1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * S390 version
4 * Copyright IBM Corp. 1999
5 * Author(s): Hartmut Penner (hp@de.ibm.com)
6 * Ulrich Weigand (uweigand@de.ibm.com)
7 *
8 * Derived from "arch/i386/mm/fault.c"
9 * Copyright (C) 1995 Linus Torvalds
10 */
11
12 #include <linux/kernel_stat.h>
13 #include <linux/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/types.h>
21 #include <linux/ptrace.h>
22 #include <linux/mman.h>
23 #include <linux/mm.h>
24 #include <linux/compat.h>
25 #include <linux/smp.h>
26 #include <linux/kdebug.h>
27 #include <linux/init.h>
28 #include <linux/console.h>
29 #include <linux/extable.h>
30 #include <linux/hardirq.h>
31 #include <linux/kprobes.h>
32 #include <linux/uaccess.h>
33 #include <linux/hugetlb.h>
34 #include <linux/kfence.h>
35 #include <asm/asm-offsets.h>
36 #include <asm/diag.h>
37 #include <asm/gmap.h>
38 #include <asm/irq.h>
39 #include <asm/mmu_context.h>
40 #include <asm/facility.h>
41 #include <asm/uv.h>
42 #include "../kernel/entry.h"
43
44 #define __FAIL_ADDR_MASK -4096L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47
48 #define VM_FAULT_BADCONTEXT ((__force vm_fault_t) 0x010000)
49 #define VM_FAULT_BADMAP ((__force vm_fault_t) 0x020000)
50 #define VM_FAULT_BADACCESS ((__force vm_fault_t) 0x040000)
51 #define VM_FAULT_SIGNAL ((__force vm_fault_t) 0x080000)
52 #define VM_FAULT_PFAULT ((__force vm_fault_t) 0x100000)
53
54 enum fault_type {
55 KERNEL_FAULT,
56 USER_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 (user_mode(regs))
81 return USER_FAULT;
82 if (!IS_ENABLED(CONFIG_PGSTE))
83 return KERNEL_FAULT;
84 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
85 return GMAP_FAULT;
86 return KERNEL_FAULT;
87 }
88 if (trans_exc_code == 2)
89 return USER_FAULT;
90 if (trans_exc_code == 1) {
91 /* access register mode, not used in the kernel */
92 return USER_FAULT;
93 }
94 /* home space exception -> access via kernel ASCE */
95 return KERNEL_FAULT;
96 }
97
bad_address(void * p)98 static int bad_address(void *p)
99 {
100 unsigned long dummy;
101
102 return get_kernel_nofault(dummy, (unsigned long *)p);
103 }
104
dump_pagetable(unsigned long asce,unsigned long address)105 static void dump_pagetable(unsigned long asce, unsigned long address)
106 {
107 unsigned long *table = __va(asce & _ASCE_ORIGIN);
108
109 pr_alert("AS:%016lx ", asce);
110 switch (asce & _ASCE_TYPE_MASK) {
111 case _ASCE_TYPE_REGION1:
112 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
113 if (bad_address(table))
114 goto bad;
115 pr_cont("R1:%016lx ", *table);
116 if (*table & _REGION_ENTRY_INVALID)
117 goto out;
118 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
119 fallthrough;
120 case _ASCE_TYPE_REGION2:
121 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
122 if (bad_address(table))
123 goto bad;
124 pr_cont("R2:%016lx ", *table);
125 if (*table & _REGION_ENTRY_INVALID)
126 goto out;
127 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
128 fallthrough;
129 case _ASCE_TYPE_REGION3:
130 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
131 if (bad_address(table))
132 goto bad;
133 pr_cont("R3:%016lx ", *table);
134 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
135 goto out;
136 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
137 fallthrough;
138 case _ASCE_TYPE_SEGMENT:
139 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
140 if (bad_address(table))
141 goto bad;
142 pr_cont("S:%016lx ", *table);
143 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
144 goto out;
145 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
146 }
147 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
148 if (bad_address(table))
149 goto bad;
150 pr_cont("P:%016lx ", *table);
151 out:
152 pr_cont("\n");
153 return;
154 bad:
155 pr_cont("BAD\n");
156 }
157
dump_fault_info(struct pt_regs * regs)158 static void dump_fault_info(struct pt_regs *regs)
159 {
160 unsigned long asce;
161
162 pr_alert("Failing address: %016lx TEID: %016lx\n",
163 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
164 pr_alert("Fault in ");
165 switch (regs->int_parm_long & 3) {
166 case 3:
167 pr_cont("home space ");
168 break;
169 case 2:
170 pr_cont("secondary space ");
171 break;
172 case 1:
173 pr_cont("access register ");
174 break;
175 case 0:
176 pr_cont("primary space ");
177 break;
178 }
179 pr_cont("mode while using ");
180 switch (get_fault_type(regs)) {
181 case USER_FAULT:
182 asce = S390_lowcore.user_asce;
183 pr_cont("user ");
184 break;
185 case GMAP_FAULT:
186 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
187 pr_cont("gmap ");
188 break;
189 case KERNEL_FAULT:
190 asce = S390_lowcore.kernel_asce;
191 pr_cont("kernel ");
192 break;
193 default:
194 unreachable();
195 }
196 pr_cont("ASCE.\n");
197 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
198 }
199
200 int show_unhandled_signals = 1;
201
report_user_fault(struct pt_regs * regs,long signr,int is_mm_fault)202 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
203 {
204 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
205 return;
206 if (!unhandled_signal(current, signr))
207 return;
208 if (!printk_ratelimit())
209 return;
210 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
211 regs->int_code & 0xffff, regs->int_code >> 17);
212 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
213 printk(KERN_CONT "\n");
214 if (is_mm_fault)
215 dump_fault_info(regs);
216 show_regs(regs);
217 }
218
219 /*
220 * Send SIGSEGV to task. This is an external routine
221 * to keep the stack usage of do_page_fault small.
222 */
do_sigsegv(struct pt_regs * regs,int si_code)223 static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
224 {
225 report_user_fault(regs, SIGSEGV, 1);
226 force_sig_fault(SIGSEGV, si_code,
227 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
228 }
229
s390_search_extables(unsigned long addr)230 const struct exception_table_entry *s390_search_extables(unsigned long addr)
231 {
232 const struct exception_table_entry *fixup;
233
234 fixup = search_extable(__start_amode31_ex_table,
235 __stop_amode31_ex_table - __start_amode31_ex_table,
236 addr);
237 if (!fixup)
238 fixup = search_exception_tables(addr);
239 return fixup;
240 }
241
do_no_context(struct pt_regs * regs)242 static noinline void do_no_context(struct pt_regs *regs)
243 {
244 const struct exception_table_entry *fixup;
245
246 /* Are we prepared to handle this kernel fault? */
247 fixup = s390_search_extables(regs->psw.addr);
248 if (fixup && ex_handle(fixup, regs))
249 return;
250
251 /*
252 * Oops. The kernel tried to access some bad page. We'll have to
253 * terminate things with extreme prejudice.
254 */
255 if (get_fault_type(regs) == KERNEL_FAULT)
256 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
257 " in virtual kernel address space\n");
258 else
259 printk(KERN_ALERT "Unable to handle kernel paging request"
260 " in virtual user address space\n");
261 dump_fault_info(regs);
262 die(regs, "Oops");
263 do_exit(SIGKILL);
264 }
265
do_low_address(struct pt_regs * regs)266 static noinline void do_low_address(struct pt_regs *regs)
267 {
268 /* Low-address protection hit in kernel mode means
269 NULL pointer write access in kernel mode. */
270 if (regs->psw.mask & PSW_MASK_PSTATE) {
271 /* Low-address protection hit in user mode 'cannot happen'. */
272 die (regs, "Low-address protection");
273 do_exit(SIGKILL);
274 }
275
276 do_no_context(regs);
277 }
278
do_sigbus(struct pt_regs * regs)279 static noinline void do_sigbus(struct pt_regs *regs)
280 {
281 /*
282 * Send a sigbus, regardless of whether we were in kernel
283 * or user mode.
284 */
285 force_sig_fault(SIGBUS, BUS_ADRERR,
286 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
287 }
288
do_fault_error(struct pt_regs * regs,int access,vm_fault_t fault)289 static noinline void do_fault_error(struct pt_regs *regs, int access,
290 vm_fault_t fault)
291 {
292 int si_code;
293
294 switch (fault) {
295 case VM_FAULT_BADACCESS:
296 case VM_FAULT_BADMAP:
297 /* Bad memory access. Check if it is kernel or user space. */
298 if (user_mode(regs)) {
299 /* User mode accesses just cause a SIGSEGV */
300 si_code = (fault == VM_FAULT_BADMAP) ?
301 SEGV_MAPERR : SEGV_ACCERR;
302 do_sigsegv(regs, si_code);
303 break;
304 }
305 fallthrough;
306 case VM_FAULT_BADCONTEXT:
307 case VM_FAULT_PFAULT:
308 do_no_context(regs);
309 break;
310 case VM_FAULT_SIGNAL:
311 if (!user_mode(regs))
312 do_no_context(regs);
313 break;
314 default: /* fault & VM_FAULT_ERROR */
315 if (fault & VM_FAULT_OOM) {
316 if (!user_mode(regs))
317 do_no_context(regs);
318 else
319 pagefault_out_of_memory();
320 } else if (fault & VM_FAULT_SIGSEGV) {
321 /* Kernel mode? Handle exceptions or die */
322 if (!user_mode(regs))
323 do_no_context(regs);
324 else
325 do_sigsegv(regs, SEGV_MAPERR);
326 } else if (fault & VM_FAULT_SIGBUS) {
327 /* Kernel mode? Handle exceptions or die */
328 if (!user_mode(regs))
329 do_no_context(regs);
330 else
331 do_sigbus(regs);
332 } else
333 BUG();
334 break;
335 }
336 }
337
338 /*
339 * This routine handles page faults. It determines the address,
340 * and the problem, and then passes it off to one of the appropriate
341 * routines.
342 *
343 * interruption code (int_code):
344 * 04 Protection -> Write-Protection (suppression)
345 * 10 Segment translation -> Not present (nullification)
346 * 11 Page translation -> Not present (nullification)
347 * 3b Region third trans. -> Not present (nullification)
348 */
do_exception(struct pt_regs * regs,int access)349 static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
350 {
351 struct gmap *gmap;
352 struct task_struct *tsk;
353 struct mm_struct *mm;
354 struct vm_area_struct *vma;
355 enum fault_type type;
356 unsigned long trans_exc_code;
357 unsigned long address;
358 unsigned int flags;
359 vm_fault_t fault;
360 bool is_write;
361
362 tsk = current;
363 /*
364 * The instruction that caused the program check has
365 * been nullified. Don't signal single step via SIGTRAP.
366 */
367 clear_thread_flag(TIF_PER_TRAP);
368
369 if (kprobe_page_fault(regs, 14))
370 return 0;
371
372 mm = tsk->mm;
373 trans_exc_code = regs->int_parm_long;
374 address = trans_exc_code & __FAIL_ADDR_MASK;
375 is_write = (trans_exc_code & store_indication) == 0x400;
376
377 /*
378 * Verify that the fault happened in user space, that
379 * we are not in an interrupt and that there is a
380 * user context.
381 */
382 fault = VM_FAULT_BADCONTEXT;
383 type = get_fault_type(regs);
384 switch (type) {
385 case KERNEL_FAULT:
386 if (kfence_handle_page_fault(address, is_write, regs))
387 return 0;
388 goto out;
389 case USER_FAULT:
390 case GMAP_FAULT:
391 if (faulthandler_disabled() || !mm)
392 goto out;
393 break;
394 }
395
396 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
397 flags = FAULT_FLAG_DEFAULT;
398 if (user_mode(regs))
399 flags |= FAULT_FLAG_USER;
400 if (is_write)
401 access = VM_WRITE;
402 if (access == VM_WRITE)
403 flags |= FAULT_FLAG_WRITE;
404 mmap_read_lock(mm);
405
406 gmap = NULL;
407 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
408 gmap = (struct gmap *) S390_lowcore.gmap;
409 current->thread.gmap_addr = address;
410 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
411 current->thread.gmap_int_code = regs->int_code & 0xffff;
412 address = __gmap_translate(gmap, address);
413 if (address == -EFAULT) {
414 fault = VM_FAULT_BADMAP;
415 goto out_up;
416 }
417 if (gmap->pfault_enabled)
418 flags |= FAULT_FLAG_RETRY_NOWAIT;
419 }
420
421 retry:
422 fault = VM_FAULT_BADMAP;
423 vma = find_vma(mm, address);
424 if (!vma)
425 goto out_up;
426
427 if (unlikely(vma->vm_start > address)) {
428 if (!(vma->vm_flags & VM_GROWSDOWN))
429 goto out_up;
430 if (expand_stack(vma, address))
431 goto out_up;
432 }
433
434 /*
435 * Ok, we have a good vm_area for this memory access, so
436 * we can handle it..
437 */
438 fault = VM_FAULT_BADACCESS;
439 if (unlikely(!(vma->vm_flags & access)))
440 goto out_up;
441
442 if (is_vm_hugetlb_page(vma))
443 address &= HPAGE_MASK;
444 /*
445 * If for any reason at all we couldn't handle the fault,
446 * make sure we exit gracefully rather than endlessly redo
447 * the fault.
448 */
449 fault = handle_mm_fault(vma, address, flags, regs);
450 if (fault_signal_pending(fault, regs)) {
451 fault = VM_FAULT_SIGNAL;
452 if (flags & FAULT_FLAG_RETRY_NOWAIT)
453 goto out_up;
454 goto out;
455 }
456 if (unlikely(fault & VM_FAULT_ERROR))
457 goto out_up;
458
459 if (flags & FAULT_FLAG_ALLOW_RETRY) {
460 if (fault & VM_FAULT_RETRY) {
461 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
462 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
463 /* FAULT_FLAG_RETRY_NOWAIT has been set,
464 * mmap_lock has not been released */
465 current->thread.gmap_pfault = 1;
466 fault = VM_FAULT_PFAULT;
467 goto out_up;
468 }
469 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
470 flags |= FAULT_FLAG_TRIED;
471 mmap_read_lock(mm);
472 goto retry;
473 }
474 }
475 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
476 address = __gmap_link(gmap, current->thread.gmap_addr,
477 address);
478 if (address == -EFAULT) {
479 fault = VM_FAULT_BADMAP;
480 goto out_up;
481 }
482 if (address == -ENOMEM) {
483 fault = VM_FAULT_OOM;
484 goto out_up;
485 }
486 }
487 fault = 0;
488 out_up:
489 mmap_read_unlock(mm);
490 out:
491 return fault;
492 }
493
do_protection_exception(struct pt_regs * regs)494 void do_protection_exception(struct pt_regs *regs)
495 {
496 unsigned long trans_exc_code;
497 int access;
498 vm_fault_t fault;
499
500 trans_exc_code = regs->int_parm_long;
501 /*
502 * Protection exceptions are suppressing, decrement psw address.
503 * The exception to this rule are aborted transactions, for these
504 * the PSW already points to the correct location.
505 */
506 if (!(regs->int_code & 0x200))
507 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
508 /*
509 * Check for low-address protection. This needs to be treated
510 * as a special case because the translation exception code
511 * field is not guaranteed to contain valid data in this case.
512 */
513 if (unlikely(!(trans_exc_code & 4))) {
514 do_low_address(regs);
515 return;
516 }
517 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
518 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
519 (regs->psw.addr & PAGE_MASK);
520 access = VM_EXEC;
521 fault = VM_FAULT_BADACCESS;
522 } else {
523 access = VM_WRITE;
524 fault = do_exception(regs, access);
525 }
526 if (unlikely(fault))
527 do_fault_error(regs, access, fault);
528 }
529 NOKPROBE_SYMBOL(do_protection_exception);
530
do_dat_exception(struct pt_regs * regs)531 void do_dat_exception(struct pt_regs *regs)
532 {
533 int access;
534 vm_fault_t fault;
535
536 access = VM_ACCESS_FLAGS;
537 fault = do_exception(regs, access);
538 if (unlikely(fault))
539 do_fault_error(regs, access, fault);
540 }
541 NOKPROBE_SYMBOL(do_dat_exception);
542
543 #ifdef CONFIG_PFAULT
544 /*
545 * 'pfault' pseudo page faults routines.
546 */
547 static int pfault_disable;
548
nopfault(char * str)549 static int __init nopfault(char *str)
550 {
551 pfault_disable = 1;
552 return 1;
553 }
554
555 __setup("nopfault", nopfault);
556
557 struct pfault_refbk {
558 u16 refdiagc;
559 u16 reffcode;
560 u16 refdwlen;
561 u16 refversn;
562 u64 refgaddr;
563 u64 refselmk;
564 u64 refcmpmk;
565 u64 reserved;
566 } __attribute__ ((packed, aligned(8)));
567
568 static struct pfault_refbk pfault_init_refbk = {
569 .refdiagc = 0x258,
570 .reffcode = 0,
571 .refdwlen = 5,
572 .refversn = 2,
573 .refgaddr = __LC_LPP,
574 .refselmk = 1ULL << 48,
575 .refcmpmk = 1ULL << 48,
576 .reserved = __PF_RES_FIELD
577 };
578
pfault_init(void)579 int pfault_init(void)
580 {
581 int rc;
582
583 if (pfault_disable)
584 return -1;
585 diag_stat_inc(DIAG_STAT_X258);
586 asm volatile(
587 " diag %1,%0,0x258\n"
588 "0: j 2f\n"
589 "1: la %0,8\n"
590 "2:\n"
591 EX_TABLE(0b,1b)
592 : "=d" (rc)
593 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
594 return rc;
595 }
596
597 static struct pfault_refbk pfault_fini_refbk = {
598 .refdiagc = 0x258,
599 .reffcode = 1,
600 .refdwlen = 5,
601 .refversn = 2,
602 };
603
pfault_fini(void)604 void pfault_fini(void)
605 {
606
607 if (pfault_disable)
608 return;
609 diag_stat_inc(DIAG_STAT_X258);
610 asm volatile(
611 " diag %0,0,0x258\n"
612 "0: nopr %%r7\n"
613 EX_TABLE(0b,0b)
614 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
615 }
616
617 static DEFINE_SPINLOCK(pfault_lock);
618 static LIST_HEAD(pfault_list);
619
620 #define PF_COMPLETE 0x0080
621
622 /*
623 * The mechanism of our pfault code: if Linux is running as guest, runs a user
624 * space process and the user space process accesses a page that the host has
625 * paged out we get a pfault interrupt.
626 *
627 * This allows us, within the guest, to schedule a different process. Without
628 * this mechanism the host would have to suspend the whole virtual cpu until
629 * the page has been paged in.
630 *
631 * So when we get such an interrupt then we set the state of the current task
632 * to uninterruptible and also set the need_resched flag. Both happens within
633 * interrupt context(!). If we later on want to return to user space we
634 * recognize the need_resched flag and then call schedule(). It's not very
635 * obvious how this works...
636 *
637 * Of course we have a lot of additional fun with the completion interrupt (->
638 * host signals that a page of a process has been paged in and the process can
639 * continue to run). This interrupt can arrive on any cpu and, since we have
640 * virtual cpus, actually appear before the interrupt that signals that a page
641 * is missing.
642 */
pfault_interrupt(struct ext_code ext_code,unsigned int param32,unsigned long param64)643 static void pfault_interrupt(struct ext_code ext_code,
644 unsigned int param32, unsigned long param64)
645 {
646 struct task_struct *tsk;
647 __u16 subcode;
648 pid_t pid;
649
650 /*
651 * Get the external interruption subcode & pfault initial/completion
652 * signal bit. VM stores this in the 'cpu address' field associated
653 * with the external interrupt.
654 */
655 subcode = ext_code.subcode;
656 if ((subcode & 0xff00) != __SUBCODE_MASK)
657 return;
658 inc_irq_stat(IRQEXT_PFL);
659 /* Get the token (= pid of the affected task). */
660 pid = param64 & LPP_PID_MASK;
661 rcu_read_lock();
662 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
663 if (tsk)
664 get_task_struct(tsk);
665 rcu_read_unlock();
666 if (!tsk)
667 return;
668 spin_lock(&pfault_lock);
669 if (subcode & PF_COMPLETE) {
670 /* signal bit is set -> a page has been swapped in by VM */
671 if (tsk->thread.pfault_wait == 1) {
672 /* Initial interrupt was faster than the completion
673 * interrupt. pfault_wait is valid. Set pfault_wait
674 * back to zero and wake up the process. This can
675 * safely be done because the task is still sleeping
676 * and can't produce new pfaults. */
677 tsk->thread.pfault_wait = 0;
678 list_del(&tsk->thread.list);
679 wake_up_process(tsk);
680 put_task_struct(tsk);
681 } else {
682 /* Completion interrupt was faster than initial
683 * interrupt. Set pfault_wait to -1 so the initial
684 * interrupt doesn't put the task to sleep.
685 * If the task is not running, ignore the completion
686 * interrupt since it must be a leftover of a PFAULT
687 * CANCEL operation which didn't remove all pending
688 * completion interrupts. */
689 if (task_is_running(tsk))
690 tsk->thread.pfault_wait = -1;
691 }
692 } else {
693 /* signal bit not set -> a real page is missing. */
694 if (WARN_ON_ONCE(tsk != current))
695 goto out;
696 if (tsk->thread.pfault_wait == 1) {
697 /* Already on the list with a reference: put to sleep */
698 goto block;
699 } else if (tsk->thread.pfault_wait == -1) {
700 /* Completion interrupt was faster than the initial
701 * interrupt (pfault_wait == -1). Set pfault_wait
702 * back to zero and exit. */
703 tsk->thread.pfault_wait = 0;
704 } else {
705 /* Initial interrupt arrived before completion
706 * interrupt. Let the task sleep.
707 * An extra task reference is needed since a different
708 * cpu may set the task state to TASK_RUNNING again
709 * before the scheduler is reached. */
710 get_task_struct(tsk);
711 tsk->thread.pfault_wait = 1;
712 list_add(&tsk->thread.list, &pfault_list);
713 block:
714 /* Since this must be a userspace fault, there
715 * is no kernel task state to trample. Rely on the
716 * return to userspace schedule() to block. */
717 __set_current_state(TASK_UNINTERRUPTIBLE);
718 set_tsk_need_resched(tsk);
719 set_preempt_need_resched();
720 }
721 }
722 out:
723 spin_unlock(&pfault_lock);
724 put_task_struct(tsk);
725 }
726
pfault_cpu_dead(unsigned int cpu)727 static int pfault_cpu_dead(unsigned int cpu)
728 {
729 struct thread_struct *thread, *next;
730 struct task_struct *tsk;
731
732 spin_lock_irq(&pfault_lock);
733 list_for_each_entry_safe(thread, next, &pfault_list, list) {
734 thread->pfault_wait = 0;
735 list_del(&thread->list);
736 tsk = container_of(thread, struct task_struct, thread);
737 wake_up_process(tsk);
738 put_task_struct(tsk);
739 }
740 spin_unlock_irq(&pfault_lock);
741 return 0;
742 }
743
pfault_irq_init(void)744 static int __init pfault_irq_init(void)
745 {
746 int rc;
747
748 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
749 if (rc)
750 goto out_extint;
751 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
752 if (rc)
753 goto out_pfault;
754 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
755 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
756 NULL, pfault_cpu_dead);
757 return 0;
758
759 out_pfault:
760 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
761 out_extint:
762 pfault_disable = 1;
763 return rc;
764 }
765 early_initcall(pfault_irq_init);
766
767 #endif /* CONFIG_PFAULT */
768
769 #if IS_ENABLED(CONFIG_PGSTE)
770
do_secure_storage_access(struct pt_regs * regs)771 void do_secure_storage_access(struct pt_regs *regs)
772 {
773 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
774 struct vm_area_struct *vma;
775 struct mm_struct *mm;
776 struct page *page;
777 int rc;
778
779 /*
780 * bit 61 tells us if the address is valid, if it's not we
781 * have a major problem and should stop the kernel or send a
782 * SIGSEGV to the process. Unfortunately bit 61 is not
783 * reliable without the misc UV feature so we need to check
784 * for that as well.
785 */
786 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
787 !test_bit_inv(61, ®s->int_parm_long)) {
788 /*
789 * When this happens, userspace did something that it
790 * was not supposed to do, e.g. branching into secure
791 * memory. Trigger a segmentation fault.
792 */
793 if (user_mode(regs)) {
794 send_sig(SIGSEGV, current, 0);
795 return;
796 }
797
798 /*
799 * The kernel should never run into this case and we
800 * have no way out of this situation.
801 */
802 panic("Unexpected PGM 0x3d with TEID bit 61=0");
803 }
804
805 switch (get_fault_type(regs)) {
806 case USER_FAULT:
807 mm = current->mm;
808 mmap_read_lock(mm);
809 vma = find_vma(mm, addr);
810 if (!vma) {
811 mmap_read_unlock(mm);
812 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
813 break;
814 }
815 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
816 if (IS_ERR_OR_NULL(page)) {
817 mmap_read_unlock(mm);
818 break;
819 }
820 if (arch_make_page_accessible(page))
821 send_sig(SIGSEGV, current, 0);
822 put_page(page);
823 mmap_read_unlock(mm);
824 break;
825 case KERNEL_FAULT:
826 page = phys_to_page(addr);
827 if (unlikely(!try_get_page(page)))
828 break;
829 rc = arch_make_page_accessible(page);
830 put_page(page);
831 if (rc)
832 BUG();
833 break;
834 case GMAP_FAULT:
835 default:
836 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
837 WARN_ON_ONCE(1);
838 }
839 }
840 NOKPROBE_SYMBOL(do_secure_storage_access);
841
do_non_secure_storage_access(struct pt_regs * regs)842 void do_non_secure_storage_access(struct pt_regs *regs)
843 {
844 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
845 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
846
847 if (get_fault_type(regs) != GMAP_FAULT) {
848 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
849 WARN_ON_ONCE(1);
850 return;
851 }
852
853 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
854 send_sig(SIGSEGV, current, 0);
855 }
856 NOKPROBE_SYMBOL(do_non_secure_storage_access);
857
do_secure_storage_violation(struct pt_regs * regs)858 void do_secure_storage_violation(struct pt_regs *regs)
859 {
860 /*
861 * Either KVM messed up the secure guest mapping or the same
862 * page is mapped into multiple secure guests.
863 *
864 * This exception is only triggered when a guest 2 is running
865 * and can therefore never occur in kernel context.
866 */
867 printk_ratelimited(KERN_WARNING
868 "Secure storage violation in task: %s, pid %d\n",
869 current->comm, current->pid);
870 send_sig(SIGSEGV, current, 0);
871 }
872
873 #endif /* CONFIG_PGSTE */
874