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 <linux/kfence.h>
35#include <asm/asm-extable.h>
36#include <asm/asm-offsets.h>
37#include <asm/diag.h>
38#include <asm/gmap.h>
39#include <asm/irq.h>
40#include <asm/mmu_context.h>
41#include <asm/facility.h>
42#include <asm/uv.h>
43#include "../kernel/entry.h"
44
45#define __FAIL_ADDR_MASK -4096L
46#define __SUBCODE_MASK 0x0600
47#define __PF_RES_FIELD 0x8000000000000000ULL
48
49#define VM_FAULT_BADCONTEXT ((__force vm_fault_t) 0x010000)
50#define VM_FAULT_BADMAP ((__force vm_fault_t) 0x020000)
51#define VM_FAULT_BADACCESS ((__force vm_fault_t) 0x040000)
52#define VM_FAULT_SIGNAL ((__force vm_fault_t) 0x080000)
53#define VM_FAULT_PFAULT ((__force vm_fault_t) 0x100000)
54
55enum fault_type {
56 KERNEL_FAULT,
57 USER_FAULT,
58 GMAP_FAULT,
59};
60
61static unsigned long store_indication __read_mostly;
62
63static int __init fault_init(void)
64{
65 if (test_facility(75))
66 store_indication = 0xc00;
67 return 0;
68}
69early_initcall(fault_init);
70
71/*
72 * Find out which address space caused the exception.
73 */
74static enum fault_type get_fault_type(struct pt_regs *regs)
75{
76 unsigned long trans_exc_code;
77
78 trans_exc_code = regs->int_parm_long & 3;
79 if (likely(trans_exc_code == 0)) {
80 /* primary space exception */
81 if (user_mode(regs))
82 return USER_FAULT;
83 if (!IS_ENABLED(CONFIG_PGSTE))
84 return KERNEL_FAULT;
85 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT))
86 return GMAP_FAULT;
87 return KERNEL_FAULT;
88 }
89 if (trans_exc_code == 2)
90 return USER_FAULT;
91 if (trans_exc_code == 1) {
92 /* access register mode, not used in the kernel */
93 return USER_FAULT;
94 }
95 /* home space exception -> access via kernel ASCE */
96 return KERNEL_FAULT;
97}
98
99static int bad_address(void *p)
100{
101 unsigned long dummy;
102
103 return get_kernel_nofault(dummy, (unsigned long *)p);
104}
105
106static void dump_pagetable(unsigned long asce, unsigned long address)
107{
108 unsigned long *table = __va(asce & _ASCE_ORIGIN);
109
110 pr_alert("AS:%016lx ", asce);
111 switch (asce & _ASCE_TYPE_MASK) {
112 case _ASCE_TYPE_REGION1:
113 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
114 if (bad_address(table))
115 goto bad;
116 pr_cont("R1:%016lx ", *table);
117 if (*table & _REGION_ENTRY_INVALID)
118 goto out;
119 table = __va(*table & _REGION_ENTRY_ORIGIN);
120 fallthrough;
121 case _ASCE_TYPE_REGION2:
122 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
123 if (bad_address(table))
124 goto bad;
125 pr_cont("R2:%016lx ", *table);
126 if (*table & _REGION_ENTRY_INVALID)
127 goto out;
128 table = __va(*table & _REGION_ENTRY_ORIGIN);
129 fallthrough;
130 case _ASCE_TYPE_REGION3:
131 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
132 if (bad_address(table))
133 goto bad;
134 pr_cont("R3:%016lx ", *table);
135 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
136 goto out;
137 table = __va(*table & _REGION_ENTRY_ORIGIN);
138 fallthrough;
139 case _ASCE_TYPE_SEGMENT:
140 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
141 if (bad_address(table))
142 goto bad;
143 pr_cont("S:%016lx ", *table);
144 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
145 goto out;
146 table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
147 }
148 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
149 if (bad_address(table))
150 goto bad;
151 pr_cont("P:%016lx ", *table);
152out:
153 pr_cont("\n");
154 return;
155bad:
156 pr_cont("BAD\n");
157}
158
159static void dump_fault_info(struct pt_regs *regs)
160{
161 unsigned long asce;
162
163 pr_alert("Failing address: %016lx TEID: %016lx\n",
164 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
165 pr_alert("Fault in ");
166 switch (regs->int_parm_long & 3) {
167 case 3:
168 pr_cont("home space ");
169 break;
170 case 2:
171 pr_cont("secondary space ");
172 break;
173 case 1:
174 pr_cont("access register ");
175 break;
176 case 0:
177 pr_cont("primary space ");
178 break;
179 }
180 pr_cont("mode while using ");
181 switch (get_fault_type(regs)) {
182 case USER_FAULT:
183 asce = S390_lowcore.user_asce;
184 pr_cont("user ");
185 break;
186 case GMAP_FAULT:
187 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
188 pr_cont("gmap ");
189 break;
190 case KERNEL_FAULT:
191 asce = S390_lowcore.kernel_asce;
192 pr_cont("kernel ");
193 break;
194 default:
195 unreachable();
196 }
197 pr_cont("ASCE.\n");
198 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
199}
200
201int show_unhandled_signals = 1;
202
203void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
204{
205 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
206 return;
207 if (!unhandled_signal(current, signr))
208 return;
209 if (!printk_ratelimit())
210 return;
211 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
212 regs->int_code & 0xffff, regs->int_code >> 17);
213 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
214 printk(KERN_CONT "\n");
215 if (is_mm_fault)
216 dump_fault_info(regs);
217 show_regs(regs);
218}
219
220/*
221 * Send SIGSEGV to task. This is an external routine
222 * to keep the stack usage of do_page_fault small.
223 */
224static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
225{
226 report_user_fault(regs, SIGSEGV, 1);
227 force_sig_fault(SIGSEGV, si_code,
228 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
229}
230
231static noinline void do_no_context(struct pt_regs *regs)
232{
233 if (fixup_exception(regs))
234 return;
235 /*
236 * Oops. The kernel tried to access some bad page. We'll have to
237 * terminate things with extreme prejudice.
238 */
239 if (get_fault_type(regs) == KERNEL_FAULT)
240 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
241 " in virtual kernel address space\n");
242 else
243 printk(KERN_ALERT "Unable to handle kernel paging request"
244 " in virtual user address space\n");
245 dump_fault_info(regs);
246 die(regs, "Oops");
247}
248
249static noinline void do_low_address(struct pt_regs *regs)
250{
251 /* Low-address protection hit in kernel mode means
252 NULL pointer write access in kernel mode. */
253 if (regs->psw.mask & PSW_MASK_PSTATE) {
254 /* Low-address protection hit in user mode 'cannot happen'. */
255 die (regs, "Low-address protection");
256 }
257
258 do_no_context(regs);
259}
260
261static noinline void do_sigbus(struct pt_regs *regs)
262{
263 /*
264 * Send a sigbus, regardless of whether we were in kernel
265 * or user mode.
266 */
267 force_sig_fault(SIGBUS, BUS_ADRERR,
268 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK));
269}
270
271static noinline void do_fault_error(struct pt_regs *regs, vm_fault_t fault)
272{
273 int si_code;
274
275 switch (fault) {
276 case VM_FAULT_BADACCESS:
277 case VM_FAULT_BADMAP:
278 /* Bad memory access. Check if it is kernel or user space. */
279 if (user_mode(regs)) {
280 /* User mode accesses just cause a SIGSEGV */
281 si_code = (fault == VM_FAULT_BADMAP) ?
282 SEGV_MAPERR : SEGV_ACCERR;
283 do_sigsegv(regs, si_code);
284 break;
285 }
286 fallthrough;
287 case VM_FAULT_BADCONTEXT:
288 case VM_FAULT_PFAULT:
289 do_no_context(regs);
290 break;
291 case VM_FAULT_SIGNAL:
292 if (!user_mode(regs))
293 do_no_context(regs);
294 break;
295 default: /* fault & VM_FAULT_ERROR */
296 if (fault & VM_FAULT_OOM) {
297 if (!user_mode(regs))
298 do_no_context(regs);
299 else
300 pagefault_out_of_memory();
301 } else if (fault & VM_FAULT_SIGSEGV) {
302 /* Kernel mode? Handle exceptions or die */
303 if (!user_mode(regs))
304 do_no_context(regs);
305 else
306 do_sigsegv(regs, SEGV_MAPERR);
307 } else if (fault & VM_FAULT_SIGBUS) {
308 /* Kernel mode? Handle exceptions or die */
309 if (!user_mode(regs))
310 do_no_context(regs);
311 else
312 do_sigbus(regs);
313 } else
314 BUG();
315 break;
316 }
317}
318
319/*
320 * This routine handles page faults. It determines the address,
321 * and the problem, and then passes it off to one of the appropriate
322 * routines.
323 *
324 * interruption code (int_code):
325 * 04 Protection -> Write-Protection (suppression)
326 * 10 Segment translation -> Not present (nullification)
327 * 11 Page translation -> Not present (nullification)
328 * 3b Region third trans. -> Not present (nullification)
329 */
330static inline vm_fault_t do_exception(struct pt_regs *regs, int access)
331{
332 struct gmap *gmap;
333 struct task_struct *tsk;
334 struct mm_struct *mm;
335 struct vm_area_struct *vma;
336 enum fault_type type;
337 unsigned long trans_exc_code;
338 unsigned long address;
339 unsigned int flags;
340 vm_fault_t fault;
341 bool is_write;
342
343 tsk = current;
344 /*
345 * The instruction that caused the program check has
346 * been nullified. Don't signal single step via SIGTRAP.
347 */
348 clear_thread_flag(TIF_PER_TRAP);
349
350 if (kprobe_page_fault(regs, 14))
351 return 0;
352
353 mm = tsk->mm;
354 trans_exc_code = regs->int_parm_long;
355 address = trans_exc_code & __FAIL_ADDR_MASK;
356 is_write = (trans_exc_code & store_indication) == 0x400;
357
358 /*
359 * Verify that the fault happened in user space, that
360 * we are not in an interrupt and that there is a
361 * user context.
362 */
363 fault = VM_FAULT_BADCONTEXT;
364 type = get_fault_type(regs);
365 switch (type) {
366 case KERNEL_FAULT:
367 if (kfence_handle_page_fault(address, is_write, regs))
368 return 0;
369 goto out;
370 case USER_FAULT:
371 case GMAP_FAULT:
372 if (faulthandler_disabled() || !mm)
373 goto out;
374 break;
375 }
376
377 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
378 flags = FAULT_FLAG_DEFAULT;
379 if (user_mode(regs))
380 flags |= FAULT_FLAG_USER;
381 if (is_write)
382 access = VM_WRITE;
383 if (access == VM_WRITE)
384 flags |= FAULT_FLAG_WRITE;
385 mmap_read_lock(mm);
386
387 gmap = NULL;
388 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
389 gmap = (struct gmap *) S390_lowcore.gmap;
390 current->thread.gmap_addr = address;
391 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
392 current->thread.gmap_int_code = regs->int_code & 0xffff;
393 address = __gmap_translate(gmap, address);
394 if (address == -EFAULT) {
395 fault = VM_FAULT_BADMAP;
396 goto out_up;
397 }
398 if (gmap->pfault_enabled)
399 flags |= FAULT_FLAG_RETRY_NOWAIT;
400 }
401
402retry:
403 fault = VM_FAULT_BADMAP;
404 vma = find_vma(mm, address);
405 if (!vma)
406 goto out_up;
407
408 if (unlikely(vma->vm_start > address)) {
409 if (!(vma->vm_flags & VM_GROWSDOWN))
410 goto out_up;
411 if (expand_stack(vma, address))
412 goto out_up;
413 }
414
415 /*
416 * Ok, we have a good vm_area for this memory access, so
417 * we can handle it..
418 */
419 fault = VM_FAULT_BADACCESS;
420 if (unlikely(!(vma->vm_flags & access)))
421 goto out_up;
422
423 /*
424 * If for any reason at all we couldn't handle the fault,
425 * make sure we exit gracefully rather than endlessly redo
426 * the fault.
427 */
428 fault = handle_mm_fault(vma, address, flags, regs);
429 if (fault_signal_pending(fault, regs)) {
430 fault = VM_FAULT_SIGNAL;
431 if (flags & FAULT_FLAG_RETRY_NOWAIT)
432 goto out_up;
433 goto out;
434 }
435
436 /* The fault is fully completed (including releasing mmap lock) */
437 if (fault & VM_FAULT_COMPLETED) {
438 if (gmap) {
439 mmap_read_lock(mm);
440 goto out_gmap;
441 }
442 fault = 0;
443 goto out;
444 }
445
446 if (unlikely(fault & VM_FAULT_ERROR))
447 goto out_up;
448
449 if (fault & VM_FAULT_RETRY) {
450 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
451 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
452 /*
453 * FAULT_FLAG_RETRY_NOWAIT has been set, mmap_lock has
454 * not been released
455 */
456 current->thread.gmap_pfault = 1;
457 fault = VM_FAULT_PFAULT;
458 goto out_up;
459 }
460 flags &= ~FAULT_FLAG_RETRY_NOWAIT;
461 flags |= FAULT_FLAG_TRIED;
462 mmap_read_lock(mm);
463 goto retry;
464 }
465out_gmap:
466 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
467 address = __gmap_link(gmap, current->thread.gmap_addr,
468 address);
469 if (address == -EFAULT) {
470 fault = VM_FAULT_BADMAP;
471 goto out_up;
472 }
473 if (address == -ENOMEM) {
474 fault = VM_FAULT_OOM;
475 goto out_up;
476 }
477 }
478 fault = 0;
479out_up:
480 mmap_read_unlock(mm);
481out:
482 return fault;
483}
484
485void do_protection_exception(struct pt_regs *regs)
486{
487 unsigned long trans_exc_code;
488 int access;
489 vm_fault_t fault;
490
491 trans_exc_code = regs->int_parm_long;
492 /*
493 * Protection exceptions are suppressing, decrement psw address.
494 * The exception to this rule are aborted transactions, for these
495 * the PSW already points to the correct location.
496 */
497 if (!(regs->int_code & 0x200))
498 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
499 /*
500 * Check for low-address protection. This needs to be treated
501 * as a special case because the translation exception code
502 * field is not guaranteed to contain valid data in this case.
503 */
504 if (unlikely(!(trans_exc_code & 4))) {
505 do_low_address(regs);
506 return;
507 }
508 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
509 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
510 (regs->psw.addr & PAGE_MASK);
511 access = VM_EXEC;
512 fault = VM_FAULT_BADACCESS;
513 } else {
514 access = VM_WRITE;
515 fault = do_exception(regs, access);
516 }
517 if (unlikely(fault))
518 do_fault_error(regs, fault);
519}
520NOKPROBE_SYMBOL(do_protection_exception);
521
522void do_dat_exception(struct pt_regs *regs)
523{
524 int access;
525 vm_fault_t fault;
526
527 access = VM_ACCESS_FLAGS;
528 fault = do_exception(regs, access);
529 if (unlikely(fault))
530 do_fault_error(regs, fault);
531}
532NOKPROBE_SYMBOL(do_dat_exception);
533
534#ifdef CONFIG_PFAULT
535/*
536 * 'pfault' pseudo page faults routines.
537 */
538static int pfault_disable;
539
540static int __init nopfault(char *str)
541{
542 pfault_disable = 1;
543 return 1;
544}
545
546__setup("nopfault", nopfault);
547
548struct pfault_refbk {
549 u16 refdiagc;
550 u16 reffcode;
551 u16 refdwlen;
552 u16 refversn;
553 u64 refgaddr;
554 u64 refselmk;
555 u64 refcmpmk;
556 u64 reserved;
557} __attribute__ ((packed, aligned(8)));
558
559static struct pfault_refbk pfault_init_refbk = {
560 .refdiagc = 0x258,
561 .reffcode = 0,
562 .refdwlen = 5,
563 .refversn = 2,
564 .refgaddr = __LC_LPP,
565 .refselmk = 1ULL << 48,
566 .refcmpmk = 1ULL << 48,
567 .reserved = __PF_RES_FIELD
568};
569
570int pfault_init(void)
571{
572 int rc;
573
574 if (pfault_disable)
575 return -1;
576 diag_stat_inc(DIAG_STAT_X258);
577 asm volatile(
578 " diag %1,%0,0x258\n"
579 "0: j 2f\n"
580 "1: la %0,8\n"
581 "2:\n"
582 EX_TABLE(0b,1b)
583 : "=d" (rc)
584 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc");
585 return rc;
586}
587
588static struct pfault_refbk pfault_fini_refbk = {
589 .refdiagc = 0x258,
590 .reffcode = 1,
591 .refdwlen = 5,
592 .refversn = 2,
593};
594
595void pfault_fini(void)
596{
597
598 if (pfault_disable)
599 return;
600 diag_stat_inc(DIAG_STAT_X258);
601 asm volatile(
602 " diag %0,0,0x258\n"
603 "0: nopr %%r7\n"
604 EX_TABLE(0b,0b)
605 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc");
606}
607
608static DEFINE_SPINLOCK(pfault_lock);
609static LIST_HEAD(pfault_list);
610
611#define PF_COMPLETE 0x0080
612
613/*
614 * The mechanism of our pfault code: if Linux is running as guest, runs a user
615 * space process and the user space process accesses a page that the host has
616 * paged out we get a pfault interrupt.
617 *
618 * This allows us, within the guest, to schedule a different process. Without
619 * this mechanism the host would have to suspend the whole virtual cpu until
620 * the page has been paged in.
621 *
622 * So when we get such an interrupt then we set the state of the current task
623 * to uninterruptible and also set the need_resched flag. Both happens within
624 * interrupt context(!). If we later on want to return to user space we
625 * recognize the need_resched flag and then call schedule(). It's not very
626 * obvious how this works...
627 *
628 * Of course we have a lot of additional fun with the completion interrupt (->
629 * host signals that a page of a process has been paged in and the process can
630 * continue to run). This interrupt can arrive on any cpu and, since we have
631 * virtual cpus, actually appear before the interrupt that signals that a page
632 * is missing.
633 */
634static void pfault_interrupt(struct ext_code ext_code,
635 unsigned int param32, unsigned long param64)
636{
637 struct task_struct *tsk;
638 __u16 subcode;
639 pid_t pid;
640
641 /*
642 * Get the external interruption subcode & pfault initial/completion
643 * signal bit. VM stores this in the 'cpu address' field associated
644 * with the external interrupt.
645 */
646 subcode = ext_code.subcode;
647 if ((subcode & 0xff00) != __SUBCODE_MASK)
648 return;
649 inc_irq_stat(IRQEXT_PFL);
650 /* Get the token (= pid of the affected task). */
651 pid = param64 & LPP_PID_MASK;
652 rcu_read_lock();
653 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
654 if (tsk)
655 get_task_struct(tsk);
656 rcu_read_unlock();
657 if (!tsk)
658 return;
659 spin_lock(&pfault_lock);
660 if (subcode & PF_COMPLETE) {
661 /* signal bit is set -> a page has been swapped in by VM */
662 if (tsk->thread.pfault_wait == 1) {
663 /* Initial interrupt was faster than the completion
664 * interrupt. pfault_wait is valid. Set pfault_wait
665 * back to zero and wake up the process. This can
666 * safely be done because the task is still sleeping
667 * and can't produce new pfaults. */
668 tsk->thread.pfault_wait = 0;
669 list_del(&tsk->thread.list);
670 wake_up_process(tsk);
671 put_task_struct(tsk);
672 } else {
673 /* Completion interrupt was faster than initial
674 * interrupt. Set pfault_wait to -1 so the initial
675 * interrupt doesn't put the task to sleep.
676 * If the task is not running, ignore the completion
677 * interrupt since it must be a leftover of a PFAULT
678 * CANCEL operation which didn't remove all pending
679 * completion interrupts. */
680 if (task_is_running(tsk))
681 tsk->thread.pfault_wait = -1;
682 }
683 } else {
684 /* signal bit not set -> a real page is missing. */
685 if (WARN_ON_ONCE(tsk != current))
686 goto out;
687 if (tsk->thread.pfault_wait == 1) {
688 /* Already on the list with a reference: put to sleep */
689 goto block;
690 } else if (tsk->thread.pfault_wait == -1) {
691 /* Completion interrupt was faster than the initial
692 * interrupt (pfault_wait == -1). Set pfault_wait
693 * back to zero and exit. */
694 tsk->thread.pfault_wait = 0;
695 } else {
696 /* Initial interrupt arrived before completion
697 * interrupt. Let the task sleep.
698 * An extra task reference is needed since a different
699 * cpu may set the task state to TASK_RUNNING again
700 * before the scheduler is reached. */
701 get_task_struct(tsk);
702 tsk->thread.pfault_wait = 1;
703 list_add(&tsk->thread.list, &pfault_list);
704block:
705 /* Since this must be a userspace fault, there
706 * is no kernel task state to trample. Rely on the
707 * return to userspace schedule() to block. */
708 __set_current_state(TASK_UNINTERRUPTIBLE);
709 set_tsk_need_resched(tsk);
710 set_preempt_need_resched();
711 }
712 }
713out:
714 spin_unlock(&pfault_lock);
715 put_task_struct(tsk);
716}
717
718static int pfault_cpu_dead(unsigned int cpu)
719{
720 struct thread_struct *thread, *next;
721 struct task_struct *tsk;
722
723 spin_lock_irq(&pfault_lock);
724 list_for_each_entry_safe(thread, next, &pfault_list, list) {
725 thread->pfault_wait = 0;
726 list_del(&thread->list);
727 tsk = container_of(thread, struct task_struct, thread);
728 wake_up_process(tsk);
729 put_task_struct(tsk);
730 }
731 spin_unlock_irq(&pfault_lock);
732 return 0;
733}
734
735static int __init pfault_irq_init(void)
736{
737 int rc;
738
739 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
740 if (rc)
741 goto out_extint;
742 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
743 if (rc)
744 goto out_pfault;
745 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
746 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
747 NULL, pfault_cpu_dead);
748 return 0;
749
750out_pfault:
751 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
752out_extint:
753 pfault_disable = 1;
754 return rc;
755}
756early_initcall(pfault_irq_init);
757
758#endif /* CONFIG_PFAULT */
759
760#if IS_ENABLED(CONFIG_PGSTE)
761
762void do_secure_storage_access(struct pt_regs *regs)
763{
764 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK;
765 struct vm_area_struct *vma;
766 struct mm_struct *mm;
767 struct page *page;
768 struct gmap *gmap;
769 int rc;
770
771 /*
772 * bit 61 tells us if the address is valid, if it's not we
773 * have a major problem and should stop the kernel or send a
774 * SIGSEGV to the process. Unfortunately bit 61 is not
775 * reliable without the misc UV feature so we need to check
776 * for that as well.
777 */
778 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
779 !test_bit_inv(61, ®s->int_parm_long)) {
780 /*
781 * When this happens, userspace did something that it
782 * was not supposed to do, e.g. branching into secure
783 * memory. Trigger a segmentation fault.
784 */
785 if (user_mode(regs)) {
786 send_sig(SIGSEGV, current, 0);
787 return;
788 }
789
790 /*
791 * The kernel should never run into this case and we
792 * have no way out of this situation.
793 */
794 panic("Unexpected PGM 0x3d with TEID bit 61=0");
795 }
796
797 switch (get_fault_type(regs)) {
798 case GMAP_FAULT:
799 mm = current->mm;
800 gmap = (struct gmap *)S390_lowcore.gmap;
801 mmap_read_lock(mm);
802 addr = __gmap_translate(gmap, addr);
803 mmap_read_unlock(mm);
804 if (IS_ERR_VALUE(addr)) {
805 do_fault_error(regs, VM_FAULT_BADMAP);
806 break;
807 }
808 fallthrough;
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_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 default:
838 do_fault_error(regs, VM_FAULT_BADMAP);
839 WARN_ON_ONCE(1);
840 }
841}
842NOKPROBE_SYMBOL(do_secure_storage_access);
843
844void do_non_secure_storage_access(struct pt_regs *regs)
845{
846 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
847 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
848
849 if (get_fault_type(regs) != GMAP_FAULT) {
850 do_fault_error(regs, VM_FAULT_BADMAP);
851 WARN_ON_ONCE(1);
852 return;
853 }
854
855 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL)
856 send_sig(SIGSEGV, current, 0);
857}
858NOKPROBE_SYMBOL(do_non_secure_storage_access);
859
860void do_secure_storage_violation(struct pt_regs *regs)
861{
862 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK;
863 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap;
864
865 /*
866 * If the VM has been rebooted, its address space might still contain
867 * secure pages from the previous boot.
868 * Clear the page so it can be reused.
869 */
870 if (!gmap_destroy_page(gmap, gaddr))
871 return;
872 /*
873 * Either KVM messed up the secure guest mapping or the same
874 * page is mapped into multiple secure guests.
875 *
876 * This exception is only triggered when a guest 2 is running
877 * and can therefore never occur in kernel context.
878 */
879 printk_ratelimited(KERN_WARNING
880 "Secure storage violation in task: %s, pid %d\n",
881 current->comm, current->pid);
882 send_sig(SIGSEGV, current, 0);
883}
884
885#endif /* CONFIG_PGSTE */
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * S390 version
4 * Copyright IBM Corp. 1999
5 * Author(s): Hartmut Penner (hp@de.ibm.com)
6 * Ulrich Weigand (uweigand@de.ibm.com)
7 *
8 * Derived from "arch/i386/mm/fault.c"
9 * Copyright (C) 1995 Linus Torvalds
10 */
11
12#include <linux/kernel_stat.h>
13#include <linux/perf_event.h>
14#include <linux/signal.h>
15#include <linux/sched.h>
16#include <linux/sched/debug.h>
17#include <linux/kernel.h>
18#include <linux/errno.h>
19#include <linux/string.h>
20#include <linux/types.h>
21#include <linux/ptrace.h>
22#include <linux/mman.h>
23#include <linux/mm.h>
24#include <linux/compat.h>
25#include <linux/smp.h>
26#include <linux/kdebug.h>
27#include <linux/init.h>
28#include <linux/console.h>
29#include <linux/extable.h>
30#include <linux/hardirq.h>
31#include <linux/kprobes.h>
32#include <linux/uaccess.h>
33#include <linux/hugetlb.h>
34#include <asm/asm-offsets.h>
35#include <asm/diag.h>
36#include <asm/pgtable.h>
37#include <asm/gmap.h>
38#include <asm/irq.h>
39#include <asm/mmu_context.h>
40#include <asm/facility.h>
41#include "../kernel/entry.h"
42
43#define __FAIL_ADDR_MASK -4096L
44#define __SUBCODE_MASK 0x0600
45#define __PF_RES_FIELD 0x8000000000000000ULL
46
47#define VM_FAULT_BADCONTEXT 0x010000
48#define VM_FAULT_BADMAP 0x020000
49#define VM_FAULT_BADACCESS 0x040000
50#define VM_FAULT_SIGNAL 0x080000
51#define VM_FAULT_PFAULT 0x100000
52
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
70static inline int notify_page_fault(struct pt_regs *regs)
71{
72 int ret = 0;
73
74 /* kprobe_running() needs smp_processor_id() */
75 if (kprobes_built_in() && !user_mode(regs)) {
76 preempt_disable();
77 if (kprobe_running() && kprobe_fault_handler(regs, 14))
78 ret = 1;
79 preempt_enable();
80 }
81 return ret;
82}
83
84
85/*
86 * Unlock any spinlocks which will prevent us from getting the
87 * message out.
88 */
89void bust_spinlocks(int yes)
90{
91 if (yes) {
92 oops_in_progress = 1;
93 } else {
94 int loglevel_save = console_loglevel;
95 console_unblank();
96 oops_in_progress = 0;
97 /*
98 * OK, the message is on the console. Now we call printk()
99 * without oops_in_progress set so that printk will give klogd
100 * a poke. Hold onto your hats...
101 */
102 console_loglevel = 15;
103 printk(" ");
104 console_loglevel = loglevel_save;
105 }
106}
107
108/*
109 * Find out which address space caused the exception.
110 * Access register mode is impossible, ignore space == 3.
111 */
112static inline enum fault_type get_fault_type(struct pt_regs *regs)
113{
114 unsigned long trans_exc_code;
115
116 trans_exc_code = regs->int_parm_long & 3;
117 if (likely(trans_exc_code == 0)) {
118 /* primary space exception */
119 if (IS_ENABLED(CONFIG_PGSTE) &&
120 test_pt_regs_flag(regs, PIF_GUEST_FAULT))
121 return GMAP_FAULT;
122 if (current->thread.mm_segment == USER_DS)
123 return USER_FAULT;
124 return KERNEL_FAULT;
125 }
126 if (trans_exc_code == 2) {
127 /* secondary space exception */
128 if (current->thread.mm_segment & 1) {
129 if (current->thread.mm_segment == USER_DS_SACF)
130 return USER_FAULT;
131 return KERNEL_FAULT;
132 }
133 return VDSO_FAULT;
134 }
135 /* home space exception -> access via kernel ASCE */
136 return KERNEL_FAULT;
137}
138
139static int bad_address(void *p)
140{
141 unsigned long dummy;
142
143 return probe_kernel_address((unsigned long *)p, dummy);
144}
145
146static void dump_pagetable(unsigned long asce, unsigned long address)
147{
148 unsigned long *table = __va(asce & _ASCE_ORIGIN);
149
150 pr_alert("AS:%016lx ", asce);
151 switch (asce & _ASCE_TYPE_MASK) {
152 case _ASCE_TYPE_REGION1:
153 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
154 if (bad_address(table))
155 goto bad;
156 pr_cont("R1:%016lx ", *table);
157 if (*table & _REGION_ENTRY_INVALID)
158 goto out;
159 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
160 /* fallthrough */
161 case _ASCE_TYPE_REGION2:
162 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
163 if (bad_address(table))
164 goto bad;
165 pr_cont("R2:%016lx ", *table);
166 if (*table & _REGION_ENTRY_INVALID)
167 goto out;
168 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
169 /* fallthrough */
170 case _ASCE_TYPE_REGION3:
171 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
172 if (bad_address(table))
173 goto bad;
174 pr_cont("R3:%016lx ", *table);
175 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
176 goto out;
177 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
178 /* fallthrough */
179 case _ASCE_TYPE_SEGMENT:
180 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
181 if (bad_address(table))
182 goto bad;
183 pr_cont("S:%016lx ", *table);
184 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
185 goto out;
186 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
187 }
188 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT;
189 if (bad_address(table))
190 goto bad;
191 pr_cont("P:%016lx ", *table);
192out:
193 pr_cont("\n");
194 return;
195bad:
196 pr_cont("BAD\n");
197}
198
199static void dump_fault_info(struct pt_regs *regs)
200{
201 unsigned long asce;
202
203 pr_alert("Failing address: %016lx TEID: %016lx\n",
204 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long);
205 pr_alert("Fault in ");
206 switch (regs->int_parm_long & 3) {
207 case 3:
208 pr_cont("home space ");
209 break;
210 case 2:
211 pr_cont("secondary space ");
212 break;
213 case 1:
214 pr_cont("access register ");
215 break;
216 case 0:
217 pr_cont("primary space ");
218 break;
219 }
220 pr_cont("mode while using ");
221 switch (get_fault_type(regs)) {
222 case USER_FAULT:
223 asce = S390_lowcore.user_asce;
224 pr_cont("user ");
225 break;
226 case VDSO_FAULT:
227 asce = S390_lowcore.vdso_asce;
228 pr_cont("vdso ");
229 break;
230 case GMAP_FAULT:
231 asce = ((struct gmap *) S390_lowcore.gmap)->asce;
232 pr_cont("gmap ");
233 break;
234 case KERNEL_FAULT:
235 asce = S390_lowcore.kernel_asce;
236 pr_cont("kernel ");
237 break;
238 }
239 pr_cont("ASCE.\n");
240 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK);
241}
242
243int show_unhandled_signals = 1;
244
245void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
246{
247 if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
248 return;
249 if (!unhandled_signal(current, signr))
250 return;
251 if (!printk_ratelimit())
252 return;
253 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ",
254 regs->int_code & 0xffff, regs->int_code >> 17);
255 print_vma_addr(KERN_CONT "in ", regs->psw.addr);
256 printk(KERN_CONT "\n");
257 if (is_mm_fault)
258 dump_fault_info(regs);
259 show_regs(regs);
260}
261
262/*
263 * Send SIGSEGV to task. This is an external routine
264 * to keep the stack usage of do_page_fault small.
265 */
266static noinline void do_sigsegv(struct pt_regs *regs, int si_code)
267{
268 struct siginfo si;
269
270 report_user_fault(regs, SIGSEGV, 1);
271 si.si_signo = SIGSEGV;
272 si.si_errno = 0;
273 si.si_code = si_code;
274 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
275 force_sig_info(SIGSEGV, &si, current);
276}
277
278static noinline void do_no_context(struct pt_regs *regs)
279{
280 const struct exception_table_entry *fixup;
281
282 /* Are we prepared to handle this kernel fault? */
283 fixup = search_exception_tables(regs->psw.addr);
284 if (fixup) {
285 regs->psw.addr = extable_fixup(fixup);
286 return;
287 }
288
289 /*
290 * Oops. The kernel tried to access some bad page. We'll have to
291 * terminate things with extreme prejudice.
292 */
293 if (get_fault_type(regs) == KERNEL_FAULT)
294 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
295 " in virtual kernel address space\n");
296 else
297 printk(KERN_ALERT "Unable to handle kernel paging request"
298 " in virtual user address space\n");
299 dump_fault_info(regs);
300 die(regs, "Oops");
301 do_exit(SIGKILL);
302}
303
304static noinline void do_low_address(struct pt_regs *regs)
305{
306 /* Low-address protection hit in kernel mode means
307 NULL pointer write access in kernel mode. */
308 if (regs->psw.mask & PSW_MASK_PSTATE) {
309 /* Low-address protection hit in user mode 'cannot happen'. */
310 die (regs, "Low-address protection");
311 do_exit(SIGKILL);
312 }
313
314 do_no_context(regs);
315}
316
317static noinline void do_sigbus(struct pt_regs *regs)
318{
319 struct task_struct *tsk = current;
320 struct siginfo si;
321
322 /*
323 * Send a sigbus, regardless of whether we were in kernel
324 * or user mode.
325 */
326 si.si_signo = SIGBUS;
327 si.si_errno = 0;
328 si.si_code = BUS_ADRERR;
329 si.si_addr = (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK);
330 force_sig_info(SIGBUS, &si, tsk);
331}
332
333static noinline int signal_return(struct pt_regs *regs)
334{
335 u16 instruction;
336 int rc;
337
338 rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
339 if (rc)
340 return rc;
341 if (instruction == 0x0a77) {
342 set_pt_regs_flag(regs, PIF_SYSCALL);
343 regs->int_code = 0x00040077;
344 return 0;
345 } else if (instruction == 0x0aad) {
346 set_pt_regs_flag(regs, PIF_SYSCALL);
347 regs->int_code = 0x000400ad;
348 return 0;
349 }
350 return -EACCES;
351}
352
353static noinline void do_fault_error(struct pt_regs *regs, int access, int fault)
354{
355 int si_code;
356
357 switch (fault) {
358 case VM_FAULT_BADACCESS:
359 if (access == VM_EXEC && signal_return(regs) == 0)
360 break;
361 case VM_FAULT_BADMAP:
362 /* Bad memory access. Check if it is kernel or user space. */
363 if (user_mode(regs)) {
364 /* User mode accesses just cause a SIGSEGV */
365 si_code = (fault == VM_FAULT_BADMAP) ?
366 SEGV_MAPERR : SEGV_ACCERR;
367 do_sigsegv(regs, si_code);
368 break;
369 }
370 case VM_FAULT_BADCONTEXT:
371 case VM_FAULT_PFAULT:
372 do_no_context(regs);
373 break;
374 case VM_FAULT_SIGNAL:
375 if (!user_mode(regs))
376 do_no_context(regs);
377 break;
378 default: /* fault & VM_FAULT_ERROR */
379 if (fault & VM_FAULT_OOM) {
380 if (!user_mode(regs))
381 do_no_context(regs);
382 else
383 pagefault_out_of_memory();
384 } else if (fault & VM_FAULT_SIGSEGV) {
385 /* Kernel mode? Handle exceptions or die */
386 if (!user_mode(regs))
387 do_no_context(regs);
388 else
389 do_sigsegv(regs, SEGV_MAPERR);
390 } else if (fault & VM_FAULT_SIGBUS) {
391 /* Kernel mode? Handle exceptions or die */
392 if (!user_mode(regs))
393 do_no_context(regs);
394 else
395 do_sigbus(regs);
396 } else
397 BUG();
398 break;
399 }
400}
401
402/*
403 * This routine handles page faults. It determines the address,
404 * and the problem, and then passes it off to one of the appropriate
405 * routines.
406 *
407 * interruption code (int_code):
408 * 04 Protection -> Write-Protection (suprression)
409 * 10 Segment translation -> Not present (nullification)
410 * 11 Page translation -> Not present (nullification)
411 * 3b Region third trans. -> Not present (nullification)
412 */
413static inline int do_exception(struct pt_regs *regs, int access)
414{
415 struct gmap *gmap;
416 struct task_struct *tsk;
417 struct mm_struct *mm;
418 struct vm_area_struct *vma;
419 enum fault_type type;
420 unsigned long trans_exc_code;
421 unsigned long address;
422 unsigned int flags;
423 int fault;
424
425 tsk = current;
426 /*
427 * The instruction that caused the program check has
428 * been nullified. Don't signal single step via SIGTRAP.
429 */
430 clear_pt_regs_flag(regs, PIF_PER_TRAP);
431
432 if (notify_page_fault(regs))
433 return 0;
434
435 mm = tsk->mm;
436 trans_exc_code = regs->int_parm_long;
437
438 /*
439 * Verify that the fault happened in user space, that
440 * we are not in an interrupt and that there is a
441 * user context.
442 */
443 fault = VM_FAULT_BADCONTEXT;
444 type = get_fault_type(regs);
445 switch (type) {
446 case KERNEL_FAULT:
447 goto out;
448 case VDSO_FAULT:
449 fault = VM_FAULT_BADMAP;
450 goto out;
451 case USER_FAULT:
452 case GMAP_FAULT:
453 if (faulthandler_disabled() || !mm)
454 goto out;
455 break;
456 }
457
458 address = trans_exc_code & __FAIL_ADDR_MASK;
459 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
460 flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
461 if (user_mode(regs))
462 flags |= FAULT_FLAG_USER;
463 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
464 flags |= FAULT_FLAG_WRITE;
465 down_read(&mm->mmap_sem);
466
467 gmap = NULL;
468 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) {
469 gmap = (struct gmap *) S390_lowcore.gmap;
470 current->thread.gmap_addr = address;
471 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE);
472 current->thread.gmap_int_code = regs->int_code & 0xffff;
473 address = __gmap_translate(gmap, address);
474 if (address == -EFAULT) {
475 fault = VM_FAULT_BADMAP;
476 goto out_up;
477 }
478 if (gmap->pfault_enabled)
479 flags |= FAULT_FLAG_RETRY_NOWAIT;
480 }
481
482retry:
483 fault = VM_FAULT_BADMAP;
484 vma = find_vma(mm, address);
485 if (!vma)
486 goto out_up;
487
488 if (unlikely(vma->vm_start > address)) {
489 if (!(vma->vm_flags & VM_GROWSDOWN))
490 goto out_up;
491 if (expand_stack(vma, address))
492 goto out_up;
493 }
494
495 /*
496 * Ok, we have a good vm_area for this memory access, so
497 * we can handle it..
498 */
499 fault = VM_FAULT_BADACCESS;
500 if (unlikely(!(vma->vm_flags & access)))
501 goto out_up;
502
503 if (is_vm_hugetlb_page(vma))
504 address &= HPAGE_MASK;
505 /*
506 * If for any reason at all we couldn't handle the fault,
507 * make sure we exit gracefully rather than endlessly redo
508 * the fault.
509 */
510 fault = handle_mm_fault(vma, address, flags);
511 /* No reason to continue if interrupted by SIGKILL. */
512 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)) {
513 fault = VM_FAULT_SIGNAL;
514 goto out;
515 }
516 if (unlikely(fault & VM_FAULT_ERROR))
517 goto out_up;
518
519 /*
520 * Major/minor page fault accounting is only done on the
521 * initial attempt. If we go through a retry, it is extremely
522 * likely that the page will be found in page cache at that point.
523 */
524 if (flags & FAULT_FLAG_ALLOW_RETRY) {
525 if (fault & VM_FAULT_MAJOR) {
526 tsk->maj_flt++;
527 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
528 regs, address);
529 } else {
530 tsk->min_flt++;
531 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
532 regs, address);
533 }
534 if (fault & VM_FAULT_RETRY) {
535 if (IS_ENABLED(CONFIG_PGSTE) && gmap &&
536 (flags & FAULT_FLAG_RETRY_NOWAIT)) {
537 /* FAULT_FLAG_RETRY_NOWAIT has been set,
538 * mmap_sem has not been released */
539 current->thread.gmap_pfault = 1;
540 fault = VM_FAULT_PFAULT;
541 goto out_up;
542 }
543 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
544 * of starvation. */
545 flags &= ~(FAULT_FLAG_ALLOW_RETRY |
546 FAULT_FLAG_RETRY_NOWAIT);
547 flags |= FAULT_FLAG_TRIED;
548 down_read(&mm->mmap_sem);
549 goto retry;
550 }
551 }
552 if (IS_ENABLED(CONFIG_PGSTE) && gmap) {
553 address = __gmap_link(gmap, current->thread.gmap_addr,
554 address);
555 if (address == -EFAULT) {
556 fault = VM_FAULT_BADMAP;
557 goto out_up;
558 }
559 if (address == -ENOMEM) {
560 fault = VM_FAULT_OOM;
561 goto out_up;
562 }
563 }
564 fault = 0;
565out_up:
566 up_read(&mm->mmap_sem);
567out:
568 return fault;
569}
570
571void do_protection_exception(struct pt_regs *regs)
572{
573 unsigned long trans_exc_code;
574 int access, fault;
575
576 trans_exc_code = regs->int_parm_long;
577 /*
578 * Protection exceptions are suppressing, decrement psw address.
579 * The exception to this rule are aborted transactions, for these
580 * the PSW already points to the correct location.
581 */
582 if (!(regs->int_code & 0x200))
583 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
584 /*
585 * Check for low-address protection. This needs to be treated
586 * as a special case because the translation exception code
587 * field is not guaranteed to contain valid data in this case.
588 */
589 if (unlikely(!(trans_exc_code & 4))) {
590 do_low_address(regs);
591 return;
592 }
593 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) {
594 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) |
595 (regs->psw.addr & PAGE_MASK);
596 access = VM_EXEC;
597 fault = VM_FAULT_BADACCESS;
598 } else {
599 access = VM_WRITE;
600 fault = do_exception(regs, access);
601 }
602 if (unlikely(fault))
603 do_fault_error(regs, access, fault);
604}
605NOKPROBE_SYMBOL(do_protection_exception);
606
607void do_dat_exception(struct pt_regs *regs)
608{
609 int access, fault;
610
611 access = VM_READ | VM_EXEC | VM_WRITE;
612 fault = do_exception(regs, access);
613 if (unlikely(fault))
614 do_fault_error(regs, access, fault);
615}
616NOKPROBE_SYMBOL(do_dat_exception);
617
618#ifdef CONFIG_PFAULT
619/*
620 * 'pfault' pseudo page faults routines.
621 */
622static int pfault_disable;
623
624static int __init nopfault(char *str)
625{
626 pfault_disable = 1;
627 return 1;
628}
629
630__setup("nopfault", nopfault);
631
632struct pfault_refbk {
633 u16 refdiagc;
634 u16 reffcode;
635 u16 refdwlen;
636 u16 refversn;
637 u64 refgaddr;
638 u64 refselmk;
639 u64 refcmpmk;
640 u64 reserved;
641} __attribute__ ((packed, aligned(8)));
642
643int pfault_init(void)
644{
645 struct pfault_refbk refbk = {
646 .refdiagc = 0x258,
647 .reffcode = 0,
648 .refdwlen = 5,
649 .refversn = 2,
650 .refgaddr = __LC_LPP,
651 .refselmk = 1ULL << 48,
652 .refcmpmk = 1ULL << 48,
653 .reserved = __PF_RES_FIELD };
654 int rc;
655
656 if (pfault_disable)
657 return -1;
658 diag_stat_inc(DIAG_STAT_X258);
659 asm volatile(
660 " diag %1,%0,0x258\n"
661 "0: j 2f\n"
662 "1: la %0,8\n"
663 "2:\n"
664 EX_TABLE(0b,1b)
665 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
666 return rc;
667}
668
669void pfault_fini(void)
670{
671 struct pfault_refbk refbk = {
672 .refdiagc = 0x258,
673 .reffcode = 1,
674 .refdwlen = 5,
675 .refversn = 2,
676 };
677
678 if (pfault_disable)
679 return;
680 diag_stat_inc(DIAG_STAT_X258);
681 asm volatile(
682 " diag %0,0,0x258\n"
683 "0: nopr %%r7\n"
684 EX_TABLE(0b,0b)
685 : : "a" (&refbk), "m" (refbk) : "cc");
686}
687
688static DEFINE_SPINLOCK(pfault_lock);
689static LIST_HEAD(pfault_list);
690
691#define PF_COMPLETE 0x0080
692
693/*
694 * The mechanism of our pfault code: if Linux is running as guest, runs a user
695 * space process and the user space process accesses a page that the host has
696 * paged out we get a pfault interrupt.
697 *
698 * This allows us, within the guest, to schedule a different process. Without
699 * this mechanism the host would have to suspend the whole virtual cpu until
700 * the page has been paged in.
701 *
702 * So when we get such an interrupt then we set the state of the current task
703 * to uninterruptible and also set the need_resched flag. Both happens within
704 * interrupt context(!). If we later on want to return to user space we
705 * recognize the need_resched flag and then call schedule(). It's not very
706 * obvious how this works...
707 *
708 * Of course we have a lot of additional fun with the completion interrupt (->
709 * host signals that a page of a process has been paged in and the process can
710 * continue to run). This interrupt can arrive on any cpu and, since we have
711 * virtual cpus, actually appear before the interrupt that signals that a page
712 * is missing.
713 */
714static void pfault_interrupt(struct ext_code ext_code,
715 unsigned int param32, unsigned long param64)
716{
717 struct task_struct *tsk;
718 __u16 subcode;
719 pid_t pid;
720
721 /*
722 * Get the external interruption subcode & pfault initial/completion
723 * signal bit. VM stores this in the 'cpu address' field associated
724 * with the external interrupt.
725 */
726 subcode = ext_code.subcode;
727 if ((subcode & 0xff00) != __SUBCODE_MASK)
728 return;
729 inc_irq_stat(IRQEXT_PFL);
730 /* Get the token (= pid of the affected task). */
731 pid = param64 & LPP_PID_MASK;
732 rcu_read_lock();
733 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
734 if (tsk)
735 get_task_struct(tsk);
736 rcu_read_unlock();
737 if (!tsk)
738 return;
739 spin_lock(&pfault_lock);
740 if (subcode & PF_COMPLETE) {
741 /* signal bit is set -> a page has been swapped in by VM */
742 if (tsk->thread.pfault_wait == 1) {
743 /* Initial interrupt was faster than the completion
744 * interrupt. pfault_wait is valid. Set pfault_wait
745 * back to zero and wake up the process. This can
746 * safely be done because the task is still sleeping
747 * and can't produce new pfaults. */
748 tsk->thread.pfault_wait = 0;
749 list_del(&tsk->thread.list);
750 wake_up_process(tsk);
751 put_task_struct(tsk);
752 } else {
753 /* Completion interrupt was faster than initial
754 * interrupt. Set pfault_wait to -1 so the initial
755 * interrupt doesn't put the task to sleep.
756 * If the task is not running, ignore the completion
757 * interrupt since it must be a leftover of a PFAULT
758 * CANCEL operation which didn't remove all pending
759 * completion interrupts. */
760 if (tsk->state == TASK_RUNNING)
761 tsk->thread.pfault_wait = -1;
762 }
763 } else {
764 /* signal bit not set -> a real page is missing. */
765 if (WARN_ON_ONCE(tsk != current))
766 goto out;
767 if (tsk->thread.pfault_wait == 1) {
768 /* Already on the list with a reference: put to sleep */
769 goto block;
770 } else if (tsk->thread.pfault_wait == -1) {
771 /* Completion interrupt was faster than the initial
772 * interrupt (pfault_wait == -1). Set pfault_wait
773 * back to zero and exit. */
774 tsk->thread.pfault_wait = 0;
775 } else {
776 /* Initial interrupt arrived before completion
777 * interrupt. Let the task sleep.
778 * An extra task reference is needed since a different
779 * cpu may set the task state to TASK_RUNNING again
780 * before the scheduler is reached. */
781 get_task_struct(tsk);
782 tsk->thread.pfault_wait = 1;
783 list_add(&tsk->thread.list, &pfault_list);
784block:
785 /* Since this must be a userspace fault, there
786 * is no kernel task state to trample. Rely on the
787 * return to userspace schedule() to block. */
788 __set_current_state(TASK_UNINTERRUPTIBLE);
789 set_tsk_need_resched(tsk);
790 set_preempt_need_resched();
791 }
792 }
793out:
794 spin_unlock(&pfault_lock);
795 put_task_struct(tsk);
796}
797
798static int pfault_cpu_dead(unsigned int cpu)
799{
800 struct thread_struct *thread, *next;
801 struct task_struct *tsk;
802
803 spin_lock_irq(&pfault_lock);
804 list_for_each_entry_safe(thread, next, &pfault_list, list) {
805 thread->pfault_wait = 0;
806 list_del(&thread->list);
807 tsk = container_of(thread, struct task_struct, thread);
808 wake_up_process(tsk);
809 put_task_struct(tsk);
810 }
811 spin_unlock_irq(&pfault_lock);
812 return 0;
813}
814
815static int __init pfault_irq_init(void)
816{
817 int rc;
818
819 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
820 if (rc)
821 goto out_extint;
822 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP;
823 if (rc)
824 goto out_pfault;
825 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
826 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead",
827 NULL, pfault_cpu_dead);
828 return 0;
829
830out_pfault:
831 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt);
832out_extint:
833 pfault_disable = 1;
834 return rc;
835}
836early_initcall(pfault_irq_init);
837
838#endif /* CONFIG_PFAULT */