xref: /openbmc/linux/arch/s390/mm/fault.c (revision 113094f7)
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  */
87 static enum fault_type get_fault_type(struct pt_regs *regs)
88 {
89 	unsigned long trans_exc_code;
90 
91 	trans_exc_code = regs->int_parm_long & 3;
92 	if (likely(trans_exc_code == 0)) {
93 		/* primary space exception */
94 		if (IS_ENABLED(CONFIG_PGSTE) &&
95 		    test_pt_regs_flag(regs, PIF_GUEST_FAULT))
96 			return GMAP_FAULT;
97 		if (current->thread.mm_segment == USER_DS)
98 			return USER_FAULT;
99 		return KERNEL_FAULT;
100 	}
101 	if (trans_exc_code == 2) {
102 		/* secondary space exception */
103 		if (current->thread.mm_segment & 1) {
104 			if (current->thread.mm_segment == USER_DS_SACF)
105 				return USER_FAULT;
106 			return KERNEL_FAULT;
107 		}
108 		return VDSO_FAULT;
109 	}
110 	if (trans_exc_code == 1) {
111 		/* access register mode, not used in the kernel */
112 		return USER_FAULT;
113 	}
114 	/* home space exception -> access via kernel ASCE */
115 	return KERNEL_FAULT;
116 }
117 
118 static int bad_address(void *p)
119 {
120 	unsigned long dummy;
121 
122 	return probe_kernel_address((unsigned long *)p, dummy);
123 }
124 
125 static void dump_pagetable(unsigned long asce, unsigned long address)
126 {
127 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
128 
129 	pr_alert("AS:%016lx ", asce);
130 	switch (asce & _ASCE_TYPE_MASK) {
131 	case _ASCE_TYPE_REGION1:
132 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
133 		if (bad_address(table))
134 			goto bad;
135 		pr_cont("R1:%016lx ", *table);
136 		if (*table & _REGION_ENTRY_INVALID)
137 			goto out;
138 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
139 		/* fallthrough */
140 	case _ASCE_TYPE_REGION2:
141 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
142 		if (bad_address(table))
143 			goto bad;
144 		pr_cont("R2:%016lx ", *table);
145 		if (*table & _REGION_ENTRY_INVALID)
146 			goto out;
147 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
148 		/* fallthrough */
149 	case _ASCE_TYPE_REGION3:
150 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
151 		if (bad_address(table))
152 			goto bad;
153 		pr_cont("R3:%016lx ", *table);
154 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
155 			goto out;
156 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
157 		/* fallthrough */
158 	case _ASCE_TYPE_SEGMENT:
159 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
160 		if (bad_address(table))
161 			goto bad;
162 		pr_cont("S:%016lx ", *table);
163 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
164 			goto out;
165 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
166 	}
167 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
168 	if (bad_address(table))
169 		goto bad;
170 	pr_cont("P:%016lx ", *table);
171 out:
172 	pr_cont("\n");
173 	return;
174 bad:
175 	pr_cont("BAD\n");
176 }
177 
178 static void dump_fault_info(struct pt_regs *regs)
179 {
180 	unsigned long asce;
181 
182 	pr_alert("Failing address: %016lx TEID: %016lx\n",
183 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
184 	pr_alert("Fault in ");
185 	switch (regs->int_parm_long & 3) {
186 	case 3:
187 		pr_cont("home space ");
188 		break;
189 	case 2:
190 		pr_cont("secondary space ");
191 		break;
192 	case 1:
193 		pr_cont("access register ");
194 		break;
195 	case 0:
196 		pr_cont("primary space ");
197 		break;
198 	}
199 	pr_cont("mode while using ");
200 	switch (get_fault_type(regs)) {
201 	case USER_FAULT:
202 		asce = S390_lowcore.user_asce;
203 		pr_cont("user ");
204 		break;
205 	case VDSO_FAULT:
206 		asce = S390_lowcore.vdso_asce;
207 		pr_cont("vdso ");
208 		break;
209 	case GMAP_FAULT:
210 		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
211 		pr_cont("gmap ");
212 		break;
213 	case KERNEL_FAULT:
214 		asce = S390_lowcore.kernel_asce;
215 		pr_cont("kernel ");
216 		break;
217 	default:
218 		unreachable();
219 	}
220 	pr_cont("ASCE.\n");
221 	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
222 }
223 
224 int show_unhandled_signals = 1;
225 
226 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
227 {
228 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
229 		return;
230 	if (!unhandled_signal(current, signr))
231 		return;
232 	if (!printk_ratelimit())
233 		return;
234 	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
235 	       regs->int_code & 0xffff, regs->int_code >> 17);
236 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
237 	printk(KERN_CONT "\n");
238 	if (is_mm_fault)
239 		dump_fault_info(regs);
240 	show_regs(regs);
241 }
242 
243 /*
244  * Send SIGSEGV to task.  This is an external routine
245  * to keep the stack usage of do_page_fault small.
246  */
247 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
248 {
249 	report_user_fault(regs, SIGSEGV, 1);
250 	force_sig_fault(SIGSEGV, si_code,
251 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
252 }
253 
254 const struct exception_table_entry *s390_search_extables(unsigned long addr)
255 {
256 	const struct exception_table_entry *fixup;
257 
258 	fixup = search_extable(__start_dma_ex_table,
259 			       __stop_dma_ex_table - __start_dma_ex_table,
260 			       addr);
261 	if (!fixup)
262 		fixup = search_exception_tables(addr);
263 	return fixup;
264 }
265 
266 static noinline void do_no_context(struct pt_regs *regs)
267 {
268 	const struct exception_table_entry *fixup;
269 
270 	/* Are we prepared to handle this kernel fault?  */
271 	fixup = s390_search_extables(regs->psw.addr);
272 	if (fixup) {
273 		regs->psw.addr = extable_fixup(fixup);
274 		return;
275 	}
276 
277 	/*
278 	 * Oops. The kernel tried to access some bad page. We'll have to
279 	 * terminate things with extreme prejudice.
280 	 */
281 	if (get_fault_type(regs) == KERNEL_FAULT)
282 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
283 		       " in virtual kernel address space\n");
284 	else
285 		printk(KERN_ALERT "Unable to handle kernel paging request"
286 		       " in virtual user address space\n");
287 	dump_fault_info(regs);
288 	die(regs, "Oops");
289 	do_exit(SIGKILL);
290 }
291 
292 static noinline void do_low_address(struct pt_regs *regs)
293 {
294 	/* Low-address protection hit in kernel mode means
295 	   NULL pointer write access in kernel mode.  */
296 	if (regs->psw.mask & PSW_MASK_PSTATE) {
297 		/* Low-address protection hit in user mode 'cannot happen'. */
298 		die (regs, "Low-address protection");
299 		do_exit(SIGKILL);
300 	}
301 
302 	do_no_context(regs);
303 }
304 
305 static noinline void do_sigbus(struct pt_regs *regs)
306 {
307 	/*
308 	 * Send a sigbus, regardless of whether we were in kernel
309 	 * or user mode.
310 	 */
311 	force_sig_fault(SIGBUS, BUS_ADRERR,
312 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
313 }
314 
315 static noinline int signal_return(struct pt_regs *regs)
316 {
317 	u16 instruction;
318 	int rc;
319 
320 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
321 	if (rc)
322 		return rc;
323 	if (instruction == 0x0a77) {
324 		set_pt_regs_flag(regs, PIF_SYSCALL);
325 		regs->int_code = 0x00040077;
326 		return 0;
327 	} else if (instruction == 0x0aad) {
328 		set_pt_regs_flag(regs, PIF_SYSCALL);
329 		regs->int_code = 0x000400ad;
330 		return 0;
331 	}
332 	return -EACCES;
333 }
334 
335 static noinline void do_fault_error(struct pt_regs *regs, int access,
336 					vm_fault_t 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 vm_fault_t do_exception(struct pt_regs *regs, int access)
397 {
398 	struct gmap *gmap;
399 	struct task_struct *tsk;
400 	struct mm_struct *mm;
401 	struct vm_area_struct *vma;
402 	enum fault_type type;
403 	unsigned long trans_exc_code;
404 	unsigned long address;
405 	unsigned int flags;
406 	vm_fault_t fault;
407 
408 	tsk = current;
409 	/*
410 	 * The instruction that caused the program check has
411 	 * been nullified. Don't signal single step via SIGTRAP.
412 	 */
413 	clear_pt_regs_flag(regs, PIF_PER_TRAP);
414 
415 	if (notify_page_fault(regs))
416 		return 0;
417 
418 	mm = tsk->mm;
419 	trans_exc_code = regs->int_parm_long;
420 
421 	/*
422 	 * Verify that the fault happened in user space, that
423 	 * we are not in an interrupt and that there is a
424 	 * user context.
425 	 */
426 	fault = VM_FAULT_BADCONTEXT;
427 	type = get_fault_type(regs);
428 	switch (type) {
429 	case KERNEL_FAULT:
430 		goto out;
431 	case VDSO_FAULT:
432 		fault = VM_FAULT_BADMAP;
433 		goto out;
434 	case USER_FAULT:
435 	case GMAP_FAULT:
436 		if (faulthandler_disabled() || !mm)
437 			goto out;
438 		break;
439 	}
440 
441 	address = trans_exc_code & __FAIL_ADDR_MASK;
442 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
443 	flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
444 	if (user_mode(regs))
445 		flags |= FAULT_FLAG_USER;
446 	if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
447 		flags |= FAULT_FLAG_WRITE;
448 	down_read(&mm->mmap_sem);
449 
450 	gmap = NULL;
451 	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
452 		gmap = (struct gmap *) S390_lowcore.gmap;
453 		current->thread.gmap_addr = address;
454 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
455 		current->thread.gmap_int_code = regs->int_code & 0xffff;
456 		address = __gmap_translate(gmap, address);
457 		if (address == -EFAULT) {
458 			fault = VM_FAULT_BADMAP;
459 			goto out_up;
460 		}
461 		if (gmap->pfault_enabled)
462 			flags |= FAULT_FLAG_RETRY_NOWAIT;
463 	}
464 
465 retry:
466 	fault = VM_FAULT_BADMAP;
467 	vma = find_vma(mm, address);
468 	if (!vma)
469 		goto out_up;
470 
471 	if (unlikely(vma->vm_start > address)) {
472 		if (!(vma->vm_flags & VM_GROWSDOWN))
473 			goto out_up;
474 		if (expand_stack(vma, address))
475 			goto out_up;
476 	}
477 
478 	/*
479 	 * Ok, we have a good vm_area for this memory access, so
480 	 * we can handle it..
481 	 */
482 	fault = VM_FAULT_BADACCESS;
483 	if (unlikely(!(vma->vm_flags & access)))
484 		goto out_up;
485 
486 	if (is_vm_hugetlb_page(vma))
487 		address &= HPAGE_MASK;
488 	/*
489 	 * If for any reason at all we couldn't handle the fault,
490 	 * make sure we exit gracefully rather than endlessly redo
491 	 * the fault.
492 	 */
493 	fault = handle_mm_fault(vma, address, flags);
494 	/* No reason to continue if interrupted by SIGKILL. */
495 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
496 		fault = VM_FAULT_SIGNAL;
497 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
498 			goto out_up;
499 		goto out;
500 	}
501 	if (unlikely(fault & VM_FAULT_ERROR))
502 		goto out_up;
503 
504 	/*
505 	 * Major/minor page fault accounting is only done on the
506 	 * initial attempt. If we go through a retry, it is extremely
507 	 * likely that the page will be found in page cache at that point.
508 	 */
509 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
510 		if (fault & VM_FAULT_MAJOR) {
511 			tsk->maj_flt++;
512 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
513 				      regs, address);
514 		} else {
515 			tsk->min_flt++;
516 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
517 				      regs, address);
518 		}
519 		if (fault & VM_FAULT_RETRY) {
520 			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
521 			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
522 				/* FAULT_FLAG_RETRY_NOWAIT has been set,
523 				 * mmap_sem has not been released */
524 				current->thread.gmap_pfault = 1;
525 				fault = VM_FAULT_PFAULT;
526 				goto out_up;
527 			}
528 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
529 			 * of starvation. */
530 			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
531 				   FAULT_FLAG_RETRY_NOWAIT);
532 			flags |= FAULT_FLAG_TRIED;
533 			down_read(&mm->mmap_sem);
534 			goto retry;
535 		}
536 	}
537 	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
538 		address =  __gmap_link(gmap, current->thread.gmap_addr,
539 				       address);
540 		if (address == -EFAULT) {
541 			fault = VM_FAULT_BADMAP;
542 			goto out_up;
543 		}
544 		if (address == -ENOMEM) {
545 			fault = VM_FAULT_OOM;
546 			goto out_up;
547 		}
548 	}
549 	fault = 0;
550 out_up:
551 	up_read(&mm->mmap_sem);
552 out:
553 	return fault;
554 }
555 
556 void do_protection_exception(struct pt_regs *regs)
557 {
558 	unsigned long trans_exc_code;
559 	int access;
560 	vm_fault_t fault;
561 
562 	trans_exc_code = regs->int_parm_long;
563 	/*
564 	 * Protection exceptions are suppressing, decrement psw address.
565 	 * The exception to this rule are aborted transactions, for these
566 	 * the PSW already points to the correct location.
567 	 */
568 	if (!(regs->int_code & 0x200))
569 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
570 	/*
571 	 * Check for low-address protection.  This needs to be treated
572 	 * as a special case because the translation exception code
573 	 * field is not guaranteed to contain valid data in this case.
574 	 */
575 	if (unlikely(!(trans_exc_code & 4))) {
576 		do_low_address(regs);
577 		return;
578 	}
579 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
580 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
581 					(regs->psw.addr & PAGE_MASK);
582 		access = VM_EXEC;
583 		fault = VM_FAULT_BADACCESS;
584 	} else {
585 		access = VM_WRITE;
586 		fault = do_exception(regs, access);
587 	}
588 	if (unlikely(fault))
589 		do_fault_error(regs, access, fault);
590 }
591 NOKPROBE_SYMBOL(do_protection_exception);
592 
593 void do_dat_exception(struct pt_regs *regs)
594 {
595 	int access;
596 	vm_fault_t fault;
597 
598 	access = VM_READ | VM_EXEC | VM_WRITE;
599 	fault = do_exception(regs, access);
600 	if (unlikely(fault))
601 		do_fault_error(regs, access, fault);
602 }
603 NOKPROBE_SYMBOL(do_dat_exception);
604 
605 #ifdef CONFIG_PFAULT
606 /*
607  * 'pfault' pseudo page faults routines.
608  */
609 static int pfault_disable;
610 
611 static int __init nopfault(char *str)
612 {
613 	pfault_disable = 1;
614 	return 1;
615 }
616 
617 __setup("nopfault", nopfault);
618 
619 struct pfault_refbk {
620 	u16 refdiagc;
621 	u16 reffcode;
622 	u16 refdwlen;
623 	u16 refversn;
624 	u64 refgaddr;
625 	u64 refselmk;
626 	u64 refcmpmk;
627 	u64 reserved;
628 } __attribute__ ((packed, aligned(8)));
629 
630 static struct pfault_refbk pfault_init_refbk = {
631 	.refdiagc = 0x258,
632 	.reffcode = 0,
633 	.refdwlen = 5,
634 	.refversn = 2,
635 	.refgaddr = __LC_LPP,
636 	.refselmk = 1ULL << 48,
637 	.refcmpmk = 1ULL << 48,
638 	.reserved = __PF_RES_FIELD
639 };
640 
641 int pfault_init(void)
642 {
643         int rc;
644 
645 	if (pfault_disable)
646 		return -1;
647 	diag_stat_inc(DIAG_STAT_X258);
648 	asm volatile(
649 		"	diag	%1,%0,0x258\n"
650 		"0:	j	2f\n"
651 		"1:	la	%0,8\n"
652 		"2:\n"
653 		EX_TABLE(0b,1b)
654 		: "=d" (rc)
655 		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
656         return rc;
657 }
658 
659 static struct pfault_refbk pfault_fini_refbk = {
660 	.refdiagc = 0x258,
661 	.reffcode = 1,
662 	.refdwlen = 5,
663 	.refversn = 2,
664 };
665 
666 void pfault_fini(void)
667 {
668 
669 	if (pfault_disable)
670 		return;
671 	diag_stat_inc(DIAG_STAT_X258);
672 	asm volatile(
673 		"	diag	%0,0,0x258\n"
674 		"0:	nopr	%%r7\n"
675 		EX_TABLE(0b,0b)
676 		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
677 }
678 
679 static DEFINE_SPINLOCK(pfault_lock);
680 static LIST_HEAD(pfault_list);
681 
682 #define PF_COMPLETE	0x0080
683 
684 /*
685  * The mechanism of our pfault code: if Linux is running as guest, runs a user
686  * space process and the user space process accesses a page that the host has
687  * paged out we get a pfault interrupt.
688  *
689  * This allows us, within the guest, to schedule a different process. Without
690  * this mechanism the host would have to suspend the whole virtual cpu until
691  * the page has been paged in.
692  *
693  * So when we get such an interrupt then we set the state of the current task
694  * to uninterruptible and also set the need_resched flag. Both happens within
695  * interrupt context(!). If we later on want to return to user space we
696  * recognize the need_resched flag and then call schedule().  It's not very
697  * obvious how this works...
698  *
699  * Of course we have a lot of additional fun with the completion interrupt (->
700  * host signals that a page of a process has been paged in and the process can
701  * continue to run). This interrupt can arrive on any cpu and, since we have
702  * virtual cpus, actually appear before the interrupt that signals that a page
703  * is missing.
704  */
705 static void pfault_interrupt(struct ext_code ext_code,
706 			     unsigned int param32, unsigned long param64)
707 {
708 	struct task_struct *tsk;
709 	__u16 subcode;
710 	pid_t pid;
711 
712 	/*
713 	 * Get the external interruption subcode & pfault initial/completion
714 	 * signal bit. VM stores this in the 'cpu address' field associated
715 	 * with the external interrupt.
716 	 */
717 	subcode = ext_code.subcode;
718 	if ((subcode & 0xff00) != __SUBCODE_MASK)
719 		return;
720 	inc_irq_stat(IRQEXT_PFL);
721 	/* Get the token (= pid of the affected task). */
722 	pid = param64 & LPP_PID_MASK;
723 	rcu_read_lock();
724 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
725 	if (tsk)
726 		get_task_struct(tsk);
727 	rcu_read_unlock();
728 	if (!tsk)
729 		return;
730 	spin_lock(&pfault_lock);
731 	if (subcode & PF_COMPLETE) {
732 		/* signal bit is set -> a page has been swapped in by VM */
733 		if (tsk->thread.pfault_wait == 1) {
734 			/* Initial interrupt was faster than the completion
735 			 * interrupt. pfault_wait is valid. Set pfault_wait
736 			 * back to zero and wake up the process. This can
737 			 * safely be done because the task is still sleeping
738 			 * and can't produce new pfaults. */
739 			tsk->thread.pfault_wait = 0;
740 			list_del(&tsk->thread.list);
741 			wake_up_process(tsk);
742 			put_task_struct(tsk);
743 		} else {
744 			/* Completion interrupt was faster than initial
745 			 * interrupt. Set pfault_wait to -1 so the initial
746 			 * interrupt doesn't put the task to sleep.
747 			 * If the task is not running, ignore the completion
748 			 * interrupt since it must be a leftover of a PFAULT
749 			 * CANCEL operation which didn't remove all pending
750 			 * completion interrupts. */
751 			if (tsk->state == TASK_RUNNING)
752 				tsk->thread.pfault_wait = -1;
753 		}
754 	} else {
755 		/* signal bit not set -> a real page is missing. */
756 		if (WARN_ON_ONCE(tsk != current))
757 			goto out;
758 		if (tsk->thread.pfault_wait == 1) {
759 			/* Already on the list with a reference: put to sleep */
760 			goto block;
761 		} else if (tsk->thread.pfault_wait == -1) {
762 			/* Completion interrupt was faster than the initial
763 			 * interrupt (pfault_wait == -1). Set pfault_wait
764 			 * back to zero and exit. */
765 			tsk->thread.pfault_wait = 0;
766 		} else {
767 			/* Initial interrupt arrived before completion
768 			 * interrupt. Let the task sleep.
769 			 * An extra task reference is needed since a different
770 			 * cpu may set the task state to TASK_RUNNING again
771 			 * before the scheduler is reached. */
772 			get_task_struct(tsk);
773 			tsk->thread.pfault_wait = 1;
774 			list_add(&tsk->thread.list, &pfault_list);
775 block:
776 			/* Since this must be a userspace fault, there
777 			 * is no kernel task state to trample. Rely on the
778 			 * return to userspace schedule() to block. */
779 			__set_current_state(TASK_UNINTERRUPTIBLE);
780 			set_tsk_need_resched(tsk);
781 			set_preempt_need_resched();
782 		}
783 	}
784 out:
785 	spin_unlock(&pfault_lock);
786 	put_task_struct(tsk);
787 }
788 
789 static int pfault_cpu_dead(unsigned int cpu)
790 {
791 	struct thread_struct *thread, *next;
792 	struct task_struct *tsk;
793 
794 	spin_lock_irq(&pfault_lock);
795 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
796 		thread->pfault_wait = 0;
797 		list_del(&thread->list);
798 		tsk = container_of(thread, struct task_struct, thread);
799 		wake_up_process(tsk);
800 		put_task_struct(tsk);
801 	}
802 	spin_unlock_irq(&pfault_lock);
803 	return 0;
804 }
805 
806 static int __init pfault_irq_init(void)
807 {
808 	int rc;
809 
810 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
811 	if (rc)
812 		goto out_extint;
813 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
814 	if (rc)
815 		goto out_pfault;
816 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
817 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
818 				  NULL, pfault_cpu_dead);
819 	return 0;
820 
821 out_pfault:
822 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
823 out_extint:
824 	pfault_disable = 1;
825 	return rc;
826 }
827 early_initcall(pfault_irq_init);
828 
829 #endif /* CONFIG_PFAULT */
830