out_up;
334 	}
335 
336 	/*
337 	 * Ok, we have a good vm_area for this memory access, so
338 	 * we can handle it..
339 	 */
340 	fault = VM_FAULT_BADACCESS;
341 	if (unlikely(!(vma->vm_flags & access)))
342 		goto out_up;
343 
344 	if (is_vm_hugetlb_page(vma))
345 		address &= HPAGE_MASK;
346 	/*
347 	 * If for any reason at all we couldn't handle the fault,
348 	 * make sure we exit gracefully rather than endlessly redo
349 	 * the fault.
350 	 */
351 	fault = handle_mm_fault(mm, vma, address, flags);
352 	if (unlikely(fault & VM_FAULT_ERROR))
353 		goto out_up;
354 
355 	/*
356 	 * Major/minor page fault accounting is only done on the
357 	 * initial attempt. If we go through a retry, it is extremely
358 	 * likely that the page will be found in page cache at that point.
359 	 */
360 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
361 		if (fault & VM_FAULT_MAJOR) {
362 			tsk->maj_flt++;
363 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
364 				      regs, address);
365 		} else {
366 			tsk->min_flt++;
367 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
368 				      regs, address);
369 		}
370 		if (fault & VM_FAULT_RETRY) {
371 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
372 			 * of starvation. */
373 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
374 			down_read(&mm->mmap_sem);
375 			goto retry;
376 		}
377 	}
378 	/*
379 	 * The instruction that caused the program check will
380 	 * be repeated. Don't signal single step via SIGTRAP.
381 	 */
382 	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
383 	fault = 0;
384 out_up:
385 	up_read(&mm->mmap_sem);
386 out:
387 	return fault;
388 }
389 
390 void __kprobes do_protection_exception(struct pt_regs *regs, long pgm_int_code,
391 				       unsigned long trans_exc_code)
392 {
393 	int fault;
394 
395 	/* Protection exception is suppressing, decrement psw address. */
396 	regs->psw.addr -= (pgm_int_code >> 16);
397 	/*
398 	 * Check for low-address protection.  This needs to be treated
399 	 * as a special case because the translation exception code
400 	 * field is not guaranteed to contain valid data in this case.
401 	 */
402 	if (unlikely(!(trans_exc_code & 4))) {
403 		do_low_address(regs, pgm_int_code, trans_exc_code);
404 		return;
405 	}
406 	fault = do_exception(regs, VM_WRITE, trans_exc_code);
407 	if (unlikely(fault))
408 		do_fault_error(regs, 4, trans_exc_code, fault);
409 }
410 
411 void __kprobes do_dat_exception(struct pt_regs *regs, long pgm_int_code,
412 				unsigned long trans_exc_code)
413 {
414 	int access, fault;
415 
416 	access = VM_READ | VM_EXEC | VM_WRITE;
417 	fault = do_exception(regs, access, trans_exc_code);
418 	if (unlikely(fault))
419 		do_fault_error(regs, pgm_int_code & 255, trans_exc_code, fault);
420 }
421 
422 #ifdef CONFIG_64BIT
423 void __kprobes do_asce_exception(struct pt_regs *regs, long pgm_int_code,
424 				 unsigned long trans_exc_code)
425 {
426 	struct mm_struct *mm = current->mm;
427 	struct vm_area_struct *vma;
428 
429 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
430 		goto no_context;
431 
432 	down_read(&mm->mmap_sem);
433 	vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
434 	up_read(&mm->mmap_sem);
435 
436 	if (vma) {
437 		update_mm(mm, current);
438 		return;
439 	}
440 
441 	/* User mode accesses just cause a SIGSEGV */
442 	if (regs->psw.mask & PSW_MASK_PSTATE) {
443 		do_sigsegv(regs, pgm_int_code, SEGV_MAPERR, trans_exc_code);
444 		return;
445 	}
446 
447 no_context:
448 	do_no_context(regs, pgm_int_code, trans_exc_code);
449 }
450 #endif
451 
452 int __handle_fault(unsigned long uaddr, unsigned long pgm_int_code, int write)
453 {
454 	struct pt_regs regs;
455 	int access, fault;
456 
457 	regs.psw.mask = psw_kernel_bits;
458 	if (!irqs_disabled())
459 		regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
460 	regs.psw.addr = (unsigned long) __builtin_return_address(0);
461 	regs.psw.addr |= PSW_ADDR_AMODE;
462 	uaddr &= PAGE_MASK;
463 	access = write ? VM_WRITE : VM_READ;
464 	fault = do_exception(&regs, access, uaddr | 2);
465 	if (unlikely(fault)) {
466 		if (fault & VM_FAULT_OOM)
467 			return -EFAULT;
468 		else if (fault & VM_FAULT_SIGBUS)
469 			do_sigbus(&regs, pgm_int_code, uaddr);
470 	}
471 	return fault ? -EFAULT : 0;
472 }
473 
474 #ifdef CONFIG_PFAULT
475 /*
476  * 'pfault' pseudo page faults routines.
477  */
478 static int pfault_disable;
479 
480 static int __init nopfault(char *str)
481 {
482 	pfault_disable = 1;
483 	return 1;
484 }
485 
486 __setup("nopfault", nopfault);
487 
488 struct pfault_refbk {
489 	u16 refdiagc;
490 	u16 reffcode;
491 	u16 refdwlen;
492 	u16 refversn;
493 	u64 refgaddr;
494 	u64 refselmk;
495 	u64 refcmpmk;
496 	u64 reserved;
497 } __attribute__ ((packed, aligned(8)));
498 
499 int pfault_init(void)
500 {
501 	struct pfault_refbk refbk = {
502 		.refdiagc = 0x258,
503 		.reffcode = 0,
504 		.refdwlen = 5,
505 		.refversn = 2,
506 		.refgaddr = __LC_CURRENT_PID,
507 		.refselmk = 1ULL << 48,
508 		.refcmpmk = 1ULL << 48,
509 		.reserved = __PF_RES_FIELD };
510         int rc;
511 
512 	if (!MACHINE_IS_VM || pfault_disable)
513 		return -1;
514 	asm volatile(
515 		"	diag	%1,%0,0x258\n"
516 		"0:	j	2f\n"
517 		"1:	la	%0,8\n"
518 		"2:\n"
519 		EX_TABLE(0b,1b)
520 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
521         return rc;
522 }
523 
524 void pfault_fini(void)
525 {
526 	struct pfault_refbk refbk = {
527 		.refdiagc = 0x258,
528 		.reffcode = 1,
529 		.refdwlen = 5,
530 		.refversn = 2,
531 	};
532 
533 	if (!MACHINE_IS_VM || pfault_disable)
534 		return;
535 	asm volatile(
536 		"	diag	%0,0,0x258\n"
537 		"0:\n"
538 		EX_TABLE(0b,0b)
539 		: : "a" (&refbk), "m" (refbk) : "cc");
540 }
541 
542 static DEFINE_SPINLOCK(pfault_lock);
543 static LIST_HEAD(pfault_list);
544 
545 static void pfault_interrupt(unsigned int ext_int_code,
546 			     unsigned int param32, unsigned long param64)
547 {
548 	struct task_struct *tsk;
549 	__u16 subcode;
550 	pid_t pid;
551 
552 	/*
553 	 * Get the external interruption subcode & pfault
554 	 * initial/completion signal bit. VM stores this
555 	 * in the 'cpu address' field associated with the
556          * external interrupt.
557 	 */
558 	subcode = ext_int_code >> 16;
559 	if ((subcode & 0xff00) != __SUBCODE_MASK)
560 		return;
561 	kstat_cpu(smp_processor_id()).irqs[EXTINT_PFL]++;
562 	if (subcode & 0x0080) {
563 		/* Get the token (= pid of the affected task). */
564 		pid = sizeof(void *) == 4 ? param32 : param64;
565 		rcu_read_lock();
566 		tsk = find_task_by_pid_ns(pid, &init_pid_ns);
567 		if (tsk)
568 			get_task_struct(tsk);
569 		rcu_read_unlock();
570 		if (!tsk)
571 			return;
572 	} else {
573 		tsk = current;
574 	}
575 	spin_lock(&pfault_lock);
576 	if (subcode & 0x0080) {
577 		/* signal bit is set -> a page has been swapped in by VM */
578 		if (tsk->thread.pfault_wait == 1) {
579 			/* Initial interrupt was faster than the completion
580 			 * interrupt. pfault_wait is valid. Set pfault_wait
581 			 * back to zero and wake up the process. This can
582 			 * safely be done because the task is still sleeping
583 			 * and can't produce new pfaults. */
584 			tsk->thread.pfault_wait = 0;
585 			list_del(&tsk->thread.list);
586 			wake_up_process(tsk);
587 		} else {
588 			/* Completion interrupt was faster than initial
589 			 * interrupt. Set pfault_wait to -1 so the initial
590 			 * interrupt doesn't put the task to sleep. */
591 			tsk->thread.pfault_wait = -1;
592 		}
593 		put_task_struct(tsk);
594 	} else {
595 		/* signal bit not set -> a real page is missing. */
596 		if (tsk->thread.pfault_wait == -1) {
597 			/* Completion interrupt was faster than the initial
598 			 * interrupt (pfault_wait == -1). Set pfault_wait
599 			 * back to zero and exit. */
600 			tsk->thread.pfault_wait = 0;
601 		} else {
602 			/* Initial interrupt arrived before completion
603 			 * interrupt. Let the task sleep. */
604 			tsk->thread.pfault_wait = 1;
605 			list_add(&tsk->thread.list, &pfault_list);
606 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
607 			set_tsk_need_resched(tsk);
608 		}
609 	}
610 	spin_unlock(&pfault_lock);
611 }
612 
613 static int __cpuinit pfault_cpu_notify(struct notifier_block *self,
614 				       unsigned long action, void *hcpu)
615 {
616 	struct thread_struct *thread, *next;
617 	struct task_struct *tsk;
618 
619 	switch (action) {
620 	case CPU_DEAD:
621 	case CPU_DEAD_FROZEN:
622 		spin_lock_irq(&pfault_lock);
623 		list_for_each_entry_safe(thread, next, &pfault_list, list) {
624 			thread->pfault_wait = 0;
625 			list_del(&thread->list);
626 			tsk = container_of(thread, struct task_struct, thread);
627 			wake_up_process(tsk);
628 		}
629 		spin_unlock_irq(&pfault_lock);
630 		break;
631 	default:
632 		break;
633 	}
634 	return NOTIFY_OK;
635 }
636 
637 static int __init pfault_irq_init(void)
638 {
639 	int rc;
640 
641 	if (!MACHINE_IS_VM)
642 		return 0;
643 	rc = register_external_interrupt(0x2603, pfault_interrupt);
644 	if (rc)
645 		goto out_extint;
646 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
647 	if (rc)
648 		goto out_pfault;
649 	service_subclass_irq_register();
650 	hotcpu_notifier(pfault_cpu_notify, 0);
651 	return 0;
652 
653 out_pfault:
654 	unregister_external_interrupt(0x2603, pfault_interrupt);
655 out_extint:
656 	pfault_disable = 1;
657 	return rc;
658 }
659 early_initcall(pfault_irq_init);
660 
661 #endif /* CONFIG_PFAULT */
662