Loading...
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
53enum fault_type {
54 KERNEL_FAULT,
55 USER_FAULT,
56 VDSO_FAULT,
57 GMAP_FAULT,
58};
59
60static unsigned long store_indication __read_mostly;
61
62static int __init fault_init(void)
63{
64 if (test_facility(75))
65 store_indication = 0xc00;
66 return 0;
67}
68early_initcall(fault_init);
69
70/*
71 * Find out which address space caused the exception.
72 */
73static 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
104static int bad_address(void *p)
105{
106 unsigned long dummy;
107
108 return get_kernel_nofault(dummy, (unsigned long *)p);
109}
110
111static 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);
157out:
158 pr_cont("\n");
159 return;
160bad:
161 pr_cont("BAD\n");
162}
163
164static 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
210int show_unhandled_signals = 1;
211
212void 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 */
233static 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
240const 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
252static 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
276static 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
289static 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
299static 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
319static 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 */
382static 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 (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
433 flags |= FAULT_FLAG_WRITE;
434 mmap_read_lock(mm);
435
436 gmap = NULL;
437 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
438 gmap = (struct gmap *) S390_lowcore.gmap;
439 current->thread.gmap_addr = address;
440 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
441 current->thread.gmap_int_code = regs->int_code & 0xffff;
442 address = __gmap_translate(gmap, address);
443 if (address == -EFAULT) {
444 fault = VM_FAULT_BADMAP;
445 goto out_up;
446 }
447 if (gmap->pfault_enabled)
448 flags |= FAULT_FLAG_RETRY_NOWAIT;
449 }
450
451retry:
452 fault = VM_FAULT_BADMAP;
453 vma = find_vma(mm, address);
454 if (!vma)
455 goto out_up;
456
457 if (unlikely(vma->vm_start > address)) {
458 if (!(vma->vm_flags & VM_GROWSDOWN))
459 goto out_up;
460 if (expand_stack(vma, address))
461 goto out_up;
462 }
463
464 /*
465 * Ok, we have a good vm_area for this memory access, so
466 * we can handle it..
467 */
468 fault = VM_FAULT_BADACCESS;
469 if (unlikely(!(vma->vm_flags & access)))
470 goto out_up;
471
472 if (is_vm_hugetlb_page(vma))
473 address &= HPAGE_MASK;
474 /*
475 * If for any reason at all we couldn't handle the fault,
476 * make sure we exit gracefully rather than endlessly redo
477 * the fault.
478 */
479 fault = handle_mm_fault(vma, address, flags, regs);
480 if (fault_signal_pending(fault, regs)) {
481 fault = VM_FAULT_SIGNAL;
482 if (flags & FAULT_FLAG_RETRY_NOWAIT)
483 goto out_up;
484 goto out;
485 }
486 if (unlikely(fault & VM_FAULT_ERROR))
487 goto out_up;
488
489 if (flags & FAULT_FLAG_ALLOW_RETRY) {
490 if (fault & VM_FAULT_RETRY) {
491 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
492 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
493 /* FAULT_FLAG_RETRY_NOWAIT has been set,
494 * mmap_lock has not been released */
495 current->thread.gmap_pfault = 1;
496 fault = VM_FAULT_PFAULT;
497 goto out_up;
498 }
499 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
500 flags |= FAULT_FLAG_TRIED;
501 mmap_read_lock(mm);
502 goto retry;
503 }
504 }
505 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
506 address = __gmap_link(gmap, current->thread.gmap_addr,
507 address);
508 if (address == -EFAULT) {
509 fault = VM_FAULT_BADMAP;
510 goto out_up;
511 }
512 if (address == -ENOMEM) {
513 fault = VM_FAULT_OOM;
514 goto out_up;
515 }
516 }
517 fault = 0;
518out_up:
519 mmap_read_unlock(mm);
520out:
521 return fault;
522}
523
524void do_protection_exception(struct pt_regs *regs)
525{
526 unsigned long trans_exc_code;
527 int access;
528 vm_fault_t fault;
529
530 trans_exc_code = regs->int_parm_long;
531 /*
532 * Protection exceptions are suppressing, decrement psw address.
533 * The exception to this rule are aborted transactions, for these
534 * the PSW already points to the correct location.
535 */
536 if (!(regs->int_code & 0x200))
537 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
538 /*
539 * Check for low-address protection. This needs to be treated
540 * as a special case because the translation exception code
541 * field is not guaranteed to contain valid data in this case.
542 */
543 if (unlikely(!(trans_exc_code & 4))) {
544 do_low_address(regs);
545 return;
546 }
547 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
548 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
549 (regs->psw.addr & PAGE_MASK);
550 access = VM_EXEC;
551 fault = VM_FAULT_BADACCESS;
552 } else {
553 access = VM_WRITE;
554 fault = do_exception(regs, access);
555 }
556 if (unlikely(fault))
557 do_fault_error(regs, access, fault);
558}
559NOKPROBE_SYMBOL(do_protection_exception);
560
561void do_dat_exception(struct pt_regs *regs)
562{
563 int access;
564 vm_fault_t fault;
565
566 access = VM_ACCESS_FLAGS;
567 fault = do_exception(regs, access);
568 if (unlikely(fault))
569 do_fault_error(regs, access, fault);
570}
571NOKPROBE_SYMBOL(do_dat_exception);
572
573#ifdef CONFIG_PFAULT
574/*
575 * 'pfault' pseudo page faults routines.
576 */
577static int pfault_disable;
578
579static int __init nopfault(char *str)
580{
581 pfault_disable = 1;
582 return 1;
583}
584
585__setup("nopfault", nopfault);
586
587struct pfault_refbk {
588 u16 refdiagc;
589 u16 reffcode;
590 u16 refdwlen;
591 u16 refversn;
592 u64 refgaddr;
593 u64 refselmk;
594 u64 refcmpmk;
595 u64 reserved;
596} __attribute__ ((packed, aligned(8)));
597
598static struct pfault_refbk pfault_init_refbk = {
599 .refdiagc = 0x258,
600 .reffcode = 0,
601 .refdwlen = 5,
602 .refversn = 2,
603 .refgaddr = __LC_LPP,
604 .refselmk = 1ULL << 48,
605 .refcmpmk = 1ULL << 48,
606 .reserved = __PF_RES_FIELD
607};
608
609int pfault_init(void)
610{
611 int rc;
612
613 if (pfault_disable)
614 return -1;
615 diag_stat_inc(DIAG_STAT_X258);
616 asm volatile(
617 " diag %1,%0,0x258\n"
618 "0: j 2f\n"
619 "1: la %0,8\n"
620 "2:\n"
621 EX_TABLE(0b,1b)
622 : "=d" (rc)
623 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
624 return rc;
625}
626
627static struct pfault_refbk pfault_fini_refbk = {
628 .refdiagc = 0x258,
629 .reffcode = 1,
630 .refdwlen = 5,
631 .refversn = 2,
632};
633
634void pfault_fini(void)
635{
636
637 if (pfault_disable)
638 return;
639 diag_stat_inc(DIAG_STAT_X258);
640 asm volatile(
641 " diag %0,0,0x258\n"
642 "0: nopr %%r7\n"
643 EX_TABLE(0b,0b)
644 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
645}
646
647static DEFINE_SPINLOCK(pfault_lock);
648static LIST_HEAD(pfault_list);
649
650#define PF_COMPLETE 0x0080
651
652/*
653 * The mechanism of our pfault code: if Linux is running as guest, runs a user
654 * space process and the user space process accesses a page that the host has
655 * paged out we get a pfault interrupt.
656 *
657 * This allows us, within the guest, to schedule a different process. Without
658 * this mechanism the host would have to suspend the whole virtual cpu until
659 * the page has been paged in.
660 *
661 * So when we get such an interrupt then we set the state of the current task
662 * to uninterruptible and also set the need_resched flag. Both happens within
663 * interrupt context(!). If we later on want to return to user space we
664 * recognize the need_resched flag and then call schedule(). It's not very
665 * obvious how this works...
666 *
667 * Of course we have a lot of additional fun with the completion interrupt (->
668 * host signals that a page of a process has been paged in and the process can
669 * continue to run). This interrupt can arrive on any cpu and, since we have
670 * virtual cpus, actually appear before the interrupt that signals that a page
671 * is missing.
672 */
673static void pfault_interrupt(struct ext_code ext_code,
674 unsigned int param32, unsigned long param64)
675{
676 struct task_struct *tsk;
677 __u16 subcode;
678 pid_t pid;
679
680 /*
681 * Get the external interruption subcode & pfault initial/completion
682 * signal bit. VM stores this in the 'cpu address' field associated
683 * with the external interrupt.
684 */
685 subcode = ext_code.subcode;
686 if ((subcode & 0xff00) != __SUBCODE_MASK)
687 return;
688 inc_irq_stat(IRQEXT_PFL);
689 /* Get the token (= pid of the affected task). */
690 pid = param64 & LPP_PID_MASK;
691 rcu_read_lock();
692 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
693 if (tsk)
694 get_task_struct(tsk);
695 rcu_read_unlock();
696 if (!tsk)
697 return;
698 spin_lock(&pfault_lock);
699 if (subcode & PF_COMPLETE) {
700 /* signal bit is set -> a page has been swapped in by VM */
701 if (tsk->thread.pfault_wait == 1) {
702 /* Initial interrupt was faster than the completion
703 * interrupt. pfault_wait is valid. Set pfault_wait
704 * back to zero and wake up the process. This can
705 * safely be done because the task is still sleeping
706 * and can't produce new pfaults. */
707 tsk->thread.pfault_wait = 0;
708 list_del(&tsk->thread.list);
709 wake_up_process(tsk);
710 put_task_struct(tsk);
711 } else {
712 /* Completion interrupt was faster than initial
713 * interrupt. Set pfault_wait to -1 so the initial
714 * interrupt doesn't put the task to sleep.
715 * If the task is not running, ignore the completion
716 * interrupt since it must be a leftover of a PFAULT
717 * CANCEL operation which didn't remove all pending
718 * completion interrupts. */
719 if (tsk->state == TASK_RUNNING)
720 tsk->thread.pfault_wait = -1;
721 }
722 } else {
723 /* signal bit not set -> a real page is missing. */
724 if (WARN_ON_ONCE(tsk != current))
725 goto out;
726 if (tsk->thread.pfault_wait == 1) {
727 /* Already on the list with a reference: put to sleep */
728 goto block;
729 } else if (tsk->thread.pfault_wait == -1) {
730 /* Completion interrupt was faster than the initial
731 * interrupt (pfault_wait == -1). Set pfault_wait
732 * back to zero and exit. */
733 tsk->thread.pfault_wait = 0;
734 } else {
735 /* Initial interrupt arrived before completion
736 * interrupt. Let the task sleep.
737 * An extra task reference is needed since a different
738 * cpu may set the task state to TASK_RUNNING again
739 * before the scheduler is reached. */
740 get_task_struct(tsk);
741 tsk->thread.pfault_wait = 1;
742 list_add(&tsk->thread.list, &pfault_list);
743block:
744 /* Since this must be a userspace fault, there
745 * is no kernel task state to trample. Rely on the
746 * return to userspace schedule() to block. */
747 __set_current_state(TASK_UNINTERRUPTIBLE);
748 set_tsk_need_resched(tsk);
749 set_preempt_need_resched();
750 }
751 }
752out:
753 spin_unlock(&pfault_lock);
754 put_task_struct(tsk);
755}
756
757static int pfault_cpu_dead(unsigned int cpu)
758{
759 struct thread_struct *thread, *next;
760 struct task_struct *tsk;
761
762 spin_lock_irq(&pfault_lock);
763 list_for_each_entry_safe(thread, next, &pfault_list, list) {
764 thread->pfault_wait = 0;
765 list_del(&thread->list);
766 tsk = container_of(thread, struct task_struct, thread);
767 wake_up_process(tsk);
768 put_task_struct(tsk);
769 }
770 spin_unlock_irq(&pfault_lock);
771 return 0;
772}
773
774static int __init pfault_irq_init(void)
775{
776 int rc;
777
778 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
779 if (rc)
780 goto out_extint;
781 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
782 if (rc)
783 goto out_pfault;
784 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
785 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
786 NULL, pfault_cpu_dead);
787 return 0;
788
789out_pfault:
790 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
791out_extint:
792 pfault_disable = 1;
793 return rc;
794}
795early_initcall(pfault_irq_init);
796
797#endif /* CONFIG_PFAULT */
798
799#if IS_ENABLED(CONFIG_PGSTE)
800void do_secure_storage_access(struct pt_regs *regs)
801{
802 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
803 struct vm_area_struct *vma;
804 struct mm_struct *mm;
805 struct page *page;
806 int rc;
807
808 switch (get_fault_type(regs)) {
809 case USER_FAULT:
810 mm = current->mm;
811 mmap_read_lock(mm);
812 vma = find_vma(mm, addr);
813 if (!vma) {
814 mmap_read_unlock(mm);
815 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
816 break;
817 }
818 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET);
819 if (IS_ERR_OR_NULL(page)) {
820 mmap_read_unlock(mm);
821 break;
822 }
823 if (arch_make_page_accessible(page))
824 send_sig(SIGSEGV, current, 0);
825 put_page(page);
826 mmap_read_unlock(mm);
827 break;
828 case KERNEL_FAULT:
829 page = phys_to_page(addr);
830 if (unlikely(!try_get_page(page)))
831 break;
832 rc = arch_make_page_accessible(page);
833 put_page(page);
834 if (rc)
835 BUG();
836 break;
837 case VDSO_FAULT:
838 case GMAP_FAULT:
839 default:
840 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
841 WARN_ON_ONCE(1);
842 }
843}
844NOKPROBE_SYMBOL(do_secure_storage_access);
845
846void do_non_secure_storage_access(struct pt_regs *regs)
847{
848 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
849 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
850
851 if (get_fault_type(regs) != GMAP_FAULT) {
852 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
853 WARN_ON_ONCE(1);
854 return;
855 }
856
857 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
858 send_sig(SIGSEGV, current, 0);
859}
860NOKPROBE_SYMBOL(do_non_secure_storage_access);
861
862void do_secure_storage_violation(struct pt_regs *regs)
863{
864 /*
865 * Either KVM messed up the secure guest mapping or the same
866 * page is mapped into multiple secure guests.
867 *
868 * This exception is only triggered when a guest 2 is running
869 * and can therefore never occur in kernel context.
870 */
871 printk_ratelimited(KERN_WARNING
872 "Secure storage violation in task: %s, pid %d\n",
873 current->comm, current->pid);
874 send_sig(SIGSEGV, current, 0);
875}
876
877#else
878void do_secure_storage_access(struct pt_regs *regs)
879{
880 default_trap_handler(regs);
881}
882
883void do_non_secure_storage_access(struct pt_regs *regs)
884{
885 default_trap_handler(regs);
886}
887
888void do_secure_storage_violation(struct pt_regs *regs)
889{
890 default_trap_handler(regs);
891}
892#endif
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
49static DEFINE_STATIC_KEY_FALSE(have_store_indication);
50
51static int __init fault_init(void)
52{
53 if (test_facility(75))
54 static_branch_enable(&have_store_indication);
55 return 0;
56}
57early_initcall(fault_init);
58
59/*
60 * Find out which address space caused the exception.
61 */
62static bool is_kernel_fault(struct pt_regs *regs)
63{
64 union teid teid = { .val = regs->int_parm_long };
65
66 if (user_mode(regs))
67 return false;
68 if (teid.as == PSW_BITS_AS_SECONDARY)
69 return false;
70 return true;
71}
72
73static unsigned long get_fault_address(struct pt_regs *regs)
74{
75 union teid teid = { .val = regs->int_parm_long };
76
77 return teid.addr * PAGE_SIZE;
78}
79
80static __always_inline bool fault_is_write(struct pt_regs *regs)
81{
82 union teid teid = { .val = regs->int_parm_long };
83
84 if (static_branch_likely(&have_store_indication))
85 return teid.fsi == TEID_FSI_STORE;
86 return false;
87}
88
89static void dump_pagetable(unsigned long asce, unsigned long address)
90{
91 unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
92
93 pr_alert("AS:%016lx ", asce);
94 switch (asce & _ASCE_TYPE_MASK) {
95 case _ASCE_TYPE_REGION1:
96 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
97 if (get_kernel_nofault(entry, table))
98 goto bad;
99 pr_cont("R1:%016lx ", entry);
100 if (entry & _REGION_ENTRY_INVALID)
101 goto out;
102 table = __va(entry & _REGION_ENTRY_ORIGIN);
103 fallthrough;
104 case _ASCE_TYPE_REGION2:
105 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
106 if (get_kernel_nofault(entry, table))
107 goto bad;
108 pr_cont("R2:%016lx ", entry);
109 if (entry & _REGION_ENTRY_INVALID)
110 goto out;
111 table = __va(entry & _REGION_ENTRY_ORIGIN);
112 fallthrough;
113 case _ASCE_TYPE_REGION3:
114 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
115 if (get_kernel_nofault(entry, table))
116 goto bad;
117 pr_cont("R3:%016lx ", entry);
118 if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
119 goto out;
120 table = __va(entry & _REGION_ENTRY_ORIGIN);
121 fallthrough;
122 case _ASCE_TYPE_SEGMENT:
123 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
124 if (get_kernel_nofault(entry, table))
125 goto bad;
126 pr_cont("S:%016lx ", entry);
127 if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
128 goto out;
129 table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
130 }
131 table += (address & _PAGE_INDEX) >> PAGE_SHIFT;
132 if (get_kernel_nofault(entry, table))
133 goto bad;
134 pr_cont("P:%016lx ", entry);
135out:
136 pr_cont("\n");
137 return;
138bad:
139 pr_cont("BAD\n");
140}
141
142static void dump_fault_info(struct pt_regs *regs)
143{
144 union teid teid = { .val = regs->int_parm_long };
145 unsigned long asce;
146
147 pr_alert("Failing address: %016lx TEID: %016lx\n",
148 get_fault_address(regs), teid.val);
149 pr_alert("Fault in ");
150 switch (teid.as) {
151 case PSW_BITS_AS_HOME:
152 pr_cont("home space ");
153 break;
154 case PSW_BITS_AS_SECONDARY:
155 pr_cont("secondary space ");
156 break;
157 case PSW_BITS_AS_ACCREG:
158 pr_cont("access register ");
159 break;
160 case PSW_BITS_AS_PRIMARY:
161 pr_cont("primary space ");
162 break;
163 }
164 pr_cont("mode while using ");
165 if (is_kernel_fault(regs)) {
166 asce = get_lowcore()->kernel_asce.val;
167 pr_cont("kernel ");
168 } else {
169 asce = get_lowcore()->user_asce.val;
170 pr_cont("user ");
171 }
172 pr_cont("ASCE.\n");
173 dump_pagetable(asce, get_fault_address(regs));
174}
175
176int show_unhandled_signals = 1;
177
178void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
179{
180 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
181
182 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
183 return;
184 if (!unhandled_signal(current, signr))
185 return;
186 if (!__ratelimit(&rs))
187 return;
188 pr_alert("User process fault: interruption code %04x ilc:%d ",
189 regs->int_code & 0xffff, regs->int_code >> 17);
190 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
191 pr_cont("\n");
192 if (is_mm_fault)
193 dump_fault_info(regs);
194 show_regs(regs);
195}
196
197static void do_sigsegv(struct pt_regs *regs, int si_code)
198{
199 report_user_fault(regs, SIGSEGV, 1);
200 force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
201}
202
203static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
204{
205 unsigned long address;
206 bool is_write;
207
208 if (user_mode(regs)) {
209 if (WARN_ON_ONCE(!si_code))
210 si_code = SEGV_MAPERR;
211 return do_sigsegv(regs, si_code);
212 }
213 if (fixup_exception(regs))
214 return;
215 if (is_kernel_fault(regs)) {
216 address = get_fault_address(regs);
217 is_write = fault_is_write(regs);
218 if (kfence_handle_page_fault(address, is_write, regs))
219 return;
220 pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
221 } else {
222 pr_alert("Unable to handle kernel paging request in virtual user address space\n");
223 }
224 dump_fault_info(regs);
225 die(regs, "Oops");
226}
227
228static void handle_fault_error(struct pt_regs *regs, int si_code)
229{
230 struct mm_struct *mm = current->mm;
231
232 mmap_read_unlock(mm);
233 handle_fault_error_nolock(regs, si_code);
234}
235
236static void do_sigbus(struct pt_regs *regs)
237{
238 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
239}
240
241/*
242 * This routine handles page faults. It determines the address,
243 * and the problem, and then passes it off to one of the appropriate
244 * routines.
245 *
246 * interruption code (int_code):
247 * 04 Protection -> Write-Protection (suppression)
248 * 10 Segment translation -> Not present (nullification)
249 * 11 Page translation -> Not present (nullification)
250 * 3b Region third trans. -> Not present (nullification)
251 */
252static void do_exception(struct pt_regs *regs, int access)
253{
254 struct vm_area_struct *vma;
255 unsigned long address;
256 struct mm_struct *mm;
257 unsigned int flags;
258 vm_fault_t fault;
259 bool is_write;
260
261 /*
262 * The instruction that caused the program check has
263 * been nullified. Don't signal single step via SIGTRAP.
264 */
265 clear_thread_flag(TIF_PER_TRAP);
266 if (kprobe_page_fault(regs, 14))
267 return;
268 mm = current->mm;
269 address = get_fault_address(regs);
270 is_write = fault_is_write(regs);
271 if (is_kernel_fault(regs) || faulthandler_disabled() || !mm)
272 return handle_fault_error_nolock(regs, 0);
273 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
274 flags = FAULT_FLAG_DEFAULT;
275 if (user_mode(regs))
276 flags |= FAULT_FLAG_USER;
277 if (is_write)
278 access = VM_WRITE;
279 if (access == VM_WRITE)
280 flags |= FAULT_FLAG_WRITE;
281 if (!(flags & FAULT_FLAG_USER))
282 goto lock_mmap;
283 vma = lock_vma_under_rcu(mm, address);
284 if (!vma)
285 goto lock_mmap;
286 if (!(vma->vm_flags & access)) {
287 vma_end_read(vma);
288 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
289 return handle_fault_error_nolock(regs, SEGV_ACCERR);
290 }
291 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
292 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
293 vma_end_read(vma);
294 if (!(fault & VM_FAULT_RETRY)) {
295 count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
296 goto done;
297 }
298 count_vm_vma_lock_event(VMA_LOCK_RETRY);
299 if (fault & VM_FAULT_MAJOR)
300 flags |= FAULT_FLAG_TRIED;
301 /* Quick path to respond to signals */
302 if (fault_signal_pending(fault, regs)) {
303 if (!user_mode(regs))
304 handle_fault_error_nolock(regs, 0);
305 return;
306 }
307lock_mmap:
308retry:
309 vma = lock_mm_and_find_vma(mm, address, regs);
310 if (!vma)
311 return handle_fault_error_nolock(regs, SEGV_MAPERR);
312 if (unlikely(!(vma->vm_flags & access)))
313 return handle_fault_error(regs, SEGV_ACCERR);
314 fault = handle_mm_fault(vma, address, flags, regs);
315 if (fault_signal_pending(fault, regs)) {
316 if (!user_mode(regs))
317 handle_fault_error_nolock(regs, 0);
318 return;
319 }
320 /* The fault is fully completed (including releasing mmap lock) */
321 if (fault & VM_FAULT_COMPLETED)
322 return;
323 if (fault & VM_FAULT_RETRY) {
324 flags |= FAULT_FLAG_TRIED;
325 goto retry;
326 }
327 mmap_read_unlock(mm);
328done:
329 if (!(fault & VM_FAULT_ERROR))
330 return;
331 if (fault & VM_FAULT_OOM) {
332 if (!user_mode(regs))
333 handle_fault_error_nolock(regs, 0);
334 else
335 pagefault_out_of_memory();
336 } else if (fault & VM_FAULT_SIGSEGV) {
337 if (!user_mode(regs))
338 handle_fault_error_nolock(regs, 0);
339 else
340 do_sigsegv(regs, SEGV_MAPERR);
341 } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
342 VM_FAULT_HWPOISON_LARGE)) {
343 if (!user_mode(regs))
344 handle_fault_error_nolock(regs, 0);
345 else
346 do_sigbus(regs);
347 } else {
348 pr_emerg("Unexpected fault flags: %08x\n", fault);
349 BUG();
350 }
351}
352
353void do_protection_exception(struct pt_regs *regs)
354{
355 union teid teid = { .val = regs->int_parm_long };
356
357 /*
358 * Protection exceptions are suppressing, decrement psw address.
359 * The exception to this rule are aborted transactions, for these
360 * the PSW already points to the correct location.
361 */
362 if (!(regs->int_code & 0x200))
363 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
364 /*
365 * Check for low-address protection. This needs to be treated
366 * as a special case because the translation exception code
367 * field is not guaranteed to contain valid data in this case.
368 */
369 if (unlikely(!teid.b61)) {
370 if (user_mode(regs)) {
371 /* Low-address protection in user mode: cannot happen */
372 die(regs, "Low-address protection");
373 }
374 /*
375 * Low-address protection in kernel mode means
376 * NULL pointer write access in kernel mode.
377 */
378 return handle_fault_error_nolock(regs, 0);
379 }
380 if (unlikely(MACHINE_HAS_NX && teid.b56)) {
381 regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
382 return handle_fault_error_nolock(regs, SEGV_ACCERR);
383 }
384 do_exception(regs, VM_WRITE);
385}
386NOKPROBE_SYMBOL(do_protection_exception);
387
388void do_dat_exception(struct pt_regs *regs)
389{
390 do_exception(regs, VM_ACCESS_FLAGS);
391}
392NOKPROBE_SYMBOL(do_dat_exception);
393
394#if IS_ENABLED(CONFIG_PGSTE)
395
396void do_secure_storage_access(struct pt_regs *regs)
397{
398 union teid teid = { .val = regs->int_parm_long };
399 unsigned long addr = get_fault_address(regs);
400 struct vm_area_struct *vma;
401 struct folio_walk fw;
402 struct mm_struct *mm;
403 struct folio *folio;
404 int rc;
405
406 /*
407 * Bit 61 indicates if the address is valid, if it is not the
408 * kernel should be stopped or SIGSEGV should be sent to the
409 * process. Bit 61 is not reliable without the misc UV feature,
410 * therefore this needs to be checked too.
411 */
412 if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
413 /*
414 * When this happens, userspace did something that it
415 * was not supposed to do, e.g. branching into secure
416 * memory. Trigger a segmentation fault.
417 */
418 if (user_mode(regs)) {
419 send_sig(SIGSEGV, current, 0);
420 return;
421 }
422 /*
423 * The kernel should never run into this case and
424 * there is no way out of this situation.
425 */
426 panic("Unexpected PGM 0x3d with TEID bit 61=0");
427 }
428 if (is_kernel_fault(regs)) {
429 folio = phys_to_folio(addr);
430 if (unlikely(!folio_try_get(folio)))
431 return;
432 rc = arch_make_folio_accessible(folio);
433 folio_put(folio);
434 if (rc)
435 BUG();
436 } else {
437 mm = current->mm;
438 mmap_read_lock(mm);
439 vma = find_vma(mm, addr);
440 if (!vma)
441 return handle_fault_error(regs, SEGV_MAPERR);
442 folio = folio_walk_start(&fw, vma, addr, 0);
443 if (!folio) {
444 mmap_read_unlock(mm);
445 return;
446 }
447 /* arch_make_folio_accessible() needs a raised refcount. */
448 folio_get(folio);
449 rc = arch_make_folio_accessible(folio);
450 folio_put(folio);
451 folio_walk_end(&fw, vma);
452 if (rc)
453 send_sig(SIGSEGV, current, 0);
454 mmap_read_unlock(mm);
455 }
456}
457NOKPROBE_SYMBOL(do_secure_storage_access);
458
459#endif /* CONFIG_PGSTE */