xref: /openbmc/linux/arch/s390/mm/fault.c (revision 63705da3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  S390 version
4  *    Copyright IBM Corp. 1999
5  *    Author(s): Hartmut Penner (hp@de.ibm.com)
6  *               Ulrich Weigand (uweigand@de.ibm.com)
7  *
8  *  Derived from "arch/i386/mm/fault.c"
9  *    Copyright (C) 1995  Linus Torvalds
10  */
11 
12 #include <linux/kernel_stat.h>
13 #include <linux/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.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/extable.h>
30 #include <linux/hardirq.h>
31 #include <linux/kprobes.h>
32 #include <linux/uaccess.h>
33 #include <linux/hugetlb.h>
34 #include <linux/kfence.h>
35 #include <asm/asm-offsets.h>
36 #include <asm/diag.h>
37 #include <asm/gmap.h>
38 #include <asm/irq.h>
39 #include <asm/mmu_context.h>
40 #include <asm/facility.h>
41 #include <asm/uv.h>
42 #include "../kernel/entry.h"
43 
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 
48 #define VM_FAULT_BADCONTEXT	((__force vm_fault_t) 0x010000)
49 #define VM_FAULT_BADMAP		((__force vm_fault_t) 0x020000)
50 #define VM_FAULT_BADACCESS	((__force vm_fault_t) 0x040000)
51 #define VM_FAULT_SIGNAL		((__force vm_fault_t) 0x080000)
52 #define VM_FAULT_PFAULT		((__force vm_fault_t) 0x100000)
53 
54 enum fault_type {
55 	KERNEL_FAULT,
56 	USER_FAULT,
57 	GMAP_FAULT,
58 };
59 
60 static unsigned long store_indication __read_mostly;
61 
62 static int __init fault_init(void)
63 {
64 	if (test_facility(75))
65 		store_indication = 0xc00;
66 	return 0;
67 }
68 early_initcall(fault_init);
69 
70 /*
71  * Find out which address space caused the exception.
72  */
73 static enum fault_type get_fault_type(struct pt_regs *regs)
74 {
75 	unsigned long trans_exc_code;
76 
77 	trans_exc_code = regs->int_parm_long & 3;
78 	if (likely(trans_exc_code == 0)) {
79 		/* primary space exception */
80 		if (user_mode(regs))
81 			return USER_FAULT;
82 		if (!IS_ENABLED(CONFIG_PGSTE))
83 			return KERNEL_FAULT;
84 		if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
85 			return GMAP_FAULT;
86 		return KERNEL_FAULT;
87 	}
88 	if (trans_exc_code == 2)
89 		return USER_FAULT;
90 	if (trans_exc_code == 1) {
91 		/* access register mode, not used in the kernel */
92 		return USER_FAULT;
93 	}
94 	/* home space exception -> access via kernel ASCE */
95 	return KERNEL_FAULT;
96 }
97 
98 static int bad_address(void *p)
99 {
100 	unsigned long dummy;
101 
102 	return get_kernel_nofault(dummy, (unsigned long *)p);
103 }
104 
105 static void dump_pagetable(unsigned long asce, unsigned long address)
106 {
107 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
108 
109 	pr_alert("AS:%016lx ", asce);
110 	switch (asce & _ASCE_TYPE_MASK) {
111 	case _ASCE_TYPE_REGION1:
112 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
113 		if (bad_address(table))
114 			goto bad;
115 		pr_cont("R1:%016lx ", *table);
116 		if (*table & _REGION_ENTRY_INVALID)
117 			goto out;
118 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
119 		fallthrough;
120 	case _ASCE_TYPE_REGION2:
121 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
122 		if (bad_address(table))
123 			goto bad;
124 		pr_cont("R2:%016lx ", *table);
125 		if (*table & _REGION_ENTRY_INVALID)
126 			goto out;
127 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
128 		fallthrough;
129 	case _ASCE_TYPE_REGION3:
130 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
131 		if (bad_address(table))
132 			goto bad;
133 		pr_cont("R3:%016lx ", *table);
134 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
135 			goto out;
136 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
137 		fallthrough;
138 	case _ASCE_TYPE_SEGMENT:
139 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
140 		if (bad_address(table))
141 			goto bad;
142 		pr_cont("S:%016lx ", *table);
143 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
144 			goto out;
145 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
146 	}
147 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
148 	if (bad_address(table))
149 		goto bad;
150 	pr_cont("P:%016lx ", *table);
151 out:
152 	pr_cont("\n");
153 	return;
154 bad:
155 	pr_cont("BAD\n");
156 }
157 
158 static void dump_fault_info(struct pt_regs *regs)
159 {
160 	unsigned long asce;
161 
162 	pr_alert("Failing address: %016lx TEID: %016lx\n",
163 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
164 	pr_alert("Fault in ");
165 	switch (regs->int_parm_long & 3) {
166 	case 3:
167 		pr_cont("home space ");
168 		break;
169 	case 2:
170 		pr_cont("secondary space ");
171 		break;
172 	case 1:
173 		pr_cont("access register ");
174 		break;
175 	case 0:
176 		pr_cont("primary space ");
177 		break;
178 	}
179 	pr_cont("mode while using ");
180 	switch (get_fault_type(regs)) {
181 	case USER_FAULT:
182 		asce = S390_lowcore.user_asce;
183 		pr_cont("user ");
184 		break;
185 	case GMAP_FAULT:
186 		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
187 		pr_cont("gmap ");
188 		break;
189 	case KERNEL_FAULT:
190 		asce = S390_lowcore.kernel_asce;
191 		pr_cont("kernel ");
192 		break;
193 	default:
194 		unreachable();
195 	}
196 	pr_cont("ASCE.\n");
197 	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
198 }
199 
200 int show_unhandled_signals = 1;
201 
202 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
203 {
204 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
205 		return;
206 	if (!unhandled_signal(current, signr))
207 		return;
208 	if (!printk_ratelimit())
209 		return;
210 	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
211 	       regs->int_code & 0xffff, regs->int_code >> 17);
212 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
213 	printk(KERN_CONT "\n");
214 	if (is_mm_fault)
215 		dump_fault_info(regs);
216 	show_regs(regs);
217 }
218 
219 /*
220  * Send SIGSEGV to task.  This is an external routine
221  * to keep the stack usage of do_page_fault small.
222  */
223 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
224 {
225 	report_user_fault(regs, SIGSEGV, 1);
226 	force_sig_fault(SIGSEGV, si_code,
227 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
228 }
229 
230 const struct exception_table_entry *s390_search_extables(unsigned long addr)
231 {
232 	const struct exception_table_entry *fixup;
233 
234 	fixup = search_extable(__start_amode31_ex_table,
235 			       __stop_amode31_ex_table - __start_amode31_ex_table,
236 			       addr);
237 	if (!fixup)
238 		fixup = search_exception_tables(addr);
239 	return fixup;
240 }
241 
242 static noinline void do_no_context(struct pt_regs *regs)
243 {
244 	const struct exception_table_entry *fixup;
245 
246 	/* Are we prepared to handle this kernel fault?  */
247 	fixup = s390_search_extables(regs->psw.addr);
248 	if (fixup && ex_handle(fixup, regs))
249 		return;
250 
251 	/*
252 	 * Oops. The kernel tried to access some bad page. We'll have to
253 	 * terminate things with extreme prejudice.
254 	 */
255 	if (get_fault_type(regs) == KERNEL_FAULT)
256 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
257 		       " in virtual kernel address space\n");
258 	else
259 		printk(KERN_ALERT "Unable to handle kernel paging request"
260 		       " in virtual user address space\n");
261 	dump_fault_info(regs);
262 	die(regs, "Oops");
263 }
264 
265 static noinline void do_low_address(struct pt_regs *regs)
266 {
267 	/* Low-address protection hit in kernel mode means
268 	   NULL pointer write access in kernel mode.  */
269 	if (regs->psw.mask & PSW_MASK_PSTATE) {
270 		/* Low-address protection hit in user mode 'cannot happen'. */
271 		die (regs, "Low-address protection");
272 	}
273 
274 	do_no_context(regs);
275 }
276 
277 static noinline void do_sigbus(struct pt_regs *regs)
278 {
279 	/*
280 	 * Send a sigbus, regardless of whether we were in kernel
281 	 * or user mode.
282 	 */
283 	force_sig_fault(SIGBUS, BUS_ADRERR,
284 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
285 }
286 
287 static noinline void do_fault_error(struct pt_regs *regs, int access,
288 					vm_fault_t fault)
289 {
290 	int si_code;
291 
292 	switch (fault) {
293 	case VM_FAULT_BADACCESS:
294 	case VM_FAULT_BADMAP:
295 		/* Bad memory access. Check if it is kernel or user space. */
296 		if (user_mode(regs)) {
297 			/* User mode accesses just cause a SIGSEGV */
298 			si_code = (fault == VM_FAULT_BADMAP) ?
299 				SEGV_MAPERR : SEGV_ACCERR;
300 			do_sigsegv(regs, si_code);
301 			break;
302 		}
303 		fallthrough;
304 	case VM_FAULT_BADCONTEXT:
305 	case VM_FAULT_PFAULT:
306 		do_no_context(regs);
307 		break;
308 	case VM_FAULT_SIGNAL:
309 		if (!user_mode(regs))
310 			do_no_context(regs);
311 		break;
312 	default: /* fault & VM_FAULT_ERROR */
313 		if (fault & VM_FAULT_OOM) {
314 			if (!user_mode(regs))
315 				do_no_context(regs);
316 			else
317 				pagefault_out_of_memory();
318 		} else if (fault & VM_FAULT_SIGSEGV) {
319 			/* Kernel mode? Handle exceptions or die */
320 			if (!user_mode(regs))
321 				do_no_context(regs);
322 			else
323 				do_sigsegv(regs, SEGV_MAPERR);
324 		} else if (fault & VM_FAULT_SIGBUS) {
325 			/* Kernel mode? Handle exceptions or die */
326 			if (!user_mode(regs))
327 				do_no_context(regs);
328 			else
329 				do_sigbus(regs);
330 		} else
331 			BUG();
332 		break;
333 	}
334 }
335 
336 /*
337  * This routine handles page faults.  It determines the address,
338  * and the problem, and then passes it off to one of the appropriate
339  * routines.
340  *
341  * interruption code (int_code):
342  *   04       Protection           ->  Write-Protection  (suppression)
343  *   10       Segment translation  ->  Not present       (nullification)
344  *   11       Page translation     ->  Not present       (nullification)
345  *   3b       Region third trans.  ->  Not present       (nullification)
346  */
347 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
348 {
349 	struct gmap *gmap;
350 	struct task_struct *tsk;
351 	struct mm_struct *mm;
352 	struct vm_area_struct *vma;
353 	enum fault_type type;
354 	unsigned long trans_exc_code;
355 	unsigned long address;
356 	unsigned int flags;
357 	vm_fault_t fault;
358 	bool is_write;
359 
360 	tsk = current;
361 	/*
362 	 * The instruction that caused the program check has
363 	 * been nullified. Don't signal single step via SIGTRAP.
364 	 */
365 	clear_thread_flag(TIF_PER_TRAP);
366 
367 	if (kprobe_page_fault(regs, 14))
368 		return 0;
369 
370 	mm = tsk->mm;
371 	trans_exc_code = regs->int_parm_long;
372 	address = trans_exc_code & __FAIL_ADDR_MASK;
373 	is_write = (trans_exc_code & store_indication) == 0x400;
374 
375 	/*
376 	 * Verify that the fault happened in user space, that
377 	 * we are not in an interrupt and that there is a
378 	 * user context.
379 	 */
380 	fault = VM_FAULT_BADCONTEXT;
381 	type = get_fault_type(regs);
382 	switch (type) {
383 	case KERNEL_FAULT:
384 		if (kfence_handle_page_fault(address, is_write, regs))
385 			return 0;
386 		goto out;
387 	case USER_FAULT:
388 	case GMAP_FAULT:
389 		if (faulthandler_disabled() || !mm)
390 			goto out;
391 		break;
392 	}
393 
394 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
395 	flags = FAULT_FLAG_DEFAULT;
396 	if (user_mode(regs))
397 		flags |= FAULT_FLAG_USER;
398 	if (access == VM_WRITE || is_write)
399 		flags |= FAULT_FLAG_WRITE;
400 	mmap_read_lock(mm);
401 
402 	gmap = NULL;
403 	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
404 		gmap = (struct gmap *) S390_lowcore.gmap;
405 		current->thread.gmap_addr = address;
406 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
407 		current->thread.gmap_int_code = regs->int_code & 0xffff;
408 		address = __gmap_translate(gmap, address);
409 		if (address == -EFAULT) {
410 			fault = VM_FAULT_BADMAP;
411 			goto out_up;
412 		}
413 		if (gmap->pfault_enabled)
414 			flags |= FAULT_FLAG_RETRY_NOWAIT;
415 	}
416 
417 retry:
418 	fault = VM_FAULT_BADMAP;
419 	vma = find_vma(mm, address);
420 	if (!vma)
421 		goto out_up;
422 
423 	if (unlikely(vma->vm_start > address)) {
424 		if (!(vma->vm_flags & VM_GROWSDOWN))
425 			goto out_up;
426 		if (expand_stack(vma, address))
427 			goto out_up;
428 	}
429 
430 	/*
431 	 * Ok, we have a good vm_area for this memory access, so
432 	 * we can handle it..
433 	 */
434 	fault = VM_FAULT_BADACCESS;
435 	if (unlikely(!(vma->vm_flags & access)))
436 		goto out_up;
437 
438 	if (is_vm_hugetlb_page(vma))
439 		address &= HPAGE_MASK;
440 	/*
441 	 * If for any reason at all we couldn't handle the fault,
442 	 * make sure we exit gracefully rather than endlessly redo
443 	 * the fault.
444 	 */
445 	fault = handle_mm_fault(vma, address, flags, regs);
446 	if (fault_signal_pending(fault, regs)) {
447 		fault = VM_FAULT_SIGNAL;
448 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
449 			goto out_up;
450 		goto out;
451 	}
452 	if (unlikely(fault & VM_FAULT_ERROR))
453 		goto out_up;
454 
455 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
456 		if (fault & VM_FAULT_RETRY) {
457 			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
458 			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
459 				/* FAULT_FLAG_RETRY_NOWAIT has been set,
460 				 * mmap_lock has not been released */
461 				current->thread.gmap_pfault = 1;
462 				fault = VM_FAULT_PFAULT;
463 				goto out_up;
464 			}
465 			flags &= ~FAULT_FLAG_RETRY_NOWAIT;
466 			flags |= FAULT_FLAG_TRIED;
467 			mmap_read_lock(mm);
468 			goto retry;
469 		}
470 	}
471 	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
472 		address =  __gmap_link(gmap, current->thread.gmap_addr,
473 				       address);
474 		if (address == -EFAULT) {
475 			fault = VM_FAULT_BADMAP;
476 			goto out_up;
477 		}
478 		if (address == -ENOMEM) {
479 			fault = VM_FAULT_OOM;
480 			goto out_up;
481 		}
482 	}
483 	fault = 0;
484 out_up:
485 	mmap_read_unlock(mm);
486 out:
487 	return fault;
488 }
489 
490 void do_protection_exception(struct pt_regs *regs)
491 {
492 	unsigned long trans_exc_code;
493 	int access;
494 	vm_fault_t fault;
495 
496 	trans_exc_code = regs->int_parm_long;
497 	/*
498 	 * Protection exceptions are suppressing, decrement psw address.
499 	 * The exception to this rule are aborted transactions, for these
500 	 * the PSW already points to the correct location.
501 	 */
502 	if (!(regs->int_code & 0x200))
503 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
504 	/*
505 	 * Check for low-address protection.  This needs to be treated
506 	 * as a special case because the translation exception code
507 	 * field is not guaranteed to contain valid data in this case.
508 	 */
509 	if (unlikely(!(trans_exc_code & 4))) {
510 		do_low_address(regs);
511 		return;
512 	}
513 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
514 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
515 					(regs->psw.addr & PAGE_MASK);
516 		access = VM_EXEC;
517 		fault = VM_FAULT_BADACCESS;
518 	} else {
519 		access = VM_WRITE;
520 		fault = do_exception(regs, access);
521 	}
522 	if (unlikely(fault))
523 		do_fault_error(regs, access, fault);
524 }
525 NOKPROBE_SYMBOL(do_protection_exception);
526 
527 void do_dat_exception(struct pt_regs *regs)
528 {
529 	int access;
530 	vm_fault_t fault;
531 
532 	access = VM_ACCESS_FLAGS;
533 	fault = do_exception(regs, access);
534 	if (unlikely(fault))
535 		do_fault_error(regs, access, fault);
536 }
537 NOKPROBE_SYMBOL(do_dat_exception);
538 
539 #ifdef CONFIG_PFAULT
540 /*
541  * 'pfault' pseudo page faults routines.
542  */
543 static int pfault_disable;
544 
545 static int __init nopfault(char *str)
546 {
547 	pfault_disable = 1;
548 	return 1;
549 }
550 
551 __setup("nopfault", nopfault);
552 
553 struct pfault_refbk {
554 	u16 refdiagc;
555 	u16 reffcode;
556 	u16 refdwlen;
557 	u16 refversn;
558 	u64 refgaddr;
559 	u64 refselmk;
560 	u64 refcmpmk;
561 	u64 reserved;
562 } __attribute__ ((packed, aligned(8)));
563 
564 static struct pfault_refbk pfault_init_refbk = {
565 	.refdiagc = 0x258,
566 	.reffcode = 0,
567 	.refdwlen = 5,
568 	.refversn = 2,
569 	.refgaddr = __LC_LPP,
570 	.refselmk = 1ULL << 48,
571 	.refcmpmk = 1ULL << 48,
572 	.reserved = __PF_RES_FIELD
573 };
574 
575 int pfault_init(void)
576 {
577         int rc;
578 
579 	if (pfault_disable)
580 		return -1;
581 	diag_stat_inc(DIAG_STAT_X258);
582 	asm volatile(
583 		"	diag	%1,%0,0x258\n"
584 		"0:	j	2f\n"
585 		"1:	la	%0,8\n"
586 		"2:\n"
587 		EX_TABLE(0b,1b)
588 		: "=d" (rc)
589 		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
590         return rc;
591 }
592 
593 static struct pfault_refbk pfault_fini_refbk = {
594 	.refdiagc = 0x258,
595 	.reffcode = 1,
596 	.refdwlen = 5,
597 	.refversn = 2,
598 };
599 
600 void pfault_fini(void)
601 {
602 
603 	if (pfault_disable)
604 		return;
605 	diag_stat_inc(DIAG_STAT_X258);
606 	asm volatile(
607 		"	diag	%0,0,0x258\n"
608 		"0:	nopr	%%r7\n"
609 		EX_TABLE(0b,0b)
610 		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
611 }
612 
613 static DEFINE_SPINLOCK(pfault_lock);
614 static LIST_HEAD(pfault_list);
615 
616 #define PF_COMPLETE	0x0080
617 
618 /*
619  * The mechanism of our pfault code: if Linux is running as guest, runs a user
620  * space process and the user space process accesses a page that the host has
621  * paged out we get a pfault interrupt.
622  *
623  * This allows us, within the guest, to schedule a different process. Without
624  * this mechanism the host would have to suspend the whole virtual cpu until
625  * the page has been paged in.
626  *
627  * So when we get such an interrupt then we set the state of the current task
628  * to uninterruptible and also set the need_resched flag. Both happens within
629  * interrupt context(!). If we later on want to return to user space we
630  * recognize the need_resched flag and then call schedule().  It's not very
631  * obvious how this works...
632  *
633  * Of course we have a lot of additional fun with the completion interrupt (->
634  * host signals that a page of a process has been paged in and the process can
635  * continue to run). This interrupt can arrive on any cpu and, since we have
636  * virtual cpus, actually appear before the interrupt that signals that a page
637  * is missing.
638  */
639 static void pfault_interrupt(struct ext_code ext_code,
640 			     unsigned int param32, unsigned long param64)
641 {
642 	struct task_struct *tsk;
643 	__u16 subcode;
644 	pid_t pid;
645 
646 	/*
647 	 * Get the external interruption subcode & pfault initial/completion
648 	 * signal bit. VM stores this in the 'cpu address' field associated
649 	 * with the external interrupt.
650 	 */
651 	subcode = ext_code.subcode;
652 	if ((subcode & 0xff00) != __SUBCODE_MASK)
653 		return;
654 	inc_irq_stat(IRQEXT_PFL);
655 	/* Get the token (= pid of the affected task). */
656 	pid = param64 & LPP_PID_MASK;
657 	rcu_read_lock();
658 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
659 	if (tsk)
660 		get_task_struct(tsk);
661 	rcu_read_unlock();
662 	if (!tsk)
663 		return;
664 	spin_lock(&pfault_lock);
665 	if (subcode & PF_COMPLETE) {
666 		/* signal bit is set -> a page has been swapped in by VM */
667 		if (tsk->thread.pfault_wait == 1) {
668 			/* Initial interrupt was faster than the completion
669 			 * interrupt. pfault_wait is valid. Set pfault_wait
670 			 * back to zero and wake up the process. This can
671 			 * safely be done because the task is still sleeping
672 			 * and can't produce new pfaults. */
673 			tsk->thread.pfault_wait = 0;
674 			list_del(&tsk->thread.list);
675 			wake_up_process(tsk);
676 			put_task_struct(tsk);
677 		} else {
678 			/* Completion interrupt was faster than initial
679 			 * interrupt. Set pfault_wait to -1 so the initial
680 			 * interrupt doesn't put the task to sleep.
681 			 * If the task is not running, ignore the completion
682 			 * interrupt since it must be a leftover of a PFAULT
683 			 * CANCEL operation which didn't remove all pending
684 			 * completion interrupts. */
685 			if (task_is_running(tsk))
686 				tsk->thread.pfault_wait = -1;
687 		}
688 	} else {
689 		/* signal bit not set -> a real page is missing. */
690 		if (WARN_ON_ONCE(tsk != current))
691 			goto out;
692 		if (tsk->thread.pfault_wait == 1) {
693 			/* Already on the list with a reference: put to sleep */
694 			goto block;
695 		} else if (tsk->thread.pfault_wait == -1) {
696 			/* Completion interrupt was faster than the initial
697 			 * interrupt (pfault_wait == -1). Set pfault_wait
698 			 * back to zero and exit. */
699 			tsk->thread.pfault_wait = 0;
700 		} else {
701 			/* Initial interrupt arrived before completion
702 			 * interrupt. Let the task sleep.
703 			 * An extra task reference is needed since a different
704 			 * cpu may set the task state to TASK_RUNNING again
705 			 * before the scheduler is reached. */
706 			get_task_struct(tsk);
707 			tsk->thread.pfault_wait = 1;
708 			list_add(&tsk->thread.list, &pfault_list);
709 block:
710 			/* Since this must be a userspace fault, there
711 			 * is no kernel task state to trample. Rely on the
712 			 * return to userspace schedule() to block. */
713 			__set_current_state(TASK_UNINTERRUPTIBLE);
714 			set_tsk_need_resched(tsk);
715 			set_preempt_need_resched();
716 		}
717 	}
718 out:
719 	spin_unlock(&pfault_lock);
720 	put_task_struct(tsk);
721 }
722 
723 static int pfault_cpu_dead(unsigned int cpu)
724 {
725 	struct thread_struct *thread, *next;
726 	struct task_struct *tsk;
727 
728 	spin_lock_irq(&pfault_lock);
729 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
730 		thread->pfault_wait = 0;
731 		list_del(&thread->list);
732 		tsk = container_of(thread, struct task_struct, thread);
733 		wake_up_process(tsk);
734 		put_task_struct(tsk);
735 	}
736 	spin_unlock_irq(&pfault_lock);
737 	return 0;
738 }
739 
740 static int __init pfault_irq_init(void)
741 {
742 	int rc;
743 
744 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
745 	if (rc)
746 		goto out_extint;
747 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
748 	if (rc)
749 		goto out_pfault;
750 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
751 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
752 				  NULL, pfault_cpu_dead);
753 	return 0;
754 
755 out_pfault:
756 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
757 out_extint:
758 	pfault_disable = 1;
759 	return rc;
760 }
761 early_initcall(pfault_irq_init);
762 
763 #endif /* CONFIG_PFAULT */
764 
765 #if IS_ENABLED(CONFIG_PGSTE)
766 
767 void do_secure_storage_access(struct pt_regs *regs)
768 {
769 	unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
770 	struct vm_area_struct *vma;
771 	struct mm_struct *mm;
772 	struct page *page;
773 	int rc;
774 
775 	/*
776 	 * bit 61 tells us if the address is valid, if it's not we
777 	 * have a major problem and should stop the kernel or send a
778 	 * SIGSEGV to the process. Unfortunately bit 61 is not
779 	 * reliable without the misc UV feature so we need to check
780 	 * for that as well.
781 	 */
782 	if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
783 	    !test_bit_inv(61, &regs->int_parm_long)) {
784 		/*
785 		 * When this happens, userspace did something that it
786 		 * was not supposed to do, e.g. branching into secure
787 		 * memory. Trigger a segmentation fault.
788 		 */
789 		if (user_mode(regs)) {
790 			send_sig(SIGSEGV, current, 0);
791 			return;
792 		}
793 
794 		/*
795 		 * The kernel should never run into this case and we
796 		 * have no way out of this situation.
797 		 */
798 		panic("Unexpected PGM 0x3d with TEID bit 61=0");
799 	}
800 
801 	switch (get_fault_type(regs)) {
802 	case USER_FAULT:
803 		mm = current->mm;
804 		mmap_read_lock(mm);
805 		vma = find_vma(mm, addr);
806 		if (!vma) {
807 			mmap_read_unlock(mm);
808 			do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
809 			break;
810 		}
811 		page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
812 		if (IS_ERR_OR_NULL(page)) {
813 			mmap_read_unlock(mm);
814 			break;
815 		}
816 		if (arch_make_page_accessible(page))
817 			send_sig(SIGSEGV, current, 0);
818 		put_page(page);
819 		mmap_read_unlock(mm);
820 		break;
821 	case KERNEL_FAULT:
822 		page = phys_to_page(addr);
823 		if (unlikely(!try_get_page(page)))
824 			break;
825 		rc = arch_make_page_accessible(page);
826 		put_page(page);
827 		if (rc)
828 			BUG();
829 		break;
830 	case GMAP_FAULT:
831 	default:
832 		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
833 		WARN_ON_ONCE(1);
834 	}
835 }
836 NOKPROBE_SYMBOL(do_secure_storage_access);
837 
838 void do_non_secure_storage_access(struct pt_regs *regs)
839 {
840 	unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
841 	struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
842 
843 	if (get_fault_type(regs) != GMAP_FAULT) {
844 		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
845 		WARN_ON_ONCE(1);
846 		return;
847 	}
848 
849 	if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
850 		send_sig(SIGSEGV, current, 0);
851 }
852 NOKPROBE_SYMBOL(do_non_secure_storage_access);
853 
854 void do_secure_storage_violation(struct pt_regs *regs)
855 {
856 	/*
857 	 * Either KVM messed up the secure guest mapping or the same
858 	 * page is mapped into multiple secure guests.
859 	 *
860 	 * This exception is only triggered when a guest 2 is running
861 	 * and can therefore never occur in kernel context.
862 	 */
863 	printk_ratelimited(KERN_WARNING
864 			   "Secure storage violation in task: %s, pid %d\n",
865 			   current->comm, current->pid);
866 	send_sig(SIGSEGV, current, 0);
867 }
868 
869 #endif /* CONFIG_PGSTE */
870