• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/gmap.h>
37 #include <asm/irq.h>
38 #include <asm/mmu_context.h>
39 #include <asm/facility.h>
40 #include <asm/uv.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	((__force vm_fault_t) 0x010000)
48 #define VM_FAULT_BADMAP		((__force vm_fault_t) 0x020000)
49 #define VM_FAULT_BADACCESS	((__force vm_fault_t) 0x040000)
50 #define VM_FAULT_SIGNAL		((__force vm_fault_t) 0x080000)
51 #define VM_FAULT_PFAULT		((__force vm_fault_t) 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 
fault_init(void)62 static int __init fault_init(void)
63 {
64 	if (test_facility(75))
65 		store_indication = 0xc00;
66 	return 0;
67 }
68 early_initcall(fault_init);
69 
70 /*
71  * Find out which address space caused the exception.
72  */
get_fault_type(struct pt_regs * regs)73 static enum fault_type get_fault_type(struct pt_regs *regs)
74 {
75 	unsigned long trans_exc_code;
76 
77 	trans_exc_code = regs->int_parm_long & 3;
78 	if (likely(trans_exc_code == 0)) {
79 		/* primary space exception */
80 		if (IS_ENABLED(CONFIG_PGSTE) &&
81 		    test_pt_regs_flag(regs, PIF_GUEST_FAULT))
82 			return GMAP_FAULT;
83 		if (current->thread.mm_segment == USER_DS)
84 			return USER_FAULT;
85 		return KERNEL_FAULT;
86 	}
87 	if (trans_exc_code == 2) {
88 		/* secondary space exception */
89 		if (current->thread.mm_segment & 1) {
90 			if (current->thread.mm_segment == USER_DS_SACF)
91 				return USER_FAULT;
92 			return KERNEL_FAULT;
93 		}
94 		return VDSO_FAULT;
95 	}
96 	if (trans_exc_code == 1) {
97 		/* access register mode, not used in the kernel */
98 		return USER_FAULT;
99 	}
100 	/* home space exception -> access via kernel ASCE */
101 	return KERNEL_FAULT;
102 }
103 
bad_address(void * p)104 static int bad_address(void *p)
105 {
106 	unsigned long dummy;
107 
108 	return get_kernel_nofault(dummy, (unsigned long *)p);
109 }
110 
dump_pagetable(unsigned long asce,unsigned long address)111 static void dump_pagetable(unsigned long asce, unsigned long address)
112 {
113 	unsigned long *table = __va(asce & _ASCE_ORIGIN);
114 
115 	pr_alert("AS:%016lx ", asce);
116 	switch (asce & _ASCE_TYPE_MASK) {
117 	case _ASCE_TYPE_REGION1:
118 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
119 		if (bad_address(table))
120 			goto bad;
121 		pr_cont("R1:%016lx ", *table);
122 		if (*table & _REGION_ENTRY_INVALID)
123 			goto out;
124 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
125 		fallthrough;
126 	case _ASCE_TYPE_REGION2:
127 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
128 		if (bad_address(table))
129 			goto bad;
130 		pr_cont("R2:%016lx ", *table);
131 		if (*table & _REGION_ENTRY_INVALID)
132 			goto out;
133 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
134 		fallthrough;
135 	case _ASCE_TYPE_REGION3:
136 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
137 		if (bad_address(table))
138 			goto bad;
139 		pr_cont("R3:%016lx ", *table);
140 		if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
141 			goto out;
142 		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
143 		fallthrough;
144 	case _ASCE_TYPE_SEGMENT:
145 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
146 		if (bad_address(table))
147 			goto bad;
148 		pr_cont("S:%016lx ", *table);
149 		if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
150 			goto out;
151 		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
152 	}
153 	table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
154 	if (bad_address(table))
155 		goto bad;
156 	pr_cont("P:%016lx ", *table);
157 out:
158 	pr_cont("\n");
159 	return;
160 bad:
161 	pr_cont("BAD\n");
162 }
163 
dump_fault_info(struct pt_regs * regs)164 static void dump_fault_info(struct pt_regs *regs)
165 {
166 	unsigned long asce;
167 
168 	pr_alert("Failing address: %016lx TEID: %016lx\n",
169 		 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
170 	pr_alert("Fault in ");
171 	switch (regs->int_parm_long & 3) {
172 	case 3:
173 		pr_cont("home space ");
174 		break;
175 	case 2:
176 		pr_cont("secondary space ");
177 		break;
178 	case 1:
179 		pr_cont("access register ");
180 		break;
181 	case 0:
182 		pr_cont("primary space ");
183 		break;
184 	}
185 	pr_cont("mode while using ");
186 	switch (get_fault_type(regs)) {
187 	case USER_FAULT:
188 		asce = S390_lowcore.user_asce;
189 		pr_cont("user ");
190 		break;
191 	case VDSO_FAULT:
192 		asce = S390_lowcore.vdso_asce;
193 		pr_cont("vdso ");
194 		break;
195 	case GMAP_FAULT:
196 		asce = ((struct gmap *) S390_lowcore.gmap)->asce;
197 		pr_cont("gmap ");
198 		break;
199 	case KERNEL_FAULT:
200 		asce = S390_lowcore.kernel_asce;
201 		pr_cont("kernel ");
202 		break;
203 	default:
204 		unreachable();
205 	}
206 	pr_cont("ASCE.\n");
207 	dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
208 }
209 
210 int show_unhandled_signals = 1;
211 
report_user_fault(struct pt_regs * regs,long signr,int is_mm_fault)212 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
213 {
214 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
215 		return;
216 	if (!unhandled_signal(current, signr))
217 		return;
218 	if (!printk_ratelimit())
219 		return;
220 	printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
221 	       regs->int_code & 0xffff, regs->int_code >> 17);
222 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
223 	printk(KERN_CONT "\n");
224 	if (is_mm_fault)
225 		dump_fault_info(regs);
226 	show_regs(regs);
227 }
228 
229 /*
230  * Send SIGSEGV to task.  This is an external routine
231  * to keep the stack usage of do_page_fault small.
232  */
do_sigsegv(struct pt_regs * regs,int si_code)233 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
234 {
235 	report_user_fault(regs, SIGSEGV, 1);
236 	force_sig_fault(SIGSEGV, si_code,
237 			(void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
238 }
239 
s390_search_extables(unsigned long addr)240 const struct exception_table_entry *s390_search_extables(unsigned long addr)
241 {
242 	const struct exception_table_entry *fixup;
243 
244 	fixup = search_extable(__start_dma_ex_table,
245 			       __stop_dma_ex_table - __start_dma_ex_table,
246 			       addr);
247 	if (!fixup)
248 		fixup = search_exception_tables(addr);
249 	return fixup;
250 }
251 
do_no_context(struct pt_regs * regs)252 static noinline void do_no_context(struct pt_regs *regs)
253 {
254 	const struct exception_table_entry *fixup;
255 
256 	/* Are we prepared to handle this kernel fault?  */
257 	fixup = s390_search_extables(regs->psw.addr);
258 	if (fixup && ex_handle(fixup, regs))
259 		return;
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 
do_low_address(struct pt_regs * regs)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 
do_sigbus(struct pt_regs * regs)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 }
298 
signal_return(struct pt_regs * regs)299 static noinline int signal_return(struct pt_regs *regs)
300 {
301 	u16 instruction;
302 	int rc;
303 
304 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
305 	if (rc)
306 		return rc;
307 	if (instruction == 0x0a77) {
308 		set_pt_regs_flag(regs, PIF_SYSCALL);
309 		regs->int_code = 0x00040077;
310 		return 0;
311 	} else if (instruction == 0x0aad) {
312 		set_pt_regs_flag(regs, PIF_SYSCALL);
313 		regs->int_code = 0x000400ad;
314 		return 0;
315 	}
316 	return -EACCES;
317 }
318 
do_fault_error(struct pt_regs * regs,int access,vm_fault_t fault)319 static noinline void do_fault_error(struct pt_regs *regs, int access,
320 					vm_fault_t fault)
321 {
322 	int si_code;
323 
324 	switch (fault) {
325 	case VM_FAULT_BADACCESS:
326 		if (access == VM_EXEC && signal_return(regs) == 0)
327 			break;
328 		fallthrough;
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 		fallthrough;
339 	case VM_FAULT_BADCONTEXT:
340 	case VM_FAULT_PFAULT:
341 		do_no_context(regs);
342 		break;
343 	case VM_FAULT_SIGNAL:
344 		if (!user_mode(regs))
345 			do_no_context(regs);
346 		break;
347 	default: /* fault & VM_FAULT_ERROR */
348 		if (fault & VM_FAULT_OOM) {
349 			if (!user_mode(regs))
350 				do_no_context(regs);
351 			else
352 				pagefault_out_of_memory();
353 		} else if (fault & VM_FAULT_SIGSEGV) {
354 			/* Kernel mode? Handle exceptions or die */
355 			if (!user_mode(regs))
356 				do_no_context(regs);
357 			else
358 				do_sigsegv(regs, SEGV_MAPERR);
359 		} else if (fault & VM_FAULT_SIGBUS) {
360 			/* Kernel mode? Handle exceptions or die */
361 			if (!user_mode(regs))
362 				do_no_context(regs);
363 			else
364 				do_sigbus(regs);
365 		} else
366 			BUG();
367 		break;
368 	}
369 }
370 
371 /*
372  * This routine handles page faults.  It determines the address,
373  * and the problem, and then passes it off to one of the appropriate
374  * routines.
375  *
376  * interruption code (int_code):
377  *   04       Protection           ->  Write-Protection  (suppression)
378  *   10       Segment translation  ->  Not present       (nullification)
379  *   11       Page translation     ->  Not present       (nullification)
380  *   3b       Region third trans.  ->  Not present       (nullification)
381  */
do_exception(struct pt_regs * regs,int access)382 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
383 {
384 	struct gmap *gmap;
385 	struct task_struct *tsk;
386 	struct mm_struct *mm;
387 	struct vm_area_struct *vma;
388 	enum fault_type type;
389 	unsigned long trans_exc_code;
390 	unsigned long address;
391 	unsigned int flags;
392 	vm_fault_t fault;
393 
394 	tsk = current;
395 	/*
396 	 * The instruction that caused the program check has
397 	 * been nullified. Don't signal single step via SIGTRAP.
398 	 */
399 	clear_pt_regs_flag(regs, PIF_PER_TRAP);
400 
401 	if (kprobe_page_fault(regs, 14))
402 		return 0;
403 
404 	mm = tsk->mm;
405 	trans_exc_code = regs->int_parm_long;
406 
407 	/*
408 	 * Verify that the fault happened in user space, that
409 	 * we are not in an interrupt and that there is a
410 	 * user context.
411 	 */
412 	fault = VM_FAULT_BADCONTEXT;
413 	type = get_fault_type(regs);
414 	switch (type) {
415 	case KERNEL_FAULT:
416 		goto out;
417 	case VDSO_FAULT:
418 		fault = VM_FAULT_BADMAP;
419 		goto out;
420 	case USER_FAULT:
421 	case GMAP_FAULT:
422 		if (faulthandler_disabled() || !mm)
423 			goto out;
424 		break;
425 	}
426 
427 	address = trans_exc_code & __FAIL_ADDR_MASK;
428 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
429 	flags = FAULT_FLAG_DEFAULT;
430 	if (user_mode(regs))
431 		flags |= FAULT_FLAG_USER;
432 	if ((trans_exc_code & store_indication) == 0x400)
433 		access = VM_WRITE;
434 	if (access == VM_WRITE)
435 		flags |= FAULT_FLAG_WRITE;
436 	mmap_read_lock(mm);
437 
438 	gmap = NULL;
439 	if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
440 		gmap = (struct gmap *) S390_lowcore.gmap;
441 		current->thread.gmap_addr = address;
442 		current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
443 		current->thread.gmap_int_code = regs->int_code & 0xffff;
444 		address = __gmap_translate(gmap, address);
445 		if (address == -EFAULT) {
446 			fault = VM_FAULT_BADMAP;
447 			goto out_up;
448 		}
449 		if (gmap->pfault_enabled)
450 			flags |= FAULT_FLAG_RETRY_NOWAIT;
451 	}
452 
453 retry:
454 	fault = VM_FAULT_BADMAP;
455 	vma = find_vma(mm, address);
456 	if (!vma)
457 		goto out_up;
458 
459 	if (unlikely(vma->vm_start > address)) {
460 		if (!(vma->vm_flags & VM_GROWSDOWN))
461 			goto out_up;
462 		if (expand_stack(vma, address))
463 			goto out_up;
464 	}
465 
466 	/*
467 	 * Ok, we have a good vm_area for this memory access, so
468 	 * we can handle it..
469 	 */
470 	fault = VM_FAULT_BADACCESS;
471 	if (unlikely(!(vma->vm_flags & access)))
472 		goto out_up;
473 
474 	if (is_vm_hugetlb_page(vma))
475 		address &= HPAGE_MASK;
476 	/*
477 	 * If for any reason at all we couldn't handle the fault,
478 	 * make sure we exit gracefully rather than endlessly redo
479 	 * the fault.
480 	 */
481 	fault = handle_mm_fault(vma, address, flags, regs);
482 	if (fault_signal_pending(fault, regs)) {
483 		fault = VM_FAULT_SIGNAL;
484 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
485 			goto out_up;
486 		goto out;
487 	}
488 	if (unlikely(fault & VM_FAULT_ERROR))
489 		goto out_up;
490 
491 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
492 		if (fault & VM_FAULT_RETRY) {
493 			if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
494 			    (flags & FAULT_FLAG_RETRY_NOWAIT)) {
495 				/* FAULT_FLAG_RETRY_NOWAIT has been set,
496 				 * mmap_lock has not been released */
497 				current->thread.gmap_pfault = 1;
498 				fault = VM_FAULT_PFAULT;
499 				goto out_up;
500 			}
501 			flags &= ~FAULT_FLAG_RETRY_NOWAIT;
502 			flags |= FAULT_FLAG_TRIED;
503 			mmap_read_lock(mm);
504 			goto retry;
505 		}
506 	}
507 	if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
508 		address =  __gmap_link(gmap, current->thread.gmap_addr,
509 				       address);
510 		if (address == -EFAULT) {
511 			fault = VM_FAULT_BADMAP;
512 			goto out_up;
513 		}
514 		if (address == -ENOMEM) {
515 			fault = VM_FAULT_OOM;
516 			goto out_up;
517 		}
518 	}
519 	fault = 0;
520 out_up:
521 	mmap_read_unlock(mm);
522 out:
523 	return fault;
524 }
525 
do_protection_exception(struct pt_regs * regs)526 void do_protection_exception(struct pt_regs *regs)
527 {
528 	unsigned long trans_exc_code;
529 	int access;
530 	vm_fault_t fault;
531 
532 	trans_exc_code = regs->int_parm_long;
533 	/*
534 	 * Protection exceptions are suppressing, decrement psw address.
535 	 * The exception to this rule are aborted transactions, for these
536 	 * the PSW already points to the correct location.
537 	 */
538 	if (!(regs->int_code & 0x200))
539 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
540 	/*
541 	 * Check for low-address protection.  This needs to be treated
542 	 * as a special case because the translation exception code
543 	 * field is not guaranteed to contain valid data in this case.
544 	 */
545 	if (unlikely(!(trans_exc_code & 4))) {
546 		do_low_address(regs);
547 		return;
548 	}
549 	if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
550 		regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
551 					(regs->psw.addr & PAGE_MASK);
552 		access = VM_EXEC;
553 		fault = VM_FAULT_BADACCESS;
554 	} else {
555 		access = VM_WRITE;
556 		fault = do_exception(regs, access);
557 	}
558 	if (unlikely(fault))
559 		do_fault_error(regs, access, fault);
560 }
561 NOKPROBE_SYMBOL(do_protection_exception);
562 
do_dat_exception(struct pt_regs * regs)563 void do_dat_exception(struct pt_regs *regs)
564 {
565 	int access;
566 	vm_fault_t fault;
567 
568 	access = VM_ACCESS_FLAGS;
569 	fault = do_exception(regs, access);
570 	if (unlikely(fault))
571 		do_fault_error(regs, access, fault);
572 }
573 NOKPROBE_SYMBOL(do_dat_exception);
574 
575 #ifdef CONFIG_PFAULT
576 /*
577  * 'pfault' pseudo page faults routines.
578  */
579 static int pfault_disable;
580 
nopfault(char * str)581 static int __init nopfault(char *str)
582 {
583 	pfault_disable = 1;
584 	return 1;
585 }
586 
587 __setup("nopfault", nopfault);
588 
589 struct pfault_refbk {
590 	u16 refdiagc;
591 	u16 reffcode;
592 	u16 refdwlen;
593 	u16 refversn;
594 	u64 refgaddr;
595 	u64 refselmk;
596 	u64 refcmpmk;
597 	u64 reserved;
598 } __attribute__ ((packed, aligned(8)));
599 
600 static struct pfault_refbk pfault_init_refbk = {
601 	.refdiagc = 0x258,
602 	.reffcode = 0,
603 	.refdwlen = 5,
604 	.refversn = 2,
605 	.refgaddr = __LC_LPP,
606 	.refselmk = 1ULL << 48,
607 	.refcmpmk = 1ULL << 48,
608 	.reserved = __PF_RES_FIELD
609 };
610 
pfault_init(void)611 int pfault_init(void)
612 {
613         int rc;
614 
615 	if (pfault_disable)
616 		return -1;
617 	diag_stat_inc(DIAG_STAT_X258);
618 	asm volatile(
619 		"	diag	%1,%0,0x258\n"
620 		"0:	j	2f\n"
621 		"1:	la	%0,8\n"
622 		"2:\n"
623 		EX_TABLE(0b,1b)
624 		: "=d" (rc)
625 		: "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
626         return rc;
627 }
628 
629 static struct pfault_refbk pfault_fini_refbk = {
630 	.refdiagc = 0x258,
631 	.reffcode = 1,
632 	.refdwlen = 5,
633 	.refversn = 2,
634 };
635 
pfault_fini(void)636 void pfault_fini(void)
637 {
638 
639 	if (pfault_disable)
640 		return;
641 	diag_stat_inc(DIAG_STAT_X258);
642 	asm volatile(
643 		"	diag	%0,0,0x258\n"
644 		"0:	nopr	%%r7\n"
645 		EX_TABLE(0b,0b)
646 		: : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
647 }
648 
649 static DEFINE_SPINLOCK(pfault_lock);
650 static LIST_HEAD(pfault_list);
651 
652 #define PF_COMPLETE	0x0080
653 
654 /*
655  * The mechanism of our pfault code: if Linux is running as guest, runs a user
656  * space process and the user space process accesses a page that the host has
657  * paged out we get a pfault interrupt.
658  *
659  * This allows us, within the guest, to schedule a different process. Without
660  * this mechanism the host would have to suspend the whole virtual cpu until
661  * the page has been paged in.
662  *
663  * So when we get such an interrupt then we set the state of the current task
664  * to uninterruptible and also set the need_resched flag. Both happens within
665  * interrupt context(!). If we later on want to return to user space we
666  * recognize the need_resched flag and then call schedule().  It's not very
667  * obvious how this works...
668  *
669  * Of course we have a lot of additional fun with the completion interrupt (->
670  * host signals that a page of a process has been paged in and the process can
671  * continue to run). This interrupt can arrive on any cpu and, since we have
672  * virtual cpus, actually appear before the interrupt that signals that a page
673  * is missing.
674  */
pfault_interrupt(struct ext_code ext_code,unsigned int param32,unsigned long param64)675 static void pfault_interrupt(struct ext_code ext_code,
676 			     unsigned int param32, unsigned long param64)
677 {
678 	struct task_struct *tsk;
679 	__u16 subcode;
680 	pid_t pid;
681 
682 	/*
683 	 * Get the external interruption subcode & pfault initial/completion
684 	 * signal bit. VM stores this in the 'cpu address' field associated
685 	 * with the external interrupt.
686 	 */
687 	subcode = ext_code.subcode;
688 	if ((subcode & 0xff00) != __SUBCODE_MASK)
689 		return;
690 	inc_irq_stat(IRQEXT_PFL);
691 	/* Get the token (= pid of the affected task). */
692 	pid = param64 & LPP_PID_MASK;
693 	rcu_read_lock();
694 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
695 	if (tsk)
696 		get_task_struct(tsk);
697 	rcu_read_unlock();
698 	if (!tsk)
699 		return;
700 	spin_lock(&pfault_lock);
701 	if (subcode & PF_COMPLETE) {
702 		/* signal bit is set -> a page has been swapped in by VM */
703 		if (tsk->thread.pfault_wait == 1) {
704 			/* Initial interrupt was faster than the completion
705 			 * interrupt. pfault_wait is valid. Set pfault_wait
706 			 * back to zero and wake up the process. This can
707 			 * safely be done because the task is still sleeping
708 			 * and can't produce new pfaults. */
709 			tsk->thread.pfault_wait = 0;
710 			list_del(&tsk->thread.list);
711 			wake_up_process(tsk);
712 			put_task_struct(tsk);
713 		} else {
714 			/* Completion interrupt was faster than initial
715 			 * interrupt. Set pfault_wait to -1 so the initial
716 			 * interrupt doesn't put the task to sleep.
717 			 * If the task is not running, ignore the completion
718 			 * interrupt since it must be a leftover of a PFAULT
719 			 * CANCEL operation which didn't remove all pending
720 			 * completion interrupts. */
721 			if (tsk->state == TASK_RUNNING)
722 				tsk->thread.pfault_wait = -1;
723 		}
724 	} else {
725 		/* signal bit not set -> a real page is missing. */
726 		if (WARN_ON_ONCE(tsk != current))
727 			goto out;
728 		if (tsk->thread.pfault_wait == 1) {
729 			/* Already on the list with a reference: put to sleep */
730 			goto block;
731 		} else if (tsk->thread.pfault_wait == -1) {
732 			/* Completion interrupt was faster than the initial
733 			 * interrupt (pfault_wait == -1). Set pfault_wait
734 			 * back to zero and exit. */
735 			tsk->thread.pfault_wait = 0;
736 		} else {
737 			/* Initial interrupt arrived before completion
738 			 * interrupt. Let the task sleep.
739 			 * An extra task reference is needed since a different
740 			 * cpu may set the task state to TASK_RUNNING again
741 			 * before the scheduler is reached. */
742 			get_task_struct(tsk);
743 			tsk->thread.pfault_wait = 1;
744 			list_add(&tsk->thread.list, &pfault_list);
745 block:
746 			/* Since this must be a userspace fault, there
747 			 * is no kernel task state to trample. Rely on the
748 			 * return to userspace schedule() to block. */
749 			__set_current_state(TASK_UNINTERRUPTIBLE);
750 			set_tsk_need_resched(tsk);
751 			set_preempt_need_resched();
752 		}
753 	}
754 out:
755 	spin_unlock(&pfault_lock);
756 	put_task_struct(tsk);
757 }
758 
pfault_cpu_dead(unsigned int cpu)759 static int pfault_cpu_dead(unsigned int cpu)
760 {
761 	struct thread_struct *thread, *next;
762 	struct task_struct *tsk;
763 
764 	spin_lock_irq(&pfault_lock);
765 	list_for_each_entry_safe(thread, next, &pfault_list, list) {
766 		thread->pfault_wait = 0;
767 		list_del(&thread->list);
768 		tsk = container_of(thread, struct task_struct, thread);
769 		wake_up_process(tsk);
770 		put_task_struct(tsk);
771 	}
772 	spin_unlock_irq(&pfault_lock);
773 	return 0;
774 }
775 
pfault_irq_init(void)776 static int __init pfault_irq_init(void)
777 {
778 	int rc;
779 
780 	rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
781 	if (rc)
782 		goto out_extint;
783 	rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
784 	if (rc)
785 		goto out_pfault;
786 	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
787 	cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
788 				  NULL, pfault_cpu_dead);
789 	return 0;
790 
791 out_pfault:
792 	unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
793 out_extint:
794 	pfault_disable = 1;
795 	return rc;
796 }
797 early_initcall(pfault_irq_init);
798 
799 #endif /* CONFIG_PFAULT */
800 
801 #if IS_ENABLED(CONFIG_PGSTE)
do_secure_storage_access(struct pt_regs * regs)802 void do_secure_storage_access(struct pt_regs *regs)
803 {
804 	unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
805 	struct vm_area_struct *vma;
806 	struct mm_struct *mm;
807 	struct page *page;
808 	int rc;
809 
810 	/*
811 	 * bit 61 tells us if the address is valid, if it's not we
812 	 * have a major problem and should stop the kernel or send a
813 	 * SIGSEGV to the process. Unfortunately bit 61 is not
814 	 * reliable without the misc UV feature so we need to check
815 	 * for that as well.
816 	 */
817 	if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
818 	    !test_bit_inv(61, &regs->int_parm_long)) {
819 		/*
820 		 * When this happens, userspace did something that it
821 		 * was not supposed to do, e.g. branching into secure
822 		 * memory. Trigger a segmentation fault.
823 		 */
824 		if (user_mode(regs)) {
825 			send_sig(SIGSEGV, current, 0);
826 			return;
827 		}
828 
829 		/*
830 		 * The kernel should never run into this case and we
831 		 * have no way out of this situation.
832 		 */
833 		panic("Unexpected PGM 0x3d with TEID bit 61=0");
834 	}
835 
836 	switch (get_fault_type(regs)) {
837 	case USER_FAULT:
838 		mm = current->mm;
839 		mmap_read_lock(mm);
840 		vma = find_vma(mm, addr);
841 		if (!vma) {
842 			mmap_read_unlock(mm);
843 			do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
844 			break;
845 		}
846 		page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
847 		if (IS_ERR_OR_NULL(page)) {
848 			mmap_read_unlock(mm);
849 			break;
850 		}
851 		if (arch_make_page_accessible(page))
852 			send_sig(SIGSEGV, current, 0);
853 		put_page(page);
854 		mmap_read_unlock(mm);
855 		break;
856 	case KERNEL_FAULT:
857 		page = phys_to_page(addr);
858 		if (unlikely(!try_get_page(page)))
859 			break;
860 		rc = arch_make_page_accessible(page);
861 		put_page(page);
862 		if (rc)
863 			BUG();
864 		break;
865 	case VDSO_FAULT:
866 	case GMAP_FAULT:
867 	default:
868 		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
869 		WARN_ON_ONCE(1);
870 	}
871 }
872 NOKPROBE_SYMBOL(do_secure_storage_access);
873 
do_non_secure_storage_access(struct pt_regs * regs)874 void do_non_secure_storage_access(struct pt_regs *regs)
875 {
876 	unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
877 	struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
878 
879 	if (get_fault_type(regs) != GMAP_FAULT) {
880 		do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
881 		WARN_ON_ONCE(1);
882 		return;
883 	}
884 
885 	if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
886 		send_sig(SIGSEGV, current, 0);
887 }
888 NOKPROBE_SYMBOL(do_non_secure_storage_access);
889 
do_secure_storage_violation(struct pt_regs * regs)890 void do_secure_storage_violation(struct pt_regs *regs)
891 {
892 	/*
893 	 * Either KVM messed up the secure guest mapping or the same
894 	 * page is mapped into multiple secure guests.
895 	 *
896 	 * This exception is only triggered when a guest 2 is running
897 	 * and can therefore never occur in kernel context.
898 	 */
899 	printk_ratelimited(KERN_WARNING
900 			   "Secure storage violation in task: %s, pid %d\n",
901 			   current->comm, current->pid);
902 	send_sig(SIGSEGV, current, 0);
903 }
904 
905 #else
do_secure_storage_access(struct pt_regs * regs)906 void do_secure_storage_access(struct pt_regs *regs)
907 {
908 	default_trap_handler(regs);
909 }
910 
do_non_secure_storage_access(struct pt_regs * regs)911 void do_non_secure_storage_access(struct pt_regs *regs)
912 {
913 	default_trap_handler(regs);
914 }
915 
do_secure_storage_violation(struct pt_regs * regs)916 void do_secure_storage_violation(struct pt_regs *regs)
917 {
918 	default_trap_handler(regs);
919 }
920 #endif
921