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/mmu_context.h>
14 #include <linux/perf_event.h>
15 #include <linux/signal.h>
16 #include <linux/sched.h>
17 #include <linux/sched/debug.h>
18 #include <linux/jump_label.h>
19 #include <linux/kernel.h>
20 #include <linux/errno.h>
21 #include <linux/string.h>
22 #include <linux/types.h>
23 #include <linux/ptrace.h>
24 #include <linux/mman.h>
25 #include <linux/mm.h>
26 #include <linux/compat.h>
27 #include <linux/smp.h>
28 #include <linux/kdebug.h>
29 #include <linux/init.h>
30 #include <linux/console.h>
31 #include <linux/extable.h>
32 #include <linux/hardirq.h>
33 #include <linux/kprobes.h>
34 #include <linux/uaccess.h>
35 #include <linux/hugetlb.h>
36 #include <linux/kfence.h>
37 #include <linux/pagewalk.h>
38 #include <asm/asm-extable.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/ptrace.h>
41 #include <asm/fault.h>
42 #include <asm/diag.h>
43 #include <asm/gmap.h>
44 #include <asm/irq.h>
45 #include <asm/facility.h>
46 #include <asm/uv.h>
47 #include "../kernel/entry.h"
48
49 enum fault_type {
50 KERNEL_FAULT,
51 USER_FAULT,
52 GMAP_FAULT,
53 };
54
55 static DEFINE_STATIC_KEY_FALSE(have_store_indication);
56
fault_init(void)57 static int __init fault_init(void)
58 {
59 if (test_facility(75))
60 static_branch_enable(&have_store_indication);
61 return 0;
62 }
63 early_initcall(fault_init);
64
65 /*
66 * Find out which address space caused the exception.
67 */
get_fault_type(struct pt_regs * regs)68 static enum fault_type get_fault_type(struct pt_regs *regs)
69 {
70 union teid teid = { .val = regs->int_parm_long };
71 struct gmap *gmap;
72
73 if (likely(teid.as == PSW_BITS_AS_PRIMARY)) {
74 if (user_mode(regs))
75 return USER_FAULT;
76 if (!IS_ENABLED(CONFIG_PGSTE))
77 return KERNEL_FAULT;
78 gmap = (struct gmap *)get_lowcore()->gmap;
79 if (gmap && gmap->asce == regs->cr1)
80 return GMAP_FAULT;
81 return KERNEL_FAULT;
82 }
83 if (teid.as == PSW_BITS_AS_SECONDARY)
84 return USER_FAULT;
85 /* Access register mode, not used in the kernel */
86 if (teid.as == PSW_BITS_AS_ACCREG)
87 return USER_FAULT;
88 /* Home space -> access via kernel ASCE */
89 return KERNEL_FAULT;
90 }
91
get_fault_address(struct pt_regs * regs)92 static unsigned long get_fault_address(struct pt_regs *regs)
93 {
94 union teid teid = { .val = regs->int_parm_long };
95
96 return teid.addr * PAGE_SIZE;
97 }
98
fault_is_write(struct pt_regs * regs)99 static __always_inline bool fault_is_write(struct pt_regs *regs)
100 {
101 union teid teid = { .val = regs->int_parm_long };
102
103 if (static_branch_likely(&have_store_indication))
104 return teid.fsi == TEID_FSI_STORE;
105 return false;
106 }
107
dump_pagetable(unsigned long asce,unsigned long address)108 static void dump_pagetable(unsigned long asce, unsigned long address)
109 {
110 unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
111
112 pr_alert("AS:%016lx ", asce);
113 switch (asce & _ASCE_TYPE_MASK) {
114 case _ASCE_TYPE_REGION1:
115 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
116 if (get_kernel_nofault(entry, table))
117 goto bad;
118 pr_cont("R1:%016lx ", entry);
119 if (entry & _REGION_ENTRY_INVALID)
120 goto out;
121 table = __va(entry & _REGION_ENTRY_ORIGIN);
122 fallthrough;
123 case _ASCE_TYPE_REGION2:
124 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
125 if (get_kernel_nofault(entry, table))
126 goto bad;
127 pr_cont("R2:%016lx ", entry);
128 if (entry & _REGION_ENTRY_INVALID)
129 goto out;
130 table = __va(entry & _REGION_ENTRY_ORIGIN);
131 fallthrough;
132 case _ASCE_TYPE_REGION3:
133 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
134 if (get_kernel_nofault(entry, table))
135 goto bad;
136 pr_cont("R3:%016lx ", entry);
137 if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
138 goto out;
139 table = __va(entry & _REGION_ENTRY_ORIGIN);
140 fallthrough;
141 case _ASCE_TYPE_SEGMENT:
142 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
143 if (get_kernel_nofault(entry, table))
144 goto bad;
145 pr_cont("S:%016lx ", entry);
146 if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
147 goto out;
148 table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
149 }
150 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
151 if (get_kernel_nofault(entry, table))
152 goto bad;
153 pr_cont("P:%016lx ", entry);
154 out:
155 pr_cont("\n");
156 return;
157 bad:
158 pr_cont("BAD\n");
159 }
160
dump_fault_info(struct pt_regs * regs)161 static void dump_fault_info(struct pt_regs *regs)
162 {
163 union teid teid = { .val = regs->int_parm_long };
164 unsigned long asce;
165
166 pr_alert("Failing address: %016lx TEID: %016lx\n",
167 get_fault_address(regs), teid.val);
168 pr_alert("Fault in ");
169 switch (teid.as) {
170 case PSW_BITS_AS_HOME:
171 pr_cont("home space ");
172 break;
173 case PSW_BITS_AS_SECONDARY:
174 pr_cont("secondary space ");
175 break;
176 case PSW_BITS_AS_ACCREG:
177 pr_cont("access register ");
178 break;
179 case PSW_BITS_AS_PRIMARY:
180 pr_cont("primary space ");
181 break;
182 }
183 pr_cont("mode while using ");
184 switch (get_fault_type(regs)) {
185 case USER_FAULT:
186 asce = get_lowcore()->user_asce.val;
187 pr_cont("user ");
188 break;
189 case GMAP_FAULT:
190 asce = ((struct gmap *)get_lowcore()->gmap)->asce;
191 pr_cont("gmap ");
192 break;
193 case KERNEL_FAULT:
194 asce = get_lowcore()->kernel_asce.val;
195 pr_cont("kernel ");
196 break;
197 default:
198 unreachable();
199 }
200 pr_cont("ASCE.\n");
201 dump_pagetable(asce, get_fault_address(regs));
202 }
203
204 int show_unhandled_signals = 1;
205
report_user_fault(struct pt_regs * regs,long signr,int is_mm_fault)206 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
207 {
208 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
209
210 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
211 return;
212 if (!unhandled_signal(current, signr))
213 return;
214 if (!__ratelimit(&rs))
215 return;
216 pr_alert("User process fault: interruption code %04x ilc:%d ",
217 regs->int_code & 0xffff, regs->int_code >> 17);
218 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
219 pr_cont("\n");
220 if (is_mm_fault)
221 dump_fault_info(regs);
222 show_regs(regs);
223 }
224
do_sigsegv(struct pt_regs * regs,int si_code)225 static void do_sigsegv(struct pt_regs *regs, int si_code)
226 {
227 report_user_fault(regs, SIGSEGV, 1);
228 force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
229 }
230
handle_fault_error_nolock(struct pt_regs * regs,int si_code)231 static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
232 {
233 enum fault_type fault_type;
234 unsigned long address;
235 bool is_write;
236
237 if (user_mode(regs)) {
238 if (WARN_ON_ONCE(!si_code))
239 si_code = SEGV_MAPERR;
240 return do_sigsegv(regs, si_code);
241 }
242 if (fixup_exception(regs))
243 return;
244 fault_type = get_fault_type(regs);
245 if (fault_type == KERNEL_FAULT) {
246 address = get_fault_address(regs);
247 is_write = fault_is_write(regs);
248 if (kfence_handle_page_fault(address, is_write, regs))
249 return;
250 }
251 if (fault_type == KERNEL_FAULT)
252 pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
253 else
254 pr_alert("Unable to handle kernel paging request in virtual user address space\n");
255 dump_fault_info(regs);
256 die(regs, "Oops");
257 }
258
handle_fault_error(struct pt_regs * regs,int si_code)259 static void handle_fault_error(struct pt_regs *regs, int si_code)
260 {
261 struct mm_struct *mm = current->mm;
262
263 mmap_read_unlock(mm);
264 handle_fault_error_nolock(regs, si_code);
265 }
266
do_sigbus(struct pt_regs * regs)267 static void do_sigbus(struct pt_regs *regs)
268 {
269 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
270 }
271
272 /*
273 * This routine handles page faults. It determines the address,
274 * and the problem, and then passes it off to one of the appropriate
275 * routines.
276 *
277 * interruption code (int_code):
278 * 04 Protection -> Write-Protection (suppression)
279 * 10 Segment translation -> Not present (nullification)
280 * 11 Page translation -> Not present (nullification)
281 * 3b Region third trans. -> Not present (nullification)
282 */
do_exception(struct pt_regs * regs,int access)283 static void do_exception(struct pt_regs *regs, int access)
284 {
285 struct vm_area_struct *vma;
286 unsigned long address;
287 struct mm_struct *mm;
288 enum fault_type type;
289 unsigned int flags;
290 struct gmap *gmap;
291 vm_fault_t fault;
292 bool is_write;
293
294 /*
295 * The instruction that caused the program check has
296 * been nullified. Don't signal single step via SIGTRAP.
297 */
298 clear_thread_flag(TIF_PER_TRAP);
299 if (kprobe_page_fault(regs, 14))
300 return;
301 mm = current->mm;
302 address = get_fault_address(regs);
303 is_write = fault_is_write(regs);
304 type = get_fault_type(regs);
305 switch (type) {
306 case KERNEL_FAULT:
307 return handle_fault_error_nolock(regs, 0);
308 case USER_FAULT:
309 case GMAP_FAULT:
310 if (faulthandler_disabled() || !mm)
311 return handle_fault_error_nolock(regs, 0);
312 break;
313 }
314 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
315 flags = FAULT_FLAG_DEFAULT;
316 if (user_mode(regs))
317 flags |= FAULT_FLAG_USER;
318 if (is_write)
319 access = VM_WRITE;
320 if (access == VM_WRITE)
321 flags |= FAULT_FLAG_WRITE;
322 if (!(flags & FAULT_FLAG_USER))
323 goto lock_mmap;
324 vma = lock_vma_under_rcu(mm, address);
325 if (!vma)
326 goto lock_mmap;
327 if (!(vma->vm_flags & access)) {
328 vma_end_read(vma);
329 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
330 return handle_fault_error_nolock(regs, SEGV_ACCERR);
331 }
332 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
333 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
334 vma_end_read(vma);
335 if (!(fault & VM_FAULT_RETRY)) {
336 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
337 if (unlikely(fault & VM_FAULT_ERROR))
338 goto error;
339 return;
340 }
341 count_vm_vma_lock_event(VMA_LOCK_RETRY);
342 if (fault & VM_FAULT_MAJOR)
343 flags |= FAULT_FLAG_TRIED;
344
345 /* Quick path to respond to signals */
346 if (fault_signal_pending(fault, regs)) {
347 if (!user_mode(regs))
348 handle_fault_error_nolock(regs, 0);
349 return;
350 }
351 lock_mmap:
352 mmap_read_lock(mm);
353 gmap = NULL;
354 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
355 gmap = (struct gmap *)get_lowcore()->gmap;
356 current->thread.gmap_addr = address;
357 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
358 current->thread.gmap_int_code = regs->int_code & 0xffff;
359 address = __gmap_translate(gmap, address);
360 if (address == -EFAULT)
361 return handle_fault_error(regs, SEGV_MAPERR);
362 if (gmap->pfault_enabled)
363 flags |= FAULT_FLAG_RETRY_NOWAIT;
364 }
365 retry:
366 vma = find_vma(mm, address);
367 if (!vma)
368 return handle_fault_error(regs, SEGV_MAPERR);
369 if (unlikely(vma->vm_start > address)) {
370 if (!(vma->vm_flags & VM_GROWSDOWN))
371 return handle_fault_error(regs, SEGV_MAPERR);
372 vma = expand_stack(mm, address);
373 if (!vma)
374 return handle_fault_error_nolock(regs, SEGV_MAPERR);
375 }
376 if (unlikely(!(vma->vm_flags & access)))
377 return handle_fault_error(regs, SEGV_ACCERR);
378 fault = handle_mm_fault(vma, address, flags, regs);
379 if (fault_signal_pending(fault, regs)) {
380 if (flags & FAULT_FLAG_RETRY_NOWAIT)
381 mmap_read_unlock(mm);
382 if (!user_mode(regs))
383 handle_fault_error_nolock(regs, 0);
384 return;
385 }
386 /* The fault is fully completed (including releasing mmap lock) */
387 if (fault & VM_FAULT_COMPLETED) {
388 if (gmap) {
389 mmap_read_lock(mm);
390 goto gmap;
391 }
392 return;
393 }
394 if (unlikely(fault & VM_FAULT_ERROR)) {
395 mmap_read_unlock(mm);
396 goto error;
397 }
398 if (fault & VM_FAULT_RETRY) {
399 if (IS_ENABLED(CONFIG_PGSTE) && gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
400 /*
401 * FAULT_FLAG_RETRY_NOWAIT has been set,
402 * mmap_lock has not been released
403 */
404 current->thread.gmap_pfault = 1;
405 return handle_fault_error(regs, 0);
406 }
407 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
408 flags |= FAULT_FLAG_TRIED;
409 mmap_read_lock(mm);
410 goto retry;
411 }
412 gmap:
413 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
414 address = __gmap_link(gmap, current->thread.gmap_addr,
415 address);
416 if (address == -EFAULT)
417 return handle_fault_error(regs, SEGV_MAPERR);
418 if (address == -ENOMEM) {
419 fault = VM_FAULT_OOM;
420 mmap_read_unlock(mm);
421 goto error;
422 }
423 }
424 mmap_read_unlock(mm);
425 return;
426 error:
427 if (fault & VM_FAULT_OOM) {
428 if (!user_mode(regs))
429 handle_fault_error_nolock(regs, 0);
430 else
431 pagefault_out_of_memory();
432 } else if (fault & VM_FAULT_SIGSEGV) {
433 if (!user_mode(regs))
434 handle_fault_error_nolock(regs, 0);
435 else
436 do_sigsegv(regs, SEGV_MAPERR);
437 } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON)) {
438 if (!user_mode(regs))
439 handle_fault_error_nolock(regs, 0);
440 else
441 do_sigbus(regs);
442 } else {
443 pr_emerg("Unexpected fault flags: %08x\n", fault);
444 BUG();
445 }
446 }
447
do_protection_exception(struct pt_regs * regs)448 void do_protection_exception(struct pt_regs *regs)
449 {
450 union teid teid = { .val = regs->int_parm_long };
451
452 /*
453 * Protection exceptions are suppressing, decrement psw address.
454 * The exception to this rule are aborted transactions, for these
455 * the PSW already points to the correct location.
456 */
457 if (!(regs->int_code & 0x200))
458 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
459 /*
460 * Check for low-address protection. This needs to be treated
461 * as a special case because the translation exception code
462 * field is not guaranteed to contain valid data in this case.
463 */
464 if (unlikely(!teid.b61)) {
465 if (user_mode(regs)) {
466 /* Low-address protection in user mode: cannot happen */
467 die(regs, "Low-address protection");
468 }
469 /*
470 * Low-address protection in kernel mode means
471 * NULL pointer write access in kernel mode.
472 */
473 return handle_fault_error_nolock(regs, 0);
474 }
475 if (unlikely(MACHINE_HAS_NX && teid.b56)) {
476 regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
477 return handle_fault_error_nolock(regs, SEGV_ACCERR);
478 }
479 do_exception(regs, VM_WRITE);
480 }
481 NOKPROBE_SYMBOL(do_protection_exception);
482
do_dat_exception(struct pt_regs * regs)483 void do_dat_exception(struct pt_regs *regs)
484 {
485 do_exception(regs, VM_ACCESS_FLAGS);
486 }
487 NOKPROBE_SYMBOL(do_dat_exception);
488
489 #if IS_ENABLED(CONFIG_PGSTE)
490
do_secure_storage_access(struct pt_regs * regs)491 void do_secure_storage_access(struct pt_regs *regs)
492 {
493 union teid teid = { .val = regs->int_parm_long };
494 unsigned long addr = get_fault_address(regs);
495 struct vm_area_struct *vma;
496 struct folio_walk fw;
497 struct mm_struct *mm;
498 struct folio *folio;
499 struct gmap *gmap;
500 int rc;
501
502 /*
503 * Bit 61 indicates if the address is valid, if it is not the
504 * kernel should be stopped or SIGSEGV should be sent to the
505 * process. Bit 61 is not reliable without the misc UV feature,
506 * therefore this needs to be checked too.
507 */
508 if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
509 /*
510 * When this happens, userspace did something that it
511 * was not supposed to do, e.g. branching into secure
512 * memory. Trigger a segmentation fault.
513 */
514 if (user_mode(regs)) {
515 send_sig(SIGSEGV, current, 0);
516 return;
517 }
518 /*
519 * The kernel should never run into this case and
520 * there is no way out of this situation.
521 */
522 panic("Unexpected PGM 0x3d with TEID bit 61=0");
523 }
524 switch (get_fault_type(regs)) {
525 case GMAP_FAULT:
526 mm = current->mm;
527 gmap = (struct gmap *)get_lowcore()->gmap;
528 mmap_read_lock(mm);
529 addr = __gmap_translate(gmap, addr);
530 mmap_read_unlock(mm);
531 if (IS_ERR_VALUE(addr))
532 return handle_fault_error_nolock(regs, SEGV_MAPERR);
533 fallthrough;
534 case USER_FAULT:
535 mm = current->mm;
536 mmap_read_lock(mm);
537 vma = find_vma(mm, addr);
538 if (!vma)
539 return handle_fault_error(regs, SEGV_MAPERR);
540 folio = folio_walk_start(&fw, vma, addr, 0);
541 if (!folio) {
542 mmap_read_unlock(mm);
543 break;
544 }
545 /* arch_make_folio_accessible() needs a raised refcount. */
546 folio_get(folio);
547 rc = arch_make_folio_accessible(folio);
548 folio_put(folio);
549 folio_walk_end(&fw, vma);
550 if (rc)
551 send_sig(SIGSEGV, current, 0);
552 mmap_read_unlock(mm);
553 break;
554 case KERNEL_FAULT:
555 folio = phys_to_folio(addr);
556 if (unlikely(!folio_try_get(folio)))
557 break;
558 rc = arch_make_folio_accessible(folio);
559 folio_put(folio);
560 if (rc)
561 BUG();
562 break;
563 default:
564 unreachable();
565 }
566 }
567 NOKPROBE_SYMBOL(do_secure_storage_access);
568
do_non_secure_storage_access(struct pt_regs * regs)569 void do_non_secure_storage_access(struct pt_regs *regs)
570 {
571 struct gmap *gmap = (struct gmap *)get_lowcore()->gmap;
572 unsigned long gaddr = get_fault_address(regs);
573
574 if (WARN_ON_ONCE(get_fault_type(regs) != GMAP_FAULT))
575 return handle_fault_error_nolock(regs, SEGV_MAPERR);
576 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
577 send_sig(SIGSEGV, current, 0);
578 }
579 NOKPROBE_SYMBOL(do_non_secure_storage_access);
580
do_secure_storage_violation(struct pt_regs * regs)581 void do_secure_storage_violation(struct pt_regs *regs)
582 {
583 struct gmap *gmap = (struct gmap *)get_lowcore()->gmap;
584 unsigned long gaddr = get_fault_address(regs);
585
586 /*
587 * If the VM has been rebooted, its address space might still contain
588 * secure pages from the previous boot.
589 * Clear the page so it can be reused.
590 */
591 if (!gmap_destroy_page(gmap, gaddr))
592 return;
593 /*
594 * Either KVM messed up the secure guest mapping or the same
595 * page is mapped into multiple secure guests.
596 *
597 * This exception is only triggered when a guest 2 is running
598 * and can therefore never occur in kernel context.
599 */
600 pr_warn_ratelimited("Secure storage violation in task: %s, pid %d\n",
601 current->comm, current->pid);
602 send_sig(SIGSEGV, current, 0);
603 }
604
605 #endif /* CONFIG_PGSTE */
606