xref: /openbmc/linux/arch/s390/mm/fault.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
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 <asm/asm-offsets.h>
35 #include <asm/diag.h>
36 #include <asm/pgtable.h>
37 #include <asm/gmap.h>
38 #include <asm/irq.h>
39 #include <asm/mmu_context.h>
40 #include <asm/facility.h>
41 #include "../kernel/entry.h"
42 
43 #define __FAIL_ADDR_MASK -4096L
44 #define __SUBCODE_MASK 0x0600
45 #define __PF_RES_FIELD 0x8000000000000000ULL
46 
47 #define VM_FAULT_BADCONTEXT	0x010000
48 #define VM_FAULT_BADMAP		0x020000
49 #define VM_FAULT_BADACCESS	0x040000
50 #define VM_FAULT_SIGNAL		0x080000
51 #define VM_FAULT_PFAULT		0x100000
52 
53 static unsigned long store_indication __read_mostly;
54 
55 static int __init fault_init(void)
56 {
57 	if (test_facility(75))
58 		store_indication = 0xc00;
59 	return 0;
60 }
61 early_initcall(fault_init);
62 
63 static inline int notify_page_fault(struct pt_regs *regs)
64 {
65 	int ret = 0;
66 
67 	/* kprobe_running() needs smp_processor_id() */
68 	if (kprobes_built_in() && !user_mode(regs)) {
69 		preempt_disable();
70 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
71 			ret = 1;
72 		preempt_enable();
73 	}
74 	return ret;
75 }
76 
77 
78 /*
79  * Unlock any spinlocks which will prevent us from getting the
80  * message out.
81  */
82 void bust_spinlocks(int yes)
83 {
84 	if (yes) {
85 		oops_in_progress = 1;
86 	} else {
87 		int loglevel_save = console_loglevel;
88 		console_unblank();
89 		oops_in_progress = 0;
90 		/*
91 		 * OK, the message is on the console.  Now we call printk()
92 		 * without oops_in_progress set so that printk will give klogd
93 		 * a poke.  Hold onto your hats...
94 		 */
95 		console_loglevel = 15;
96 		printk(" ");
97 		console_loglevel = loglevel_save;
98 	}
99 }
100 
101 /*
102  * Returns the address space associated with the fault.
103  * Returns 0 for kernel space and 1 for user space.
104  */
105 static inline int user_space_fault(struct pt_regs *regs)
106 {
107 	unsigned long trans_exc_code;
108 
109 	/*
110 	 * The lowest two bits of the translation exception
111 	 * identification indicate which paging table was used.
112 	 */
113 	trans_exc_code = regs->int_parm_long & 3;
114 	if (trans_exc_code == 3) /* home space -> kernel */
115 		return 0;
116 	if (user_mode(regs))
117 		return 1;
118 	if (trans_exc_code == 2) /* secondary space -> set_fs */
119 		return current->thread.mm_segment.ar4;
120 	if (current->flags & PF_VCPU)
121 		return 1;
122 	return 0;
123 }
124 
125 static int bad_address(void *p)
126 {
127 	unsigned long dummy;
128 
129 	return probe_kernel_address((unsigned long *)p, dummy);
130 }
131 
132 static void dump_pagetable(unsigned long asce, unsigned long address)
133 {
134 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
135 
136 	pr_alert("AS:%016lx ", asce);
137 	switch (asce & _ASCE_TYPE_MASK) {
138 	case _ASCE_TYPE_REGION1:
139 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
140 		if (bad_address(table))
141 			goto bad;
142 		pr_cont("R1:%016lx ", *table);
143 		if (*table & _REGION_ENTRY_INVALID)
144 			goto out;
145 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
146 		/* fallthrough */
147 	case _ASCE_TYPE_REGION2:
148 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
149 		if (bad_address(table))
150 			goto bad;
151 		pr_cont("R2:%016lx ", *table);
152 		if (*table & _REGION_ENTRY_INVALID)
153 			goto out;
154 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
155 		/* fallthrough */
156 	case _ASCE_TYPE_REGION3:
157 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
158 		if (bad_address(table))
159 			goto bad;
160 		pr_cont("R3:%016lx ", *table);
161 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
162 			goto out;
163 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
164 		/* fallthrough */
165 	case _ASCE_TYPE_SEGMENT:
166 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
167 		if (bad_address(table))
168 			goto bad;
169 		pr_cont("S:%016lx ", *table);
170 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
171 			goto out;
172 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
173 	}
174 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
175 	if (bad_address(table))
176 		goto bad;
177 	pr_cont("P:%016lx ", *table);
178 out:
179 	pr_cont("\n");
180 	return;
181 bad:
182 	pr_cont("BAD\n");
183 }
184 
185 static void dump_fault_info(struct pt_regs *regs)
186 {
187 	unsigned long asce;
188 
189 	pr_alert("Failing address: %016lx TEID: %016lx\n",
190 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
191 	pr_alert("Fault in ");
192 	switch (regs->int_parm_long & 3) {
193 	case 3:
194 		pr_cont("home space ");
195 		break;
196 	case 2:
197 		pr_cont("secondary space ");
198 		break;
199 	case 1:
200 		pr_cont("access register ");
201 		break;
202 	case 0:
203 		pr_cont("primary space ");
204 		break;
205 	}
206 	pr_cont("mode while using ");
207 	if (!user_space_fault(regs)) {
208 		asce = S390_lowcore.kernel_asce;
209 		pr_cont("kernel ");
210 	}
211 #ifdef CONFIG_PGSTE
212 	else if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
213 		struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
214 		asce = gmap->asce;
215 		pr_cont("gmap ");
216 	}
217 #endif
218 	else {
219 		asce = S390_lowcore.user_asce;
220 		pr_cont("user ");
221 	}
222 	pr_cont("ASCE.\n");
223 	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
224 }
225 
226 int show_unhandled_signals = 1;
227 
228 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
229 {
230 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
231 		return;
232 	if (!unhandled_signal(current, signr))
233 		return;
234 	if (!printk_ratelimit())
235 		return;
236 	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
237 	       regs->int_code & 0xffff, regs->int_code >> 17);
238 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
239 	printk(KERN_CONT "\n");
240 	if (is_mm_fault)
241 		dump_fault_info(regs);
242 	show_regs(regs);
243 }
244 
245 /*
246  * Send SIGSEGV to task.  This is an external routine
247  * to keep the stack usage of do_page_fault small.
248  */
249 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
250 {
251 	struct siginfo si;
252 
253 	report_user_fault(regs, SIGSEGV, 1);
254 	si.si_signo = SIGSEGV;
255 	si.si_errno = 0;
256 	si.si_code = si_code;
257 	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
258 	force_sig_info(SIGSEGV, &si, current);
259 }
260 
261 static noinline void do_no_context(struct pt_regs *regs)
262 {
263 	const struct exception_table_entry *fixup;
264 
265 	/* Are we prepared to handle this kernel fault?  */
266 	fixup = search_exception_tables(regs->psw.addr);
267 	if (fixup) {
268 		regs->psw.addr = extable_fixup(fixup);
269 		return;
270 	}
271 
272 	/*
273 	 * Oops. The kernel tried to access some bad page. We'll have to
274 	 * terminate things with extreme prejudice.
275 	 */
276 	if (!user_space_fault(regs))
277 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
278 		       " in virtual kernel address space\n");
279 	else
280 		printk(KERN_ALERT "Unable to handle kernel paging request"
281 		       " in virtual user address space\n");
282 	dump_fault_info(regs);
283 	die(regs, "Oops");
284 	do_exit(SIGKILL);
285 }
286 
287 static noinline void do_low_address(struct pt_regs *regs)
288 {
289 	/* Low-address protection hit in kernel mode means
290 	   NULL pointer write access in kernel mode.  */
291 	if (regs->psw.mask & PSW_MASK_PSTATE) {
292 		/* Low-address protection hit in user mode 'cannot happen'. */
293 		die (regs, "Low-address protection");
294 		do_exit(SIGKILL);
295 	}
296 
297 	do_no_context(regs);
298 }
299 
300 static noinline void do_sigbus(struct pt_regs *regs)
301 {
302 	struct task_struct *tsk = current;
303 	struct siginfo si;
304 
305 	/*
306 	 * Send a sigbus, regardless of whether we were in kernel
307 	 * or user mode.
308 	 */
309 	si.si_signo = SIGBUS;
310 	si.si_errno = 0;
311 	si.si_code = BUS_ADRERR;
312 	si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
313 	force_sig_info(SIGBUS, &si, tsk);
314 }
315 
316 static noinline int signal_return(struct pt_regs *regs)
317 {
318 	u16 instruction;
319 	int rc;
320 
321 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
322 	if (rc)
323 		return rc;
324 	if (instruction == 0x0a77) {
325 		set_pt_regs_flag(regs, PIF_SYSCALL);
326 		regs->int_code = 0x00040077;
327 		return 0;
328 	} else if (instruction == 0x0aad) {
329 		set_pt_regs_flag(regs, PIF_SYSCALL);
330 		regs->int_code = 0x000400ad;
331 		return 0;
332 	}
333 	return -EACCES;
334 }
335 
336 static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
337 {
338 	int si_code;
339 
340 	switch (fault) {
341 	case VM_FAULT_BADACCESS:
342 		if (access == VM_EXEC && signal_return(regs) == 0)
343 			break;
344 	case VM_FAULT_BADMAP:
345 		/* Bad memory access. Check if it is kernel or user space. */
346 		if (user_mode(regs)) {
347 			/* User mode accesses just cause a SIGSEGV */
348 			si_code = (fault == VM_FAULT_BADMAP) ?
349 				SEGV_MAPERR : SEGV_ACCERR;
350 			do_sigsegv(regs, si_code);
351 			break;
352 		}
353 	case VM_FAULT_BADCONTEXT:
354 	case VM_FAULT_PFAULT:
355 		do_no_context(regs);
356 		break;
357 	case VM_FAULT_SIGNAL:
358 		if (!user_mode(regs))
359 			do_no_context(regs);
360 		break;
361 	default: /* fault & VM_FAULT_ERROR */
362 		if (fault & VM_FAULT_OOM) {
363 			if (!user_mode(regs))
364 				do_no_context(regs);
365 			else
366 				pagefault_out_of_memory();
367 		} else if (fault & VM_FAULT_SIGSEGV) {
368 			/* Kernel mode? Handle exceptions or die */
369 			if (!user_mode(regs))
370 				do_no_context(regs);
371 			else
372 				do_sigsegv(regs, SEGV_MAPERR);
373 		} else if (fault & VM_FAULT_SIGBUS) {
374 			/* Kernel mode? Handle exceptions or die */
375 			if (!user_mode(regs))
376 				do_no_context(regs);
377 			else
378 				do_sigbus(regs);
379 		} else
380 			BUG();
381 		break;
382 	}
383 }
384 
385 /*
386  * This routine handles page faults.  It determines the address,
387  * and the problem, and then passes it off to one of the appropriate
388  * routines.
389  *
390  * interruption code (int_code):
391  *   04       Protection           ->  Write-Protection  (suprression)
392  *   10       Segment translation  ->  Not present       (nullification)
393  *   11       Page translation     ->  Not present       (nullification)
394  *   3b       Region third trans.  ->  Not present       (nullification)
395  */
396 static inline int do_exception(struct pt_regs *regs, int access)
397 {
398 #ifdef CONFIG_PGSTE
399 	struct gmap *gmap;
400 #endif
401 	struct task_struct *tsk;
402 	struct mm_struct *mm;
403 	struct vm_area_struct *vma;
404 	unsigned long trans_exc_code;
405 	unsigned long address;
406 	unsigned int flags;
407 	int fault;
408 
409 	tsk = current;
410 	/*
411 	 * The instruction that caused the program check has
412 	 * been nullified. Don't signal single step via SIGTRAP.
413 	 */
414 	clear_pt_regs_flag(regs, PIF_PER_TRAP);
415 
416 	if (notify_page_fault(regs))
417 		return 0;
418 
419 	mm = tsk->mm;
420 	trans_exc_code = regs->int_parm_long;
421 
422 	/*
423 	 * Verify that the fault happened in user space, that
424 	 * we are not in an interrupt and that there is a
425 	 * user context.
426 	 */
427 	fault = VM_FAULT_BADCONTEXT;
428 	if (unlikely(!user_space_fault(regs) || faulthandler_disabled() || !mm))
429 		goto out;
430 
431 	address = trans_exc_code & __FAIL_ADDR_MASK;
432 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
433 	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
434 	if (user_mode(regs))
435 		flags |= FAULT_FLAG_USER;
436 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
437 		flags |= FAULT_FLAG_WRITE;
438 	down_read(&mm->mmap_sem);
439 
440 #ifdef CONFIG_PGSTE
441 	gmap = (current->flags & PF_VCPU) ?
442 		(struct gmap *) S390_lowcore.gmap : NULL;
443 	if (gmap) {
444 		current->thread.gmap_addr = address;
445 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
446 		current->thread.gmap_int_code = regs->int_code & 0xffff;
447 		address = __gmap_translate(gmap, address);
448 		if (address == -EFAULT) {
449 			fault = VM_FAULT_BADMAP;
450 			goto out_up;
451 		}
452 		if (gmap->pfault_enabled)
453 			flags |= FAULT_FLAG_RETRY_NOWAIT;
454 	}
455 #endif
456 
457 retry:
458 	fault = VM_FAULT_BADMAP;
459 	vma = find_vma(mm, address);
460 	if (!vma)
461 		goto out_up;
462 
463 	if (unlikely(vma->vm_start > address)) {
464 		if (!(vma->vm_flags & VM_GROWSDOWN))
465 			goto out_up;
466 		if (expand_stack(vma, address))
467 			goto out_up;
468 	}
469 
470 	/*
471 	 * Ok, we have a good vm_area for this memory access, so
472 	 * we can handle it..
473 	 */
474 	fault = VM_FAULT_BADACCESS;
475 	if (unlikely(!(vma->vm_flags & access)))
476 		goto out_up;
477 
478 	if (is_vm_hugetlb_page(vma))
479 		address &= HPAGE_MASK;
480 	/*
481 	 * If for any reason at all we couldn't handle the fault,
482 	 * make sure we exit gracefully rather than endlessly redo
483 	 * the fault.
484 	 */
485 	fault = handle_mm_fault(vma, address, flags);
486 	/* No reason to continue if interrupted by SIGKILL. */
487 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
488 		fault = VM_FAULT_SIGNAL;
489 		goto out;
490 	}
491 	if (unlikely(fault & VM_FAULT_ERROR))
492 		goto out_up;
493 
494 	/*
495 	 * Major/minor page fault accounting is only done on the
496 	 * initial attempt. If we go through a retry, it is extremely
497 	 * likely that the page will be found in page cache at that point.
498 	 */
499 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
500 		if (fault & VM_FAULT_MAJOR) {
501 			tsk->maj_flt++;
502 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
503 				      regs, address);
504 		} else {
505 			tsk->min_flt++;
506 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
507 				      regs, address);
508 		}
509 		if (fault & VM_FAULT_RETRY) {
510 #ifdef CONFIG_PGSTE
511 			if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
512 				/* FAULT_FLAG_RETRY_NOWAIT has been set,
513 				 * mmap_sem has not been released */
514 				current->thread.gmap_pfault = 1;
515 				fault = VM_FAULT_PFAULT;
516 				goto out_up;
517 			}
518 #endif
519 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
520 			 * of starvation. */
521 			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
522 				   FAULT_FLAG_RETRY_NOWAIT);
523 			flags |= FAULT_FLAG_TRIED;
524 			down_read(&mm->mmap_sem);
525 			goto retry;
526 		}
527 	}
528 #ifdef CONFIG_PGSTE
529 	if (gmap) {
530 		address =  __gmap_link(gmap, current->thread.gmap_addr,
531 				       address);
532 		if (address == -EFAULT) {
533 			fault = VM_FAULT_BADMAP;
534 			goto out_up;
535 		}
536 		if (address == -ENOMEM) {
537 			fault = VM_FAULT_OOM;
538 			goto out_up;
539 		}
540 	}
541 #endif
542 	fault = 0;
543 out_up:
544 	up_read(&mm->mmap_sem);
545 out:
546 	return fault;
547 }
548 
549 void do_protection_exception(struct pt_regs *regs)
550 {
551 	unsigned long trans_exc_code;
552 	int access, fault;
553 
554 	trans_exc_code = regs->int_parm_long;
555 	/*
556 	 * Protection exceptions are suppressing, decrement psw address.
557 	 * The exception to this rule are aborted transactions, for these
558 	 * the PSW already points to the correct location.
559 	 */
560 	if (!(regs->int_code & 0x200))
561 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
562 	/*
563 	 * Check for low-address protection.  This needs to be treated
564 	 * as a special case because the translation exception code
565 	 * field is not guaranteed to contain valid data in this case.
566 	 */
567 	if (unlikely(!(trans_exc_code & 4))) {
568 		do_low_address(regs);
569 		return;
570 	}
571 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
572 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
573 					(regs->psw.addr & PAGE_MASK);
574 		access = VM_EXEC;
575 		fault = VM_FAULT_BADACCESS;
576 	} else {
577 		access = VM_WRITE;
578 		fault = do_exception(regs, access);
579 	}
580 	if (unlikely(fault))
581 		do_fault_error(regs, access, fault);
582 }
583 NOKPROBE_SYMBOL(do_protection_exception);
584 
585 void do_dat_exception(struct pt_regs *regs)
586 {
587 	int access, fault;
588 
589 	access = VM_READ | VM_EXEC | VM_WRITE;
590 	fault = do_exception(regs, access);
591 	if (unlikely(fault))
592 		do_fault_error(regs, access, fault);
593 }
594 NOKPROBE_SYMBOL(do_dat_exception);
595 
596 #ifdef CONFIG_PFAULT
597 /*
598  * 'pfault' pseudo page faults routines.
599  */
600 static int pfault_disable;
601 
602 static int __init nopfault(char *str)
603 {
604 	pfault_disable = 1;
605 	return 1;
606 }
607 
608 __setup("nopfault", nopfault);
609 
610 struct pfault_refbk {
611 	u16 refdiagc;
612 	u16 reffcode;
613 	u16 refdwlen;
614 	u16 refversn;
615 	u64 refgaddr;
616 	u64 refselmk;
617 	u64 refcmpmk;
618 	u64 reserved;
619 } __attribute__ ((packed, aligned(8)));
620 
621 int pfault_init(void)
622 {
623 	struct pfault_refbk refbk = {
624 		.refdiagc = 0x258,
625 		.reffcode = 0,
626 		.refdwlen = 5,
627 		.refversn = 2,
628 		.refgaddr = __LC_LPP,
629 		.refselmk = 1ULL << 48,
630 		.refcmpmk = 1ULL << 48,
631 		.reserved = __PF_RES_FIELD };
632         int rc;
633 
634 	if (pfault_disable)
635 		return -1;
636 	diag_stat_inc(DIAG_STAT_X258);
637 	asm volatile(
638 		"	diag	%1,%0,0x258\n"
639 		"0:	j	2f\n"
640 		"1:	la	%0,8\n"
641 		"2:\n"
642 		EX_TABLE(0b,1b)
643 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
644         return rc;
645 }
646 
647 void pfault_fini(void)
648 {
649 	struct pfault_refbk refbk = {
650 		.refdiagc = 0x258,
651 		.reffcode = 1,
652 		.refdwlen = 5,
653 		.refversn = 2,
654 	};
655 
656 	if (pfault_disable)
657 		return;
658 	diag_stat_inc(DIAG_STAT_X258);
659 	asm volatile(
660 		"	diag	%0,0,0x258\n"
661 		"0:	nopr	%%r7\n"
662 		EX_TABLE(0b,0b)
663 		: : "a" (&refbk), "m" (refbk) : "cc");
664 }
665 
666 static DEFINE_SPINLOCK(pfault_lock);
667 static LIST_HEAD(pfault_list);
668 
669 #define PF_COMPLETE	0x0080
670 
671 /*
672  * The mechanism of our pfault code: if Linux is running as guest, runs a user
673  * space process and the user space process accesses a page that the host has
674  * paged out we get a pfault interrupt.
675  *
676  * This allows us, within the guest, to schedule a different process. Without
677  * this mechanism the host would have to suspend the whole virtual cpu until
678  * the page has been paged in.
679  *
680  * So when we get such an interrupt then we set the state of the current task
681  * to uninterruptible and also set the need_resched flag. Both happens within
682  * interrupt context(!). If we later on want to return to user space we
683  * recognize the need_resched flag and then call schedule().  It's not very
684  * obvious how this works...
685  *
686  * Of course we have a lot of additional fun with the completion interrupt (->
687  * host signals that a page of a process has been paged in and the process can
688  * continue to run). This interrupt can arrive on any cpu and, since we have
689  * virtual cpus, actually appear before the interrupt that signals that a page
690  * is missing.
691  */
692 static void pfault_interrupt(struct ext_code ext_code,
693 			     unsigned int param32, unsigned long param64)
694 {
695 	struct task_struct *tsk;
696 	__u16 subcode;
697 	pid_t pid;
698 
699 	/*
700 	 * Get the external interruption subcode & pfault initial/completion
701 	 * signal bit. VM stores this in the 'cpu address' field associated
702 	 * with the external interrupt.
703 	 */
704 	subcode = ext_code.subcode;
705 	if ((subcode & 0xff00) != __SUBCODE_MASK)
706 		return;
707 	inc_irq_stat(IRQEXT_PFL);
708 	/* Get the token (= pid of the affected task). */
709 	pid = param64 & LPP_PFAULT_PID_MASK;
710 	rcu_read_lock();
711 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
712 	if (tsk)
713 		get_task_struct(tsk);
714 	rcu_read_unlock();
715 	if (!tsk)
716 		return;
717 	spin_lock(&pfault_lock);
718 	if (subcode & PF_COMPLETE) {
719 		/* signal bit is set -> a page has been swapped in by VM */
720 		if (tsk->thread.pfault_wait == 1) {
721 			/* Initial interrupt was faster than the completion
722 			 * interrupt. pfault_wait is valid. Set pfault_wait
723 			 * back to zero and wake up the process. This can
724 			 * safely be done because the task is still sleeping
725 			 * and can't produce new pfaults. */
726 			tsk->thread.pfault_wait = 0;
727 			list_del(&tsk->thread.list);
728 			wake_up_process(tsk);
729 			put_task_struct(tsk);
730 		} else {
731 			/* Completion interrupt was faster than initial
732 			 * interrupt. Set pfault_wait to -1 so the initial
733 			 * interrupt doesn't put the task to sleep.
734 			 * If the task is not running, ignore the completion
735 			 * interrupt since it must be a leftover of a PFAULT
736 			 * CANCEL operation which didn't remove all pending
737 			 * completion interrupts. */
738 			if (tsk->state == TASK_RUNNING)
739 				tsk->thread.pfault_wait = -1;
740 		}
741 	} else {
742 		/* signal bit not set -> a real page is missing. */
743 		if (WARN_ON_ONCE(tsk != current))
744 			goto out;
745 		if (tsk->thread.pfault_wait == 1) {
746 			/* Already on the list with a reference: put to sleep */
747 			goto block;
748 		} else if (tsk->thread.pfault_wait == -1) {
749 			/* Completion interrupt was faster than the initial
750 			 * interrupt (pfault_wait == -1). Set pfault_wait
751 			 * back to zero and exit. */
752 			tsk->thread.pfault_wait = 0;
753 		} else {
754 			/* Initial interrupt arrived before completion
755 			 * interrupt. Let the task sleep.
756 			 * An extra task reference is needed since a different
757 			 * cpu may set the task state to TASK_RUNNING again
758 			 * before the scheduler is reached. */
759 			get_task_struct(tsk);
760 			tsk->thread.pfault_wait = 1;
761 			list_add(&tsk->thread.list, &pfault_list);
762 block:
763 			/* Since this must be a userspace fault, there
764 			 * is no kernel task state to trample. Rely on the
765 			 * return to userspace schedule() to block. */
766 			__set_current_state(TASK_UNINTERRUPTIBLE);
767 			set_tsk_need_resched(tsk);
768 			set_preempt_need_resched();
769 		}
770 	}
771 out:
772 	spin_unlock(&pfault_lock);
773 	put_task_struct(tsk);
774 }
775 
776 static int pfault_cpu_dead(unsigned int cpu)
777 {
778 	struct thread_struct *thread, *next;
779 	struct task_struct *tsk;
780 
781 	spin_lock_irq(&pfault_lock);
782 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
783 		thread->pfault_wait = 0;
784 		list_del(&thread->list);
785 		tsk = container_of(thread, struct task_struct, thread);
786 		wake_up_process(tsk);
787 		put_task_struct(tsk);
788 	}
789 	spin_unlock_irq(&pfault_lock);
790 	return 0;
791 }
792 
793 static int __init pfault_irq_init(void)
794 {
795 	int rc;
796 
797 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
798 	if (rc)
799 		goto out_extint;
800 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
801 	if (rc)
802 		goto out_pfault;
803 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
804 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
805 				  NULL, pfault_cpu_dead);
806 	return 0;
807 
808 out_pfault:
809 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
810 out_extint:
811 	pfault_disable = 1;
812 	return rc;
813 }
814 early_initcall(pfault_irq_init);
815 
816 #endif /* CONFIG_PFAULT */
817