xref: /openbmc/linux/arch/s390/mm/fault.c (revision f9a82c48)
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 enum fault_type {
54 	KERNEL_FAULT,
55 	USER_FAULT,
56 	VDSO_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 static inline int notify_page_fault(struct pt_regs *regs)
71 {
72 	int ret = 0;
73 
74 	/* kprobe_running() needs smp_processor_id() */
75 	if (kprobes_built_in() && !user_mode(regs)) {
76 		preempt_disable();
77 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
78 			ret = 1;
79 		preempt_enable();
80 	}
81 	return ret;
82 }
83 
84 /*
85  * Find out which address space caused the exception.
86  * Access register mode is impossible, ignore space == 3.
87  */
88 static inline enum fault_type get_fault_type(struct pt_regs *regs)
89 {
90 	unsigned long trans_exc_code;
91 
92 	trans_exc_code = regs->int_parm_long & 3;
93 	if (likely(trans_exc_code == 0)) {
94 		/* primary space exception */
95 		if (IS_ENABLED(CONFIG_PGSTE) &&
96 		    test_pt_regs_flag(regs, PIF_GUEST_FAULT))
97 			return GMAP_FAULT;
98 		if (current->thread.mm_segment == USER_DS)
99 			return USER_FAULT;
100 		return KERNEL_FAULT;
101 	}
102 	if (trans_exc_code == 2) {
103 		/* secondary space exception */
104 		if (current->thread.mm_segment & 1) {
105 			if (current->thread.mm_segment == USER_DS_SACF)
106 				return USER_FAULT;
107 			return KERNEL_FAULT;
108 		}
109 		return VDSO_FAULT;
110 	}
111 	/* home space exception -> access via kernel ASCE */
112 	return KERNEL_FAULT;
113 }
114 
115 static int bad_address(void *p)
116 {
117 	unsigned long dummy;
118 
119 	return probe_kernel_address((unsigned long *)p, dummy);
120 }
121 
122 static void dump_pagetable(unsigned long asce, unsigned long address)
123 {
124 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
125 
126 	pr_alert("AS:%016lx ", asce);
127 	switch (asce & _ASCE_TYPE_MASK) {
128 	case _ASCE_TYPE_REGION1:
129 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
130 		if (bad_address(table))
131 			goto bad;
132 		pr_cont("R1:%016lx ", *table);
133 		if (*table & _REGION_ENTRY_INVALID)
134 			goto out;
135 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
136 		/* fallthrough */
137 	case _ASCE_TYPE_REGION2:
138 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
139 		if (bad_address(table))
140 			goto bad;
141 		pr_cont("R2:%016lx ", *table);
142 		if (*table & _REGION_ENTRY_INVALID)
143 			goto out;
144 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
145 		/* fallthrough */
146 	case _ASCE_TYPE_REGION3:
147 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
148 		if (bad_address(table))
149 			goto bad;
150 		pr_cont("R3:%016lx ", *table);
151 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
152 			goto out;
153 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
154 		/* fallthrough */
155 	case _ASCE_TYPE_SEGMENT:
156 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
157 		if (bad_address(table))
158 			goto bad;
159 		pr_cont("S:%016lx ", *table);
160 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
161 			goto out;
162 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
163 	}
164 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
165 	if (bad_address(table))
166 		goto bad;
167 	pr_cont("P:%016lx ", *table);
168 out:
169 	pr_cont("\n");
170 	return;
171 bad:
172 	pr_cont("BAD\n");
173 }
174 
175 static void dump_fault_info(struct pt_regs *regs)
176 {
177 	unsigned long asce;
178 
179 	pr_alert("Failing address: %016lx TEID: %016lx\n",
180 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
181 	pr_alert("Fault in ");
182 	switch (regs->int_parm_long & 3) {
183 	case 3:
184 		pr_cont("home space ");
185 		break;
186 	case 2:
187 		pr_cont("secondary space ");
188 		break;
189 	case 1:
190 		pr_cont("access register ");
191 		break;
192 	case 0:
193 		pr_cont("primary space ");
194 		break;
195 	}
196 	pr_cont("mode while using ");
197 	switch (get_fault_type(regs)) {
198 	case USER_FAULT:
199 		asce = S390_lowcore.user_asce;
200 		pr_cont("user ");
201 		break;
202 	case VDSO_FAULT:
203 		asce = S390_lowcore.vdso_asce;
204 		pr_cont("vdso ");
205 		break;
206 	case GMAP_FAULT:
207 		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
208 		pr_cont("gmap ");
209 		break;
210 	case KERNEL_FAULT:
211 		asce = S390_lowcore.kernel_asce;
212 		pr_cont("kernel ");
213 		break;
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 			current);
248 }
249 
250 static noinline void do_no_context(struct pt_regs *regs)
251 {
252 	const struct exception_table_entry *fixup;
253 
254 	/* Are we prepared to handle this kernel fault?  */
255 	fixup = search_exception_tables(regs->psw.addr);
256 	if (fixup) {
257 		regs->psw.addr = extable_fixup(fixup);
258 		return;
259 	}
260 
261 	/*
262 	 * Oops. The kernel tried to access some bad page. We'll have to
263 	 * terminate things with extreme prejudice.
264 	 */
265 	if (get_fault_type(regs) == KERNEL_FAULT)
266 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
267 		       " in virtual kernel address space\n");
268 	else
269 		printk(KERN_ALERT "Unable to handle kernel paging request"
270 		       " in virtual user address space\n");
271 	dump_fault_info(regs);
272 	die(regs, "Oops");
273 	do_exit(SIGKILL);
274 }
275 
276 static noinline void do_low_address(struct pt_regs *regs)
277 {
278 	/* Low-address protection hit in kernel mode means
279 	   NULL pointer write access in kernel mode.  */
280 	if (regs->psw.mask & PSW_MASK_PSTATE) {
281 		/* Low-address protection hit in user mode 'cannot happen'. */
282 		die (regs, "Low-address protection");
283 		do_exit(SIGKILL);
284 	}
285 
286 	do_no_context(regs);
287 }
288 
289 static noinline void do_sigbus(struct pt_regs *regs)
290 {
291 	/*
292 	 * Send a sigbus, regardless of whether we were in kernel
293 	 * or user mode.
294 	 */
295 	force_sig_fault(SIGBUS, BUS_ADRERR,
296 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK),
297 			current);
298 }
299 
300 static noinline int signal_return(struct pt_regs *regs)
301 {
302 	u16 instruction;
303 	int rc;
304 
305 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
306 	if (rc)
307 		return rc;
308 	if (instruction == 0x0a77) {
309 		set_pt_regs_flag(regs, PIF_SYSCALL);
310 		regs->int_code = 0x00040077;
311 		return 0;
312 	} else if (instruction == 0x0aad) {
313 		set_pt_regs_flag(regs, PIF_SYSCALL);
314 		regs->int_code = 0x000400ad;
315 		return 0;
316 	}
317 	return -EACCES;
318 }
319 
320 static noinline void do_fault_error(struct pt_regs *regs, int access,
321 					vm_fault_t fault)
322 {
323 	int si_code;
324 
325 	switch (fault) {
326 	case VM_FAULT_BADACCESS:
327 		if (access == VM_EXEC && signal_return(regs) == 0)
328 			break;
329 	case VM_FAULT_BADMAP:
330 		/* Bad memory access. Check if it is kernel or user space. */
331 		if (user_mode(regs)) {
332 			/* User mode accesses just cause a SIGSEGV */
333 			si_code = (fault == VM_FAULT_BADMAP) ?
334 				SEGV_MAPERR : SEGV_ACCERR;
335 			do_sigsegv(regs, si_code);
336 			break;
337 		}
338 	case VM_FAULT_BADCONTEXT:
339 	case VM_FAULT_PFAULT:
340 		do_no_context(regs);
341 		break;
342 	case VM_FAULT_SIGNAL:
343 		if (!user_mode(regs))
344 			do_no_context(regs);
345 		break;
346 	default: /* fault & VM_FAULT_ERROR */
347 		if (fault & VM_FAULT_OOM) {
348 			if (!user_mode(regs))
349 				do_no_context(regs);
350 			else
351 				pagefault_out_of_memory();
352 		} else if (fault & VM_FAULT_SIGSEGV) {
353 			/* Kernel mode? Handle exceptions or die */
354 			if (!user_mode(regs))
355 				do_no_context(regs);
356 			else
357 				do_sigsegv(regs, SEGV_MAPERR);
358 		} else if (fault & VM_FAULT_SIGBUS) {
359 			/* Kernel mode? Handle exceptions or die */
360 			if (!user_mode(regs))
361 				do_no_context(regs);
362 			else
363 				do_sigbus(regs);
364 		} else
365 			BUG();
366 		break;
367 	}
368 }
369 
370 /*
371  * This routine handles page faults.  It determines the address,
372  * and the problem, and then passes it off to one of the appropriate
373  * routines.
374  *
375  * interruption code (int_code):
376  *   04       Protection           ->  Write-Protection  (suprression)
377  *   10       Segment translation  ->  Not present       (nullification)
378  *   11       Page translation     ->  Not present       (nullification)
379  *   3b       Region third trans.  ->  Not present       (nullification)
380  */
381 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
382 {
383 	struct gmap *gmap;
384 	struct task_struct *tsk;
385 	struct mm_struct *mm;
386 	struct vm_area_struct *vma;
387 	enum fault_type type;
388 	unsigned long trans_exc_code;
389 	unsigned long address;
390 	unsigned int flags;
391 	vm_fault_t fault;
392 
393 	tsk = current;
394 	/*
395 	 * The instruction that caused the program check has
396 	 * been nullified. Don't signal single step via SIGTRAP.
397 	 */
398 	clear_pt_regs_flag(regs, PIF_PER_TRAP);
399 
400 	if (notify_page_fault(regs))
401 		return 0;
402 
403 	mm = tsk->mm;
404 	trans_exc_code = regs->int_parm_long;
405 
406 	/*
407 	 * Verify that the fault happened in user space, that
408 	 * we are not in an interrupt and that there is a
409 	 * user context.
410 	 */
411 	fault = VM_FAULT_BADCONTEXT;
412 	type = get_fault_type(regs);
413 	switch (type) {
414 	case KERNEL_FAULT:
415 		goto out;
416 	case VDSO_FAULT:
417 		fault = VM_FAULT_BADMAP;
418 		goto out;
419 	case USER_FAULT:
420 	case GMAP_FAULT:
421 		if (faulthandler_disabled() || !mm)
422 			goto out;
423 		break;
424 	}
425 
426 	address = trans_exc_code & __FAIL_ADDR_MASK;
427 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
428 	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
429 	if (user_mode(regs))
430 		flags |= FAULT_FLAG_USER;
431 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
432 		flags |= FAULT_FLAG_WRITE;
433 	down_read(&mm->mmap_sem);
434 
435 	gmap = NULL;
436 	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
437 		gmap = (struct gmap *) S390_lowcore.gmap;
438 		current->thread.gmap_addr = address;
439 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
440 		current->thread.gmap_int_code = regs->int_code & 0xffff;
441 		address = __gmap_translate(gmap, address);
442 		if (address == -EFAULT) {
443 			fault = VM_FAULT_BADMAP;
444 			goto out_up;
445 		}
446 		if (gmap->pfault_enabled)
447 			flags |= FAULT_FLAG_RETRY_NOWAIT;
448 	}
449 
450 retry:
451 	fault = VM_FAULT_BADMAP;
452 	vma = find_vma(mm, address);
453 	if (!vma)
454 		goto out_up;
455 
456 	if (unlikely(vma->vm_start > address)) {
457 		if (!(vma->vm_flags & VM_GROWSDOWN))
458 			goto out_up;
459 		if (expand_stack(vma, address))
460 			goto out_up;
461 	}
462 
463 	/*
464 	 * Ok, we have a good vm_area for this memory access, so
465 	 * we can handle it..
466 	 */
467 	fault = VM_FAULT_BADACCESS;
468 	if (unlikely(!(vma->vm_flags & access)))
469 		goto out_up;
470 
471 	if (is_vm_hugetlb_page(vma))
472 		address &= HPAGE_MASK;
473 	/*
474 	 * If for any reason at all we couldn't handle the fault,
475 	 * make sure we exit gracefully rather than endlessly redo
476 	 * the fault.
477 	 */
478 	fault = handle_mm_fault(vma, address, flags);
479 	/* No reason to continue if interrupted by SIGKILL. */
480 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
481 		fault = VM_FAULT_SIGNAL;
482 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
483 			goto out_up;
484 		goto out;
485 	}
486 	if (unlikely(fault & VM_FAULT_ERROR))
487 		goto out_up;
488 
489 	/*
490 	 * Major/minor page fault accounting is only done on the
491 	 * initial attempt. If we go through a retry, it is extremely
492 	 * likely that the page will be found in page cache at that point.
493 	 */
494 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
495 		if (fault & VM_FAULT_MAJOR) {
496 			tsk->maj_flt++;
497 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
498 				      regs, address);
499 		} else {
500 			tsk->min_flt++;
501 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
502 				      regs, address);
503 		}
504 		if (fault & VM_FAULT_RETRY) {
505 			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
506 			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
507 				/* FAULT_FLAG_RETRY_NOWAIT has been set,
508 				 * mmap_sem has not been released */
509 				current->thread.gmap_pfault = 1;
510 				fault = VM_FAULT_PFAULT;
511 				goto out_up;
512 			}
513 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
514 			 * of starvation. */
515 			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
516 				   FAULT_FLAG_RETRY_NOWAIT);
517 			flags |= FAULT_FLAG_TRIED;
518 			down_read(&mm->mmap_sem);
519 			goto retry;
520 		}
521 	}
522 	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
523 		address =  __gmap_link(gmap, current->thread.gmap_addr,
524 				       address);
525 		if (address == -EFAULT) {
526 			fault = VM_FAULT_BADMAP;
527 			goto out_up;
528 		}
529 		if (address == -ENOMEM) {
530 			fault = VM_FAULT_OOM;
531 			goto out_up;
532 		}
533 	}
534 	fault = 0;
535 out_up:
536 	up_read(&mm->mmap_sem);
537 out:
538 	return fault;
539 }
540 
541 void do_protection_exception(struct pt_regs *regs)
542 {
543 	unsigned long trans_exc_code;
544 	int access;
545 	vm_fault_t fault;
546 
547 	trans_exc_code = regs->int_parm_long;
548 	/*
549 	 * Protection exceptions are suppressing, decrement psw address.
550 	 * The exception to this rule are aborted transactions, for these
551 	 * the PSW already points to the correct location.
552 	 */
553 	if (!(regs->int_code & 0x200))
554 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
555 	/*
556 	 * Check for low-address protection.  This needs to be treated
557 	 * as a special case because the translation exception code
558 	 * field is not guaranteed to contain valid data in this case.
559 	 */
560 	if (unlikely(!(trans_exc_code & 4))) {
561 		do_low_address(regs);
562 		return;
563 	}
564 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
565 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
566 					(regs->psw.addr & PAGE_MASK);
567 		access = VM_EXEC;
568 		fault = VM_FAULT_BADACCESS;
569 	} else {
570 		access = VM_WRITE;
571 		fault = do_exception(regs, access);
572 	}
573 	if (unlikely(fault))
574 		do_fault_error(regs, access, fault);
575 }
576 NOKPROBE_SYMBOL(do_protection_exception);
577 
578 void do_dat_exception(struct pt_regs *regs)
579 {
580 	int access;
581 	vm_fault_t fault;
582 
583 	access = VM_READ | VM_EXEC | VM_WRITE;
584 	fault = do_exception(regs, access);
585 	if (unlikely(fault))
586 		do_fault_error(regs, access, fault);
587 }
588 NOKPROBE_SYMBOL(do_dat_exception);
589 
590 #ifdef CONFIG_PFAULT
591 /*
592  * 'pfault' pseudo page faults routines.
593  */
594 static int pfault_disable;
595 
596 static int __init nopfault(char *str)
597 {
598 	pfault_disable = 1;
599 	return 1;
600 }
601 
602 __setup("nopfault", nopfault);
603 
604 struct pfault_refbk {
605 	u16 refdiagc;
606 	u16 reffcode;
607 	u16 refdwlen;
608 	u16 refversn;
609 	u64 refgaddr;
610 	u64 refselmk;
611 	u64 refcmpmk;
612 	u64 reserved;
613 } __attribute__ ((packed, aligned(8)));
614 
615 static struct pfault_refbk pfault_init_refbk = {
616 	.refdiagc = 0x258,
617 	.reffcode = 0,
618 	.refdwlen = 5,
619 	.refversn = 2,
620 	.refgaddr = __LC_LPP,
621 	.refselmk = 1ULL << 48,
622 	.refcmpmk = 1ULL << 48,
623 	.reserved = __PF_RES_FIELD
624 };
625 
626 int pfault_init(void)
627 {
628         int rc;
629 
630 	if (pfault_disable)
631 		return -1;
632 	diag_stat_inc(DIAG_STAT_X258);
633 	asm volatile(
634 		"	diag	%1,%0,0x258\n"
635 		"0:	j	2f\n"
636 		"1:	la	%0,8\n"
637 		"2:\n"
638 		EX_TABLE(0b,1b)
639 		: "=d" (rc)
640 		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
641         return rc;
642 }
643 
644 static struct pfault_refbk pfault_fini_refbk = {
645 	.refdiagc = 0x258,
646 	.reffcode = 1,
647 	.refdwlen = 5,
648 	.refversn = 2,
649 };
650 
651 void pfault_fini(void)
652 {
653 
654 	if (pfault_disable)
655 		return;
656 	diag_stat_inc(DIAG_STAT_X258);
657 	asm volatile(
658 		"	diag	%0,0,0x258\n"
659 		"0:	nopr	%%r7\n"
660 		EX_TABLE(0b,0b)
661 		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
662 }
663 
664 static DEFINE_SPINLOCK(pfault_lock);
665 static LIST_HEAD(pfault_list);
666 
667 #define PF_COMPLETE	0x0080
668 
669 /*
670  * The mechanism of our pfault code: if Linux is running as guest, runs a user
671  * space process and the user space process accesses a page that the host has
672  * paged out we get a pfault interrupt.
673  *
674  * This allows us, within the guest, to schedule a different process. Without
675  * this mechanism the host would have to suspend the whole virtual cpu until
676  * the page has been paged in.
677  *
678  * So when we get such an interrupt then we set the state of the current task
679  * to uninterruptible and also set the need_resched flag. Both happens within
680  * interrupt context(!). If we later on want to return to user space we
681  * recognize the need_resched flag and then call schedule().  It's not very
682  * obvious how this works...
683  *
684  * Of course we have a lot of additional fun with the completion interrupt (->
685  * host signals that a page of a process has been paged in and the process can
686  * continue to run). This interrupt can arrive on any cpu and, since we have
687  * virtual cpus, actually appear before the interrupt that signals that a page
688  * is missing.
689  */
690 static void pfault_interrupt(struct ext_code ext_code,
691 			     unsigned int param32, unsigned long param64)
692 {
693 	struct task_struct *tsk;
694 	__u16 subcode;
695 	pid_t pid;
696 
697 	/*
698 	 * Get the external interruption subcode & pfault initial/completion
699 	 * signal bit. VM stores this in the 'cpu address' field associated
700 	 * with the external interrupt.
701 	 */
702 	subcode = ext_code.subcode;
703 	if ((subcode & 0xff00) != __SUBCODE_MASK)
704 		return;
705 	inc_irq_stat(IRQEXT_PFL);
706 	/* Get the token (= pid of the affected task). */
707 	pid = param64 & LPP_PID_MASK;
708 	rcu_read_lock();
709 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
710 	if (tsk)
711 		get_task_struct(tsk);
712 	rcu_read_unlock();
713 	if (!tsk)
714 		return;
715 	spin_lock(&pfault_lock);
716 	if (subcode & PF_COMPLETE) {
717 		/* signal bit is set -> a page has been swapped in by VM */
718 		if (tsk->thread.pfault_wait == 1) {
719 			/* Initial interrupt was faster than the completion
720 			 * interrupt. pfault_wait is valid. Set pfault_wait
721 			 * back to zero and wake up the process. This can
722 			 * safely be done because the task is still sleeping
723 			 * and can't produce new pfaults. */
724 			tsk->thread.pfault_wait = 0;
725 			list_del(&tsk->thread.list);
726 			wake_up_process(tsk);
727 			put_task_struct(tsk);
728 		} else {
729 			/* Completion interrupt was faster than initial
730 			 * interrupt. Set pfault_wait to -1 so the initial
731 			 * interrupt doesn't put the task to sleep.
732 			 * If the task is not running, ignore the completion
733 			 * interrupt since it must be a leftover of a PFAULT
734 			 * CANCEL operation which didn't remove all pending
735 			 * completion interrupts. */
736 			if (tsk->state == TASK_RUNNING)
737 				tsk->thread.pfault_wait = -1;
738 		}
739 	} else {
740 		/* signal bit not set -> a real page is missing. */
741 		if (WARN_ON_ONCE(tsk != current))
742 			goto out;
743 		if (tsk->thread.pfault_wait == 1) {
744 			/* Already on the list with a reference: put to sleep */
745 			goto block;
746 		} else if (tsk->thread.pfault_wait == -1) {
747 			/* Completion interrupt was faster than the initial
748 			 * interrupt (pfault_wait == -1). Set pfault_wait
749 			 * back to zero and exit. */
750 			tsk->thread.pfault_wait = 0;
751 		} else {
752 			/* Initial interrupt arrived before completion
753 			 * interrupt. Let the task sleep.
754 			 * An extra task reference is needed since a different
755 			 * cpu may set the task state to TASK_RUNNING again
756 			 * before the scheduler is reached. */
757 			get_task_struct(tsk);
758 			tsk->thread.pfault_wait = 1;
759 			list_add(&tsk->thread.list, &pfault_list);
760 block:
761 			/* Since this must be a userspace fault, there
762 			 * is no kernel task state to trample. Rely on the
763 			 * return to userspace schedule() to block. */
764 			__set_current_state(TASK_UNINTERRUPTIBLE);
765 			set_tsk_need_resched(tsk);
766 			set_preempt_need_resched();
767 		}
768 	}
769 out:
770 	spin_unlock(&pfault_lock);
771 	put_task_struct(tsk);
772 }
773 
774 static int pfault_cpu_dead(unsigned int cpu)
775 {
776 	struct thread_struct *thread, *next;
777 	struct task_struct *tsk;
778 
779 	spin_lock_irq(&pfault_lock);
780 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
781 		thread->pfault_wait = 0;
782 		list_del(&thread->list);
783 		tsk = container_of(thread, struct task_struct, thread);
784 		wake_up_process(tsk);
785 		put_task_struct(tsk);
786 	}
787 	spin_unlock_irq(&pfault_lock);
788 	return 0;
789 }
790 
791 static int __init pfault_irq_init(void)
792 {
793 	int rc;
794 
795 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
796 	if (rc)
797 		goto out_extint;
798 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
799 	if (rc)
800 		goto out_pfault;
801 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
802 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
803 				  NULL, pfault_cpu_dead);
804 	return 0;
805 
806 out_pfault:
807 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
808 out_extint:
809 	pfault_disable = 1;
810 	return rc;
811 }
812 early_initcall(pfault_irq_init);
813 
814 #endif /* CONFIG_PFAULT */
815