xref: /linux/arch/s390/mm/fault.c (revision 0e685c3e7158d35626d6d76b9f859eae806d87fa)
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-extable.h>
36 #include <asm/asm-offsets.h>
37 #include <asm/diag.h>
38 #include <asm/gmap.h>
39 #include <asm/irq.h>
40 #include <asm/mmu_context.h>
41 #include <asm/facility.h>
42 #include <asm/uv.h>
43 #include "../kernel/entry.h"
44 
45 #define __FAIL_ADDR_MASK -4096L
46 #define __SUBCODE_MASK 0x0600
47 #define __PF_RES_FIELD 0x8000000000000000ULL
48 
49 /*
50  * Allocate private vm_fault_reason from top.  Please make sure it won't
51  * collide with vm_fault_reason.
52  */
53 #define VM_FAULT_BADCONTEXT	((__force vm_fault_t)0x80000000)
54 #define VM_FAULT_BADMAP		((__force vm_fault_t)0x40000000)
55 #define VM_FAULT_BADACCESS	((__force vm_fault_t)0x20000000)
56 #define VM_FAULT_SIGNAL		((__force vm_fault_t)0x10000000)
57 #define VM_FAULT_PFAULT		((__force vm_fault_t)0x8000000)
58 
59 enum fault_type {
60 	KERNEL_FAULT,
61 	USER_FAULT,
62 	GMAP_FAULT,
63 };
64 
65 static unsigned long store_indication __read_mostly;
66 
67 static int __init fault_init(void)
68 {
69 	if (test_facility(75))
70 		store_indication = 0xc00;
71 	return 0;
72 }
73 early_initcall(fault_init);
74 
75 /*
76  * Find out which address space caused the exception.
77  */
78 static enum fault_type get_fault_type(struct pt_regs *regs)
79 {
80 	unsigned long trans_exc_code;
81 
82 	trans_exc_code = regs->int_parm_long & 3;
83 	if (likely(trans_exc_code == 0)) {
84 		/* primary space exception */
85 		if (user_mode(regs))
86 			return USER_FAULT;
87 		if (!IS_ENABLED(CONFIG_PGSTE))
88 			return KERNEL_FAULT;
89 		if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
90 			return GMAP_FAULT;
91 		return KERNEL_FAULT;
92 	}
93 	if (trans_exc_code == 2)
94 		return USER_FAULT;
95 	if (trans_exc_code == 1) {
96 		/* access register mode, not used in the kernel */
97 		return USER_FAULT;
98 	}
99 	/* home space exception -> access via kernel ASCE */
100 	return KERNEL_FAULT;
101 }
102 
103 static unsigned long get_fault_address(struct pt_regs *regs)
104 {
105 	unsigned long trans_exc_code = regs->int_parm_long;
106 
107 	return trans_exc_code & __FAIL_ADDR_MASK;
108 }
109 
110 static bool fault_is_write(struct pt_regs *regs)
111 {
112 	unsigned long trans_exc_code = regs->int_parm_long;
113 
114 	return (trans_exc_code & store_indication) == 0x400;
115 }
116 
117 static int bad_address(void *p)
118 {
119 	unsigned long dummy;
120 
121 	return get_kernel_nofault(dummy, (unsigned long *)p);
122 }
123 
124 static void dump_pagetable(unsigned long asce, unsigned long address)
125 {
126 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
127 
128 	pr_alert("AS:%016lx ", asce);
129 	switch (asce & _ASCE_TYPE_MASK) {
130 	case _ASCE_TYPE_REGION1:
131 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
132 		if (bad_address(table))
133 			goto bad;
134 		pr_cont("R1:%016lx ", *table);
135 		if (*table & _REGION_ENTRY_INVALID)
136 			goto out;
137 		table = __va(*table & _REGION_ENTRY_ORIGIN);
138 		fallthrough;
139 	case _ASCE_TYPE_REGION2:
140 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
141 		if (bad_address(table))
142 			goto bad;
143 		pr_cont("R2:%016lx ", *table);
144 		if (*table & _REGION_ENTRY_INVALID)
145 			goto out;
146 		table = __va(*table & _REGION_ENTRY_ORIGIN);
147 		fallthrough;
148 	case _ASCE_TYPE_REGION3:
149 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
150 		if (bad_address(table))
151 			goto bad;
152 		pr_cont("R3:%016lx ", *table);
153 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
154 			goto out;
155 		table = __va(*table & _REGION_ENTRY_ORIGIN);
156 		fallthrough;
157 	case _ASCE_TYPE_SEGMENT:
158 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
159 		if (bad_address(table))
160 			goto bad;
161 		pr_cont("S:%016lx ", *table);
162 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
163 			goto out;
164 		table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
165 	}
166 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
167 	if (bad_address(table))
168 		goto bad;
169 	pr_cont("P:%016lx ", *table);
170 out:
171 	pr_cont("\n");
172 	return;
173 bad:
174 	pr_cont("BAD\n");
175 }
176 
177 static void dump_fault_info(struct pt_regs *regs)
178 {
179 	unsigned long asce;
180 
181 	pr_alert("Failing address: %016lx TEID: %016lx\n",
182 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
183 	pr_alert("Fault in ");
184 	switch (regs->int_parm_long & 3) {
185 	case 3:
186 		pr_cont("home space ");
187 		break;
188 	case 2:
189 		pr_cont("secondary space ");
190 		break;
191 	case 1:
192 		pr_cont("access register ");
193 		break;
194 	case 0:
195 		pr_cont("primary space ");
196 		break;
197 	}
198 	pr_cont("mode while using ");
199 	switch (get_fault_type(regs)) {
200 	case USER_FAULT:
201 		asce = S390_lowcore.user_asce;
202 		pr_cont("user ");
203 		break;
204 	case GMAP_FAULT:
205 		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
206 		pr_cont("gmap ");
207 		break;
208 	case KERNEL_FAULT:
209 		asce = S390_lowcore.kernel_asce;
210 		pr_cont("kernel ");
211 		break;
212 	default:
213 		unreachable();
214 	}
215 	pr_cont("ASCE.\n");
216 	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
217 }
218 
219 int show_unhandled_signals = 1;
220 
221 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
222 {
223 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
224 		return;
225 	if (!unhandled_signal(current, signr))
226 		return;
227 	if (!printk_ratelimit())
228 		return;
229 	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
230 	       regs->int_code & 0xffff, regs->int_code >> 17);
231 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
232 	printk(KERN_CONT "\n");
233 	if (is_mm_fault)
234 		dump_fault_info(regs);
235 	show_regs(regs);
236 }
237 
238 /*
239  * Send SIGSEGV to task.  This is an external routine
240  * to keep the stack usage of do_page_fault small.
241  */
242 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
243 {
244 	report_user_fault(regs, SIGSEGV, 1);
245 	force_sig_fault(SIGSEGV, si_code,
246 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
247 }
248 
249 static noinline void do_no_context(struct pt_regs *regs, vm_fault_t fault)
250 {
251 	enum fault_type fault_type;
252 	unsigned long address;
253 	bool is_write;
254 
255 	if (fixup_exception(regs))
256 		return;
257 	fault_type = get_fault_type(regs);
258 	if ((fault_type == KERNEL_FAULT) && (fault == VM_FAULT_BADCONTEXT)) {
259 		address = get_fault_address(regs);
260 		is_write = fault_is_write(regs);
261 		if (kfence_handle_page_fault(address, is_write, regs))
262 			return;
263 	}
264 	/*
265 	 * Oops. The kernel tried to access some bad page. We'll have to
266 	 * terminate things with extreme prejudice.
267 	 */
268 	if (fault_type == KERNEL_FAULT)
269 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
270 		       " in virtual kernel address space\n");
271 	else
272 		printk(KERN_ALERT "Unable to handle kernel paging request"
273 		       " in virtual user address space\n");
274 	dump_fault_info(regs);
275 	die(regs, "Oops");
276 }
277 
278 static noinline void do_low_address(struct pt_regs *regs)
279 {
280 	/* Low-address protection hit in kernel mode means
281 	   NULL pointer write access in kernel mode.  */
282 	if (regs->psw.mask & PSW_MASK_PSTATE) {
283 		/* Low-address protection hit in user mode 'cannot happen'. */
284 		die (regs, "Low-address protection");
285 	}
286 
287 	do_no_context(regs, VM_FAULT_BADACCESS);
288 }
289 
290 static noinline void do_sigbus(struct pt_regs *regs)
291 {
292 	/*
293 	 * Send a sigbus, regardless of whether we were in kernel
294 	 * or user mode.
295 	 */
296 	force_sig_fault(SIGBUS, BUS_ADRERR,
297 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
298 }
299 
300 static noinline void do_fault_error(struct pt_regs *regs, vm_fault_t fault)
301 {
302 	int si_code;
303 
304 	switch (fault) {
305 	case VM_FAULT_BADACCESS:
306 	case VM_FAULT_BADMAP:
307 		/* Bad memory access. Check if it is kernel or user space. */
308 		if (user_mode(regs)) {
309 			/* User mode accesses just cause a SIGSEGV */
310 			si_code = (fault == VM_FAULT_BADMAP) ?
311 				SEGV_MAPERR : SEGV_ACCERR;
312 			do_sigsegv(regs, si_code);
313 			break;
314 		}
315 		fallthrough;
316 	case VM_FAULT_BADCONTEXT:
317 	case VM_FAULT_PFAULT:
318 		do_no_context(regs, fault);
319 		break;
320 	case VM_FAULT_SIGNAL:
321 		if (!user_mode(regs))
322 			do_no_context(regs, fault);
323 		break;
324 	default: /* fault & VM_FAULT_ERROR */
325 		if (fault & VM_FAULT_OOM) {
326 			if (!user_mode(regs))
327 				do_no_context(regs, fault);
328 			else
329 				pagefault_out_of_memory();
330 		} else if (fault & VM_FAULT_SIGSEGV) {
331 			/* Kernel mode? Handle exceptions or die */
332 			if (!user_mode(regs))
333 				do_no_context(regs, fault);
334 			else
335 				do_sigsegv(regs, SEGV_MAPERR);
336 		} else if (fault & VM_FAULT_SIGBUS) {
337 			/* Kernel mode? Handle exceptions or die */
338 			if (!user_mode(regs))
339 				do_no_context(regs, fault);
340 			else
341 				do_sigbus(regs);
342 		} else
343 			BUG();
344 		break;
345 	}
346 }
347 
348 /*
349  * This routine handles page faults.  It determines the address,
350  * and the problem, and then passes it off to one of the appropriate
351  * routines.
352  *
353  * interruption code (int_code):
354  *   04       Protection           ->  Write-Protection  (suppression)
355  *   10       Segment translation  ->  Not present       (nullification)
356  *   11       Page translation     ->  Not present       (nullification)
357  *   3b       Region third trans.  ->  Not present       (nullification)
358  */
359 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
360 {
361 	struct gmap *gmap;
362 	struct task_struct *tsk;
363 	struct mm_struct *mm;
364 	struct vm_area_struct *vma;
365 	enum fault_type type;
366 	unsigned long address;
367 	unsigned int flags;
368 	vm_fault_t fault;
369 	bool is_write;
370 
371 	tsk = current;
372 	/*
373 	 * The instruction that caused the program check has
374 	 * been nullified. Don't signal single step via SIGTRAP.
375 	 */
376 	clear_thread_flag(TIF_PER_TRAP);
377 
378 	if (kprobe_page_fault(regs, 14))
379 		return 0;
380 
381 	mm = tsk->mm;
382 	address = get_fault_address(regs);
383 	is_write = fault_is_write(regs);
384 
385 	/*
386 	 * Verify that the fault happened in user space, that
387 	 * we are not in an interrupt and that there is a
388 	 * user context.
389 	 */
390 	fault = VM_FAULT_BADCONTEXT;
391 	type = get_fault_type(regs);
392 	switch (type) {
393 	case KERNEL_FAULT:
394 		goto out;
395 	case USER_FAULT:
396 	case GMAP_FAULT:
397 		if (faulthandler_disabled() || !mm)
398 			goto out;
399 		break;
400 	}
401 
402 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
403 	flags = FAULT_FLAG_DEFAULT;
404 	if (user_mode(regs))
405 		flags |= FAULT_FLAG_USER;
406 	if (is_write)
407 		access = VM_WRITE;
408 	if (access == VM_WRITE)
409 		flags |= FAULT_FLAG_WRITE;
410 	mmap_read_lock(mm);
411 
412 	gmap = NULL;
413 	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
414 		gmap = (struct gmap *) S390_lowcore.gmap;
415 		current->thread.gmap_addr = address;
416 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
417 		current->thread.gmap_int_code = regs->int_code & 0xffff;
418 		address = __gmap_translate(gmap, address);
419 		if (address == -EFAULT) {
420 			fault = VM_FAULT_BADMAP;
421 			goto out_up;
422 		}
423 		if (gmap->pfault_enabled)
424 			flags |= FAULT_FLAG_RETRY_NOWAIT;
425 	}
426 
427 retry:
428 	fault = VM_FAULT_BADMAP;
429 	vma = find_vma(mm, address);
430 	if (!vma)
431 		goto out_up;
432 
433 	if (unlikely(vma->vm_start > address)) {
434 		if (!(vma->vm_flags & VM_GROWSDOWN))
435 			goto out_up;
436 		if (expand_stack(vma, address))
437 			goto out_up;
438 	}
439 
440 	/*
441 	 * Ok, we have a good vm_area for this memory access, so
442 	 * we can handle it..
443 	 */
444 	fault = VM_FAULT_BADACCESS;
445 	if (unlikely(!(vma->vm_flags & access)))
446 		goto out_up;
447 
448 	/*
449 	 * If for any reason at all we couldn't handle the fault,
450 	 * make sure we exit gracefully rather than endlessly redo
451 	 * the fault.
452 	 */
453 	fault = handle_mm_fault(vma, address, flags, regs);
454 	if (fault_signal_pending(fault, regs)) {
455 		fault = VM_FAULT_SIGNAL;
456 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
457 			goto out_up;
458 		goto out;
459 	}
460 
461 	/* The fault is fully completed (including releasing mmap lock) */
462 	if (fault & VM_FAULT_COMPLETED) {
463 		if (gmap) {
464 			mmap_read_lock(mm);
465 			goto out_gmap;
466 		}
467 		fault = 0;
468 		goto out;
469 	}
470 
471 	if (unlikely(fault & VM_FAULT_ERROR))
472 		goto out_up;
473 
474 	if (fault & VM_FAULT_RETRY) {
475 		if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
476 			(flags & FAULT_FLAG_RETRY_NOWAIT)) {
477 			/*
478 			 * FAULT_FLAG_RETRY_NOWAIT has been set, mmap_lock has
479 			 * not been released
480 			 */
481 			current->thread.gmap_pfault = 1;
482 			fault = VM_FAULT_PFAULT;
483 			goto out_up;
484 		}
485 		flags &= ~FAULT_FLAG_RETRY_NOWAIT;
486 		flags |= FAULT_FLAG_TRIED;
487 		mmap_read_lock(mm);
488 		goto retry;
489 	}
490 out_gmap:
491 	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
492 		address =  __gmap_link(gmap, current->thread.gmap_addr,
493 				       address);
494 		if (address == -EFAULT) {
495 			fault = VM_FAULT_BADMAP;
496 			goto out_up;
497 		}
498 		if (address == -ENOMEM) {
499 			fault = VM_FAULT_OOM;
500 			goto out_up;
501 		}
502 	}
503 	fault = 0;
504 out_up:
505 	mmap_read_unlock(mm);
506 out:
507 	return fault;
508 }
509 
510 void do_protection_exception(struct pt_regs *regs)
511 {
512 	unsigned long trans_exc_code;
513 	int access;
514 	vm_fault_t fault;
515 
516 	trans_exc_code = regs->int_parm_long;
517 	/*
518 	 * Protection exceptions are suppressing, decrement psw address.
519 	 * The exception to this rule are aborted transactions, for these
520 	 * the PSW already points to the correct location.
521 	 */
522 	if (!(regs->int_code & 0x200))
523 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
524 	/*
525 	 * Check for low-address protection.  This needs to be treated
526 	 * as a special case because the translation exception code
527 	 * field is not guaranteed to contain valid data in this case.
528 	 */
529 	if (unlikely(!(trans_exc_code & 4))) {
530 		do_low_address(regs);
531 		return;
532 	}
533 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
534 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
535 					(regs->psw.addr & PAGE_MASK);
536 		access = VM_EXEC;
537 		fault = VM_FAULT_BADACCESS;
538 	} else {
539 		access = VM_WRITE;
540 		fault = do_exception(regs, access);
541 	}
542 	if (unlikely(fault))
543 		do_fault_error(regs, fault);
544 }
545 NOKPROBE_SYMBOL(do_protection_exception);
546 
547 void do_dat_exception(struct pt_regs *regs)
548 {
549 	int access;
550 	vm_fault_t fault;
551 
552 	access = VM_ACCESS_FLAGS;
553 	fault = do_exception(regs, access);
554 	if (unlikely(fault))
555 		do_fault_error(regs, fault);
556 }
557 NOKPROBE_SYMBOL(do_dat_exception);
558 
559 #ifdef CONFIG_PFAULT
560 /*
561  * 'pfault' pseudo page faults routines.
562  */
563 static int pfault_disable;
564 
565 static int __init nopfault(char *str)
566 {
567 	pfault_disable = 1;
568 	return 1;
569 }
570 
571 __setup("nopfault", nopfault);
572 
573 struct pfault_refbk {
574 	u16 refdiagc;
575 	u16 reffcode;
576 	u16 refdwlen;
577 	u16 refversn;
578 	u64 refgaddr;
579 	u64 refselmk;
580 	u64 refcmpmk;
581 	u64 reserved;
582 } __attribute__ ((packed, aligned(8)));
583 
584 static struct pfault_refbk pfault_init_refbk = {
585 	.refdiagc = 0x258,
586 	.reffcode = 0,
587 	.refdwlen = 5,
588 	.refversn = 2,
589 	.refgaddr = __LC_LPP,
590 	.refselmk = 1ULL << 48,
591 	.refcmpmk = 1ULL << 48,
592 	.reserved = __PF_RES_FIELD
593 };
594 
595 int pfault_init(void)
596 {
597         int rc;
598 
599 	if (pfault_disable)
600 		return -1;
601 	diag_stat_inc(DIAG_STAT_X258);
602 	asm volatile(
603 		"	diag	%1,%0,0x258\n"
604 		"0:	j	2f\n"
605 		"1:	la	%0,8\n"
606 		"2:\n"
607 		EX_TABLE(0b,1b)
608 		: "=d" (rc)
609 		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
610         return rc;
611 }
612 
613 static struct pfault_refbk pfault_fini_refbk = {
614 	.refdiagc = 0x258,
615 	.reffcode = 1,
616 	.refdwlen = 5,
617 	.refversn = 2,
618 };
619 
620 void pfault_fini(void)
621 {
622 
623 	if (pfault_disable)
624 		return;
625 	diag_stat_inc(DIAG_STAT_X258);
626 	asm volatile(
627 		"	diag	%0,0,0x258\n"
628 		"0:	nopr	%%r7\n"
629 		EX_TABLE(0b,0b)
630 		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
631 }
632 
633 static DEFINE_SPINLOCK(pfault_lock);
634 static LIST_HEAD(pfault_list);
635 
636 #define PF_COMPLETE	0x0080
637 
638 /*
639  * The mechanism of our pfault code: if Linux is running as guest, runs a user
640  * space process and the user space process accesses a page that the host has
641  * paged out we get a pfault interrupt.
642  *
643  * This allows us, within the guest, to schedule a different process. Without
644  * this mechanism the host would have to suspend the whole virtual cpu until
645  * the page has been paged in.
646  *
647  * So when we get such an interrupt then we set the state of the current task
648  * to uninterruptible and also set the need_resched flag. Both happens within
649  * interrupt context(!). If we later on want to return to user space we
650  * recognize the need_resched flag and then call schedule().  It's not very
651  * obvious how this works...
652  *
653  * Of course we have a lot of additional fun with the completion interrupt (->
654  * host signals that a page of a process has been paged in and the process can
655  * continue to run). This interrupt can arrive on any cpu and, since we have
656  * virtual cpus, actually appear before the interrupt that signals that a page
657  * is missing.
658  */
659 static void pfault_interrupt(struct ext_code ext_code,
660 			     unsigned int param32, unsigned long param64)
661 {
662 	struct task_struct *tsk;
663 	__u16 subcode;
664 	pid_t pid;
665 
666 	/*
667 	 * Get the external interruption subcode & pfault initial/completion
668 	 * signal bit. VM stores this in the 'cpu address' field associated
669 	 * with the external interrupt.
670 	 */
671 	subcode = ext_code.subcode;
672 	if ((subcode & 0xff00) != __SUBCODE_MASK)
673 		return;
674 	inc_irq_stat(IRQEXT_PFL);
675 	/* Get the token (= pid of the affected task). */
676 	pid = param64 & LPP_PID_MASK;
677 	rcu_read_lock();
678 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
679 	if (tsk)
680 		get_task_struct(tsk);
681 	rcu_read_unlock();
682 	if (!tsk)
683 		return;
684 	spin_lock(&pfault_lock);
685 	if (subcode & PF_COMPLETE) {
686 		/* signal bit is set -> a page has been swapped in by VM */
687 		if (tsk->thread.pfault_wait == 1) {
688 			/* Initial interrupt was faster than the completion
689 			 * interrupt. pfault_wait is valid. Set pfault_wait
690 			 * back to zero and wake up the process. This can
691 			 * safely be done because the task is still sleeping
692 			 * and can't produce new pfaults. */
693 			tsk->thread.pfault_wait = 0;
694 			list_del(&tsk->thread.list);
695 			wake_up_process(tsk);
696 			put_task_struct(tsk);
697 		} else {
698 			/* Completion interrupt was faster than initial
699 			 * interrupt. Set pfault_wait to -1 so the initial
700 			 * interrupt doesn't put the task to sleep.
701 			 * If the task is not running, ignore the completion
702 			 * interrupt since it must be a leftover of a PFAULT
703 			 * CANCEL operation which didn't remove all pending
704 			 * completion interrupts. */
705 			if (task_is_running(tsk))
706 				tsk->thread.pfault_wait = -1;
707 		}
708 	} else {
709 		/* signal bit not set -> a real page is missing. */
710 		if (WARN_ON_ONCE(tsk != current))
711 			goto out;
712 		if (tsk->thread.pfault_wait == 1) {
713 			/* Already on the list with a reference: put to sleep */
714 			goto block;
715 		} else if (tsk->thread.pfault_wait == -1) {
716 			/* Completion interrupt was faster than the initial
717 			 * interrupt (pfault_wait == -1). Set pfault_wait
718 			 * back to zero and exit. */
719 			tsk->thread.pfault_wait = 0;
720 		} else {
721 			/* Initial interrupt arrived before completion
722 			 * interrupt. Let the task sleep.
723 			 * An extra task reference is needed since a different
724 			 * cpu may set the task state to TASK_RUNNING again
725 			 * before the scheduler is reached. */
726 			get_task_struct(tsk);
727 			tsk->thread.pfault_wait = 1;
728 			list_add(&tsk->thread.list, &pfault_list);
729 block:
730 			/* Since this must be a userspace fault, there
731 			 * is no kernel task state to trample. Rely on the
732 			 * return to userspace schedule() to block. */
733 			__set_current_state(TASK_UNINTERRUPTIBLE);
734 			set_tsk_need_resched(tsk);
735 			set_preempt_need_resched();
736 		}
737 	}
738 out:
739 	spin_unlock(&pfault_lock);
740 	put_task_struct(tsk);
741 }
742 
743 static int pfault_cpu_dead(unsigned int cpu)
744 {
745 	struct thread_struct *thread, *next;
746 	struct task_struct *tsk;
747 
748 	spin_lock_irq(&pfault_lock);
749 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
750 		thread->pfault_wait = 0;
751 		list_del(&thread->list);
752 		tsk = container_of(thread, struct task_struct, thread);
753 		wake_up_process(tsk);
754 		put_task_struct(tsk);
755 	}
756 	spin_unlock_irq(&pfault_lock);
757 	return 0;
758 }
759 
760 static int __init pfault_irq_init(void)
761 {
762 	int rc;
763 
764 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
765 	if (rc)
766 		goto out_extint;
767 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
768 	if (rc)
769 		goto out_pfault;
770 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
771 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
772 				  NULL, pfault_cpu_dead);
773 	return 0;
774 
775 out_pfault:
776 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
777 out_extint:
778 	pfault_disable = 1;
779 	return rc;
780 }
781 early_initcall(pfault_irq_init);
782 
783 #endif /* CONFIG_PFAULT */
784 
785 #if IS_ENABLED(CONFIG_PGSTE)
786 
787 void do_secure_storage_access(struct pt_regs *regs)
788 {
789 	unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
790 	struct vm_area_struct *vma;
791 	struct mm_struct *mm;
792 	struct page *page;
793 	struct gmap *gmap;
794 	int rc;
795 
796 	/*
797 	 * bit 61 tells us if the address is valid, if it's not we
798 	 * have a major problem and should stop the kernel or send a
799 	 * SIGSEGV to the process. Unfortunately bit 61 is not
800 	 * reliable without the misc UV feature so we need to check
801 	 * for that as well.
802 	 */
803 	if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
804 	    !test_bit_inv(61, &regs->int_parm_long)) {
805 		/*
806 		 * When this happens, userspace did something that it
807 		 * was not supposed to do, e.g. branching into secure
808 		 * memory. Trigger a segmentation fault.
809 		 */
810 		if (user_mode(regs)) {
811 			send_sig(SIGSEGV, current, 0);
812 			return;
813 		}
814 
815 		/*
816 		 * The kernel should never run into this case and we
817 		 * have no way out of this situation.
818 		 */
819 		panic("Unexpected PGM 0x3d with TEID bit 61=0");
820 	}
821 
822 	switch (get_fault_type(regs)) {
823 	case GMAP_FAULT:
824 		mm = current->mm;
825 		gmap = (struct gmap *)S390_lowcore.gmap;
826 		mmap_read_lock(mm);
827 		addr = __gmap_translate(gmap, addr);
828 		mmap_read_unlock(mm);
829 		if (IS_ERR_VALUE(addr)) {
830 			do_fault_error(regs, VM_FAULT_BADMAP);
831 			break;
832 		}
833 		fallthrough;
834 	case USER_FAULT:
835 		mm = current->mm;
836 		mmap_read_lock(mm);
837 		vma = find_vma(mm, addr);
838 		if (!vma) {
839 			mmap_read_unlock(mm);
840 			do_fault_error(regs, VM_FAULT_BADMAP);
841 			break;
842 		}
843 		page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
844 		if (IS_ERR_OR_NULL(page)) {
845 			mmap_read_unlock(mm);
846 			break;
847 		}
848 		if (arch_make_page_accessible(page))
849 			send_sig(SIGSEGV, current, 0);
850 		put_page(page);
851 		mmap_read_unlock(mm);
852 		break;
853 	case KERNEL_FAULT:
854 		page = phys_to_page(addr);
855 		if (unlikely(!try_get_page(page)))
856 			break;
857 		rc = arch_make_page_accessible(page);
858 		put_page(page);
859 		if (rc)
860 			BUG();
861 		break;
862 	default:
863 		do_fault_error(regs, VM_FAULT_BADMAP);
864 		WARN_ON_ONCE(1);
865 	}
866 }
867 NOKPROBE_SYMBOL(do_secure_storage_access);
868 
869 void do_non_secure_storage_access(struct pt_regs *regs)
870 {
871 	unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
872 	struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
873 
874 	if (get_fault_type(regs) != GMAP_FAULT) {
875 		do_fault_error(regs, VM_FAULT_BADMAP);
876 		WARN_ON_ONCE(1);
877 		return;
878 	}
879 
880 	if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
881 		send_sig(SIGSEGV, current, 0);
882 }
883 NOKPROBE_SYMBOL(do_non_secure_storage_access);
884 
885 void do_secure_storage_violation(struct pt_regs *regs)
886 {
887 	unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
888 	struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
889 
890 	/*
891 	 * If the VM has been rebooted, its address space might still contain
892 	 * secure pages from the previous boot.
893 	 * Clear the page so it can be reused.
894 	 */
895 	if (!gmap_destroy_page(gmap, gaddr))
896 		return;
897 	/*
898 	 * Either KVM messed up the secure guest mapping or the same
899 	 * page is mapped into multiple secure guests.
900 	 *
901 	 * This exception is only triggered when a guest 2 is running
902 	 * and can therefore never occur in kernel context.
903 	 */
904 	printk_ratelimited(KERN_WARNING
905 			   "Secure storage violation in task: %s, pid %d\n",
906 			   current->comm, current->pid);
907 	send_sig(SIGSEGV, current, 0);
908 }
909 
910 #endif /* CONFIG_PGSTE */
911